/src/netcdf-c/build/libsrc/ncx.c
Line | Count | Source |
1 | | /* Do not edit this file. It is produced from the corresponding .m4 source */ |
2 | | /* |
3 | | * Copyright (C) 2014, Northwestern University and Argonne National Laboratory |
4 | | * See COPYRIGHT notice in top-level directory. |
5 | | */ |
6 | | /* $Id: ncx.m4 2601 2016-11-07 04:54:42Z wkliao $ */ |
7 | | |
8 | | #ifdef __GNUC__ |
9 | | #pragma GCC diagnostic ignored "-Wunused-parameter" |
10 | | #endif |
11 | | |
12 | | |
13 | | |
14 | | |
15 | | |
16 | | |
17 | | #if HAVE_CONFIG_H |
18 | | #include <config.h> |
19 | | #endif |
20 | | #include <config.h> |
21 | | #include <stdio.h> |
22 | | #include <stdlib.h> |
23 | | #include <string.h> |
24 | | #include <limits.h> |
25 | | |
26 | | |
27 | | #pragma GCC diagnostic ignored "-Wdeprecated" |
28 | | #include "ncx.h" |
29 | | #include "nc3dispatch.h" |
30 | | |
31 | | |
32 | | |
33 | | |
34 | | |
35 | | #ifdef HAVE_INTTYPES_H |
36 | | #include <inttypes.h> /* uint16_t, uint32_t, uint64_t */ |
37 | | #elif defined(HAVE_STDINT_H) |
38 | | #include <stdint.h> /* uint16_t, uint32_t, uint64_t */ |
39 | | #endif |
40 | | |
41 | | |
42 | | |
43 | | /* |
44 | | * The only error code returned from subroutines in this file is NC_ERANGE, |
45 | | * if errors are detected. |
46 | | */ |
47 | | |
48 | | /* |
49 | | * An external data representation interface. |
50 | | */ |
51 | | |
52 | | /* alias poorly named limits.h macros */ |
53 | 0 | #define SHORT_MAX SHRT_MAX |
54 | 0 | #define SHORT_MIN SHRT_MIN |
55 | 0 | #define USHORT_MAX USHRT_MAX |
56 | | #ifndef LLONG_MAX |
57 | | # define LLONG_MAX 9223372036854775807LL |
58 | | # define LLONG_MIN (-LLONG_MAX - 1LL) |
59 | | # define ULLONG_MAX 18446744073709551615ULL |
60 | | #endif |
61 | | #ifndef LONG_LONG_MAX |
62 | 0 | #define LONG_LONG_MAX LLONG_MAX |
63 | | #endif |
64 | | #ifndef LONGLONG_MAX |
65 | 0 | #define LONGLONG_MAX LONG_LONG_MAX |
66 | | #endif |
67 | | #ifndef LONG_LONG_MIN |
68 | 0 | #define LONG_LONG_MIN LLONG_MIN |
69 | | #endif |
70 | | #ifndef LONGLONG_MIN |
71 | 0 | #define LONGLONG_MIN LONG_LONG_MIN |
72 | | #endif |
73 | | #ifndef ULONG_LONG_MAX |
74 | 0 | #define ULONG_LONG_MAX ULLONG_MAX |
75 | | #endif |
76 | | #ifndef ULONGLONG_MAX |
77 | 0 | #define ULONGLONG_MAX ULONG_LONG_MAX |
78 | | #endif |
79 | | #include <float.h> |
80 | | #ifndef FLT_MAX /* This POSIX macro missing on some systems */ |
81 | | # ifndef NO_IEEE_FLOAT |
82 | | # define FLT_MAX 3.40282347e+38f |
83 | | # else |
84 | | # error "You will need to define FLT_MAX" |
85 | | # endif |
86 | | #endif |
87 | | /* alias poorly named float.h macros */ |
88 | | #define FLOAT_MAX FLT_MAX |
89 | | #define FLOAT_MIN (-FLT_MAX) |
90 | | #define DOUBLE_MAX DBL_MAX |
91 | | #define DOUBLE_MIN (-DBL_MAX) |
92 | | #define FLOAT_MAX_EXP FLT_MAX_EXP |
93 | | #define DOUBLE_MAX_EXP DBL_MAX_EXP |
94 | | #include <assert.h> |
95 | | #define UCHAR_MIN 0 |
96 | | #define Min(a,b) ((a) < (b) ? (a) : (b)) |
97 | | #define Max(a,b) ((a) > (b) ? (a) : (b)) |
98 | | |
99 | | #ifndef SIZEOF_UCHAR |
100 | | #ifdef SIZEOF_UNSIGNED_CHAR |
101 | | #define SIZEOF_UCHAR SIZEOF_UNSIGNED_CHAR |
102 | | #else |
103 | | #error "unknown SIZEOF_UCHAR" |
104 | | #endif |
105 | | #endif |
106 | | |
107 | | #ifndef SIZEOF_USHORT |
108 | | #ifdef SIZEOF_UNSIGNED_SHORT_INT |
109 | | #define SIZEOF_USHORT SIZEOF_UNSIGNED_SHORT_INT |
110 | | #elif defined(SIZEOF_UNSIGNED_SHORT) |
111 | | #define SIZEOF_USHORT SIZEOF_UNSIGNED_SHORT |
112 | | #else |
113 | | #error "unknown SIZEOF_USHORT" |
114 | | #endif |
115 | | #endif |
116 | | |
117 | | #ifndef SIZEOF_UINT |
118 | | #ifdef SIZEOF_UNSIGNED_INT |
119 | | #define SIZEOF_UINT SIZEOF_UNSIGNED_INT |
120 | | #else |
121 | | #error "unknown SIZEOF_UINT" |
122 | | #endif |
123 | | #endif |
124 | | |
125 | | #ifndef SIZEOF_LONGLONG |
126 | | #ifdef SIZEOF_LONG_LONG |
127 | | #define SIZEOF_LONGLONG SIZEOF_LONG_LONG |
128 | | #else |
129 | | #error "unknown SIZEOF_LONGLONG" |
130 | | #endif |
131 | | #endif |
132 | | |
133 | | #ifndef SIZEOF_INT64 |
134 | | #ifdef SIZEOF_LONG_LONG |
135 | | #define SIZEOF_INT64 SIZEOF_LONG_LONG |
136 | | #elif defined(SIZEOF_LONGLONG) |
137 | | #define SIZEOF_INT64 SIZEOF_LONGLONG |
138 | | #else |
139 | | #error "unknown SIZEOF_INT64" |
140 | | #endif |
141 | | #endif |
142 | | |
143 | | #ifndef SIZEOF_ULONGLONG |
144 | | #ifdef SIZEOF_UNSIGNED_LONG_LONG |
145 | | #define SIZEOF_ULONGLONG SIZEOF_UNSIGNED_LONG_LONG |
146 | | #else |
147 | | #error "unknown SIZEOF_ULONGLONG" |
148 | | #endif |
149 | | #endif |
150 | | |
151 | | #ifndef SIZEOF_UINT64 |
152 | | #ifdef SIZEOF_UNSIGNED_LONG_LONG |
153 | | #define SIZEOF_UINT64 SIZEOF_UNSIGNED_LONG_LONG |
154 | | #elif defined(SIZEOF_ULONGLONG) |
155 | | #define SIZEOF_UINT64 SIZEOF_ULONGLONG |
156 | | #else |
157 | | #error "unknown SIZEOF_UINT64" |
158 | | #endif |
159 | | #endif |
160 | | |
161 | | /* |
162 | | * If the machine's float domain is "smaller" than the external one |
163 | | * use the machine domain |
164 | | */ |
165 | | #if defined(FLT_MAX_EXP) && FLT_MAX_EXP < 128 /* 128 is X_FLT_MAX_EXP */ |
166 | | #undef X_FLOAT_MAX |
167 | | # define X_FLOAT_MAX FLT_MAX |
168 | | #undef X_FLOAT_MIN |
169 | | # define X_FLOAT_MIN (-X_FLOAT_MAX) |
170 | | #endif |
171 | | |
172 | | #if defined(_SX) && _SX != 0 /* NEC SUPER UX */ |
173 | | #define LOOPCNT 256 /* must be no longer than hardware vector length */ |
174 | | #if _INT64 |
175 | | #undef INT_MAX /* workaround cpp bug */ |
176 | | #define INT_MAX X_INT_MAX |
177 | | #undef INT_MIN /* workaround cpp bug */ |
178 | | #define INT_MIN X_INT_MIN |
179 | | #undef LONG_MAX /* workaround cpp bug */ |
180 | | #define LONG_MAX X_INT_MAX |
181 | | #undef LONG_MIN /* workaround cpp bug */ |
182 | | #define LONG_MIN X_INT_MIN |
183 | | #elif _LONG64 |
184 | | #undef LONG_MAX /* workaround cpp bug */ |
185 | | #define LONG_MAX 4294967295L |
186 | | #undef LONG_MIN /* workaround cpp bug */ |
187 | | #define LONG_MIN -4294967295L |
188 | | #endif |
189 | | #if !_FLOAT0 |
190 | | #error "FLOAT1 and FLOAT2 not supported" |
191 | | #endif |
192 | | #endif /* _SX */ |
193 | | |
194 | | static const char nada[X_ALIGN] = {0, 0, 0, 0}; |
195 | | |
196 | | #ifndef WORDS_BIGENDIAN |
197 | | /* LITTLE_ENDIAN: DEC and intel */ |
198 | | /* |
199 | | * Routines to convert to BIG ENDIAN. |
200 | | * Optimize the swapn?b() and swap?b() routines aggressively. |
201 | | */ |
202 | | |
203 | 0 | #define SWAP2(a) ( (((a) & 0xff) << 8) | \ |
204 | 0 | (((a) >> 8) & 0xff) ) |
205 | | |
206 | 39.2k | #define SWAP4(a) ( ((a) << 24) | \ |
207 | 39.2k | (((a) << 8) & 0x00ff0000) | \ |
208 | 39.2k | (((a) >> 8) & 0x0000ff00) | \ |
209 | 39.2k | (((a) >> 24) & 0x000000ff) ) |
210 | | |
211 | 0 | #define SWAP8(a) ( (((a) & 0x00000000000000FFULL) << 56) | \ |
212 | 0 | (((a) & 0x000000000000FF00ULL) << 40) | \ |
213 | 0 | (((a) & 0x0000000000FF0000ULL) << 24) | \ |
214 | 0 | (((a) & 0x00000000FF000000ULL) << 8) | \ |
215 | 0 | (((a) & 0x000000FF00000000ULL) >> 8) | \ |
216 | 0 | (((a) & 0x0000FF0000000000ULL) >> 24) | \ |
217 | 0 | (((a) & 0x00FF000000000000ULL) >> 40) | \ |
218 | 0 | (((a) & 0xFF00000000000000ULL) >> 56) ) |
219 | | |
220 | | #if defined(_MSC_VER) && _MSC_VER < 1900 |
221 | | #define inline __inline |
222 | | #endif |
223 | | |
224 | | inline static void |
225 | | swapn2b(void *dst, const void *src, size_t nn) |
226 | 0 | { |
227 | | /* it is OK if dst == src */ |
228 | 0 | size_t i; |
229 | 0 | char *op = (char*) dst; |
230 | 0 | char *ip = (char*) src; |
231 | 0 | uint16_t tmp; |
232 | 0 | for (i=0; i<nn; i++) { |
233 | | /* memcpy is used to handle the case of unaligned memory */ |
234 | 0 | memcpy(&tmp, ip, sizeof(tmp)); |
235 | 0 | tmp = SWAP2(tmp); |
236 | 0 | memcpy(op, &tmp, sizeof(tmp)); |
237 | 0 | ip += sizeof(uint16_t); |
238 | 0 | op += sizeof(uint16_t); |
239 | 0 | } |
240 | 0 | } |
241 | | |
242 | | # ifndef vax |
243 | | inline static void |
244 | | swap4b(void *dst, const void *src) |
245 | 0 | { |
246 | 0 | uint32_t tmp; |
247 | | /* memcpy is used to handle the case of unaligned memory */ |
248 | 0 | memcpy(&tmp, src, sizeof(tmp)); |
249 | 0 | tmp = SWAP4(tmp); |
250 | 0 | memcpy(dst, &tmp, 4); |
251 | 0 | } |
252 | | # endif /* !vax */ |
253 | | |
254 | | inline static void |
255 | | swapn4b(void *dst, const void *src, size_t nn) |
256 | 131k | { |
257 | 131k | size_t i; |
258 | 131k | char *op = (char*) dst; |
259 | 131k | char *ip = (char*) src; |
260 | 131k | uint32_t tmp; |
261 | 170k | for (i=0; i<nn; i++) { |
262 | | /* memcpy is used to handle the case of unaligned memory */ |
263 | 39.2k | memcpy(&tmp, ip, sizeof(tmp)); |
264 | 39.2k | tmp = SWAP4(tmp); |
265 | 39.2k | memcpy(op, &tmp, sizeof(tmp)); |
266 | 39.2k | ip += sizeof(uint32_t); |
267 | 39.2k | op += sizeof(uint32_t); |
268 | 39.2k | } |
269 | 131k | } |
270 | | |
271 | | # ifndef vax |
272 | | inline static void |
273 | | swap8b(void *dst, const void *src) |
274 | 0 | { |
275 | 0 | uint64_t tmp; |
276 | | /* memcpy is used to handle the case of unaligned memory */ |
277 | 0 | memcpy(&tmp, src, sizeof(tmp)); |
278 | 0 | tmp = SWAP8(tmp); |
279 | 0 | memcpy(dst, &tmp, 8); |
280 | 0 | } |
281 | | # endif /* !vax */ |
282 | | |
283 | | # ifndef vax |
284 | | inline static void |
285 | | swapn8b(void *dst, const void *src, size_t nn) |
286 | 0 | { |
287 | 0 | size_t i; |
288 | 0 | char *op = (char*) dst; |
289 | 0 | char *ip = (char*) src; |
290 | 0 | uint64_t tmp; |
291 | 0 | for (i=0; i<nn; i++) { |
292 | | /* memcpy is used to handle the case of unaligned memory */ |
293 | 0 | memcpy(&tmp, ip, sizeof(tmp)); |
294 | 0 | tmp = SWAP8(tmp); |
295 | 0 | memcpy(op, &tmp, sizeof(tmp)); |
296 | 0 | ip += sizeof(uint64_t); |
297 | 0 | op += sizeof(uint64_t); |
298 | 0 | } |
299 | 0 | } |
300 | | # endif /* !vax */ |
301 | | |
302 | | #endif /* LITTLE_ENDIAN */ |
303 | | |
304 | | |
305 | | |
306 | | |
307 | | |
308 | | |
309 | | /* |
310 | | * Primitive numeric conversion functions. |
311 | | */ |
312 | | |
313 | | |
314 | | |
315 | | |
316 | | |
317 | | /* x_schar */ |
318 | | /* x_uchar */ |
319 | | |
320 | | /* We don't implement any x_schar and x_uchar primitives. */ |
321 | | |
322 | | |
323 | | /* external NC_SHORT --------------------------------------------------------*/ |
324 | | |
325 | | #if SHORT_MAX == X_SHORT_MAX |
326 | | typedef short ix_short; |
327 | | #define SIZEOF_IX_SHORT SIZEOF_SHORT |
328 | 0 | #define IX_SHORT_MAX SHORT_MAX |
329 | | #elif INT_MAX >= X_SHORT_MAX |
330 | | typedef int ix_short; |
331 | | #define SIZEOF_IX_SHORT SIZEOF_INT |
332 | | #define IX_SHORT_MAX INT_MAX |
333 | | #elif LONG_MAX >= X_SHORT_MAX |
334 | | typedef long ix_short; |
335 | | #define SIZEOF_IX_SHORT SIZEOF_LONG |
336 | | #define IX_SHORT_MAX LONG_MAX |
337 | | #elif LLONG_MAX >= X_SHORT_MAX |
338 | | typedef long long ix_short; |
339 | | #define SIZEOF_IX_SHORT SIZEOF_LONGLONG |
340 | | #define IX_SHORT_MAX LLONG_MAX |
341 | | #else |
342 | | #error "ix_short implementation" |
343 | | #endif |
344 | | |
345 | | static void |
346 | | get_ix_short(const void *xp, ix_short *ip) |
347 | 0 | { |
348 | 0 | const uchar *cp = (const uchar *) xp; |
349 | 0 | *ip = (ix_short)(*cp++ << 8); |
350 | | #if SIZEOF_IX_SHORT > X_SIZEOF_SHORT |
351 | | if (*ip & 0x8000) |
352 | | { |
353 | | /* extern is negative */ |
354 | | *ip |= (~(0xffff)); /* N.B. Assumes "twos complement" */ |
355 | | } |
356 | | #endif |
357 | 0 | *ip = (ix_short)(*ip | *cp); |
358 | 0 | } |
359 | | |
360 | | static void |
361 | | put_ix_short(void *xp, const ix_short *ip) |
362 | 0 | { |
363 | 0 | uchar *cp = (uchar *) xp; |
364 | 0 | *cp++ = (uchar)((*ip) >> 8); |
365 | 0 | *cp = (uchar)((*ip) & 0xff); |
366 | 0 | } |
367 | | |
368 | | static int |
369 | | ncx_get_short_schar(const void *xp, schar *ip) |
370 | 0 | { |
371 | 0 | int err=NC_NOERR; |
372 | 0 | ix_short xx = 0; |
373 | 0 | get_ix_short(xp, &xx); |
374 | |
|
375 | 0 | #if IX_SHORT_MAX > SCHAR_MAX |
376 | 0 | if (xx > SCHAR_MAX || xx < SCHAR_MIN) { |
377 | 0 | #ifdef ERANGE_FILL |
378 | 0 | *ip = NC_FILL_BYTE; |
379 | 0 | return NC_ERANGE; |
380 | | #else |
381 | | err = NC_ERANGE; |
382 | | #endif |
383 | 0 | } |
384 | 0 | #endif |
385 | | |
386 | | |
387 | 0 | *ip = (schar) xx; |
388 | 0 | return err; |
389 | 0 | } |
390 | | |
391 | | static int |
392 | | ncx_get_short_short(const void *xp, short *ip) |
393 | 0 | { |
394 | 0 | int err=NC_NOERR; |
395 | 0 | #if SIZEOF_IX_SHORT == SIZEOF_SHORT && IX_SHORT_MAX == SHORT_MAX |
396 | 0 | get_ix_short(xp, (ix_short *)ip); |
397 | | #else |
398 | | ix_short xx = 0; |
399 | | get_ix_short(xp, &xx); |
400 | | |
401 | | #if IX_SHORT_MAX > SHORT_MAX |
402 | | if (xx > SHORT_MAX || xx < SHORT_MIN) { |
403 | | #ifdef ERANGE_FILL |
404 | | *ip = NC_FILL_SHORT; |
405 | | return NC_ERANGE; |
406 | | #else |
407 | | err = NC_ERANGE; |
408 | | #endif |
409 | | } |
410 | | #endif |
411 | | |
412 | | |
413 | | *ip = (short) xx; |
414 | | #endif |
415 | 0 | return err; |
416 | 0 | } |
417 | | |
418 | | static int |
419 | | ncx_get_short_int(const void *xp, int *ip) |
420 | 0 | { |
421 | 0 | int err=NC_NOERR; |
422 | | #if SIZEOF_IX_SHORT == SIZEOF_INT && IX_SHORT_MAX == INT_MAX |
423 | | get_ix_short(xp, (ix_short *)ip); |
424 | | #else |
425 | 0 | ix_short xx = 0; |
426 | 0 | get_ix_short(xp, &xx); |
427 | |
|
428 | | #if IX_SHORT_MAX > INT_MAX |
429 | | if (xx > INT_MAX || xx < INT_MIN) { |
430 | | #ifdef ERANGE_FILL |
431 | | *ip = NC_FILL_INT; |
432 | | return NC_ERANGE; |
433 | | #else |
434 | | err = NC_ERANGE; |
435 | | #endif |
436 | | } |
437 | | #endif |
438 | | |
439 | |
|
440 | 0 | *ip = (int) xx; |
441 | 0 | #endif |
442 | 0 | return err; |
443 | 0 | } |
444 | | |
445 | | static int |
446 | | ncx_get_short_long(const void *xp, long *ip) |
447 | 0 | { |
448 | 0 | int err=NC_NOERR; |
449 | | #if SIZEOF_IX_SHORT == SIZEOF_LONG && IX_SHORT_MAX == LONG_MAX |
450 | | get_ix_short(xp, (ix_short *)ip); |
451 | | #else |
452 | 0 | ix_short xx = 0; |
453 | 0 | get_ix_short(xp, &xx); |
454 | |
|
455 | | #if IX_SHORT_MAX > LONG_MAX |
456 | | if (xx > LONG_MAX || xx < LONG_MIN) { |
457 | | #ifdef ERANGE_FILL |
458 | | *ip = NC_FILL_INT; |
459 | | return NC_ERANGE; |
460 | | #else |
461 | | err = NC_ERANGE; |
462 | | #endif |
463 | | } |
464 | | #endif |
465 | | |
466 | |
|
467 | 0 | *ip = (long) xx; |
468 | 0 | #endif |
469 | 0 | return err; |
470 | 0 | } |
471 | | |
472 | | static int |
473 | | ncx_get_short_longlong(const void *xp, longlong *ip) |
474 | 0 | { |
475 | 0 | int err=NC_NOERR; |
476 | | #if SIZEOF_IX_SHORT == SIZEOF_LONGLONG && IX_SHORT_MAX == LONGLONG_MAX |
477 | | get_ix_short(xp, (ix_short *)ip); |
478 | | #else |
479 | 0 | ix_short xx = 0; |
480 | 0 | get_ix_short(xp, &xx); |
481 | |
|
482 | | #if IX_SHORT_MAX > LONGLONG_MAX |
483 | | if (xx > LONGLONG_MAX || xx < LONGLONG_MIN) { |
484 | | #ifdef ERANGE_FILL |
485 | | *ip = NC_FILL_INT64; |
486 | | return NC_ERANGE; |
487 | | #else |
488 | | err = NC_ERANGE; |
489 | | #endif |
490 | | } |
491 | | #endif |
492 | | |
493 | |
|
494 | 0 | *ip = (longlong) xx; |
495 | 0 | #endif |
496 | 0 | return err; |
497 | 0 | } |
498 | | |
499 | | static int |
500 | | ncx_get_short_ushort(const void *xp, ushort *ip) |
501 | 0 | { |
502 | 0 | int err=NC_NOERR; |
503 | 0 | ix_short xx = 0; |
504 | 0 | get_ix_short(xp, &xx); |
505 | |
|
506 | | #if IX_SHORT_MAX > USHORT_MAX |
507 | | if (xx > USHORT_MAX) { |
508 | | #ifdef ERANGE_FILL |
509 | | *ip = NC_FILL_USHORT; |
510 | | return NC_ERANGE; |
511 | | #else |
512 | | err = NC_ERANGE; |
513 | | #endif |
514 | | } |
515 | | #endif |
516 | |
|
517 | 0 | if (xx < 0) { |
518 | 0 | #ifdef ERANGE_FILL |
519 | 0 | *ip = NC_FILL_USHORT; |
520 | 0 | return NC_ERANGE; |
521 | | #else |
522 | | err = NC_ERANGE; /* because ip is unsigned */ |
523 | | #endif |
524 | 0 | } |
525 | 0 | *ip = (ushort) xx; |
526 | 0 | return err; |
527 | 0 | } |
528 | | |
529 | | static int |
530 | | ncx_get_short_uchar(const void *xp, uchar *ip) |
531 | 0 | { |
532 | 0 | int err=NC_NOERR; |
533 | 0 | ix_short xx = 0; |
534 | 0 | get_ix_short(xp, &xx); |
535 | |
|
536 | 0 | #if IX_SHORT_MAX > UCHAR_MAX |
537 | 0 | if (xx > UCHAR_MAX) { |
538 | 0 | #ifdef ERANGE_FILL |
539 | 0 | *ip = NC_FILL_UBYTE; |
540 | 0 | return NC_ERANGE; |
541 | | #else |
542 | | err = NC_ERANGE; |
543 | | #endif |
544 | 0 | } |
545 | 0 | #endif |
546 | | |
547 | 0 | if (xx < 0) { |
548 | 0 | #ifdef ERANGE_FILL |
549 | 0 | *ip = NC_FILL_UBYTE; |
550 | 0 | return NC_ERANGE; |
551 | | #else |
552 | | err = NC_ERANGE; /* because ip is unsigned */ |
553 | | #endif |
554 | 0 | } |
555 | 0 | *ip = (uchar) xx; |
556 | 0 | return err; |
557 | 0 | } |
558 | | |
559 | | static int |
560 | | ncx_get_short_uint(const void *xp, uint *ip) |
561 | 0 | { |
562 | 0 | int err=NC_NOERR; |
563 | 0 | ix_short xx = 0; |
564 | 0 | get_ix_short(xp, &xx); |
565 | |
|
566 | | #if IX_SHORT_MAX > UINT_MAX |
567 | | if (xx > UINT_MAX) { |
568 | | #ifdef ERANGE_FILL |
569 | | *ip = NC_FILL_UINT; |
570 | | return NC_ERANGE; |
571 | | #else |
572 | | err = NC_ERANGE; |
573 | | #endif |
574 | | } |
575 | | #endif |
576 | |
|
577 | 0 | if (xx < 0) { |
578 | 0 | #ifdef ERANGE_FILL |
579 | 0 | *ip = NC_FILL_UINT; |
580 | 0 | return NC_ERANGE; |
581 | | #else |
582 | | err = NC_ERANGE; /* because ip is unsigned */ |
583 | | #endif |
584 | 0 | } |
585 | 0 | *ip = (uint) xx; |
586 | 0 | return err; |
587 | 0 | } |
588 | | |
589 | | static int |
590 | | ncx_get_short_ulonglong(const void *xp, ulonglong *ip) |
591 | 0 | { |
592 | 0 | int err=NC_NOERR; |
593 | 0 | ix_short xx = 0; |
594 | 0 | get_ix_short(xp, &xx); |
595 | |
|
596 | | #if IX_SHORT_MAX > ULONGLONG_MAX |
597 | | if (xx > ULONGLONG_MAX) { |
598 | | #ifdef ERANGE_FILL |
599 | | *ip = NC_FILL_UINT64; |
600 | | return NC_ERANGE; |
601 | | #else |
602 | | err = NC_ERANGE; |
603 | | #endif |
604 | | } |
605 | | #endif |
606 | |
|
607 | 0 | if (xx < 0) { |
608 | 0 | #ifdef ERANGE_FILL |
609 | 0 | *ip = NC_FILL_UINT64; |
610 | 0 | return NC_ERANGE; |
611 | | #else |
612 | | err = NC_ERANGE; /* because ip is unsigned */ |
613 | | #endif |
614 | 0 | } |
615 | 0 | *ip = (ulonglong) xx; |
616 | 0 | return err; |
617 | 0 | } |
618 | | |
619 | | static int |
620 | | ncx_get_short_float(const void *xp, float *ip) |
621 | 0 | { |
622 | 0 | ix_short xx = 0; |
623 | 0 | get_ix_short(xp, &xx); |
624 | 0 | *ip = (float)xx; |
625 | 0 | return NC_NOERR; |
626 | 0 | } |
627 | | |
628 | | static int |
629 | | ncx_get_short_double(const void *xp, double *ip) |
630 | 0 | { |
631 | 0 | ix_short xx = 0; |
632 | 0 | get_ix_short(xp, &xx); |
633 | 0 | *ip = (double)xx; |
634 | 0 | return NC_NOERR; |
635 | 0 | } |
636 | | |
637 | | |
638 | | static int |
639 | | ncx_put_short_schar(void *xp, const schar *ip, void *fillp) |
640 | 0 | { |
641 | 0 | uchar *cp = (uchar *) xp; |
642 | 0 | if (*ip & 0x80) |
643 | 0 | *cp++ = 0xff; |
644 | 0 | else |
645 | 0 | *cp++ = 0; |
646 | 0 | *cp = (uchar)*ip; |
647 | 0 | return NC_NOERR; |
648 | 0 | } |
649 | | |
650 | | static int |
651 | | ncx_put_short_uchar(void *xp, const uchar *ip, void *fillp) |
652 | 0 | { |
653 | 0 | uchar *cp = (uchar *) xp; |
654 | 0 | *cp++ = 0; |
655 | 0 | *cp = *ip; |
656 | 0 | return NC_NOERR; |
657 | 0 | } |
658 | | |
659 | | static int |
660 | | ncx_put_short_short(void *xp, const short *ip, void *fillp) |
661 | 0 | { |
662 | 0 | int err=NC_NOERR; |
663 | 0 | #if SIZEOF_IX_SHORT == SIZEOF_SHORT && IX_SHORT_MAX == SHORT_MAX |
664 | 0 | put_ix_short(xp, (const ix_short *)ip); |
665 | | #else |
666 | | ix_short xx = NC_FILL_SHORT; |
667 | | |
668 | | #if IX_SHORT_MAX < SHORT_MAX |
669 | | if (*ip > IX_SHORT_MAX || *ip < X_SHORT_MIN) { |
670 | | |
671 | | #ifdef ERANGE_FILL |
672 | | if (fillp != NULL) memcpy(&xx, fillp, 2); |
673 | | #endif |
674 | | err = NC_ERANGE; |
675 | | } |
676 | | #ifdef ERANGE_FILL |
677 | | else |
678 | | #endif |
679 | | #endif |
680 | | xx = (ix_short)*ip; |
681 | | |
682 | | put_ix_short(xp, &xx); |
683 | | #endif |
684 | 0 | return err; |
685 | 0 | } |
686 | | |
687 | | static int |
688 | | ncx_put_short_int(void *xp, const int *ip, void *fillp) |
689 | 0 | { |
690 | 0 | int err=NC_NOERR; |
691 | | #if SIZEOF_IX_SHORT == SIZEOF_INT && IX_SHORT_MAX == INT_MAX |
692 | | put_ix_short(xp, (const ix_short *)ip); |
693 | | #else |
694 | 0 | ix_short xx = NC_FILL_SHORT; |
695 | |
|
696 | 0 | #if IX_SHORT_MAX < INT_MAX |
697 | 0 | if (*ip > IX_SHORT_MAX || *ip < X_SHORT_MIN) { |
698 | | |
699 | 0 | #ifdef ERANGE_FILL |
700 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
701 | 0 | #endif |
702 | 0 | err = NC_ERANGE; |
703 | 0 | } |
704 | 0 | #ifdef ERANGE_FILL |
705 | 0 | else |
706 | 0 | #endif |
707 | 0 | #endif |
708 | 0 | xx = (ix_short)*ip; |
709 | |
|
710 | 0 | put_ix_short(xp, &xx); |
711 | 0 | #endif |
712 | 0 | return err; |
713 | 0 | } |
714 | | |
715 | | static int |
716 | | ncx_put_short_long(void *xp, const long *ip, void *fillp) |
717 | 0 | { |
718 | 0 | int err=NC_NOERR; |
719 | | #if SIZEOF_IX_SHORT == SIZEOF_LONG && IX_SHORT_MAX == LONG_MAX |
720 | | put_ix_short(xp, (const ix_short *)ip); |
721 | | #else |
722 | 0 | ix_short xx = NC_FILL_SHORT; |
723 | |
|
724 | 0 | #if IX_SHORT_MAX < LONG_MAX |
725 | 0 | if (*ip > IX_SHORT_MAX || *ip < X_SHORT_MIN) { |
726 | | |
727 | 0 | #ifdef ERANGE_FILL |
728 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
729 | 0 | #endif |
730 | 0 | err = NC_ERANGE; |
731 | 0 | } |
732 | 0 | #ifdef ERANGE_FILL |
733 | 0 | else |
734 | 0 | #endif |
735 | 0 | #endif |
736 | 0 | xx = (ix_short)*ip; |
737 | |
|
738 | 0 | put_ix_short(xp, &xx); |
739 | 0 | #endif |
740 | 0 | return err; |
741 | 0 | } |
742 | | |
743 | | static int |
744 | | ncx_put_short_longlong(void *xp, const longlong *ip, void *fillp) |
745 | 0 | { |
746 | 0 | int err=NC_NOERR; |
747 | | #if SIZEOF_IX_SHORT == SIZEOF_LONGLONG && IX_SHORT_MAX == LONGLONG_MAX |
748 | | put_ix_short(xp, (const ix_short *)ip); |
749 | | #else |
750 | 0 | ix_short xx = NC_FILL_SHORT; |
751 | |
|
752 | 0 | #if IX_SHORT_MAX < LONGLONG_MAX |
753 | 0 | if (*ip > IX_SHORT_MAX || *ip < X_SHORT_MIN) { |
754 | | |
755 | 0 | #ifdef ERANGE_FILL |
756 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
757 | 0 | #endif |
758 | 0 | err = NC_ERANGE; |
759 | 0 | } |
760 | 0 | #ifdef ERANGE_FILL |
761 | 0 | else |
762 | 0 | #endif |
763 | 0 | #endif |
764 | 0 | xx = (ix_short)*ip; |
765 | |
|
766 | 0 | put_ix_short(xp, &xx); |
767 | 0 | #endif |
768 | 0 | return err; |
769 | 0 | } |
770 | | |
771 | | static int |
772 | | ncx_put_short_ushort(void *xp, const ushort *ip, void *fillp) |
773 | 0 | { |
774 | 0 | int err=NC_NOERR; |
775 | 0 | ix_short xx = NC_FILL_SHORT; |
776 | |
|
777 | 0 | #if IX_SHORT_MAX < USHORT_MAX |
778 | 0 | if (*ip > IX_SHORT_MAX) { |
779 | | |
780 | 0 | #ifdef ERANGE_FILL |
781 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
782 | 0 | #endif |
783 | 0 | err = NC_ERANGE; |
784 | 0 | } |
785 | 0 | #ifdef ERANGE_FILL |
786 | 0 | else |
787 | 0 | #endif |
788 | 0 | #endif |
789 | 0 | xx = (ix_short)*ip; |
790 | |
|
791 | 0 | put_ix_short(xp, &xx); |
792 | 0 | return err; |
793 | 0 | } |
794 | | |
795 | | static int |
796 | | ncx_put_short_uint(void *xp, const uint *ip, void *fillp) |
797 | 0 | { |
798 | 0 | int err=NC_NOERR; |
799 | 0 | ix_short xx = NC_FILL_SHORT; |
800 | |
|
801 | 0 | #if IX_SHORT_MAX < UINT_MAX |
802 | 0 | if (*ip > IX_SHORT_MAX) { |
803 | | |
804 | 0 | #ifdef ERANGE_FILL |
805 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
806 | 0 | #endif |
807 | 0 | err = NC_ERANGE; |
808 | 0 | } |
809 | 0 | #ifdef ERANGE_FILL |
810 | 0 | else |
811 | 0 | #endif |
812 | 0 | #endif |
813 | 0 | xx = (ix_short)*ip; |
814 | |
|
815 | 0 | put_ix_short(xp, &xx); |
816 | 0 | return err; |
817 | 0 | } |
818 | | |
819 | | static int |
820 | | ncx_put_short_ulonglong(void *xp, const ulonglong *ip, void *fillp) |
821 | 0 | { |
822 | 0 | int err=NC_NOERR; |
823 | 0 | ix_short xx = NC_FILL_SHORT; |
824 | |
|
825 | 0 | #if IX_SHORT_MAX < ULONGLONG_MAX |
826 | 0 | if (*ip > IX_SHORT_MAX) { |
827 | | |
828 | 0 | #ifdef ERANGE_FILL |
829 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
830 | 0 | #endif |
831 | 0 | err = NC_ERANGE; |
832 | 0 | } |
833 | 0 | #ifdef ERANGE_FILL |
834 | 0 | else |
835 | 0 | #endif |
836 | 0 | #endif |
837 | 0 | xx = (ix_short)*ip; |
838 | |
|
839 | 0 | put_ix_short(xp, &xx); |
840 | 0 | return err; |
841 | 0 | } |
842 | | |
843 | | static int |
844 | | ncx_put_short_float(void *xp, const float *ip, void *fillp) |
845 | 0 | { |
846 | 0 | int err=NC_NOERR; |
847 | 0 | ix_short xx = NC_FILL_SHORT; |
848 | |
|
849 | 0 | if (*ip > (double)X_SHORT_MAX || *ip < (double)X_SHORT_MIN) { |
850 | | |
851 | 0 | #ifdef ERANGE_FILL |
852 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
853 | 0 | #endif |
854 | 0 | err = NC_ERANGE; |
855 | 0 | } |
856 | 0 | #ifdef ERANGE_FILL |
857 | 0 | else |
858 | 0 | #endif |
859 | 0 | xx = (ix_short)*ip; |
860 | |
|
861 | 0 | put_ix_short(xp, &xx); |
862 | 0 | return err; |
863 | 0 | } |
864 | | |
865 | | static int |
866 | | ncx_put_short_double(void *xp, const double *ip, void *fillp) |
867 | 0 | { |
868 | 0 | int err=NC_NOERR; |
869 | 0 | ix_short xx = NC_FILL_SHORT; |
870 | |
|
871 | 0 | if (*ip > X_SHORT_MAX || *ip < X_SHORT_MIN) { |
872 | | |
873 | 0 | #ifdef ERANGE_FILL |
874 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
875 | 0 | #endif |
876 | 0 | err = NC_ERANGE; |
877 | 0 | } |
878 | 0 | #ifdef ERANGE_FILL |
879 | 0 | else |
880 | 0 | #endif |
881 | 0 | xx = (ix_short)*ip; |
882 | |
|
883 | 0 | put_ix_short(xp, &xx); |
884 | 0 | return err; |
885 | 0 | } |
886 | | |
887 | | |
888 | | /* external NC_USHORT -------------------------------------------------------*/ |
889 | | |
890 | | #if USHORT_MAX == X_USHORT_MAX |
891 | | typedef unsigned short ix_ushort; |
892 | | #define SIZEOF_IX_USHORT SIZEOF_USHORT |
893 | 0 | #define IX_USHORT_MAX USHORT_MAX |
894 | | #elif UINT_MAX >= X_USHORT_MAX |
895 | | typedef unsigned int ix_ushort; |
896 | | #define SIZEOF_IX_USHORT SIZEOF_UINT |
897 | | #define IX_USHORT_MAX UINT_MAX |
898 | | #elif ULONG_MAX >= X_USHORT_MAX |
899 | | typedef unsigned long ix_ushort; |
900 | | #define SIZEOF_IX_USHORT SIZEOF_ULONG |
901 | | #define IX_USHORT_MAX ULONG_MAX |
902 | | #elif ULLONG_MAX >= X_USHORT_MAX |
903 | | typedef unsigned long long ix_ushort; |
904 | | #define SIZEOF_IX_USHORT SIZEOF_ULONGLONG |
905 | | #define IX_USHORT_MAX ULLONG_MAX |
906 | | #else |
907 | | #error "ix_ushort implementation" |
908 | | #endif |
909 | | |
910 | | static void |
911 | | get_ix_ushort(const void *xp, ix_ushort *ip) |
912 | 0 | { |
913 | 0 | const uchar *cp = (const uchar *) xp; |
914 | 0 | *ip = (ix_ushort)(*cp++ << 8); |
915 | | #if SIZEOF_IX_SHORT > X_SIZEOF_SHORT |
916 | | if (*ip & 0x8000) |
917 | | { |
918 | | /* extern is negative */ |
919 | | *ip |= (~(0xffff)); /* N.B. Assumes "twos complement" */ |
920 | | } |
921 | | #endif |
922 | 0 | *ip = (ix_ushort)(*ip | *cp); |
923 | 0 | } |
924 | | |
925 | | static void |
926 | | put_ix_ushort(void *xp, const ix_ushort *ip) |
927 | 0 | { |
928 | 0 | uchar *cp = (uchar *) xp; |
929 | 0 | *cp++ = (uchar)((*ip) >> 8); |
930 | 0 | *cp = (uchar)((*ip) & 0xff); |
931 | 0 | } |
932 | | |
933 | | static int |
934 | | ncx_get_ushort_schar(const void *xp, schar *ip) |
935 | 0 | { |
936 | 0 | int err=NC_NOERR; |
937 | 0 | ix_ushort xx = 0; |
938 | 0 | get_ix_ushort(xp, &xx); |
939 | |
|
940 | 0 | #if IX_USHORT_MAX > SCHAR_MAX |
941 | 0 | if (xx > SCHAR_MAX) { |
942 | 0 | #ifdef ERANGE_FILL |
943 | 0 | *ip = NC_FILL_BYTE; |
944 | 0 | return NC_ERANGE; |
945 | | #else |
946 | | err = NC_ERANGE; |
947 | | #endif |
948 | 0 | } |
949 | 0 | #endif |
950 | | |
951 | | |
952 | 0 | *ip = (schar) xx; |
953 | 0 | return err; |
954 | 0 | } |
955 | | |
956 | | static int |
957 | | ncx_get_ushort_short(const void *xp, short *ip) |
958 | 0 | { |
959 | 0 | int err=NC_NOERR; |
960 | 0 | ix_ushort xx = 0; |
961 | 0 | get_ix_ushort(xp, &xx); |
962 | |
|
963 | 0 | #if IX_USHORT_MAX > SHORT_MAX |
964 | 0 | if (xx > SHORT_MAX) { |
965 | 0 | #ifdef ERANGE_FILL |
966 | 0 | *ip = NC_FILL_SHORT; |
967 | 0 | return NC_ERANGE; |
968 | | #else |
969 | | err = NC_ERANGE; |
970 | | #endif |
971 | 0 | } |
972 | 0 | #endif |
973 | | |
974 | | |
975 | 0 | *ip = (short) xx; |
976 | 0 | return err; |
977 | 0 | } |
978 | | |
979 | | static int |
980 | | ncx_get_ushort_int(const void *xp, int *ip) |
981 | 0 | { |
982 | 0 | int err=NC_NOERR; |
983 | 0 | ix_ushort xx = 0; |
984 | 0 | get_ix_ushort(xp, &xx); |
985 | |
|
986 | | #if IX_USHORT_MAX > INT_MAX |
987 | | if (xx > INT_MAX) { |
988 | | #ifdef ERANGE_FILL |
989 | | *ip = NC_FILL_INT; |
990 | | return NC_ERANGE; |
991 | | #else |
992 | | err = NC_ERANGE; |
993 | | #endif |
994 | | } |
995 | | #endif |
996 | | |
997 | |
|
998 | 0 | *ip = (int) xx; |
999 | 0 | return err; |
1000 | 0 | } |
1001 | | |
1002 | | static int |
1003 | | ncx_get_ushort_long(const void *xp, long *ip) |
1004 | 0 | { |
1005 | 0 | int err=NC_NOERR; |
1006 | 0 | ix_ushort xx = 0; |
1007 | 0 | get_ix_ushort(xp, &xx); |
1008 | |
|
1009 | | #if IX_USHORT_MAX > LONG_MAX |
1010 | | if (xx > LONG_MAX) { |
1011 | | #ifdef ERANGE_FILL |
1012 | | *ip = NC_FILL_INT; |
1013 | | return NC_ERANGE; |
1014 | | #else |
1015 | | err = NC_ERANGE; |
1016 | | #endif |
1017 | | } |
1018 | | #endif |
1019 | | |
1020 | |
|
1021 | 0 | *ip = (long) xx; |
1022 | 0 | return err; |
1023 | 0 | } |
1024 | | |
1025 | | static int |
1026 | | ncx_get_ushort_longlong(const void *xp, longlong *ip) |
1027 | 0 | { |
1028 | 0 | int err=NC_NOERR; |
1029 | 0 | ix_ushort xx = 0; |
1030 | 0 | get_ix_ushort(xp, &xx); |
1031 | |
|
1032 | | #if IX_USHORT_MAX > LONGLONG_MAX |
1033 | | if (xx > LONGLONG_MAX) { |
1034 | | #ifdef ERANGE_FILL |
1035 | | *ip = NC_FILL_INT64; |
1036 | | return NC_ERANGE; |
1037 | | #else |
1038 | | err = NC_ERANGE; |
1039 | | #endif |
1040 | | } |
1041 | | #endif |
1042 | | |
1043 | |
|
1044 | 0 | *ip = (longlong) xx; |
1045 | 0 | return err; |
1046 | 0 | } |
1047 | | |
1048 | | static int |
1049 | | ncx_get_ushort_ushort(const void *xp, ushort *ip) |
1050 | 0 | { |
1051 | 0 | int err=NC_NOERR; |
1052 | 0 | #if SIZEOF_IX_USHORT == SIZEOF_USHORT && IX_USHORT_MAX == USHORT_MAX |
1053 | 0 | get_ix_ushort(xp, (ix_ushort *)ip); |
1054 | | #else |
1055 | | ix_ushort xx = 0; |
1056 | | get_ix_ushort(xp, &xx); |
1057 | | |
1058 | | #if IX_USHORT_MAX > USHORT_MAX |
1059 | | if (xx > USHORT_MAX) { |
1060 | | #ifdef ERANGE_FILL |
1061 | | *ip = NC_FILL_USHORT; |
1062 | | return NC_ERANGE; |
1063 | | #else |
1064 | | err = NC_ERANGE; |
1065 | | #endif |
1066 | | } |
1067 | | #endif |
1068 | | |
1069 | | |
1070 | | *ip = (ushort) xx; |
1071 | | #endif |
1072 | 0 | return err; |
1073 | 0 | } |
1074 | | |
1075 | | static int |
1076 | | ncx_get_ushort_uchar(const void *xp, uchar *ip) |
1077 | 0 | { |
1078 | 0 | int err=NC_NOERR; |
1079 | | #if SIZEOF_IX_USHORT == SIZEOF_UCHAR && IX_USHORT_MAX == UCHAR_MAX |
1080 | | get_ix_ushort(xp, (ix_ushort *)ip); |
1081 | | #else |
1082 | 0 | ix_ushort xx = 0; |
1083 | 0 | get_ix_ushort(xp, &xx); |
1084 | |
|
1085 | 0 | #if IX_USHORT_MAX > UCHAR_MAX |
1086 | 0 | if (xx > UCHAR_MAX) { |
1087 | 0 | #ifdef ERANGE_FILL |
1088 | 0 | *ip = NC_FILL_UBYTE; |
1089 | 0 | return NC_ERANGE; |
1090 | | #else |
1091 | | err = NC_ERANGE; |
1092 | | #endif |
1093 | 0 | } |
1094 | 0 | #endif |
1095 | | |
1096 | | |
1097 | 0 | *ip = (uchar) xx; |
1098 | 0 | #endif |
1099 | 0 | return err; |
1100 | 0 | } |
1101 | | |
1102 | | static int |
1103 | | ncx_get_ushort_uint(const void *xp, uint *ip) |
1104 | 0 | { |
1105 | 0 | int err=NC_NOERR; |
1106 | | #if SIZEOF_IX_USHORT == SIZEOF_UINT && IX_USHORT_MAX == UINT_MAX |
1107 | | get_ix_ushort(xp, (ix_ushort *)ip); |
1108 | | #else |
1109 | 0 | ix_ushort xx = 0; |
1110 | 0 | get_ix_ushort(xp, &xx); |
1111 | |
|
1112 | | #if IX_USHORT_MAX > UINT_MAX |
1113 | | if (xx > UINT_MAX) { |
1114 | | #ifdef ERANGE_FILL |
1115 | | *ip = NC_FILL_UINT; |
1116 | | return NC_ERANGE; |
1117 | | #else |
1118 | | err = NC_ERANGE; |
1119 | | #endif |
1120 | | } |
1121 | | #endif |
1122 | | |
1123 | |
|
1124 | 0 | *ip = (uint) xx; |
1125 | 0 | #endif |
1126 | 0 | return err; |
1127 | 0 | } |
1128 | | |
1129 | | static int |
1130 | | ncx_get_ushort_ulonglong(const void *xp, ulonglong *ip) |
1131 | 0 | { |
1132 | 0 | int err=NC_NOERR; |
1133 | | #if SIZEOF_IX_USHORT == SIZEOF_ULONGLONG && IX_USHORT_MAX == ULONGLONG_MAX |
1134 | | get_ix_ushort(xp, (ix_ushort *)ip); |
1135 | | #else |
1136 | 0 | ix_ushort xx = 0; |
1137 | 0 | get_ix_ushort(xp, &xx); |
1138 | |
|
1139 | | #if IX_USHORT_MAX > ULONGLONG_MAX |
1140 | | if (xx > ULONGLONG_MAX) { |
1141 | | #ifdef ERANGE_FILL |
1142 | | *ip = NC_FILL_UINT64; |
1143 | | return NC_ERANGE; |
1144 | | #else |
1145 | | err = NC_ERANGE; |
1146 | | #endif |
1147 | | } |
1148 | | #endif |
1149 | | |
1150 | |
|
1151 | 0 | *ip = (ulonglong) xx; |
1152 | 0 | #endif |
1153 | 0 | return err; |
1154 | 0 | } |
1155 | | |
1156 | | static int |
1157 | | ncx_get_ushort_float(const void *xp, float *ip) |
1158 | 0 | { |
1159 | 0 | ix_ushort xx = 0; |
1160 | 0 | get_ix_ushort(xp, &xx); |
1161 | 0 | *ip = (float)xx; |
1162 | 0 | return NC_NOERR; |
1163 | 0 | } |
1164 | | |
1165 | | static int |
1166 | | ncx_get_ushort_double(const void *xp, double *ip) |
1167 | 0 | { |
1168 | 0 | ix_ushort xx = 0; |
1169 | 0 | get_ix_ushort(xp, &xx); |
1170 | 0 | *ip = (double)xx; |
1171 | 0 | return NC_NOERR; |
1172 | 0 | } |
1173 | | |
1174 | | |
1175 | | static int |
1176 | | ncx_put_ushort_schar(void *xp, const schar *ip, void *fillp) |
1177 | 0 | { |
1178 | 0 | int err=NC_NOERR; |
1179 | 0 | uchar *cp; |
1180 | 0 | if (*ip < 0) { |
1181 | 0 | #ifdef ERANGE_FILL |
1182 | 0 | if (fillp != NULL) memcpy(xp, fillp, 2); |
1183 | 0 | #ifndef WORDS_BIGENDIAN |
1184 | 0 | swapn2b(xp, xp, 1); |
1185 | 0 | #endif |
1186 | 0 | return NC_ERANGE; |
1187 | | #else |
1188 | | err = NC_ERANGE; |
1189 | | #endif |
1190 | 0 | } |
1191 | | |
1192 | 0 | cp = (uchar *) xp; |
1193 | 0 | if (*ip & 0x80) |
1194 | 0 | *cp++ = 0xff; |
1195 | 0 | else |
1196 | 0 | *cp++ = 0; |
1197 | 0 | *cp = (uchar)*ip; |
1198 | |
|
1199 | 0 | return err; |
1200 | 0 | } |
1201 | | |
1202 | | static int |
1203 | | ncx_put_ushort_uchar(void *xp, const uchar *ip, void *fillp) |
1204 | 0 | { |
1205 | 0 | uchar *cp = (uchar *) xp; |
1206 | 0 | *cp++ = 0; |
1207 | 0 | *cp = *ip; |
1208 | 0 | return NC_NOERR; |
1209 | 0 | } |
1210 | | |
1211 | | static int |
1212 | | ncx_put_ushort_short(void *xp, const short *ip, void *fillp) |
1213 | 0 | { |
1214 | 0 | int err=NC_NOERR; |
1215 | 0 | ix_ushort xx = NC_FILL_USHORT; |
1216 | |
|
1217 | | #if IX_USHORT_MAX < SHORT_MAX |
1218 | | if (*ip > IX_USHORT_MAX) { |
1219 | | |
1220 | | #ifdef ERANGE_FILL |
1221 | | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1222 | | #endif |
1223 | | err = NC_ERANGE; |
1224 | | } |
1225 | | #ifdef ERANGE_FILL |
1226 | | else |
1227 | | #endif |
1228 | | #endif |
1229 | 0 | if (*ip < 0) { |
1230 | | |
1231 | 0 | #ifdef ERANGE_FILL |
1232 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1233 | 0 | #endif |
1234 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
1235 | 0 | } |
1236 | 0 | #ifdef ERANGE_FILL |
1237 | 0 | else |
1238 | 0 | #endif |
1239 | 0 | xx = (ix_ushort)*ip; |
1240 | |
|
1241 | 0 | put_ix_ushort(xp, &xx); |
1242 | 0 | return err; |
1243 | 0 | } |
1244 | | |
1245 | | static int |
1246 | | ncx_put_ushort_int(void *xp, const int *ip, void *fillp) |
1247 | 0 | { |
1248 | 0 | int err=NC_NOERR; |
1249 | 0 | ix_ushort xx = NC_FILL_USHORT; |
1250 | |
|
1251 | 0 | #if IX_USHORT_MAX < INT_MAX |
1252 | 0 | if (*ip > IX_USHORT_MAX) { |
1253 | | |
1254 | 0 | #ifdef ERANGE_FILL |
1255 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1256 | 0 | #endif |
1257 | 0 | err = NC_ERANGE; |
1258 | 0 | } |
1259 | 0 | #ifdef ERANGE_FILL |
1260 | 0 | else |
1261 | 0 | #endif |
1262 | 0 | #endif |
1263 | 0 | if (*ip < 0) { |
1264 | | |
1265 | 0 | #ifdef ERANGE_FILL |
1266 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1267 | 0 | #endif |
1268 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
1269 | 0 | } |
1270 | 0 | #ifdef ERANGE_FILL |
1271 | 0 | else |
1272 | 0 | #endif |
1273 | 0 | xx = (ix_ushort)*ip; |
1274 | |
|
1275 | 0 | put_ix_ushort(xp, &xx); |
1276 | 0 | return err; |
1277 | 0 | } |
1278 | | |
1279 | | static int |
1280 | | ncx_put_ushort_long(void *xp, const long *ip, void *fillp) |
1281 | 0 | { |
1282 | 0 | int err=NC_NOERR; |
1283 | 0 | ix_ushort xx = NC_FILL_USHORT; |
1284 | |
|
1285 | 0 | #if IX_USHORT_MAX < LONG_MAX |
1286 | 0 | if (*ip > IX_USHORT_MAX) { |
1287 | | |
1288 | 0 | #ifdef ERANGE_FILL |
1289 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1290 | 0 | #endif |
1291 | 0 | err = NC_ERANGE; |
1292 | 0 | } |
1293 | 0 | #ifdef ERANGE_FILL |
1294 | 0 | else |
1295 | 0 | #endif |
1296 | 0 | #endif |
1297 | 0 | if (*ip < 0) { |
1298 | | |
1299 | 0 | #ifdef ERANGE_FILL |
1300 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1301 | 0 | #endif |
1302 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
1303 | 0 | } |
1304 | 0 | #ifdef ERANGE_FILL |
1305 | 0 | else |
1306 | 0 | #endif |
1307 | 0 | xx = (ix_ushort)*ip; |
1308 | |
|
1309 | 0 | put_ix_ushort(xp, &xx); |
1310 | 0 | return err; |
1311 | 0 | } |
1312 | | |
1313 | | static int |
1314 | | ncx_put_ushort_longlong(void *xp, const longlong *ip, void *fillp) |
1315 | 0 | { |
1316 | 0 | int err=NC_NOERR; |
1317 | 0 | ix_ushort xx = NC_FILL_USHORT; |
1318 | |
|
1319 | 0 | #if IX_USHORT_MAX < LONGLONG_MAX |
1320 | 0 | if (*ip > IX_USHORT_MAX) { |
1321 | | |
1322 | 0 | #ifdef ERANGE_FILL |
1323 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1324 | 0 | #endif |
1325 | 0 | err = NC_ERANGE; |
1326 | 0 | } |
1327 | 0 | #ifdef ERANGE_FILL |
1328 | 0 | else |
1329 | 0 | #endif |
1330 | 0 | #endif |
1331 | 0 | if (*ip < 0) { |
1332 | | |
1333 | 0 | #ifdef ERANGE_FILL |
1334 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1335 | 0 | #endif |
1336 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
1337 | 0 | } |
1338 | 0 | #ifdef ERANGE_FILL |
1339 | 0 | else |
1340 | 0 | #endif |
1341 | 0 | xx = (ix_ushort)*ip; |
1342 | |
|
1343 | 0 | put_ix_ushort(xp, &xx); |
1344 | 0 | return err; |
1345 | 0 | } |
1346 | | |
1347 | | static int |
1348 | | ncx_put_ushort_ushort(void *xp, const ushort *ip, void *fillp) |
1349 | 0 | { |
1350 | 0 | int err=NC_NOERR; |
1351 | 0 | #if SIZEOF_IX_USHORT == SIZEOF_USHORT && IX_USHORT_MAX == USHORT_MAX |
1352 | 0 | put_ix_ushort(xp, (const ix_ushort *)ip); |
1353 | | #else |
1354 | | ix_ushort xx = NC_FILL_USHORT; |
1355 | | |
1356 | | #if IX_USHORT_MAX < USHORT_MAX |
1357 | | if (*ip > IX_USHORT_MAX) { |
1358 | | |
1359 | | #ifdef ERANGE_FILL |
1360 | | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1361 | | #endif |
1362 | | err = NC_ERANGE; |
1363 | | } |
1364 | | #ifdef ERANGE_FILL |
1365 | | else |
1366 | | #endif |
1367 | | #endif |
1368 | | xx = (ix_ushort)*ip; |
1369 | | |
1370 | | put_ix_ushort(xp, &xx); |
1371 | | #endif |
1372 | 0 | return err; |
1373 | 0 | } |
1374 | | |
1375 | | static int |
1376 | | ncx_put_ushort_uint(void *xp, const uint *ip, void *fillp) |
1377 | 0 | { |
1378 | 0 | int err=NC_NOERR; |
1379 | | #if SIZEOF_IX_USHORT == SIZEOF_UINT && IX_USHORT_MAX == UINT_MAX |
1380 | | put_ix_ushort(xp, (const ix_ushort *)ip); |
1381 | | #else |
1382 | 0 | ix_ushort xx = NC_FILL_USHORT; |
1383 | |
|
1384 | 0 | #if IX_USHORT_MAX < UINT_MAX |
1385 | 0 | if (*ip > IX_USHORT_MAX) { |
1386 | | |
1387 | 0 | #ifdef ERANGE_FILL |
1388 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1389 | 0 | #endif |
1390 | 0 | err = NC_ERANGE; |
1391 | 0 | } |
1392 | 0 | #ifdef ERANGE_FILL |
1393 | 0 | else |
1394 | 0 | #endif |
1395 | 0 | #endif |
1396 | 0 | xx = (ix_ushort)*ip; |
1397 | |
|
1398 | 0 | put_ix_ushort(xp, &xx); |
1399 | 0 | #endif |
1400 | 0 | return err; |
1401 | 0 | } |
1402 | | |
1403 | | static int |
1404 | | ncx_put_ushort_ulonglong(void *xp, const ulonglong *ip, void *fillp) |
1405 | 0 | { |
1406 | 0 | int err=NC_NOERR; |
1407 | | #if SIZEOF_IX_USHORT == SIZEOF_ULONGLONG && IX_USHORT_MAX == ULONGLONG_MAX |
1408 | | put_ix_ushort(xp, (const ix_ushort *)ip); |
1409 | | #else |
1410 | 0 | ix_ushort xx = NC_FILL_USHORT; |
1411 | |
|
1412 | 0 | #if IX_USHORT_MAX < ULONGLONG_MAX |
1413 | 0 | if (*ip > IX_USHORT_MAX) { |
1414 | | |
1415 | 0 | #ifdef ERANGE_FILL |
1416 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1417 | 0 | #endif |
1418 | 0 | err = NC_ERANGE; |
1419 | 0 | } |
1420 | 0 | #ifdef ERANGE_FILL |
1421 | 0 | else |
1422 | 0 | #endif |
1423 | 0 | #endif |
1424 | 0 | xx = (ix_ushort)*ip; |
1425 | |
|
1426 | 0 | put_ix_ushort(xp, &xx); |
1427 | 0 | #endif |
1428 | 0 | return err; |
1429 | 0 | } |
1430 | | |
1431 | | static int |
1432 | | ncx_put_ushort_float(void *xp, const float *ip, void *fillp) |
1433 | 0 | { |
1434 | 0 | int err=NC_NOERR; |
1435 | 0 | ix_ushort xx = NC_FILL_USHORT; |
1436 | |
|
1437 | 0 | if (*ip > (double)X_USHORT_MAX || *ip < 0) { |
1438 | | |
1439 | 0 | #ifdef ERANGE_FILL |
1440 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1441 | 0 | #endif |
1442 | 0 | err = NC_ERANGE; |
1443 | 0 | } |
1444 | 0 | #ifdef ERANGE_FILL |
1445 | 0 | else |
1446 | 0 | #endif |
1447 | 0 | xx = (ix_ushort)*ip; |
1448 | |
|
1449 | 0 | put_ix_ushort(xp, &xx); |
1450 | 0 | return err; |
1451 | 0 | } |
1452 | | |
1453 | | static int |
1454 | | ncx_put_ushort_double(void *xp, const double *ip, void *fillp) |
1455 | 0 | { |
1456 | 0 | int err=NC_NOERR; |
1457 | 0 | ix_ushort xx = NC_FILL_USHORT; |
1458 | |
|
1459 | 0 | if (*ip > X_USHORT_MAX || *ip < 0) { |
1460 | | |
1461 | 0 | #ifdef ERANGE_FILL |
1462 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 2); |
1463 | 0 | #endif |
1464 | 0 | err = NC_ERANGE; |
1465 | 0 | } |
1466 | 0 | #ifdef ERANGE_FILL |
1467 | 0 | else |
1468 | 0 | #endif |
1469 | 0 | xx = (ix_ushort)*ip; |
1470 | |
|
1471 | 0 | put_ix_ushort(xp, &xx); |
1472 | 0 | return err; |
1473 | 0 | } |
1474 | | |
1475 | | |
1476 | | /* external NC_INT ----------------------------------------------------------*/ |
1477 | | |
1478 | | #if SHORT_MAX == X_INT_MAX |
1479 | | typedef short ix_int; |
1480 | | #define SIZEOF_IX_INT SIZEOF_SHORT |
1481 | | #define IX_INT_MAX SHORT_MAX |
1482 | | #elif INT_MAX >= X_INT_MAX |
1483 | | typedef int ix_int; |
1484 | | #define SIZEOF_IX_INT SIZEOF_INT |
1485 | 0 | #define IX_INT_MAX INT_MAX |
1486 | | #elif LONG_MAX >= X_INT_MAX |
1487 | | typedef long ix_int; |
1488 | | #define SIZEOF_IX_INT SIZEOF_LONG |
1489 | | #define IX_INT_MAX LONG_MAX |
1490 | | #else |
1491 | | #error "ix_int implementation" |
1492 | | #endif |
1493 | | |
1494 | | |
1495 | | static void |
1496 | | get_ix_int(const void *xp, ix_int *ip) |
1497 | 0 | { |
1498 | 0 | const uchar *cp = (const uchar *) xp; |
1499 | |
|
1500 | 0 | #if INT_MAX >= X_INT_MAX |
1501 | 0 | *ip = (ix_int)((unsigned)(*cp++) << 24); |
1502 | | #else |
1503 | | *ip = *cp++ << 24; |
1504 | | #endif |
1505 | | #if SIZEOF_IX_INT > X_SIZEOF_INT |
1506 | | if (*ip & 0x80000000) |
1507 | | { |
1508 | | /* extern is negative */ |
1509 | | *ip |= (~(0xffffffff)); /* N.B. Assumes "twos complement" */ |
1510 | | } |
1511 | | #endif |
1512 | 0 | *ip |= (*cp++ << 16); |
1513 | 0 | *ip |= (*cp++ << 8); |
1514 | 0 | *ip |= *cp; |
1515 | 0 | } |
1516 | | |
1517 | | static void |
1518 | | put_ix_int(void *xp, const ix_int *ip) |
1519 | 0 | { |
1520 | 0 | uchar *cp = (uchar *) xp; |
1521 | |
|
1522 | 0 | *cp++ = (uchar)( (*ip) >> 24); |
1523 | 0 | *cp++ = (uchar)(((*ip) & 0x00ff0000) >> 16); |
1524 | 0 | *cp++ = (uchar)(((*ip) & 0x0000ff00) >> 8); |
1525 | 0 | *cp = (uchar)( (*ip) & 0x000000ff); |
1526 | 0 | } |
1527 | | |
1528 | | #if X_SIZEOF_INT != SIZEOF_INT |
1529 | | static int |
1530 | | ncx_get_int_int(const void *xp, int *ip) |
1531 | | { |
1532 | | int err=NC_NOERR; |
1533 | | #if SIZEOF_IX_INT == SIZEOF_INT && IX_INT_MAX == INT_MAX |
1534 | | get_ix_int(xp, (ix_int *)ip); |
1535 | | #else |
1536 | | ix_int xx = 0; |
1537 | | get_ix_int(xp, &xx); |
1538 | | |
1539 | | #if IX_INT_MAX > INT_MAX |
1540 | | if (xx > INT_MAX || xx < INT_MIN) { |
1541 | | #ifdef ERANGE_FILL |
1542 | | *ip = NC_FILL_INT; |
1543 | | return NC_ERANGE; |
1544 | | #else |
1545 | | err = NC_ERANGE; |
1546 | | #endif |
1547 | | } |
1548 | | #endif |
1549 | | |
1550 | | |
1551 | | *ip = (int) xx; |
1552 | | #endif |
1553 | | return err; |
1554 | | } |
1555 | | |
1556 | | #endif |
1557 | | static int |
1558 | | ncx_get_int_schar(const void *xp, schar *ip) |
1559 | 0 | { |
1560 | 0 | int err=NC_NOERR; |
1561 | 0 | ix_int xx = 0; |
1562 | 0 | get_ix_int(xp, &xx); |
1563 | |
|
1564 | 0 | #if IX_INT_MAX > SCHAR_MAX |
1565 | 0 | if (xx > SCHAR_MAX || xx < SCHAR_MIN) { |
1566 | 0 | #ifdef ERANGE_FILL |
1567 | 0 | *ip = NC_FILL_BYTE; |
1568 | 0 | return NC_ERANGE; |
1569 | | #else |
1570 | | err = NC_ERANGE; |
1571 | | #endif |
1572 | 0 | } |
1573 | 0 | #endif |
1574 | | |
1575 | | |
1576 | 0 | *ip = (schar) xx; |
1577 | 0 | return err; |
1578 | 0 | } |
1579 | | |
1580 | | static int |
1581 | | ncx_get_int_short(const void *xp, short *ip) |
1582 | 0 | { |
1583 | 0 | int err=NC_NOERR; |
1584 | | #if SIZEOF_IX_INT == SIZEOF_SHORT && IX_INT_MAX == SHORT_MAX |
1585 | | get_ix_int(xp, (ix_int *)ip); |
1586 | | #else |
1587 | 0 | ix_int xx = 0; |
1588 | 0 | get_ix_int(xp, &xx); |
1589 | |
|
1590 | 0 | #if IX_INT_MAX > SHORT_MAX |
1591 | 0 | if (xx > SHORT_MAX || xx < SHORT_MIN) { |
1592 | 0 | #ifdef ERANGE_FILL |
1593 | 0 | *ip = NC_FILL_SHORT; |
1594 | 0 | return NC_ERANGE; |
1595 | | #else |
1596 | | err = NC_ERANGE; |
1597 | | #endif |
1598 | 0 | } |
1599 | 0 | #endif |
1600 | | |
1601 | | |
1602 | 0 | *ip = (short) xx; |
1603 | 0 | #endif |
1604 | 0 | return err; |
1605 | 0 | } |
1606 | | |
1607 | | static int |
1608 | | ncx_get_int_long(const void *xp, long *ip) |
1609 | 0 | { |
1610 | 0 | int err=NC_NOERR; |
1611 | | #if SIZEOF_IX_INT == SIZEOF_LONG && IX_INT_MAX == LONG_MAX |
1612 | | get_ix_int(xp, (ix_int *)ip); |
1613 | | #else |
1614 | 0 | ix_int xx = 0; |
1615 | 0 | get_ix_int(xp, &xx); |
1616 | |
|
1617 | | #if IX_INT_MAX > LONG_MAX |
1618 | | if (xx > LONG_MAX || xx < LONG_MIN) { |
1619 | | #ifdef ERANGE_FILL |
1620 | | *ip = NC_FILL_INT; |
1621 | | return NC_ERANGE; |
1622 | | #else |
1623 | | err = NC_ERANGE; |
1624 | | #endif |
1625 | | } |
1626 | | #endif |
1627 | | |
1628 | |
|
1629 | 0 | *ip = (long) xx; |
1630 | 0 | #endif |
1631 | 0 | return err; |
1632 | 0 | } |
1633 | | |
1634 | | static int |
1635 | | ncx_get_int_longlong(const void *xp, longlong *ip) |
1636 | 0 | { |
1637 | 0 | int err=NC_NOERR; |
1638 | | #if SIZEOF_IX_INT == SIZEOF_LONGLONG && IX_INT_MAX == LONGLONG_MAX |
1639 | | get_ix_int(xp, (ix_int *)ip); |
1640 | | #else |
1641 | 0 | ix_int xx = 0; |
1642 | 0 | get_ix_int(xp, &xx); |
1643 | |
|
1644 | | #if IX_INT_MAX > LONGLONG_MAX |
1645 | | if (xx > LONGLONG_MAX || xx < LONGLONG_MIN) { |
1646 | | #ifdef ERANGE_FILL |
1647 | | *ip = NC_FILL_INT64; |
1648 | | return NC_ERANGE; |
1649 | | #else |
1650 | | err = NC_ERANGE; |
1651 | | #endif |
1652 | | } |
1653 | | #endif |
1654 | | |
1655 | |
|
1656 | 0 | *ip = (longlong) xx; |
1657 | 0 | #endif |
1658 | 0 | return err; |
1659 | 0 | } |
1660 | | |
1661 | | static int |
1662 | | ncx_get_int_ushort(const void *xp, ushort *ip) |
1663 | 0 | { |
1664 | 0 | int err=NC_NOERR; |
1665 | 0 | ix_int xx = 0; |
1666 | 0 | get_ix_int(xp, &xx); |
1667 | |
|
1668 | 0 | #if IX_INT_MAX > USHORT_MAX |
1669 | 0 | if (xx > USHORT_MAX) { |
1670 | 0 | #ifdef ERANGE_FILL |
1671 | 0 | *ip = NC_FILL_USHORT; |
1672 | 0 | return NC_ERANGE; |
1673 | | #else |
1674 | | err = NC_ERANGE; |
1675 | | #endif |
1676 | 0 | } |
1677 | 0 | #endif |
1678 | | |
1679 | 0 | if (xx < 0) { |
1680 | 0 | #ifdef ERANGE_FILL |
1681 | 0 | *ip = NC_FILL_USHORT; |
1682 | 0 | return NC_ERANGE; |
1683 | | #else |
1684 | | err = NC_ERANGE; /* because ip is unsigned */ |
1685 | | #endif |
1686 | 0 | } |
1687 | 0 | *ip = (ushort) xx; |
1688 | 0 | return err; |
1689 | 0 | } |
1690 | | |
1691 | | static int |
1692 | | ncx_get_int_uchar(const void *xp, uchar *ip) |
1693 | 0 | { |
1694 | 0 | int err=NC_NOERR; |
1695 | 0 | ix_int xx = 0; |
1696 | 0 | get_ix_int(xp, &xx); |
1697 | |
|
1698 | 0 | #if IX_INT_MAX > UCHAR_MAX |
1699 | 0 | if (xx > UCHAR_MAX) { |
1700 | 0 | #ifdef ERANGE_FILL |
1701 | 0 | *ip = NC_FILL_UBYTE; |
1702 | 0 | return NC_ERANGE; |
1703 | | #else |
1704 | | err = NC_ERANGE; |
1705 | | #endif |
1706 | 0 | } |
1707 | 0 | #endif |
1708 | | |
1709 | 0 | if (xx < 0) { |
1710 | 0 | #ifdef ERANGE_FILL |
1711 | 0 | *ip = NC_FILL_UBYTE; |
1712 | 0 | return NC_ERANGE; |
1713 | | #else |
1714 | | err = NC_ERANGE; /* because ip is unsigned */ |
1715 | | #endif |
1716 | 0 | } |
1717 | 0 | *ip = (uchar) xx; |
1718 | 0 | return err; |
1719 | 0 | } |
1720 | | |
1721 | | static int |
1722 | | ncx_get_int_uint(const void *xp, uint *ip) |
1723 | 0 | { |
1724 | 0 | int err=NC_NOERR; |
1725 | 0 | ix_int xx = 0; |
1726 | 0 | get_ix_int(xp, &xx); |
1727 | |
|
1728 | | #if IX_INT_MAX > UINT_MAX |
1729 | | if (xx > UINT_MAX) { |
1730 | | #ifdef ERANGE_FILL |
1731 | | *ip = NC_FILL_UINT; |
1732 | | return NC_ERANGE; |
1733 | | #else |
1734 | | err = NC_ERANGE; |
1735 | | #endif |
1736 | | } |
1737 | | #endif |
1738 | |
|
1739 | 0 | if (xx < 0) { |
1740 | 0 | #ifdef ERANGE_FILL |
1741 | 0 | *ip = NC_FILL_UINT; |
1742 | 0 | return NC_ERANGE; |
1743 | | #else |
1744 | | err = NC_ERANGE; /* because ip is unsigned */ |
1745 | | #endif |
1746 | 0 | } |
1747 | 0 | *ip = (uint) xx; |
1748 | 0 | return err; |
1749 | 0 | } |
1750 | | |
1751 | | static int |
1752 | | ncx_get_int_ulonglong(const void *xp, ulonglong *ip) |
1753 | 0 | { |
1754 | 0 | int err=NC_NOERR; |
1755 | 0 | ix_int xx = 0; |
1756 | 0 | get_ix_int(xp, &xx); |
1757 | |
|
1758 | | #if IX_INT_MAX > ULONGLONG_MAX |
1759 | | if (xx > ULONGLONG_MAX) { |
1760 | | #ifdef ERANGE_FILL |
1761 | | *ip = NC_FILL_UINT64; |
1762 | | return NC_ERANGE; |
1763 | | #else |
1764 | | err = NC_ERANGE; |
1765 | | #endif |
1766 | | } |
1767 | | #endif |
1768 | |
|
1769 | 0 | if (xx < 0) { |
1770 | 0 | #ifdef ERANGE_FILL |
1771 | 0 | *ip = NC_FILL_UINT64; |
1772 | 0 | return NC_ERANGE; |
1773 | | #else |
1774 | | err = NC_ERANGE; /* because ip is unsigned */ |
1775 | | #endif |
1776 | 0 | } |
1777 | 0 | *ip = (ulonglong) xx; |
1778 | 0 | return err; |
1779 | 0 | } |
1780 | | |
1781 | | static int |
1782 | | ncx_get_int_float(const void *xp, float *ip) |
1783 | 0 | { |
1784 | 0 | ix_int xx = 0; |
1785 | 0 | get_ix_int(xp, &xx); |
1786 | 0 | *ip = (float)xx; |
1787 | 0 | return NC_NOERR; |
1788 | 0 | } |
1789 | | |
1790 | | static int |
1791 | | ncx_get_int_double(const void *xp, double *ip) |
1792 | 0 | { |
1793 | 0 | ix_int xx = 0; |
1794 | 0 | get_ix_int(xp, &xx); |
1795 | 0 | *ip = (double)xx; |
1796 | 0 | return NC_NOERR; |
1797 | 0 | } |
1798 | | |
1799 | | |
1800 | | static int |
1801 | | ncx_put_int_schar(void *xp, const schar *ip, void *fillp) |
1802 | 0 | { |
1803 | 0 | uchar *cp = (uchar *) xp; |
1804 | 0 | if (*ip & 0x80) |
1805 | 0 | { |
1806 | 0 | *cp++ = 0xff; |
1807 | 0 | *cp++ = 0xff; |
1808 | 0 | *cp++ = 0xff; |
1809 | 0 | } |
1810 | 0 | else |
1811 | 0 | { |
1812 | 0 | *cp++ = 0x00; |
1813 | 0 | *cp++ = 0x00; |
1814 | 0 | *cp++ = 0x00; |
1815 | 0 | } |
1816 | 0 | *cp = (uchar)*ip; |
1817 | 0 | return NC_NOERR; |
1818 | 0 | } |
1819 | | |
1820 | | static int |
1821 | | ncx_put_int_uchar(void *xp, const uchar *ip, void *fillp) |
1822 | 0 | { |
1823 | 0 | uchar *cp = (uchar *) xp; |
1824 | 0 | *cp++ = 0x00; |
1825 | 0 | *cp++ = 0x00; |
1826 | 0 | *cp++ = 0x00; |
1827 | 0 | *cp = *ip; |
1828 | 0 | return NC_NOERR; |
1829 | 0 | } |
1830 | | |
1831 | | #if X_SIZEOF_INT != SIZEOF_INT |
1832 | | static int |
1833 | | ncx_put_int_int(void *xp, const int *ip, void *fillp) |
1834 | | { |
1835 | | int err=NC_NOERR; |
1836 | | #if SIZEOF_IX_INT == SIZEOF_INT && IX_INT_MAX == INT_MAX |
1837 | | put_ix_int(xp, (const ix_int *)ip); |
1838 | | #else |
1839 | | ix_int xx = NC_FILL_INT; |
1840 | | |
1841 | | #if IX_INT_MAX < INT_MAX |
1842 | | if (*ip > IX_INT_MAX || *ip < X_INT_MIN) { |
1843 | | |
1844 | | #ifdef ERANGE_FILL |
1845 | | if (fillp != NULL) memcpy(&xx, fillp, 4); |
1846 | | #endif |
1847 | | err = NC_ERANGE; |
1848 | | } |
1849 | | #ifdef ERANGE_FILL |
1850 | | else |
1851 | | #endif |
1852 | | #endif |
1853 | | xx = (ix_int)*ip; |
1854 | | |
1855 | | put_ix_int(xp, &xx); |
1856 | | #endif |
1857 | | return err; |
1858 | | } |
1859 | | |
1860 | | #endif |
1861 | | static int |
1862 | | ncx_put_int_short(void *xp, const short *ip, void *fillp) |
1863 | 0 | { |
1864 | 0 | int err=NC_NOERR; |
1865 | | #if SIZEOF_IX_INT == SIZEOF_SHORT && IX_INT_MAX == SHORT_MAX |
1866 | | put_ix_int(xp, (const ix_int *)ip); |
1867 | | #else |
1868 | 0 | ix_int xx = NC_FILL_INT; |
1869 | |
|
1870 | | #if IX_INT_MAX < SHORT_MAX |
1871 | | if (*ip > IX_INT_MAX || *ip < X_INT_MIN) { |
1872 | | |
1873 | | #ifdef ERANGE_FILL |
1874 | | if (fillp != NULL) memcpy(&xx, fillp, 4); |
1875 | | #endif |
1876 | | err = NC_ERANGE; |
1877 | | } |
1878 | | #ifdef ERANGE_FILL |
1879 | | else |
1880 | | #endif |
1881 | | #endif |
1882 | 0 | xx = (ix_int)*ip; |
1883 | |
|
1884 | 0 | put_ix_int(xp, &xx); |
1885 | 0 | #endif |
1886 | 0 | return err; |
1887 | 0 | } |
1888 | | |
1889 | | static int |
1890 | | ncx_put_int_long(void *xp, const long *ip, void *fillp) |
1891 | 0 | { |
1892 | 0 | int err=NC_NOERR; |
1893 | | #if SIZEOF_IX_INT == SIZEOF_LONG && IX_INT_MAX == LONG_MAX |
1894 | | put_ix_int(xp, (const ix_int *)ip); |
1895 | | #else |
1896 | 0 | ix_int xx = NC_FILL_INT; |
1897 | |
|
1898 | 0 | #if IX_INT_MAX < LONG_MAX |
1899 | 0 | if (*ip > IX_INT_MAX || *ip < X_INT_MIN) { |
1900 | | |
1901 | 0 | #ifdef ERANGE_FILL |
1902 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
1903 | 0 | #endif |
1904 | 0 | err = NC_ERANGE; |
1905 | 0 | } |
1906 | 0 | #ifdef ERANGE_FILL |
1907 | 0 | else |
1908 | 0 | #endif |
1909 | 0 | #endif |
1910 | 0 | xx = (ix_int)*ip; |
1911 | |
|
1912 | 0 | put_ix_int(xp, &xx); |
1913 | 0 | #endif |
1914 | 0 | return err; |
1915 | 0 | } |
1916 | | |
1917 | | static int |
1918 | | ncx_put_int_longlong(void *xp, const longlong *ip, void *fillp) |
1919 | 0 | { |
1920 | 0 | int err=NC_NOERR; |
1921 | | #if SIZEOF_IX_INT == SIZEOF_LONGLONG && IX_INT_MAX == LONGLONG_MAX |
1922 | | put_ix_int(xp, (const ix_int *)ip); |
1923 | | #else |
1924 | 0 | ix_int xx = NC_FILL_INT; |
1925 | |
|
1926 | 0 | #if IX_INT_MAX < LONGLONG_MAX |
1927 | 0 | if (*ip > IX_INT_MAX || *ip < X_INT_MIN) { |
1928 | | |
1929 | 0 | #ifdef ERANGE_FILL |
1930 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
1931 | 0 | #endif |
1932 | 0 | err = NC_ERANGE; |
1933 | 0 | } |
1934 | 0 | #ifdef ERANGE_FILL |
1935 | 0 | else |
1936 | 0 | #endif |
1937 | 0 | #endif |
1938 | 0 | xx = (ix_int)*ip; |
1939 | |
|
1940 | 0 | put_ix_int(xp, &xx); |
1941 | 0 | #endif |
1942 | 0 | return err; |
1943 | 0 | } |
1944 | | |
1945 | | static int |
1946 | | ncx_put_int_ushort(void *xp, const ushort *ip, void *fillp) |
1947 | 0 | { |
1948 | 0 | int err=NC_NOERR; |
1949 | 0 | ix_int xx = NC_FILL_INT; |
1950 | |
|
1951 | | #if IX_INT_MAX < USHORT_MAX |
1952 | | if (*ip > IX_INT_MAX) { |
1953 | | |
1954 | | #ifdef ERANGE_FILL |
1955 | | if (fillp != NULL) memcpy(&xx, fillp, 4); |
1956 | | #endif |
1957 | | err = NC_ERANGE; |
1958 | | } |
1959 | | #ifdef ERANGE_FILL |
1960 | | else |
1961 | | #endif |
1962 | | #endif |
1963 | 0 | xx = (ix_int)*ip; |
1964 | |
|
1965 | 0 | put_ix_int(xp, &xx); |
1966 | 0 | return err; |
1967 | 0 | } |
1968 | | |
1969 | | static int |
1970 | | ncx_put_int_uint(void *xp, const uint *ip, void *fillp) |
1971 | 0 | { |
1972 | 0 | int err=NC_NOERR; |
1973 | 0 | ix_int xx = NC_FILL_INT; |
1974 | |
|
1975 | 0 | #if IX_INT_MAX < UINT_MAX |
1976 | 0 | if (*ip > IX_INT_MAX) { |
1977 | | |
1978 | 0 | #ifdef ERANGE_FILL |
1979 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
1980 | 0 | #endif |
1981 | 0 | err = NC_ERANGE; |
1982 | 0 | } |
1983 | 0 | #ifdef ERANGE_FILL |
1984 | 0 | else |
1985 | 0 | #endif |
1986 | 0 | #endif |
1987 | 0 | xx = (ix_int)*ip; |
1988 | |
|
1989 | 0 | put_ix_int(xp, &xx); |
1990 | 0 | return err; |
1991 | 0 | } |
1992 | | |
1993 | | static int |
1994 | | ncx_put_int_ulonglong(void *xp, const ulonglong *ip, void *fillp) |
1995 | 0 | { |
1996 | 0 | int err=NC_NOERR; |
1997 | 0 | ix_int xx = NC_FILL_INT; |
1998 | |
|
1999 | 0 | #if IX_INT_MAX < ULONGLONG_MAX |
2000 | 0 | if (*ip > IX_INT_MAX) { |
2001 | | |
2002 | 0 | #ifdef ERANGE_FILL |
2003 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2004 | 0 | #endif |
2005 | 0 | err = NC_ERANGE; |
2006 | 0 | } |
2007 | 0 | #ifdef ERANGE_FILL |
2008 | 0 | else |
2009 | 0 | #endif |
2010 | 0 | #endif |
2011 | 0 | xx = (ix_int)*ip; |
2012 | |
|
2013 | 0 | put_ix_int(xp, &xx); |
2014 | 0 | return err; |
2015 | 0 | } |
2016 | | |
2017 | | static int |
2018 | | ncx_put_int_float(void *xp, const float *ip, void *fillp) |
2019 | 0 | { |
2020 | 0 | int err=NC_NOERR; |
2021 | 0 | ix_int xx = NC_FILL_INT; |
2022 | |
|
2023 | 0 | if (*ip > (double)X_INT_MAX || *ip < (double)X_INT_MIN) { |
2024 | | |
2025 | 0 | #ifdef ERANGE_FILL |
2026 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2027 | 0 | #endif |
2028 | 0 | err = NC_ERANGE; |
2029 | 0 | } |
2030 | 0 | #ifdef ERANGE_FILL |
2031 | 0 | else |
2032 | 0 | #endif |
2033 | 0 | xx = (ix_int)*ip; |
2034 | |
|
2035 | 0 | put_ix_int(xp, &xx); |
2036 | 0 | return err; |
2037 | 0 | } |
2038 | | |
2039 | | static int |
2040 | | ncx_put_int_double(void *xp, const double *ip, void *fillp) |
2041 | 0 | { |
2042 | 0 | int err=NC_NOERR; |
2043 | 0 | ix_int xx = NC_FILL_INT; |
2044 | |
|
2045 | 0 | if (*ip > X_INT_MAX || *ip < X_INT_MIN) { |
2046 | | |
2047 | 0 | #ifdef ERANGE_FILL |
2048 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2049 | 0 | #endif |
2050 | 0 | err = NC_ERANGE; |
2051 | 0 | } |
2052 | 0 | #ifdef ERANGE_FILL |
2053 | 0 | else |
2054 | 0 | #endif |
2055 | 0 | xx = (ix_int)*ip; |
2056 | |
|
2057 | 0 | put_ix_int(xp, &xx); |
2058 | 0 | return err; |
2059 | 0 | } |
2060 | | |
2061 | | |
2062 | | |
2063 | | /* external NC_UINT ---------------------------------------------------------*/ |
2064 | | |
2065 | | #if USHORT_MAX == X_UINT_MAX |
2066 | | typedef ushort ix_uint; |
2067 | | #define SIZEOF_IX_UINT SIZEOF_USHORT |
2068 | | #define IX_UINT_MAX USHORT_MAX |
2069 | | #elif UINT_MAX >= X_UINT_MAX |
2070 | | typedef uint ix_uint; |
2071 | | #define SIZEOF_IX_UINT SIZEOF_UINT |
2072 | 0 | #define IX_UINT_MAX UINT_MAX |
2073 | | #elif ULONG_MAX >= X_UINT_MAX |
2074 | | typedef ulong ix_uint; |
2075 | | #define SIZEOF_IX_UINT SIZEOF_ULONG |
2076 | | #define IX_UINT_MAX ULONG_MAX |
2077 | | #else |
2078 | | #error "ix_uint implementation" |
2079 | | #endif |
2080 | | |
2081 | | |
2082 | | static void |
2083 | | get_ix_uint(const void *xp, ix_uint *ip) |
2084 | 0 | { |
2085 | 0 | const uchar *cp = (const uchar *) xp; |
2086 | |
|
2087 | 0 | *ip = (ix_uint)(*cp++) << 24; |
2088 | 0 | *ip = *ip | (ix_uint)(*cp++) << 16; |
2089 | 0 | *ip = *ip | (ix_uint)(*cp++) << 8; |
2090 | 0 | *ip = *ip | (ix_uint)(*cp); |
2091 | 0 | } |
2092 | | |
2093 | | static void |
2094 | | put_ix_uint(void *xp, const ix_uint *ip) |
2095 | 0 | { |
2096 | 0 | uchar *cp = (uchar *) xp; |
2097 | |
|
2098 | 0 | *cp++ = (uchar)((*ip) >> 24); |
2099 | 0 | *cp++ = (uchar)(((*ip) & 0x00ff0000) >> 16); |
2100 | 0 | *cp++ = (uchar)(((*ip) & 0x0000ff00) >> 8); |
2101 | 0 | *cp = (uchar)( (*ip) & 0x000000ff); |
2102 | 0 | } |
2103 | | |
2104 | | #if X_SIZEOF_UINT != SIZEOF_UINT |
2105 | | static int |
2106 | | ncx_get_uint_uint(const void *xp, uint *ip) |
2107 | | { |
2108 | | int err=NC_NOERR; |
2109 | | #if SIZEOF_IX_UINT == SIZEOF_UINT && IX_UINT_MAX == UINT_MAX |
2110 | | get_ix_uint(xp, (ix_uint *)ip); |
2111 | | #else |
2112 | | ix_uint xx = 0; |
2113 | | get_ix_uint(xp, &xx); |
2114 | | |
2115 | | #if IX_UINT_MAX > UINT_MAX |
2116 | | if (xx > UINT_MAX) { |
2117 | | #ifdef ERANGE_FILL |
2118 | | *ip = NC_FILL_UINT; |
2119 | | return NC_ERANGE; |
2120 | | #else |
2121 | | err = NC_ERANGE; |
2122 | | #endif |
2123 | | } |
2124 | | #endif |
2125 | | |
2126 | | |
2127 | | *ip = (uint) xx; |
2128 | | #endif |
2129 | | return err; |
2130 | | } |
2131 | | |
2132 | | #endif |
2133 | | |
2134 | | static int |
2135 | | ncx_get_uint_schar(const void *xp, schar *ip) |
2136 | 0 | { |
2137 | 0 | int err=NC_NOERR; |
2138 | 0 | ix_uint xx = 0; |
2139 | 0 | get_ix_uint(xp, &xx); |
2140 | |
|
2141 | 0 | #if IX_UINT_MAX > SCHAR_MAX |
2142 | 0 | if (xx > SCHAR_MAX) { |
2143 | 0 | #ifdef ERANGE_FILL |
2144 | 0 | *ip = NC_FILL_BYTE; |
2145 | 0 | return NC_ERANGE; |
2146 | | #else |
2147 | | err = NC_ERANGE; |
2148 | | #endif |
2149 | 0 | } |
2150 | 0 | #endif |
2151 | | |
2152 | | |
2153 | 0 | *ip = (schar) xx; |
2154 | 0 | return err; |
2155 | 0 | } |
2156 | | |
2157 | | static int |
2158 | | ncx_get_uint_short(const void *xp, short *ip) |
2159 | 0 | { |
2160 | 0 | int err=NC_NOERR; |
2161 | 0 | ix_uint xx = 0; |
2162 | 0 | get_ix_uint(xp, &xx); |
2163 | |
|
2164 | 0 | #if IX_UINT_MAX > SHORT_MAX |
2165 | 0 | if (xx > SHORT_MAX) { |
2166 | 0 | #ifdef ERANGE_FILL |
2167 | 0 | *ip = NC_FILL_SHORT; |
2168 | 0 | return NC_ERANGE; |
2169 | | #else |
2170 | | err = NC_ERANGE; |
2171 | | #endif |
2172 | 0 | } |
2173 | 0 | #endif |
2174 | | |
2175 | | |
2176 | 0 | *ip = (short) xx; |
2177 | 0 | return err; |
2178 | 0 | } |
2179 | | |
2180 | | static int |
2181 | | ncx_get_uint_int(const void *xp, int *ip) |
2182 | 0 | { |
2183 | 0 | int err=NC_NOERR; |
2184 | 0 | ix_uint xx = 0; |
2185 | 0 | get_ix_uint(xp, &xx); |
2186 | |
|
2187 | 0 | #if IX_UINT_MAX > INT_MAX |
2188 | 0 | if (xx > INT_MAX) { |
2189 | 0 | #ifdef ERANGE_FILL |
2190 | 0 | *ip = NC_FILL_INT; |
2191 | 0 | return NC_ERANGE; |
2192 | | #else |
2193 | | err = NC_ERANGE; |
2194 | | #endif |
2195 | 0 | } |
2196 | 0 | #endif |
2197 | | |
2198 | | |
2199 | 0 | *ip = (int) xx; |
2200 | 0 | return err; |
2201 | 0 | } |
2202 | | |
2203 | | static int |
2204 | | ncx_get_uint_long(const void *xp, long *ip) |
2205 | 0 | { |
2206 | 0 | int err=NC_NOERR; |
2207 | 0 | ix_uint xx = 0; |
2208 | 0 | get_ix_uint(xp, &xx); |
2209 | |
|
2210 | | #if IX_UINT_MAX > LONG_MAX |
2211 | | if (xx > LONG_MAX) { |
2212 | | #ifdef ERANGE_FILL |
2213 | | *ip = NC_FILL_INT; |
2214 | | return NC_ERANGE; |
2215 | | #else |
2216 | | err = NC_ERANGE; |
2217 | | #endif |
2218 | | } |
2219 | | #endif |
2220 | | |
2221 | |
|
2222 | 0 | *ip = (long) xx; |
2223 | 0 | return err; |
2224 | 0 | } |
2225 | | |
2226 | | static int |
2227 | | ncx_get_uint_longlong(const void *xp, longlong *ip) |
2228 | 0 | { |
2229 | 0 | int err=NC_NOERR; |
2230 | 0 | ix_uint xx = 0; |
2231 | 0 | get_ix_uint(xp, &xx); |
2232 | |
|
2233 | | #if IX_UINT_MAX > LONGLONG_MAX |
2234 | | if (xx > LONGLONG_MAX) { |
2235 | | #ifdef ERANGE_FILL |
2236 | | *ip = NC_FILL_INT64; |
2237 | | return NC_ERANGE; |
2238 | | #else |
2239 | | err = NC_ERANGE; |
2240 | | #endif |
2241 | | } |
2242 | | #endif |
2243 | | |
2244 | |
|
2245 | 0 | *ip = (longlong) xx; |
2246 | 0 | return err; |
2247 | 0 | } |
2248 | | |
2249 | | static int |
2250 | | ncx_get_uint_ushort(const void *xp, ushort *ip) |
2251 | 0 | { |
2252 | 0 | int err=NC_NOERR; |
2253 | | #if SIZEOF_IX_UINT == SIZEOF_USHORT && IX_UINT_MAX == USHORT_MAX |
2254 | | get_ix_uint(xp, (ix_uint *)ip); |
2255 | | #else |
2256 | 0 | ix_uint xx = 0; |
2257 | 0 | get_ix_uint(xp, &xx); |
2258 | |
|
2259 | 0 | #if IX_UINT_MAX > USHORT_MAX |
2260 | 0 | if (xx > USHORT_MAX) { |
2261 | 0 | #ifdef ERANGE_FILL |
2262 | 0 | *ip = NC_FILL_USHORT; |
2263 | 0 | return NC_ERANGE; |
2264 | | #else |
2265 | | err = NC_ERANGE; |
2266 | | #endif |
2267 | 0 | } |
2268 | 0 | #endif |
2269 | | |
2270 | | |
2271 | 0 | *ip = (ushort) xx; |
2272 | 0 | #endif |
2273 | 0 | return err; |
2274 | 0 | } |
2275 | | |
2276 | | static int |
2277 | | ncx_get_uint_uchar(const void *xp, uchar *ip) |
2278 | 0 | { |
2279 | 0 | int err=NC_NOERR; |
2280 | | #if SIZEOF_IX_UINT == SIZEOF_UCHAR && IX_UINT_MAX == UCHAR_MAX |
2281 | | get_ix_uint(xp, (ix_uint *)ip); |
2282 | | #else |
2283 | 0 | ix_uint xx = 0; |
2284 | 0 | get_ix_uint(xp, &xx); |
2285 | |
|
2286 | 0 | #if IX_UINT_MAX > UCHAR_MAX |
2287 | 0 | if (xx > UCHAR_MAX) { |
2288 | 0 | #ifdef ERANGE_FILL |
2289 | 0 | *ip = NC_FILL_UBYTE; |
2290 | 0 | return NC_ERANGE; |
2291 | | #else |
2292 | | err = NC_ERANGE; |
2293 | | #endif |
2294 | 0 | } |
2295 | 0 | #endif |
2296 | | |
2297 | | |
2298 | 0 | *ip = (uchar) xx; |
2299 | 0 | #endif |
2300 | 0 | return err; |
2301 | 0 | } |
2302 | | |
2303 | | static int |
2304 | | ncx_get_uint_ulonglong(const void *xp, ulonglong *ip) |
2305 | 0 | { |
2306 | 0 | int err=NC_NOERR; |
2307 | | #if SIZEOF_IX_UINT == SIZEOF_ULONGLONG && IX_UINT_MAX == ULONGLONG_MAX |
2308 | | get_ix_uint(xp, (ix_uint *)ip); |
2309 | | #else |
2310 | 0 | ix_uint xx = 0; |
2311 | 0 | get_ix_uint(xp, &xx); |
2312 | |
|
2313 | | #if IX_UINT_MAX > ULONGLONG_MAX |
2314 | | if (xx > ULONGLONG_MAX) { |
2315 | | #ifdef ERANGE_FILL |
2316 | | *ip = NC_FILL_UINT64; |
2317 | | return NC_ERANGE; |
2318 | | #else |
2319 | | err = NC_ERANGE; |
2320 | | #endif |
2321 | | } |
2322 | | #endif |
2323 | | |
2324 | |
|
2325 | 0 | *ip = (ulonglong) xx; |
2326 | 0 | #endif |
2327 | 0 | return err; |
2328 | 0 | } |
2329 | | |
2330 | | static int |
2331 | | ncx_get_uint_float(const void *xp, float *ip) |
2332 | 0 | { |
2333 | 0 | ix_uint xx = 0; |
2334 | 0 | get_ix_uint(xp, &xx); |
2335 | 0 | *ip = (float)xx; |
2336 | 0 | return NC_NOERR; |
2337 | 0 | } |
2338 | | |
2339 | | static int |
2340 | | ncx_get_uint_double(const void *xp, double *ip) |
2341 | 0 | { |
2342 | 0 | ix_uint xx = 0; |
2343 | 0 | get_ix_uint(xp, &xx); |
2344 | 0 | *ip = (double)xx; |
2345 | 0 | return NC_NOERR; |
2346 | 0 | } |
2347 | | |
2348 | | |
2349 | | static int |
2350 | | ncx_put_uint_schar(void *xp, const schar *ip, void *fillp) |
2351 | 0 | { |
2352 | 0 | uchar *cp; |
2353 | 0 | if (*ip < 0) { |
2354 | 0 | #ifdef ERANGE_FILL |
2355 | 0 | if (fillp != NULL) memcpy(xp, fillp, 4); |
2356 | 0 | #ifndef WORDS_BIGENDIAN |
2357 | 0 | swapn4b(xp, xp, 1); |
2358 | 0 | #endif |
2359 | 0 | #endif |
2360 | 0 | return NC_ERANGE; |
2361 | 0 | } |
2362 | | |
2363 | 0 | cp = (uchar *) xp; |
2364 | 0 | *cp++ = 0x00; |
2365 | 0 | *cp++ = 0x00; |
2366 | 0 | *cp++ = 0x00; |
2367 | 0 | *cp = (uchar)*ip; |
2368 | |
|
2369 | 0 | return NC_NOERR; |
2370 | 0 | } |
2371 | | |
2372 | | static int |
2373 | | ncx_put_uint_uchar(void *xp, const uchar *ip, void *fillp) |
2374 | 0 | { |
2375 | 0 | uchar *cp = (uchar *) xp; |
2376 | 0 | *cp++ = 0x00; |
2377 | 0 | *cp++ = 0x00; |
2378 | 0 | *cp++ = 0x00; |
2379 | 0 | *cp = *ip; |
2380 | 0 | return NC_NOERR; |
2381 | 0 | } |
2382 | | |
2383 | | #if X_SIZEOF_UINT != SIZEOF_UINT |
2384 | | static int |
2385 | | ncx_put_uint_uint(void *xp, const uint *ip, void *fillp) |
2386 | | { |
2387 | | int err=NC_NOERR; |
2388 | | #if SIZEOF_IX_UINT == SIZEOF_UINT && IX_UINT_MAX == UINT_MAX |
2389 | | put_ix_uint(xp, (const ix_uint *)ip); |
2390 | | #else |
2391 | | ix_uint xx = NC_FILL_UINT; |
2392 | | |
2393 | | #if IX_UINT_MAX < UINT_MAX |
2394 | | if (*ip > IX_UINT_MAX) { |
2395 | | |
2396 | | #ifdef ERANGE_FILL |
2397 | | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2398 | | #endif |
2399 | | err = NC_ERANGE; |
2400 | | } |
2401 | | #ifdef ERANGE_FILL |
2402 | | else |
2403 | | #endif |
2404 | | #endif |
2405 | | xx = (ix_uint)*ip; |
2406 | | |
2407 | | put_ix_uint(xp, &xx); |
2408 | | #endif |
2409 | | return err; |
2410 | | } |
2411 | | |
2412 | | #endif |
2413 | | |
2414 | | static int |
2415 | | ncx_put_uint_short(void *xp, const short *ip, void *fillp) |
2416 | 0 | { |
2417 | 0 | int err=NC_NOERR; |
2418 | 0 | ix_uint xx = NC_FILL_UINT; |
2419 | |
|
2420 | | #if IX_UINT_MAX < SHORT_MAX |
2421 | | if (*ip > IX_UINT_MAX) { |
2422 | | |
2423 | | #ifdef ERANGE_FILL |
2424 | | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2425 | | #endif |
2426 | | err = NC_ERANGE; |
2427 | | } |
2428 | | #ifdef ERANGE_FILL |
2429 | | else |
2430 | | #endif |
2431 | | #endif |
2432 | 0 | if (*ip < 0) { |
2433 | | |
2434 | 0 | #ifdef ERANGE_FILL |
2435 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2436 | 0 | #endif |
2437 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
2438 | 0 | } |
2439 | 0 | #ifdef ERANGE_FILL |
2440 | 0 | else |
2441 | 0 | #endif |
2442 | 0 | xx = (ix_uint)*ip; |
2443 | |
|
2444 | 0 | put_ix_uint(xp, &xx); |
2445 | 0 | return err; |
2446 | 0 | } |
2447 | | |
2448 | | static int |
2449 | | ncx_put_uint_int(void *xp, const int *ip, void *fillp) |
2450 | 0 | { |
2451 | 0 | int err=NC_NOERR; |
2452 | 0 | ix_uint xx = NC_FILL_UINT; |
2453 | |
|
2454 | | #if IX_UINT_MAX < INT_MAX |
2455 | | if (*ip > IX_UINT_MAX) { |
2456 | | |
2457 | | #ifdef ERANGE_FILL |
2458 | | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2459 | | #endif |
2460 | | err = NC_ERANGE; |
2461 | | } |
2462 | | #ifdef ERANGE_FILL |
2463 | | else |
2464 | | #endif |
2465 | | #endif |
2466 | 0 | if (*ip < 0) { |
2467 | | |
2468 | 0 | #ifdef ERANGE_FILL |
2469 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2470 | 0 | #endif |
2471 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
2472 | 0 | } |
2473 | 0 | #ifdef ERANGE_FILL |
2474 | 0 | else |
2475 | 0 | #endif |
2476 | 0 | xx = (ix_uint)*ip; |
2477 | |
|
2478 | 0 | put_ix_uint(xp, &xx); |
2479 | 0 | return err; |
2480 | 0 | } |
2481 | | |
2482 | | static int |
2483 | | ncx_put_uint_long(void *xp, const long *ip, void *fillp) |
2484 | 0 | { |
2485 | 0 | int err=NC_NOERR; |
2486 | 0 | ix_uint xx = NC_FILL_UINT; |
2487 | |
|
2488 | 0 | #if IX_UINT_MAX < LONG_MAX |
2489 | 0 | if (*ip > IX_UINT_MAX) { |
2490 | | |
2491 | 0 | #ifdef ERANGE_FILL |
2492 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2493 | 0 | #endif |
2494 | 0 | err = NC_ERANGE; |
2495 | 0 | } |
2496 | 0 | #ifdef ERANGE_FILL |
2497 | 0 | else |
2498 | 0 | #endif |
2499 | 0 | #endif |
2500 | 0 | if (*ip < 0) { |
2501 | | |
2502 | 0 | #ifdef ERANGE_FILL |
2503 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2504 | 0 | #endif |
2505 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
2506 | 0 | } |
2507 | 0 | #ifdef ERANGE_FILL |
2508 | 0 | else |
2509 | 0 | #endif |
2510 | 0 | xx = (ix_uint)*ip; |
2511 | |
|
2512 | 0 | put_ix_uint(xp, &xx); |
2513 | 0 | return err; |
2514 | 0 | } |
2515 | | |
2516 | | static int |
2517 | | ncx_put_uint_longlong(void *xp, const longlong *ip, void *fillp) |
2518 | 0 | { |
2519 | 0 | int err=NC_NOERR; |
2520 | 0 | ix_uint xx = NC_FILL_UINT; |
2521 | |
|
2522 | 0 | #if IX_UINT_MAX < LONGLONG_MAX |
2523 | 0 | if (*ip > IX_UINT_MAX) { |
2524 | | |
2525 | 0 | #ifdef ERANGE_FILL |
2526 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2527 | 0 | #endif |
2528 | 0 | err = NC_ERANGE; |
2529 | 0 | } |
2530 | 0 | #ifdef ERANGE_FILL |
2531 | 0 | else |
2532 | 0 | #endif |
2533 | 0 | #endif |
2534 | 0 | if (*ip < 0) { |
2535 | | |
2536 | 0 | #ifdef ERANGE_FILL |
2537 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2538 | 0 | #endif |
2539 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
2540 | 0 | } |
2541 | 0 | #ifdef ERANGE_FILL |
2542 | 0 | else |
2543 | 0 | #endif |
2544 | 0 | xx = (ix_uint)*ip; |
2545 | |
|
2546 | 0 | put_ix_uint(xp, &xx); |
2547 | 0 | return err; |
2548 | 0 | } |
2549 | | |
2550 | | static int |
2551 | | ncx_put_uint_ushort(void *xp, const ushort *ip, void *fillp) |
2552 | 0 | { |
2553 | 0 | int err=NC_NOERR; |
2554 | | #if SIZEOF_IX_UINT == SIZEOF_USHORT && IX_UINT_MAX == USHORT_MAX |
2555 | | put_ix_uint(xp, (const ix_uint *)ip); |
2556 | | #else |
2557 | 0 | ix_uint xx = NC_FILL_UINT; |
2558 | |
|
2559 | | #if IX_UINT_MAX < USHORT_MAX |
2560 | | if (*ip > IX_UINT_MAX) { |
2561 | | |
2562 | | #ifdef ERANGE_FILL |
2563 | | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2564 | | #endif |
2565 | | err = NC_ERANGE; |
2566 | | } |
2567 | | #ifdef ERANGE_FILL |
2568 | | else |
2569 | | #endif |
2570 | | #endif |
2571 | 0 | xx = (ix_uint)*ip; |
2572 | |
|
2573 | 0 | put_ix_uint(xp, &xx); |
2574 | 0 | #endif |
2575 | 0 | return err; |
2576 | 0 | } |
2577 | | |
2578 | | static int |
2579 | | ncx_put_uint_ulonglong(void *xp, const ulonglong *ip, void *fillp) |
2580 | 0 | { |
2581 | 0 | int err=NC_NOERR; |
2582 | | #if SIZEOF_IX_UINT == SIZEOF_ULONGLONG && IX_UINT_MAX == ULONGLONG_MAX |
2583 | | put_ix_uint(xp, (const ix_uint *)ip); |
2584 | | #else |
2585 | 0 | ix_uint xx = NC_FILL_UINT; |
2586 | |
|
2587 | 0 | #if IX_UINT_MAX < ULONGLONG_MAX |
2588 | 0 | if (*ip > IX_UINT_MAX) { |
2589 | | |
2590 | 0 | #ifdef ERANGE_FILL |
2591 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2592 | 0 | #endif |
2593 | 0 | err = NC_ERANGE; |
2594 | 0 | } |
2595 | 0 | #ifdef ERANGE_FILL |
2596 | 0 | else |
2597 | 0 | #endif |
2598 | 0 | #endif |
2599 | 0 | xx = (ix_uint)*ip; |
2600 | |
|
2601 | 0 | put_ix_uint(xp, &xx); |
2602 | 0 | #endif |
2603 | 0 | return err; |
2604 | 0 | } |
2605 | | |
2606 | | static int |
2607 | | ncx_put_uint_float(void *xp, const float *ip, void *fillp) |
2608 | 0 | { |
2609 | 0 | int err=NC_NOERR; |
2610 | 0 | ix_uint xx = NC_FILL_UINT; |
2611 | |
|
2612 | 0 | if (*ip > (double)X_UINT_MAX || *ip < 0) { |
2613 | | |
2614 | 0 | #ifdef ERANGE_FILL |
2615 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2616 | 0 | #endif |
2617 | 0 | err = NC_ERANGE; |
2618 | 0 | } |
2619 | 0 | #ifdef ERANGE_FILL |
2620 | 0 | else |
2621 | 0 | #endif |
2622 | 0 | xx = (ix_uint)*ip; |
2623 | |
|
2624 | 0 | put_ix_uint(xp, &xx); |
2625 | 0 | return err; |
2626 | 0 | } |
2627 | | |
2628 | | static int |
2629 | | ncx_put_uint_double(void *xp, const double *ip, void *fillp) |
2630 | 0 | { |
2631 | 0 | int err=NC_NOERR; |
2632 | 0 | ix_uint xx = NC_FILL_UINT; |
2633 | |
|
2634 | 0 | if (*ip > X_UINT_MAX || *ip < 0) { |
2635 | | |
2636 | 0 | #ifdef ERANGE_FILL |
2637 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
2638 | 0 | #endif |
2639 | 0 | err = NC_ERANGE; |
2640 | 0 | } |
2641 | 0 | #ifdef ERANGE_FILL |
2642 | 0 | else |
2643 | 0 | #endif |
2644 | 0 | xx = (ix_uint)*ip; |
2645 | |
|
2646 | 0 | put_ix_uint(xp, &xx); |
2647 | 0 | return err; |
2648 | 0 | } |
2649 | | |
2650 | | |
2651 | | |
2652 | | /* external NC_FLOAT --------------------------------------------------------*/ |
2653 | | |
2654 | | #if X_SIZEOF_FLOAT == SIZEOF_FLOAT && !defined(NO_IEEE_FLOAT) |
2655 | | |
2656 | | inline static void |
2657 | | get_ix_float(const void *xp, float *ip) |
2658 | 0 | { |
2659 | | #ifdef WORDS_BIGENDIAN |
2660 | | (void) memcpy(ip, xp, SIZEOF_FLOAT); |
2661 | | #else |
2662 | 0 | swap4b(ip, xp); |
2663 | 0 | #endif |
2664 | 0 | } |
2665 | | |
2666 | | inline static void |
2667 | | put_ix_float(void *xp, const float *ip) |
2668 | 0 | { |
2669 | | #ifdef WORDS_BIGENDIAN |
2670 | | (void) memcpy(xp, ip, X_SIZEOF_FLOAT); |
2671 | | #else |
2672 | 0 | swap4b(xp, ip); |
2673 | 0 | #endif |
2674 | 0 | } |
2675 | | |
2676 | | #elif defined(vax) && vax != 0 |
2677 | | |
2678 | | /* What IEEE single precision floating point looks like on a Vax */ |
2679 | | struct ieee_single { |
2680 | | unsigned int exp_hi : 7; |
2681 | | unsigned int sign : 1; |
2682 | | unsigned int mant_hi : 7; |
2683 | | unsigned int exp_lo : 1; |
2684 | | unsigned int mant_lo_hi : 8; |
2685 | | unsigned int mant_lo_lo : 8; |
2686 | | }; |
2687 | | |
2688 | | /* Vax single precision floating point */ |
2689 | | struct vax_single { |
2690 | | unsigned int mantissa1 : 7; |
2691 | | unsigned int exp : 8; |
2692 | | unsigned int sign : 1; |
2693 | | unsigned int mantissa2 : 16; |
2694 | | }; |
2695 | | |
2696 | | #define VAX_SNG_BIAS 0x81 |
2697 | | #define IEEE_SNG_BIAS 0x7f |
2698 | | |
2699 | | static struct sgl_limits { |
2700 | | struct vax_single s; |
2701 | | struct ieee_single ieee; |
2702 | | } max = { |
2703 | | { 0x7f, 0xff, 0x0, 0xffff }, /* Max Vax */ |
2704 | | { 0x7f, 0x0, 0x0, 0x1, 0x0, 0x0 } /* Max IEEE */ |
2705 | | }; |
2706 | | static struct sgl_limits min = { |
2707 | | { 0x0, 0x0, 0x0, 0x0 }, /* Min Vax */ |
2708 | | { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 } /* Min IEEE */ |
2709 | | }; |
2710 | | |
2711 | | static void |
2712 | | get_ix_float(const void *xp, float *ip) |
2713 | | { |
2714 | | struct vax_single *const vsp = (struct vax_single *) ip; |
2715 | | const struct ieee_single *const isp = |
2716 | | (const struct ieee_single *) xp; |
2717 | | unsigned exp = isp->exp_hi << 1 | isp->exp_lo; |
2718 | | |
2719 | | switch(exp) { |
2720 | | case 0 : |
2721 | | /* ieee subnormal */ |
2722 | | if (isp->mant_hi == min.ieee.mant_hi |
2723 | | && isp->mant_lo_hi == min.ieee.mant_lo_hi |
2724 | | && isp->mant_lo_lo == min.ieee.mant_lo_lo) |
2725 | | { |
2726 | | *vsp = min.s; |
2727 | | } |
2728 | | else |
2729 | | { |
2730 | | unsigned mantissa = (isp->mant_hi << 16) |
2731 | | | isp->mant_lo_hi << 8 |
2732 | | | isp->mant_lo_lo; |
2733 | | unsigned tmp = mantissa >> 20; |
2734 | | if (tmp >= 4) { |
2735 | | vsp->exp = 2; |
2736 | | } else if (tmp >= 2) { |
2737 | | vsp->exp = 1; |
2738 | | } else { |
2739 | | *vsp = min.s; |
2740 | | break; |
2741 | | } /* else */ |
2742 | | tmp = mantissa - (1 << (20 + vsp->exp )); |
2743 | | tmp <<= 3 - vsp->exp; |
2744 | | vsp->mantissa2 = tmp; |
2745 | | vsp->mantissa1 = (tmp >> 16); |
2746 | | } |
2747 | | break; |
2748 | | case 0xfe : |
2749 | | case 0xff : |
2750 | | *vsp = max.s; |
2751 | | break; |
2752 | | default : |
2753 | | vsp->exp = exp - IEEE_SNG_BIAS + VAX_SNG_BIAS; |
2754 | | vsp->mantissa2 = isp->mant_lo_hi << 8 | isp->mant_lo_lo; |
2755 | | vsp->mantissa1 = isp->mant_hi; |
2756 | | } |
2757 | | |
2758 | | vsp->sign = isp->sign; |
2759 | | |
2760 | | } |
2761 | | |
2762 | | |
2763 | | static void |
2764 | | put_ix_float(void *xp, const float *ip) |
2765 | | { |
2766 | | const struct vax_single *const vsp = |
2767 | | (const struct vax_single *)ip; |
2768 | | struct ieee_single *const isp = (struct ieee_single *) xp; |
2769 | | |
2770 | | switch(vsp->exp){ |
2771 | | case 0 : |
2772 | | /* all vax float with zero exponent map to zero */ |
2773 | | *isp = min.ieee; |
2774 | | break; |
2775 | | case 2 : |
2776 | | case 1 : |
2777 | | { |
2778 | | /* These will map to subnormals */ |
2779 | | unsigned mantissa = (vsp->mantissa1 << 16) |
2780 | | | vsp->mantissa2; |
2781 | | mantissa >>= 3 - vsp->exp; |
2782 | | mantissa += (1 << (20 + vsp->exp)); |
2783 | | isp->mant_lo_lo = mantissa; |
2784 | | isp->mant_lo_hi = mantissa >> 8; |
2785 | | isp->mant_hi = mantissa >> 16; |
2786 | | isp->exp_lo = 0; |
2787 | | isp->exp_hi = 0; |
2788 | | } |
2789 | | break; |
2790 | | case 0xff : /* max.s.exp */ |
2791 | | if (vsp->mantissa2 == max.s.mantissa2 && |
2792 | | vsp->mantissa1 == max.s.mantissa1) |
2793 | | { |
2794 | | /* map largest vax float to ieee infinity */ |
2795 | | *isp = max.ieee; |
2796 | | break; |
2797 | | } /* else, fall thru */ |
2798 | | default : |
2799 | | { |
2800 | | unsigned exp = vsp->exp - VAX_SNG_BIAS + IEEE_SNG_BIAS; |
2801 | | isp->exp_hi = exp >> 1; |
2802 | | isp->exp_lo = exp; |
2803 | | isp->mant_lo_lo = vsp->mantissa2; |
2804 | | isp->mant_lo_hi = vsp->mantissa2 >> 8; |
2805 | | isp->mant_hi = vsp->mantissa1; |
2806 | | } |
2807 | | } |
2808 | | |
2809 | | isp->sign = vsp->sign; |
2810 | | |
2811 | | } |
2812 | | |
2813 | | /* vax */ |
2814 | | #elif defined(_CRAY) && !defined(__crayx1) |
2815 | | |
2816 | | /* |
2817 | | * Return the number of bytes until the next "word" boundary |
2818 | | * N.B. This is based on the very weird YMP address structure, |
2819 | | * which puts the address within a word in the leftmost 3 bits |
2820 | | * of the address. |
2821 | | */ |
2822 | | static size_t |
2823 | | word_align(const void *vp) |
2824 | | { |
2825 | | const size_t rem = ((size_t)vp >> (64 - 3)) & 0x7; |
2826 | | return (rem != 0); |
2827 | | } |
2828 | | |
2829 | | struct ieee_single_hi { |
2830 | | unsigned int sign : 1; |
2831 | | unsigned int exp : 8; |
2832 | | unsigned int mant :23; |
2833 | | unsigned int pad :32; |
2834 | | }; |
2835 | | typedef struct ieee_single_hi ieee_single_hi; |
2836 | | |
2837 | | struct ieee_single_lo { |
2838 | | unsigned int pad :32; |
2839 | | unsigned int sign : 1; |
2840 | | unsigned int exp : 8; |
2841 | | unsigned int mant :23; |
2842 | | }; |
2843 | | typedef struct ieee_single_lo ieee_single_lo; |
2844 | | |
2845 | | static const int ieee_single_bias = 0x7f; |
2846 | | |
2847 | | struct ieee_double { |
2848 | | unsigned int sign : 1; |
2849 | | unsigned int exp :11; |
2850 | | unsigned int mant :52; |
2851 | | }; |
2852 | | typedef struct ieee_double ieee_double; |
2853 | | |
2854 | | static const int ieee_double_bias = 0x3ff; |
2855 | | |
2856 | | #if defined(NO_IEEE_FLOAT) |
2857 | | |
2858 | | struct cray_single { |
2859 | | unsigned int sign : 1; |
2860 | | unsigned int exp :15; |
2861 | | unsigned int mant :48; |
2862 | | }; |
2863 | | typedef struct cray_single cray_single; |
2864 | | |
2865 | | static const int cs_ieis_bias = 0x4000 - 0x7f; |
2866 | | |
2867 | | static const int cs_id_bias = 0x4000 - 0x3ff; |
2868 | | |
2869 | | |
2870 | | static void |
2871 | | get_ix_float(const void *xp, float *ip) |
2872 | | { |
2873 | | |
2874 | | if (word_align(xp) == 0) |
2875 | | { |
2876 | | const ieee_single_hi *isp = (const ieee_single_hi *) xp; |
2877 | | cray_single *csp = (cray_single *) ip; |
2878 | | |
2879 | | if (isp->exp == 0) |
2880 | | { |
2881 | | /* ieee subnormal */ |
2882 | | *ip = (double)isp->mant; |
2883 | | if (isp->mant != 0) |
2884 | | { |
2885 | | csp->exp -= (ieee_single_bias + 22); |
2886 | | } |
2887 | | } |
2888 | | else |
2889 | | { |
2890 | | csp->exp = isp->exp + cs_ieis_bias + 1; |
2891 | | csp->mant = isp->mant << (48 - 1 - 23); |
2892 | | csp->mant |= (1 << (48 - 1)); |
2893 | | } |
2894 | | csp->sign = isp->sign; |
2895 | | |
2896 | | |
2897 | | } |
2898 | | else |
2899 | | { |
2900 | | const ieee_single_lo *isp = (const ieee_single_lo *) xp; |
2901 | | cray_single *csp = (cray_single *) ip; |
2902 | | |
2903 | | if (isp->exp == 0) |
2904 | | { |
2905 | | /* ieee subnormal */ |
2906 | | *ip = (double)isp->mant; |
2907 | | if (isp->mant != 0) |
2908 | | { |
2909 | | csp->exp -= (ieee_single_bias + 22); |
2910 | | } |
2911 | | } |
2912 | | else |
2913 | | { |
2914 | | csp->exp = isp->exp + cs_ieis_bias + 1; |
2915 | | csp->mant = isp->mant << (48 - 1 - 23); |
2916 | | csp->mant |= (1 << (48 - 1)); |
2917 | | } |
2918 | | csp->sign = isp->sign; |
2919 | | |
2920 | | |
2921 | | } |
2922 | | } |
2923 | | |
2924 | | static void |
2925 | | put_ix_float(void *xp, const float *ip) |
2926 | | { |
2927 | | if (word_align(xp) == 0) |
2928 | | { |
2929 | | ieee_single_hi *isp = (ieee_single_hi*)xp; |
2930 | | const cray_single *csp = (const cray_single *) ip; |
2931 | | int ieee_exp = csp->exp - cs_ieis_bias -1; |
2932 | | |
2933 | | isp->sign = csp->sign; |
2934 | | |
2935 | | if (ieee_exp >= 0xff) |
2936 | | { |
2937 | | /* NC_ERANGE => ieee Inf */ |
2938 | | isp->exp = 0xff; |
2939 | | isp->mant = 0x0; |
2940 | | } |
2941 | | else if (ieee_exp > 0) |
2942 | | { |
2943 | | /* normal ieee representation */ |
2944 | | isp->exp = ieee_exp; |
2945 | | /* assumes cray rep is in normal form */ |
2946 | | assert(csp->mant & 0x800000000000); |
2947 | | isp->mant = (((csp->mant << 1) & |
2948 | | 0xffffffffffff) >> (48 - 23)); |
2949 | | } |
2950 | | else if (ieee_exp > -23) |
2951 | | { |
2952 | | /* ieee subnormal, right shift */ |
2953 | | const int rshift = (48 - 23 - ieee_exp); |
2954 | | |
2955 | | isp->mant = csp->mant >> rshift; |
2956 | | |
2957 | | #if 0 |
2958 | | if (csp->mant & (1 << (rshift -1))) |
2959 | | { |
2960 | | /* round up */ |
2961 | | isp->mant++; |
2962 | | } |
2963 | | #endif |
2964 | | |
2965 | | isp->exp = 0; |
2966 | | } |
2967 | | else |
2968 | | { |
2969 | | /* smaller than ieee can represent */ |
2970 | | isp->exp = 0; |
2971 | | isp->mant = 0; |
2972 | | } |
2973 | | |
2974 | | } |
2975 | | else |
2976 | | { |
2977 | | ieee_single_lo *isp = (ieee_single_lo*)xp; |
2978 | | const cray_single *csp = (const cray_single *) ip; |
2979 | | int ieee_exp = csp->exp - cs_ieis_bias -1; |
2980 | | |
2981 | | isp->sign = csp->sign; |
2982 | | |
2983 | | if (ieee_exp >= 0xff) |
2984 | | { |
2985 | | /* NC_ERANGE => ieee Inf */ |
2986 | | isp->exp = 0xff; |
2987 | | isp->mant = 0x0; |
2988 | | } |
2989 | | else if (ieee_exp > 0) |
2990 | | { |
2991 | | /* normal ieee representation */ |
2992 | | isp->exp = ieee_exp; |
2993 | | /* assumes cray rep is in normal form */ |
2994 | | assert(csp->mant & 0x800000000000); |
2995 | | isp->mant = (((csp->mant << 1) & |
2996 | | 0xffffffffffff) >> (48 - 23)); |
2997 | | } |
2998 | | else if (ieee_exp > -23) |
2999 | | { |
3000 | | /* ieee subnormal, right shift */ |
3001 | | const int rshift = (48 - 23 - ieee_exp); |
3002 | | |
3003 | | isp->mant = csp->mant >> rshift; |
3004 | | |
3005 | | #if 0 |
3006 | | if (csp->mant & (1 << (rshift -1))) |
3007 | | { |
3008 | | /* round up */ |
3009 | | isp->mant++; |
3010 | | } |
3011 | | #endif |
3012 | | |
3013 | | isp->exp = 0; |
3014 | | } |
3015 | | else |
3016 | | { |
3017 | | /* smaller than ieee can represent */ |
3018 | | isp->exp = 0; |
3019 | | isp->mant = 0; |
3020 | | } |
3021 | | |
3022 | | } |
3023 | | } |
3024 | | |
3025 | | #else |
3026 | | /* IEEE Cray with only doubles */ |
3027 | | static void |
3028 | | get_ix_float(const void *xp, float *ip) |
3029 | | { |
3030 | | |
3031 | | ieee_double *idp = (ieee_double *) ip; |
3032 | | |
3033 | | if (word_align(xp) == 0) |
3034 | | { |
3035 | | const ieee_single_hi *isp = (const ieee_single_hi *) xp; |
3036 | | if (isp->exp == 0 && isp->mant == 0) |
3037 | | { |
3038 | | idp->exp = 0; |
3039 | | idp->mant = 0; |
3040 | | } |
3041 | | else |
3042 | | { |
3043 | | idp->exp = isp->exp + (ieee_double_bias - ieee_single_bias); |
3044 | | idp->mant = isp->mant << (52 - 23); |
3045 | | } |
3046 | | idp->sign = isp->sign; |
3047 | | } |
3048 | | else |
3049 | | { |
3050 | | const ieee_single_lo *isp = (const ieee_single_lo *) xp; |
3051 | | if (isp->exp == 0 && isp->mant == 0) |
3052 | | { |
3053 | | idp->exp = 0; |
3054 | | idp->mant = 0; |
3055 | | } |
3056 | | else |
3057 | | { |
3058 | | idp->exp = isp->exp + (ieee_double_bias - ieee_single_bias); |
3059 | | idp->mant = isp->mant << (52 - 23); |
3060 | | } |
3061 | | idp->sign = isp->sign; |
3062 | | } |
3063 | | } |
3064 | | |
3065 | | static void |
3066 | | put_ix_float(void *xp, const float *ip) |
3067 | | { |
3068 | | const ieee_double *idp = (const ieee_double *) ip; |
3069 | | if (word_align(xp) == 0) |
3070 | | { |
3071 | | ieee_single_hi *isp = (ieee_single_hi*)xp; |
3072 | | if (idp->exp > (ieee_double_bias - ieee_single_bias)) |
3073 | | isp->exp = idp->exp - (ieee_double_bias - ieee_single_bias); |
3074 | | else |
3075 | | isp->exp = 0; |
3076 | | isp->mant = idp->mant >> (52 - 23); |
3077 | | isp->sign = idp->sign; |
3078 | | } |
3079 | | else |
3080 | | { |
3081 | | ieee_single_lo *isp = (ieee_single_lo*)xp; |
3082 | | if (idp->exp > (ieee_double_bias - ieee_single_bias)) |
3083 | | isp->exp = idp->exp - (ieee_double_bias - ieee_single_bias); |
3084 | | else |
3085 | | isp->exp = 0; |
3086 | | isp->mant = idp->mant >> (52 - 23); |
3087 | | isp->sign = idp->sign; |
3088 | | } |
3089 | | } |
3090 | | #endif |
3091 | | |
3092 | | #else |
3093 | | #error "ix_float implementation" |
3094 | | #endif |
3095 | | |
3096 | | #if X_SIZEOF_FLOAT != SIZEOF_FLOAT || defined(NO_IEEE_FLOAT) |
3097 | | static int |
3098 | | ncx_get_float_float(const void *xp, float *ip, void *fillp) |
3099 | | { |
3100 | | /* TODO */ |
3101 | | get_ix_float(xp, ip); |
3102 | | return NC_NOERR; |
3103 | | } |
3104 | | #endif |
3105 | | |
3106 | 0 | #define ix_float float |
3107 | | |
3108 | | static int |
3109 | | ncx_get_float_schar(const void *xp, schar *ip) |
3110 | 0 | { |
3111 | 0 | ix_float xx = 0; |
3112 | 0 | get_ix_float(xp, &xx); |
3113 | 0 | if (xx > (double)SCHAR_MAX || xx < (double)SCHAR_MIN) { |
3114 | 0 | #ifdef ERANGE_FILL |
3115 | 0 | *ip = NC_FILL_BYTE; |
3116 | 0 | #endif |
3117 | 0 | return NC_ERANGE; |
3118 | 0 | } |
3119 | 0 | *ip = (schar)xx; |
3120 | 0 | return NC_NOERR; |
3121 | 0 | } |
3122 | | |
3123 | | static int |
3124 | | ncx_get_float_short(const void *xp, short *ip) |
3125 | 0 | { |
3126 | 0 | ix_float xx = 0; |
3127 | 0 | get_ix_float(xp, &xx); |
3128 | 0 | if (xx > (double)SHORT_MAX || xx < (double)SHORT_MIN) { |
3129 | 0 | #ifdef ERANGE_FILL |
3130 | 0 | *ip = NC_FILL_SHORT; |
3131 | 0 | #endif |
3132 | 0 | return NC_ERANGE; |
3133 | 0 | } |
3134 | 0 | *ip = (short)xx; |
3135 | 0 | return NC_NOERR; |
3136 | 0 | } |
3137 | | |
3138 | | static int |
3139 | | ncx_get_float_int(const void *xp, int *ip) |
3140 | 0 | { |
3141 | 0 | ix_float xx = 0; |
3142 | 0 | get_ix_float(xp, &xx); |
3143 | 0 | if (xx > (double)INT_MAX || xx < (double)INT_MIN) { |
3144 | 0 | #ifdef ERANGE_FILL |
3145 | 0 | *ip = NC_FILL_INT; |
3146 | 0 | #endif |
3147 | 0 | return NC_ERANGE; |
3148 | 0 | } |
3149 | 0 | *ip = (int)xx; |
3150 | 0 | return NC_NOERR; |
3151 | 0 | } |
3152 | | |
3153 | | static int |
3154 | | ncx_get_float_long(const void *xp, long *ip) |
3155 | 0 | { |
3156 | 0 | ix_float xx = 0; |
3157 | 0 | get_ix_float(xp, &xx); |
3158 | 0 | if (xx > (double)LONG_MAX || xx < (double)LONG_MIN) { |
3159 | 0 | #ifdef ERANGE_FILL |
3160 | 0 | *ip = NC_FILL_INT; |
3161 | 0 | #endif |
3162 | 0 | return NC_ERANGE; |
3163 | 0 | } |
3164 | 0 | *ip = (long)xx; |
3165 | 0 | return NC_NOERR; |
3166 | 0 | } |
3167 | | |
3168 | | static int |
3169 | | ncx_get_float_double(const void *xp, double *ip) |
3170 | 0 | { |
3171 | 0 | ix_float xx = 0; |
3172 | 0 | get_ix_float(xp, &xx); |
3173 | 0 | *ip = (double)xx; |
3174 | 0 | return NC_NOERR; |
3175 | 0 | } |
3176 | | |
3177 | | static int |
3178 | | ncx_get_float_longlong(const void *xp, longlong *ip) |
3179 | 0 | { |
3180 | 0 | ix_float xx = 0; |
3181 | 0 | get_ix_float(xp, &xx); |
3182 | 0 | if (xx == LONGLONG_MAX) *ip = LONGLONG_MAX; |
3183 | 0 | else if (xx == LONGLONG_MIN) *ip = LONGLONG_MIN; |
3184 | 0 | else if (xx > (double)LONGLONG_MAX || xx < (double)LONGLONG_MIN) { |
3185 | 0 | #ifdef ERANGE_FILL |
3186 | 0 | *ip = NC_FILL_INT64; |
3187 | 0 | #endif |
3188 | 0 | return NC_ERANGE; |
3189 | 0 | } |
3190 | 0 | else *ip = (longlong)xx; |
3191 | 0 | return NC_NOERR; |
3192 | 0 | } |
3193 | | |
3194 | | static int |
3195 | | ncx_get_float_uchar(const void *xp, uchar *ip) |
3196 | 0 | { |
3197 | 0 | ix_float xx = 0; |
3198 | 0 | get_ix_float(xp, &xx); |
3199 | 0 | if (xx > (double)UCHAR_MAX || xx < 0) { |
3200 | 0 | #ifdef ERANGE_FILL |
3201 | 0 | *ip = NC_FILL_UBYTE; |
3202 | 0 | #endif |
3203 | 0 | return NC_ERANGE; |
3204 | 0 | } |
3205 | 0 | *ip = (uchar)xx; |
3206 | 0 | return NC_NOERR; |
3207 | 0 | } |
3208 | | |
3209 | | static int |
3210 | | ncx_get_float_ushort(const void *xp, ushort *ip) |
3211 | 0 | { |
3212 | 0 | ix_float xx = 0; |
3213 | 0 | get_ix_float(xp, &xx); |
3214 | 0 | if (xx > (double)USHORT_MAX || xx < 0) { |
3215 | 0 | #ifdef ERANGE_FILL |
3216 | 0 | *ip = NC_FILL_USHORT; |
3217 | 0 | #endif |
3218 | 0 | return NC_ERANGE; |
3219 | 0 | } |
3220 | 0 | *ip = (ushort)xx; |
3221 | 0 | return NC_NOERR; |
3222 | 0 | } |
3223 | | |
3224 | | static int |
3225 | | ncx_get_float_uint(const void *xp, uint *ip) |
3226 | 0 | { |
3227 | 0 | ix_float xx = 0; |
3228 | 0 | get_ix_float(xp, &xx); |
3229 | 0 | if (xx > (double)UINT_MAX || xx < 0) { |
3230 | 0 | #ifdef ERANGE_FILL |
3231 | 0 | *ip = NC_FILL_UINT; |
3232 | 0 | #endif |
3233 | 0 | return NC_ERANGE; |
3234 | 0 | } |
3235 | 0 | *ip = (uint)xx; |
3236 | 0 | return NC_NOERR; |
3237 | 0 | } |
3238 | | |
3239 | | static int |
3240 | | ncx_get_float_ulonglong(const void *xp, ulonglong *ip) |
3241 | 0 | { |
3242 | 0 | ix_float xx = 0; |
3243 | 0 | get_ix_float(xp, &xx); |
3244 | 0 | if (xx == ULONGLONG_MAX) *ip = ULONGLONG_MAX; |
3245 | 0 | else if (xx > (double)ULONGLONG_MAX || xx < 0) { |
3246 | 0 | #ifdef ERANGE_FILL |
3247 | 0 | *ip = NC_FILL_UINT64; |
3248 | 0 | #endif |
3249 | 0 | return NC_ERANGE; |
3250 | 0 | } |
3251 | 0 | else *ip = (ulonglong)xx; |
3252 | 0 | return NC_NOERR; |
3253 | 0 | } |
3254 | | |
3255 | | |
3256 | | #if X_SIZEOF_FLOAT != SIZEOF_FLOAT || defined(NO_IEEE_FLOAT) |
3257 | | static int |
3258 | | ncx_put_float_float(void *xp, const float *ip, void *fillp) |
3259 | | { |
3260 | | int err=NC_NOERR; |
3261 | | float *_ip=ip; |
3262 | | #ifdef NO_IEEE_FLOAT |
3263 | | #ifdef ERANGE_FILL |
3264 | | float tmp; |
3265 | | #endif |
3266 | | if (*ip > X_FLOAT_MAX || *ip < X_FLOAT_MIN) { |
3267 | | |
3268 | | #ifdef ERANGE_FILL |
3269 | | if (fillp != NULL) memcpy(&tmp, fillp, 4); |
3270 | | #endif |
3271 | | #ifdef ERANGE_FILL |
3272 | | _ip = &tmp; |
3273 | | #endif |
3274 | | err = NC_ERANGE; |
3275 | | } |
3276 | | #endif |
3277 | | put_ix_float(xp, _ip); |
3278 | | return err; |
3279 | | } |
3280 | | #endif |
3281 | | |
3282 | | static int |
3283 | | ncx_put_float_schar(void *xp, const schar *ip, void *fillp) |
3284 | 0 | { |
3285 | 0 | int err=NC_NOERR; |
3286 | 0 | ix_float xx = NC_FILL_FLOAT; |
3287 | | |
3288 | | |
3289 | 0 | xx = (ix_float)*ip; |
3290 | |
|
3291 | 0 | put_ix_float(xp, &xx); |
3292 | 0 | return err; |
3293 | 0 | } |
3294 | | |
3295 | | static int |
3296 | | ncx_put_float_short(void *xp, const short *ip, void *fillp) |
3297 | 0 | { |
3298 | 0 | int err=NC_NOERR; |
3299 | 0 | ix_float xx = NC_FILL_FLOAT; |
3300 | | |
3301 | | |
3302 | 0 | xx = (ix_float)*ip; |
3303 | |
|
3304 | 0 | put_ix_float(xp, &xx); |
3305 | 0 | return err; |
3306 | 0 | } |
3307 | | |
3308 | | static int |
3309 | | ncx_put_float_int(void *xp, const int *ip, void *fillp) |
3310 | 0 | { |
3311 | 0 | int err=NC_NOERR; |
3312 | 0 | ix_float xx = NC_FILL_FLOAT; |
3313 | | |
3314 | | |
3315 | 0 | xx = (ix_float)*ip; |
3316 | |
|
3317 | 0 | put_ix_float(xp, &xx); |
3318 | 0 | return err; |
3319 | 0 | } |
3320 | | |
3321 | | static int |
3322 | | ncx_put_float_long(void *xp, const long *ip, void *fillp) |
3323 | 0 | { |
3324 | 0 | int err=NC_NOERR; |
3325 | 0 | ix_float xx = NC_FILL_FLOAT; |
3326 | | |
3327 | | |
3328 | 0 | xx = (ix_float)*ip; |
3329 | |
|
3330 | 0 | put_ix_float(xp, &xx); |
3331 | 0 | return err; |
3332 | 0 | } |
3333 | | |
3334 | | static int |
3335 | | ncx_put_float_double(void *xp, const double *ip, void *fillp) |
3336 | 0 | { |
3337 | 0 | int err=NC_NOERR; |
3338 | 0 | ix_float xx = NC_FILL_FLOAT; |
3339 | |
|
3340 | 0 | if (*ip > X_FLOAT_MAX || *ip < X_FLOAT_MIN) { |
3341 | | |
3342 | 0 | #ifdef ERANGE_FILL |
3343 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 4); |
3344 | 0 | #endif |
3345 | 0 | err = NC_ERANGE; |
3346 | 0 | } |
3347 | 0 | #ifdef ERANGE_FILL |
3348 | 0 | else |
3349 | 0 | #endif |
3350 | 0 | xx = (ix_float)*ip; |
3351 | |
|
3352 | 0 | put_ix_float(xp, &xx); |
3353 | 0 | return err; |
3354 | 0 | } |
3355 | | |
3356 | | static int |
3357 | | ncx_put_float_longlong(void *xp, const longlong *ip, void *fillp) |
3358 | 0 | { |
3359 | 0 | int err=NC_NOERR; |
3360 | 0 | ix_float xx = NC_FILL_FLOAT; |
3361 | | |
3362 | | |
3363 | 0 | xx = (ix_float)*ip; |
3364 | |
|
3365 | 0 | put_ix_float(xp, &xx); |
3366 | 0 | return err; |
3367 | 0 | } |
3368 | | |
3369 | | static int |
3370 | | ncx_put_float_uchar(void *xp, const uchar *ip, void *fillp) |
3371 | 0 | { |
3372 | 0 | int err=NC_NOERR; |
3373 | 0 | ix_float xx = NC_FILL_FLOAT; |
3374 | | |
3375 | | |
3376 | 0 | xx = (ix_float)*ip; |
3377 | |
|
3378 | 0 | put_ix_float(xp, &xx); |
3379 | 0 | return err; |
3380 | 0 | } |
3381 | | |
3382 | | static int |
3383 | | ncx_put_float_ushort(void *xp, const ushort *ip, void *fillp) |
3384 | 0 | { |
3385 | 0 | int err=NC_NOERR; |
3386 | 0 | ix_float xx = NC_FILL_FLOAT; |
3387 | | |
3388 | | |
3389 | 0 | xx = (ix_float)*ip; |
3390 | |
|
3391 | 0 | put_ix_float(xp, &xx); |
3392 | 0 | return err; |
3393 | 0 | } |
3394 | | |
3395 | | static int |
3396 | | ncx_put_float_uint(void *xp, const uint *ip, void *fillp) |
3397 | 0 | { |
3398 | 0 | int err=NC_NOERR; |
3399 | 0 | ix_float xx = NC_FILL_FLOAT; |
3400 | | |
3401 | | |
3402 | 0 | xx = (ix_float)*ip; |
3403 | |
|
3404 | 0 | put_ix_float(xp, &xx); |
3405 | 0 | return err; |
3406 | 0 | } |
3407 | | |
3408 | | static int |
3409 | | ncx_put_float_ulonglong(void *xp, const ulonglong *ip, void *fillp) |
3410 | 0 | { |
3411 | 0 | int err=NC_NOERR; |
3412 | 0 | ix_float xx = NC_FILL_FLOAT; |
3413 | | |
3414 | | |
3415 | 0 | xx = (ix_float)*ip; |
3416 | |
|
3417 | 0 | put_ix_float(xp, &xx); |
3418 | 0 | return err; |
3419 | 0 | } |
3420 | | |
3421 | | |
3422 | | |
3423 | | /* external NC_DOUBLE -------------------------------------------------------*/ |
3424 | | |
3425 | | #if X_SIZEOF_DOUBLE == SIZEOF_DOUBLE && !defined(NO_IEEE_FLOAT) |
3426 | | |
3427 | | static void |
3428 | | get_ix_double(const void *xp, double *ip) |
3429 | 0 | { |
3430 | | #ifdef WORDS_BIGENDIAN |
3431 | | (void) memcpy(ip, xp, SIZEOF_DOUBLE); |
3432 | | #else |
3433 | 0 | swap8b(ip, xp); |
3434 | 0 | #endif |
3435 | 0 | } |
3436 | | |
3437 | | static void |
3438 | | put_ix_double(void *xp, const double *ip) |
3439 | 0 | { |
3440 | | #ifdef WORDS_BIGENDIAN |
3441 | | (void) memcpy(xp, ip, X_SIZEOF_DOUBLE); |
3442 | | #else |
3443 | 0 | swap8b(xp, ip); |
3444 | 0 | #endif |
3445 | 0 | } |
3446 | | |
3447 | | #elif defined(vax) && vax != 0 |
3448 | | |
3449 | | /* What IEEE double precision floating point looks like on a Vax */ |
3450 | | struct ieee_double { |
3451 | | unsigned int exp_hi : 7; |
3452 | | unsigned int sign : 1; |
3453 | | unsigned int mant_6 : 4; |
3454 | | unsigned int exp_lo : 4; |
3455 | | unsigned int mant_5 : 8; |
3456 | | unsigned int mant_4 : 8; |
3457 | | |
3458 | | unsigned int mant_lo : 32; |
3459 | | }; |
3460 | | |
3461 | | /* Vax double precision floating point */ |
3462 | | struct vax_double { |
3463 | | unsigned int mantissa1 : 7; |
3464 | | unsigned int exp : 8; |
3465 | | unsigned int sign : 1; |
3466 | | unsigned int mantissa2 : 16; |
3467 | | unsigned int mantissa3 : 16; |
3468 | | unsigned int mantissa4 : 16; |
3469 | | }; |
3470 | | |
3471 | | #define VAX_DBL_BIAS 0x81 |
3472 | | #define IEEE_DBL_BIAS 0x3ff |
3473 | | #define MASK(nbits) ((1 << nbits) - 1) |
3474 | | |
3475 | | static const struct dbl_limits { |
3476 | | struct vax_double d; |
3477 | | struct ieee_double ieee; |
3478 | | } dbl_limits[2] = { |
3479 | | {{ 0x7f, 0xff, 0x0, 0xffff, 0xffff, 0xffff }, /* Max Vax */ |
3480 | | { 0x7f, 0x0, 0x0, 0xf, 0x0, 0x0, 0x0}}, /* Max IEEE */ |
3481 | | {{ 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, /* Min Vax */ |
3482 | | { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}}, /* Min IEEE */ |
3483 | | }; |
3484 | | |
3485 | | |
3486 | | static void |
3487 | | get_ix_double(const void *xp, double *ip) |
3488 | | { |
3489 | | struct vax_double *const vdp = |
3490 | | (struct vax_double *)ip; |
3491 | | const struct ieee_double *const idp = |
3492 | | (const struct ieee_double *) xp; |
3493 | | { |
3494 | | const struct dbl_limits *lim; |
3495 | | int ii; |
3496 | | for (ii = 0, lim = dbl_limits; |
3497 | | ii < sizeof(dbl_limits)/sizeof(struct dbl_limits); |
3498 | | ii++, lim++) |
3499 | | { |
3500 | | if ((idp->mant_lo == lim->ieee.mant_lo) |
3501 | | && (idp->mant_4 == lim->ieee.mant_4) |
3502 | | && (idp->mant_5 == lim->ieee.mant_5) |
3503 | | && (idp->mant_6 == lim->ieee.mant_6) |
3504 | | && (idp->exp_lo == lim->ieee.exp_lo) |
3505 | | && (idp->exp_hi == lim->ieee.exp_hi) |
3506 | | ) |
3507 | | { |
3508 | | *vdp = lim->d; |
3509 | | goto doneit; |
3510 | | } |
3511 | | } |
3512 | | } |
3513 | | { |
3514 | | unsigned exp = idp->exp_hi << 4 | idp->exp_lo; |
3515 | | vdp->exp = exp - IEEE_DBL_BIAS + VAX_DBL_BIAS; |
3516 | | } |
3517 | | { |
3518 | | unsigned mant_hi = ((idp->mant_6 << 16) |
3519 | | | (idp->mant_5 << 8) |
3520 | | | idp->mant_4); |
3521 | | unsigned mant_lo = SWAP4(idp->mant_lo); |
3522 | | vdp->mantissa1 = (mant_hi >> 13); |
3523 | | vdp->mantissa2 = ((mant_hi & MASK(13)) << 3) |
3524 | | | (mant_lo >> 29); |
3525 | | vdp->mantissa3 = (mant_lo >> 13); |
3526 | | vdp->mantissa4 = (mant_lo << 3); |
3527 | | } |
3528 | | doneit: |
3529 | | vdp->sign = idp->sign; |
3530 | | |
3531 | | } |
3532 | | |
3533 | | |
3534 | | static void |
3535 | | put_ix_double(void *xp, const double *ip) |
3536 | | { |
3537 | | const struct vax_double *const vdp = |
3538 | | (const struct vax_double *)ip; |
3539 | | struct ieee_double *const idp = |
3540 | | (struct ieee_double *) xp; |
3541 | | |
3542 | | if ((vdp->mantissa4 > (dbl_limits[0].d.mantissa4 - 3)) && |
3543 | | (vdp->mantissa3 == dbl_limits[0].d.mantissa3) && |
3544 | | (vdp->mantissa2 == dbl_limits[0].d.mantissa2) && |
3545 | | (vdp->mantissa1 == dbl_limits[0].d.mantissa1) && |
3546 | | (vdp->exp == dbl_limits[0].d.exp)) |
3547 | | { |
3548 | | *idp = dbl_limits[0].ieee; |
3549 | | goto shipit; |
3550 | | } |
3551 | | if ((vdp->mantissa4 == dbl_limits[1].d.mantissa4) && |
3552 | | (vdp->mantissa3 == dbl_limits[1].d.mantissa3) && |
3553 | | (vdp->mantissa2 == dbl_limits[1].d.mantissa2) && |
3554 | | (vdp->mantissa1 == dbl_limits[1].d.mantissa1) && |
3555 | | (vdp->exp == dbl_limits[1].d.exp)) |
3556 | | { |
3557 | | *idp = dbl_limits[1].ieee; |
3558 | | goto shipit; |
3559 | | } |
3560 | | |
3561 | | { |
3562 | | unsigned exp = vdp->exp - VAX_DBL_BIAS + IEEE_DBL_BIAS; |
3563 | | |
3564 | | unsigned mant_lo = ((vdp->mantissa2 & MASK(3)) << 29) | |
3565 | | (vdp->mantissa3 << 13) | |
3566 | | ((vdp->mantissa4 >> 3) & MASK(13)); |
3567 | | |
3568 | | unsigned mant_hi = (vdp->mantissa1 << 13) |
3569 | | | (vdp->mantissa2 >> 3); |
3570 | | |
3571 | | if ((vdp->mantissa4 & 7) > 4) |
3572 | | { |
3573 | | /* round up */ |
3574 | | mant_lo++; |
3575 | | if (mant_lo == 0) |
3576 | | { |
3577 | | mant_hi++; |
3578 | | if (mant_hi > 0xffffff) |
3579 | | { |
3580 | | mant_hi = 0; |
3581 | | exp++; |
3582 | | } |
3583 | | } |
3584 | | } |
3585 | | |
3586 | | idp->mant_lo = SWAP4(mant_lo); |
3587 | | idp->mant_6 = mant_hi >> 16; |
3588 | | idp->mant_5 = (mant_hi & 0xff00) >> 8; |
3589 | | idp->mant_4 = mant_hi; |
3590 | | idp->exp_hi = exp >> 4; |
3591 | | idp->exp_lo = exp; |
3592 | | } |
3593 | | |
3594 | | shipit: |
3595 | | idp->sign = vdp->sign; |
3596 | | |
3597 | | } |
3598 | | |
3599 | | /* vax */ |
3600 | | #elif defined(_CRAY) && !defined(__crayx1) |
3601 | | |
3602 | | static void |
3603 | | get_ix_double(const void *xp, double *ip) |
3604 | | { |
3605 | | const ieee_double *idp = (const ieee_double *) xp; |
3606 | | cray_single *csp = (cray_single *) ip; |
3607 | | |
3608 | | if (idp->exp == 0) |
3609 | | { |
3610 | | /* ieee subnormal */ |
3611 | | *ip = (double)idp->mant; |
3612 | | if (idp->mant != 0) |
3613 | | { |
3614 | | csp->exp -= (ieee_double_bias + 51); |
3615 | | } |
3616 | | } |
3617 | | else |
3618 | | { |
3619 | | csp->exp = idp->exp + cs_id_bias + 1; |
3620 | | csp->mant = idp->mant >> (52 - 48 + 1); |
3621 | | csp->mant |= (1 << (48 - 1)); |
3622 | | } |
3623 | | csp->sign = idp->sign; |
3624 | | } |
3625 | | |
3626 | | static void |
3627 | | put_ix_double(void *xp, const double *ip) |
3628 | | { |
3629 | | ieee_double *idp = (ieee_double *) xp; |
3630 | | const cray_single *csp = (const cray_single *) ip; |
3631 | | |
3632 | | int ieee_exp = csp->exp - cs_id_bias -1; |
3633 | | |
3634 | | idp->sign = csp->sign; |
3635 | | |
3636 | | if (ieee_exp >= 0x7ff) |
3637 | | { |
3638 | | /* NC_ERANGE => ieee Inf */ |
3639 | | idp->exp = 0x7ff; |
3640 | | idp->mant = 0x0; |
3641 | | } |
3642 | | else if (ieee_exp > 0) |
3643 | | { |
3644 | | /* normal ieee representation */ |
3645 | | idp->exp = ieee_exp; |
3646 | | /* assumes cray rep is in normal form */ |
3647 | | assert(csp->mant & 0x800000000000); |
3648 | | idp->mant = (((csp->mant << 1) & |
3649 | | 0xffffffffffff) << (52 - 48)); |
3650 | | } |
3651 | | else if (ieee_exp >= (-(52 -48))) |
3652 | | { |
3653 | | /* ieee subnormal, left shift */ |
3654 | | const int lshift = (52 - 48) + ieee_exp; |
3655 | | idp->mant = csp->mant << lshift; |
3656 | | idp->exp = 0; |
3657 | | } |
3658 | | else if (ieee_exp >= -52) |
3659 | | { |
3660 | | /* ieee subnormal, right shift */ |
3661 | | const int rshift = (- (52 - 48) - ieee_exp); |
3662 | | |
3663 | | idp->mant = csp->mant >> rshift; |
3664 | | |
3665 | | #if 0 |
3666 | | if (csp->mant & (1 << (rshift -1))) |
3667 | | { |
3668 | | /* round up */ |
3669 | | idp->mant++; |
3670 | | } |
3671 | | #endif |
3672 | | |
3673 | | idp->exp = 0; |
3674 | | } |
3675 | | else |
3676 | | { |
3677 | | /* smaller than ieee can represent */ |
3678 | | idp->exp = 0; |
3679 | | idp->mant = 0; |
3680 | | } |
3681 | | } |
3682 | | #else |
3683 | | #error "ix_double implementation" |
3684 | | #endif |
3685 | | |
3686 | 0 | #define ix_double double |
3687 | | |
3688 | | static int |
3689 | | ncx_get_double_schar(const void *xp, schar *ip) |
3690 | 0 | { |
3691 | 0 | ix_double xx = 0; |
3692 | 0 | get_ix_double(xp, &xx); |
3693 | 0 | if (xx > (double)SCHAR_MAX || xx < (double)SCHAR_MIN) { |
3694 | 0 | #ifdef ERANGE_FILL |
3695 | 0 | *ip = NC_FILL_BYTE; |
3696 | 0 | #endif |
3697 | 0 | return NC_ERANGE; |
3698 | 0 | } |
3699 | 0 | *ip = (schar)xx; |
3700 | 0 | return NC_NOERR; |
3701 | 0 | } |
3702 | | |
3703 | | static int |
3704 | | ncx_get_double_short(const void *xp, short *ip) |
3705 | 0 | { |
3706 | 0 | ix_double xx = 0; |
3707 | 0 | get_ix_double(xp, &xx); |
3708 | 0 | if (xx > (double)SHORT_MAX || xx < (double)SHORT_MIN) { |
3709 | 0 | #ifdef ERANGE_FILL |
3710 | 0 | *ip = NC_FILL_SHORT; |
3711 | 0 | #endif |
3712 | 0 | return NC_ERANGE; |
3713 | 0 | } |
3714 | 0 | *ip = (short)xx; |
3715 | 0 | return NC_NOERR; |
3716 | 0 | } |
3717 | | |
3718 | | static int |
3719 | | ncx_get_double_int(const void *xp, int *ip) |
3720 | 0 | { |
3721 | 0 | ix_double xx = 0; |
3722 | 0 | get_ix_double(xp, &xx); |
3723 | 0 | if (xx > (double)INT_MAX || xx < (double)INT_MIN) { |
3724 | 0 | #ifdef ERANGE_FILL |
3725 | 0 | *ip = NC_FILL_INT; |
3726 | 0 | #endif |
3727 | 0 | return NC_ERANGE; |
3728 | 0 | } |
3729 | 0 | *ip = (int)xx; |
3730 | 0 | return NC_NOERR; |
3731 | 0 | } |
3732 | | |
3733 | | static int |
3734 | | ncx_get_double_long(const void *xp, long *ip) |
3735 | 0 | { |
3736 | 0 | ix_double xx = 0; |
3737 | 0 | get_ix_double(xp, &xx); |
3738 | 0 | if (xx > (double)LONG_MAX || xx < (double)LONG_MIN) { |
3739 | 0 | #ifdef ERANGE_FILL |
3740 | 0 | *ip = NC_FILL_INT; |
3741 | 0 | #endif |
3742 | 0 | return NC_ERANGE; |
3743 | 0 | } |
3744 | 0 | *ip = (long)xx; |
3745 | 0 | return NC_NOERR; |
3746 | 0 | } |
3747 | | |
3748 | | static int |
3749 | | ncx_get_double_longlong(const void *xp, longlong *ip) |
3750 | 0 | { |
3751 | 0 | ix_double xx = 0; |
3752 | 0 | get_ix_double(xp, &xx); |
3753 | 0 | if (xx == LONGLONG_MAX) *ip = LONGLONG_MAX; |
3754 | 0 | else if (xx == LONGLONG_MIN) *ip = LONGLONG_MIN; |
3755 | 0 | else if (xx > (double)LONGLONG_MAX || xx < (double)LONGLONG_MIN) { |
3756 | 0 | #ifdef ERANGE_FILL |
3757 | 0 | *ip = NC_FILL_INT64; |
3758 | 0 | #endif |
3759 | 0 | return NC_ERANGE; |
3760 | 0 | } |
3761 | 0 | else *ip = (longlong)xx; |
3762 | 0 | return NC_NOERR; |
3763 | 0 | } |
3764 | | |
3765 | | static int |
3766 | | ncx_get_double_uchar(const void *xp, uchar *ip) |
3767 | 0 | { |
3768 | 0 | ix_double xx = 0; |
3769 | 0 | get_ix_double(xp, &xx); |
3770 | 0 | if (xx > (double)UCHAR_MAX || xx < 0) { |
3771 | 0 | #ifdef ERANGE_FILL |
3772 | 0 | *ip = NC_FILL_UBYTE; |
3773 | 0 | #endif |
3774 | 0 | return NC_ERANGE; |
3775 | 0 | } |
3776 | 0 | *ip = (uchar)xx; |
3777 | 0 | return NC_NOERR; |
3778 | 0 | } |
3779 | | |
3780 | | static int |
3781 | | ncx_get_double_ushort(const void *xp, ushort *ip) |
3782 | 0 | { |
3783 | 0 | ix_double xx = 0; |
3784 | 0 | get_ix_double(xp, &xx); |
3785 | 0 | if (xx > (double)USHORT_MAX || xx < 0) { |
3786 | 0 | #ifdef ERANGE_FILL |
3787 | 0 | *ip = NC_FILL_USHORT; |
3788 | 0 | #endif |
3789 | 0 | return NC_ERANGE; |
3790 | 0 | } |
3791 | 0 | *ip = (ushort)xx; |
3792 | 0 | return NC_NOERR; |
3793 | 0 | } |
3794 | | |
3795 | | static int |
3796 | | ncx_get_double_uint(const void *xp, uint *ip) |
3797 | 0 | { |
3798 | 0 | ix_double xx = 0; |
3799 | 0 | get_ix_double(xp, &xx); |
3800 | 0 | if (xx > (double)UINT_MAX || xx < 0) { |
3801 | 0 | #ifdef ERANGE_FILL |
3802 | 0 | *ip = NC_FILL_UINT; |
3803 | 0 | #endif |
3804 | 0 | return NC_ERANGE; |
3805 | 0 | } |
3806 | 0 | *ip = (uint)xx; |
3807 | 0 | return NC_NOERR; |
3808 | 0 | } |
3809 | | |
3810 | | static int |
3811 | | ncx_get_double_ulonglong(const void *xp, ulonglong *ip) |
3812 | 0 | { |
3813 | 0 | ix_double xx = 0; |
3814 | 0 | get_ix_double(xp, &xx); |
3815 | 0 | if (xx == ULONGLONG_MAX) *ip = ULONGLONG_MAX; |
3816 | 0 | else if (xx > (double)ULONGLONG_MAX || xx < 0) { |
3817 | 0 | #ifdef ERANGE_FILL |
3818 | 0 | *ip = NC_FILL_UINT64; |
3819 | 0 | #endif |
3820 | 0 | return NC_ERANGE; |
3821 | 0 | } |
3822 | 0 | else *ip = (ulonglong)xx; |
3823 | 0 | return NC_NOERR; |
3824 | 0 | } |
3825 | | |
3826 | | |
3827 | | static int |
3828 | | ncx_get_double_float(const void *xp, float *ip) |
3829 | 0 | { |
3830 | 0 | double xx = 0.0; |
3831 | 0 | get_ix_double(xp, &xx); |
3832 | 0 | if (xx > FLT_MAX) { |
3833 | 0 | #ifdef ERANGE_FILL |
3834 | 0 | *ip = NC_FILL_FLOAT; |
3835 | | #else |
3836 | | *ip = FLT_MAX; |
3837 | | #endif |
3838 | 0 | return NC_ERANGE; |
3839 | 0 | } |
3840 | 0 | if (xx < (-FLT_MAX)) { |
3841 | 0 | #ifdef ERANGE_FILL |
3842 | 0 | *ip = NC_FILL_FLOAT; |
3843 | | #else |
3844 | | *ip = (-FLT_MAX); |
3845 | | #endif |
3846 | 0 | return NC_ERANGE; |
3847 | 0 | } |
3848 | 0 | *ip = (float) xx; |
3849 | 0 | return NC_NOERR; |
3850 | 0 | } |
3851 | | |
3852 | | #if X_SIZEOF_DOUBLE != SIZEOF_DOUBLE || defined(NO_IEEE_FLOAT) |
3853 | | static int |
3854 | | ncx_get_double_double(const void *xp, double *ip, void *fillp) |
3855 | | { |
3856 | | /* TODO */ |
3857 | | get_ix_double(xp, ip); |
3858 | | return NC_NOERR; |
3859 | | } |
3860 | | #endif |
3861 | | |
3862 | | static int |
3863 | | ncx_put_double_schar(void *xp, const schar *ip, void *fillp) |
3864 | 0 | { |
3865 | 0 | int err=NC_NOERR; |
3866 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3867 | | |
3868 | | |
3869 | 0 | xx = (ix_double)*ip; |
3870 | |
|
3871 | 0 | put_ix_double(xp, &xx); |
3872 | 0 | return err; |
3873 | 0 | } |
3874 | | |
3875 | | static int |
3876 | | ncx_put_double_uchar(void *xp, const uchar *ip, void *fillp) |
3877 | 0 | { |
3878 | 0 | int err=NC_NOERR; |
3879 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3880 | | |
3881 | | |
3882 | 0 | xx = (ix_double)*ip; |
3883 | |
|
3884 | 0 | put_ix_double(xp, &xx); |
3885 | 0 | return err; |
3886 | 0 | } |
3887 | | |
3888 | | static int |
3889 | | ncx_put_double_short(void *xp, const short *ip, void *fillp) |
3890 | 0 | { |
3891 | 0 | int err=NC_NOERR; |
3892 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3893 | | |
3894 | | |
3895 | 0 | xx = (ix_double)*ip; |
3896 | |
|
3897 | 0 | put_ix_double(xp, &xx); |
3898 | 0 | return err; |
3899 | 0 | } |
3900 | | |
3901 | | static int |
3902 | | ncx_put_double_ushort(void *xp, const ushort *ip, void *fillp) |
3903 | 0 | { |
3904 | 0 | int err=NC_NOERR; |
3905 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3906 | | |
3907 | | |
3908 | 0 | xx = (ix_double)*ip; |
3909 | |
|
3910 | 0 | put_ix_double(xp, &xx); |
3911 | 0 | return err; |
3912 | 0 | } |
3913 | | |
3914 | | static int |
3915 | | ncx_put_double_int(void *xp, const int *ip, void *fillp) |
3916 | 0 | { |
3917 | 0 | int err=NC_NOERR; |
3918 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3919 | | |
3920 | | |
3921 | 0 | xx = (ix_double)*ip; |
3922 | |
|
3923 | 0 | put_ix_double(xp, &xx); |
3924 | 0 | return err; |
3925 | 0 | } |
3926 | | |
3927 | | static int |
3928 | | ncx_put_double_long(void *xp, const long *ip, void *fillp) |
3929 | 0 | { |
3930 | 0 | int err=NC_NOERR; |
3931 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3932 | | |
3933 | | |
3934 | 0 | xx = (ix_double)*ip; |
3935 | |
|
3936 | 0 | put_ix_double(xp, &xx); |
3937 | 0 | return err; |
3938 | 0 | } |
3939 | | |
3940 | | static int |
3941 | | ncx_put_double_uint(void *xp, const uint *ip, void *fillp) |
3942 | 0 | { |
3943 | 0 | int err=NC_NOERR; |
3944 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3945 | | |
3946 | | |
3947 | 0 | xx = (ix_double)*ip; |
3948 | |
|
3949 | 0 | put_ix_double(xp, &xx); |
3950 | 0 | return err; |
3951 | 0 | } |
3952 | | |
3953 | | static int |
3954 | | ncx_put_double_longlong(void *xp, const longlong *ip, void *fillp) |
3955 | 0 | { |
3956 | 0 | int err=NC_NOERR; |
3957 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3958 | | |
3959 | | |
3960 | 0 | xx = (ix_double)*ip; |
3961 | |
|
3962 | 0 | put_ix_double(xp, &xx); |
3963 | 0 | return err; |
3964 | 0 | } |
3965 | | |
3966 | | static int |
3967 | | ncx_put_double_ulonglong(void *xp, const ulonglong *ip, void *fillp) |
3968 | 0 | { |
3969 | 0 | int err=NC_NOERR; |
3970 | 0 | ix_double xx = NC_FILL_DOUBLE; |
3971 | | |
3972 | | |
3973 | 0 | xx = (ix_double)*ip; |
3974 | |
|
3975 | 0 | put_ix_double(xp, &xx); |
3976 | 0 | return err; |
3977 | 0 | } |
3978 | | |
3979 | | |
3980 | | static int |
3981 | | ncx_put_double_float(void *xp, const float *ip, void *fillp) |
3982 | 0 | { |
3983 | 0 | int err=NC_NOERR; |
3984 | 0 | double xx = NC_FILL_DOUBLE; |
3985 | 0 | #if 1 /* TODO: figure this out (if condition below will never be true)*/ |
3986 | 0 | if ((double)(*ip) > X_DOUBLE_MAX || (double)(*ip) < X_DOUBLE_MIN) { |
3987 | | |
3988 | 0 | #ifdef ERANGE_FILL |
3989 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
3990 | 0 | #endif |
3991 | 0 | err = NC_ERANGE; |
3992 | 0 | } |
3993 | 0 | #ifdef ERANGE_FILL |
3994 | 0 | else |
3995 | 0 | #endif |
3996 | 0 | #endif |
3997 | 0 | xx = (double) *ip; |
3998 | |
|
3999 | 0 | put_ix_double(xp, &xx); |
4000 | 0 | return err; |
4001 | 0 | } |
4002 | | |
4003 | | #if X_SIZEOF_DOUBLE != SIZEOF_DOUBLE || defined(NO_IEEE_FLOAT) |
4004 | | static int |
4005 | | ncx_put_double_double(void *xp, const double *ip, void *fillp) |
4006 | | { |
4007 | | int err=NC_NOERR; |
4008 | | double *_ip = ip; |
4009 | | #ifdef NO_IEEE_FLOAT |
4010 | | #ifdef ERANGE_FILL |
4011 | | double tmp=NC_FILL_DOUBLE; |
4012 | | #endif |
4013 | | if (*ip > X_DOUBLE_MAX || *ip < X_DOUBLE_MIN) { |
4014 | | |
4015 | | #ifdef ERANGE_FILL |
4016 | | if (fillp != NULL) memcpy(&tmp, fillp, 8); |
4017 | | #endif |
4018 | | #ifdef ERANGE_FILL |
4019 | | _ip = &tmp; |
4020 | | #endif |
4021 | | err = NC_ERANGE; |
4022 | | } |
4023 | | #endif |
4024 | | put_ix_double(xp, _ip); |
4025 | | return err; |
4026 | | } |
4027 | | #endif |
4028 | | |
4029 | | |
4030 | | /* external NC_INT64 --------------------------------------------------------*/ |
4031 | | |
4032 | | #if SHORT_MAX == X_INT64_MAX |
4033 | | typedef short ix_int64; |
4034 | | #define SIZEOF_IX_INT64 SIZEOF_SHORT |
4035 | | #define IX_INT64_MAX SHORT_MAX |
4036 | | #elif LONG_LONG_MAX >= X_INT64_MAX |
4037 | | typedef longlong ix_int64; |
4038 | | #define SIZEOF_IX_INT64 SIZEOF_LONGLONG |
4039 | 0 | #define IX_INT64_MAX LONG_LONG_MAX |
4040 | | #elif LONG_MAX >= X_INT64_MAX |
4041 | | typedef long ix_int64; |
4042 | | #define SIZEOF_IX_INT64 SIZEOF_LONG |
4043 | | #define IX_INT64_MAX LONG_MAX |
4044 | | #else |
4045 | | #error "ix_int64 implementation" |
4046 | | #endif |
4047 | | |
4048 | | |
4049 | | static void |
4050 | | get_ix_int64(const void *xp, ix_int64 *ip) |
4051 | 0 | { |
4052 | 0 | const uchar *cp = (const uchar *) xp; |
4053 | |
|
4054 | 0 | *ip = (ix_int64)((uint64_t)(*cp++) << 56); |
4055 | 0 | *ip |= (ix_int64)((uint64_t)(*cp++) << 48); |
4056 | 0 | *ip |= (ix_int64)((uint64_t)(*cp++) << 40); |
4057 | 0 | *ip |= (ix_int64)((uint64_t)(*cp++) << 32); |
4058 | 0 | *ip |= (ix_int64)((uint64_t)(*cp++) << 24); |
4059 | 0 | *ip |= (ix_int64)((uint64_t)(*cp++) << 16); |
4060 | 0 | *ip |= (ix_int64)((uint64_t)(*cp++) << 8); |
4061 | 0 | *ip |= (ix_int64)*cp; |
4062 | 0 | } |
4063 | | |
4064 | | static void |
4065 | | put_ix_int64(void *xp, const ix_int64 *ip) |
4066 | 0 | { |
4067 | 0 | uchar *cp = (uchar *) xp; |
4068 | |
|
4069 | 0 | *cp++ = (uchar)((*ip) >> 56); |
4070 | 0 | *cp++ = (uchar)(((*ip) & 0x00ff000000000000LL) >> 48); |
4071 | 0 | *cp++ = (uchar)(((*ip) & 0x0000ff0000000000LL) >> 40); |
4072 | 0 | *cp++ = (uchar)(((*ip) & 0x000000ff00000000LL) >> 32); |
4073 | 0 | *cp++ = (uchar)(((*ip) & 0x00000000ff000000LL) >> 24); |
4074 | 0 | *cp++ = (uchar)(((*ip) & 0x0000000000ff0000LL) >> 16); |
4075 | 0 | *cp++ = (uchar)(((*ip) & 0x000000000000ff00LL) >> 8); |
4076 | 0 | *cp = (uchar)( (*ip) & 0x00000000000000ffLL); |
4077 | 0 | } |
4078 | | |
4079 | | #if X_SIZEOF_INT64 != SIZEOF_LONGLONG |
4080 | | static int |
4081 | | ncx_get_longlong_longlong(const void *xp, longlong *ip) |
4082 | | { |
4083 | | int err=NC_NOERR; |
4084 | | #if SIZEOF_IX_INT64 == SIZEOF_LONGLONG && IX_INT64_MAX == LONGLONG_MAX |
4085 | | get_ix_int64(xp, (ix_int64 *)ip); |
4086 | | #else |
4087 | | ix_int64 xx = 0; |
4088 | | get_ix_int64(xp, &xx); |
4089 | | |
4090 | | #if IX_INT64_MAX > LONGLONG_MAX |
4091 | | if (xx > LONGLONG_MAX || xx < LONGLONG_MIN) { |
4092 | | #ifdef ERANGE_FILL |
4093 | | *ip = NC_FILL_INT64; |
4094 | | return NC_ERANGE; |
4095 | | #else |
4096 | | err = NC_ERANGE; |
4097 | | #endif |
4098 | | } |
4099 | | #endif |
4100 | | |
4101 | | |
4102 | | *ip = (longlong) xx; |
4103 | | #endif |
4104 | | return err; |
4105 | | } |
4106 | | |
4107 | | #endif |
4108 | | static int |
4109 | | ncx_get_longlong_schar(const void *xp, schar *ip) |
4110 | 0 | { |
4111 | 0 | int err=NC_NOERR; |
4112 | 0 | ix_int64 xx = 0; |
4113 | 0 | get_ix_int64(xp, &xx); |
4114 | |
|
4115 | 0 | #if IX_INT64_MAX > SCHAR_MAX |
4116 | 0 | if (xx > SCHAR_MAX || xx < SCHAR_MIN) { |
4117 | 0 | #ifdef ERANGE_FILL |
4118 | 0 | *ip = NC_FILL_BYTE; |
4119 | 0 | return NC_ERANGE; |
4120 | | #else |
4121 | | err = NC_ERANGE; |
4122 | | #endif |
4123 | 0 | } |
4124 | 0 | #endif |
4125 | | |
4126 | | |
4127 | 0 | *ip = (schar) xx; |
4128 | 0 | return err; |
4129 | 0 | } |
4130 | | |
4131 | | static int |
4132 | | ncx_get_longlong_short(const void *xp, short *ip) |
4133 | 0 | { |
4134 | 0 | int err=NC_NOERR; |
4135 | | #if SIZEOF_IX_INT64 == SIZEOF_SHORT && IX_INT64_MAX == SHORT_MAX |
4136 | | get_ix_int64(xp, (ix_int64 *)ip); |
4137 | | #else |
4138 | 0 | ix_int64 xx = 0; |
4139 | 0 | get_ix_int64(xp, &xx); |
4140 | |
|
4141 | 0 | #if IX_INT64_MAX > SHORT_MAX |
4142 | 0 | if (xx > SHORT_MAX || xx < SHORT_MIN) { |
4143 | 0 | #ifdef ERANGE_FILL |
4144 | 0 | *ip = NC_FILL_SHORT; |
4145 | 0 | return NC_ERANGE; |
4146 | | #else |
4147 | | err = NC_ERANGE; |
4148 | | #endif |
4149 | 0 | } |
4150 | 0 | #endif |
4151 | | |
4152 | | |
4153 | 0 | *ip = (short) xx; |
4154 | 0 | #endif |
4155 | 0 | return err; |
4156 | 0 | } |
4157 | | |
4158 | | static int |
4159 | | ncx_get_longlong_int(const void *xp, int *ip) |
4160 | 0 | { |
4161 | 0 | int err=NC_NOERR; |
4162 | | #if SIZEOF_IX_INT64 == SIZEOF_INT && IX_INT64_MAX == INT_MAX |
4163 | | get_ix_int64(xp, (ix_int64 *)ip); |
4164 | | #else |
4165 | 0 | ix_int64 xx = 0; |
4166 | 0 | get_ix_int64(xp, &xx); |
4167 | |
|
4168 | 0 | #if IX_INT64_MAX > INT_MAX |
4169 | 0 | if (xx > INT_MAX || xx < INT_MIN) { |
4170 | 0 | #ifdef ERANGE_FILL |
4171 | 0 | *ip = NC_FILL_INT; |
4172 | 0 | return NC_ERANGE; |
4173 | | #else |
4174 | | err = NC_ERANGE; |
4175 | | #endif |
4176 | 0 | } |
4177 | 0 | #endif |
4178 | | |
4179 | | |
4180 | 0 | *ip = (int) xx; |
4181 | 0 | #endif |
4182 | 0 | return err; |
4183 | 0 | } |
4184 | | |
4185 | | static int |
4186 | | ncx_get_longlong_long(const void *xp, long *ip) |
4187 | 0 | { |
4188 | 0 | int err=NC_NOERR; |
4189 | 0 | #if SIZEOF_IX_INT64 == SIZEOF_LONG && IX_INT64_MAX == LONG_MAX |
4190 | 0 | get_ix_int64(xp, (ix_int64 *)ip); |
4191 | | #else |
4192 | | ix_int64 xx = 0; |
4193 | | get_ix_int64(xp, &xx); |
4194 | | |
4195 | | #if IX_INT64_MAX > LONG_MAX |
4196 | | if (xx > LONG_MAX || xx < LONG_MIN) { |
4197 | | #ifdef ERANGE_FILL |
4198 | | *ip = NC_FILL_INT; |
4199 | | return NC_ERANGE; |
4200 | | #else |
4201 | | err = NC_ERANGE; |
4202 | | #endif |
4203 | | } |
4204 | | #endif |
4205 | | |
4206 | | |
4207 | | *ip = (long) xx; |
4208 | | #endif |
4209 | 0 | return err; |
4210 | 0 | } |
4211 | | |
4212 | | static int |
4213 | | ncx_get_longlong_ushort(const void *xp, ushort *ip) |
4214 | 0 | { |
4215 | 0 | int err=NC_NOERR; |
4216 | 0 | ix_int64 xx = 0; |
4217 | 0 | get_ix_int64(xp, &xx); |
4218 | |
|
4219 | 0 | #if IX_INT64_MAX > USHORT_MAX |
4220 | 0 | if (xx > USHORT_MAX) { |
4221 | 0 | #ifdef ERANGE_FILL |
4222 | 0 | *ip = NC_FILL_USHORT; |
4223 | 0 | return NC_ERANGE; |
4224 | | #else |
4225 | | err = NC_ERANGE; |
4226 | | #endif |
4227 | 0 | } |
4228 | 0 | #endif |
4229 | | |
4230 | 0 | if (xx < 0) { |
4231 | 0 | #ifdef ERANGE_FILL |
4232 | 0 | *ip = NC_FILL_USHORT; |
4233 | 0 | return NC_ERANGE; |
4234 | | #else |
4235 | | err = NC_ERANGE; /* because ip is unsigned */ |
4236 | | #endif |
4237 | 0 | } |
4238 | 0 | *ip = (ushort) xx; |
4239 | 0 | return err; |
4240 | 0 | } |
4241 | | |
4242 | | static int |
4243 | | ncx_get_longlong_uchar(const void *xp, uchar *ip) |
4244 | 0 | { |
4245 | 0 | int err=NC_NOERR; |
4246 | 0 | ix_int64 xx = 0; |
4247 | 0 | get_ix_int64(xp, &xx); |
4248 | |
|
4249 | 0 | #if IX_INT64_MAX > UCHAR_MAX |
4250 | 0 | if (xx > UCHAR_MAX) { |
4251 | 0 | #ifdef ERANGE_FILL |
4252 | 0 | *ip = NC_FILL_UBYTE; |
4253 | 0 | return NC_ERANGE; |
4254 | | #else |
4255 | | err = NC_ERANGE; |
4256 | | #endif |
4257 | 0 | } |
4258 | 0 | #endif |
4259 | | |
4260 | 0 | if (xx < 0) { |
4261 | 0 | #ifdef ERANGE_FILL |
4262 | 0 | *ip = NC_FILL_UBYTE; |
4263 | 0 | return NC_ERANGE; |
4264 | | #else |
4265 | | err = NC_ERANGE; /* because ip is unsigned */ |
4266 | | #endif |
4267 | 0 | } |
4268 | 0 | *ip = (uchar) xx; |
4269 | 0 | return err; |
4270 | 0 | } |
4271 | | |
4272 | | static int |
4273 | | ncx_get_longlong_uint(const void *xp, uint *ip) |
4274 | 0 | { |
4275 | 0 | int err=NC_NOERR; |
4276 | 0 | ix_int64 xx = 0; |
4277 | 0 | get_ix_int64(xp, &xx); |
4278 | |
|
4279 | 0 | #if IX_INT64_MAX > UINT_MAX |
4280 | 0 | if (xx > UINT_MAX) { |
4281 | 0 | #ifdef ERANGE_FILL |
4282 | 0 | *ip = NC_FILL_UINT; |
4283 | 0 | return NC_ERANGE; |
4284 | | #else |
4285 | | err = NC_ERANGE; |
4286 | | #endif |
4287 | 0 | } |
4288 | 0 | #endif |
4289 | | |
4290 | 0 | if (xx < 0) { |
4291 | 0 | #ifdef ERANGE_FILL |
4292 | 0 | *ip = NC_FILL_UINT; |
4293 | 0 | return NC_ERANGE; |
4294 | | #else |
4295 | | err = NC_ERANGE; /* because ip is unsigned */ |
4296 | | #endif |
4297 | 0 | } |
4298 | 0 | *ip = (uint) xx; |
4299 | 0 | return err; |
4300 | 0 | } |
4301 | | |
4302 | | static int |
4303 | | ncx_get_longlong_ulonglong(const void *xp, ulonglong *ip) |
4304 | 0 | { |
4305 | 0 | int err=NC_NOERR; |
4306 | 0 | ix_int64 xx = 0; |
4307 | 0 | get_ix_int64(xp, &xx); |
4308 | |
|
4309 | | #if IX_INT64_MAX > ULONGLONG_MAX |
4310 | | if (xx > ULONGLONG_MAX) { |
4311 | | #ifdef ERANGE_FILL |
4312 | | *ip = NC_FILL_UINT64; |
4313 | | return NC_ERANGE; |
4314 | | #else |
4315 | | err = NC_ERANGE; |
4316 | | #endif |
4317 | | } |
4318 | | #endif |
4319 | |
|
4320 | 0 | if (xx < 0) { |
4321 | 0 | #ifdef ERANGE_FILL |
4322 | 0 | *ip = NC_FILL_UINT64; |
4323 | 0 | return NC_ERANGE; |
4324 | | #else |
4325 | | err = NC_ERANGE; /* because ip is unsigned */ |
4326 | | #endif |
4327 | 0 | } |
4328 | 0 | *ip = (ulonglong) xx; |
4329 | 0 | return err; |
4330 | 0 | } |
4331 | | |
4332 | | static int |
4333 | | ncx_get_longlong_float(const void *xp, float *ip) |
4334 | 0 | { |
4335 | 0 | ix_int64 xx = 0; |
4336 | 0 | get_ix_int64(xp, &xx); |
4337 | 0 | *ip = (float)xx; |
4338 | 0 | return NC_NOERR; |
4339 | 0 | } |
4340 | | |
4341 | | static int |
4342 | | ncx_get_longlong_double(const void *xp, double *ip) |
4343 | 0 | { |
4344 | 0 | ix_int64 xx = 0; |
4345 | 0 | get_ix_int64(xp, &xx); |
4346 | 0 | *ip = (double)xx; |
4347 | 0 | return NC_NOERR; |
4348 | 0 | } |
4349 | | |
4350 | | |
4351 | | #if X_SIZEOF_INT64 != SIZEOF_LONGLONG |
4352 | | static int |
4353 | | ncx_put_longlong_longlong(void *xp, const longlong *ip, void *fillp) |
4354 | | { |
4355 | | int err=NC_NOERR; |
4356 | | #if SIZEOF_IX_INT64 == SIZEOF_LONGLONG && IX_INT64_MAX == LONGLONG_MAX |
4357 | | put_ix_int64(xp, (const ix_int64 *)ip); |
4358 | | #else |
4359 | | ix_int64 xx = NC_FILL_INT64; |
4360 | | |
4361 | | #if IX_INT64_MAX < LONGLONG_MAX |
4362 | | if (*ip > IX_INT64_MAX || *ip < X_INT64_MIN) { |
4363 | | |
4364 | | #ifdef ERANGE_FILL |
4365 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4366 | | #endif |
4367 | | err = NC_ERANGE; |
4368 | | } |
4369 | | #ifdef ERANGE_FILL |
4370 | | else |
4371 | | #endif |
4372 | | #endif |
4373 | | xx = (ix_int64)*ip; |
4374 | | |
4375 | | put_ix_int64(xp, &xx); |
4376 | | #endif |
4377 | | return err; |
4378 | | } |
4379 | | |
4380 | | #endif |
4381 | | static int |
4382 | | ncx_put_longlong_schar(void *xp, const schar *ip, void *fillp) |
4383 | 0 | { |
4384 | 0 | int err=NC_NOERR; |
4385 | 0 | ix_int64 xx = NC_FILL_INT64; |
4386 | |
|
4387 | | #if IX_INT64_MAX < SCHAR_MAX |
4388 | | if (*ip > IX_INT64_MAX || *ip < X_INT64_MIN) { |
4389 | | |
4390 | | #ifdef ERANGE_FILL |
4391 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4392 | | #endif |
4393 | | err = NC_ERANGE; |
4394 | | } |
4395 | | #ifdef ERANGE_FILL |
4396 | | else |
4397 | | #endif |
4398 | | #endif |
4399 | 0 | xx = (ix_int64)*ip; |
4400 | |
|
4401 | 0 | put_ix_int64(xp, &xx); |
4402 | 0 | return err; |
4403 | 0 | } |
4404 | | |
4405 | | static int |
4406 | | ncx_put_longlong_short(void *xp, const short *ip, void *fillp) |
4407 | 0 | { |
4408 | 0 | int err=NC_NOERR; |
4409 | | #if SIZEOF_IX_INT64 == SIZEOF_SHORT && IX_INT64_MAX == SHORT_MAX |
4410 | | put_ix_int64(xp, (const ix_int64 *)ip); |
4411 | | #else |
4412 | 0 | ix_int64 xx = NC_FILL_INT64; |
4413 | |
|
4414 | | #if IX_INT64_MAX < SHORT_MAX |
4415 | | if (*ip > IX_INT64_MAX || *ip < X_INT64_MIN) { |
4416 | | |
4417 | | #ifdef ERANGE_FILL |
4418 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4419 | | #endif |
4420 | | err = NC_ERANGE; |
4421 | | } |
4422 | | #ifdef ERANGE_FILL |
4423 | | else |
4424 | | #endif |
4425 | | #endif |
4426 | 0 | xx = (ix_int64)*ip; |
4427 | |
|
4428 | 0 | put_ix_int64(xp, &xx); |
4429 | 0 | #endif |
4430 | 0 | return err; |
4431 | 0 | } |
4432 | | |
4433 | | static int |
4434 | | ncx_put_longlong_int(void *xp, const int *ip, void *fillp) |
4435 | 0 | { |
4436 | 0 | int err=NC_NOERR; |
4437 | | #if SIZEOF_IX_INT64 == SIZEOF_INT && IX_INT64_MAX == INT_MAX |
4438 | | put_ix_int64(xp, (const ix_int64 *)ip); |
4439 | | #else |
4440 | 0 | ix_int64 xx = NC_FILL_INT64; |
4441 | |
|
4442 | | #if IX_INT64_MAX < INT_MAX |
4443 | | if (*ip > IX_INT64_MAX || *ip < X_INT64_MIN) { |
4444 | | |
4445 | | #ifdef ERANGE_FILL |
4446 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4447 | | #endif |
4448 | | err = NC_ERANGE; |
4449 | | } |
4450 | | #ifdef ERANGE_FILL |
4451 | | else |
4452 | | #endif |
4453 | | #endif |
4454 | 0 | xx = (ix_int64)*ip; |
4455 | |
|
4456 | 0 | put_ix_int64(xp, &xx); |
4457 | 0 | #endif |
4458 | 0 | return err; |
4459 | 0 | } |
4460 | | |
4461 | | static int |
4462 | | ncx_put_longlong_long(void *xp, const long *ip, void *fillp) |
4463 | 0 | { |
4464 | 0 | int err=NC_NOERR; |
4465 | 0 | #if SIZEOF_IX_INT64 == SIZEOF_LONG && IX_INT64_MAX == LONG_MAX |
4466 | 0 | put_ix_int64(xp, (const ix_int64 *)ip); |
4467 | | #else |
4468 | | ix_int64 xx = NC_FILL_INT64; |
4469 | | |
4470 | | #if IX_INT64_MAX < LONG_MAX |
4471 | | if (*ip > IX_INT64_MAX || *ip < X_INT64_MIN) { |
4472 | | |
4473 | | #ifdef ERANGE_FILL |
4474 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4475 | | #endif |
4476 | | err = NC_ERANGE; |
4477 | | } |
4478 | | #ifdef ERANGE_FILL |
4479 | | else |
4480 | | #endif |
4481 | | #endif |
4482 | | xx = (ix_int64)*ip; |
4483 | | |
4484 | | put_ix_int64(xp, &xx); |
4485 | | #endif |
4486 | 0 | return err; |
4487 | 0 | } |
4488 | | |
4489 | | static int |
4490 | | ncx_put_longlong_ushort(void *xp, const ushort *ip, void *fillp) |
4491 | 0 | { |
4492 | 0 | int err=NC_NOERR; |
4493 | 0 | ix_int64 xx = NC_FILL_INT64; |
4494 | |
|
4495 | | #if IX_INT64_MAX < USHORT_MAX |
4496 | | if (*ip > IX_INT64_MAX) { |
4497 | | |
4498 | | #ifdef ERANGE_FILL |
4499 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4500 | | #endif |
4501 | | err = NC_ERANGE; |
4502 | | } |
4503 | | #ifdef ERANGE_FILL |
4504 | | else |
4505 | | #endif |
4506 | | #endif |
4507 | 0 | xx = (ix_int64)*ip; |
4508 | |
|
4509 | 0 | put_ix_int64(xp, &xx); |
4510 | 0 | return err; |
4511 | 0 | } |
4512 | | |
4513 | | static int |
4514 | | ncx_put_longlong_uchar(void *xp, const uchar *ip, void *fillp) |
4515 | 0 | { |
4516 | 0 | int err=NC_NOERR; |
4517 | 0 | ix_int64 xx = NC_FILL_INT64; |
4518 | |
|
4519 | | #if IX_INT64_MAX < UCHAR_MAX |
4520 | | if (*ip > IX_INT64_MAX) { |
4521 | | |
4522 | | #ifdef ERANGE_FILL |
4523 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4524 | | #endif |
4525 | | err = NC_ERANGE; |
4526 | | } |
4527 | | #ifdef ERANGE_FILL |
4528 | | else |
4529 | | #endif |
4530 | | #endif |
4531 | 0 | xx = (ix_int64)*ip; |
4532 | |
|
4533 | 0 | put_ix_int64(xp, &xx); |
4534 | 0 | return err; |
4535 | 0 | } |
4536 | | |
4537 | | static int |
4538 | | ncx_put_longlong_uint(void *xp, const uint *ip, void *fillp) |
4539 | 0 | { |
4540 | 0 | int err=NC_NOERR; |
4541 | 0 | ix_int64 xx = NC_FILL_INT64; |
4542 | |
|
4543 | | #if IX_INT64_MAX < UINT_MAX |
4544 | | if (*ip > IX_INT64_MAX) { |
4545 | | |
4546 | | #ifdef ERANGE_FILL |
4547 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4548 | | #endif |
4549 | | err = NC_ERANGE; |
4550 | | } |
4551 | | #ifdef ERANGE_FILL |
4552 | | else |
4553 | | #endif |
4554 | | #endif |
4555 | 0 | xx = (ix_int64)*ip; |
4556 | |
|
4557 | 0 | put_ix_int64(xp, &xx); |
4558 | 0 | return err; |
4559 | 0 | } |
4560 | | |
4561 | | static int |
4562 | | ncx_put_longlong_ulonglong(void *xp, const ulonglong *ip, void *fillp) |
4563 | 0 | { |
4564 | 0 | int err=NC_NOERR; |
4565 | 0 | ix_int64 xx = NC_FILL_INT64; |
4566 | |
|
4567 | 0 | #if IX_INT64_MAX < ULONGLONG_MAX |
4568 | 0 | if (*ip > IX_INT64_MAX) { |
4569 | | |
4570 | 0 | #ifdef ERANGE_FILL |
4571 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4572 | 0 | #endif |
4573 | 0 | err = NC_ERANGE; |
4574 | 0 | } |
4575 | 0 | #ifdef ERANGE_FILL |
4576 | 0 | else |
4577 | 0 | #endif |
4578 | 0 | #endif |
4579 | 0 | xx = (ix_int64)*ip; |
4580 | |
|
4581 | 0 | put_ix_int64(xp, &xx); |
4582 | 0 | return err; |
4583 | 0 | } |
4584 | | |
4585 | | static int |
4586 | | ncx_put_longlong_float(void *xp, const float *ip, void *fillp) |
4587 | 0 | { |
4588 | 0 | int err=NC_NOERR; |
4589 | 0 | ix_int64 xx = NC_FILL_INT64; |
4590 | |
|
4591 | 0 | if (*ip > (double)X_INT64_MAX || *ip < (double)X_INT64_MIN) { |
4592 | | |
4593 | 0 | #ifdef ERANGE_FILL |
4594 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4595 | 0 | #endif |
4596 | 0 | err = NC_ERANGE; |
4597 | 0 | } |
4598 | 0 | #ifdef ERANGE_FILL |
4599 | 0 | else |
4600 | 0 | #endif |
4601 | 0 | xx = (ix_int64)*ip; |
4602 | |
|
4603 | 0 | put_ix_int64(xp, &xx); |
4604 | 0 | return err; |
4605 | 0 | } |
4606 | | |
4607 | | static int |
4608 | | ncx_put_longlong_double(void *xp, const double *ip, void *fillp) |
4609 | 0 | { |
4610 | 0 | int err=NC_NOERR; |
4611 | 0 | ix_int64 xx = NC_FILL_INT64; |
4612 | |
|
4613 | 0 | if (*ip > X_INT64_MAX || *ip < X_INT64_MIN) { |
4614 | | |
4615 | 0 | #ifdef ERANGE_FILL |
4616 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4617 | 0 | #endif |
4618 | 0 | err = NC_ERANGE; |
4619 | 0 | } |
4620 | 0 | #ifdef ERANGE_FILL |
4621 | 0 | else |
4622 | 0 | #endif |
4623 | 0 | xx = (ix_int64)*ip; |
4624 | |
|
4625 | 0 | put_ix_int64(xp, &xx); |
4626 | 0 | return err; |
4627 | 0 | } |
4628 | | |
4629 | | |
4630 | | |
4631 | | /* external NC_UINT64 -------------------------------------------------------*/ |
4632 | | |
4633 | | #if USHORT_MAX == X_UINT64_MAX |
4634 | | typedef ushort ix_uint64; |
4635 | | #define SIZEOF_IX_UINT64 SIZEOF_USHORT |
4636 | | #define IX_UINT64_MAX USHORT_MAX |
4637 | | #elif ULONG_LONG_MAX >= X_UINT64_MAX |
4638 | | typedef ulonglong ix_uint64; |
4639 | | #define SIZEOF_IX_UINT64 SIZEOF_ULONGLONG |
4640 | | #define IX_UINT64_MAX ULONG_LONG_MAX |
4641 | | #elif ULONG_MAX >= X_UINT64_MAX |
4642 | | typedef ulong ix_uint64; |
4643 | | #define SIZEOF_IX_UINT64 SIZEOF_ULONG |
4644 | | #define IX_UINT64_MAX ULONG_MAX |
4645 | | #else |
4646 | | #error "ix_uint64 implementation" |
4647 | | #endif |
4648 | | |
4649 | | |
4650 | | static void |
4651 | | get_ix_uint64(const void *xp, ix_uint64 *ip) |
4652 | 0 | { |
4653 | 0 | const uchar *cp = (const uchar *) xp; |
4654 | |
|
4655 | 0 | *ip = ((ix_uint64)(*cp++) << 56); |
4656 | 0 | *ip |= ((ix_uint64)(*cp++) << 48); |
4657 | 0 | *ip |= ((ix_uint64)(*cp++) << 40); |
4658 | 0 | *ip |= ((ix_uint64)(*cp++) << 32); |
4659 | 0 | *ip |= ((ix_uint64)(*cp++) << 24); |
4660 | 0 | *ip |= ((ix_uint64)(*cp++) << 16); |
4661 | 0 | *ip |= ((ix_uint64)(*cp++) << 8); |
4662 | 0 | *ip |= (ix_uint64)*cp; |
4663 | 0 | } |
4664 | | |
4665 | | static void |
4666 | | put_ix_uint64(void *xp, const ix_uint64 *ip) |
4667 | 0 | { |
4668 | 0 | uchar *cp = (uchar *) xp; |
4669 | |
|
4670 | 0 | *cp++ = (uchar)((*ip) >> 56); |
4671 | 0 | *cp++ = (uchar)(((*ip) & 0x00ff000000000000ULL) >> 48); |
4672 | 0 | *cp++ = (uchar)(((*ip) & 0x0000ff0000000000ULL) >> 40); |
4673 | 0 | *cp++ = (uchar)(((*ip) & 0x000000ff00000000ULL) >> 32); |
4674 | 0 | *cp++ = (uchar)(((*ip) & 0x00000000ff000000ULL) >> 24); |
4675 | 0 | *cp++ = (uchar)(((*ip) & 0x0000000000ff0000ULL) >> 16); |
4676 | 0 | *cp++ = (uchar)(((*ip) & 0x000000000000ff00ULL) >> 8); |
4677 | 0 | *cp = (uchar)( (*ip) & 0x00000000000000ffULL); |
4678 | 0 | } |
4679 | | |
4680 | | #if X_SIZEOF_UINT64 != SIZEOF_ULONGLONG |
4681 | | static int |
4682 | | ncx_get_ulonglong_ulonglong(const void *xp, ulonglong *ip) |
4683 | | { |
4684 | | int err=NC_NOERR; |
4685 | | #if SIZEOF_IX_UINT64 == SIZEOF_ULONGLONG && IX_UINT64_MAX == ULONGLONG_MAX |
4686 | | get_ix_uint64(xp, (ix_uint64 *)ip); |
4687 | | #else |
4688 | | ix_uint64 xx = 0; |
4689 | | get_ix_uint64(xp, &xx); |
4690 | | |
4691 | | #if IX_UINT64_MAX > ULONGLONG_MAX |
4692 | | if (xx > ULONGLONG_MAX) { |
4693 | | #ifdef ERANGE_FILL |
4694 | | *ip = NC_FILL_UINT64; |
4695 | | return NC_ERANGE; |
4696 | | #else |
4697 | | err = NC_ERANGE; |
4698 | | #endif |
4699 | | } |
4700 | | #endif |
4701 | | |
4702 | | |
4703 | | *ip = (ulonglong) xx; |
4704 | | #endif |
4705 | | return err; |
4706 | | } |
4707 | | |
4708 | | #endif |
4709 | | static int |
4710 | | ncx_get_ulonglong_schar(const void *xp, schar *ip) |
4711 | 0 | { |
4712 | 0 | int err=NC_NOERR; |
4713 | 0 | ix_uint64 xx = 0; |
4714 | 0 | get_ix_uint64(xp, &xx); |
4715 | |
|
4716 | 0 | #if IX_UINT64_MAX > SCHAR_MAX |
4717 | 0 | if (xx > SCHAR_MAX) { |
4718 | 0 | #ifdef ERANGE_FILL |
4719 | 0 | *ip = NC_FILL_BYTE; |
4720 | 0 | return NC_ERANGE; |
4721 | | #else |
4722 | | err = NC_ERANGE; |
4723 | | #endif |
4724 | 0 | } |
4725 | 0 | #endif |
4726 | | |
4727 | | |
4728 | 0 | *ip = (schar) xx; |
4729 | 0 | return err; |
4730 | 0 | } |
4731 | | |
4732 | | static int |
4733 | | ncx_get_ulonglong_short(const void *xp, short *ip) |
4734 | 0 | { |
4735 | 0 | int err=NC_NOERR; |
4736 | 0 | ix_uint64 xx = 0; |
4737 | 0 | get_ix_uint64(xp, &xx); |
4738 | |
|
4739 | 0 | #if IX_UINT64_MAX > SHORT_MAX |
4740 | 0 | if (xx > SHORT_MAX) { |
4741 | 0 | #ifdef ERANGE_FILL |
4742 | 0 | *ip = NC_FILL_SHORT; |
4743 | 0 | return NC_ERANGE; |
4744 | | #else |
4745 | | err = NC_ERANGE; |
4746 | | #endif |
4747 | 0 | } |
4748 | 0 | #endif |
4749 | | |
4750 | | |
4751 | 0 | *ip = (short) xx; |
4752 | 0 | return err; |
4753 | 0 | } |
4754 | | |
4755 | | static int |
4756 | | ncx_get_ulonglong_int(const void *xp, int *ip) |
4757 | 0 | { |
4758 | 0 | int err=NC_NOERR; |
4759 | 0 | ix_uint64 xx = 0; |
4760 | 0 | get_ix_uint64(xp, &xx); |
4761 | |
|
4762 | 0 | #if IX_UINT64_MAX > INT_MAX |
4763 | 0 | if (xx > INT_MAX) { |
4764 | 0 | #ifdef ERANGE_FILL |
4765 | 0 | *ip = NC_FILL_INT; |
4766 | 0 | return NC_ERANGE; |
4767 | | #else |
4768 | | err = NC_ERANGE; |
4769 | | #endif |
4770 | 0 | } |
4771 | 0 | #endif |
4772 | | |
4773 | | |
4774 | 0 | *ip = (int) xx; |
4775 | 0 | return err; |
4776 | 0 | } |
4777 | | |
4778 | | static int |
4779 | | ncx_get_ulonglong_long(const void *xp, long *ip) |
4780 | 0 | { |
4781 | 0 | int err=NC_NOERR; |
4782 | 0 | ix_uint64 xx = 0; |
4783 | 0 | get_ix_uint64(xp, &xx); |
4784 | |
|
4785 | 0 | #if IX_UINT64_MAX > LONG_MAX |
4786 | 0 | if (xx > LONG_MAX) { |
4787 | 0 | #ifdef ERANGE_FILL |
4788 | 0 | *ip = NC_FILL_INT; |
4789 | 0 | return NC_ERANGE; |
4790 | | #else |
4791 | | err = NC_ERANGE; |
4792 | | #endif |
4793 | 0 | } |
4794 | 0 | #endif |
4795 | | |
4796 | | |
4797 | 0 | *ip = (long) xx; |
4798 | 0 | return err; |
4799 | 0 | } |
4800 | | |
4801 | | static int |
4802 | | ncx_get_ulonglong_longlong(const void *xp, longlong *ip) |
4803 | 0 | { |
4804 | 0 | int err=NC_NOERR; |
4805 | 0 | ix_uint64 xx = 0; |
4806 | 0 | get_ix_uint64(xp, &xx); |
4807 | |
|
4808 | 0 | #if IX_UINT64_MAX > LONGLONG_MAX |
4809 | 0 | if (xx > LONGLONG_MAX) { |
4810 | 0 | #ifdef ERANGE_FILL |
4811 | 0 | *ip = NC_FILL_INT64; |
4812 | 0 | return NC_ERANGE; |
4813 | | #else |
4814 | | err = NC_ERANGE; |
4815 | | #endif |
4816 | 0 | } |
4817 | 0 | #endif |
4818 | | |
4819 | | |
4820 | 0 | *ip = (longlong) xx; |
4821 | 0 | return err; |
4822 | 0 | } |
4823 | | |
4824 | | static int |
4825 | | ncx_get_ulonglong_ushort(const void *xp, ushort *ip) |
4826 | 0 | { |
4827 | 0 | int err=NC_NOERR; |
4828 | | #if SIZEOF_IX_UINT64 == SIZEOF_USHORT && IX_UINT64_MAX == USHORT_MAX |
4829 | | get_ix_uint64(xp, (ix_uint64 *)ip); |
4830 | | #else |
4831 | 0 | ix_uint64 xx = 0; |
4832 | 0 | get_ix_uint64(xp, &xx); |
4833 | |
|
4834 | 0 | #if IX_UINT64_MAX > USHORT_MAX |
4835 | 0 | if (xx > USHORT_MAX) { |
4836 | 0 | #ifdef ERANGE_FILL |
4837 | 0 | *ip = NC_FILL_USHORT; |
4838 | 0 | return NC_ERANGE; |
4839 | | #else |
4840 | | err = NC_ERANGE; |
4841 | | #endif |
4842 | 0 | } |
4843 | 0 | #endif |
4844 | | |
4845 | | |
4846 | 0 | *ip = (ushort) xx; |
4847 | 0 | #endif |
4848 | 0 | return err; |
4849 | 0 | } |
4850 | | |
4851 | | static int |
4852 | | ncx_get_ulonglong_uchar(const void *xp, uchar *ip) |
4853 | 0 | { |
4854 | 0 | int err=NC_NOERR; |
4855 | | #if SIZEOF_IX_UINT64 == SIZEOF_UCHAR && IX_UINT64_MAX == UCHAR_MAX |
4856 | | get_ix_uint64(xp, (ix_uint64 *)ip); |
4857 | | #else |
4858 | 0 | ix_uint64 xx = 0; |
4859 | 0 | get_ix_uint64(xp, &xx); |
4860 | |
|
4861 | 0 | #if IX_UINT64_MAX > UCHAR_MAX |
4862 | 0 | if (xx > UCHAR_MAX) { |
4863 | 0 | #ifdef ERANGE_FILL |
4864 | 0 | *ip = NC_FILL_UBYTE; |
4865 | 0 | return NC_ERANGE; |
4866 | | #else |
4867 | | err = NC_ERANGE; |
4868 | | #endif |
4869 | 0 | } |
4870 | 0 | #endif |
4871 | | |
4872 | | |
4873 | 0 | *ip = (uchar) xx; |
4874 | 0 | #endif |
4875 | 0 | return err; |
4876 | 0 | } |
4877 | | |
4878 | | static int |
4879 | | ncx_get_ulonglong_uint(const void *xp, uint *ip) |
4880 | 0 | { |
4881 | 0 | int err=NC_NOERR; |
4882 | | #if SIZEOF_IX_UINT64 == SIZEOF_UINT && IX_UINT64_MAX == UINT_MAX |
4883 | | get_ix_uint64(xp, (ix_uint64 *)ip); |
4884 | | #else |
4885 | 0 | ix_uint64 xx = 0; |
4886 | 0 | get_ix_uint64(xp, &xx); |
4887 | |
|
4888 | 0 | #if IX_UINT64_MAX > UINT_MAX |
4889 | 0 | if (xx > UINT_MAX) { |
4890 | 0 | #ifdef ERANGE_FILL |
4891 | 0 | *ip = NC_FILL_UINT; |
4892 | 0 | return NC_ERANGE; |
4893 | | #else |
4894 | | err = NC_ERANGE; |
4895 | | #endif |
4896 | 0 | } |
4897 | 0 | #endif |
4898 | | |
4899 | | |
4900 | 0 | *ip = (uint) xx; |
4901 | 0 | #endif |
4902 | 0 | return err; |
4903 | 0 | } |
4904 | | |
4905 | | static int |
4906 | | ncx_get_ulonglong_float(const void *xp, float *ip) |
4907 | 0 | { |
4908 | 0 | ix_uint64 xx = 0; |
4909 | 0 | get_ix_uint64(xp, &xx); |
4910 | 0 | *ip = (float)xx; |
4911 | 0 | return NC_NOERR; |
4912 | 0 | } |
4913 | | |
4914 | | static int |
4915 | | ncx_get_ulonglong_double(const void *xp, double *ip) |
4916 | 0 | { |
4917 | 0 | ix_uint64 xx = 0; |
4918 | 0 | get_ix_uint64(xp, &xx); |
4919 | 0 | *ip = (double)xx; |
4920 | 0 | return NC_NOERR; |
4921 | 0 | } |
4922 | | |
4923 | | |
4924 | | #if X_SIZEOF_UINT64 != SIZEOF_ULONGLONG |
4925 | | static int |
4926 | | ncx_put_ulonglong_ulonglong(void *xp, const ulonglong *ip, void *fillp) |
4927 | | { |
4928 | | int err=NC_NOERR; |
4929 | | #if SIZEOF_IX_UINT64 == SIZEOF_ULONGLONG && IX_UINT64_MAX == ULONGLONG_MAX |
4930 | | put_ix_uint64(xp, (const ix_uint64 *)ip); |
4931 | | #else |
4932 | | ix_uint64 xx = NC_FILL_UINT64; |
4933 | | |
4934 | | #if IX_UINT64_MAX < ULONGLONG_MAX |
4935 | | if (*ip > IX_UINT64_MAX) { |
4936 | | |
4937 | | #ifdef ERANGE_FILL |
4938 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4939 | | #endif |
4940 | | err = NC_ERANGE; |
4941 | | } |
4942 | | #ifdef ERANGE_FILL |
4943 | | else |
4944 | | #endif |
4945 | | #endif |
4946 | | xx = (ix_uint64)*ip; |
4947 | | |
4948 | | put_ix_uint64(xp, &xx); |
4949 | | #endif |
4950 | | return err; |
4951 | | } |
4952 | | |
4953 | | #endif |
4954 | | static int |
4955 | | ncx_put_ulonglong_schar(void *xp, const schar *ip, void *fillp) |
4956 | 0 | { |
4957 | 0 | int err=NC_NOERR; |
4958 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
4959 | |
|
4960 | | #if IX_UINT64_MAX < SCHAR_MAX |
4961 | | if (*ip > IX_UINT64_MAX) { |
4962 | | |
4963 | | #ifdef ERANGE_FILL |
4964 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4965 | | #endif |
4966 | | err = NC_ERANGE; |
4967 | | } |
4968 | | #ifdef ERANGE_FILL |
4969 | | else |
4970 | | #endif |
4971 | | #endif |
4972 | 0 | if (*ip < 0) { |
4973 | | |
4974 | 0 | #ifdef ERANGE_FILL |
4975 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4976 | 0 | #endif |
4977 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
4978 | 0 | } |
4979 | 0 | #ifdef ERANGE_FILL |
4980 | 0 | else |
4981 | 0 | #endif |
4982 | 0 | xx = (ix_uint64)*ip; |
4983 | |
|
4984 | 0 | put_ix_uint64(xp, &xx); |
4985 | 0 | return err; |
4986 | 0 | } |
4987 | | |
4988 | | static int |
4989 | | ncx_put_ulonglong_short(void *xp, const short *ip, void *fillp) |
4990 | 0 | { |
4991 | 0 | int err=NC_NOERR; |
4992 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
4993 | |
|
4994 | | #if IX_UINT64_MAX < SHORT_MAX |
4995 | | if (*ip > IX_UINT64_MAX) { |
4996 | | |
4997 | | #ifdef ERANGE_FILL |
4998 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
4999 | | #endif |
5000 | | err = NC_ERANGE; |
5001 | | } |
5002 | | #ifdef ERANGE_FILL |
5003 | | else |
5004 | | #endif |
5005 | | #endif |
5006 | 0 | if (*ip < 0) { |
5007 | | |
5008 | 0 | #ifdef ERANGE_FILL |
5009 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5010 | 0 | #endif |
5011 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
5012 | 0 | } |
5013 | 0 | #ifdef ERANGE_FILL |
5014 | 0 | else |
5015 | 0 | #endif |
5016 | 0 | xx = (ix_uint64)*ip; |
5017 | |
|
5018 | 0 | put_ix_uint64(xp, &xx); |
5019 | 0 | return err; |
5020 | 0 | } |
5021 | | |
5022 | | static int |
5023 | | ncx_put_ulonglong_int(void *xp, const int *ip, void *fillp) |
5024 | 0 | { |
5025 | 0 | int err=NC_NOERR; |
5026 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
5027 | |
|
5028 | | #if IX_UINT64_MAX < INT_MAX |
5029 | | if (*ip > IX_UINT64_MAX) { |
5030 | | |
5031 | | #ifdef ERANGE_FILL |
5032 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5033 | | #endif |
5034 | | err = NC_ERANGE; |
5035 | | } |
5036 | | #ifdef ERANGE_FILL |
5037 | | else |
5038 | | #endif |
5039 | | #endif |
5040 | 0 | if (*ip < 0) { |
5041 | | |
5042 | 0 | #ifdef ERANGE_FILL |
5043 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5044 | 0 | #endif |
5045 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
5046 | 0 | } |
5047 | 0 | #ifdef ERANGE_FILL |
5048 | 0 | else |
5049 | 0 | #endif |
5050 | 0 | xx = (ix_uint64)*ip; |
5051 | |
|
5052 | 0 | put_ix_uint64(xp, &xx); |
5053 | 0 | return err; |
5054 | 0 | } |
5055 | | |
5056 | | static int |
5057 | | ncx_put_ulonglong_long(void *xp, const long *ip, void *fillp) |
5058 | 0 | { |
5059 | 0 | int err=NC_NOERR; |
5060 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
5061 | |
|
5062 | | #if IX_UINT64_MAX < LONG_MAX |
5063 | | if (*ip > IX_UINT64_MAX) { |
5064 | | |
5065 | | #ifdef ERANGE_FILL |
5066 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5067 | | #endif |
5068 | | err = NC_ERANGE; |
5069 | | } |
5070 | | #ifdef ERANGE_FILL |
5071 | | else |
5072 | | #endif |
5073 | | #endif |
5074 | 0 | if (*ip < 0) { |
5075 | | |
5076 | 0 | #ifdef ERANGE_FILL |
5077 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5078 | 0 | #endif |
5079 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
5080 | 0 | } |
5081 | 0 | #ifdef ERANGE_FILL |
5082 | 0 | else |
5083 | 0 | #endif |
5084 | 0 | xx = (ix_uint64)*ip; |
5085 | |
|
5086 | 0 | put_ix_uint64(xp, &xx); |
5087 | 0 | return err; |
5088 | 0 | } |
5089 | | |
5090 | | static int |
5091 | | ncx_put_ulonglong_longlong(void *xp, const longlong *ip, void *fillp) |
5092 | 0 | { |
5093 | 0 | int err=NC_NOERR; |
5094 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
5095 | |
|
5096 | | #if IX_UINT64_MAX < LONGLONG_MAX |
5097 | | if (*ip > IX_UINT64_MAX) { |
5098 | | |
5099 | | #ifdef ERANGE_FILL |
5100 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5101 | | #endif |
5102 | | err = NC_ERANGE; |
5103 | | } |
5104 | | #ifdef ERANGE_FILL |
5105 | | else |
5106 | | #endif |
5107 | | #endif |
5108 | 0 | if (*ip < 0) { |
5109 | | |
5110 | 0 | #ifdef ERANGE_FILL |
5111 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5112 | 0 | #endif |
5113 | 0 | err = NC_ERANGE; /* because xp is unsigned */ |
5114 | 0 | } |
5115 | 0 | #ifdef ERANGE_FILL |
5116 | 0 | else |
5117 | 0 | #endif |
5118 | 0 | xx = (ix_uint64)*ip; |
5119 | |
|
5120 | 0 | put_ix_uint64(xp, &xx); |
5121 | 0 | return err; |
5122 | 0 | } |
5123 | | |
5124 | | static int |
5125 | | ncx_put_ulonglong_uchar(void *xp, const uchar *ip, void *fillp) |
5126 | 0 | { |
5127 | 0 | int err=NC_NOERR; |
5128 | | #if SIZEOF_IX_UINT64 == SIZEOF_UCHAR && IX_UINT64_MAX == UCHAR_MAX |
5129 | | put_ix_uint64(xp, (const ix_uint64 *)ip); |
5130 | | #else |
5131 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
5132 | |
|
5133 | | #if IX_UINT64_MAX < UCHAR_MAX |
5134 | | if (*ip > IX_UINT64_MAX) { |
5135 | | |
5136 | | #ifdef ERANGE_FILL |
5137 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5138 | | #endif |
5139 | | err = NC_ERANGE; |
5140 | | } |
5141 | | #ifdef ERANGE_FILL |
5142 | | else |
5143 | | #endif |
5144 | | #endif |
5145 | 0 | xx = (ix_uint64)*ip; |
5146 | |
|
5147 | 0 | put_ix_uint64(xp, &xx); |
5148 | 0 | #endif |
5149 | 0 | return err; |
5150 | 0 | } |
5151 | | |
5152 | | static int |
5153 | | ncx_put_ulonglong_ushort(void *xp, const ushort *ip, void *fillp) |
5154 | 0 | { |
5155 | 0 | int err=NC_NOERR; |
5156 | | #if SIZEOF_IX_UINT64 == SIZEOF_USHORT && IX_UINT64_MAX == USHORT_MAX |
5157 | | put_ix_uint64(xp, (const ix_uint64 *)ip); |
5158 | | #else |
5159 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
5160 | |
|
5161 | | #if IX_UINT64_MAX < USHORT_MAX |
5162 | | if (*ip > IX_UINT64_MAX) { |
5163 | | |
5164 | | #ifdef ERANGE_FILL |
5165 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5166 | | #endif |
5167 | | err = NC_ERANGE; |
5168 | | } |
5169 | | #ifdef ERANGE_FILL |
5170 | | else |
5171 | | #endif |
5172 | | #endif |
5173 | 0 | xx = (ix_uint64)*ip; |
5174 | |
|
5175 | 0 | put_ix_uint64(xp, &xx); |
5176 | 0 | #endif |
5177 | 0 | return err; |
5178 | 0 | } |
5179 | | |
5180 | | static int |
5181 | | ncx_put_ulonglong_uint(void *xp, const uint *ip, void *fillp) |
5182 | 0 | { |
5183 | 0 | int err=NC_NOERR; |
5184 | | #if SIZEOF_IX_UINT64 == SIZEOF_UINT && IX_UINT64_MAX == UINT_MAX |
5185 | | put_ix_uint64(xp, (const ix_uint64 *)ip); |
5186 | | #else |
5187 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
5188 | |
|
5189 | | #if IX_UINT64_MAX < UINT_MAX |
5190 | | if (*ip > IX_UINT64_MAX) { |
5191 | | |
5192 | | #ifdef ERANGE_FILL |
5193 | | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5194 | | #endif |
5195 | | err = NC_ERANGE; |
5196 | | } |
5197 | | #ifdef ERANGE_FILL |
5198 | | else |
5199 | | #endif |
5200 | | #endif |
5201 | 0 | xx = (ix_uint64)*ip; |
5202 | |
|
5203 | 0 | put_ix_uint64(xp, &xx); |
5204 | 0 | #endif |
5205 | 0 | return err; |
5206 | 0 | } |
5207 | | |
5208 | | static int |
5209 | | ncx_put_ulonglong_float(void *xp, const float *ip, void *fillp) |
5210 | 0 | { |
5211 | 0 | int err=NC_NOERR; |
5212 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
5213 | |
|
5214 | 0 | if (*ip > (double)X_UINT64_MAX || *ip < 0) { |
5215 | | |
5216 | 0 | #ifdef ERANGE_FILL |
5217 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5218 | 0 | #endif |
5219 | 0 | err = NC_ERANGE; |
5220 | 0 | } |
5221 | 0 | #ifdef ERANGE_FILL |
5222 | 0 | else |
5223 | 0 | #endif |
5224 | 0 | xx = (ix_uint64)*ip; |
5225 | |
|
5226 | 0 | put_ix_uint64(xp, &xx); |
5227 | 0 | return err; |
5228 | 0 | } |
5229 | | |
5230 | | static int |
5231 | | ncx_put_ulonglong_double(void *xp, const double *ip, void *fillp) |
5232 | 0 | { |
5233 | 0 | int err=NC_NOERR; |
5234 | 0 | ix_uint64 xx = NC_FILL_UINT64; |
5235 | |
|
5236 | 0 | if (*ip > X_UINT64_MAX || *ip < 0) { |
5237 | | |
5238 | 0 | #ifdef ERANGE_FILL |
5239 | 0 | if (fillp != NULL) memcpy(&xx, fillp, 8); |
5240 | 0 | #endif |
5241 | 0 | err = NC_ERANGE; |
5242 | 0 | } |
5243 | 0 | #ifdef ERANGE_FILL |
5244 | 0 | else |
5245 | 0 | #endif |
5246 | 0 | xx = (ix_uint64)*ip; |
5247 | |
|
5248 | 0 | put_ix_uint64(xp, &xx); |
5249 | 0 | return err; |
5250 | 0 | } |
5251 | | |
5252 | | |
5253 | | |
5254 | | /* x_size_t */ |
5255 | | |
5256 | | #if SIZEOF_SIZE_T < X_SIZEOF_SIZE_T |
5257 | | #error "x_size_t implementation" |
5258 | | /* netcdf requires size_t which can hold a values from 0 to 2^32 -1 */ |
5259 | | #endif |
5260 | | |
5261 | | int |
5262 | | ncx_put_size_t(void **xpp, const size_t *ulp) |
5263 | 0 | { |
5264 | | /* similar to put_ix_int() */ |
5265 | 0 | uchar *cp = (uchar *) *xpp; |
5266 | 0 | assert(*ulp <= X_SIZE_MAX); |
5267 | |
|
5268 | 0 | *cp++ = (uchar)((*ulp) >> 24); |
5269 | 0 | *cp++ = (uchar)(((*ulp) & 0x00ff0000) >> 16); |
5270 | 0 | *cp++ = (uchar)(((*ulp) & 0x0000ff00) >> 8); |
5271 | 0 | *cp = (uchar)((*ulp) & 0x000000ff); |
5272 | |
|
5273 | 0 | *xpp = (void *)((char *)(*xpp) + X_SIZEOF_SIZE_T); |
5274 | 0 | return NC_NOERR; |
5275 | 0 | } |
5276 | | |
5277 | | int |
5278 | | ncx_get_size_t(const void **xpp, size_t *ulp) |
5279 | 1.17M | { |
5280 | | /* similar to get_ix_int */ |
5281 | 1.17M | const uchar *cp = (const uchar *) *xpp; |
5282 | | |
5283 | 1.17M | *ulp = (unsigned)(*cp++) << 24; |
5284 | 1.17M | *ulp |= (*cp++ << 16); |
5285 | 1.17M | *ulp |= (*cp++ << 8); |
5286 | 1.17M | *ulp |= *cp; |
5287 | | |
5288 | 1.17M | *xpp = (const void *)((const char *)(*xpp) + X_SIZEOF_SIZE_T); |
5289 | 1.17M | return NC_NOERR; |
5290 | 1.17M | } |
5291 | | |
5292 | | /* x_off_t */ |
5293 | | |
5294 | | int |
5295 | | ncx_put_off_t(void **xpp, const off_t *lp, size_t sizeof_off_t) |
5296 | 0 | { |
5297 | | /* similar to put_ix_int() */ |
5298 | 0 | uchar *cp = (uchar *) *xpp; |
5299 | | |
5300 | | /* No negative offsets stored in netcdf */ |
5301 | 0 | if (*lp < 0) { |
5302 | | /* Assume this is an overflow of a 32-bit int... */ |
5303 | 0 | return NC_ERANGE; |
5304 | 0 | } |
5305 | | |
5306 | 0 | assert(sizeof_off_t == 4 || sizeof_off_t == 8); |
5307 | |
|
5308 | 0 | if (sizeof_off_t == 4) { |
5309 | 0 | *cp++ = (uchar) ((*lp) >> 24); |
5310 | 0 | *cp++ = (uchar)(((*lp) & 0x00ff0000) >> 16); |
5311 | 0 | *cp++ = (uchar)(((*lp) & 0x0000ff00) >> 8); |
5312 | 0 | *cp = (uchar)( (*lp) & 0x000000ff); |
5313 | 0 | } else { |
5314 | | #if SIZEOF_OFF_T == 4 |
5315 | | /* Write a 64-bit offset on a system with only a 32-bit offset */ |
5316 | | *cp++ = (uchar)0; |
5317 | | *cp++ = (uchar)0; |
5318 | | *cp++ = (uchar)0; |
5319 | | *cp++ = (uchar)0; |
5320 | | |
5321 | | *cp++ = (uchar)(((*lp) & 0xff000000) >> 24); |
5322 | | *cp++ = (uchar)(((*lp) & 0x00ff0000) >> 16); |
5323 | | *cp++ = (uchar)(((*lp) & 0x0000ff00) >> 8); |
5324 | | *cp = (uchar)( (*lp) & 0x000000ff); |
5325 | | #else |
5326 | 0 | *cp++ = (uchar) ((*lp) >> 56); |
5327 | 0 | *cp++ = (uchar)(((*lp) & 0x00ff000000000000LL) >> 48); |
5328 | 0 | *cp++ = (uchar)(((*lp) & 0x0000ff0000000000LL) >> 40); |
5329 | 0 | *cp++ = (uchar)(((*lp) & 0x000000ff00000000LL) >> 32); |
5330 | 0 | *cp++ = (uchar)(((*lp) & 0x00000000ff000000LL) >> 24); |
5331 | 0 | *cp++ = (uchar)(((*lp) & 0x0000000000ff0000LL) >> 16); |
5332 | 0 | *cp++ = (uchar)(((*lp) & 0x000000000000ff00LL) >> 8); |
5333 | 0 | *cp = (uchar)( (*lp) & 0x00000000000000ffLL); |
5334 | 0 | #endif |
5335 | 0 | } |
5336 | 0 | *xpp = (void *)((char *)(*xpp) + sizeof_off_t); |
5337 | 0 | return NC_NOERR; |
5338 | 0 | } |
5339 | | |
5340 | | int |
5341 | | ncx_get_off_t(const void **xpp, off_t *lp, size_t sizeof_off_t) |
5342 | 131k | { |
5343 | | /* similar to get_ix_int() */ |
5344 | 131k | const uchar *cp = (const uchar *) *xpp; |
5345 | 131k | assert(sizeof_off_t == 4 || sizeof_off_t == 8); |
5346 | | |
5347 | 131k | if (sizeof_off_t == 4) { |
5348 | 0 | *lp = (off_t)(*cp++ << 24); |
5349 | 0 | *lp |= (off_t)(*cp++ << 16); |
5350 | 0 | *lp |= (off_t)(*cp++ << 8); |
5351 | 0 | *lp |= (off_t)*cp; |
5352 | 131k | } else { |
5353 | | #if SIZEOF_OFF_T == 4 |
5354 | | /* Read a 64-bit offset on a system with only a 32-bit offset */ |
5355 | | /* If the offset overflows, set an error code and return */ |
5356 | | *lp = ((off_t)(*cp++) << 24); |
5357 | | *lp |= ((off_t)(*cp++) << 16); |
5358 | | *lp |= ((off_t)(*cp++) << 8); |
5359 | | *lp |= ((off_t)(*cp++)); |
5360 | | /* |
5361 | | * lp now contains the upper 32-bits of the 64-bit offset. if lp is |
5362 | | * not zero, then the dataset is larger than can be represented |
5363 | | * on this system. Set an error code and return. |
5364 | | */ |
5365 | | if (*lp != 0) { |
5366 | | return NC_ERANGE; |
5367 | | } |
5368 | | |
5369 | | *lp = ((off_t)(*cp++) << 24); |
5370 | | *lp |= ((off_t)(*cp++) << 16); |
5371 | | *lp |= ((off_t)(*cp++) << 8); |
5372 | | *lp |= (off_t)*cp; |
5373 | | |
5374 | | if (*lp < 0) { |
5375 | | /* |
5376 | | * If this fails, then the offset is >2^31, but less |
5377 | | * than 2^32 which is not allowed, but is not caught |
5378 | | * by the previous check |
5379 | | */ |
5380 | | return NC_ERANGE; |
5381 | | } |
5382 | | #else |
5383 | 131k | *lp = ((off_t)(*cp++) << 56); |
5384 | 131k | *lp |= ((off_t)(*cp++) << 48); |
5385 | 131k | *lp |= ((off_t)(*cp++) << 40); |
5386 | 131k | *lp |= ((off_t)(*cp++) << 32); |
5387 | 131k | *lp |= ((off_t)(*cp++) << 24); |
5388 | 131k | *lp |= ((off_t)(*cp++) << 16); |
5389 | 131k | *lp |= ((off_t)(*cp++) << 8); |
5390 | 131k | *lp |= (off_t)*cp; |
5391 | 131k | #endif |
5392 | 131k | } |
5393 | 131k | *xpp = (const void *)((const char *)(*xpp) + sizeof_off_t); |
5394 | 131k | return NC_NOERR; |
5395 | 131k | } |
5396 | | |
5397 | | /*----< ncx_get_uint32() >------------------------------------------*/ |
5398 | | int |
5399 | | ncx_get_uint32(const void **xpp, uint *ip) |
5400 | 369k | { |
5401 | | #ifdef WORDS_BIGENDIAN |
5402 | | /* use memcpy instead of assignment to avoid BUS_ADRALN alignment error on |
5403 | | * some system, such as HPUX */ |
5404 | | (void) memcpy(ip, *xpp, SIZEOF_UINT); |
5405 | | #else |
5406 | 369k | const uchar *cp = (const uchar *) *xpp; |
5407 | | |
5408 | 369k | *ip = (uint)(*cp++ << 24); |
5409 | 369k | *ip = (uint)(*ip | (uint)(*cp++ << 16)); |
5410 | 369k | *ip = (uint)(*ip | (uint)(*cp++ << 8)); |
5411 | 369k | *ip = (uint)(*ip | *cp); |
5412 | 369k | #endif |
5413 | | /* advance *xpp 4 bytes */ |
5414 | 369k | *xpp = (void *)((const char *)(*xpp) + 4); |
5415 | | |
5416 | 369k | return NC_NOERR; |
5417 | 369k | } |
5418 | | |
5419 | | /*----< ncx_get_uint64() >------------------------------------------*/ |
5420 | | int |
5421 | | ncx_get_uint64(const void **xpp, unsigned long long *ullp) |
5422 | 287k | { |
5423 | | #ifdef WORDS_BIGENDIAN |
5424 | | /* use memcpy instead of assignment to avoid BUS_ADRALN alignment error on |
5425 | | * some system, such as HPUX */ |
5426 | | (void) memcpy(ullp, *xpp, SIZEOF_UINT64); |
5427 | | #else |
5428 | 287k | const uchar *cp = (const uchar *) *xpp; |
5429 | | |
5430 | | /* below is the same as calling swap8b(ullp, *xpp) */ |
5431 | 287k | *ullp = (unsigned long long)(*cp++) << 56; |
5432 | 287k | *ullp = (unsigned long long)(*ullp | (unsigned long long)(*cp++) << 48); |
5433 | 287k | *ullp = (unsigned long long)(*ullp | (unsigned long long)(*cp++) << 40); |
5434 | 287k | *ullp = (unsigned long long)(*ullp | (unsigned long long)(*cp++) << 32); |
5435 | 287k | *ullp = (unsigned long long)(*ullp | (unsigned long long)(*cp++) << 24); |
5436 | 287k | *ullp = (unsigned long long)(*ullp | (unsigned long long)(*cp++) << 16); |
5437 | 287k | *ullp = (unsigned long long)(*ullp | (unsigned long long)(*cp++) << 8); |
5438 | 287k | *ullp = (unsigned long long)(*ullp | (unsigned long long)(*cp)); |
5439 | 287k | #endif |
5440 | | /* advance *xpp 8 bytes */ |
5441 | 287k | *xpp = (void *)((const char *)(*xpp) + 8); |
5442 | | |
5443 | 287k | return NC_NOERR; |
5444 | 287k | } |
5445 | | |
5446 | | /*---< ncx_put_uint32() >-------------------------------------------*/ |
5447 | | /* copy the contents of ip (an unsigned 32-bit integer) to xpp in Big Endian |
5448 | | * form and advance *xpp 4 bytes |
5449 | | */ |
5450 | | int |
5451 | | ncx_put_uint32(void **xpp, const unsigned int ip) |
5452 | 0 | { |
5453 | | #ifdef WORDS_BIGENDIAN |
5454 | | /* use memcpy instead of assignment to avoid BUS_ADRALN alignment error on |
5455 | | * some system, such as HPUX */ |
5456 | | (void) memcpy(*xpp, &ip, X_SIZEOF_UINT); |
5457 | | #else |
5458 | | /* bitwise shifts below are to produce an integer in Big Endian */ |
5459 | 0 | uchar *cp = (uchar *) *xpp; |
5460 | 0 | *cp++ = (uchar)((ip & 0xff000000) >> 24); |
5461 | 0 | *cp++ = (uchar)((ip & 0x00ff0000) >> 16); |
5462 | 0 | *cp++ = (uchar)((ip & 0x0000ff00) >> 8); |
5463 | 0 | *cp = (uchar)( ip & 0x000000ff); |
5464 | 0 | #endif |
5465 | | /* advance *xpp 4 bytes */ |
5466 | 0 | *xpp = (void *)((char *)(*xpp) + 4); |
5467 | |
|
5468 | 0 | return NC_NOERR; |
5469 | 0 | } |
5470 | | |
5471 | | /*---< ncx_put_uint64() >-------------------------------------------*/ |
5472 | | /* copy the contents of ip (an unsigned 64-bit integer) to xpp in Big Endian |
5473 | | * form and advance *xpp 8 bytes |
5474 | | */ |
5475 | | int |
5476 | | ncx_put_uint64(void **xpp, const unsigned long long ip) |
5477 | 0 | { |
5478 | | #ifdef WORDS_BIGENDIAN |
5479 | | /* use memcpy instead of assignment to avoid BUS_ADRALN alignment error on |
5480 | | * some system, such as HPUX */ |
5481 | | (void) memcpy(*xpp, &ip, X_SIZEOF_UINT64); |
5482 | | #else |
5483 | 0 | uchar *cp = (uchar *) *xpp; |
5484 | | /* below is the same as calling swap8b(*xpp, &ip) */ |
5485 | 0 | *cp++ = (uchar) (ip >> 56); |
5486 | 0 | *cp++ = (uchar)((ip & 0x00ff000000000000LL) >> 48); |
5487 | 0 | *cp++ = (uchar)((ip & 0x0000ff0000000000LL) >> 40); |
5488 | 0 | *cp++ = (uchar)((ip & 0x000000ff00000000LL) >> 32); |
5489 | 0 | *cp++ = (uchar)((ip & 0x00000000ff000000LL) >> 24); |
5490 | 0 | *cp++ = (uchar)((ip & 0x0000000000ff0000LL) >> 16); |
5491 | 0 | *cp++ = (uchar)((ip & 0x000000000000ff00LL) >> 8); |
5492 | 0 | *cp = (uchar) (ip & 0x00000000000000ffLL); |
5493 | 0 | #endif |
5494 | | /* advance *xpp 8 bytes */ |
5495 | 0 | *xpp = (void *)((char *)(*xpp) + 8); |
5496 | |
|
5497 | 0 | return NC_NOERR; |
5498 | 0 | } |
5499 | | |
5500 | | |
5501 | | /* |
5502 | | * Aggregate numeric conversion functions. |
5503 | | */ |
5504 | | |
5505 | | |
5506 | | |
5507 | | /* schar ---------------------------------------------------------------------*/ |
5508 | | |
5509 | | int |
5510 | | ncx_getn_schar_schar(const void **xpp, size_t nelems, schar *tp) |
5511 | 97 | { |
5512 | 97 | (void) memcpy(tp, *xpp, (size_t)nelems); |
5513 | 97 | *xpp = (void *)((char *)(*xpp) + nelems); |
5514 | 97 | return NC_NOERR; |
5515 | | |
5516 | 97 | } |
5517 | | int |
5518 | | ncx_getn_schar_uchar(const void **xpp, size_t nelems, uchar *tp) |
5519 | 0 | { |
5520 | 0 | int status = NC_NOERR; |
5521 | 0 | schar *xp = (schar *)(*xpp); |
5522 | |
|
5523 | 0 | while (nelems-- != 0) { |
5524 | | |
5525 | 0 | if (*xp < 0) { |
5526 | 0 | #ifdef ERANGE_FILL |
5527 | 0 | *tp = NC_FILL_UBYTE; |
5528 | 0 | #endif |
5529 | 0 | status = NC_ERANGE; /* because tp is unsigned */ |
5530 | | |
5531 | 0 | #ifdef ERANGE_FILL |
5532 | 0 | xp++; tp++; continue; |
5533 | 0 | #endif |
5534 | 0 | } |
5535 | 0 | *tp++ = (uchar) (signed) (*xp++); /* type cast from schar to uchar */ |
5536 | 0 | } |
5537 | |
|
5538 | 0 | *xpp = (const void *)xp; |
5539 | 0 | return status; |
5540 | 0 | } |
5541 | | |
5542 | | int |
5543 | | ncx_getn_schar_short(const void **xpp, size_t nelems, short *tp) |
5544 | 0 | { |
5545 | 0 | int status = NC_NOERR; |
5546 | 0 | schar *xp = (schar *)(*xpp); |
5547 | |
|
5548 | 0 | while (nelems-- != 0) { |
5549 | | |
5550 | 0 | *tp++ = (short) (*xp++); /* type cast from schar to short */ |
5551 | 0 | } |
5552 | |
|
5553 | 0 | *xpp = (const void *)xp; |
5554 | 0 | return status; |
5555 | 0 | } |
5556 | | |
5557 | | int |
5558 | | ncx_getn_schar_int(const void **xpp, size_t nelems, int *tp) |
5559 | 0 | { |
5560 | 0 | int status = NC_NOERR; |
5561 | 0 | schar *xp = (schar *)(*xpp); |
5562 | |
|
5563 | 0 | while (nelems-- != 0) { |
5564 | | |
5565 | 0 | *tp++ = (int) (*xp++); /* type cast from schar to int */ |
5566 | 0 | } |
5567 | |
|
5568 | 0 | *xpp = (const void *)xp; |
5569 | 0 | return status; |
5570 | 0 | } |
5571 | | |
5572 | | int |
5573 | | ncx_getn_schar_long(const void **xpp, size_t nelems, long *tp) |
5574 | 0 | { |
5575 | 0 | int status = NC_NOERR; |
5576 | 0 | schar *xp = (schar *)(*xpp); |
5577 | |
|
5578 | 0 | while (nelems-- != 0) { |
5579 | | |
5580 | 0 | *tp++ = (long) (*xp++); /* type cast from schar to long */ |
5581 | 0 | } |
5582 | |
|
5583 | 0 | *xpp = (const void *)xp; |
5584 | 0 | return status; |
5585 | 0 | } |
5586 | | |
5587 | | int |
5588 | | ncx_getn_schar_float(const void **xpp, size_t nelems, float *tp) |
5589 | 0 | { |
5590 | 0 | int status = NC_NOERR; |
5591 | 0 | schar *xp = (schar *)(*xpp); |
5592 | |
|
5593 | 0 | while (nelems-- != 0) { |
5594 | | |
5595 | 0 | *tp++ = (float) (*xp++); /* type cast from schar to float */ |
5596 | 0 | } |
5597 | |
|
5598 | 0 | *xpp = (const void *)xp; |
5599 | 0 | return status; |
5600 | 0 | } |
5601 | | |
5602 | | int |
5603 | | ncx_getn_schar_double(const void **xpp, size_t nelems, double *tp) |
5604 | 0 | { |
5605 | 0 | int status = NC_NOERR; |
5606 | 0 | schar *xp = (schar *)(*xpp); |
5607 | |
|
5608 | 0 | while (nelems-- != 0) { |
5609 | | |
5610 | 0 | *tp++ = (double) (*xp++); /* type cast from schar to double */ |
5611 | 0 | } |
5612 | |
|
5613 | 0 | *xpp = (const void *)xp; |
5614 | 0 | return status; |
5615 | 0 | } |
5616 | | |
5617 | | int |
5618 | | ncx_getn_schar_longlong(const void **xpp, size_t nelems, longlong *tp) |
5619 | 0 | { |
5620 | 0 | int status = NC_NOERR; |
5621 | 0 | schar *xp = (schar *)(*xpp); |
5622 | |
|
5623 | 0 | while (nelems-- != 0) { |
5624 | | |
5625 | 0 | *tp++ = (longlong) (*xp++); /* type cast from schar to longlong */ |
5626 | 0 | } |
5627 | |
|
5628 | 0 | *xpp = (const void *)xp; |
5629 | 0 | return status; |
5630 | 0 | } |
5631 | | |
5632 | | int |
5633 | | ncx_getn_schar_ushort(const void **xpp, size_t nelems, ushort *tp) |
5634 | 0 | { |
5635 | 0 | int status = NC_NOERR; |
5636 | 0 | schar *xp = (schar *)(*xpp); |
5637 | |
|
5638 | 0 | while (nelems-- != 0) { |
5639 | | |
5640 | 0 | if (*xp < 0) { |
5641 | 0 | #ifdef ERANGE_FILL |
5642 | 0 | *tp = NC_FILL_USHORT; |
5643 | 0 | #endif |
5644 | 0 | status = NC_ERANGE; /* because tp is unsigned */ |
5645 | | |
5646 | 0 | #ifdef ERANGE_FILL |
5647 | 0 | xp++; tp++; continue; |
5648 | 0 | #endif |
5649 | 0 | } |
5650 | 0 | *tp++ = (ushort) (signed) (*xp++); /* type cast from schar to ushort */ |
5651 | 0 | } |
5652 | |
|
5653 | 0 | *xpp = (const void *)xp; |
5654 | 0 | return status; |
5655 | 0 | } |
5656 | | |
5657 | | int |
5658 | | ncx_getn_schar_uint(const void **xpp, size_t nelems, uint *tp) |
5659 | 0 | { |
5660 | 0 | int status = NC_NOERR; |
5661 | 0 | schar *xp = (schar *)(*xpp); |
5662 | |
|
5663 | 0 | while (nelems-- != 0) { |
5664 | | |
5665 | 0 | if (*xp < 0) { |
5666 | 0 | #ifdef ERANGE_FILL |
5667 | 0 | *tp = NC_FILL_UINT; |
5668 | 0 | #endif |
5669 | 0 | status = NC_ERANGE; /* because tp is unsigned */ |
5670 | | |
5671 | 0 | #ifdef ERANGE_FILL |
5672 | 0 | xp++; tp++; continue; |
5673 | 0 | #endif |
5674 | 0 | } |
5675 | 0 | *tp++ = (uint) (signed) (*xp++); /* type cast from schar to uint */ |
5676 | 0 | } |
5677 | |
|
5678 | 0 | *xpp = (const void *)xp; |
5679 | 0 | return status; |
5680 | 0 | } |
5681 | | |
5682 | | int |
5683 | | ncx_getn_schar_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
5684 | 0 | { |
5685 | 0 | int status = NC_NOERR; |
5686 | 0 | schar *xp = (schar *)(*xpp); |
5687 | |
|
5688 | 0 | while (nelems-- != 0) { |
5689 | | |
5690 | 0 | if (*xp < 0) { |
5691 | 0 | #ifdef ERANGE_FILL |
5692 | 0 | *tp = NC_FILL_UINT64; |
5693 | 0 | #endif |
5694 | 0 | status = NC_ERANGE; /* because tp is unsigned */ |
5695 | | |
5696 | 0 | #ifdef ERANGE_FILL |
5697 | 0 | xp++; tp++; continue; |
5698 | 0 | #endif |
5699 | 0 | } |
5700 | 0 | *tp++ = (ulonglong) (signed) (*xp++); /* type cast from schar to ulonglong */ |
5701 | 0 | } |
5702 | |
|
5703 | 0 | *xpp = (const void *)xp; |
5704 | 0 | return status; |
5705 | 0 | } |
5706 | | |
5707 | | |
5708 | | int |
5709 | | ncx_pad_getn_schar_schar(const void **xpp, size_t nelems, schar *tp) |
5710 | 0 | { |
5711 | 0 | size_t rndup = nelems % X_ALIGN; |
5712 | |
|
5713 | 0 | if (rndup) |
5714 | 0 | rndup = X_ALIGN - rndup; |
5715 | |
|
5716 | 0 | (void) memcpy(tp, *xpp, (size_t)nelems); |
5717 | 0 | *xpp = (void *)((char *)(*xpp) + nelems + rndup); |
5718 | |
|
5719 | 0 | return NC_NOERR; |
5720 | |
|
5721 | 0 | } |
5722 | | int |
5723 | | ncx_pad_getn_schar_uchar(const void **xpp, size_t nelems, uchar *tp) |
5724 | 0 | { |
5725 | 0 | int status = NC_NOERR; |
5726 | 0 | size_t rndup = nelems % X_ALIGN; |
5727 | 0 | schar *xp = (schar *) *xpp; |
5728 | |
|
5729 | 0 | if (rndup) |
5730 | 0 | rndup = X_ALIGN - rndup; |
5731 | |
|
5732 | 0 | while (nelems-- != 0) { |
5733 | | |
5734 | 0 | if (*xp < 0) { |
5735 | 0 | #ifdef ERANGE_FILL |
5736 | 0 | *tp = NC_FILL_UBYTE; |
5737 | 0 | #endif |
5738 | 0 | status = NC_ERANGE; /* because tp is unsigned */ |
5739 | | |
5740 | 0 | #ifdef ERANGE_FILL |
5741 | 0 | xp++; tp++; continue; |
5742 | 0 | #endif |
5743 | 0 | } |
5744 | 0 | *tp++ = (uchar) (signed) (*xp++); /* type cast from schar to uchar */ |
5745 | 0 | } |
5746 | |
|
5747 | 0 | *xpp = (void *)(xp + rndup); |
5748 | 0 | return status; |
5749 | 0 | } |
5750 | | |
5751 | | int |
5752 | | ncx_pad_getn_schar_short(const void **xpp, size_t nelems, short *tp) |
5753 | 0 | { |
5754 | 0 | int status = NC_NOERR; |
5755 | 0 | size_t rndup = nelems % X_ALIGN; |
5756 | 0 | schar *xp = (schar *) *xpp; |
5757 | |
|
5758 | 0 | if (rndup) |
5759 | 0 | rndup = X_ALIGN - rndup; |
5760 | |
|
5761 | 0 | while (nelems-- != 0) { |
5762 | | |
5763 | 0 | *tp++ = (short) (*xp++); /* type cast from schar to short */ |
5764 | 0 | } |
5765 | |
|
5766 | 0 | *xpp = (void *)(xp + rndup); |
5767 | 0 | return status; |
5768 | 0 | } |
5769 | | |
5770 | | int |
5771 | | ncx_pad_getn_schar_int(const void **xpp, size_t nelems, int *tp) |
5772 | 0 | { |
5773 | 0 | int status = NC_NOERR; |
5774 | 0 | size_t rndup = nelems % X_ALIGN; |
5775 | 0 | schar *xp = (schar *) *xpp; |
5776 | |
|
5777 | 0 | if (rndup) |
5778 | 0 | rndup = X_ALIGN - rndup; |
5779 | |
|
5780 | 0 | while (nelems-- != 0) { |
5781 | | |
5782 | 0 | *tp++ = (int) (*xp++); /* type cast from schar to int */ |
5783 | 0 | } |
5784 | |
|
5785 | 0 | *xpp = (void *)(xp + rndup); |
5786 | 0 | return status; |
5787 | 0 | } |
5788 | | |
5789 | | int |
5790 | | ncx_pad_getn_schar_long(const void **xpp, size_t nelems, long *tp) |
5791 | 0 | { |
5792 | 0 | int status = NC_NOERR; |
5793 | 0 | size_t rndup = nelems % X_ALIGN; |
5794 | 0 | schar *xp = (schar *) *xpp; |
5795 | |
|
5796 | 0 | if (rndup) |
5797 | 0 | rndup = X_ALIGN - rndup; |
5798 | |
|
5799 | 0 | while (nelems-- != 0) { |
5800 | | |
5801 | 0 | *tp++ = (long) (*xp++); /* type cast from schar to long */ |
5802 | 0 | } |
5803 | |
|
5804 | 0 | *xpp = (void *)(xp + rndup); |
5805 | 0 | return status; |
5806 | 0 | } |
5807 | | |
5808 | | int |
5809 | | ncx_pad_getn_schar_float(const void **xpp, size_t nelems, float *tp) |
5810 | 0 | { |
5811 | 0 | int status = NC_NOERR; |
5812 | 0 | size_t rndup = nelems % X_ALIGN; |
5813 | 0 | schar *xp = (schar *) *xpp; |
5814 | |
|
5815 | 0 | if (rndup) |
5816 | 0 | rndup = X_ALIGN - rndup; |
5817 | |
|
5818 | 0 | while (nelems-- != 0) { |
5819 | | |
5820 | 0 | *tp++ = (float) (*xp++); /* type cast from schar to float */ |
5821 | 0 | } |
5822 | |
|
5823 | 0 | *xpp = (void *)(xp + rndup); |
5824 | 0 | return status; |
5825 | 0 | } |
5826 | | |
5827 | | int |
5828 | | ncx_pad_getn_schar_double(const void **xpp, size_t nelems, double *tp) |
5829 | 0 | { |
5830 | 0 | int status = NC_NOERR; |
5831 | 0 | size_t rndup = nelems % X_ALIGN; |
5832 | 0 | schar *xp = (schar *) *xpp; |
5833 | |
|
5834 | 0 | if (rndup) |
5835 | 0 | rndup = X_ALIGN - rndup; |
5836 | |
|
5837 | 0 | while (nelems-- != 0) { |
5838 | | |
5839 | 0 | *tp++ = (double) (*xp++); /* type cast from schar to double */ |
5840 | 0 | } |
5841 | |
|
5842 | 0 | *xpp = (void *)(xp + rndup); |
5843 | 0 | return status; |
5844 | 0 | } |
5845 | | |
5846 | | int |
5847 | | ncx_pad_getn_schar_longlong(const void **xpp, size_t nelems, longlong *tp) |
5848 | 0 | { |
5849 | 0 | int status = NC_NOERR; |
5850 | 0 | size_t rndup = nelems % X_ALIGN; |
5851 | 0 | schar *xp = (schar *) *xpp; |
5852 | |
|
5853 | 0 | if (rndup) |
5854 | 0 | rndup = X_ALIGN - rndup; |
5855 | |
|
5856 | 0 | while (nelems-- != 0) { |
5857 | | |
5858 | 0 | *tp++ = (longlong) (*xp++); /* type cast from schar to longlong */ |
5859 | 0 | } |
5860 | |
|
5861 | 0 | *xpp = (void *)(xp + rndup); |
5862 | 0 | return status; |
5863 | 0 | } |
5864 | | |
5865 | | int |
5866 | | ncx_pad_getn_schar_ushort(const void **xpp, size_t nelems, ushort *tp) |
5867 | 0 | { |
5868 | 0 | int status = NC_NOERR; |
5869 | 0 | size_t rndup = nelems % X_ALIGN; |
5870 | 0 | schar *xp = (schar *) *xpp; |
5871 | |
|
5872 | 0 | if (rndup) |
5873 | 0 | rndup = X_ALIGN - rndup; |
5874 | |
|
5875 | 0 | while (nelems-- != 0) { |
5876 | | |
5877 | 0 | if (*xp < 0) { |
5878 | 0 | #ifdef ERANGE_FILL |
5879 | 0 | *tp = NC_FILL_USHORT; |
5880 | 0 | #endif |
5881 | 0 | status = NC_ERANGE; /* because tp is unsigned */ |
5882 | | |
5883 | 0 | #ifdef ERANGE_FILL |
5884 | 0 | xp++; tp++; continue; |
5885 | 0 | #endif |
5886 | 0 | } |
5887 | 0 | *tp++ = (ushort) (signed) (*xp++); /* type cast from schar to ushort */ |
5888 | 0 | } |
5889 | |
|
5890 | 0 | *xpp = (void *)(xp + rndup); |
5891 | 0 | return status; |
5892 | 0 | } |
5893 | | |
5894 | | int |
5895 | | ncx_pad_getn_schar_uint(const void **xpp, size_t nelems, uint *tp) |
5896 | 0 | { |
5897 | 0 | int status = NC_NOERR; |
5898 | 0 | size_t rndup = nelems % X_ALIGN; |
5899 | 0 | schar *xp = (schar *) *xpp; |
5900 | |
|
5901 | 0 | if (rndup) |
5902 | 0 | rndup = X_ALIGN - rndup; |
5903 | |
|
5904 | 0 | while (nelems-- != 0) { |
5905 | | |
5906 | 0 | if (*xp < 0) { |
5907 | 0 | #ifdef ERANGE_FILL |
5908 | 0 | *tp = NC_FILL_UINT; |
5909 | 0 | #endif |
5910 | 0 | status = NC_ERANGE; /* because tp is unsigned */ |
5911 | | |
5912 | 0 | #ifdef ERANGE_FILL |
5913 | 0 | xp++; tp++; continue; |
5914 | 0 | #endif |
5915 | 0 | } |
5916 | 0 | *tp++ = (uint) (signed) (*xp++); /* type cast from schar to uint */ |
5917 | 0 | } |
5918 | |
|
5919 | 0 | *xpp = (void *)(xp + rndup); |
5920 | 0 | return status; |
5921 | 0 | } |
5922 | | |
5923 | | int |
5924 | | ncx_pad_getn_schar_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
5925 | 0 | { |
5926 | 0 | int status = NC_NOERR; |
5927 | 0 | size_t rndup = nelems % X_ALIGN; |
5928 | 0 | schar *xp = (schar *) *xpp; |
5929 | |
|
5930 | 0 | if (rndup) |
5931 | 0 | rndup = X_ALIGN - rndup; |
5932 | |
|
5933 | 0 | while (nelems-- != 0) { |
5934 | | |
5935 | 0 | if (*xp < 0) { |
5936 | 0 | #ifdef ERANGE_FILL |
5937 | 0 | *tp = NC_FILL_UINT64; |
5938 | 0 | #endif |
5939 | 0 | status = NC_ERANGE; /* because tp is unsigned */ |
5940 | | |
5941 | 0 | #ifdef ERANGE_FILL |
5942 | 0 | xp++; tp++; continue; |
5943 | 0 | #endif |
5944 | 0 | } |
5945 | 0 | *tp++ = (ulonglong) (signed) (*xp++); /* type cast from schar to ulonglong */ |
5946 | 0 | } |
5947 | |
|
5948 | 0 | *xpp = (void *)(xp + rndup); |
5949 | 0 | return status; |
5950 | 0 | } |
5951 | | |
5952 | | |
5953 | | int |
5954 | | ncx_putn_schar_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
5955 | 0 | { |
5956 | 0 | (void) memcpy(*xpp, tp, (size_t)nelems); |
5957 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
5958 | |
|
5959 | 0 | return NC_NOERR; |
5960 | |
|
5961 | 0 | } |
5962 | | int |
5963 | | ncx_putn_schar_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
5964 | 0 | { |
5965 | 0 | int status = NC_NOERR; |
5966 | 0 | schar *xp = (schar *) *xpp; |
5967 | |
|
5968 | 0 | while (nelems-- != 0) { |
5969 | 0 | if (*tp > (uchar)X_SCHAR_MAX ) { |
5970 | | |
5971 | 0 | #ifdef ERANGE_FILL |
5972 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
5973 | 0 | #endif |
5974 | 0 | status = NC_ERANGE; |
5975 | | |
5976 | 0 | #ifdef ERANGE_FILL |
5977 | 0 | xp++; tp++; continue; |
5978 | 0 | #endif |
5979 | 0 | } |
5980 | 0 | *xp++ = (schar) *tp++; /* type cast from uchar to schar */ |
5981 | 0 | } |
5982 | |
|
5983 | 0 | *xpp = (void *)xp; |
5984 | 0 | return status; |
5985 | 0 | } |
5986 | | |
5987 | | int |
5988 | | ncx_putn_schar_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
5989 | 0 | { |
5990 | 0 | int status = NC_NOERR; |
5991 | 0 | schar *xp = (schar *) *xpp; |
5992 | |
|
5993 | 0 | while (nelems-- != 0) { |
5994 | 0 | if (*tp > (short)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
5995 | | |
5996 | 0 | #ifdef ERANGE_FILL |
5997 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
5998 | 0 | #endif |
5999 | 0 | status = NC_ERANGE; |
6000 | | |
6001 | 0 | #ifdef ERANGE_FILL |
6002 | 0 | xp++; tp++; continue; |
6003 | 0 | #endif |
6004 | 0 | } |
6005 | 0 | *xp++ = (schar) *tp++; /* type cast from short to schar */ |
6006 | 0 | } |
6007 | |
|
6008 | 0 | *xpp = (void *)xp; |
6009 | 0 | return status; |
6010 | 0 | } |
6011 | | |
6012 | | int |
6013 | | ncx_putn_schar_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
6014 | 0 | { |
6015 | 0 | int status = NC_NOERR; |
6016 | 0 | schar *xp = (schar *) *xpp; |
6017 | |
|
6018 | 0 | while (nelems-- != 0) { |
6019 | 0 | if (*tp > (int)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6020 | | |
6021 | 0 | #ifdef ERANGE_FILL |
6022 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6023 | 0 | #endif |
6024 | 0 | status = NC_ERANGE; |
6025 | | |
6026 | 0 | #ifdef ERANGE_FILL |
6027 | 0 | xp++; tp++; continue; |
6028 | 0 | #endif |
6029 | 0 | } |
6030 | 0 | *xp++ = (schar) *tp++; /* type cast from int to schar */ |
6031 | 0 | } |
6032 | |
|
6033 | 0 | *xpp = (void *)xp; |
6034 | 0 | return status; |
6035 | 0 | } |
6036 | | |
6037 | | int |
6038 | | ncx_putn_schar_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
6039 | 0 | { |
6040 | 0 | int status = NC_NOERR; |
6041 | 0 | schar *xp = (schar *) *xpp; |
6042 | |
|
6043 | 0 | while (nelems-- != 0) { |
6044 | 0 | if (*tp > (long)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6045 | | |
6046 | 0 | #ifdef ERANGE_FILL |
6047 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6048 | 0 | #endif |
6049 | 0 | status = NC_ERANGE; |
6050 | | |
6051 | 0 | #ifdef ERANGE_FILL |
6052 | 0 | xp++; tp++; continue; |
6053 | 0 | #endif |
6054 | 0 | } |
6055 | 0 | *xp++ = (schar) *tp++; /* type cast from long to schar */ |
6056 | 0 | } |
6057 | |
|
6058 | 0 | *xpp = (void *)xp; |
6059 | 0 | return status; |
6060 | 0 | } |
6061 | | |
6062 | | int |
6063 | | ncx_putn_schar_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
6064 | 0 | { |
6065 | 0 | int status = NC_NOERR; |
6066 | 0 | schar *xp = (schar *) *xpp; |
6067 | |
|
6068 | 0 | while (nelems-- != 0) { |
6069 | 0 | if (*tp > (float)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6070 | | |
6071 | 0 | #ifdef ERANGE_FILL |
6072 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6073 | 0 | #endif |
6074 | 0 | status = NC_ERANGE; |
6075 | | |
6076 | 0 | #ifdef ERANGE_FILL |
6077 | 0 | xp++; tp++; continue; |
6078 | 0 | #endif |
6079 | 0 | } |
6080 | 0 | *xp++ = (schar) *tp++; /* type cast from float to schar */ |
6081 | 0 | } |
6082 | |
|
6083 | 0 | *xpp = (void *)xp; |
6084 | 0 | return status; |
6085 | 0 | } |
6086 | | |
6087 | | int |
6088 | | ncx_putn_schar_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
6089 | 0 | { |
6090 | 0 | int status = NC_NOERR; |
6091 | 0 | schar *xp = (schar *) *xpp; |
6092 | |
|
6093 | 0 | while (nelems-- != 0) { |
6094 | 0 | if (*tp > (double)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6095 | | |
6096 | 0 | #ifdef ERANGE_FILL |
6097 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6098 | 0 | #endif |
6099 | 0 | status = NC_ERANGE; |
6100 | | |
6101 | 0 | #ifdef ERANGE_FILL |
6102 | 0 | xp++; tp++; continue; |
6103 | 0 | #endif |
6104 | 0 | } |
6105 | 0 | *xp++ = (schar) *tp++; /* type cast from double to schar */ |
6106 | 0 | } |
6107 | |
|
6108 | 0 | *xpp = (void *)xp; |
6109 | 0 | return status; |
6110 | 0 | } |
6111 | | |
6112 | | int |
6113 | | ncx_putn_schar_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
6114 | 0 | { |
6115 | 0 | int status = NC_NOERR; |
6116 | 0 | schar *xp = (schar *) *xpp; |
6117 | |
|
6118 | 0 | while (nelems-- != 0) { |
6119 | 0 | if (*tp > (longlong)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6120 | | |
6121 | 0 | #ifdef ERANGE_FILL |
6122 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6123 | 0 | #endif |
6124 | 0 | status = NC_ERANGE; |
6125 | | |
6126 | 0 | #ifdef ERANGE_FILL |
6127 | 0 | xp++; tp++; continue; |
6128 | 0 | #endif |
6129 | 0 | } |
6130 | 0 | *xp++ = (schar) *tp++; /* type cast from longlong to schar */ |
6131 | 0 | } |
6132 | |
|
6133 | 0 | *xpp = (void *)xp; |
6134 | 0 | return status; |
6135 | 0 | } |
6136 | | |
6137 | | int |
6138 | | ncx_putn_schar_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
6139 | 0 | { |
6140 | 0 | int status = NC_NOERR; |
6141 | 0 | schar *xp = (schar *) *xpp; |
6142 | |
|
6143 | 0 | while (nelems-- != 0) { |
6144 | 0 | if (*tp > (ushort)X_SCHAR_MAX ) { |
6145 | | |
6146 | 0 | #ifdef ERANGE_FILL |
6147 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6148 | 0 | #endif |
6149 | 0 | status = NC_ERANGE; |
6150 | | |
6151 | 0 | #ifdef ERANGE_FILL |
6152 | 0 | xp++; tp++; continue; |
6153 | 0 | #endif |
6154 | 0 | } |
6155 | 0 | *xp++ = (schar) *tp++; /* type cast from ushort to schar */ |
6156 | 0 | } |
6157 | |
|
6158 | 0 | *xpp = (void *)xp; |
6159 | 0 | return status; |
6160 | 0 | } |
6161 | | |
6162 | | int |
6163 | | ncx_putn_schar_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
6164 | 0 | { |
6165 | 0 | int status = NC_NOERR; |
6166 | 0 | schar *xp = (schar *) *xpp; |
6167 | |
|
6168 | 0 | while (nelems-- != 0) { |
6169 | 0 | if (*tp > (uint)X_SCHAR_MAX ) { |
6170 | | |
6171 | 0 | #ifdef ERANGE_FILL |
6172 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6173 | 0 | #endif |
6174 | 0 | status = NC_ERANGE; |
6175 | | |
6176 | 0 | #ifdef ERANGE_FILL |
6177 | 0 | xp++; tp++; continue; |
6178 | 0 | #endif |
6179 | 0 | } |
6180 | 0 | *xp++ = (schar) *tp++; /* type cast from uint to schar */ |
6181 | 0 | } |
6182 | |
|
6183 | 0 | *xpp = (void *)xp; |
6184 | 0 | return status; |
6185 | 0 | } |
6186 | | |
6187 | | int |
6188 | | ncx_putn_schar_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
6189 | 0 | { |
6190 | 0 | int status = NC_NOERR; |
6191 | 0 | schar *xp = (schar *) *xpp; |
6192 | |
|
6193 | 0 | while (nelems-- != 0) { |
6194 | 0 | if (*tp > (ulonglong)X_SCHAR_MAX ) { |
6195 | | |
6196 | 0 | #ifdef ERANGE_FILL |
6197 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6198 | 0 | #endif |
6199 | 0 | status = NC_ERANGE; |
6200 | | |
6201 | 0 | #ifdef ERANGE_FILL |
6202 | 0 | xp++; tp++; continue; |
6203 | 0 | #endif |
6204 | 0 | } |
6205 | 0 | *xp++ = (schar) *tp++; /* type cast from ulonglong to schar */ |
6206 | 0 | } |
6207 | |
|
6208 | 0 | *xpp = (void *)xp; |
6209 | 0 | return status; |
6210 | 0 | } |
6211 | | |
6212 | | |
6213 | | int |
6214 | | ncx_pad_putn_schar_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
6215 | 0 | { |
6216 | 0 | size_t rndup = nelems % X_ALIGN; |
6217 | |
|
6218 | 0 | if (rndup) |
6219 | 0 | rndup = X_ALIGN - rndup; |
6220 | |
|
6221 | 0 | (void) memcpy(*xpp, tp, (size_t)nelems); |
6222 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
6223 | |
|
6224 | 0 | if (rndup) |
6225 | 0 | { |
6226 | 0 | (void) memcpy(*xpp, nada, (size_t)rndup); |
6227 | 0 | *xpp = (void *)((char *)(*xpp) + rndup); |
6228 | 0 | } |
6229 | |
|
6230 | 0 | return NC_NOERR; |
6231 | |
|
6232 | 0 | } |
6233 | | int |
6234 | | ncx_pad_putn_schar_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
6235 | 0 | { |
6236 | 0 | int status = NC_NOERR; |
6237 | 0 | size_t rndup = nelems % X_ALIGN; |
6238 | 0 | schar *xp = (schar *) *xpp; |
6239 | |
|
6240 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6241 | |
|
6242 | 0 | while (nelems-- != 0) { |
6243 | 0 | if (*tp > (uchar)X_SCHAR_MAX ) { |
6244 | | |
6245 | 0 | #ifdef ERANGE_FILL |
6246 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6247 | 0 | #endif |
6248 | 0 | status = NC_ERANGE; |
6249 | | |
6250 | 0 | #ifdef ERANGE_FILL |
6251 | 0 | xp++; tp++; continue; |
6252 | 0 | #endif |
6253 | 0 | } |
6254 | 0 | *xp++ = (schar) *tp++; /* type cast from uchar to schar */ |
6255 | 0 | } |
6256 | | |
6257 | |
|
6258 | 0 | if (rndup) { |
6259 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6260 | 0 | xp += rndup; |
6261 | 0 | } |
6262 | |
|
6263 | 0 | *xpp = (void *)xp; |
6264 | 0 | return status; |
6265 | 0 | } |
6266 | | |
6267 | | int |
6268 | | ncx_pad_putn_schar_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
6269 | 0 | { |
6270 | 0 | int status = NC_NOERR; |
6271 | 0 | size_t rndup = nelems % X_ALIGN; |
6272 | 0 | schar *xp = (schar *) *xpp; |
6273 | |
|
6274 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6275 | |
|
6276 | 0 | while (nelems-- != 0) { |
6277 | 0 | if (*tp > (short)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6278 | | |
6279 | 0 | #ifdef ERANGE_FILL |
6280 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6281 | 0 | #endif |
6282 | 0 | status = NC_ERANGE; |
6283 | | |
6284 | 0 | #ifdef ERANGE_FILL |
6285 | 0 | xp++; tp++; continue; |
6286 | 0 | #endif |
6287 | 0 | } |
6288 | 0 | *xp++ = (schar) *tp++; /* type cast from short to schar */ |
6289 | 0 | } |
6290 | | |
6291 | |
|
6292 | 0 | if (rndup) { |
6293 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6294 | 0 | xp += rndup; |
6295 | 0 | } |
6296 | |
|
6297 | 0 | *xpp = (void *)xp; |
6298 | 0 | return status; |
6299 | 0 | } |
6300 | | |
6301 | | int |
6302 | | ncx_pad_putn_schar_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
6303 | 0 | { |
6304 | 0 | int status = NC_NOERR; |
6305 | 0 | size_t rndup = nelems % X_ALIGN; |
6306 | 0 | schar *xp = (schar *) *xpp; |
6307 | |
|
6308 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6309 | |
|
6310 | 0 | while (nelems-- != 0) { |
6311 | 0 | if (*tp > (int)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6312 | | |
6313 | 0 | #ifdef ERANGE_FILL |
6314 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6315 | 0 | #endif |
6316 | 0 | status = NC_ERANGE; |
6317 | | |
6318 | 0 | #ifdef ERANGE_FILL |
6319 | 0 | xp++; tp++; continue; |
6320 | 0 | #endif |
6321 | 0 | } |
6322 | 0 | *xp++ = (schar) *tp++; /* type cast from int to schar */ |
6323 | 0 | } |
6324 | | |
6325 | |
|
6326 | 0 | if (rndup) { |
6327 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6328 | 0 | xp += rndup; |
6329 | 0 | } |
6330 | |
|
6331 | 0 | *xpp = (void *)xp; |
6332 | 0 | return status; |
6333 | 0 | } |
6334 | | |
6335 | | int |
6336 | | ncx_pad_putn_schar_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
6337 | 0 | { |
6338 | 0 | int status = NC_NOERR; |
6339 | 0 | size_t rndup = nelems % X_ALIGN; |
6340 | 0 | schar *xp = (schar *) *xpp; |
6341 | |
|
6342 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6343 | |
|
6344 | 0 | while (nelems-- != 0) { |
6345 | 0 | if (*tp > (long)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6346 | | |
6347 | 0 | #ifdef ERANGE_FILL |
6348 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6349 | 0 | #endif |
6350 | 0 | status = NC_ERANGE; |
6351 | | |
6352 | 0 | #ifdef ERANGE_FILL |
6353 | 0 | xp++; tp++; continue; |
6354 | 0 | #endif |
6355 | 0 | } |
6356 | 0 | *xp++ = (schar) *tp++; /* type cast from long to schar */ |
6357 | 0 | } |
6358 | | |
6359 | |
|
6360 | 0 | if (rndup) { |
6361 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6362 | 0 | xp += rndup; |
6363 | 0 | } |
6364 | |
|
6365 | 0 | *xpp = (void *)xp; |
6366 | 0 | return status; |
6367 | 0 | } |
6368 | | |
6369 | | int |
6370 | | ncx_pad_putn_schar_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
6371 | 0 | { |
6372 | 0 | int status = NC_NOERR; |
6373 | 0 | size_t rndup = nelems % X_ALIGN; |
6374 | 0 | schar *xp = (schar *) *xpp; |
6375 | |
|
6376 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6377 | |
|
6378 | 0 | while (nelems-- != 0) { |
6379 | 0 | if (*tp > (float)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6380 | | |
6381 | 0 | #ifdef ERANGE_FILL |
6382 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6383 | 0 | #endif |
6384 | 0 | status = NC_ERANGE; |
6385 | | |
6386 | 0 | #ifdef ERANGE_FILL |
6387 | 0 | xp++; tp++; continue; |
6388 | 0 | #endif |
6389 | 0 | } |
6390 | 0 | *xp++ = (schar) *tp++; /* type cast from float to schar */ |
6391 | 0 | } |
6392 | | |
6393 | |
|
6394 | 0 | if (rndup) { |
6395 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6396 | 0 | xp += rndup; |
6397 | 0 | } |
6398 | |
|
6399 | 0 | *xpp = (void *)xp; |
6400 | 0 | return status; |
6401 | 0 | } |
6402 | | |
6403 | | int |
6404 | | ncx_pad_putn_schar_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
6405 | 0 | { |
6406 | 0 | int status = NC_NOERR; |
6407 | 0 | size_t rndup = nelems % X_ALIGN; |
6408 | 0 | schar *xp = (schar *) *xpp; |
6409 | |
|
6410 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6411 | |
|
6412 | 0 | while (nelems-- != 0) { |
6413 | 0 | if (*tp > (double)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6414 | | |
6415 | 0 | #ifdef ERANGE_FILL |
6416 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6417 | 0 | #endif |
6418 | 0 | status = NC_ERANGE; |
6419 | | |
6420 | 0 | #ifdef ERANGE_FILL |
6421 | 0 | xp++; tp++; continue; |
6422 | 0 | #endif |
6423 | 0 | } |
6424 | 0 | *xp++ = (schar) *tp++; /* type cast from double to schar */ |
6425 | 0 | } |
6426 | | |
6427 | |
|
6428 | 0 | if (rndup) { |
6429 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6430 | 0 | xp += rndup; |
6431 | 0 | } |
6432 | |
|
6433 | 0 | *xpp = (void *)xp; |
6434 | 0 | return status; |
6435 | 0 | } |
6436 | | |
6437 | | int |
6438 | | ncx_pad_putn_schar_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
6439 | 0 | { |
6440 | 0 | int status = NC_NOERR; |
6441 | 0 | size_t rndup = nelems % X_ALIGN; |
6442 | 0 | schar *xp = (schar *) *xpp; |
6443 | |
|
6444 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6445 | |
|
6446 | 0 | while (nelems-- != 0) { |
6447 | 0 | if (*tp > (longlong)X_SCHAR_MAX || *tp < X_SCHAR_MIN) { |
6448 | | |
6449 | 0 | #ifdef ERANGE_FILL |
6450 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6451 | 0 | #endif |
6452 | 0 | status = NC_ERANGE; |
6453 | | |
6454 | 0 | #ifdef ERANGE_FILL |
6455 | 0 | xp++; tp++; continue; |
6456 | 0 | #endif |
6457 | 0 | } |
6458 | 0 | *xp++ = (schar) *tp++; /* type cast from longlong to schar */ |
6459 | 0 | } |
6460 | | |
6461 | |
|
6462 | 0 | if (rndup) { |
6463 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6464 | 0 | xp += rndup; |
6465 | 0 | } |
6466 | |
|
6467 | 0 | *xpp = (void *)xp; |
6468 | 0 | return status; |
6469 | 0 | } |
6470 | | |
6471 | | int |
6472 | | ncx_pad_putn_schar_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
6473 | 0 | { |
6474 | 0 | int status = NC_NOERR; |
6475 | 0 | size_t rndup = nelems % X_ALIGN; |
6476 | 0 | schar *xp = (schar *) *xpp; |
6477 | |
|
6478 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6479 | |
|
6480 | 0 | while (nelems-- != 0) { |
6481 | 0 | if (*tp > (ushort)X_SCHAR_MAX ) { |
6482 | | |
6483 | 0 | #ifdef ERANGE_FILL |
6484 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6485 | 0 | #endif |
6486 | 0 | status = NC_ERANGE; |
6487 | | |
6488 | 0 | #ifdef ERANGE_FILL |
6489 | 0 | xp++; tp++; continue; |
6490 | 0 | #endif |
6491 | 0 | } |
6492 | 0 | *xp++ = (schar) *tp++; /* type cast from ushort to schar */ |
6493 | 0 | } |
6494 | | |
6495 | |
|
6496 | 0 | if (rndup) { |
6497 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6498 | 0 | xp += rndup; |
6499 | 0 | } |
6500 | |
|
6501 | 0 | *xpp = (void *)xp; |
6502 | 0 | return status; |
6503 | 0 | } |
6504 | | |
6505 | | int |
6506 | | ncx_pad_putn_schar_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
6507 | 0 | { |
6508 | 0 | int status = NC_NOERR; |
6509 | 0 | size_t rndup = nelems % X_ALIGN; |
6510 | 0 | schar *xp = (schar *) *xpp; |
6511 | |
|
6512 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6513 | |
|
6514 | 0 | while (nelems-- != 0) { |
6515 | 0 | if (*tp > (uint)X_SCHAR_MAX ) { |
6516 | | |
6517 | 0 | #ifdef ERANGE_FILL |
6518 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6519 | 0 | #endif |
6520 | 0 | status = NC_ERANGE; |
6521 | | |
6522 | 0 | #ifdef ERANGE_FILL |
6523 | 0 | xp++; tp++; continue; |
6524 | 0 | #endif |
6525 | 0 | } |
6526 | 0 | *xp++ = (schar) *tp++; /* type cast from uint to schar */ |
6527 | 0 | } |
6528 | | |
6529 | |
|
6530 | 0 | if (rndup) { |
6531 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6532 | 0 | xp += rndup; |
6533 | 0 | } |
6534 | |
|
6535 | 0 | *xpp = (void *)xp; |
6536 | 0 | return status; |
6537 | 0 | } |
6538 | | |
6539 | | int |
6540 | | ncx_pad_putn_schar_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
6541 | 0 | { |
6542 | 0 | int status = NC_NOERR; |
6543 | 0 | size_t rndup = nelems % X_ALIGN; |
6544 | 0 | schar *xp = (schar *) *xpp; |
6545 | |
|
6546 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6547 | |
|
6548 | 0 | while (nelems-- != 0) { |
6549 | 0 | if (*tp > (ulonglong)X_SCHAR_MAX ) { |
6550 | | |
6551 | 0 | #ifdef ERANGE_FILL |
6552 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6553 | 0 | #endif |
6554 | 0 | status = NC_ERANGE; |
6555 | | |
6556 | 0 | #ifdef ERANGE_FILL |
6557 | 0 | xp++; tp++; continue; |
6558 | 0 | #endif |
6559 | 0 | } |
6560 | 0 | *xp++ = (schar) *tp++; /* type cast from ulonglong to schar */ |
6561 | 0 | } |
6562 | | |
6563 | |
|
6564 | 0 | if (rndup) { |
6565 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
6566 | 0 | xp += rndup; |
6567 | 0 | } |
6568 | |
|
6569 | 0 | *xpp = (void *)xp; |
6570 | 0 | return status; |
6571 | 0 | } |
6572 | | |
6573 | | |
6574 | | |
6575 | | /* uchar ---------------------------------------------------------------------*/ |
6576 | | int |
6577 | | ncx_getn_uchar_schar(const void **xpp, size_t nelems, schar *tp) |
6578 | 0 | { |
6579 | 0 | int status = NC_NOERR; |
6580 | 0 | uchar *xp = (uchar *)(*xpp); |
6581 | |
|
6582 | 0 | while (nelems-- != 0) { |
6583 | 0 | if (*xp > SCHAR_MAX) { |
6584 | 0 | *tp = NC_FILL_BYTE; |
6585 | 0 | status = NC_ERANGE; |
6586 | | |
6587 | 0 | #ifdef ERANGE_FILL |
6588 | 0 | xp++; tp++; continue; |
6589 | 0 | #endif |
6590 | 0 | } |
6591 | 0 | *tp++ = (schar) *xp++; /* type cast from uchar to schar */ |
6592 | 0 | } |
6593 | |
|
6594 | 0 | *xpp = (const void *)xp; |
6595 | 0 | return status; |
6596 | 0 | } |
6597 | | int |
6598 | | ncx_getn_uchar_uchar(const void **xpp, size_t nelems, uchar *tp) |
6599 | 0 | { |
6600 | 0 | (void) memcpy(tp, *xpp, (size_t)nelems); |
6601 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
6602 | 0 | return NC_NOERR; |
6603 | |
|
6604 | 0 | } |
6605 | | int |
6606 | | ncx_getn_uchar_short(const void **xpp, size_t nelems, short *tp) |
6607 | 0 | { |
6608 | 0 | int status = NC_NOERR; |
6609 | 0 | uchar *xp = (uchar *)(*xpp); |
6610 | |
|
6611 | 0 | while (nelems-- != 0) { |
6612 | | |
6613 | 0 | *tp++ = (short) (*xp++); /* type cast from uchar to short */ |
6614 | 0 | } |
6615 | |
|
6616 | 0 | *xpp = (const void *)xp; |
6617 | 0 | return status; |
6618 | 0 | } |
6619 | | |
6620 | | int |
6621 | | ncx_getn_uchar_int(const void **xpp, size_t nelems, int *tp) |
6622 | 0 | { |
6623 | 0 | int status = NC_NOERR; |
6624 | 0 | uchar *xp = (uchar *)(*xpp); |
6625 | |
|
6626 | 0 | while (nelems-- != 0) { |
6627 | | |
6628 | 0 | *tp++ = (int) (*xp++); /* type cast from uchar to int */ |
6629 | 0 | } |
6630 | |
|
6631 | 0 | *xpp = (const void *)xp; |
6632 | 0 | return status; |
6633 | 0 | } |
6634 | | |
6635 | | int |
6636 | | ncx_getn_uchar_long(const void **xpp, size_t nelems, long *tp) |
6637 | 0 | { |
6638 | 0 | int status = NC_NOERR; |
6639 | 0 | uchar *xp = (uchar *)(*xpp); |
6640 | |
|
6641 | 0 | while (nelems-- != 0) { |
6642 | | |
6643 | 0 | *tp++ = (long) (*xp++); /* type cast from uchar to long */ |
6644 | 0 | } |
6645 | |
|
6646 | 0 | *xpp = (const void *)xp; |
6647 | 0 | return status; |
6648 | 0 | } |
6649 | | |
6650 | | int |
6651 | | ncx_getn_uchar_float(const void **xpp, size_t nelems, float *tp) |
6652 | 0 | { |
6653 | 0 | int status = NC_NOERR; |
6654 | 0 | uchar *xp = (uchar *)(*xpp); |
6655 | |
|
6656 | 0 | while (nelems-- != 0) { |
6657 | | |
6658 | 0 | *tp++ = (float) (*xp++); /* type cast from uchar to float */ |
6659 | 0 | } |
6660 | |
|
6661 | 0 | *xpp = (const void *)xp; |
6662 | 0 | return status; |
6663 | 0 | } |
6664 | | |
6665 | | int |
6666 | | ncx_getn_uchar_double(const void **xpp, size_t nelems, double *tp) |
6667 | 0 | { |
6668 | 0 | int status = NC_NOERR; |
6669 | 0 | uchar *xp = (uchar *)(*xpp); |
6670 | |
|
6671 | 0 | while (nelems-- != 0) { |
6672 | | |
6673 | 0 | *tp++ = (double) (*xp++); /* type cast from uchar to double */ |
6674 | 0 | } |
6675 | |
|
6676 | 0 | *xpp = (const void *)xp; |
6677 | 0 | return status; |
6678 | 0 | } |
6679 | | |
6680 | | int |
6681 | | ncx_getn_uchar_longlong(const void **xpp, size_t nelems, longlong *tp) |
6682 | 0 | { |
6683 | 0 | int status = NC_NOERR; |
6684 | 0 | uchar *xp = (uchar *)(*xpp); |
6685 | |
|
6686 | 0 | while (nelems-- != 0) { |
6687 | | |
6688 | 0 | *tp++ = (longlong) (*xp++); /* type cast from uchar to longlong */ |
6689 | 0 | } |
6690 | |
|
6691 | 0 | *xpp = (const void *)xp; |
6692 | 0 | return status; |
6693 | 0 | } |
6694 | | |
6695 | | int |
6696 | | ncx_getn_uchar_ushort(const void **xpp, size_t nelems, ushort *tp) |
6697 | 0 | { |
6698 | 0 | int status = NC_NOERR; |
6699 | 0 | uchar *xp = (uchar *)(*xpp); |
6700 | |
|
6701 | 0 | while (nelems-- != 0) { |
6702 | | |
6703 | 0 | *tp++ = (ushort) (*xp++); /* type cast from uchar to ushort */ |
6704 | 0 | } |
6705 | |
|
6706 | 0 | *xpp = (const void *)xp; |
6707 | 0 | return status; |
6708 | 0 | } |
6709 | | |
6710 | | int |
6711 | | ncx_getn_uchar_uint(const void **xpp, size_t nelems, uint *tp) |
6712 | 0 | { |
6713 | 0 | int status = NC_NOERR; |
6714 | 0 | uchar *xp = (uchar *)(*xpp); |
6715 | |
|
6716 | 0 | while (nelems-- != 0) { |
6717 | | |
6718 | 0 | *tp++ = (uint) (*xp++); /* type cast from uchar to uint */ |
6719 | 0 | } |
6720 | |
|
6721 | 0 | *xpp = (const void *)xp; |
6722 | 0 | return status; |
6723 | 0 | } |
6724 | | |
6725 | | int |
6726 | | ncx_getn_uchar_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
6727 | 0 | { |
6728 | 0 | int status = NC_NOERR; |
6729 | 0 | uchar *xp = (uchar *)(*xpp); |
6730 | |
|
6731 | 0 | while (nelems-- != 0) { |
6732 | | |
6733 | 0 | *tp++ = (ulonglong) (*xp++); /* type cast from uchar to ulonglong */ |
6734 | 0 | } |
6735 | |
|
6736 | 0 | *xpp = (const void *)xp; |
6737 | 0 | return status; |
6738 | 0 | } |
6739 | | |
6740 | | |
6741 | | int |
6742 | | ncx_pad_getn_uchar_schar(const void **xpp, size_t nelems, schar *tp) |
6743 | 0 | { |
6744 | 0 | int status = NC_NOERR; |
6745 | 0 | size_t rndup = nelems % X_ALIGN; |
6746 | 0 | uchar *xp = (uchar *) *xpp; |
6747 | |
|
6748 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
6749 | |
|
6750 | 0 | while (nelems-- != 0) { |
6751 | 0 | if (*xp > SCHAR_MAX) { |
6752 | 0 | *tp = NC_FILL_BYTE; |
6753 | 0 | status = NC_ERANGE; |
6754 | | |
6755 | 0 | #ifdef ERANGE_FILL |
6756 | 0 | xp++; tp++; continue; |
6757 | 0 | #endif |
6758 | 0 | } |
6759 | 0 | *tp++ = (schar) *xp++; /* type cast from uchar to schar */ |
6760 | 0 | } |
6761 | |
|
6762 | 0 | *xpp = (void *)(xp + rndup); |
6763 | 0 | return status; |
6764 | 0 | } |
6765 | | int |
6766 | | ncx_pad_getn_uchar_uchar(const void **xpp, size_t nelems, uchar *tp) |
6767 | 0 | { |
6768 | 0 | size_t rndup = nelems % X_ALIGN; |
6769 | |
|
6770 | 0 | if (rndup) |
6771 | 0 | rndup = X_ALIGN - rndup; |
6772 | |
|
6773 | 0 | (void) memcpy(tp, *xpp, (size_t)nelems); |
6774 | 0 | *xpp = (void *)((char *)(*xpp) + nelems + rndup); |
6775 | |
|
6776 | 0 | return NC_NOERR; |
6777 | |
|
6778 | 0 | } |
6779 | | int |
6780 | | ncx_pad_getn_uchar_short(const void **xpp, size_t nelems, short *tp) |
6781 | 0 | { |
6782 | 0 | int status = NC_NOERR; |
6783 | 0 | size_t rndup = nelems % X_ALIGN; |
6784 | 0 | uchar *xp = (uchar *) *xpp; |
6785 | |
|
6786 | 0 | if (rndup) |
6787 | 0 | rndup = X_ALIGN - rndup; |
6788 | |
|
6789 | 0 | while (nelems-- != 0) { |
6790 | | |
6791 | 0 | *tp++ = (short) (*xp++); /* type cast from uchar to short */ |
6792 | 0 | } |
6793 | |
|
6794 | 0 | *xpp = (void *)(xp + rndup); |
6795 | 0 | return status; |
6796 | 0 | } |
6797 | | |
6798 | | int |
6799 | | ncx_pad_getn_uchar_int(const void **xpp, size_t nelems, int *tp) |
6800 | 0 | { |
6801 | 0 | int status = NC_NOERR; |
6802 | 0 | size_t rndup = nelems % X_ALIGN; |
6803 | 0 | uchar *xp = (uchar *) *xpp; |
6804 | |
|
6805 | 0 | if (rndup) |
6806 | 0 | rndup = X_ALIGN - rndup; |
6807 | |
|
6808 | 0 | while (nelems-- != 0) { |
6809 | | |
6810 | 0 | *tp++ = (int) (*xp++); /* type cast from uchar to int */ |
6811 | 0 | } |
6812 | |
|
6813 | 0 | *xpp = (void *)(xp + rndup); |
6814 | 0 | return status; |
6815 | 0 | } |
6816 | | |
6817 | | int |
6818 | | ncx_pad_getn_uchar_long(const void **xpp, size_t nelems, long *tp) |
6819 | 0 | { |
6820 | 0 | int status = NC_NOERR; |
6821 | 0 | size_t rndup = nelems % X_ALIGN; |
6822 | 0 | uchar *xp = (uchar *) *xpp; |
6823 | |
|
6824 | 0 | if (rndup) |
6825 | 0 | rndup = X_ALIGN - rndup; |
6826 | |
|
6827 | 0 | while (nelems-- != 0) { |
6828 | | |
6829 | 0 | *tp++ = (long) (*xp++); /* type cast from uchar to long */ |
6830 | 0 | } |
6831 | |
|
6832 | 0 | *xpp = (void *)(xp + rndup); |
6833 | 0 | return status; |
6834 | 0 | } |
6835 | | |
6836 | | int |
6837 | | ncx_pad_getn_uchar_float(const void **xpp, size_t nelems, float *tp) |
6838 | 0 | { |
6839 | 0 | int status = NC_NOERR; |
6840 | 0 | size_t rndup = nelems % X_ALIGN; |
6841 | 0 | uchar *xp = (uchar *) *xpp; |
6842 | |
|
6843 | 0 | if (rndup) |
6844 | 0 | rndup = X_ALIGN - rndup; |
6845 | |
|
6846 | 0 | while (nelems-- != 0) { |
6847 | | |
6848 | 0 | *tp++ = (float) (*xp++); /* type cast from uchar to float */ |
6849 | 0 | } |
6850 | |
|
6851 | 0 | *xpp = (void *)(xp + rndup); |
6852 | 0 | return status; |
6853 | 0 | } |
6854 | | |
6855 | | int |
6856 | | ncx_pad_getn_uchar_double(const void **xpp, size_t nelems, double *tp) |
6857 | 0 | { |
6858 | 0 | int status = NC_NOERR; |
6859 | 0 | size_t rndup = nelems % X_ALIGN; |
6860 | 0 | uchar *xp = (uchar *) *xpp; |
6861 | |
|
6862 | 0 | if (rndup) |
6863 | 0 | rndup = X_ALIGN - rndup; |
6864 | |
|
6865 | 0 | while (nelems-- != 0) { |
6866 | | |
6867 | 0 | *tp++ = (double) (*xp++); /* type cast from uchar to double */ |
6868 | 0 | } |
6869 | |
|
6870 | 0 | *xpp = (void *)(xp + rndup); |
6871 | 0 | return status; |
6872 | 0 | } |
6873 | | |
6874 | | int |
6875 | | ncx_pad_getn_uchar_longlong(const void **xpp, size_t nelems, longlong *tp) |
6876 | 0 | { |
6877 | 0 | int status = NC_NOERR; |
6878 | 0 | size_t rndup = nelems % X_ALIGN; |
6879 | 0 | uchar *xp = (uchar *) *xpp; |
6880 | |
|
6881 | 0 | if (rndup) |
6882 | 0 | rndup = X_ALIGN - rndup; |
6883 | |
|
6884 | 0 | while (nelems-- != 0) { |
6885 | | |
6886 | 0 | *tp++ = (longlong) (*xp++); /* type cast from uchar to longlong */ |
6887 | 0 | } |
6888 | |
|
6889 | 0 | *xpp = (void *)(xp + rndup); |
6890 | 0 | return status; |
6891 | 0 | } |
6892 | | |
6893 | | int |
6894 | | ncx_pad_getn_uchar_ushort(const void **xpp, size_t nelems, ushort *tp) |
6895 | 0 | { |
6896 | 0 | int status = NC_NOERR; |
6897 | 0 | size_t rndup = nelems % X_ALIGN; |
6898 | 0 | uchar *xp = (uchar *) *xpp; |
6899 | |
|
6900 | 0 | if (rndup) |
6901 | 0 | rndup = X_ALIGN - rndup; |
6902 | |
|
6903 | 0 | while (nelems-- != 0) { |
6904 | | |
6905 | 0 | *tp++ = (ushort) (*xp++); /* type cast from uchar to ushort */ |
6906 | 0 | } |
6907 | |
|
6908 | 0 | *xpp = (void *)(xp + rndup); |
6909 | 0 | return status; |
6910 | 0 | } |
6911 | | |
6912 | | int |
6913 | | ncx_pad_getn_uchar_uint(const void **xpp, size_t nelems, uint *tp) |
6914 | 0 | { |
6915 | 0 | int status = NC_NOERR; |
6916 | 0 | size_t rndup = nelems % X_ALIGN; |
6917 | 0 | uchar *xp = (uchar *) *xpp; |
6918 | |
|
6919 | 0 | if (rndup) |
6920 | 0 | rndup = X_ALIGN - rndup; |
6921 | |
|
6922 | 0 | while (nelems-- != 0) { |
6923 | | |
6924 | 0 | *tp++ = (uint) (*xp++); /* type cast from uchar to uint */ |
6925 | 0 | } |
6926 | |
|
6927 | 0 | *xpp = (void *)(xp + rndup); |
6928 | 0 | return status; |
6929 | 0 | } |
6930 | | |
6931 | | int |
6932 | | ncx_pad_getn_uchar_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
6933 | 0 | { |
6934 | 0 | int status = NC_NOERR; |
6935 | 0 | size_t rndup = nelems % X_ALIGN; |
6936 | 0 | uchar *xp = (uchar *) *xpp; |
6937 | |
|
6938 | 0 | if (rndup) |
6939 | 0 | rndup = X_ALIGN - rndup; |
6940 | |
|
6941 | 0 | while (nelems-- != 0) { |
6942 | | |
6943 | 0 | *tp++ = (ulonglong) (*xp++); /* type cast from uchar to ulonglong */ |
6944 | 0 | } |
6945 | |
|
6946 | 0 | *xpp = (void *)(xp + rndup); |
6947 | 0 | return status; |
6948 | 0 | } |
6949 | | |
6950 | | |
6951 | | int |
6952 | | ncx_putn_uchar_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
6953 | 0 | { |
6954 | 0 | int status = NC_NOERR; |
6955 | 0 | uchar *xp = (uchar *) *xpp; |
6956 | |
|
6957 | 0 | while (nelems-- != 0) { |
6958 | 0 | if (*tp < 0) { |
6959 | | |
6960 | 0 | #ifdef ERANGE_FILL |
6961 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6962 | 0 | #endif |
6963 | 0 | status = NC_ERANGE; |
6964 | | |
6965 | 0 | #ifdef ERANGE_FILL |
6966 | 0 | xp++; tp++; continue; |
6967 | 0 | #endif |
6968 | 0 | } |
6969 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from schar to uchar */ |
6970 | 0 | } |
6971 | |
|
6972 | 0 | *xpp = (void *)xp; |
6973 | 0 | return status; |
6974 | 0 | } |
6975 | | int |
6976 | | ncx_putn_uchar_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
6977 | 0 | { |
6978 | 0 | (void) memcpy(*xpp, tp, (size_t)nelems); |
6979 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
6980 | |
|
6981 | 0 | return NC_NOERR; |
6982 | |
|
6983 | 0 | } |
6984 | | int |
6985 | | ncx_putn_uchar_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
6986 | 0 | { |
6987 | 0 | int status = NC_NOERR; |
6988 | 0 | uchar *xp = (uchar *) *xpp; |
6989 | |
|
6990 | 0 | while (nelems-- != 0) { |
6991 | 0 | if (*tp > (short)X_UCHAR_MAX || *tp < 0) { |
6992 | | |
6993 | 0 | #ifdef ERANGE_FILL |
6994 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
6995 | 0 | #endif |
6996 | 0 | status = NC_ERANGE; |
6997 | | |
6998 | 0 | #ifdef ERANGE_FILL |
6999 | 0 | xp++; tp++; continue; |
7000 | 0 | #endif |
7001 | 0 | } |
7002 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from short to uchar */ |
7003 | 0 | } |
7004 | |
|
7005 | 0 | *xpp = (void *)xp; |
7006 | 0 | return status; |
7007 | 0 | } |
7008 | | |
7009 | | int |
7010 | | ncx_putn_uchar_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
7011 | 0 | { |
7012 | 0 | int status = NC_NOERR; |
7013 | 0 | uchar *xp = (uchar *) *xpp; |
7014 | |
|
7015 | 0 | while (nelems-- != 0) { |
7016 | 0 | if (*tp > (int)X_UCHAR_MAX || *tp < 0) { |
7017 | | |
7018 | 0 | #ifdef ERANGE_FILL |
7019 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7020 | 0 | #endif |
7021 | 0 | status = NC_ERANGE; |
7022 | | |
7023 | 0 | #ifdef ERANGE_FILL |
7024 | 0 | xp++; tp++; continue; |
7025 | 0 | #endif |
7026 | 0 | } |
7027 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from int to uchar */ |
7028 | 0 | } |
7029 | |
|
7030 | 0 | *xpp = (void *)xp; |
7031 | 0 | return status; |
7032 | 0 | } |
7033 | | |
7034 | | int |
7035 | | ncx_putn_uchar_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
7036 | 0 | { |
7037 | 0 | int status = NC_NOERR; |
7038 | 0 | uchar *xp = (uchar *) *xpp; |
7039 | |
|
7040 | 0 | while (nelems-- != 0) { |
7041 | 0 | if (*tp > (long)X_UCHAR_MAX || *tp < 0) { |
7042 | | |
7043 | 0 | #ifdef ERANGE_FILL |
7044 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7045 | 0 | #endif |
7046 | 0 | status = NC_ERANGE; |
7047 | | |
7048 | 0 | #ifdef ERANGE_FILL |
7049 | 0 | xp++; tp++; continue; |
7050 | 0 | #endif |
7051 | 0 | } |
7052 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from long to uchar */ |
7053 | 0 | } |
7054 | |
|
7055 | 0 | *xpp = (void *)xp; |
7056 | 0 | return status; |
7057 | 0 | } |
7058 | | |
7059 | | int |
7060 | | ncx_putn_uchar_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
7061 | 0 | { |
7062 | 0 | int status = NC_NOERR; |
7063 | 0 | uchar *xp = (uchar *) *xpp; |
7064 | |
|
7065 | 0 | while (nelems-- != 0) { |
7066 | 0 | if (*tp > (float)X_UCHAR_MAX || *tp < 0) { |
7067 | | |
7068 | 0 | #ifdef ERANGE_FILL |
7069 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7070 | 0 | #endif |
7071 | 0 | status = NC_ERANGE; |
7072 | | |
7073 | 0 | #ifdef ERANGE_FILL |
7074 | 0 | xp++; tp++; continue; |
7075 | 0 | #endif |
7076 | 0 | } |
7077 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from float to uchar */ |
7078 | 0 | } |
7079 | |
|
7080 | 0 | *xpp = (void *)xp; |
7081 | 0 | return status; |
7082 | 0 | } |
7083 | | |
7084 | | int |
7085 | | ncx_putn_uchar_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
7086 | 0 | { |
7087 | 0 | int status = NC_NOERR; |
7088 | 0 | uchar *xp = (uchar *) *xpp; |
7089 | |
|
7090 | 0 | while (nelems-- != 0) { |
7091 | 0 | if (*tp > (double)X_UCHAR_MAX || *tp < 0) { |
7092 | | |
7093 | 0 | #ifdef ERANGE_FILL |
7094 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7095 | 0 | #endif |
7096 | 0 | status = NC_ERANGE; |
7097 | | |
7098 | 0 | #ifdef ERANGE_FILL |
7099 | 0 | xp++; tp++; continue; |
7100 | 0 | #endif |
7101 | 0 | } |
7102 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from double to uchar */ |
7103 | 0 | } |
7104 | |
|
7105 | 0 | *xpp = (void *)xp; |
7106 | 0 | return status; |
7107 | 0 | } |
7108 | | |
7109 | | int |
7110 | | ncx_putn_uchar_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
7111 | 0 | { |
7112 | 0 | int status = NC_NOERR; |
7113 | 0 | uchar *xp = (uchar *) *xpp; |
7114 | |
|
7115 | 0 | while (nelems-- != 0) { |
7116 | 0 | if (*tp > (longlong)X_UCHAR_MAX || *tp < 0) { |
7117 | | |
7118 | 0 | #ifdef ERANGE_FILL |
7119 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7120 | 0 | #endif |
7121 | 0 | status = NC_ERANGE; |
7122 | | |
7123 | 0 | #ifdef ERANGE_FILL |
7124 | 0 | xp++; tp++; continue; |
7125 | 0 | #endif |
7126 | 0 | } |
7127 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from longlong to uchar */ |
7128 | 0 | } |
7129 | |
|
7130 | 0 | *xpp = (void *)xp; |
7131 | 0 | return status; |
7132 | 0 | } |
7133 | | |
7134 | | int |
7135 | | ncx_putn_uchar_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
7136 | 0 | { |
7137 | 0 | int status = NC_NOERR; |
7138 | 0 | uchar *xp = (uchar *) *xpp; |
7139 | |
|
7140 | 0 | while (nelems-- != 0) { |
7141 | 0 | if (*tp > (ushort)X_UCHAR_MAX ) { |
7142 | | |
7143 | 0 | #ifdef ERANGE_FILL |
7144 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7145 | 0 | #endif |
7146 | 0 | status = NC_ERANGE; |
7147 | | |
7148 | 0 | #ifdef ERANGE_FILL |
7149 | 0 | xp++; tp++; continue; |
7150 | 0 | #endif |
7151 | 0 | } |
7152 | 0 | *xp++ = (uchar) *tp++; /* type cast from ushort to uchar */ |
7153 | 0 | } |
7154 | |
|
7155 | 0 | *xpp = (void *)xp; |
7156 | 0 | return status; |
7157 | 0 | } |
7158 | | |
7159 | | int |
7160 | | ncx_putn_uchar_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
7161 | 0 | { |
7162 | 0 | int status = NC_NOERR; |
7163 | 0 | uchar *xp = (uchar *) *xpp; |
7164 | |
|
7165 | 0 | while (nelems-- != 0) { |
7166 | 0 | if (*tp > (uint)X_UCHAR_MAX ) { |
7167 | | |
7168 | 0 | #ifdef ERANGE_FILL |
7169 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7170 | 0 | #endif |
7171 | 0 | status = NC_ERANGE; |
7172 | | |
7173 | 0 | #ifdef ERANGE_FILL |
7174 | 0 | xp++; tp++; continue; |
7175 | 0 | #endif |
7176 | 0 | } |
7177 | 0 | *xp++ = (uchar) *tp++; /* type cast from uint to uchar */ |
7178 | 0 | } |
7179 | |
|
7180 | 0 | *xpp = (void *)xp; |
7181 | 0 | return status; |
7182 | 0 | } |
7183 | | |
7184 | | int |
7185 | | ncx_putn_uchar_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
7186 | 0 | { |
7187 | 0 | int status = NC_NOERR; |
7188 | 0 | uchar *xp = (uchar *) *xpp; |
7189 | |
|
7190 | 0 | while (nelems-- != 0) { |
7191 | 0 | if (*tp > (ulonglong)X_UCHAR_MAX ) { |
7192 | | |
7193 | 0 | #ifdef ERANGE_FILL |
7194 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7195 | 0 | #endif |
7196 | 0 | status = NC_ERANGE; |
7197 | | |
7198 | 0 | #ifdef ERANGE_FILL |
7199 | 0 | xp++; tp++; continue; |
7200 | 0 | #endif |
7201 | 0 | } |
7202 | 0 | *xp++ = (uchar) *tp++; /* type cast from ulonglong to uchar */ |
7203 | 0 | } |
7204 | |
|
7205 | 0 | *xpp = (void *)xp; |
7206 | 0 | return status; |
7207 | 0 | } |
7208 | | |
7209 | | |
7210 | | int |
7211 | | ncx_pad_putn_uchar_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
7212 | 0 | { |
7213 | 0 | int status = NC_NOERR; |
7214 | 0 | size_t rndup = nelems % X_ALIGN; |
7215 | 0 | uchar *xp = (uchar *) *xpp; |
7216 | |
|
7217 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7218 | |
|
7219 | 0 | while (nelems-- != 0) { |
7220 | 0 | if (*tp < 0) { |
7221 | | |
7222 | 0 | #ifdef ERANGE_FILL |
7223 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7224 | 0 | #endif |
7225 | 0 | status = NC_ERANGE; |
7226 | | |
7227 | 0 | #ifdef ERANGE_FILL |
7228 | 0 | xp++; tp++; continue; |
7229 | 0 | #endif |
7230 | 0 | } |
7231 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from schar to uchar */ |
7232 | 0 | } |
7233 | |
|
7234 | 0 | if (rndup) { |
7235 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7236 | 0 | xp += rndup; |
7237 | 0 | } |
7238 | |
|
7239 | 0 | *xpp = (void *)xp; |
7240 | 0 | return status; |
7241 | 0 | } |
7242 | | int |
7243 | | ncx_pad_putn_uchar_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
7244 | 0 | { |
7245 | 0 | size_t rndup = nelems % X_ALIGN; |
7246 | |
|
7247 | 0 | if (rndup) |
7248 | 0 | rndup = X_ALIGN - rndup; |
7249 | |
|
7250 | 0 | (void) memcpy(*xpp, tp, (size_t)nelems); |
7251 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
7252 | |
|
7253 | 0 | if (rndup) |
7254 | 0 | { |
7255 | 0 | (void) memcpy(*xpp, nada, (size_t)rndup); |
7256 | 0 | *xpp = (void *)((char *)(*xpp) + rndup); |
7257 | 0 | } |
7258 | |
|
7259 | 0 | return NC_NOERR; |
7260 | |
|
7261 | 0 | } |
7262 | | int |
7263 | | ncx_pad_putn_uchar_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
7264 | 0 | { |
7265 | 0 | int status = NC_NOERR; |
7266 | 0 | size_t rndup = nelems % X_ALIGN; |
7267 | 0 | uchar *xp = (uchar *) *xpp; |
7268 | |
|
7269 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7270 | |
|
7271 | 0 | while (nelems-- != 0) { |
7272 | 0 | if (*tp > (short)X_UCHAR_MAX || *tp < 0) { |
7273 | | |
7274 | 0 | #ifdef ERANGE_FILL |
7275 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7276 | 0 | #endif |
7277 | 0 | status = NC_ERANGE; |
7278 | | |
7279 | 0 | #ifdef ERANGE_FILL |
7280 | 0 | xp++; tp++; continue; |
7281 | 0 | #endif |
7282 | 0 | } |
7283 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from short to uchar */ |
7284 | 0 | } |
7285 | | |
7286 | |
|
7287 | 0 | if (rndup) { |
7288 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7289 | 0 | xp += rndup; |
7290 | 0 | } |
7291 | |
|
7292 | 0 | *xpp = (void *)xp; |
7293 | 0 | return status; |
7294 | 0 | } |
7295 | | |
7296 | | int |
7297 | | ncx_pad_putn_uchar_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
7298 | 0 | { |
7299 | 0 | int status = NC_NOERR; |
7300 | 0 | size_t rndup = nelems % X_ALIGN; |
7301 | 0 | uchar *xp = (uchar *) *xpp; |
7302 | |
|
7303 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7304 | |
|
7305 | 0 | while (nelems-- != 0) { |
7306 | 0 | if (*tp > (int)X_UCHAR_MAX || *tp < 0) { |
7307 | | |
7308 | 0 | #ifdef ERANGE_FILL |
7309 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7310 | 0 | #endif |
7311 | 0 | status = NC_ERANGE; |
7312 | | |
7313 | 0 | #ifdef ERANGE_FILL |
7314 | 0 | xp++; tp++; continue; |
7315 | 0 | #endif |
7316 | 0 | } |
7317 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from int to uchar */ |
7318 | 0 | } |
7319 | | |
7320 | |
|
7321 | 0 | if (rndup) { |
7322 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7323 | 0 | xp += rndup; |
7324 | 0 | } |
7325 | |
|
7326 | 0 | *xpp = (void *)xp; |
7327 | 0 | return status; |
7328 | 0 | } |
7329 | | |
7330 | | int |
7331 | | ncx_pad_putn_uchar_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
7332 | 0 | { |
7333 | 0 | int status = NC_NOERR; |
7334 | 0 | size_t rndup = nelems % X_ALIGN; |
7335 | 0 | uchar *xp = (uchar *) *xpp; |
7336 | |
|
7337 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7338 | |
|
7339 | 0 | while (nelems-- != 0) { |
7340 | 0 | if (*tp > (long)X_UCHAR_MAX || *tp < 0) { |
7341 | | |
7342 | 0 | #ifdef ERANGE_FILL |
7343 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7344 | 0 | #endif |
7345 | 0 | status = NC_ERANGE; |
7346 | | |
7347 | 0 | #ifdef ERANGE_FILL |
7348 | 0 | xp++; tp++; continue; |
7349 | 0 | #endif |
7350 | 0 | } |
7351 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from long to uchar */ |
7352 | 0 | } |
7353 | | |
7354 | |
|
7355 | 0 | if (rndup) { |
7356 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7357 | 0 | xp += rndup; |
7358 | 0 | } |
7359 | |
|
7360 | 0 | *xpp = (void *)xp; |
7361 | 0 | return status; |
7362 | 0 | } |
7363 | | |
7364 | | int |
7365 | | ncx_pad_putn_uchar_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
7366 | 0 | { |
7367 | 0 | int status = NC_NOERR; |
7368 | 0 | size_t rndup = nelems % X_ALIGN; |
7369 | 0 | uchar *xp = (uchar *) *xpp; |
7370 | |
|
7371 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7372 | |
|
7373 | 0 | while (nelems-- != 0) { |
7374 | 0 | if (*tp > (float)X_UCHAR_MAX || *tp < 0) { |
7375 | | |
7376 | 0 | #ifdef ERANGE_FILL |
7377 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7378 | 0 | #endif |
7379 | 0 | status = NC_ERANGE; |
7380 | | |
7381 | 0 | #ifdef ERANGE_FILL |
7382 | 0 | xp++; tp++; continue; |
7383 | 0 | #endif |
7384 | 0 | } |
7385 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from float to uchar */ |
7386 | 0 | } |
7387 | | |
7388 | |
|
7389 | 0 | if (rndup) { |
7390 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7391 | 0 | xp += rndup; |
7392 | 0 | } |
7393 | |
|
7394 | 0 | *xpp = (void *)xp; |
7395 | 0 | return status; |
7396 | 0 | } |
7397 | | |
7398 | | int |
7399 | | ncx_pad_putn_uchar_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
7400 | 0 | { |
7401 | 0 | int status = NC_NOERR; |
7402 | 0 | size_t rndup = nelems % X_ALIGN; |
7403 | 0 | uchar *xp = (uchar *) *xpp; |
7404 | |
|
7405 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7406 | |
|
7407 | 0 | while (nelems-- != 0) { |
7408 | 0 | if (*tp > (double)X_UCHAR_MAX || *tp < 0) { |
7409 | | |
7410 | 0 | #ifdef ERANGE_FILL |
7411 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7412 | 0 | #endif |
7413 | 0 | status = NC_ERANGE; |
7414 | | |
7415 | 0 | #ifdef ERANGE_FILL |
7416 | 0 | xp++; tp++; continue; |
7417 | 0 | #endif |
7418 | 0 | } |
7419 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from double to uchar */ |
7420 | 0 | } |
7421 | | |
7422 | |
|
7423 | 0 | if (rndup) { |
7424 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7425 | 0 | xp += rndup; |
7426 | 0 | } |
7427 | |
|
7428 | 0 | *xpp = (void *)xp; |
7429 | 0 | return status; |
7430 | 0 | } |
7431 | | |
7432 | | int |
7433 | | ncx_pad_putn_uchar_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
7434 | 0 | { |
7435 | 0 | int status = NC_NOERR; |
7436 | 0 | size_t rndup = nelems % X_ALIGN; |
7437 | 0 | uchar *xp = (uchar *) *xpp; |
7438 | |
|
7439 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7440 | |
|
7441 | 0 | while (nelems-- != 0) { |
7442 | 0 | if (*tp > (longlong)X_UCHAR_MAX || *tp < 0) { |
7443 | | |
7444 | 0 | #ifdef ERANGE_FILL |
7445 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7446 | 0 | #endif |
7447 | 0 | status = NC_ERANGE; |
7448 | | |
7449 | 0 | #ifdef ERANGE_FILL |
7450 | 0 | xp++; tp++; continue; |
7451 | 0 | #endif |
7452 | 0 | } |
7453 | 0 | *xp++ = (uchar) (signed) *tp++; /* type cast from longlong to uchar */ |
7454 | 0 | } |
7455 | | |
7456 | |
|
7457 | 0 | if (rndup) { |
7458 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7459 | 0 | xp += rndup; |
7460 | 0 | } |
7461 | |
|
7462 | 0 | *xpp = (void *)xp; |
7463 | 0 | return status; |
7464 | 0 | } |
7465 | | |
7466 | | int |
7467 | | ncx_pad_putn_uchar_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
7468 | 0 | { |
7469 | 0 | int status = NC_NOERR; |
7470 | 0 | size_t rndup = nelems % X_ALIGN; |
7471 | 0 | uchar *xp = (uchar *) *xpp; |
7472 | |
|
7473 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7474 | |
|
7475 | 0 | while (nelems-- != 0) { |
7476 | 0 | if (*tp > (ushort)X_UCHAR_MAX ) { |
7477 | | |
7478 | 0 | #ifdef ERANGE_FILL |
7479 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7480 | 0 | #endif |
7481 | 0 | status = NC_ERANGE; |
7482 | | |
7483 | 0 | #ifdef ERANGE_FILL |
7484 | 0 | xp++; tp++; continue; |
7485 | 0 | #endif |
7486 | 0 | } |
7487 | 0 | *xp++ = (uchar) *tp++; /* type cast from ushort to uchar */ |
7488 | 0 | } |
7489 | | |
7490 | |
|
7491 | 0 | if (rndup) { |
7492 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7493 | 0 | xp += rndup; |
7494 | 0 | } |
7495 | |
|
7496 | 0 | *xpp = (void *)xp; |
7497 | 0 | return status; |
7498 | 0 | } |
7499 | | |
7500 | | int |
7501 | | ncx_pad_putn_uchar_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
7502 | 0 | { |
7503 | 0 | int status = NC_NOERR; |
7504 | 0 | size_t rndup = nelems % X_ALIGN; |
7505 | 0 | uchar *xp = (uchar *) *xpp; |
7506 | |
|
7507 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7508 | |
|
7509 | 0 | while (nelems-- != 0) { |
7510 | 0 | if (*tp > (uint)X_UCHAR_MAX ) { |
7511 | | |
7512 | 0 | #ifdef ERANGE_FILL |
7513 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7514 | 0 | #endif |
7515 | 0 | status = NC_ERANGE; |
7516 | | |
7517 | 0 | #ifdef ERANGE_FILL |
7518 | 0 | xp++; tp++; continue; |
7519 | 0 | #endif |
7520 | 0 | } |
7521 | 0 | *xp++ = (uchar) *tp++; /* type cast from uint to uchar */ |
7522 | 0 | } |
7523 | | |
7524 | |
|
7525 | 0 | if (rndup) { |
7526 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7527 | 0 | xp += rndup; |
7528 | 0 | } |
7529 | |
|
7530 | 0 | *xpp = (void *)xp; |
7531 | 0 | return status; |
7532 | 0 | } |
7533 | | |
7534 | | int |
7535 | | ncx_pad_putn_uchar_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
7536 | 0 | { |
7537 | 0 | int status = NC_NOERR; |
7538 | 0 | size_t rndup = nelems % X_ALIGN; |
7539 | 0 | uchar *xp = (uchar *) *xpp; |
7540 | |
|
7541 | 0 | if (rndup) rndup = X_ALIGN - rndup; |
7542 | |
|
7543 | 0 | while (nelems-- != 0) { |
7544 | 0 | if (*tp > (ulonglong)X_UCHAR_MAX ) { |
7545 | | |
7546 | 0 | #ifdef ERANGE_FILL |
7547 | 0 | if (fillp != NULL) memcpy(xp, fillp, 1); |
7548 | 0 | #endif |
7549 | 0 | status = NC_ERANGE; |
7550 | | |
7551 | 0 | #ifdef ERANGE_FILL |
7552 | 0 | xp++; tp++; continue; |
7553 | 0 | #endif |
7554 | 0 | } |
7555 | 0 | *xp++ = (uchar) *tp++; /* type cast from ulonglong to uchar */ |
7556 | 0 | } |
7557 | | |
7558 | |
|
7559 | 0 | if (rndup) { |
7560 | 0 | (void) memcpy(xp, nada, (size_t)rndup); |
7561 | 0 | xp += rndup; |
7562 | 0 | } |
7563 | |
|
7564 | 0 | *xpp = (void *)xp; |
7565 | 0 | return status; |
7566 | 0 | } |
7567 | | |
7568 | | |
7569 | | /* short ---------------------------------------------------------------------*/ |
7570 | | |
7571 | | #if X_SIZEOF_SHORT == SIZEOF_SHORT |
7572 | | /* optimized version */ |
7573 | | int |
7574 | | ncx_getn_short_short(const void **xpp, size_t nelems, short *tp) |
7575 | 0 | { |
7576 | | #ifdef WORDS_BIGENDIAN |
7577 | | (void) memcpy(tp, *xpp, (size_t)nelems * SIZEOF_SHORT); |
7578 | | # else |
7579 | 0 | swapn2b(tp, *xpp, nelems); |
7580 | 0 | # endif |
7581 | 0 | *xpp = (const void *)((const char *)(*xpp) + nelems * X_SIZEOF_SHORT); |
7582 | 0 | return NC_NOERR; |
7583 | 0 | } |
7584 | | #else |
7585 | | int |
7586 | | ncx_getn_short_short(const void **xpp, size_t nelems, short *tp) |
7587 | | { |
7588 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
7589 | | |
7590 | | /* basic algorithm is: |
7591 | | * - ensure sane alignment of input data |
7592 | | * - copy (conversion happens automatically) input data |
7593 | | * to output |
7594 | | * - update xpp to point at next unconverted input, and tp to point |
7595 | | * at next location for converted output |
7596 | | */ |
7597 | | long i, j, ni; |
7598 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
7599 | | short *xp; |
7600 | | int nrange = 0; /* number of range errors */ |
7601 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
7602 | | long cxp = (long) *((char**)xpp); |
7603 | | |
7604 | | realign = (cxp & 7) % SIZEOF_SHORT; |
7605 | | /* sjl: manually stripmine so we can limit amount of |
7606 | | * vector work space reserved to LOOPCNT elements. Also |
7607 | | * makes vectorisation easy */ |
7608 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
7609 | | ni=Min(nelems-j,LOOPCNT); |
7610 | | if (realign) { |
7611 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
7612 | | xp = tmp; |
7613 | | } else { |
7614 | | xp = (short *) *xpp; |
7615 | | } |
7616 | | /* copy the next block */ |
7617 | | #pragma cdir loopcnt=LOOPCNT |
7618 | | #pragma cdir shortloop |
7619 | | for (i=0; i<ni; i++) { |
7620 | | tp[i] = (short) Max( SHORT_MIN, Min(SHORT_MAX, (short) xp[i])); |
7621 | | /* test for range errors (not always needed but do it anyway) */ |
7622 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
7623 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
7624 | | nrange += xp[i] > SHORT_MAX || xp[i] < SHORT_MIN; |
7625 | | } |
7626 | | /* update xpp and tp */ |
7627 | | if (realign) xp = (short *) *xpp; |
7628 | | xp += ni; |
7629 | | tp += ni; |
7630 | | *xpp = (void*)xp; |
7631 | | } |
7632 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
7633 | | |
7634 | | #else /* not SX */ |
7635 | | const char *xp = (const char *) *xpp; |
7636 | | int status = NC_NOERR; |
7637 | | |
7638 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
7639 | | { |
7640 | | const int lstatus = ncx_get_short_short(xp, tp); |
7641 | | if (status == NC_NOERR) /* report the first encountered error */ |
7642 | | status = lstatus; |
7643 | | } |
7644 | | |
7645 | | *xpp = (const void *)xp; |
7646 | | return status; |
7647 | | #endif |
7648 | | } |
7649 | | |
7650 | | #endif |
7651 | | int |
7652 | | ncx_getn_short_schar(const void **xpp, size_t nelems, schar *tp) |
7653 | 0 | { |
7654 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
7655 | | |
7656 | | /* basic algorithm is: |
7657 | | * - ensure sane alignment of input data |
7658 | | * - copy (conversion happens automatically) input data |
7659 | | * to output |
7660 | | * - update xpp to point at next unconverted input, and tp to point |
7661 | | * at next location for converted output |
7662 | | */ |
7663 | | long i, j, ni; |
7664 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
7665 | | short *xp; |
7666 | | int nrange = 0; /* number of range errors */ |
7667 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
7668 | | long cxp = (long) *((char**)xpp); |
7669 | | |
7670 | | realign = (cxp & 7) % SIZEOF_SHORT; |
7671 | | /* sjl: manually stripmine so we can limit amount of |
7672 | | * vector work space reserved to LOOPCNT elements. Also |
7673 | | * makes vectorisation easy */ |
7674 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
7675 | | ni=Min(nelems-j,LOOPCNT); |
7676 | | if (realign) { |
7677 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
7678 | | xp = tmp; |
7679 | | } else { |
7680 | | xp = (short *) *xpp; |
7681 | | } |
7682 | | /* copy the next block */ |
7683 | | #pragma cdir loopcnt=LOOPCNT |
7684 | | #pragma cdir shortloop |
7685 | | for (i=0; i<ni; i++) { |
7686 | | tp[i] = (schar) Max( SCHAR_MIN, Min(SCHAR_MAX, (schar) xp[i])); |
7687 | | /* test for range errors (not always needed but do it anyway) */ |
7688 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
7689 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
7690 | | nrange += xp[i] > SCHAR_MAX || xp[i] < SCHAR_MIN; |
7691 | | } |
7692 | | /* update xpp and tp */ |
7693 | | if (realign) xp = (short *) *xpp; |
7694 | | xp += ni; |
7695 | | tp += ni; |
7696 | | *xpp = (void*)xp; |
7697 | | } |
7698 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
7699 | | |
7700 | | #else /* not SX */ |
7701 | 0 | const char *xp = (const char *) *xpp; |
7702 | 0 | int status = NC_NOERR; |
7703 | |
|
7704 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
7705 | 0 | { |
7706 | 0 | const int lstatus = ncx_get_short_schar(xp, tp); |
7707 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
7708 | 0 | status = lstatus; |
7709 | 0 | } |
7710 | |
|
7711 | 0 | *xpp = (const void *)xp; |
7712 | 0 | return status; |
7713 | 0 | #endif |
7714 | 0 | } |
7715 | | |
7716 | | int |
7717 | | ncx_getn_short_int(const void **xpp, size_t nelems, int *tp) |
7718 | 0 | { |
7719 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
7720 | | |
7721 | | /* basic algorithm is: |
7722 | | * - ensure sane alignment of input data |
7723 | | * - copy (conversion happens automatically) input data |
7724 | | * to output |
7725 | | * - update xpp to point at next unconverted input, and tp to point |
7726 | | * at next location for converted output |
7727 | | */ |
7728 | | long i, j, ni; |
7729 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
7730 | | short *xp; |
7731 | | int nrange = 0; /* number of range errors */ |
7732 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
7733 | | long cxp = (long) *((char**)xpp); |
7734 | | |
7735 | | realign = (cxp & 7) % SIZEOF_SHORT; |
7736 | | /* sjl: manually stripmine so we can limit amount of |
7737 | | * vector work space reserved to LOOPCNT elements. Also |
7738 | | * makes vectorisation easy */ |
7739 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
7740 | | ni=Min(nelems-j,LOOPCNT); |
7741 | | if (realign) { |
7742 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
7743 | | xp = tmp; |
7744 | | } else { |
7745 | | xp = (short *) *xpp; |
7746 | | } |
7747 | | /* copy the next block */ |
7748 | | #pragma cdir loopcnt=LOOPCNT |
7749 | | #pragma cdir shortloop |
7750 | | for (i=0; i<ni; i++) { |
7751 | | tp[i] = (int) Max( INT_MIN, Min(INT_MAX, (int) xp[i])); |
7752 | | /* test for range errors (not always needed but do it anyway) */ |
7753 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
7754 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
7755 | | nrange += xp[i] > INT_MAX || xp[i] < INT_MIN; |
7756 | | } |
7757 | | /* update xpp and tp */ |
7758 | | if (realign) xp = (short *) *xpp; |
7759 | | xp += ni; |
7760 | | tp += ni; |
7761 | | *xpp = (void*)xp; |
7762 | | } |
7763 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
7764 | | |
7765 | | #else /* not SX */ |
7766 | 0 | const char *xp = (const char *) *xpp; |
7767 | 0 | int status = NC_NOERR; |
7768 | |
|
7769 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
7770 | 0 | { |
7771 | 0 | const int lstatus = ncx_get_short_int(xp, tp); |
7772 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
7773 | 0 | status = lstatus; |
7774 | 0 | } |
7775 | |
|
7776 | 0 | *xpp = (const void *)xp; |
7777 | 0 | return status; |
7778 | 0 | #endif |
7779 | 0 | } |
7780 | | |
7781 | | int |
7782 | | ncx_getn_short_long(const void **xpp, size_t nelems, long *tp) |
7783 | 0 | { |
7784 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
7785 | | |
7786 | | /* basic algorithm is: |
7787 | | * - ensure sane alignment of input data |
7788 | | * - copy (conversion happens automatically) input data |
7789 | | * to output |
7790 | | * - update xpp to point at next unconverted input, and tp to point |
7791 | | * at next location for converted output |
7792 | | */ |
7793 | | long i, j, ni; |
7794 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
7795 | | short *xp; |
7796 | | int nrange = 0; /* number of range errors */ |
7797 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
7798 | | long cxp = (long) *((char**)xpp); |
7799 | | |
7800 | | realign = (cxp & 7) % SIZEOF_SHORT; |
7801 | | /* sjl: manually stripmine so we can limit amount of |
7802 | | * vector work space reserved to LOOPCNT elements. Also |
7803 | | * makes vectorisation easy */ |
7804 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
7805 | | ni=Min(nelems-j,LOOPCNT); |
7806 | | if (realign) { |
7807 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
7808 | | xp = tmp; |
7809 | | } else { |
7810 | | xp = (short *) *xpp; |
7811 | | } |
7812 | | /* copy the next block */ |
7813 | | #pragma cdir loopcnt=LOOPCNT |
7814 | | #pragma cdir shortloop |
7815 | | for (i=0; i<ni; i++) { |
7816 | | tp[i] = (long) Max( LONG_MIN, Min(LONG_MAX, (long) xp[i])); |
7817 | | /* test for range errors (not always needed but do it anyway) */ |
7818 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
7819 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
7820 | | nrange += xp[i] > LONG_MAX || xp[i] < LONG_MIN; |
7821 | | } |
7822 | | /* update xpp and tp */ |
7823 | | if (realign) xp = (short *) *xpp; |
7824 | | xp += ni; |
7825 | | tp += ni; |
7826 | | *xpp = (void*)xp; |
7827 | | } |
7828 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
7829 | | |
7830 | | #else /* not SX */ |
7831 | 0 | const char *xp = (const char *) *xpp; |
7832 | 0 | int status = NC_NOERR; |
7833 | |
|
7834 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
7835 | 0 | { |
7836 | 0 | const int lstatus = ncx_get_short_long(xp, tp); |
7837 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
7838 | 0 | status = lstatus; |
7839 | 0 | } |
7840 | |
|
7841 | 0 | *xpp = (const void *)xp; |
7842 | 0 | return status; |
7843 | 0 | #endif |
7844 | 0 | } |
7845 | | |
7846 | | int |
7847 | | ncx_getn_short_float(const void **xpp, size_t nelems, float *tp) |
7848 | 0 | { |
7849 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
7850 | | |
7851 | | /* basic algorithm is: |
7852 | | * - ensure sane alignment of input data |
7853 | | * - copy (conversion happens automatically) input data |
7854 | | * to output |
7855 | | * - update xpp to point at next unconverted input, and tp to point |
7856 | | * at next location for converted output |
7857 | | */ |
7858 | | long i, j, ni; |
7859 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
7860 | | short *xp; |
7861 | | int nrange = 0; /* number of range errors */ |
7862 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
7863 | | long cxp = (long) *((char**)xpp); |
7864 | | |
7865 | | realign = (cxp & 7) % SIZEOF_SHORT; |
7866 | | /* sjl: manually stripmine so we can limit amount of |
7867 | | * vector work space reserved to LOOPCNT elements. Also |
7868 | | * makes vectorisation easy */ |
7869 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
7870 | | ni=Min(nelems-j,LOOPCNT); |
7871 | | if (realign) { |
7872 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
7873 | | xp = tmp; |
7874 | | } else { |
7875 | | xp = (short *) *xpp; |
7876 | | } |
7877 | | /* copy the next block */ |
7878 | | #pragma cdir loopcnt=LOOPCNT |
7879 | | #pragma cdir shortloop |
7880 | | for (i=0; i<ni; i++) { |
7881 | | tp[i] = (float) Max( FLOAT_MIN, Min(FLOAT_MAX, (float) xp[i])); |
7882 | | /* test for range errors (not always needed but do it anyway) */ |
7883 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
7884 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
7885 | | nrange += xp[i] > FLOAT_MAX || xp[i] < FLOAT_MIN; |
7886 | | } |
7887 | | /* update xpp and tp */ |
7888 | | if (realign) xp = (short *) *xpp; |
7889 | | xp += ni; |
7890 | | tp += ni; |
7891 | | *xpp = (void*)xp; |
7892 | | } |
7893 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
7894 | | |
7895 | | #else /* not SX */ |
7896 | 0 | const char *xp = (const char *) *xpp; |
7897 | 0 | int status = NC_NOERR; |
7898 | |
|
7899 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
7900 | 0 | { |
7901 | 0 | const int lstatus = ncx_get_short_float(xp, tp); |
7902 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
7903 | 0 | status = lstatus; |
7904 | 0 | } |
7905 | |
|
7906 | 0 | *xpp = (const void *)xp; |
7907 | 0 | return status; |
7908 | 0 | #endif |
7909 | 0 | } |
7910 | | |
7911 | | int |
7912 | | ncx_getn_short_double(const void **xpp, size_t nelems, double *tp) |
7913 | 0 | { |
7914 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
7915 | | |
7916 | | /* basic algorithm is: |
7917 | | * - ensure sane alignment of input data |
7918 | | * - copy (conversion happens automatically) input data |
7919 | | * to output |
7920 | | * - update xpp to point at next unconverted input, and tp to point |
7921 | | * at next location for converted output |
7922 | | */ |
7923 | | long i, j, ni; |
7924 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
7925 | | short *xp; |
7926 | | int nrange = 0; /* number of range errors */ |
7927 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
7928 | | long cxp = (long) *((char**)xpp); |
7929 | | |
7930 | | realign = (cxp & 7) % SIZEOF_SHORT; |
7931 | | /* sjl: manually stripmine so we can limit amount of |
7932 | | * vector work space reserved to LOOPCNT elements. Also |
7933 | | * makes vectorisation easy */ |
7934 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
7935 | | ni=Min(nelems-j,LOOPCNT); |
7936 | | if (realign) { |
7937 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
7938 | | xp = tmp; |
7939 | | } else { |
7940 | | xp = (short *) *xpp; |
7941 | | } |
7942 | | /* copy the next block */ |
7943 | | #pragma cdir loopcnt=LOOPCNT |
7944 | | #pragma cdir shortloop |
7945 | | for (i=0; i<ni; i++) { |
7946 | | tp[i] = (double) Max( DOUBLE_MIN, Min(DOUBLE_MAX, (double) xp[i])); |
7947 | | /* test for range errors (not always needed but do it anyway) */ |
7948 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
7949 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
7950 | | nrange += xp[i] > DOUBLE_MAX || xp[i] < DOUBLE_MIN; |
7951 | | } |
7952 | | /* update xpp and tp */ |
7953 | | if (realign) xp = (short *) *xpp; |
7954 | | xp += ni; |
7955 | | tp += ni; |
7956 | | *xpp = (void*)xp; |
7957 | | } |
7958 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
7959 | | |
7960 | | #else /* not SX */ |
7961 | 0 | const char *xp = (const char *) *xpp; |
7962 | 0 | int status = NC_NOERR; |
7963 | |
|
7964 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
7965 | 0 | { |
7966 | 0 | const int lstatus = ncx_get_short_double(xp, tp); |
7967 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
7968 | 0 | status = lstatus; |
7969 | 0 | } |
7970 | |
|
7971 | 0 | *xpp = (const void *)xp; |
7972 | 0 | return status; |
7973 | 0 | #endif |
7974 | 0 | } |
7975 | | |
7976 | | int |
7977 | | ncx_getn_short_longlong(const void **xpp, size_t nelems, longlong *tp) |
7978 | 0 | { |
7979 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
7980 | | |
7981 | | /* basic algorithm is: |
7982 | | * - ensure sane alignment of input data |
7983 | | * - copy (conversion happens automatically) input data |
7984 | | * to output |
7985 | | * - update xpp to point at next unconverted input, and tp to point |
7986 | | * at next location for converted output |
7987 | | */ |
7988 | | long i, j, ni; |
7989 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
7990 | | short *xp; |
7991 | | int nrange = 0; /* number of range errors */ |
7992 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
7993 | | long cxp = (long) *((char**)xpp); |
7994 | | |
7995 | | realign = (cxp & 7) % SIZEOF_SHORT; |
7996 | | /* sjl: manually stripmine so we can limit amount of |
7997 | | * vector work space reserved to LOOPCNT elements. Also |
7998 | | * makes vectorisation easy */ |
7999 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8000 | | ni=Min(nelems-j,LOOPCNT); |
8001 | | if (realign) { |
8002 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
8003 | | xp = tmp; |
8004 | | } else { |
8005 | | xp = (short *) *xpp; |
8006 | | } |
8007 | | /* copy the next block */ |
8008 | | #pragma cdir loopcnt=LOOPCNT |
8009 | | #pragma cdir shortloop |
8010 | | for (i=0; i<ni; i++) { |
8011 | | tp[i] = (longlong) Max( LONGLONG_MIN, Min(LONGLONG_MAX, (longlong) xp[i])); |
8012 | | /* test for range errors (not always needed but do it anyway) */ |
8013 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
8014 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
8015 | | nrange += xp[i] > LONGLONG_MAX || xp[i] < LONGLONG_MIN; |
8016 | | } |
8017 | | /* update xpp and tp */ |
8018 | | if (realign) xp = (short *) *xpp; |
8019 | | xp += ni; |
8020 | | tp += ni; |
8021 | | *xpp = (void*)xp; |
8022 | | } |
8023 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8024 | | |
8025 | | #else /* not SX */ |
8026 | 0 | const char *xp = (const char *) *xpp; |
8027 | 0 | int status = NC_NOERR; |
8028 | |
|
8029 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8030 | 0 | { |
8031 | 0 | const int lstatus = ncx_get_short_longlong(xp, tp); |
8032 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8033 | 0 | status = lstatus; |
8034 | 0 | } |
8035 | |
|
8036 | 0 | *xpp = (const void *)xp; |
8037 | 0 | return status; |
8038 | 0 | #endif |
8039 | 0 | } |
8040 | | |
8041 | | int |
8042 | | ncx_getn_short_uchar(const void **xpp, size_t nelems, uchar *tp) |
8043 | 0 | { |
8044 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8045 | | |
8046 | | /* basic algorithm is: |
8047 | | * - ensure sane alignment of input data |
8048 | | * - copy (conversion happens automatically) input data |
8049 | | * to output |
8050 | | * - update xpp to point at next unconverted input, and tp to point |
8051 | | * at next location for converted output |
8052 | | */ |
8053 | | long i, j, ni; |
8054 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8055 | | short *xp; |
8056 | | int nrange = 0; /* number of range errors */ |
8057 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8058 | | long cxp = (long) *((char**)xpp); |
8059 | | |
8060 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8061 | | /* sjl: manually stripmine so we can limit amount of |
8062 | | * vector work space reserved to LOOPCNT elements. Also |
8063 | | * makes vectorisation easy */ |
8064 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8065 | | ni=Min(nelems-j,LOOPCNT); |
8066 | | if (realign) { |
8067 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
8068 | | xp = tmp; |
8069 | | } else { |
8070 | | xp = (short *) *xpp; |
8071 | | } |
8072 | | /* copy the next block */ |
8073 | | #pragma cdir loopcnt=LOOPCNT |
8074 | | #pragma cdir shortloop |
8075 | | for (i=0; i<ni; i++) { |
8076 | | tp[i] = (uchar) Max( UCHAR_MIN, Min(UCHAR_MAX, (uchar) xp[i])); |
8077 | | /* test for range errors (not always needed but do it anyway) */ |
8078 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
8079 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
8080 | | nrange += xp[i] > UCHAR_MAX || xp[i] < 0; |
8081 | | } |
8082 | | /* update xpp and tp */ |
8083 | | if (realign) xp = (short *) *xpp; |
8084 | | xp += ni; |
8085 | | tp += ni; |
8086 | | *xpp = (void*)xp; |
8087 | | } |
8088 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8089 | | |
8090 | | #else /* not SX */ |
8091 | 0 | const char *xp = (const char *) *xpp; |
8092 | 0 | int status = NC_NOERR; |
8093 | |
|
8094 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8095 | 0 | { |
8096 | 0 | const int lstatus = ncx_get_short_uchar(xp, tp); |
8097 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8098 | 0 | status = lstatus; |
8099 | 0 | } |
8100 | |
|
8101 | 0 | *xpp = (const void *)xp; |
8102 | 0 | return status; |
8103 | 0 | #endif |
8104 | 0 | } |
8105 | | |
8106 | | int |
8107 | | ncx_getn_short_ushort(const void **xpp, size_t nelems, ushort *tp) |
8108 | 0 | { |
8109 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8110 | | |
8111 | | /* basic algorithm is: |
8112 | | * - ensure sane alignment of input data |
8113 | | * - copy (conversion happens automatically) input data |
8114 | | * to output |
8115 | | * - update xpp to point at next unconverted input, and tp to point |
8116 | | * at next location for converted output |
8117 | | */ |
8118 | | long i, j, ni; |
8119 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8120 | | short *xp; |
8121 | | int nrange = 0; /* number of range errors */ |
8122 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8123 | | long cxp = (long) *((char**)xpp); |
8124 | | |
8125 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8126 | | /* sjl: manually stripmine so we can limit amount of |
8127 | | * vector work space reserved to LOOPCNT elements. Also |
8128 | | * makes vectorisation easy */ |
8129 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8130 | | ni=Min(nelems-j,LOOPCNT); |
8131 | | if (realign) { |
8132 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
8133 | | xp = tmp; |
8134 | | } else { |
8135 | | xp = (short *) *xpp; |
8136 | | } |
8137 | | /* copy the next block */ |
8138 | | #pragma cdir loopcnt=LOOPCNT |
8139 | | #pragma cdir shortloop |
8140 | | for (i=0; i<ni; i++) { |
8141 | | tp[i] = (ushort) Max( USHORT_MIN, Min(USHORT_MAX, (ushort) xp[i])); |
8142 | | /* test for range errors (not always needed but do it anyway) */ |
8143 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
8144 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
8145 | | nrange += xp[i] > USHORT_MAX || xp[i] < 0; |
8146 | | } |
8147 | | /* update xpp and tp */ |
8148 | | if (realign) xp = (short *) *xpp; |
8149 | | xp += ni; |
8150 | | tp += ni; |
8151 | | *xpp = (void*)xp; |
8152 | | } |
8153 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8154 | | |
8155 | | #else /* not SX */ |
8156 | 0 | const char *xp = (const char *) *xpp; |
8157 | 0 | int status = NC_NOERR; |
8158 | |
|
8159 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8160 | 0 | { |
8161 | 0 | const int lstatus = ncx_get_short_ushort(xp, tp); |
8162 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8163 | 0 | status = lstatus; |
8164 | 0 | } |
8165 | |
|
8166 | 0 | *xpp = (const void *)xp; |
8167 | 0 | return status; |
8168 | 0 | #endif |
8169 | 0 | } |
8170 | | |
8171 | | int |
8172 | | ncx_getn_short_uint(const void **xpp, size_t nelems, uint *tp) |
8173 | 0 | { |
8174 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8175 | | |
8176 | | /* basic algorithm is: |
8177 | | * - ensure sane alignment of input data |
8178 | | * - copy (conversion happens automatically) input data |
8179 | | * to output |
8180 | | * - update xpp to point at next unconverted input, and tp to point |
8181 | | * at next location for converted output |
8182 | | */ |
8183 | | long i, j, ni; |
8184 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8185 | | short *xp; |
8186 | | int nrange = 0; /* number of range errors */ |
8187 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8188 | | long cxp = (long) *((char**)xpp); |
8189 | | |
8190 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8191 | | /* sjl: manually stripmine so we can limit amount of |
8192 | | * vector work space reserved to LOOPCNT elements. Also |
8193 | | * makes vectorisation easy */ |
8194 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8195 | | ni=Min(nelems-j,LOOPCNT); |
8196 | | if (realign) { |
8197 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
8198 | | xp = tmp; |
8199 | | } else { |
8200 | | xp = (short *) *xpp; |
8201 | | } |
8202 | | /* copy the next block */ |
8203 | | #pragma cdir loopcnt=LOOPCNT |
8204 | | #pragma cdir shortloop |
8205 | | for (i=0; i<ni; i++) { |
8206 | | tp[i] = (uint) Max( UINT_MIN, Min(UINT_MAX, (uint) xp[i])); |
8207 | | /* test for range errors (not always needed but do it anyway) */ |
8208 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
8209 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
8210 | | nrange += xp[i] > UINT_MAX || xp[i] < 0; |
8211 | | } |
8212 | | /* update xpp and tp */ |
8213 | | if (realign) xp = (short *) *xpp; |
8214 | | xp += ni; |
8215 | | tp += ni; |
8216 | | *xpp = (void*)xp; |
8217 | | } |
8218 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8219 | | |
8220 | | #else /* not SX */ |
8221 | 0 | const char *xp = (const char *) *xpp; |
8222 | 0 | int status = NC_NOERR; |
8223 | |
|
8224 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8225 | 0 | { |
8226 | 0 | const int lstatus = ncx_get_short_uint(xp, tp); |
8227 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8228 | 0 | status = lstatus; |
8229 | 0 | } |
8230 | |
|
8231 | 0 | *xpp = (const void *)xp; |
8232 | 0 | return status; |
8233 | 0 | #endif |
8234 | 0 | } |
8235 | | |
8236 | | int |
8237 | | ncx_getn_short_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
8238 | 0 | { |
8239 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8240 | | |
8241 | | /* basic algorithm is: |
8242 | | * - ensure sane alignment of input data |
8243 | | * - copy (conversion happens automatically) input data |
8244 | | * to output |
8245 | | * - update xpp to point at next unconverted input, and tp to point |
8246 | | * at next location for converted output |
8247 | | */ |
8248 | | long i, j, ni; |
8249 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8250 | | short *xp; |
8251 | | int nrange = 0; /* number of range errors */ |
8252 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8253 | | long cxp = (long) *((char**)xpp); |
8254 | | |
8255 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8256 | | /* sjl: manually stripmine so we can limit amount of |
8257 | | * vector work space reserved to LOOPCNT elements. Also |
8258 | | * makes vectorisation easy */ |
8259 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8260 | | ni=Min(nelems-j,LOOPCNT); |
8261 | | if (realign) { |
8262 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_SHORT)); |
8263 | | xp = tmp; |
8264 | | } else { |
8265 | | xp = (short *) *xpp; |
8266 | | } |
8267 | | /* copy the next block */ |
8268 | | #pragma cdir loopcnt=LOOPCNT |
8269 | | #pragma cdir shortloop |
8270 | | for (i=0; i<ni; i++) { |
8271 | | tp[i] = (ulonglong) Max( ULONGLONG_MIN, Min(ULONGLONG_MAX, (ulonglong) xp[i])); |
8272 | | /* test for range errors (not always needed but do it anyway) */ |
8273 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
8274 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
8275 | | nrange += xp[i] > ULONGLONG_MAX || xp[i] < 0; |
8276 | | } |
8277 | | /* update xpp and tp */ |
8278 | | if (realign) xp = (short *) *xpp; |
8279 | | xp += ni; |
8280 | | tp += ni; |
8281 | | *xpp = (void*)xp; |
8282 | | } |
8283 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8284 | | |
8285 | | #else /* not SX */ |
8286 | 0 | const char *xp = (const char *) *xpp; |
8287 | 0 | int status = NC_NOERR; |
8288 | |
|
8289 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8290 | 0 | { |
8291 | 0 | const int lstatus = ncx_get_short_ulonglong(xp, tp); |
8292 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8293 | 0 | status = lstatus; |
8294 | 0 | } |
8295 | |
|
8296 | 0 | *xpp = (const void *)xp; |
8297 | 0 | return status; |
8298 | 0 | #endif |
8299 | 0 | } |
8300 | | |
8301 | | |
8302 | | int |
8303 | | ncx_pad_getn_short_schar(const void **xpp, size_t nelems, schar *tp) |
8304 | 0 | { |
8305 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8306 | |
|
8307 | 0 | const char *xp = (const char *) *xpp; |
8308 | 0 | int status = NC_NOERR; |
8309 | |
|
8310 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8311 | 0 | { |
8312 | 0 | const int lstatus = ncx_get_short_schar(xp, tp); |
8313 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8314 | 0 | status = lstatus; |
8315 | 0 | } |
8316 | |
|
8317 | 0 | if (rndup != 0) |
8318 | 0 | xp += X_SIZEOF_SHORT; |
8319 | |
|
8320 | 0 | *xpp = (void *)xp; |
8321 | 0 | return status; |
8322 | 0 | } |
8323 | | |
8324 | | int |
8325 | | ncx_pad_getn_short_uchar(const void **xpp, size_t nelems, uchar *tp) |
8326 | 0 | { |
8327 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8328 | |
|
8329 | 0 | const char *xp = (const char *) *xpp; |
8330 | 0 | int status = NC_NOERR; |
8331 | |
|
8332 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8333 | 0 | { |
8334 | 0 | const int lstatus = ncx_get_short_uchar(xp, tp); |
8335 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8336 | 0 | status = lstatus; |
8337 | 0 | } |
8338 | |
|
8339 | 0 | if (rndup != 0) |
8340 | 0 | xp += X_SIZEOF_SHORT; |
8341 | |
|
8342 | 0 | *xpp = (void *)xp; |
8343 | 0 | return status; |
8344 | 0 | } |
8345 | | |
8346 | | int |
8347 | | ncx_pad_getn_short_short(const void **xpp, size_t nelems, short *tp) |
8348 | 0 | { |
8349 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8350 | |
|
8351 | 0 | const char *xp = (const char *) *xpp; |
8352 | 0 | int status = NC_NOERR; |
8353 | |
|
8354 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8355 | 0 | { |
8356 | 0 | const int lstatus = ncx_get_short_short(xp, tp); |
8357 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8358 | 0 | status = lstatus; |
8359 | 0 | } |
8360 | |
|
8361 | 0 | if (rndup != 0) |
8362 | 0 | xp += X_SIZEOF_SHORT; |
8363 | |
|
8364 | 0 | *xpp = (void *)xp; |
8365 | 0 | return status; |
8366 | 0 | } |
8367 | | |
8368 | | int |
8369 | | ncx_pad_getn_short_int(const void **xpp, size_t nelems, int *tp) |
8370 | 0 | { |
8371 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8372 | |
|
8373 | 0 | const char *xp = (const char *) *xpp; |
8374 | 0 | int status = NC_NOERR; |
8375 | |
|
8376 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8377 | 0 | { |
8378 | 0 | const int lstatus = ncx_get_short_int(xp, tp); |
8379 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8380 | 0 | status = lstatus; |
8381 | 0 | } |
8382 | |
|
8383 | 0 | if (rndup != 0) |
8384 | 0 | xp += X_SIZEOF_SHORT; |
8385 | |
|
8386 | 0 | *xpp = (void *)xp; |
8387 | 0 | return status; |
8388 | 0 | } |
8389 | | |
8390 | | int |
8391 | | ncx_pad_getn_short_long(const void **xpp, size_t nelems, long *tp) |
8392 | 0 | { |
8393 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8394 | |
|
8395 | 0 | const char *xp = (const char *) *xpp; |
8396 | 0 | int status = NC_NOERR; |
8397 | |
|
8398 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8399 | 0 | { |
8400 | 0 | const int lstatus = ncx_get_short_long(xp, tp); |
8401 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8402 | 0 | status = lstatus; |
8403 | 0 | } |
8404 | |
|
8405 | 0 | if (rndup != 0) |
8406 | 0 | xp += X_SIZEOF_SHORT; |
8407 | |
|
8408 | 0 | *xpp = (void *)xp; |
8409 | 0 | return status; |
8410 | 0 | } |
8411 | | |
8412 | | int |
8413 | | ncx_pad_getn_short_float(const void **xpp, size_t nelems, float *tp) |
8414 | 0 | { |
8415 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8416 | |
|
8417 | 0 | const char *xp = (const char *) *xpp; |
8418 | 0 | int status = NC_NOERR; |
8419 | |
|
8420 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8421 | 0 | { |
8422 | 0 | const int lstatus = ncx_get_short_float(xp, tp); |
8423 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8424 | 0 | status = lstatus; |
8425 | 0 | } |
8426 | |
|
8427 | 0 | if (rndup != 0) |
8428 | 0 | xp += X_SIZEOF_SHORT; |
8429 | |
|
8430 | 0 | *xpp = (void *)xp; |
8431 | 0 | return status; |
8432 | 0 | } |
8433 | | |
8434 | | int |
8435 | | ncx_pad_getn_short_double(const void **xpp, size_t nelems, double *tp) |
8436 | 0 | { |
8437 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8438 | |
|
8439 | 0 | const char *xp = (const char *) *xpp; |
8440 | 0 | int status = NC_NOERR; |
8441 | |
|
8442 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8443 | 0 | { |
8444 | 0 | const int lstatus = ncx_get_short_double(xp, tp); |
8445 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8446 | 0 | status = lstatus; |
8447 | 0 | } |
8448 | |
|
8449 | 0 | if (rndup != 0) |
8450 | 0 | xp += X_SIZEOF_SHORT; |
8451 | |
|
8452 | 0 | *xpp = (void *)xp; |
8453 | 0 | return status; |
8454 | 0 | } |
8455 | | |
8456 | | int |
8457 | | ncx_pad_getn_short_uint(const void **xpp, size_t nelems, uint *tp) |
8458 | 0 | { |
8459 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8460 | |
|
8461 | 0 | const char *xp = (const char *) *xpp; |
8462 | 0 | int status = NC_NOERR; |
8463 | |
|
8464 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8465 | 0 | { |
8466 | 0 | const int lstatus = ncx_get_short_uint(xp, tp); |
8467 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8468 | 0 | status = lstatus; |
8469 | 0 | } |
8470 | |
|
8471 | 0 | if (rndup != 0) |
8472 | 0 | xp += X_SIZEOF_SHORT; |
8473 | |
|
8474 | 0 | *xpp = (void *)xp; |
8475 | 0 | return status; |
8476 | 0 | } |
8477 | | |
8478 | | int |
8479 | | ncx_pad_getn_short_longlong(const void **xpp, size_t nelems, longlong *tp) |
8480 | 0 | { |
8481 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8482 | |
|
8483 | 0 | const char *xp = (const char *) *xpp; |
8484 | 0 | int status = NC_NOERR; |
8485 | |
|
8486 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8487 | 0 | { |
8488 | 0 | const int lstatus = ncx_get_short_longlong(xp, tp); |
8489 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8490 | 0 | status = lstatus; |
8491 | 0 | } |
8492 | |
|
8493 | 0 | if (rndup != 0) |
8494 | 0 | xp += X_SIZEOF_SHORT; |
8495 | |
|
8496 | 0 | *xpp = (void *)xp; |
8497 | 0 | return status; |
8498 | 0 | } |
8499 | | |
8500 | | int |
8501 | | ncx_pad_getn_short_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
8502 | 0 | { |
8503 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8504 | |
|
8505 | 0 | const char *xp = (const char *) *xpp; |
8506 | 0 | int status = NC_NOERR; |
8507 | |
|
8508 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8509 | 0 | { |
8510 | 0 | const int lstatus = ncx_get_short_ulonglong(xp, tp); |
8511 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8512 | 0 | status = lstatus; |
8513 | 0 | } |
8514 | |
|
8515 | 0 | if (rndup != 0) |
8516 | 0 | xp += X_SIZEOF_SHORT; |
8517 | |
|
8518 | 0 | *xpp = (void *)xp; |
8519 | 0 | return status; |
8520 | 0 | } |
8521 | | |
8522 | | int |
8523 | | ncx_pad_getn_short_ushort(const void **xpp, size_t nelems, ushort *tp) |
8524 | 0 | { |
8525 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
8526 | |
|
8527 | 0 | const char *xp = (const char *) *xpp; |
8528 | 0 | int status = NC_NOERR; |
8529 | |
|
8530 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8531 | 0 | { |
8532 | 0 | const int lstatus = ncx_get_short_ushort(xp, tp); |
8533 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8534 | 0 | status = lstatus; |
8535 | 0 | } |
8536 | |
|
8537 | 0 | if (rndup != 0) |
8538 | 0 | xp += X_SIZEOF_SHORT; |
8539 | |
|
8540 | 0 | *xpp = (void *)xp; |
8541 | 0 | return status; |
8542 | 0 | } |
8543 | | |
8544 | | |
8545 | | #if X_SIZEOF_SHORT == SIZEOF_SHORT |
8546 | | /* optimized version */ |
8547 | | int |
8548 | | ncx_putn_short_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
8549 | 0 | { |
8550 | | #ifdef WORDS_BIGENDIAN |
8551 | | (void) memcpy(*xpp, tp, (size_t)nelems * X_SIZEOF_SHORT); |
8552 | | # else |
8553 | 0 | swapn2b(*xpp, tp, nelems); |
8554 | 0 | # endif |
8555 | 0 | *xpp = (void *)((char *)(*xpp) + nelems * X_SIZEOF_SHORT); |
8556 | 0 | return NC_NOERR; |
8557 | 0 | } |
8558 | | #else |
8559 | | int |
8560 | | ncx_putn_short_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
8561 | | { |
8562 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8563 | | |
8564 | | /* basic algorithm is: |
8565 | | * - ensure sane alignment of output data |
8566 | | * - copy (conversion happens automatically) input data |
8567 | | * to output |
8568 | | * - update tp to point at next unconverted input, and xpp to point |
8569 | | * at next location for converted output |
8570 | | */ |
8571 | | long i, j, ni; |
8572 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8573 | | short *xp; |
8574 | | int nrange = 0; /* number of range errors */ |
8575 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8576 | | long cxp = (long) *((char**)xpp); |
8577 | | |
8578 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8579 | | /* sjl: manually stripmine so we can limit amount of |
8580 | | * vector work space reserved to LOOPCNT elements. Also |
8581 | | * makes vectorisation easy */ |
8582 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8583 | | ni=Min(nelems-j,LOOPCNT); |
8584 | | if (realign) { |
8585 | | xp = tmp; |
8586 | | } else { |
8587 | | xp = (short *) *xpp; |
8588 | | } |
8589 | | /* copy the next block */ |
8590 | | #pragma cdir loopcnt=LOOPCNT |
8591 | | #pragma cdir shortloop |
8592 | | for (i=0; i<ni; i++) { |
8593 | | /* the normal case: */ |
8594 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
8595 | | /* test for range errors (not always needed but do it anyway) */ |
8596 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
8597 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
8598 | | nrange += tp[i] > X_SHORT_MAX || tp[i] < X_SHORT_MIN; |
8599 | | } |
8600 | | /* copy workspace back if necessary */ |
8601 | | if (realign) { |
8602 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
8603 | | xp = (short *) *xpp; |
8604 | | } |
8605 | | /* update xpp and tp */ |
8606 | | xp += ni; |
8607 | | tp += ni; |
8608 | | *xpp = (void*)xp; |
8609 | | } |
8610 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8611 | | |
8612 | | #else /* not SX */ |
8613 | | |
8614 | | char *xp = (char *) *xpp; |
8615 | | int status = NC_NOERR; |
8616 | | |
8617 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8618 | | { |
8619 | | int lstatus = ncx_put_short_short(xp, tp, fillp); |
8620 | | if (status == NC_NOERR) /* report the first encountered error */ |
8621 | | status = lstatus; |
8622 | | } |
8623 | | |
8624 | | *xpp = (void *)xp; |
8625 | | return status; |
8626 | | #endif |
8627 | | } |
8628 | | |
8629 | | #endif |
8630 | | int |
8631 | | ncx_putn_short_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
8632 | 0 | { |
8633 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8634 | | |
8635 | | /* basic algorithm is: |
8636 | | * - ensure sane alignment of output data |
8637 | | * - copy (conversion happens automatically) input data |
8638 | | * to output |
8639 | | * - update tp to point at next unconverted input, and xpp to point |
8640 | | * at next location for converted output |
8641 | | */ |
8642 | | long i, j, ni; |
8643 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8644 | | short *xp; |
8645 | | int nrange = 0; /* number of range errors */ |
8646 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8647 | | long cxp = (long) *((char**)xpp); |
8648 | | |
8649 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8650 | | /* sjl: manually stripmine so we can limit amount of |
8651 | | * vector work space reserved to LOOPCNT elements. Also |
8652 | | * makes vectorisation easy */ |
8653 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8654 | | ni=Min(nelems-j,LOOPCNT); |
8655 | | if (realign) { |
8656 | | xp = tmp; |
8657 | | } else { |
8658 | | xp = (short *) *xpp; |
8659 | | } |
8660 | | /* copy the next block */ |
8661 | | #pragma cdir loopcnt=LOOPCNT |
8662 | | #pragma cdir shortloop |
8663 | | for (i=0; i<ni; i++) { |
8664 | | /* the normal case: */ |
8665 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
8666 | | /* test for range errors (not always needed but do it anyway) */ |
8667 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
8668 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
8669 | | nrange += tp[i] > X_SHORT_MAX || tp[i] < X_SHORT_MIN; |
8670 | | } |
8671 | | /* copy workspace back if necessary */ |
8672 | | if (realign) { |
8673 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
8674 | | xp = (short *) *xpp; |
8675 | | } |
8676 | | /* update xpp and tp */ |
8677 | | xp += ni; |
8678 | | tp += ni; |
8679 | | *xpp = (void*)xp; |
8680 | | } |
8681 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8682 | | |
8683 | | #else /* not SX */ |
8684 | |
|
8685 | 0 | char *xp = (char *) *xpp; |
8686 | 0 | int status = NC_NOERR; |
8687 | |
|
8688 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8689 | 0 | { |
8690 | 0 | int lstatus = ncx_put_short_schar(xp, tp, fillp); |
8691 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8692 | 0 | status = lstatus; |
8693 | 0 | } |
8694 | |
|
8695 | 0 | *xpp = (void *)xp; |
8696 | 0 | return status; |
8697 | 0 | #endif |
8698 | 0 | } |
8699 | | |
8700 | | int |
8701 | | ncx_putn_short_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
8702 | 0 | { |
8703 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8704 | | |
8705 | | /* basic algorithm is: |
8706 | | * - ensure sane alignment of output data |
8707 | | * - copy (conversion happens automatically) input data |
8708 | | * to output |
8709 | | * - update tp to point at next unconverted input, and xpp to point |
8710 | | * at next location for converted output |
8711 | | */ |
8712 | | long i, j, ni; |
8713 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8714 | | short *xp; |
8715 | | int nrange = 0; /* number of range errors */ |
8716 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8717 | | long cxp = (long) *((char**)xpp); |
8718 | | |
8719 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8720 | | /* sjl: manually stripmine so we can limit amount of |
8721 | | * vector work space reserved to LOOPCNT elements. Also |
8722 | | * makes vectorisation easy */ |
8723 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8724 | | ni=Min(nelems-j,LOOPCNT); |
8725 | | if (realign) { |
8726 | | xp = tmp; |
8727 | | } else { |
8728 | | xp = (short *) *xpp; |
8729 | | } |
8730 | | /* copy the next block */ |
8731 | | #pragma cdir loopcnt=LOOPCNT |
8732 | | #pragma cdir shortloop |
8733 | | for (i=0; i<ni; i++) { |
8734 | | /* the normal case: */ |
8735 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
8736 | | /* test for range errors (not always needed but do it anyway) */ |
8737 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
8738 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
8739 | | nrange += tp[i] > X_SHORT_MAX || tp[i] < X_SHORT_MIN; |
8740 | | } |
8741 | | /* copy workspace back if necessary */ |
8742 | | if (realign) { |
8743 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
8744 | | xp = (short *) *xpp; |
8745 | | } |
8746 | | /* update xpp and tp */ |
8747 | | xp += ni; |
8748 | | tp += ni; |
8749 | | *xpp = (void*)xp; |
8750 | | } |
8751 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8752 | | |
8753 | | #else /* not SX */ |
8754 | |
|
8755 | 0 | char *xp = (char *) *xpp; |
8756 | 0 | int status = NC_NOERR; |
8757 | |
|
8758 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8759 | 0 | { |
8760 | 0 | int lstatus = ncx_put_short_int(xp, tp, fillp); |
8761 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8762 | 0 | status = lstatus; |
8763 | 0 | } |
8764 | |
|
8765 | 0 | *xpp = (void *)xp; |
8766 | 0 | return status; |
8767 | 0 | #endif |
8768 | 0 | } |
8769 | | |
8770 | | int |
8771 | | ncx_putn_short_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
8772 | 0 | { |
8773 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8774 | | |
8775 | | /* basic algorithm is: |
8776 | | * - ensure sane alignment of output data |
8777 | | * - copy (conversion happens automatically) input data |
8778 | | * to output |
8779 | | * - update tp to point at next unconverted input, and xpp to point |
8780 | | * at next location for converted output |
8781 | | */ |
8782 | | long i, j, ni; |
8783 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8784 | | short *xp; |
8785 | | int nrange = 0; /* number of range errors */ |
8786 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8787 | | long cxp = (long) *((char**)xpp); |
8788 | | |
8789 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8790 | | /* sjl: manually stripmine so we can limit amount of |
8791 | | * vector work space reserved to LOOPCNT elements. Also |
8792 | | * makes vectorisation easy */ |
8793 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8794 | | ni=Min(nelems-j,LOOPCNT); |
8795 | | if (realign) { |
8796 | | xp = tmp; |
8797 | | } else { |
8798 | | xp = (short *) *xpp; |
8799 | | } |
8800 | | /* copy the next block */ |
8801 | | #pragma cdir loopcnt=LOOPCNT |
8802 | | #pragma cdir shortloop |
8803 | | for (i=0; i<ni; i++) { |
8804 | | /* the normal case: */ |
8805 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
8806 | | /* test for range errors (not always needed but do it anyway) */ |
8807 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
8808 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
8809 | | nrange += tp[i] > X_SHORT_MAX || tp[i] < X_SHORT_MIN; |
8810 | | } |
8811 | | /* copy workspace back if necessary */ |
8812 | | if (realign) { |
8813 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
8814 | | xp = (short *) *xpp; |
8815 | | } |
8816 | | /* update xpp and tp */ |
8817 | | xp += ni; |
8818 | | tp += ni; |
8819 | | *xpp = (void*)xp; |
8820 | | } |
8821 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8822 | | |
8823 | | #else /* not SX */ |
8824 | |
|
8825 | 0 | char *xp = (char *) *xpp; |
8826 | 0 | int status = NC_NOERR; |
8827 | |
|
8828 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8829 | 0 | { |
8830 | 0 | int lstatus = ncx_put_short_long(xp, tp, fillp); |
8831 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8832 | 0 | status = lstatus; |
8833 | 0 | } |
8834 | |
|
8835 | 0 | *xpp = (void *)xp; |
8836 | 0 | return status; |
8837 | 0 | #endif |
8838 | 0 | } |
8839 | | |
8840 | | int |
8841 | | ncx_putn_short_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
8842 | 0 | { |
8843 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8844 | | |
8845 | | /* basic algorithm is: |
8846 | | * - ensure sane alignment of output data |
8847 | | * - copy (conversion happens automatically) input data |
8848 | | * to output |
8849 | | * - update tp to point at next unconverted input, and xpp to point |
8850 | | * at next location for converted output |
8851 | | */ |
8852 | | long i, j, ni; |
8853 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8854 | | short *xp; |
8855 | | int nrange = 0; /* number of range errors */ |
8856 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8857 | | long cxp = (long) *((char**)xpp); |
8858 | | |
8859 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8860 | | /* sjl: manually stripmine so we can limit amount of |
8861 | | * vector work space reserved to LOOPCNT elements. Also |
8862 | | * makes vectorisation easy */ |
8863 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8864 | | ni=Min(nelems-j,LOOPCNT); |
8865 | | if (realign) { |
8866 | | xp = tmp; |
8867 | | } else { |
8868 | | xp = (short *) *xpp; |
8869 | | } |
8870 | | /* copy the next block */ |
8871 | | #pragma cdir loopcnt=LOOPCNT |
8872 | | #pragma cdir shortloop |
8873 | | for (i=0; i<ni; i++) { |
8874 | | /* the normal case: */ |
8875 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
8876 | | /* test for range errors (not always needed but do it anyway) */ |
8877 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
8878 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
8879 | | nrange += tp[i] > X_SHORT_MAX || tp[i] < X_SHORT_MIN; |
8880 | | } |
8881 | | /* copy workspace back if necessary */ |
8882 | | if (realign) { |
8883 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
8884 | | xp = (short *) *xpp; |
8885 | | } |
8886 | | /* update xpp and tp */ |
8887 | | xp += ni; |
8888 | | tp += ni; |
8889 | | *xpp = (void*)xp; |
8890 | | } |
8891 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8892 | | |
8893 | | #else /* not SX */ |
8894 | |
|
8895 | 0 | char *xp = (char *) *xpp; |
8896 | 0 | int status = NC_NOERR; |
8897 | |
|
8898 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8899 | 0 | { |
8900 | 0 | int lstatus = ncx_put_short_float(xp, tp, fillp); |
8901 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8902 | 0 | status = lstatus; |
8903 | 0 | } |
8904 | |
|
8905 | 0 | *xpp = (void *)xp; |
8906 | 0 | return status; |
8907 | 0 | #endif |
8908 | 0 | } |
8909 | | |
8910 | | int |
8911 | | ncx_putn_short_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
8912 | 0 | { |
8913 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8914 | | |
8915 | | /* basic algorithm is: |
8916 | | * - ensure sane alignment of output data |
8917 | | * - copy (conversion happens automatically) input data |
8918 | | * to output |
8919 | | * - update tp to point at next unconverted input, and xpp to point |
8920 | | * at next location for converted output |
8921 | | */ |
8922 | | long i, j, ni; |
8923 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8924 | | short *xp; |
8925 | | int nrange = 0; /* number of range errors */ |
8926 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8927 | | long cxp = (long) *((char**)xpp); |
8928 | | |
8929 | | realign = (cxp & 7) % SIZEOF_SHORT; |
8930 | | /* sjl: manually stripmine so we can limit amount of |
8931 | | * vector work space reserved to LOOPCNT elements. Also |
8932 | | * makes vectorisation easy */ |
8933 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
8934 | | ni=Min(nelems-j,LOOPCNT); |
8935 | | if (realign) { |
8936 | | xp = tmp; |
8937 | | } else { |
8938 | | xp = (short *) *xpp; |
8939 | | } |
8940 | | /* copy the next block */ |
8941 | | #pragma cdir loopcnt=LOOPCNT |
8942 | | #pragma cdir shortloop |
8943 | | for (i=0; i<ni; i++) { |
8944 | | /* the normal case: */ |
8945 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
8946 | | /* test for range errors (not always needed but do it anyway) */ |
8947 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
8948 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
8949 | | nrange += tp[i] > X_SHORT_MAX || tp[i] < X_SHORT_MIN; |
8950 | | } |
8951 | | /* copy workspace back if necessary */ |
8952 | | if (realign) { |
8953 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
8954 | | xp = (short *) *xpp; |
8955 | | } |
8956 | | /* update xpp and tp */ |
8957 | | xp += ni; |
8958 | | tp += ni; |
8959 | | *xpp = (void*)xp; |
8960 | | } |
8961 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
8962 | | |
8963 | | #else /* not SX */ |
8964 | |
|
8965 | 0 | char *xp = (char *) *xpp; |
8966 | 0 | int status = NC_NOERR; |
8967 | |
|
8968 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
8969 | 0 | { |
8970 | 0 | int lstatus = ncx_put_short_double(xp, tp, fillp); |
8971 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
8972 | 0 | status = lstatus; |
8973 | 0 | } |
8974 | |
|
8975 | 0 | *xpp = (void *)xp; |
8976 | 0 | return status; |
8977 | 0 | #endif |
8978 | 0 | } |
8979 | | |
8980 | | int |
8981 | | ncx_putn_short_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
8982 | 0 | { |
8983 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
8984 | | |
8985 | | /* basic algorithm is: |
8986 | | * - ensure sane alignment of output data |
8987 | | * - copy (conversion happens automatically) input data |
8988 | | * to output |
8989 | | * - update tp to point at next unconverted input, and xpp to point |
8990 | | * at next location for converted output |
8991 | | */ |
8992 | | long i, j, ni; |
8993 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
8994 | | short *xp; |
8995 | | int nrange = 0; /* number of range errors */ |
8996 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
8997 | | long cxp = (long) *((char**)xpp); |
8998 | | |
8999 | | realign = (cxp & 7) % SIZEOF_SHORT; |
9000 | | /* sjl: manually stripmine so we can limit amount of |
9001 | | * vector work space reserved to LOOPCNT elements. Also |
9002 | | * makes vectorisation easy */ |
9003 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9004 | | ni=Min(nelems-j,LOOPCNT); |
9005 | | if (realign) { |
9006 | | xp = tmp; |
9007 | | } else { |
9008 | | xp = (short *) *xpp; |
9009 | | } |
9010 | | /* copy the next block */ |
9011 | | #pragma cdir loopcnt=LOOPCNT |
9012 | | #pragma cdir shortloop |
9013 | | for (i=0; i<ni; i++) { |
9014 | | /* the normal case: */ |
9015 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
9016 | | /* test for range errors (not always needed but do it anyway) */ |
9017 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
9018 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
9019 | | nrange += tp[i] > X_SHORT_MAX || tp[i] < X_SHORT_MIN; |
9020 | | } |
9021 | | /* copy workspace back if necessary */ |
9022 | | if (realign) { |
9023 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
9024 | | xp = (short *) *xpp; |
9025 | | } |
9026 | | /* update xpp and tp */ |
9027 | | xp += ni; |
9028 | | tp += ni; |
9029 | | *xpp = (void*)xp; |
9030 | | } |
9031 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9032 | | |
9033 | | #else /* not SX */ |
9034 | |
|
9035 | 0 | char *xp = (char *) *xpp; |
9036 | 0 | int status = NC_NOERR; |
9037 | |
|
9038 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9039 | 0 | { |
9040 | 0 | int lstatus = ncx_put_short_longlong(xp, tp, fillp); |
9041 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9042 | 0 | status = lstatus; |
9043 | 0 | } |
9044 | |
|
9045 | 0 | *xpp = (void *)xp; |
9046 | 0 | return status; |
9047 | 0 | #endif |
9048 | 0 | } |
9049 | | |
9050 | | int |
9051 | | ncx_putn_short_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
9052 | 0 | { |
9053 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
9054 | | |
9055 | | /* basic algorithm is: |
9056 | | * - ensure sane alignment of output data |
9057 | | * - copy (conversion happens automatically) input data |
9058 | | * to output |
9059 | | * - update tp to point at next unconverted input, and xpp to point |
9060 | | * at next location for converted output |
9061 | | */ |
9062 | | long i, j, ni; |
9063 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
9064 | | short *xp; |
9065 | | int nrange = 0; /* number of range errors */ |
9066 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9067 | | long cxp = (long) *((char**)xpp); |
9068 | | |
9069 | | realign = (cxp & 7) % SIZEOF_SHORT; |
9070 | | /* sjl: manually stripmine so we can limit amount of |
9071 | | * vector work space reserved to LOOPCNT elements. Also |
9072 | | * makes vectorisation easy */ |
9073 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9074 | | ni=Min(nelems-j,LOOPCNT); |
9075 | | if (realign) { |
9076 | | xp = tmp; |
9077 | | } else { |
9078 | | xp = (short *) *xpp; |
9079 | | } |
9080 | | /* copy the next block */ |
9081 | | #pragma cdir loopcnt=LOOPCNT |
9082 | | #pragma cdir shortloop |
9083 | | for (i=0; i<ni; i++) { |
9084 | | /* the normal case: */ |
9085 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
9086 | | /* test for range errors (not always needed but do it anyway) */ |
9087 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
9088 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
9089 | | nrange += tp[i] > X_SHORT_MAX ; |
9090 | | } |
9091 | | /* copy workspace back if necessary */ |
9092 | | if (realign) { |
9093 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
9094 | | xp = (short *) *xpp; |
9095 | | } |
9096 | | /* update xpp and tp */ |
9097 | | xp += ni; |
9098 | | tp += ni; |
9099 | | *xpp = (void*)xp; |
9100 | | } |
9101 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9102 | | |
9103 | | #else /* not SX */ |
9104 | |
|
9105 | 0 | char *xp = (char *) *xpp; |
9106 | 0 | int status = NC_NOERR; |
9107 | |
|
9108 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9109 | 0 | { |
9110 | 0 | int lstatus = ncx_put_short_uchar(xp, tp, fillp); |
9111 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9112 | 0 | status = lstatus; |
9113 | 0 | } |
9114 | |
|
9115 | 0 | *xpp = (void *)xp; |
9116 | 0 | return status; |
9117 | 0 | #endif |
9118 | 0 | } |
9119 | | |
9120 | | int |
9121 | | ncx_putn_short_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
9122 | 0 | { |
9123 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
9124 | | |
9125 | | /* basic algorithm is: |
9126 | | * - ensure sane alignment of output data |
9127 | | * - copy (conversion happens automatically) input data |
9128 | | * to output |
9129 | | * - update tp to point at next unconverted input, and xpp to point |
9130 | | * at next location for converted output |
9131 | | */ |
9132 | | long i, j, ni; |
9133 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
9134 | | short *xp; |
9135 | | int nrange = 0; /* number of range errors */ |
9136 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9137 | | long cxp = (long) *((char**)xpp); |
9138 | | |
9139 | | realign = (cxp & 7) % SIZEOF_SHORT; |
9140 | | /* sjl: manually stripmine so we can limit amount of |
9141 | | * vector work space reserved to LOOPCNT elements. Also |
9142 | | * makes vectorisation easy */ |
9143 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9144 | | ni=Min(nelems-j,LOOPCNT); |
9145 | | if (realign) { |
9146 | | xp = tmp; |
9147 | | } else { |
9148 | | xp = (short *) *xpp; |
9149 | | } |
9150 | | /* copy the next block */ |
9151 | | #pragma cdir loopcnt=LOOPCNT |
9152 | | #pragma cdir shortloop |
9153 | | for (i=0; i<ni; i++) { |
9154 | | /* the normal case: */ |
9155 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
9156 | | /* test for range errors (not always needed but do it anyway) */ |
9157 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
9158 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
9159 | | nrange += tp[i] > X_SHORT_MAX ; |
9160 | | } |
9161 | | /* copy workspace back if necessary */ |
9162 | | if (realign) { |
9163 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
9164 | | xp = (short *) *xpp; |
9165 | | } |
9166 | | /* update xpp and tp */ |
9167 | | xp += ni; |
9168 | | tp += ni; |
9169 | | *xpp = (void*)xp; |
9170 | | } |
9171 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9172 | | |
9173 | | #else /* not SX */ |
9174 | |
|
9175 | 0 | char *xp = (char *) *xpp; |
9176 | 0 | int status = NC_NOERR; |
9177 | |
|
9178 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9179 | 0 | { |
9180 | 0 | int lstatus = ncx_put_short_uint(xp, tp, fillp); |
9181 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9182 | 0 | status = lstatus; |
9183 | 0 | } |
9184 | |
|
9185 | 0 | *xpp = (void *)xp; |
9186 | 0 | return status; |
9187 | 0 | #endif |
9188 | 0 | } |
9189 | | |
9190 | | int |
9191 | | ncx_putn_short_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
9192 | 0 | { |
9193 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
9194 | | |
9195 | | /* basic algorithm is: |
9196 | | * - ensure sane alignment of output data |
9197 | | * - copy (conversion happens automatically) input data |
9198 | | * to output |
9199 | | * - update tp to point at next unconverted input, and xpp to point |
9200 | | * at next location for converted output |
9201 | | */ |
9202 | | long i, j, ni; |
9203 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
9204 | | short *xp; |
9205 | | int nrange = 0; /* number of range errors */ |
9206 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9207 | | long cxp = (long) *((char**)xpp); |
9208 | | |
9209 | | realign = (cxp & 7) % SIZEOF_SHORT; |
9210 | | /* sjl: manually stripmine so we can limit amount of |
9211 | | * vector work space reserved to LOOPCNT elements. Also |
9212 | | * makes vectorisation easy */ |
9213 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9214 | | ni=Min(nelems-j,LOOPCNT); |
9215 | | if (realign) { |
9216 | | xp = tmp; |
9217 | | } else { |
9218 | | xp = (short *) *xpp; |
9219 | | } |
9220 | | /* copy the next block */ |
9221 | | #pragma cdir loopcnt=LOOPCNT |
9222 | | #pragma cdir shortloop |
9223 | | for (i=0; i<ni; i++) { |
9224 | | /* the normal case: */ |
9225 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
9226 | | /* test for range errors (not always needed but do it anyway) */ |
9227 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
9228 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
9229 | | nrange += tp[i] > X_SHORT_MAX ; |
9230 | | } |
9231 | | /* copy workspace back if necessary */ |
9232 | | if (realign) { |
9233 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
9234 | | xp = (short *) *xpp; |
9235 | | } |
9236 | | /* update xpp and tp */ |
9237 | | xp += ni; |
9238 | | tp += ni; |
9239 | | *xpp = (void*)xp; |
9240 | | } |
9241 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9242 | | |
9243 | | #else /* not SX */ |
9244 | |
|
9245 | 0 | char *xp = (char *) *xpp; |
9246 | 0 | int status = NC_NOERR; |
9247 | |
|
9248 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9249 | 0 | { |
9250 | 0 | int lstatus = ncx_put_short_ulonglong(xp, tp, fillp); |
9251 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9252 | 0 | status = lstatus; |
9253 | 0 | } |
9254 | |
|
9255 | 0 | *xpp = (void *)xp; |
9256 | 0 | return status; |
9257 | 0 | #endif |
9258 | 0 | } |
9259 | | |
9260 | | int |
9261 | | ncx_putn_short_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
9262 | 0 | { |
9263 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_SHORT == SIZEOF_SHORT |
9264 | | |
9265 | | /* basic algorithm is: |
9266 | | * - ensure sane alignment of output data |
9267 | | * - copy (conversion happens automatically) input data |
9268 | | * to output |
9269 | | * - update tp to point at next unconverted input, and xpp to point |
9270 | | * at next location for converted output |
9271 | | */ |
9272 | | long i, j, ni; |
9273 | | short tmp[LOOPCNT]; /* in case input is misaligned */ |
9274 | | short *xp; |
9275 | | int nrange = 0; /* number of range errors */ |
9276 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9277 | | long cxp = (long) *((char**)xpp); |
9278 | | |
9279 | | realign = (cxp & 7) % SIZEOF_SHORT; |
9280 | | /* sjl: manually stripmine so we can limit amount of |
9281 | | * vector work space reserved to LOOPCNT elements. Also |
9282 | | * makes vectorisation easy */ |
9283 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9284 | | ni=Min(nelems-j,LOOPCNT); |
9285 | | if (realign) { |
9286 | | xp = tmp; |
9287 | | } else { |
9288 | | xp = (short *) *xpp; |
9289 | | } |
9290 | | /* copy the next block */ |
9291 | | #pragma cdir loopcnt=LOOPCNT |
9292 | | #pragma cdir shortloop |
9293 | | for (i=0; i<ni; i++) { |
9294 | | /* the normal case: */ |
9295 | | xp[i] = (short) Max( X_SHORT_MIN, Min(X_SHORT_MAX, (short) tp[i])); |
9296 | | /* test for range errors (not always needed but do it anyway) */ |
9297 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
9298 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
9299 | | nrange += tp[i] > X_SHORT_MAX ; |
9300 | | } |
9301 | | /* copy workspace back if necessary */ |
9302 | | if (realign) { |
9303 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_SHORT); |
9304 | | xp = (short *) *xpp; |
9305 | | } |
9306 | | /* update xpp and tp */ |
9307 | | xp += ni; |
9308 | | tp += ni; |
9309 | | *xpp = (void*)xp; |
9310 | | } |
9311 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9312 | | |
9313 | | #else /* not SX */ |
9314 | |
|
9315 | 0 | char *xp = (char *) *xpp; |
9316 | 0 | int status = NC_NOERR; |
9317 | |
|
9318 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9319 | 0 | { |
9320 | 0 | int lstatus = ncx_put_short_ushort(xp, tp, fillp); |
9321 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9322 | 0 | status = lstatus; |
9323 | 0 | } |
9324 | |
|
9325 | 0 | *xpp = (void *)xp; |
9326 | 0 | return status; |
9327 | 0 | #endif |
9328 | 0 | } |
9329 | | |
9330 | | |
9331 | | int |
9332 | | ncx_pad_putn_short_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
9333 | 0 | { |
9334 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9335 | |
|
9336 | 0 | char *xp = (char *) *xpp; |
9337 | 0 | int status = NC_NOERR; |
9338 | |
|
9339 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9340 | 0 | { |
9341 | 0 | int lstatus = ncx_put_short_schar(xp, tp, fillp); |
9342 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9343 | 0 | status = lstatus; |
9344 | 0 | } |
9345 | |
|
9346 | 0 | if (rndup != 0) |
9347 | 0 | { |
9348 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9349 | 0 | xp += X_SIZEOF_SHORT; |
9350 | 0 | } |
9351 | |
|
9352 | 0 | *xpp = (void *)xp; |
9353 | 0 | return status; |
9354 | 0 | } |
9355 | | |
9356 | | int |
9357 | | ncx_pad_putn_short_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
9358 | 0 | { |
9359 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9360 | |
|
9361 | 0 | char *xp = (char *) *xpp; |
9362 | 0 | int status = NC_NOERR; |
9363 | |
|
9364 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9365 | 0 | { |
9366 | 0 | int lstatus = ncx_put_short_uchar(xp, tp, fillp); |
9367 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9368 | 0 | status = lstatus; |
9369 | 0 | } |
9370 | |
|
9371 | 0 | if (rndup != 0) |
9372 | 0 | { |
9373 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9374 | 0 | xp += X_SIZEOF_SHORT; |
9375 | 0 | } |
9376 | |
|
9377 | 0 | *xpp = (void *)xp; |
9378 | 0 | return status; |
9379 | 0 | } |
9380 | | |
9381 | | int |
9382 | | ncx_pad_putn_short_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
9383 | 0 | { |
9384 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9385 | |
|
9386 | 0 | char *xp = (char *) *xpp; |
9387 | 0 | int status = NC_NOERR; |
9388 | |
|
9389 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9390 | 0 | { |
9391 | 0 | int lstatus = ncx_put_short_short(xp, tp, fillp); |
9392 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9393 | 0 | status = lstatus; |
9394 | 0 | } |
9395 | |
|
9396 | 0 | if (rndup != 0) |
9397 | 0 | { |
9398 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9399 | 0 | xp += X_SIZEOF_SHORT; |
9400 | 0 | } |
9401 | |
|
9402 | 0 | *xpp = (void *)xp; |
9403 | 0 | return status; |
9404 | 0 | } |
9405 | | |
9406 | | int |
9407 | | ncx_pad_putn_short_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
9408 | 0 | { |
9409 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9410 | |
|
9411 | 0 | char *xp = (char *) *xpp; |
9412 | 0 | int status = NC_NOERR; |
9413 | |
|
9414 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9415 | 0 | { |
9416 | 0 | int lstatus = ncx_put_short_int(xp, tp, fillp); |
9417 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9418 | 0 | status = lstatus; |
9419 | 0 | } |
9420 | |
|
9421 | 0 | if (rndup != 0) |
9422 | 0 | { |
9423 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9424 | 0 | xp += X_SIZEOF_SHORT; |
9425 | 0 | } |
9426 | |
|
9427 | 0 | *xpp = (void *)xp; |
9428 | 0 | return status; |
9429 | 0 | } |
9430 | | |
9431 | | int |
9432 | | ncx_pad_putn_short_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
9433 | 0 | { |
9434 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9435 | |
|
9436 | 0 | char *xp = (char *) *xpp; |
9437 | 0 | int status = NC_NOERR; |
9438 | |
|
9439 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9440 | 0 | { |
9441 | 0 | int lstatus = ncx_put_short_long(xp, tp, fillp); |
9442 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9443 | 0 | status = lstatus; |
9444 | 0 | } |
9445 | |
|
9446 | 0 | if (rndup != 0) |
9447 | 0 | { |
9448 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9449 | 0 | xp += X_SIZEOF_SHORT; |
9450 | 0 | } |
9451 | |
|
9452 | 0 | *xpp = (void *)xp; |
9453 | 0 | return status; |
9454 | 0 | } |
9455 | | |
9456 | | int |
9457 | | ncx_pad_putn_short_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
9458 | 0 | { |
9459 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9460 | |
|
9461 | 0 | char *xp = (char *) *xpp; |
9462 | 0 | int status = NC_NOERR; |
9463 | |
|
9464 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9465 | 0 | { |
9466 | 0 | int lstatus = ncx_put_short_float(xp, tp, fillp); |
9467 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9468 | 0 | status = lstatus; |
9469 | 0 | } |
9470 | |
|
9471 | 0 | if (rndup != 0) |
9472 | 0 | { |
9473 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9474 | 0 | xp += X_SIZEOF_SHORT; |
9475 | 0 | } |
9476 | |
|
9477 | 0 | *xpp = (void *)xp; |
9478 | 0 | return status; |
9479 | 0 | } |
9480 | | |
9481 | | int |
9482 | | ncx_pad_putn_short_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
9483 | 0 | { |
9484 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9485 | |
|
9486 | 0 | char *xp = (char *) *xpp; |
9487 | 0 | int status = NC_NOERR; |
9488 | |
|
9489 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9490 | 0 | { |
9491 | 0 | int lstatus = ncx_put_short_double(xp, tp, fillp); |
9492 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9493 | 0 | status = lstatus; |
9494 | 0 | } |
9495 | |
|
9496 | 0 | if (rndup != 0) |
9497 | 0 | { |
9498 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9499 | 0 | xp += X_SIZEOF_SHORT; |
9500 | 0 | } |
9501 | |
|
9502 | 0 | *xpp = (void *)xp; |
9503 | 0 | return status; |
9504 | 0 | } |
9505 | | |
9506 | | int |
9507 | | ncx_pad_putn_short_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
9508 | 0 | { |
9509 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9510 | |
|
9511 | 0 | char *xp = (char *) *xpp; |
9512 | 0 | int status = NC_NOERR; |
9513 | |
|
9514 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9515 | 0 | { |
9516 | 0 | int lstatus = ncx_put_short_uint(xp, tp, fillp); |
9517 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9518 | 0 | status = lstatus; |
9519 | 0 | } |
9520 | |
|
9521 | 0 | if (rndup != 0) |
9522 | 0 | { |
9523 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9524 | 0 | xp += X_SIZEOF_SHORT; |
9525 | 0 | } |
9526 | |
|
9527 | 0 | *xpp = (void *)xp; |
9528 | 0 | return status; |
9529 | 0 | } |
9530 | | |
9531 | | int |
9532 | | ncx_pad_putn_short_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
9533 | 0 | { |
9534 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9535 | |
|
9536 | 0 | char *xp = (char *) *xpp; |
9537 | 0 | int status = NC_NOERR; |
9538 | |
|
9539 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9540 | 0 | { |
9541 | 0 | int lstatus = ncx_put_short_longlong(xp, tp, fillp); |
9542 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9543 | 0 | status = lstatus; |
9544 | 0 | } |
9545 | |
|
9546 | 0 | if (rndup != 0) |
9547 | 0 | { |
9548 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9549 | 0 | xp += X_SIZEOF_SHORT; |
9550 | 0 | } |
9551 | |
|
9552 | 0 | *xpp = (void *)xp; |
9553 | 0 | return status; |
9554 | 0 | } |
9555 | | |
9556 | | int |
9557 | | ncx_pad_putn_short_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
9558 | 0 | { |
9559 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9560 | |
|
9561 | 0 | char *xp = (char *) *xpp; |
9562 | 0 | int status = NC_NOERR; |
9563 | |
|
9564 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9565 | 0 | { |
9566 | 0 | int lstatus = ncx_put_short_ulonglong(xp, tp, fillp); |
9567 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9568 | 0 | status = lstatus; |
9569 | 0 | } |
9570 | |
|
9571 | 0 | if (rndup != 0) |
9572 | 0 | { |
9573 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9574 | 0 | xp += X_SIZEOF_SHORT; |
9575 | 0 | } |
9576 | |
|
9577 | 0 | *xpp = (void *)xp; |
9578 | 0 | return status; |
9579 | 0 | } |
9580 | | |
9581 | | int |
9582 | | ncx_pad_putn_short_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
9583 | 0 | { |
9584 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
9585 | |
|
9586 | 0 | char *xp = (char *) *xpp; |
9587 | 0 | int status = NC_NOERR; |
9588 | |
|
9589 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_SHORT, tp++) |
9590 | 0 | { |
9591 | 0 | int lstatus = ncx_put_short_ushort(xp, tp, fillp); |
9592 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9593 | 0 | status = lstatus; |
9594 | 0 | } |
9595 | |
|
9596 | 0 | if (rndup != 0) |
9597 | 0 | { |
9598 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_SHORT)); |
9599 | 0 | xp += X_SIZEOF_SHORT; |
9600 | 0 | } |
9601 | |
|
9602 | 0 | *xpp = (void *)xp; |
9603 | 0 | return status; |
9604 | 0 | } |
9605 | | |
9606 | | |
9607 | | |
9608 | | /* ushort --------------------------------------------------------------------*/ |
9609 | | |
9610 | | #if X_SIZEOF_USHORT == SIZEOF_USHORT |
9611 | | /* optimized version */ |
9612 | | int |
9613 | | ncx_getn_ushort_ushort(const void **xpp, size_t nelems, unsigned short *tp) |
9614 | 0 | { |
9615 | | #ifdef WORDS_BIGENDIAN |
9616 | | (void) memcpy(tp, *xpp, (size_t)nelems * SIZEOF_USHORT); |
9617 | | # else |
9618 | 0 | swapn2b(tp, *xpp, nelems); |
9619 | 0 | # endif |
9620 | 0 | *xpp = (const void *)((const char *)(*xpp) + nelems * X_SIZEOF_USHORT); |
9621 | 0 | return NC_NOERR; |
9622 | 0 | } |
9623 | | #else |
9624 | | int |
9625 | | ncx_getn_ushort_ushort(const void **xpp, size_t nelems, ushort *tp) |
9626 | | { |
9627 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
9628 | | |
9629 | | /* basic algorithm is: |
9630 | | * - ensure sane alignment of input data |
9631 | | * - copy (conversion happens automatically) input data |
9632 | | * to output |
9633 | | * - update xpp to point at next unconverted input, and tp to point |
9634 | | * at next location for converted output |
9635 | | */ |
9636 | | long i, j, ni; |
9637 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
9638 | | ushort *xp; |
9639 | | int nrange = 0; /* number of range errors */ |
9640 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9641 | | long cxp = (long) *((char**)xpp); |
9642 | | |
9643 | | realign = (cxp & 7) % SIZEOF_USHORT; |
9644 | | /* sjl: manually stripmine so we can limit amount of |
9645 | | * vector work space reserved to LOOPCNT elements. Also |
9646 | | * makes vectorisation easy */ |
9647 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9648 | | ni=Min(nelems-j,LOOPCNT); |
9649 | | if (realign) { |
9650 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
9651 | | xp = tmp; |
9652 | | } else { |
9653 | | xp = (ushort *) *xpp; |
9654 | | } |
9655 | | /* copy the next block */ |
9656 | | #pragma cdir loopcnt=LOOPCNT |
9657 | | #pragma cdir shortloop |
9658 | | for (i=0; i<ni; i++) { |
9659 | | tp[i] = (ushort) Max( USHORT_MIN, Min(USHORT_MAX, (ushort) xp[i])); |
9660 | | /* test for range errors (not always needed but do it anyway) */ |
9661 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
9662 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
9663 | | nrange += xp[i] > USHORT_MAX ; |
9664 | | } |
9665 | | /* update xpp and tp */ |
9666 | | if (realign) xp = (ushort *) *xpp; |
9667 | | xp += ni; |
9668 | | tp += ni; |
9669 | | *xpp = (void*)xp; |
9670 | | } |
9671 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9672 | | |
9673 | | #else /* not SX */ |
9674 | | const char *xp = (const char *) *xpp; |
9675 | | int status = NC_NOERR; |
9676 | | |
9677 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
9678 | | { |
9679 | | const int lstatus = ncx_get_ushort_ushort(xp, tp); |
9680 | | if (status == NC_NOERR) /* report the first encountered error */ |
9681 | | status = lstatus; |
9682 | | } |
9683 | | |
9684 | | *xpp = (const void *)xp; |
9685 | | return status; |
9686 | | #endif |
9687 | | } |
9688 | | |
9689 | | #endif |
9690 | | int |
9691 | | ncx_getn_ushort_schar(const void **xpp, size_t nelems, schar *tp) |
9692 | 0 | { |
9693 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
9694 | | |
9695 | | /* basic algorithm is: |
9696 | | * - ensure sane alignment of input data |
9697 | | * - copy (conversion happens automatically) input data |
9698 | | * to output |
9699 | | * - update xpp to point at next unconverted input, and tp to point |
9700 | | * at next location for converted output |
9701 | | */ |
9702 | | long i, j, ni; |
9703 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
9704 | | ushort *xp; |
9705 | | int nrange = 0; /* number of range errors */ |
9706 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9707 | | long cxp = (long) *((char**)xpp); |
9708 | | |
9709 | | realign = (cxp & 7) % SIZEOF_USHORT; |
9710 | | /* sjl: manually stripmine so we can limit amount of |
9711 | | * vector work space reserved to LOOPCNT elements. Also |
9712 | | * makes vectorisation easy */ |
9713 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9714 | | ni=Min(nelems-j,LOOPCNT); |
9715 | | if (realign) { |
9716 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
9717 | | xp = tmp; |
9718 | | } else { |
9719 | | xp = (ushort *) *xpp; |
9720 | | } |
9721 | | /* copy the next block */ |
9722 | | #pragma cdir loopcnt=LOOPCNT |
9723 | | #pragma cdir shortloop |
9724 | | for (i=0; i<ni; i++) { |
9725 | | tp[i] = (schar) Max( SCHAR_MIN, Min(SCHAR_MAX, (schar) xp[i])); |
9726 | | /* test for range errors (not always needed but do it anyway) */ |
9727 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
9728 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
9729 | | nrange += xp[i] > SCHAR_MAX ; |
9730 | | } |
9731 | | /* update xpp and tp */ |
9732 | | if (realign) xp = (ushort *) *xpp; |
9733 | | xp += ni; |
9734 | | tp += ni; |
9735 | | *xpp = (void*)xp; |
9736 | | } |
9737 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9738 | | |
9739 | | #else /* not SX */ |
9740 | 0 | const char *xp = (const char *) *xpp; |
9741 | 0 | int status = NC_NOERR; |
9742 | |
|
9743 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
9744 | 0 | { |
9745 | 0 | const int lstatus = ncx_get_ushort_schar(xp, tp); |
9746 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9747 | 0 | status = lstatus; |
9748 | 0 | } |
9749 | |
|
9750 | 0 | *xpp = (const void *)xp; |
9751 | 0 | return status; |
9752 | 0 | #endif |
9753 | 0 | } |
9754 | | |
9755 | | int |
9756 | | ncx_getn_ushort_short(const void **xpp, size_t nelems, short *tp) |
9757 | 0 | { |
9758 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
9759 | | |
9760 | | /* basic algorithm is: |
9761 | | * - ensure sane alignment of input data |
9762 | | * - copy (conversion happens automatically) input data |
9763 | | * to output |
9764 | | * - update xpp to point at next unconverted input, and tp to point |
9765 | | * at next location for converted output |
9766 | | */ |
9767 | | long i, j, ni; |
9768 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
9769 | | ushort *xp; |
9770 | | int nrange = 0; /* number of range errors */ |
9771 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9772 | | long cxp = (long) *((char**)xpp); |
9773 | | |
9774 | | realign = (cxp & 7) % SIZEOF_USHORT; |
9775 | | /* sjl: manually stripmine so we can limit amount of |
9776 | | * vector work space reserved to LOOPCNT elements. Also |
9777 | | * makes vectorisation easy */ |
9778 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9779 | | ni=Min(nelems-j,LOOPCNT); |
9780 | | if (realign) { |
9781 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
9782 | | xp = tmp; |
9783 | | } else { |
9784 | | xp = (ushort *) *xpp; |
9785 | | } |
9786 | | /* copy the next block */ |
9787 | | #pragma cdir loopcnt=LOOPCNT |
9788 | | #pragma cdir shortloop |
9789 | | for (i=0; i<ni; i++) { |
9790 | | tp[i] = (short) Max( SHORT_MIN, Min(SHORT_MAX, (short) xp[i])); |
9791 | | /* test for range errors (not always needed but do it anyway) */ |
9792 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
9793 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
9794 | | nrange += xp[i] > SHORT_MAX ; |
9795 | | } |
9796 | | /* update xpp and tp */ |
9797 | | if (realign) xp = (ushort *) *xpp; |
9798 | | xp += ni; |
9799 | | tp += ni; |
9800 | | *xpp = (void*)xp; |
9801 | | } |
9802 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9803 | | |
9804 | | #else /* not SX */ |
9805 | 0 | const char *xp = (const char *) *xpp; |
9806 | 0 | int status = NC_NOERR; |
9807 | |
|
9808 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
9809 | 0 | { |
9810 | 0 | const int lstatus = ncx_get_ushort_short(xp, tp); |
9811 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9812 | 0 | status = lstatus; |
9813 | 0 | } |
9814 | |
|
9815 | 0 | *xpp = (const void *)xp; |
9816 | 0 | return status; |
9817 | 0 | #endif |
9818 | 0 | } |
9819 | | |
9820 | | int |
9821 | | ncx_getn_ushort_int(const void **xpp, size_t nelems, int *tp) |
9822 | 0 | { |
9823 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
9824 | | |
9825 | | /* basic algorithm is: |
9826 | | * - ensure sane alignment of input data |
9827 | | * - copy (conversion happens automatically) input data |
9828 | | * to output |
9829 | | * - update xpp to point at next unconverted input, and tp to point |
9830 | | * at next location for converted output |
9831 | | */ |
9832 | | long i, j, ni; |
9833 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
9834 | | ushort *xp; |
9835 | | int nrange = 0; /* number of range errors */ |
9836 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9837 | | long cxp = (long) *((char**)xpp); |
9838 | | |
9839 | | realign = (cxp & 7) % SIZEOF_USHORT; |
9840 | | /* sjl: manually stripmine so we can limit amount of |
9841 | | * vector work space reserved to LOOPCNT elements. Also |
9842 | | * makes vectorisation easy */ |
9843 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9844 | | ni=Min(nelems-j,LOOPCNT); |
9845 | | if (realign) { |
9846 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
9847 | | xp = tmp; |
9848 | | } else { |
9849 | | xp = (ushort *) *xpp; |
9850 | | } |
9851 | | /* copy the next block */ |
9852 | | #pragma cdir loopcnt=LOOPCNT |
9853 | | #pragma cdir shortloop |
9854 | | for (i=0; i<ni; i++) { |
9855 | | tp[i] = (int) Max( INT_MIN, Min(INT_MAX, (int) xp[i])); |
9856 | | /* test for range errors (not always needed but do it anyway) */ |
9857 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
9858 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
9859 | | nrange += xp[i] > INT_MAX ; |
9860 | | } |
9861 | | /* update xpp and tp */ |
9862 | | if (realign) xp = (ushort *) *xpp; |
9863 | | xp += ni; |
9864 | | tp += ni; |
9865 | | *xpp = (void*)xp; |
9866 | | } |
9867 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9868 | | |
9869 | | #else /* not SX */ |
9870 | 0 | const char *xp = (const char *) *xpp; |
9871 | 0 | int status = NC_NOERR; |
9872 | |
|
9873 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
9874 | 0 | { |
9875 | 0 | const int lstatus = ncx_get_ushort_int(xp, tp); |
9876 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9877 | 0 | status = lstatus; |
9878 | 0 | } |
9879 | |
|
9880 | 0 | *xpp = (const void *)xp; |
9881 | 0 | return status; |
9882 | 0 | #endif |
9883 | 0 | } |
9884 | | |
9885 | | int |
9886 | | ncx_getn_ushort_long(const void **xpp, size_t nelems, long *tp) |
9887 | 0 | { |
9888 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
9889 | | |
9890 | | /* basic algorithm is: |
9891 | | * - ensure sane alignment of input data |
9892 | | * - copy (conversion happens automatically) input data |
9893 | | * to output |
9894 | | * - update xpp to point at next unconverted input, and tp to point |
9895 | | * at next location for converted output |
9896 | | */ |
9897 | | long i, j, ni; |
9898 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
9899 | | ushort *xp; |
9900 | | int nrange = 0; /* number of range errors */ |
9901 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9902 | | long cxp = (long) *((char**)xpp); |
9903 | | |
9904 | | realign = (cxp & 7) % SIZEOF_USHORT; |
9905 | | /* sjl: manually stripmine so we can limit amount of |
9906 | | * vector work space reserved to LOOPCNT elements. Also |
9907 | | * makes vectorisation easy */ |
9908 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9909 | | ni=Min(nelems-j,LOOPCNT); |
9910 | | if (realign) { |
9911 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
9912 | | xp = tmp; |
9913 | | } else { |
9914 | | xp = (ushort *) *xpp; |
9915 | | } |
9916 | | /* copy the next block */ |
9917 | | #pragma cdir loopcnt=LOOPCNT |
9918 | | #pragma cdir shortloop |
9919 | | for (i=0; i<ni; i++) { |
9920 | | tp[i] = (long) Max( LONG_MIN, Min(LONG_MAX, (long) xp[i])); |
9921 | | /* test for range errors (not always needed but do it anyway) */ |
9922 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
9923 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
9924 | | nrange += xp[i] > LONG_MAX ; |
9925 | | } |
9926 | | /* update xpp and tp */ |
9927 | | if (realign) xp = (ushort *) *xpp; |
9928 | | xp += ni; |
9929 | | tp += ni; |
9930 | | *xpp = (void*)xp; |
9931 | | } |
9932 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9933 | | |
9934 | | #else /* not SX */ |
9935 | 0 | const char *xp = (const char *) *xpp; |
9936 | 0 | int status = NC_NOERR; |
9937 | |
|
9938 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
9939 | 0 | { |
9940 | 0 | const int lstatus = ncx_get_ushort_long(xp, tp); |
9941 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
9942 | 0 | status = lstatus; |
9943 | 0 | } |
9944 | |
|
9945 | 0 | *xpp = (const void *)xp; |
9946 | 0 | return status; |
9947 | 0 | #endif |
9948 | 0 | } |
9949 | | |
9950 | | int |
9951 | | ncx_getn_ushort_float(const void **xpp, size_t nelems, float *tp) |
9952 | 0 | { |
9953 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
9954 | | |
9955 | | /* basic algorithm is: |
9956 | | * - ensure sane alignment of input data |
9957 | | * - copy (conversion happens automatically) input data |
9958 | | * to output |
9959 | | * - update xpp to point at next unconverted input, and tp to point |
9960 | | * at next location for converted output |
9961 | | */ |
9962 | | long i, j, ni; |
9963 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
9964 | | ushort *xp; |
9965 | | int nrange = 0; /* number of range errors */ |
9966 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
9967 | | long cxp = (long) *((char**)xpp); |
9968 | | |
9969 | | realign = (cxp & 7) % SIZEOF_USHORT; |
9970 | | /* sjl: manually stripmine so we can limit amount of |
9971 | | * vector work space reserved to LOOPCNT elements. Also |
9972 | | * makes vectorisation easy */ |
9973 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
9974 | | ni=Min(nelems-j,LOOPCNT); |
9975 | | if (realign) { |
9976 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
9977 | | xp = tmp; |
9978 | | } else { |
9979 | | xp = (ushort *) *xpp; |
9980 | | } |
9981 | | /* copy the next block */ |
9982 | | #pragma cdir loopcnt=LOOPCNT |
9983 | | #pragma cdir shortloop |
9984 | | for (i=0; i<ni; i++) { |
9985 | | tp[i] = (float) Max( FLOAT_MIN, Min(FLOAT_MAX, (float) xp[i])); |
9986 | | /* test for range errors (not always needed but do it anyway) */ |
9987 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
9988 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
9989 | | nrange += xp[i] > FLOAT_MAX ; |
9990 | | } |
9991 | | /* update xpp and tp */ |
9992 | | if (realign) xp = (ushort *) *xpp; |
9993 | | xp += ni; |
9994 | | tp += ni; |
9995 | | *xpp = (void*)xp; |
9996 | | } |
9997 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
9998 | | |
9999 | | #else /* not SX */ |
10000 | 0 | const char *xp = (const char *) *xpp; |
10001 | 0 | int status = NC_NOERR; |
10002 | |
|
10003 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10004 | 0 | { |
10005 | 0 | const int lstatus = ncx_get_ushort_float(xp, tp); |
10006 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10007 | 0 | status = lstatus; |
10008 | 0 | } |
10009 | |
|
10010 | 0 | *xpp = (const void *)xp; |
10011 | 0 | return status; |
10012 | 0 | #endif |
10013 | 0 | } |
10014 | | |
10015 | | int |
10016 | | ncx_getn_ushort_double(const void **xpp, size_t nelems, double *tp) |
10017 | 0 | { |
10018 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10019 | | |
10020 | | /* basic algorithm is: |
10021 | | * - ensure sane alignment of input data |
10022 | | * - copy (conversion happens automatically) input data |
10023 | | * to output |
10024 | | * - update xpp to point at next unconverted input, and tp to point |
10025 | | * at next location for converted output |
10026 | | */ |
10027 | | long i, j, ni; |
10028 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10029 | | ushort *xp; |
10030 | | int nrange = 0; /* number of range errors */ |
10031 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10032 | | long cxp = (long) *((char**)xpp); |
10033 | | |
10034 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10035 | | /* sjl: manually stripmine so we can limit amount of |
10036 | | * vector work space reserved to LOOPCNT elements. Also |
10037 | | * makes vectorisation easy */ |
10038 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10039 | | ni=Min(nelems-j,LOOPCNT); |
10040 | | if (realign) { |
10041 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
10042 | | xp = tmp; |
10043 | | } else { |
10044 | | xp = (ushort *) *xpp; |
10045 | | } |
10046 | | /* copy the next block */ |
10047 | | #pragma cdir loopcnt=LOOPCNT |
10048 | | #pragma cdir shortloop |
10049 | | for (i=0; i<ni; i++) { |
10050 | | tp[i] = (double) Max( DOUBLE_MIN, Min(DOUBLE_MAX, (double) xp[i])); |
10051 | | /* test for range errors (not always needed but do it anyway) */ |
10052 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
10053 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
10054 | | nrange += xp[i] > DOUBLE_MAX ; |
10055 | | } |
10056 | | /* update xpp and tp */ |
10057 | | if (realign) xp = (ushort *) *xpp; |
10058 | | xp += ni; |
10059 | | tp += ni; |
10060 | | *xpp = (void*)xp; |
10061 | | } |
10062 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10063 | | |
10064 | | #else /* not SX */ |
10065 | 0 | const char *xp = (const char *) *xpp; |
10066 | 0 | int status = NC_NOERR; |
10067 | |
|
10068 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10069 | 0 | { |
10070 | 0 | const int lstatus = ncx_get_ushort_double(xp, tp); |
10071 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10072 | 0 | status = lstatus; |
10073 | 0 | } |
10074 | |
|
10075 | 0 | *xpp = (const void *)xp; |
10076 | 0 | return status; |
10077 | 0 | #endif |
10078 | 0 | } |
10079 | | |
10080 | | int |
10081 | | ncx_getn_ushort_longlong(const void **xpp, size_t nelems, longlong *tp) |
10082 | 0 | { |
10083 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10084 | | |
10085 | | /* basic algorithm is: |
10086 | | * - ensure sane alignment of input data |
10087 | | * - copy (conversion happens automatically) input data |
10088 | | * to output |
10089 | | * - update xpp to point at next unconverted input, and tp to point |
10090 | | * at next location for converted output |
10091 | | */ |
10092 | | long i, j, ni; |
10093 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10094 | | ushort *xp; |
10095 | | int nrange = 0; /* number of range errors */ |
10096 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10097 | | long cxp = (long) *((char**)xpp); |
10098 | | |
10099 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10100 | | /* sjl: manually stripmine so we can limit amount of |
10101 | | * vector work space reserved to LOOPCNT elements. Also |
10102 | | * makes vectorisation easy */ |
10103 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10104 | | ni=Min(nelems-j,LOOPCNT); |
10105 | | if (realign) { |
10106 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
10107 | | xp = tmp; |
10108 | | } else { |
10109 | | xp = (ushort *) *xpp; |
10110 | | } |
10111 | | /* copy the next block */ |
10112 | | #pragma cdir loopcnt=LOOPCNT |
10113 | | #pragma cdir shortloop |
10114 | | for (i=0; i<ni; i++) { |
10115 | | tp[i] = (longlong) Max( LONGLONG_MIN, Min(LONGLONG_MAX, (longlong) xp[i])); |
10116 | | /* test for range errors (not always needed but do it anyway) */ |
10117 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
10118 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
10119 | | nrange += xp[i] > LONGLONG_MAX ; |
10120 | | } |
10121 | | /* update xpp and tp */ |
10122 | | if (realign) xp = (ushort *) *xpp; |
10123 | | xp += ni; |
10124 | | tp += ni; |
10125 | | *xpp = (void*)xp; |
10126 | | } |
10127 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10128 | | |
10129 | | #else /* not SX */ |
10130 | 0 | const char *xp = (const char *) *xpp; |
10131 | 0 | int status = NC_NOERR; |
10132 | |
|
10133 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10134 | 0 | { |
10135 | 0 | const int lstatus = ncx_get_ushort_longlong(xp, tp); |
10136 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10137 | 0 | status = lstatus; |
10138 | 0 | } |
10139 | |
|
10140 | 0 | *xpp = (const void *)xp; |
10141 | 0 | return status; |
10142 | 0 | #endif |
10143 | 0 | } |
10144 | | |
10145 | | int |
10146 | | ncx_getn_ushort_uchar(const void **xpp, size_t nelems, uchar *tp) |
10147 | 0 | { |
10148 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10149 | | |
10150 | | /* basic algorithm is: |
10151 | | * - ensure sane alignment of input data |
10152 | | * - copy (conversion happens automatically) input data |
10153 | | * to output |
10154 | | * - update xpp to point at next unconverted input, and tp to point |
10155 | | * at next location for converted output |
10156 | | */ |
10157 | | long i, j, ni; |
10158 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10159 | | ushort *xp; |
10160 | | int nrange = 0; /* number of range errors */ |
10161 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10162 | | long cxp = (long) *((char**)xpp); |
10163 | | |
10164 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10165 | | /* sjl: manually stripmine so we can limit amount of |
10166 | | * vector work space reserved to LOOPCNT elements. Also |
10167 | | * makes vectorisation easy */ |
10168 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10169 | | ni=Min(nelems-j,LOOPCNT); |
10170 | | if (realign) { |
10171 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
10172 | | xp = tmp; |
10173 | | } else { |
10174 | | xp = (ushort *) *xpp; |
10175 | | } |
10176 | | /* copy the next block */ |
10177 | | #pragma cdir loopcnt=LOOPCNT |
10178 | | #pragma cdir shortloop |
10179 | | for (i=0; i<ni; i++) { |
10180 | | tp[i] = (uchar) Max( UCHAR_MIN, Min(UCHAR_MAX, (uchar) xp[i])); |
10181 | | /* test for range errors (not always needed but do it anyway) */ |
10182 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
10183 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
10184 | | nrange += xp[i] > UCHAR_MAX ; |
10185 | | } |
10186 | | /* update xpp and tp */ |
10187 | | if (realign) xp = (ushort *) *xpp; |
10188 | | xp += ni; |
10189 | | tp += ni; |
10190 | | *xpp = (void*)xp; |
10191 | | } |
10192 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10193 | | |
10194 | | #else /* not SX */ |
10195 | 0 | const char *xp = (const char *) *xpp; |
10196 | 0 | int status = NC_NOERR; |
10197 | |
|
10198 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10199 | 0 | { |
10200 | 0 | const int lstatus = ncx_get_ushort_uchar(xp, tp); |
10201 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10202 | 0 | status = lstatus; |
10203 | 0 | } |
10204 | |
|
10205 | 0 | *xpp = (const void *)xp; |
10206 | 0 | return status; |
10207 | 0 | #endif |
10208 | 0 | } |
10209 | | |
10210 | | int |
10211 | | ncx_getn_ushort_uint(const void **xpp, size_t nelems, uint *tp) |
10212 | 0 | { |
10213 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10214 | | |
10215 | | /* basic algorithm is: |
10216 | | * - ensure sane alignment of input data |
10217 | | * - copy (conversion happens automatically) input data |
10218 | | * to output |
10219 | | * - update xpp to point at next unconverted input, and tp to point |
10220 | | * at next location for converted output |
10221 | | */ |
10222 | | long i, j, ni; |
10223 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10224 | | ushort *xp; |
10225 | | int nrange = 0; /* number of range errors */ |
10226 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10227 | | long cxp = (long) *((char**)xpp); |
10228 | | |
10229 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10230 | | /* sjl: manually stripmine so we can limit amount of |
10231 | | * vector work space reserved to LOOPCNT elements. Also |
10232 | | * makes vectorisation easy */ |
10233 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10234 | | ni=Min(nelems-j,LOOPCNT); |
10235 | | if (realign) { |
10236 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
10237 | | xp = tmp; |
10238 | | } else { |
10239 | | xp = (ushort *) *xpp; |
10240 | | } |
10241 | | /* copy the next block */ |
10242 | | #pragma cdir loopcnt=LOOPCNT |
10243 | | #pragma cdir shortloop |
10244 | | for (i=0; i<ni; i++) { |
10245 | | tp[i] = (uint) Max( UINT_MIN, Min(UINT_MAX, (uint) xp[i])); |
10246 | | /* test for range errors (not always needed but do it anyway) */ |
10247 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
10248 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
10249 | | nrange += xp[i] > UINT_MAX ; |
10250 | | } |
10251 | | /* update xpp and tp */ |
10252 | | if (realign) xp = (ushort *) *xpp; |
10253 | | xp += ni; |
10254 | | tp += ni; |
10255 | | *xpp = (void*)xp; |
10256 | | } |
10257 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10258 | | |
10259 | | #else /* not SX */ |
10260 | 0 | const char *xp = (const char *) *xpp; |
10261 | 0 | int status = NC_NOERR; |
10262 | |
|
10263 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10264 | 0 | { |
10265 | 0 | const int lstatus = ncx_get_ushort_uint(xp, tp); |
10266 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10267 | 0 | status = lstatus; |
10268 | 0 | } |
10269 | |
|
10270 | 0 | *xpp = (const void *)xp; |
10271 | 0 | return status; |
10272 | 0 | #endif |
10273 | 0 | } |
10274 | | |
10275 | | int |
10276 | | ncx_getn_ushort_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
10277 | 0 | { |
10278 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10279 | | |
10280 | | /* basic algorithm is: |
10281 | | * - ensure sane alignment of input data |
10282 | | * - copy (conversion happens automatically) input data |
10283 | | * to output |
10284 | | * - update xpp to point at next unconverted input, and tp to point |
10285 | | * at next location for converted output |
10286 | | */ |
10287 | | long i, j, ni; |
10288 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10289 | | ushort *xp; |
10290 | | int nrange = 0; /* number of range errors */ |
10291 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10292 | | long cxp = (long) *((char**)xpp); |
10293 | | |
10294 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10295 | | /* sjl: manually stripmine so we can limit amount of |
10296 | | * vector work space reserved to LOOPCNT elements. Also |
10297 | | * makes vectorisation easy */ |
10298 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10299 | | ni=Min(nelems-j,LOOPCNT); |
10300 | | if (realign) { |
10301 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_USHORT)); |
10302 | | xp = tmp; |
10303 | | } else { |
10304 | | xp = (ushort *) *xpp; |
10305 | | } |
10306 | | /* copy the next block */ |
10307 | | #pragma cdir loopcnt=LOOPCNT |
10308 | | #pragma cdir shortloop |
10309 | | for (i=0; i<ni; i++) { |
10310 | | tp[i] = (ulonglong) Max( ULONGLONG_MIN, Min(ULONGLONG_MAX, (ulonglong) xp[i])); |
10311 | | /* test for range errors (not always needed but do it anyway) */ |
10312 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
10313 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
10314 | | nrange += xp[i] > ULONGLONG_MAX ; |
10315 | | } |
10316 | | /* update xpp and tp */ |
10317 | | if (realign) xp = (ushort *) *xpp; |
10318 | | xp += ni; |
10319 | | tp += ni; |
10320 | | *xpp = (void*)xp; |
10321 | | } |
10322 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10323 | | |
10324 | | #else /* not SX */ |
10325 | 0 | const char *xp = (const char *) *xpp; |
10326 | 0 | int status = NC_NOERR; |
10327 | |
|
10328 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10329 | 0 | { |
10330 | 0 | const int lstatus = ncx_get_ushort_ulonglong(xp, tp); |
10331 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10332 | 0 | status = lstatus; |
10333 | 0 | } |
10334 | |
|
10335 | 0 | *xpp = (const void *)xp; |
10336 | 0 | return status; |
10337 | 0 | #endif |
10338 | 0 | } |
10339 | | |
10340 | | |
10341 | | int |
10342 | | ncx_pad_getn_ushort_schar(const void **xpp, size_t nelems, schar *tp) |
10343 | 0 | { |
10344 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10345 | |
|
10346 | 0 | const char *xp = (const char *) *xpp; |
10347 | 0 | int status = NC_NOERR; |
10348 | |
|
10349 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10350 | 0 | { |
10351 | 0 | const int lstatus = ncx_get_ushort_schar(xp, tp); |
10352 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10353 | 0 | status = lstatus; |
10354 | 0 | } |
10355 | |
|
10356 | 0 | if (rndup != 0) |
10357 | 0 | xp += X_SIZEOF_USHORT; |
10358 | |
|
10359 | 0 | *xpp = (void *)xp; |
10360 | 0 | return status; |
10361 | 0 | } |
10362 | | |
10363 | | int |
10364 | | ncx_pad_getn_ushort_short(const void **xpp, size_t nelems, short *tp) |
10365 | 0 | { |
10366 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10367 | |
|
10368 | 0 | const char *xp = (const char *) *xpp; |
10369 | 0 | int status = NC_NOERR; |
10370 | |
|
10371 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10372 | 0 | { |
10373 | 0 | const int lstatus = ncx_get_ushort_short(xp, tp); |
10374 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10375 | 0 | status = lstatus; |
10376 | 0 | } |
10377 | |
|
10378 | 0 | if (rndup != 0) |
10379 | 0 | xp += X_SIZEOF_USHORT; |
10380 | |
|
10381 | 0 | *xpp = (void *)xp; |
10382 | 0 | return status; |
10383 | 0 | } |
10384 | | |
10385 | | int |
10386 | | ncx_pad_getn_ushort_int(const void **xpp, size_t nelems, int *tp) |
10387 | 0 | { |
10388 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10389 | |
|
10390 | 0 | const char *xp = (const char *) *xpp; |
10391 | 0 | int status = NC_NOERR; |
10392 | |
|
10393 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10394 | 0 | { |
10395 | 0 | const int lstatus = ncx_get_ushort_int(xp, tp); |
10396 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10397 | 0 | status = lstatus; |
10398 | 0 | } |
10399 | |
|
10400 | 0 | if (rndup != 0) |
10401 | 0 | xp += X_SIZEOF_USHORT; |
10402 | |
|
10403 | 0 | *xpp = (void *)xp; |
10404 | 0 | return status; |
10405 | 0 | } |
10406 | | |
10407 | | int |
10408 | | ncx_pad_getn_ushort_long(const void **xpp, size_t nelems, long *tp) |
10409 | 0 | { |
10410 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10411 | |
|
10412 | 0 | const char *xp = (const char *) *xpp; |
10413 | 0 | int status = NC_NOERR; |
10414 | |
|
10415 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10416 | 0 | { |
10417 | 0 | const int lstatus = ncx_get_ushort_long(xp, tp); |
10418 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10419 | 0 | status = lstatus; |
10420 | 0 | } |
10421 | |
|
10422 | 0 | if (rndup != 0) |
10423 | 0 | xp += X_SIZEOF_USHORT; |
10424 | |
|
10425 | 0 | *xpp = (void *)xp; |
10426 | 0 | return status; |
10427 | 0 | } |
10428 | | |
10429 | | int |
10430 | | ncx_pad_getn_ushort_float(const void **xpp, size_t nelems, float *tp) |
10431 | 0 | { |
10432 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10433 | |
|
10434 | 0 | const char *xp = (const char *) *xpp; |
10435 | 0 | int status = NC_NOERR; |
10436 | |
|
10437 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10438 | 0 | { |
10439 | 0 | const int lstatus = ncx_get_ushort_float(xp, tp); |
10440 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10441 | 0 | status = lstatus; |
10442 | 0 | } |
10443 | |
|
10444 | 0 | if (rndup != 0) |
10445 | 0 | xp += X_SIZEOF_USHORT; |
10446 | |
|
10447 | 0 | *xpp = (void *)xp; |
10448 | 0 | return status; |
10449 | 0 | } |
10450 | | |
10451 | | int |
10452 | | ncx_pad_getn_ushort_double(const void **xpp, size_t nelems, double *tp) |
10453 | 0 | { |
10454 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10455 | |
|
10456 | 0 | const char *xp = (const char *) *xpp; |
10457 | 0 | int status = NC_NOERR; |
10458 | |
|
10459 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10460 | 0 | { |
10461 | 0 | const int lstatus = ncx_get_ushort_double(xp, tp); |
10462 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10463 | 0 | status = lstatus; |
10464 | 0 | } |
10465 | |
|
10466 | 0 | if (rndup != 0) |
10467 | 0 | xp += X_SIZEOF_USHORT; |
10468 | |
|
10469 | 0 | *xpp = (void *)xp; |
10470 | 0 | return status; |
10471 | 0 | } |
10472 | | |
10473 | | int |
10474 | | ncx_pad_getn_ushort_uchar(const void **xpp, size_t nelems, uchar *tp) |
10475 | 0 | { |
10476 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10477 | |
|
10478 | 0 | const char *xp = (const char *) *xpp; |
10479 | 0 | int status = NC_NOERR; |
10480 | |
|
10481 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10482 | 0 | { |
10483 | 0 | const int lstatus = ncx_get_ushort_uchar(xp, tp); |
10484 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10485 | 0 | status = lstatus; |
10486 | 0 | } |
10487 | |
|
10488 | 0 | if (rndup != 0) |
10489 | 0 | xp += X_SIZEOF_USHORT; |
10490 | |
|
10491 | 0 | *xpp = (void *)xp; |
10492 | 0 | return status; |
10493 | 0 | } |
10494 | | |
10495 | | int |
10496 | | ncx_pad_getn_ushort_ushort(const void **xpp, size_t nelems, ushort *tp) |
10497 | 0 | { |
10498 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10499 | |
|
10500 | 0 | const char *xp = (const char *) *xpp; |
10501 | 0 | int status = NC_NOERR; |
10502 | |
|
10503 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10504 | 0 | { |
10505 | 0 | const int lstatus = ncx_get_ushort_ushort(xp, tp); |
10506 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10507 | 0 | status = lstatus; |
10508 | 0 | } |
10509 | |
|
10510 | 0 | if (rndup != 0) |
10511 | 0 | xp += X_SIZEOF_USHORT; |
10512 | |
|
10513 | 0 | *xpp = (void *)xp; |
10514 | 0 | return status; |
10515 | 0 | } |
10516 | | |
10517 | | int |
10518 | | ncx_pad_getn_ushort_uint(const void **xpp, size_t nelems, uint *tp) |
10519 | 0 | { |
10520 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10521 | |
|
10522 | 0 | const char *xp = (const char *) *xpp; |
10523 | 0 | int status = NC_NOERR; |
10524 | |
|
10525 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10526 | 0 | { |
10527 | 0 | const int lstatus = ncx_get_ushort_uint(xp, tp); |
10528 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10529 | 0 | status = lstatus; |
10530 | 0 | } |
10531 | |
|
10532 | 0 | if (rndup != 0) |
10533 | 0 | xp += X_SIZEOF_USHORT; |
10534 | |
|
10535 | 0 | *xpp = (void *)xp; |
10536 | 0 | return status; |
10537 | 0 | } |
10538 | | |
10539 | | int |
10540 | | ncx_pad_getn_ushort_longlong(const void **xpp, size_t nelems, longlong *tp) |
10541 | 0 | { |
10542 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10543 | |
|
10544 | 0 | const char *xp = (const char *) *xpp; |
10545 | 0 | int status = NC_NOERR; |
10546 | |
|
10547 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10548 | 0 | { |
10549 | 0 | const int lstatus = ncx_get_ushort_longlong(xp, tp); |
10550 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10551 | 0 | status = lstatus; |
10552 | 0 | } |
10553 | |
|
10554 | 0 | if (rndup != 0) |
10555 | 0 | xp += X_SIZEOF_USHORT; |
10556 | |
|
10557 | 0 | *xpp = (void *)xp; |
10558 | 0 | return status; |
10559 | 0 | } |
10560 | | |
10561 | | int |
10562 | | ncx_pad_getn_ushort_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
10563 | 0 | { |
10564 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
10565 | |
|
10566 | 0 | const char *xp = (const char *) *xpp; |
10567 | 0 | int status = NC_NOERR; |
10568 | |
|
10569 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10570 | 0 | { |
10571 | 0 | const int lstatus = ncx_get_ushort_ulonglong(xp, tp); |
10572 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10573 | 0 | status = lstatus; |
10574 | 0 | } |
10575 | |
|
10576 | 0 | if (rndup != 0) |
10577 | 0 | xp += X_SIZEOF_USHORT; |
10578 | |
|
10579 | 0 | *xpp = (void *)xp; |
10580 | 0 | return status; |
10581 | 0 | } |
10582 | | |
10583 | | |
10584 | | #if X_SIZEOF_USHORT == SIZEOF_USHORT |
10585 | | /* optimized version */ |
10586 | | int |
10587 | | ncx_putn_ushort_ushort(void **xpp, size_t nelems, const unsigned short *tp, void *fillp) |
10588 | 0 | { |
10589 | | #ifdef WORDS_BIGENDIAN |
10590 | | (void) memcpy(*xpp, tp, (size_t)nelems * X_SIZEOF_USHORT); |
10591 | | # else |
10592 | 0 | swapn2b(*xpp, tp, nelems); |
10593 | 0 | # endif |
10594 | 0 | *xpp = (void *)((char *)(*xpp) + nelems * X_SIZEOF_USHORT); |
10595 | 0 | return NC_NOERR; |
10596 | 0 | } |
10597 | | #else |
10598 | | int |
10599 | | ncx_putn_ushort_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
10600 | | { |
10601 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10602 | | |
10603 | | /* basic algorithm is: |
10604 | | * - ensure sane alignment of output data |
10605 | | * - copy (conversion happens automatically) input data |
10606 | | * to output |
10607 | | * - update tp to point at next unconverted input, and xpp to point |
10608 | | * at next location for converted output |
10609 | | */ |
10610 | | long i, j, ni; |
10611 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10612 | | ushort *xp; |
10613 | | int nrange = 0; /* number of range errors */ |
10614 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10615 | | long cxp = (long) *((char**)xpp); |
10616 | | |
10617 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10618 | | /* sjl: manually stripmine so we can limit amount of |
10619 | | * vector work space reserved to LOOPCNT elements. Also |
10620 | | * makes vectorisation easy */ |
10621 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10622 | | ni=Min(nelems-j,LOOPCNT); |
10623 | | if (realign) { |
10624 | | xp = tmp; |
10625 | | } else { |
10626 | | xp = (ushort *) *xpp; |
10627 | | } |
10628 | | /* copy the next block */ |
10629 | | #pragma cdir loopcnt=LOOPCNT |
10630 | | #pragma cdir shortloop |
10631 | | for (i=0; i<ni; i++) { |
10632 | | /* the normal case: */ |
10633 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
10634 | | /* test for range errors (not always needed but do it anyway) */ |
10635 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
10636 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
10637 | | nrange += tp[i] > X_USHORT_MAX ; |
10638 | | } |
10639 | | /* copy workspace back if necessary */ |
10640 | | if (realign) { |
10641 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
10642 | | xp = (ushort *) *xpp; |
10643 | | } |
10644 | | /* update xpp and tp */ |
10645 | | xp += ni; |
10646 | | tp += ni; |
10647 | | *xpp = (void*)xp; |
10648 | | } |
10649 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10650 | | |
10651 | | #else /* not SX */ |
10652 | | |
10653 | | char *xp = (char *) *xpp; |
10654 | | int status = NC_NOERR; |
10655 | | |
10656 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10657 | | { |
10658 | | int lstatus = ncx_put_ushort_ushort(xp, tp, fillp); |
10659 | | if (status == NC_NOERR) /* report the first encountered error */ |
10660 | | status = lstatus; |
10661 | | } |
10662 | | |
10663 | | *xpp = (void *)xp; |
10664 | | return status; |
10665 | | #endif |
10666 | | } |
10667 | | |
10668 | | #endif |
10669 | | int |
10670 | | ncx_putn_ushort_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
10671 | 0 | { |
10672 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10673 | | |
10674 | | /* basic algorithm is: |
10675 | | * - ensure sane alignment of output data |
10676 | | * - copy (conversion happens automatically) input data |
10677 | | * to output |
10678 | | * - update tp to point at next unconverted input, and xpp to point |
10679 | | * at next location for converted output |
10680 | | */ |
10681 | | long i, j, ni; |
10682 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10683 | | ushort *xp; |
10684 | | int nrange = 0; /* number of range errors */ |
10685 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10686 | | long cxp = (long) *((char**)xpp); |
10687 | | |
10688 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10689 | | /* sjl: manually stripmine so we can limit amount of |
10690 | | * vector work space reserved to LOOPCNT elements. Also |
10691 | | * makes vectorisation easy */ |
10692 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10693 | | ni=Min(nelems-j,LOOPCNT); |
10694 | | if (realign) { |
10695 | | xp = tmp; |
10696 | | } else { |
10697 | | xp = (ushort *) *xpp; |
10698 | | } |
10699 | | /* copy the next block */ |
10700 | | #pragma cdir loopcnt=LOOPCNT |
10701 | | #pragma cdir shortloop |
10702 | | for (i=0; i<ni; i++) { |
10703 | | /* the normal case: */ |
10704 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
10705 | | /* test for range errors (not always needed but do it anyway) */ |
10706 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
10707 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
10708 | | nrange += tp[i] > X_USHORT_MAX || tp[i] < 0; |
10709 | | } |
10710 | | /* copy workspace back if necessary */ |
10711 | | if (realign) { |
10712 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
10713 | | xp = (ushort *) *xpp; |
10714 | | } |
10715 | | /* update xpp and tp */ |
10716 | | xp += ni; |
10717 | | tp += ni; |
10718 | | *xpp = (void*)xp; |
10719 | | } |
10720 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10721 | | |
10722 | | #else /* not SX */ |
10723 | |
|
10724 | 0 | char *xp = (char *) *xpp; |
10725 | 0 | int status = NC_NOERR; |
10726 | |
|
10727 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10728 | 0 | { |
10729 | 0 | int lstatus = ncx_put_ushort_schar(xp, tp, fillp); |
10730 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10731 | 0 | status = lstatus; |
10732 | 0 | } |
10733 | |
|
10734 | 0 | *xpp = (void *)xp; |
10735 | 0 | return status; |
10736 | 0 | #endif |
10737 | 0 | } |
10738 | | |
10739 | | int |
10740 | | ncx_putn_ushort_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
10741 | 0 | { |
10742 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10743 | | |
10744 | | /* basic algorithm is: |
10745 | | * - ensure sane alignment of output data |
10746 | | * - copy (conversion happens automatically) input data |
10747 | | * to output |
10748 | | * - update tp to point at next unconverted input, and xpp to point |
10749 | | * at next location for converted output |
10750 | | */ |
10751 | | long i, j, ni; |
10752 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10753 | | ushort *xp; |
10754 | | int nrange = 0; /* number of range errors */ |
10755 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10756 | | long cxp = (long) *((char**)xpp); |
10757 | | |
10758 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10759 | | /* sjl: manually stripmine so we can limit amount of |
10760 | | * vector work space reserved to LOOPCNT elements. Also |
10761 | | * makes vectorisation easy */ |
10762 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10763 | | ni=Min(nelems-j,LOOPCNT); |
10764 | | if (realign) { |
10765 | | xp = tmp; |
10766 | | } else { |
10767 | | xp = (ushort *) *xpp; |
10768 | | } |
10769 | | /* copy the next block */ |
10770 | | #pragma cdir loopcnt=LOOPCNT |
10771 | | #pragma cdir shortloop |
10772 | | for (i=0; i<ni; i++) { |
10773 | | /* the normal case: */ |
10774 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
10775 | | /* test for range errors (not always needed but do it anyway) */ |
10776 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
10777 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
10778 | | nrange += tp[i] > X_USHORT_MAX || tp[i] < 0; |
10779 | | } |
10780 | | /* copy workspace back if necessary */ |
10781 | | if (realign) { |
10782 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
10783 | | xp = (ushort *) *xpp; |
10784 | | } |
10785 | | /* update xpp and tp */ |
10786 | | xp += ni; |
10787 | | tp += ni; |
10788 | | *xpp = (void*)xp; |
10789 | | } |
10790 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10791 | | |
10792 | | #else /* not SX */ |
10793 | |
|
10794 | 0 | char *xp = (char *) *xpp; |
10795 | 0 | int status = NC_NOERR; |
10796 | |
|
10797 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10798 | 0 | { |
10799 | 0 | int lstatus = ncx_put_ushort_short(xp, tp, fillp); |
10800 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10801 | 0 | status = lstatus; |
10802 | 0 | } |
10803 | |
|
10804 | 0 | *xpp = (void *)xp; |
10805 | 0 | return status; |
10806 | 0 | #endif |
10807 | 0 | } |
10808 | | |
10809 | | int |
10810 | | ncx_putn_ushort_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
10811 | 0 | { |
10812 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10813 | | |
10814 | | /* basic algorithm is: |
10815 | | * - ensure sane alignment of output data |
10816 | | * - copy (conversion happens automatically) input data |
10817 | | * to output |
10818 | | * - update tp to point at next unconverted input, and xpp to point |
10819 | | * at next location for converted output |
10820 | | */ |
10821 | | long i, j, ni; |
10822 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10823 | | ushort *xp; |
10824 | | int nrange = 0; /* number of range errors */ |
10825 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10826 | | long cxp = (long) *((char**)xpp); |
10827 | | |
10828 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10829 | | /* sjl: manually stripmine so we can limit amount of |
10830 | | * vector work space reserved to LOOPCNT elements. Also |
10831 | | * makes vectorisation easy */ |
10832 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10833 | | ni=Min(nelems-j,LOOPCNT); |
10834 | | if (realign) { |
10835 | | xp = tmp; |
10836 | | } else { |
10837 | | xp = (ushort *) *xpp; |
10838 | | } |
10839 | | /* copy the next block */ |
10840 | | #pragma cdir loopcnt=LOOPCNT |
10841 | | #pragma cdir shortloop |
10842 | | for (i=0; i<ni; i++) { |
10843 | | /* the normal case: */ |
10844 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
10845 | | /* test for range errors (not always needed but do it anyway) */ |
10846 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
10847 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
10848 | | nrange += tp[i] > X_USHORT_MAX || tp[i] < 0; |
10849 | | } |
10850 | | /* copy workspace back if necessary */ |
10851 | | if (realign) { |
10852 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
10853 | | xp = (ushort *) *xpp; |
10854 | | } |
10855 | | /* update xpp and tp */ |
10856 | | xp += ni; |
10857 | | tp += ni; |
10858 | | *xpp = (void*)xp; |
10859 | | } |
10860 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10861 | | |
10862 | | #else /* not SX */ |
10863 | |
|
10864 | 0 | char *xp = (char *) *xpp; |
10865 | 0 | int status = NC_NOERR; |
10866 | |
|
10867 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10868 | 0 | { |
10869 | 0 | int lstatus = ncx_put_ushort_int(xp, tp, fillp); |
10870 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10871 | 0 | status = lstatus; |
10872 | 0 | } |
10873 | |
|
10874 | 0 | *xpp = (void *)xp; |
10875 | 0 | return status; |
10876 | 0 | #endif |
10877 | 0 | } |
10878 | | |
10879 | | int |
10880 | | ncx_putn_ushort_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
10881 | 0 | { |
10882 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10883 | | |
10884 | | /* basic algorithm is: |
10885 | | * - ensure sane alignment of output data |
10886 | | * - copy (conversion happens automatically) input data |
10887 | | * to output |
10888 | | * - update tp to point at next unconverted input, and xpp to point |
10889 | | * at next location for converted output |
10890 | | */ |
10891 | | long i, j, ni; |
10892 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10893 | | ushort *xp; |
10894 | | int nrange = 0; /* number of range errors */ |
10895 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10896 | | long cxp = (long) *((char**)xpp); |
10897 | | |
10898 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10899 | | /* sjl: manually stripmine so we can limit amount of |
10900 | | * vector work space reserved to LOOPCNT elements. Also |
10901 | | * makes vectorisation easy */ |
10902 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10903 | | ni=Min(nelems-j,LOOPCNT); |
10904 | | if (realign) { |
10905 | | xp = tmp; |
10906 | | } else { |
10907 | | xp = (ushort *) *xpp; |
10908 | | } |
10909 | | /* copy the next block */ |
10910 | | #pragma cdir loopcnt=LOOPCNT |
10911 | | #pragma cdir shortloop |
10912 | | for (i=0; i<ni; i++) { |
10913 | | /* the normal case: */ |
10914 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
10915 | | /* test for range errors (not always needed but do it anyway) */ |
10916 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
10917 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
10918 | | nrange += tp[i] > X_USHORT_MAX || tp[i] < 0; |
10919 | | } |
10920 | | /* copy workspace back if necessary */ |
10921 | | if (realign) { |
10922 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
10923 | | xp = (ushort *) *xpp; |
10924 | | } |
10925 | | /* update xpp and tp */ |
10926 | | xp += ni; |
10927 | | tp += ni; |
10928 | | *xpp = (void*)xp; |
10929 | | } |
10930 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
10931 | | |
10932 | | #else /* not SX */ |
10933 | |
|
10934 | 0 | char *xp = (char *) *xpp; |
10935 | 0 | int status = NC_NOERR; |
10936 | |
|
10937 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
10938 | 0 | { |
10939 | 0 | int lstatus = ncx_put_ushort_long(xp, tp, fillp); |
10940 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
10941 | 0 | status = lstatus; |
10942 | 0 | } |
10943 | |
|
10944 | 0 | *xpp = (void *)xp; |
10945 | 0 | return status; |
10946 | 0 | #endif |
10947 | 0 | } |
10948 | | |
10949 | | int |
10950 | | ncx_putn_ushort_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
10951 | 0 | { |
10952 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
10953 | | |
10954 | | /* basic algorithm is: |
10955 | | * - ensure sane alignment of output data |
10956 | | * - copy (conversion happens automatically) input data |
10957 | | * to output |
10958 | | * - update tp to point at next unconverted input, and xpp to point |
10959 | | * at next location for converted output |
10960 | | */ |
10961 | | long i, j, ni; |
10962 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
10963 | | ushort *xp; |
10964 | | int nrange = 0; /* number of range errors */ |
10965 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
10966 | | long cxp = (long) *((char**)xpp); |
10967 | | |
10968 | | realign = (cxp & 7) % SIZEOF_USHORT; |
10969 | | /* sjl: manually stripmine so we can limit amount of |
10970 | | * vector work space reserved to LOOPCNT elements. Also |
10971 | | * makes vectorisation easy */ |
10972 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
10973 | | ni=Min(nelems-j,LOOPCNT); |
10974 | | if (realign) { |
10975 | | xp = tmp; |
10976 | | } else { |
10977 | | xp = (ushort *) *xpp; |
10978 | | } |
10979 | | /* copy the next block */ |
10980 | | #pragma cdir loopcnt=LOOPCNT |
10981 | | #pragma cdir shortloop |
10982 | | for (i=0; i<ni; i++) { |
10983 | | /* the normal case: */ |
10984 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
10985 | | /* test for range errors (not always needed but do it anyway) */ |
10986 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
10987 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
10988 | | nrange += tp[i] > X_USHORT_MAX || tp[i] < 0; |
10989 | | } |
10990 | | /* copy workspace back if necessary */ |
10991 | | if (realign) { |
10992 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
10993 | | xp = (ushort *) *xpp; |
10994 | | } |
10995 | | /* update xpp and tp */ |
10996 | | xp += ni; |
10997 | | tp += ni; |
10998 | | *xpp = (void*)xp; |
10999 | | } |
11000 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11001 | | |
11002 | | #else /* not SX */ |
11003 | |
|
11004 | 0 | char *xp = (char *) *xpp; |
11005 | 0 | int status = NC_NOERR; |
11006 | |
|
11007 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11008 | 0 | { |
11009 | 0 | int lstatus = ncx_put_ushort_float(xp, tp, fillp); |
11010 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11011 | 0 | status = lstatus; |
11012 | 0 | } |
11013 | |
|
11014 | 0 | *xpp = (void *)xp; |
11015 | 0 | return status; |
11016 | 0 | #endif |
11017 | 0 | } |
11018 | | |
11019 | | int |
11020 | | ncx_putn_ushort_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
11021 | 0 | { |
11022 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
11023 | | |
11024 | | /* basic algorithm is: |
11025 | | * - ensure sane alignment of output data |
11026 | | * - copy (conversion happens automatically) input data |
11027 | | * to output |
11028 | | * - update tp to point at next unconverted input, and xpp to point |
11029 | | * at next location for converted output |
11030 | | */ |
11031 | | long i, j, ni; |
11032 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
11033 | | ushort *xp; |
11034 | | int nrange = 0; /* number of range errors */ |
11035 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11036 | | long cxp = (long) *((char**)xpp); |
11037 | | |
11038 | | realign = (cxp & 7) % SIZEOF_USHORT; |
11039 | | /* sjl: manually stripmine so we can limit amount of |
11040 | | * vector work space reserved to LOOPCNT elements. Also |
11041 | | * makes vectorisation easy */ |
11042 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11043 | | ni=Min(nelems-j,LOOPCNT); |
11044 | | if (realign) { |
11045 | | xp = tmp; |
11046 | | } else { |
11047 | | xp = (ushort *) *xpp; |
11048 | | } |
11049 | | /* copy the next block */ |
11050 | | #pragma cdir loopcnt=LOOPCNT |
11051 | | #pragma cdir shortloop |
11052 | | for (i=0; i<ni; i++) { |
11053 | | /* the normal case: */ |
11054 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
11055 | | /* test for range errors (not always needed but do it anyway) */ |
11056 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
11057 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
11058 | | nrange += tp[i] > X_USHORT_MAX || tp[i] < 0; |
11059 | | } |
11060 | | /* copy workspace back if necessary */ |
11061 | | if (realign) { |
11062 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
11063 | | xp = (ushort *) *xpp; |
11064 | | } |
11065 | | /* update xpp and tp */ |
11066 | | xp += ni; |
11067 | | tp += ni; |
11068 | | *xpp = (void*)xp; |
11069 | | } |
11070 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11071 | | |
11072 | | #else /* not SX */ |
11073 | |
|
11074 | 0 | char *xp = (char *) *xpp; |
11075 | 0 | int status = NC_NOERR; |
11076 | |
|
11077 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11078 | 0 | { |
11079 | 0 | int lstatus = ncx_put_ushort_double(xp, tp, fillp); |
11080 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11081 | 0 | status = lstatus; |
11082 | 0 | } |
11083 | |
|
11084 | 0 | *xpp = (void *)xp; |
11085 | 0 | return status; |
11086 | 0 | #endif |
11087 | 0 | } |
11088 | | |
11089 | | int |
11090 | | ncx_putn_ushort_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
11091 | 0 | { |
11092 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
11093 | | |
11094 | | /* basic algorithm is: |
11095 | | * - ensure sane alignment of output data |
11096 | | * - copy (conversion happens automatically) input data |
11097 | | * to output |
11098 | | * - update tp to point at next unconverted input, and xpp to point |
11099 | | * at next location for converted output |
11100 | | */ |
11101 | | long i, j, ni; |
11102 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
11103 | | ushort *xp; |
11104 | | int nrange = 0; /* number of range errors */ |
11105 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11106 | | long cxp = (long) *((char**)xpp); |
11107 | | |
11108 | | realign = (cxp & 7) % SIZEOF_USHORT; |
11109 | | /* sjl: manually stripmine so we can limit amount of |
11110 | | * vector work space reserved to LOOPCNT elements. Also |
11111 | | * makes vectorisation easy */ |
11112 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11113 | | ni=Min(nelems-j,LOOPCNT); |
11114 | | if (realign) { |
11115 | | xp = tmp; |
11116 | | } else { |
11117 | | xp = (ushort *) *xpp; |
11118 | | } |
11119 | | /* copy the next block */ |
11120 | | #pragma cdir loopcnt=LOOPCNT |
11121 | | #pragma cdir shortloop |
11122 | | for (i=0; i<ni; i++) { |
11123 | | /* the normal case: */ |
11124 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
11125 | | /* test for range errors (not always needed but do it anyway) */ |
11126 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
11127 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
11128 | | nrange += tp[i] > X_USHORT_MAX || tp[i] < 0; |
11129 | | } |
11130 | | /* copy workspace back if necessary */ |
11131 | | if (realign) { |
11132 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
11133 | | xp = (ushort *) *xpp; |
11134 | | } |
11135 | | /* update xpp and tp */ |
11136 | | xp += ni; |
11137 | | tp += ni; |
11138 | | *xpp = (void*)xp; |
11139 | | } |
11140 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11141 | | |
11142 | | #else /* not SX */ |
11143 | |
|
11144 | 0 | char *xp = (char *) *xpp; |
11145 | 0 | int status = NC_NOERR; |
11146 | |
|
11147 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11148 | 0 | { |
11149 | 0 | int lstatus = ncx_put_ushort_longlong(xp, tp, fillp); |
11150 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11151 | 0 | status = lstatus; |
11152 | 0 | } |
11153 | |
|
11154 | 0 | *xpp = (void *)xp; |
11155 | 0 | return status; |
11156 | 0 | #endif |
11157 | 0 | } |
11158 | | |
11159 | | int |
11160 | | ncx_putn_ushort_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
11161 | 0 | { |
11162 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
11163 | | |
11164 | | /* basic algorithm is: |
11165 | | * - ensure sane alignment of output data |
11166 | | * - copy (conversion happens automatically) input data |
11167 | | * to output |
11168 | | * - update tp to point at next unconverted input, and xpp to point |
11169 | | * at next location for converted output |
11170 | | */ |
11171 | | long i, j, ni; |
11172 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
11173 | | ushort *xp; |
11174 | | int nrange = 0; /* number of range errors */ |
11175 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11176 | | long cxp = (long) *((char**)xpp); |
11177 | | |
11178 | | realign = (cxp & 7) % SIZEOF_USHORT; |
11179 | | /* sjl: manually stripmine so we can limit amount of |
11180 | | * vector work space reserved to LOOPCNT elements. Also |
11181 | | * makes vectorisation easy */ |
11182 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11183 | | ni=Min(nelems-j,LOOPCNT); |
11184 | | if (realign) { |
11185 | | xp = tmp; |
11186 | | } else { |
11187 | | xp = (ushort *) *xpp; |
11188 | | } |
11189 | | /* copy the next block */ |
11190 | | #pragma cdir loopcnt=LOOPCNT |
11191 | | #pragma cdir shortloop |
11192 | | for (i=0; i<ni; i++) { |
11193 | | /* the normal case: */ |
11194 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
11195 | | /* test for range errors (not always needed but do it anyway) */ |
11196 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
11197 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
11198 | | nrange += tp[i] > X_USHORT_MAX ; |
11199 | | } |
11200 | | /* copy workspace back if necessary */ |
11201 | | if (realign) { |
11202 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
11203 | | xp = (ushort *) *xpp; |
11204 | | } |
11205 | | /* update xpp and tp */ |
11206 | | xp += ni; |
11207 | | tp += ni; |
11208 | | *xpp = (void*)xp; |
11209 | | } |
11210 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11211 | | |
11212 | | #else /* not SX */ |
11213 | |
|
11214 | 0 | char *xp = (char *) *xpp; |
11215 | 0 | int status = NC_NOERR; |
11216 | |
|
11217 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11218 | 0 | { |
11219 | 0 | int lstatus = ncx_put_ushort_uchar(xp, tp, fillp); |
11220 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11221 | 0 | status = lstatus; |
11222 | 0 | } |
11223 | |
|
11224 | 0 | *xpp = (void *)xp; |
11225 | 0 | return status; |
11226 | 0 | #endif |
11227 | 0 | } |
11228 | | |
11229 | | int |
11230 | | ncx_putn_ushort_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
11231 | 0 | { |
11232 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
11233 | | |
11234 | | /* basic algorithm is: |
11235 | | * - ensure sane alignment of output data |
11236 | | * - copy (conversion happens automatically) input data |
11237 | | * to output |
11238 | | * - update tp to point at next unconverted input, and xpp to point |
11239 | | * at next location for converted output |
11240 | | */ |
11241 | | long i, j, ni; |
11242 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
11243 | | ushort *xp; |
11244 | | int nrange = 0; /* number of range errors */ |
11245 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11246 | | long cxp = (long) *((char**)xpp); |
11247 | | |
11248 | | realign = (cxp & 7) % SIZEOF_USHORT; |
11249 | | /* sjl: manually stripmine so we can limit amount of |
11250 | | * vector work space reserved to LOOPCNT elements. Also |
11251 | | * makes vectorisation easy */ |
11252 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11253 | | ni=Min(nelems-j,LOOPCNT); |
11254 | | if (realign) { |
11255 | | xp = tmp; |
11256 | | } else { |
11257 | | xp = (ushort *) *xpp; |
11258 | | } |
11259 | | /* copy the next block */ |
11260 | | #pragma cdir loopcnt=LOOPCNT |
11261 | | #pragma cdir shortloop |
11262 | | for (i=0; i<ni; i++) { |
11263 | | /* the normal case: */ |
11264 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
11265 | | /* test for range errors (not always needed but do it anyway) */ |
11266 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
11267 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
11268 | | nrange += tp[i] > X_USHORT_MAX ; |
11269 | | } |
11270 | | /* copy workspace back if necessary */ |
11271 | | if (realign) { |
11272 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
11273 | | xp = (ushort *) *xpp; |
11274 | | } |
11275 | | /* update xpp and tp */ |
11276 | | xp += ni; |
11277 | | tp += ni; |
11278 | | *xpp = (void*)xp; |
11279 | | } |
11280 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11281 | | |
11282 | | #else /* not SX */ |
11283 | |
|
11284 | 0 | char *xp = (char *) *xpp; |
11285 | 0 | int status = NC_NOERR; |
11286 | |
|
11287 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11288 | 0 | { |
11289 | 0 | int lstatus = ncx_put_ushort_uint(xp, tp, fillp); |
11290 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11291 | 0 | status = lstatus; |
11292 | 0 | } |
11293 | |
|
11294 | 0 | *xpp = (void *)xp; |
11295 | 0 | return status; |
11296 | 0 | #endif |
11297 | 0 | } |
11298 | | |
11299 | | int |
11300 | | ncx_putn_ushort_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
11301 | 0 | { |
11302 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_USHORT == SIZEOF_USHORT |
11303 | | |
11304 | | /* basic algorithm is: |
11305 | | * - ensure sane alignment of output data |
11306 | | * - copy (conversion happens automatically) input data |
11307 | | * to output |
11308 | | * - update tp to point at next unconverted input, and xpp to point |
11309 | | * at next location for converted output |
11310 | | */ |
11311 | | long i, j, ni; |
11312 | | ushort tmp[LOOPCNT]; /* in case input is misaligned */ |
11313 | | ushort *xp; |
11314 | | int nrange = 0; /* number of range errors */ |
11315 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11316 | | long cxp = (long) *((char**)xpp); |
11317 | | |
11318 | | realign = (cxp & 7) % SIZEOF_USHORT; |
11319 | | /* sjl: manually stripmine so we can limit amount of |
11320 | | * vector work space reserved to LOOPCNT elements. Also |
11321 | | * makes vectorisation easy */ |
11322 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11323 | | ni=Min(nelems-j,LOOPCNT); |
11324 | | if (realign) { |
11325 | | xp = tmp; |
11326 | | } else { |
11327 | | xp = (ushort *) *xpp; |
11328 | | } |
11329 | | /* copy the next block */ |
11330 | | #pragma cdir loopcnt=LOOPCNT |
11331 | | #pragma cdir shortloop |
11332 | | for (i=0; i<ni; i++) { |
11333 | | /* the normal case: */ |
11334 | | xp[i] = (ushort) Max( X_USHORT_MIN, Min(X_USHORT_MAX, (ushort) tp[i])); |
11335 | | /* test for range errors (not always needed but do it anyway) */ |
11336 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
11337 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
11338 | | nrange += tp[i] > X_USHORT_MAX ; |
11339 | | } |
11340 | | /* copy workspace back if necessary */ |
11341 | | if (realign) { |
11342 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_USHORT); |
11343 | | xp = (ushort *) *xpp; |
11344 | | } |
11345 | | /* update xpp and tp */ |
11346 | | xp += ni; |
11347 | | tp += ni; |
11348 | | *xpp = (void*)xp; |
11349 | | } |
11350 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11351 | | |
11352 | | #else /* not SX */ |
11353 | |
|
11354 | 0 | char *xp = (char *) *xpp; |
11355 | 0 | int status = NC_NOERR; |
11356 | |
|
11357 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11358 | 0 | { |
11359 | 0 | int lstatus = ncx_put_ushort_ulonglong(xp, tp, fillp); |
11360 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11361 | 0 | status = lstatus; |
11362 | 0 | } |
11363 | |
|
11364 | 0 | *xpp = (void *)xp; |
11365 | 0 | return status; |
11366 | 0 | #endif |
11367 | 0 | } |
11368 | | |
11369 | | |
11370 | | int |
11371 | | ncx_pad_putn_ushort_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
11372 | 0 | { |
11373 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11374 | |
|
11375 | 0 | char *xp = (char *) *xpp; |
11376 | 0 | int status = NC_NOERR; |
11377 | |
|
11378 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11379 | 0 | { |
11380 | 0 | int lstatus = ncx_put_ushort_schar(xp, tp, fillp); |
11381 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11382 | 0 | status = lstatus; |
11383 | 0 | } |
11384 | |
|
11385 | 0 | if (rndup != 0) |
11386 | 0 | { |
11387 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11388 | 0 | xp += X_SIZEOF_USHORT; |
11389 | 0 | } |
11390 | |
|
11391 | 0 | *xpp = (void *)xp; |
11392 | 0 | return status; |
11393 | 0 | } |
11394 | | |
11395 | | int |
11396 | | ncx_pad_putn_ushort_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
11397 | 0 | { |
11398 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11399 | |
|
11400 | 0 | char *xp = (char *) *xpp; |
11401 | 0 | int status = NC_NOERR; |
11402 | |
|
11403 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11404 | 0 | { |
11405 | 0 | int lstatus = ncx_put_ushort_uchar(xp, tp, fillp); |
11406 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11407 | 0 | status = lstatus; |
11408 | 0 | } |
11409 | |
|
11410 | 0 | if (rndup != 0) |
11411 | 0 | { |
11412 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11413 | 0 | xp += X_SIZEOF_USHORT; |
11414 | 0 | } |
11415 | |
|
11416 | 0 | *xpp = (void *)xp; |
11417 | 0 | return status; |
11418 | 0 | } |
11419 | | |
11420 | | int |
11421 | | ncx_pad_putn_ushort_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
11422 | 0 | { |
11423 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11424 | |
|
11425 | 0 | char *xp = (char *) *xpp; |
11426 | 0 | int status = NC_NOERR; |
11427 | |
|
11428 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11429 | 0 | { |
11430 | 0 | int lstatus = ncx_put_ushort_short(xp, tp, fillp); |
11431 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11432 | 0 | status = lstatus; |
11433 | 0 | } |
11434 | |
|
11435 | 0 | if (rndup != 0) |
11436 | 0 | { |
11437 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11438 | 0 | xp += X_SIZEOF_USHORT; |
11439 | 0 | } |
11440 | |
|
11441 | 0 | *xpp = (void *)xp; |
11442 | 0 | return status; |
11443 | 0 | } |
11444 | | |
11445 | | int |
11446 | | ncx_pad_putn_ushort_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
11447 | 0 | { |
11448 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11449 | |
|
11450 | 0 | char *xp = (char *) *xpp; |
11451 | 0 | int status = NC_NOERR; |
11452 | |
|
11453 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11454 | 0 | { |
11455 | 0 | int lstatus = ncx_put_ushort_int(xp, tp, fillp); |
11456 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11457 | 0 | status = lstatus; |
11458 | 0 | } |
11459 | |
|
11460 | 0 | if (rndup != 0) |
11461 | 0 | { |
11462 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11463 | 0 | xp += X_SIZEOF_USHORT; |
11464 | 0 | } |
11465 | |
|
11466 | 0 | *xpp = (void *)xp; |
11467 | 0 | return status; |
11468 | 0 | } |
11469 | | |
11470 | | int |
11471 | | ncx_pad_putn_ushort_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
11472 | 0 | { |
11473 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11474 | |
|
11475 | 0 | char *xp = (char *) *xpp; |
11476 | 0 | int status = NC_NOERR; |
11477 | |
|
11478 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11479 | 0 | { |
11480 | 0 | int lstatus = ncx_put_ushort_long(xp, tp, fillp); |
11481 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11482 | 0 | status = lstatus; |
11483 | 0 | } |
11484 | |
|
11485 | 0 | if (rndup != 0) |
11486 | 0 | { |
11487 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11488 | 0 | xp += X_SIZEOF_USHORT; |
11489 | 0 | } |
11490 | |
|
11491 | 0 | *xpp = (void *)xp; |
11492 | 0 | return status; |
11493 | 0 | } |
11494 | | |
11495 | | int |
11496 | | ncx_pad_putn_ushort_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
11497 | 0 | { |
11498 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11499 | |
|
11500 | 0 | char *xp = (char *) *xpp; |
11501 | 0 | int status = NC_NOERR; |
11502 | |
|
11503 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11504 | 0 | { |
11505 | 0 | int lstatus = ncx_put_ushort_float(xp, tp, fillp); |
11506 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11507 | 0 | status = lstatus; |
11508 | 0 | } |
11509 | |
|
11510 | 0 | if (rndup != 0) |
11511 | 0 | { |
11512 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11513 | 0 | xp += X_SIZEOF_USHORT; |
11514 | 0 | } |
11515 | |
|
11516 | 0 | *xpp = (void *)xp; |
11517 | 0 | return status; |
11518 | 0 | } |
11519 | | |
11520 | | int |
11521 | | ncx_pad_putn_ushort_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
11522 | 0 | { |
11523 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11524 | |
|
11525 | 0 | char *xp = (char *) *xpp; |
11526 | 0 | int status = NC_NOERR; |
11527 | |
|
11528 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11529 | 0 | { |
11530 | 0 | int lstatus = ncx_put_ushort_double(xp, tp, fillp); |
11531 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11532 | 0 | status = lstatus; |
11533 | 0 | } |
11534 | |
|
11535 | 0 | if (rndup != 0) |
11536 | 0 | { |
11537 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11538 | 0 | xp += X_SIZEOF_USHORT; |
11539 | 0 | } |
11540 | |
|
11541 | 0 | *xpp = (void *)xp; |
11542 | 0 | return status; |
11543 | 0 | } |
11544 | | |
11545 | | int |
11546 | | ncx_pad_putn_ushort_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
11547 | 0 | { |
11548 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11549 | |
|
11550 | 0 | char *xp = (char *) *xpp; |
11551 | 0 | int status = NC_NOERR; |
11552 | |
|
11553 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11554 | 0 | { |
11555 | 0 | int lstatus = ncx_put_ushort_uint(xp, tp, fillp); |
11556 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11557 | 0 | status = lstatus; |
11558 | 0 | } |
11559 | |
|
11560 | 0 | if (rndup != 0) |
11561 | 0 | { |
11562 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11563 | 0 | xp += X_SIZEOF_USHORT; |
11564 | 0 | } |
11565 | |
|
11566 | 0 | *xpp = (void *)xp; |
11567 | 0 | return status; |
11568 | 0 | } |
11569 | | |
11570 | | int |
11571 | | ncx_pad_putn_ushort_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
11572 | 0 | { |
11573 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11574 | |
|
11575 | 0 | char *xp = (char *) *xpp; |
11576 | 0 | int status = NC_NOERR; |
11577 | |
|
11578 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11579 | 0 | { |
11580 | 0 | int lstatus = ncx_put_ushort_longlong(xp, tp, fillp); |
11581 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11582 | 0 | status = lstatus; |
11583 | 0 | } |
11584 | |
|
11585 | 0 | if (rndup != 0) |
11586 | 0 | { |
11587 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11588 | 0 | xp += X_SIZEOF_USHORT; |
11589 | 0 | } |
11590 | |
|
11591 | 0 | *xpp = (void *)xp; |
11592 | 0 | return status; |
11593 | 0 | } |
11594 | | |
11595 | | int |
11596 | | ncx_pad_putn_ushort_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
11597 | 0 | { |
11598 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11599 | |
|
11600 | 0 | char *xp = (char *) *xpp; |
11601 | 0 | int status = NC_NOERR; |
11602 | |
|
11603 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11604 | 0 | { |
11605 | 0 | int lstatus = ncx_put_ushort_ulonglong(xp, tp, fillp); |
11606 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11607 | 0 | status = lstatus; |
11608 | 0 | } |
11609 | |
|
11610 | 0 | if (rndup != 0) |
11611 | 0 | { |
11612 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11613 | 0 | xp += X_SIZEOF_USHORT; |
11614 | 0 | } |
11615 | |
|
11616 | 0 | *xpp = (void *)xp; |
11617 | 0 | return status; |
11618 | 0 | } |
11619 | | |
11620 | | int |
11621 | | ncx_pad_putn_ushort_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
11622 | 0 | { |
11623 | 0 | const size_t rndup = nelems % X_SIZEOF_SHORT; |
11624 | |
|
11625 | 0 | char *xp = (char *) *xpp; |
11626 | 0 | int status = NC_NOERR; |
11627 | |
|
11628 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_USHORT, tp++) |
11629 | 0 | { |
11630 | 0 | int lstatus = ncx_put_ushort_ushort(xp, tp, fillp); |
11631 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11632 | 0 | status = lstatus; |
11633 | 0 | } |
11634 | |
|
11635 | 0 | if (rndup != 0) |
11636 | 0 | { |
11637 | 0 | (void) memcpy(xp, nada, (size_t)(X_SIZEOF_USHORT)); |
11638 | 0 | xp += X_SIZEOF_USHORT; |
11639 | 0 | } |
11640 | |
|
11641 | 0 | *xpp = (void *)xp; |
11642 | 0 | return status; |
11643 | 0 | } |
11644 | | |
11645 | | |
11646 | | |
11647 | | /* int -----------------------------------------------------------------------*/ |
11648 | | |
11649 | | #if X_SIZEOF_INT == SIZEOF_INT |
11650 | | /* optimized version */ |
11651 | | int |
11652 | | ncx_getn_int_int(const void **xpp, size_t nelems, int *tp) |
11653 | 131k | { |
11654 | | #ifdef WORDS_BIGENDIAN |
11655 | | (void) memcpy(tp, *xpp, (size_t)nelems * SIZEOF_INT); |
11656 | | # else |
11657 | 131k | swapn4b(tp, *xpp, nelems); |
11658 | 131k | # endif |
11659 | 131k | *xpp = (const void *)((const char *)(*xpp) + nelems * X_SIZEOF_INT); |
11660 | 131k | return NC_NOERR; |
11661 | 131k | } |
11662 | | #else |
11663 | | int |
11664 | | ncx_getn_int_int(const void **xpp, size_t nelems, int *tp) |
11665 | | { |
11666 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
11667 | | |
11668 | | /* basic algorithm is: |
11669 | | * - ensure sane alignment of input data |
11670 | | * - copy (conversion happens automatically) input data |
11671 | | * to output |
11672 | | * - update xpp to point at next unconverted input, and tp to point |
11673 | | * at next location for converted output |
11674 | | */ |
11675 | | long i, j, ni; |
11676 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
11677 | | int *xp; |
11678 | | int nrange = 0; /* number of range errors */ |
11679 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11680 | | long cxp = (long) *((char**)xpp); |
11681 | | |
11682 | | realign = (cxp & 7) % SIZEOF_INT; |
11683 | | /* sjl: manually stripmine so we can limit amount of |
11684 | | * vector work space reserved to LOOPCNT elements. Also |
11685 | | * makes vectorisation easy */ |
11686 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11687 | | ni=Min(nelems-j,LOOPCNT); |
11688 | | if (realign) { |
11689 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
11690 | | xp = tmp; |
11691 | | } else { |
11692 | | xp = (int *) *xpp; |
11693 | | } |
11694 | | /* copy the next block */ |
11695 | | #pragma cdir loopcnt=LOOPCNT |
11696 | | #pragma cdir shortloop |
11697 | | for (i=0; i<ni; i++) { |
11698 | | tp[i] = (int) Max( INT_MIN, Min(INT_MAX, (int) xp[i])); |
11699 | | /* test for range errors (not always needed but do it anyway) */ |
11700 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
11701 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
11702 | | nrange += xp[i] > INT_MAX || xp[i] < INT_MIN; |
11703 | | } |
11704 | | /* update xpp and tp */ |
11705 | | if (realign) xp = (int *) *xpp; |
11706 | | xp += ni; |
11707 | | tp += ni; |
11708 | | *xpp = (void*)xp; |
11709 | | } |
11710 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11711 | | |
11712 | | #else /* not SX */ |
11713 | | const char *xp = (const char *) *xpp; |
11714 | | int status = NC_NOERR; |
11715 | | |
11716 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
11717 | | { |
11718 | | const int lstatus = ncx_get_int_int(xp, tp); |
11719 | | if (status == NC_NOERR) /* report the first encountered error */ |
11720 | | status = lstatus; |
11721 | | } |
11722 | | |
11723 | | *xpp = (const void *)xp; |
11724 | | return status; |
11725 | | #endif |
11726 | | } |
11727 | | |
11728 | | #endif |
11729 | | int |
11730 | | ncx_getn_int_schar(const void **xpp, size_t nelems, schar *tp) |
11731 | 0 | { |
11732 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
11733 | | |
11734 | | /* basic algorithm is: |
11735 | | * - ensure sane alignment of input data |
11736 | | * - copy (conversion happens automatically) input data |
11737 | | * to output |
11738 | | * - update xpp to point at next unconverted input, and tp to point |
11739 | | * at next location for converted output |
11740 | | */ |
11741 | | long i, j, ni; |
11742 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
11743 | | int *xp; |
11744 | | int nrange = 0; /* number of range errors */ |
11745 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11746 | | long cxp = (long) *((char**)xpp); |
11747 | | |
11748 | | realign = (cxp & 7) % SIZEOF_INT; |
11749 | | /* sjl: manually stripmine so we can limit amount of |
11750 | | * vector work space reserved to LOOPCNT elements. Also |
11751 | | * makes vectorisation easy */ |
11752 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11753 | | ni=Min(nelems-j,LOOPCNT); |
11754 | | if (realign) { |
11755 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
11756 | | xp = tmp; |
11757 | | } else { |
11758 | | xp = (int *) *xpp; |
11759 | | } |
11760 | | /* copy the next block */ |
11761 | | #pragma cdir loopcnt=LOOPCNT |
11762 | | #pragma cdir shortloop |
11763 | | for (i=0; i<ni; i++) { |
11764 | | tp[i] = (schar) Max( SCHAR_MIN, Min(SCHAR_MAX, (schar) xp[i])); |
11765 | | /* test for range errors (not always needed but do it anyway) */ |
11766 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
11767 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
11768 | | nrange += xp[i] > SCHAR_MAX || xp[i] < SCHAR_MIN; |
11769 | | } |
11770 | | /* update xpp and tp */ |
11771 | | if (realign) xp = (int *) *xpp; |
11772 | | xp += ni; |
11773 | | tp += ni; |
11774 | | *xpp = (void*)xp; |
11775 | | } |
11776 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11777 | | |
11778 | | #else /* not SX */ |
11779 | 0 | const char *xp = (const char *) *xpp; |
11780 | 0 | int status = NC_NOERR; |
11781 | |
|
11782 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
11783 | 0 | { |
11784 | 0 | const int lstatus = ncx_get_int_schar(xp, tp); |
11785 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11786 | 0 | status = lstatus; |
11787 | 0 | } |
11788 | |
|
11789 | 0 | *xpp = (const void *)xp; |
11790 | 0 | return status; |
11791 | 0 | #endif |
11792 | 0 | } |
11793 | | |
11794 | | int |
11795 | | ncx_getn_int_short(const void **xpp, size_t nelems, short *tp) |
11796 | 0 | { |
11797 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
11798 | | |
11799 | | /* basic algorithm is: |
11800 | | * - ensure sane alignment of input data |
11801 | | * - copy (conversion happens automatically) input data |
11802 | | * to output |
11803 | | * - update xpp to point at next unconverted input, and tp to point |
11804 | | * at next location for converted output |
11805 | | */ |
11806 | | long i, j, ni; |
11807 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
11808 | | int *xp; |
11809 | | int nrange = 0; /* number of range errors */ |
11810 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11811 | | long cxp = (long) *((char**)xpp); |
11812 | | |
11813 | | realign = (cxp & 7) % SIZEOF_INT; |
11814 | | /* sjl: manually stripmine so we can limit amount of |
11815 | | * vector work space reserved to LOOPCNT elements. Also |
11816 | | * makes vectorisation easy */ |
11817 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11818 | | ni=Min(nelems-j,LOOPCNT); |
11819 | | if (realign) { |
11820 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
11821 | | xp = tmp; |
11822 | | } else { |
11823 | | xp = (int *) *xpp; |
11824 | | } |
11825 | | /* copy the next block */ |
11826 | | #pragma cdir loopcnt=LOOPCNT |
11827 | | #pragma cdir shortloop |
11828 | | for (i=0; i<ni; i++) { |
11829 | | tp[i] = (short) Max( SHORT_MIN, Min(SHORT_MAX, (short) xp[i])); |
11830 | | /* test for range errors (not always needed but do it anyway) */ |
11831 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
11832 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
11833 | | nrange += xp[i] > SHORT_MAX || xp[i] < SHORT_MIN; |
11834 | | } |
11835 | | /* update xpp and tp */ |
11836 | | if (realign) xp = (int *) *xpp; |
11837 | | xp += ni; |
11838 | | tp += ni; |
11839 | | *xpp = (void*)xp; |
11840 | | } |
11841 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11842 | | |
11843 | | #else /* not SX */ |
11844 | 0 | const char *xp = (const char *) *xpp; |
11845 | 0 | int status = NC_NOERR; |
11846 | |
|
11847 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
11848 | 0 | { |
11849 | 0 | const int lstatus = ncx_get_int_short(xp, tp); |
11850 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11851 | 0 | status = lstatus; |
11852 | 0 | } |
11853 | |
|
11854 | 0 | *xpp = (const void *)xp; |
11855 | 0 | return status; |
11856 | 0 | #endif |
11857 | 0 | } |
11858 | | |
11859 | | int |
11860 | | ncx_getn_int_long(const void **xpp, size_t nelems, long *tp) |
11861 | 0 | { |
11862 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
11863 | | |
11864 | | /* basic algorithm is: |
11865 | | * - ensure sane alignment of input data |
11866 | | * - copy (conversion happens automatically) input data |
11867 | | * to output |
11868 | | * - update xpp to point at next unconverted input, and tp to point |
11869 | | * at next location for converted output |
11870 | | */ |
11871 | | long i, j, ni; |
11872 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
11873 | | int *xp; |
11874 | | int nrange = 0; /* number of range errors */ |
11875 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11876 | | long cxp = (long) *((char**)xpp); |
11877 | | |
11878 | | realign = (cxp & 7) % SIZEOF_INT; |
11879 | | /* sjl: manually stripmine so we can limit amount of |
11880 | | * vector work space reserved to LOOPCNT elements. Also |
11881 | | * makes vectorisation easy */ |
11882 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11883 | | ni=Min(nelems-j,LOOPCNT); |
11884 | | if (realign) { |
11885 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
11886 | | xp = tmp; |
11887 | | } else { |
11888 | | xp = (int *) *xpp; |
11889 | | } |
11890 | | /* copy the next block */ |
11891 | | #pragma cdir loopcnt=LOOPCNT |
11892 | | #pragma cdir shortloop |
11893 | | for (i=0; i<ni; i++) { |
11894 | | tp[i] = (long) Max( LONG_MIN, Min(LONG_MAX, (long) xp[i])); |
11895 | | /* test for range errors (not always needed but do it anyway) */ |
11896 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
11897 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
11898 | | nrange += xp[i] > LONG_MAX || xp[i] < LONG_MIN; |
11899 | | } |
11900 | | /* update xpp and tp */ |
11901 | | if (realign) xp = (int *) *xpp; |
11902 | | xp += ni; |
11903 | | tp += ni; |
11904 | | *xpp = (void*)xp; |
11905 | | } |
11906 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11907 | | |
11908 | | #else /* not SX */ |
11909 | 0 | const char *xp = (const char *) *xpp; |
11910 | 0 | int status = NC_NOERR; |
11911 | |
|
11912 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
11913 | 0 | { |
11914 | 0 | const int lstatus = ncx_get_int_long(xp, tp); |
11915 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11916 | 0 | status = lstatus; |
11917 | 0 | } |
11918 | |
|
11919 | 0 | *xpp = (const void *)xp; |
11920 | 0 | return status; |
11921 | 0 | #endif |
11922 | 0 | } |
11923 | | |
11924 | | int |
11925 | | ncx_getn_int_float(const void **xpp, size_t nelems, float *tp) |
11926 | 0 | { |
11927 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
11928 | | |
11929 | | /* basic algorithm is: |
11930 | | * - ensure sane alignment of input data |
11931 | | * - copy (conversion happens automatically) input data |
11932 | | * to output |
11933 | | * - update xpp to point at next unconverted input, and tp to point |
11934 | | * at next location for converted output |
11935 | | */ |
11936 | | long i, j, ni; |
11937 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
11938 | | int *xp; |
11939 | | int nrange = 0; /* number of range errors */ |
11940 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
11941 | | long cxp = (long) *((char**)xpp); |
11942 | | |
11943 | | realign = (cxp & 7) % SIZEOF_INT; |
11944 | | /* sjl: manually stripmine so we can limit amount of |
11945 | | * vector work space reserved to LOOPCNT elements. Also |
11946 | | * makes vectorisation easy */ |
11947 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
11948 | | ni=Min(nelems-j,LOOPCNT); |
11949 | | if (realign) { |
11950 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
11951 | | xp = tmp; |
11952 | | } else { |
11953 | | xp = (int *) *xpp; |
11954 | | } |
11955 | | /* copy the next block */ |
11956 | | #pragma cdir loopcnt=LOOPCNT |
11957 | | #pragma cdir shortloop |
11958 | | for (i=0; i<ni; i++) { |
11959 | | tp[i] = (float) Max( FLOAT_MIN, Min(FLOAT_MAX, (float) xp[i])); |
11960 | | /* test for range errors (not always needed but do it anyway) */ |
11961 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
11962 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
11963 | | nrange += xp[i] > FLOAT_MAX || xp[i] < FLOAT_MIN; |
11964 | | } |
11965 | | /* update xpp and tp */ |
11966 | | if (realign) xp = (int *) *xpp; |
11967 | | xp += ni; |
11968 | | tp += ni; |
11969 | | *xpp = (void*)xp; |
11970 | | } |
11971 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
11972 | | |
11973 | | #else /* not SX */ |
11974 | 0 | const char *xp = (const char *) *xpp; |
11975 | 0 | int status = NC_NOERR; |
11976 | |
|
11977 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
11978 | 0 | { |
11979 | 0 | const int lstatus = ncx_get_int_float(xp, tp); |
11980 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
11981 | 0 | status = lstatus; |
11982 | 0 | } |
11983 | |
|
11984 | 0 | *xpp = (const void *)xp; |
11985 | 0 | return status; |
11986 | 0 | #endif |
11987 | 0 | } |
11988 | | |
11989 | | int |
11990 | | ncx_getn_int_double(const void **xpp, size_t nelems, double *tp) |
11991 | 0 | { |
11992 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
11993 | | |
11994 | | /* basic algorithm is: |
11995 | | * - ensure sane alignment of input data |
11996 | | * - copy (conversion happens automatically) input data |
11997 | | * to output |
11998 | | * - update xpp to point at next unconverted input, and tp to point |
11999 | | * at next location for converted output |
12000 | | */ |
12001 | | long i, j, ni; |
12002 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12003 | | int *xp; |
12004 | | int nrange = 0; /* number of range errors */ |
12005 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12006 | | long cxp = (long) *((char**)xpp); |
12007 | | |
12008 | | realign = (cxp & 7) % SIZEOF_INT; |
12009 | | /* sjl: manually stripmine so we can limit amount of |
12010 | | * vector work space reserved to LOOPCNT elements. Also |
12011 | | * makes vectorisation easy */ |
12012 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12013 | | ni=Min(nelems-j,LOOPCNT); |
12014 | | if (realign) { |
12015 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
12016 | | xp = tmp; |
12017 | | } else { |
12018 | | xp = (int *) *xpp; |
12019 | | } |
12020 | | /* copy the next block */ |
12021 | | #pragma cdir loopcnt=LOOPCNT |
12022 | | #pragma cdir shortloop |
12023 | | for (i=0; i<ni; i++) { |
12024 | | tp[i] = (double) Max( DOUBLE_MIN, Min(DOUBLE_MAX, (double) xp[i])); |
12025 | | /* test for range errors (not always needed but do it anyway) */ |
12026 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
12027 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
12028 | | nrange += xp[i] > DOUBLE_MAX || xp[i] < DOUBLE_MIN; |
12029 | | } |
12030 | | /* update xpp and tp */ |
12031 | | if (realign) xp = (int *) *xpp; |
12032 | | xp += ni; |
12033 | | tp += ni; |
12034 | | *xpp = (void*)xp; |
12035 | | } |
12036 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12037 | | |
12038 | | #else /* not SX */ |
12039 | 0 | const char *xp = (const char *) *xpp; |
12040 | 0 | int status = NC_NOERR; |
12041 | |
|
12042 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12043 | 0 | { |
12044 | 0 | const int lstatus = ncx_get_int_double(xp, tp); |
12045 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12046 | 0 | status = lstatus; |
12047 | 0 | } |
12048 | |
|
12049 | 0 | *xpp = (const void *)xp; |
12050 | 0 | return status; |
12051 | 0 | #endif |
12052 | 0 | } |
12053 | | |
12054 | | int |
12055 | | ncx_getn_int_longlong(const void **xpp, size_t nelems, longlong *tp) |
12056 | 0 | { |
12057 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12058 | | |
12059 | | /* basic algorithm is: |
12060 | | * - ensure sane alignment of input data |
12061 | | * - copy (conversion happens automatically) input data |
12062 | | * to output |
12063 | | * - update xpp to point at next unconverted input, and tp to point |
12064 | | * at next location for converted output |
12065 | | */ |
12066 | | long i, j, ni; |
12067 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12068 | | int *xp; |
12069 | | int nrange = 0; /* number of range errors */ |
12070 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12071 | | long cxp = (long) *((char**)xpp); |
12072 | | |
12073 | | realign = (cxp & 7) % SIZEOF_INT; |
12074 | | /* sjl: manually stripmine so we can limit amount of |
12075 | | * vector work space reserved to LOOPCNT elements. Also |
12076 | | * makes vectorisation easy */ |
12077 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12078 | | ni=Min(nelems-j,LOOPCNT); |
12079 | | if (realign) { |
12080 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
12081 | | xp = tmp; |
12082 | | } else { |
12083 | | xp = (int *) *xpp; |
12084 | | } |
12085 | | /* copy the next block */ |
12086 | | #pragma cdir loopcnt=LOOPCNT |
12087 | | #pragma cdir shortloop |
12088 | | for (i=0; i<ni; i++) { |
12089 | | tp[i] = (longlong) Max( LONGLONG_MIN, Min(LONGLONG_MAX, (longlong) xp[i])); |
12090 | | /* test for range errors (not always needed but do it anyway) */ |
12091 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
12092 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
12093 | | nrange += xp[i] > LONGLONG_MAX || xp[i] < LONGLONG_MIN; |
12094 | | } |
12095 | | /* update xpp and tp */ |
12096 | | if (realign) xp = (int *) *xpp; |
12097 | | xp += ni; |
12098 | | tp += ni; |
12099 | | *xpp = (void*)xp; |
12100 | | } |
12101 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12102 | | |
12103 | | #else /* not SX */ |
12104 | 0 | const char *xp = (const char *) *xpp; |
12105 | 0 | int status = NC_NOERR; |
12106 | |
|
12107 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12108 | 0 | { |
12109 | 0 | const int lstatus = ncx_get_int_longlong(xp, tp); |
12110 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12111 | 0 | status = lstatus; |
12112 | 0 | } |
12113 | |
|
12114 | 0 | *xpp = (const void *)xp; |
12115 | 0 | return status; |
12116 | 0 | #endif |
12117 | 0 | } |
12118 | | |
12119 | | int |
12120 | | ncx_getn_int_uchar(const void **xpp, size_t nelems, uchar *tp) |
12121 | 0 | { |
12122 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12123 | | |
12124 | | /* basic algorithm is: |
12125 | | * - ensure sane alignment of input data |
12126 | | * - copy (conversion happens automatically) input data |
12127 | | * to output |
12128 | | * - update xpp to point at next unconverted input, and tp to point |
12129 | | * at next location for converted output |
12130 | | */ |
12131 | | long i, j, ni; |
12132 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12133 | | int *xp; |
12134 | | int nrange = 0; /* number of range errors */ |
12135 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12136 | | long cxp = (long) *((char**)xpp); |
12137 | | |
12138 | | realign = (cxp & 7) % SIZEOF_INT; |
12139 | | /* sjl: manually stripmine so we can limit amount of |
12140 | | * vector work space reserved to LOOPCNT elements. Also |
12141 | | * makes vectorisation easy */ |
12142 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12143 | | ni=Min(nelems-j,LOOPCNT); |
12144 | | if (realign) { |
12145 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
12146 | | xp = tmp; |
12147 | | } else { |
12148 | | xp = (int *) *xpp; |
12149 | | } |
12150 | | /* copy the next block */ |
12151 | | #pragma cdir loopcnt=LOOPCNT |
12152 | | #pragma cdir shortloop |
12153 | | for (i=0; i<ni; i++) { |
12154 | | tp[i] = (uchar) Max( UCHAR_MIN, Min(UCHAR_MAX, (uchar) xp[i])); |
12155 | | /* test for range errors (not always needed but do it anyway) */ |
12156 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
12157 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
12158 | | nrange += xp[i] > UCHAR_MAX || xp[i] < 0; |
12159 | | } |
12160 | | /* update xpp and tp */ |
12161 | | if (realign) xp = (int *) *xpp; |
12162 | | xp += ni; |
12163 | | tp += ni; |
12164 | | *xpp = (void*)xp; |
12165 | | } |
12166 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12167 | | |
12168 | | #else /* not SX */ |
12169 | 0 | const char *xp = (const char *) *xpp; |
12170 | 0 | int status = NC_NOERR; |
12171 | |
|
12172 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12173 | 0 | { |
12174 | 0 | const int lstatus = ncx_get_int_uchar(xp, tp); |
12175 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12176 | 0 | status = lstatus; |
12177 | 0 | } |
12178 | |
|
12179 | 0 | *xpp = (const void *)xp; |
12180 | 0 | return status; |
12181 | 0 | #endif |
12182 | 0 | } |
12183 | | |
12184 | | int |
12185 | | ncx_getn_int_ushort(const void **xpp, size_t nelems, ushort *tp) |
12186 | 0 | { |
12187 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12188 | | |
12189 | | /* basic algorithm is: |
12190 | | * - ensure sane alignment of input data |
12191 | | * - copy (conversion happens automatically) input data |
12192 | | * to output |
12193 | | * - update xpp to point at next unconverted input, and tp to point |
12194 | | * at next location for converted output |
12195 | | */ |
12196 | | long i, j, ni; |
12197 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12198 | | int *xp; |
12199 | | int nrange = 0; /* number of range errors */ |
12200 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12201 | | long cxp = (long) *((char**)xpp); |
12202 | | |
12203 | | realign = (cxp & 7) % SIZEOF_INT; |
12204 | | /* sjl: manually stripmine so we can limit amount of |
12205 | | * vector work space reserved to LOOPCNT elements. Also |
12206 | | * makes vectorisation easy */ |
12207 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12208 | | ni=Min(nelems-j,LOOPCNT); |
12209 | | if (realign) { |
12210 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
12211 | | xp = tmp; |
12212 | | } else { |
12213 | | xp = (int *) *xpp; |
12214 | | } |
12215 | | /* copy the next block */ |
12216 | | #pragma cdir loopcnt=LOOPCNT |
12217 | | #pragma cdir shortloop |
12218 | | for (i=0; i<ni; i++) { |
12219 | | tp[i] = (ushort) Max( USHORT_MIN, Min(USHORT_MAX, (ushort) xp[i])); |
12220 | | /* test for range errors (not always needed but do it anyway) */ |
12221 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
12222 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
12223 | | nrange += xp[i] > USHORT_MAX || xp[i] < 0; |
12224 | | } |
12225 | | /* update xpp and tp */ |
12226 | | if (realign) xp = (int *) *xpp; |
12227 | | xp += ni; |
12228 | | tp += ni; |
12229 | | *xpp = (void*)xp; |
12230 | | } |
12231 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12232 | | |
12233 | | #else /* not SX */ |
12234 | 0 | const char *xp = (const char *) *xpp; |
12235 | 0 | int status = NC_NOERR; |
12236 | |
|
12237 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12238 | 0 | { |
12239 | 0 | const int lstatus = ncx_get_int_ushort(xp, tp); |
12240 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12241 | 0 | status = lstatus; |
12242 | 0 | } |
12243 | |
|
12244 | 0 | *xpp = (const void *)xp; |
12245 | 0 | return status; |
12246 | 0 | #endif |
12247 | 0 | } |
12248 | | |
12249 | | int |
12250 | | ncx_getn_int_uint(const void **xpp, size_t nelems, uint *tp) |
12251 | 0 | { |
12252 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12253 | | |
12254 | | /* basic algorithm is: |
12255 | | * - ensure sane alignment of input data |
12256 | | * - copy (conversion happens automatically) input data |
12257 | | * to output |
12258 | | * - update xpp to point at next unconverted input, and tp to point |
12259 | | * at next location for converted output |
12260 | | */ |
12261 | | long i, j, ni; |
12262 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12263 | | int *xp; |
12264 | | int nrange = 0; /* number of range errors */ |
12265 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12266 | | long cxp = (long) *((char**)xpp); |
12267 | | |
12268 | | realign = (cxp & 7) % SIZEOF_INT; |
12269 | | /* sjl: manually stripmine so we can limit amount of |
12270 | | * vector work space reserved to LOOPCNT elements. Also |
12271 | | * makes vectorisation easy */ |
12272 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12273 | | ni=Min(nelems-j,LOOPCNT); |
12274 | | if (realign) { |
12275 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
12276 | | xp = tmp; |
12277 | | } else { |
12278 | | xp = (int *) *xpp; |
12279 | | } |
12280 | | /* copy the next block */ |
12281 | | #pragma cdir loopcnt=LOOPCNT |
12282 | | #pragma cdir shortloop |
12283 | | for (i=0; i<ni; i++) { |
12284 | | tp[i] = (uint) Max( UINT_MIN, Min(UINT_MAX, (uint) xp[i])); |
12285 | | /* test for range errors (not always needed but do it anyway) */ |
12286 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
12287 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
12288 | | nrange += xp[i] > UINT_MAX || xp[i] < 0; |
12289 | | } |
12290 | | /* update xpp and tp */ |
12291 | | if (realign) xp = (int *) *xpp; |
12292 | | xp += ni; |
12293 | | tp += ni; |
12294 | | *xpp = (void*)xp; |
12295 | | } |
12296 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12297 | | |
12298 | | #else /* not SX */ |
12299 | 0 | const char *xp = (const char *) *xpp; |
12300 | 0 | int status = NC_NOERR; |
12301 | |
|
12302 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12303 | 0 | { |
12304 | 0 | const int lstatus = ncx_get_int_uint(xp, tp); |
12305 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12306 | 0 | status = lstatus; |
12307 | 0 | } |
12308 | |
|
12309 | 0 | *xpp = (const void *)xp; |
12310 | 0 | return status; |
12311 | 0 | #endif |
12312 | 0 | } |
12313 | | |
12314 | | int |
12315 | | ncx_getn_int_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
12316 | 0 | { |
12317 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12318 | | |
12319 | | /* basic algorithm is: |
12320 | | * - ensure sane alignment of input data |
12321 | | * - copy (conversion happens automatically) input data |
12322 | | * to output |
12323 | | * - update xpp to point at next unconverted input, and tp to point |
12324 | | * at next location for converted output |
12325 | | */ |
12326 | | long i, j, ni; |
12327 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12328 | | int *xp; |
12329 | | int nrange = 0; /* number of range errors */ |
12330 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12331 | | long cxp = (long) *((char**)xpp); |
12332 | | |
12333 | | realign = (cxp & 7) % SIZEOF_INT; |
12334 | | /* sjl: manually stripmine so we can limit amount of |
12335 | | * vector work space reserved to LOOPCNT elements. Also |
12336 | | * makes vectorisation easy */ |
12337 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12338 | | ni=Min(nelems-j,LOOPCNT); |
12339 | | if (realign) { |
12340 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT)); |
12341 | | xp = tmp; |
12342 | | } else { |
12343 | | xp = (int *) *xpp; |
12344 | | } |
12345 | | /* copy the next block */ |
12346 | | #pragma cdir loopcnt=LOOPCNT |
12347 | | #pragma cdir shortloop |
12348 | | for (i=0; i<ni; i++) { |
12349 | | tp[i] = (ulonglong) Max( ULONGLONG_MIN, Min(ULONGLONG_MAX, (ulonglong) xp[i])); |
12350 | | /* test for range errors (not always needed but do it anyway) */ |
12351 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
12352 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
12353 | | nrange += xp[i] > ULONGLONG_MAX || xp[i] < 0; |
12354 | | } |
12355 | | /* update xpp and tp */ |
12356 | | if (realign) xp = (int *) *xpp; |
12357 | | xp += ni; |
12358 | | tp += ni; |
12359 | | *xpp = (void*)xp; |
12360 | | } |
12361 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12362 | | |
12363 | | #else /* not SX */ |
12364 | 0 | const char *xp = (const char *) *xpp; |
12365 | 0 | int status = NC_NOERR; |
12366 | |
|
12367 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12368 | 0 | { |
12369 | 0 | const int lstatus = ncx_get_int_ulonglong(xp, tp); |
12370 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12371 | 0 | status = lstatus; |
12372 | 0 | } |
12373 | |
|
12374 | 0 | *xpp = (const void *)xp; |
12375 | 0 | return status; |
12376 | 0 | #endif |
12377 | 0 | } |
12378 | | |
12379 | | |
12380 | | #if X_SIZEOF_INT == SIZEOF_INT |
12381 | | /* optimized version */ |
12382 | | int |
12383 | | ncx_putn_int_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
12384 | 0 | { |
12385 | | #ifdef WORDS_BIGENDIAN |
12386 | | (void) memcpy(*xpp, tp, (size_t)nelems * X_SIZEOF_INT); |
12387 | | # else |
12388 | 0 | swapn4b(*xpp, tp, nelems); |
12389 | 0 | # endif |
12390 | 0 | *xpp = (void *)((char *)(*xpp) + nelems * X_SIZEOF_INT); |
12391 | 0 | return NC_NOERR; |
12392 | 0 | } |
12393 | | #else |
12394 | | int |
12395 | | ncx_putn_int_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
12396 | | { |
12397 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12398 | | |
12399 | | /* basic algorithm is: |
12400 | | * - ensure sane alignment of output data |
12401 | | * - copy (conversion happens automatically) input data |
12402 | | * to output |
12403 | | * - update tp to point at next unconverted input, and xpp to point |
12404 | | * at next location for converted output |
12405 | | */ |
12406 | | long i, j, ni; |
12407 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12408 | | int *xp; |
12409 | | int nrange = 0; /* number of range errors */ |
12410 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12411 | | long cxp = (long) *((char**)xpp); |
12412 | | |
12413 | | realign = (cxp & 7) % SIZEOF_INT; |
12414 | | /* sjl: manually stripmine so we can limit amount of |
12415 | | * vector work space reserved to LOOPCNT elements. Also |
12416 | | * makes vectorisation easy */ |
12417 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12418 | | ni=Min(nelems-j,LOOPCNT); |
12419 | | if (realign) { |
12420 | | xp = tmp; |
12421 | | } else { |
12422 | | xp = (int *) *xpp; |
12423 | | } |
12424 | | /* copy the next block */ |
12425 | | #pragma cdir loopcnt=LOOPCNT |
12426 | | #pragma cdir shortloop |
12427 | | for (i=0; i<ni; i++) { |
12428 | | /* the normal case: */ |
12429 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
12430 | | /* test for range errors (not always needed but do it anyway) */ |
12431 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
12432 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
12433 | | nrange += tp[i] > X_INT_MAX || tp[i] < X_INT_MIN; |
12434 | | } |
12435 | | /* copy workspace back if necessary */ |
12436 | | if (realign) { |
12437 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12438 | | xp = (int *) *xpp; |
12439 | | } |
12440 | | /* update xpp and tp */ |
12441 | | xp += ni; |
12442 | | tp += ni; |
12443 | | *xpp = (void*)xp; |
12444 | | } |
12445 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12446 | | |
12447 | | #else /* not SX */ |
12448 | | |
12449 | | char *xp = (char *) *xpp; |
12450 | | int status = NC_NOERR; |
12451 | | |
12452 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12453 | | { |
12454 | | int lstatus = ncx_put_int_int(xp, tp, fillp); |
12455 | | if (status == NC_NOERR) /* report the first encountered error */ |
12456 | | status = lstatus; |
12457 | | } |
12458 | | |
12459 | | *xpp = (void *)xp; |
12460 | | return status; |
12461 | | #endif |
12462 | | } |
12463 | | |
12464 | | #endif |
12465 | | int |
12466 | | ncx_putn_int_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
12467 | 0 | { |
12468 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12469 | | |
12470 | | /* basic algorithm is: |
12471 | | * - ensure sane alignment of output data |
12472 | | * - copy (conversion happens automatically) input data |
12473 | | * to output |
12474 | | * - update tp to point at next unconverted input, and xpp to point |
12475 | | * at next location for converted output |
12476 | | */ |
12477 | | long i, j, ni; |
12478 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12479 | | int *xp; |
12480 | | int nrange = 0; /* number of range errors */ |
12481 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12482 | | long cxp = (long) *((char**)xpp); |
12483 | | |
12484 | | realign = (cxp & 7) % SIZEOF_INT; |
12485 | | /* sjl: manually stripmine so we can limit amount of |
12486 | | * vector work space reserved to LOOPCNT elements. Also |
12487 | | * makes vectorisation easy */ |
12488 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12489 | | ni=Min(nelems-j,LOOPCNT); |
12490 | | if (realign) { |
12491 | | xp = tmp; |
12492 | | } else { |
12493 | | xp = (int *) *xpp; |
12494 | | } |
12495 | | /* copy the next block */ |
12496 | | #pragma cdir loopcnt=LOOPCNT |
12497 | | #pragma cdir shortloop |
12498 | | for (i=0; i<ni; i++) { |
12499 | | /* the normal case: */ |
12500 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
12501 | | /* test for range errors (not always needed but do it anyway) */ |
12502 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
12503 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
12504 | | nrange += tp[i] > X_INT_MAX || tp[i] < X_INT_MIN; |
12505 | | } |
12506 | | /* copy workspace back if necessary */ |
12507 | | if (realign) { |
12508 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12509 | | xp = (int *) *xpp; |
12510 | | } |
12511 | | /* update xpp and tp */ |
12512 | | xp += ni; |
12513 | | tp += ni; |
12514 | | *xpp = (void*)xp; |
12515 | | } |
12516 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12517 | | |
12518 | | #else /* not SX */ |
12519 | |
|
12520 | 0 | char *xp = (char *) *xpp; |
12521 | 0 | int status = NC_NOERR; |
12522 | |
|
12523 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12524 | 0 | { |
12525 | 0 | int lstatus = ncx_put_int_schar(xp, tp, fillp); |
12526 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12527 | 0 | status = lstatus; |
12528 | 0 | } |
12529 | |
|
12530 | 0 | *xpp = (void *)xp; |
12531 | 0 | return status; |
12532 | 0 | #endif |
12533 | 0 | } |
12534 | | |
12535 | | int |
12536 | | ncx_putn_int_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
12537 | 0 | { |
12538 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12539 | | |
12540 | | /* basic algorithm is: |
12541 | | * - ensure sane alignment of output data |
12542 | | * - copy (conversion happens automatically) input data |
12543 | | * to output |
12544 | | * - update tp to point at next unconverted input, and xpp to point |
12545 | | * at next location for converted output |
12546 | | */ |
12547 | | long i, j, ni; |
12548 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12549 | | int *xp; |
12550 | | int nrange = 0; /* number of range errors */ |
12551 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12552 | | long cxp = (long) *((char**)xpp); |
12553 | | |
12554 | | realign = (cxp & 7) % SIZEOF_INT; |
12555 | | /* sjl: manually stripmine so we can limit amount of |
12556 | | * vector work space reserved to LOOPCNT elements. Also |
12557 | | * makes vectorisation easy */ |
12558 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12559 | | ni=Min(nelems-j,LOOPCNT); |
12560 | | if (realign) { |
12561 | | xp = tmp; |
12562 | | } else { |
12563 | | xp = (int *) *xpp; |
12564 | | } |
12565 | | /* copy the next block */ |
12566 | | #pragma cdir loopcnt=LOOPCNT |
12567 | | #pragma cdir shortloop |
12568 | | for (i=0; i<ni; i++) { |
12569 | | /* the normal case: */ |
12570 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
12571 | | /* test for range errors (not always needed but do it anyway) */ |
12572 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
12573 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
12574 | | nrange += tp[i] > X_INT_MAX || tp[i] < X_INT_MIN; |
12575 | | } |
12576 | | /* copy workspace back if necessary */ |
12577 | | if (realign) { |
12578 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12579 | | xp = (int *) *xpp; |
12580 | | } |
12581 | | /* update xpp and tp */ |
12582 | | xp += ni; |
12583 | | tp += ni; |
12584 | | *xpp = (void*)xp; |
12585 | | } |
12586 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12587 | | |
12588 | | #else /* not SX */ |
12589 | |
|
12590 | 0 | char *xp = (char *) *xpp; |
12591 | 0 | int status = NC_NOERR; |
12592 | |
|
12593 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12594 | 0 | { |
12595 | 0 | int lstatus = ncx_put_int_short(xp, tp, fillp); |
12596 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12597 | 0 | status = lstatus; |
12598 | 0 | } |
12599 | |
|
12600 | 0 | *xpp = (void *)xp; |
12601 | 0 | return status; |
12602 | 0 | #endif |
12603 | 0 | } |
12604 | | |
12605 | | int |
12606 | | ncx_putn_int_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
12607 | 0 | { |
12608 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12609 | | |
12610 | | /* basic algorithm is: |
12611 | | * - ensure sane alignment of output data |
12612 | | * - copy (conversion happens automatically) input data |
12613 | | * to output |
12614 | | * - update tp to point at next unconverted input, and xpp to point |
12615 | | * at next location for converted output |
12616 | | */ |
12617 | | long i, j, ni; |
12618 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12619 | | int *xp; |
12620 | | int nrange = 0; /* number of range errors */ |
12621 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12622 | | long cxp = (long) *((char**)xpp); |
12623 | | |
12624 | | realign = (cxp & 7) % SIZEOF_INT; |
12625 | | /* sjl: manually stripmine so we can limit amount of |
12626 | | * vector work space reserved to LOOPCNT elements. Also |
12627 | | * makes vectorisation easy */ |
12628 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12629 | | ni=Min(nelems-j,LOOPCNT); |
12630 | | if (realign) { |
12631 | | xp = tmp; |
12632 | | } else { |
12633 | | xp = (int *) *xpp; |
12634 | | } |
12635 | | /* copy the next block */ |
12636 | | #pragma cdir loopcnt=LOOPCNT |
12637 | | #pragma cdir shortloop |
12638 | | for (i=0; i<ni; i++) { |
12639 | | /* the normal case: */ |
12640 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
12641 | | /* test for range errors (not always needed but do it anyway) */ |
12642 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
12643 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
12644 | | nrange += tp[i] > X_INT_MAX || tp[i] < X_INT_MIN; |
12645 | | } |
12646 | | /* copy workspace back if necessary */ |
12647 | | if (realign) { |
12648 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12649 | | xp = (int *) *xpp; |
12650 | | } |
12651 | | /* update xpp and tp */ |
12652 | | xp += ni; |
12653 | | tp += ni; |
12654 | | *xpp = (void*)xp; |
12655 | | } |
12656 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12657 | | |
12658 | | #else /* not SX */ |
12659 | |
|
12660 | 0 | char *xp = (char *) *xpp; |
12661 | 0 | int status = NC_NOERR; |
12662 | |
|
12663 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12664 | 0 | { |
12665 | 0 | int lstatus = ncx_put_int_long(xp, tp, fillp); |
12666 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12667 | 0 | status = lstatus; |
12668 | 0 | } |
12669 | |
|
12670 | 0 | *xpp = (void *)xp; |
12671 | 0 | return status; |
12672 | 0 | #endif |
12673 | 0 | } |
12674 | | |
12675 | | int |
12676 | | ncx_putn_int_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
12677 | 0 | { |
12678 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12679 | | |
12680 | | /* basic algorithm is: |
12681 | | * - ensure sane alignment of output data |
12682 | | * - copy (conversion happens automatically) input data |
12683 | | * to output |
12684 | | * - update tp to point at next unconverted input, and xpp to point |
12685 | | * at next location for converted output |
12686 | | */ |
12687 | | long i, j, ni; |
12688 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12689 | | int *xp; |
12690 | | double d; /* special case for ncx_putn_int_float */ |
12691 | | int nrange = 0; /* number of range errors */ |
12692 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12693 | | long cxp = (long) *((char**)xpp); |
12694 | | |
12695 | | realign = (cxp & 7) % SIZEOF_INT; |
12696 | | /* sjl: manually stripmine so we can limit amount of |
12697 | | * vector work space reserved to LOOPCNT elements. Also |
12698 | | * makes vectorisation easy */ |
12699 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12700 | | ni=Min(nelems-j,LOOPCNT); |
12701 | | if (realign) { |
12702 | | xp = tmp; |
12703 | | } else { |
12704 | | xp = (int *) *xpp; |
12705 | | } |
12706 | | /* copy the next block */ |
12707 | | #pragma cdir loopcnt=LOOPCNT |
12708 | | #pragma cdir shortloop |
12709 | | for (i=0; i<ni; i++) { |
12710 | | /* for some reason int to float, for putn, requires a special case */ |
12711 | | d = tp[i]; |
12712 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) d)); |
12713 | | nrange += tp[i] > X_INT_MAX || tp[i] < X_INT_MIN; |
12714 | | } |
12715 | | /* copy workspace back if necessary */ |
12716 | | if (realign) { |
12717 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12718 | | xp = (int *) *xpp; |
12719 | | } |
12720 | | /* update xpp and tp */ |
12721 | | xp += ni; |
12722 | | tp += ni; |
12723 | | *xpp = (void*)xp; |
12724 | | } |
12725 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12726 | | |
12727 | | #else /* not SX */ |
12728 | |
|
12729 | 0 | char *xp = (char *) *xpp; |
12730 | 0 | int status = NC_NOERR; |
12731 | |
|
12732 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12733 | 0 | { |
12734 | 0 | int lstatus = ncx_put_int_float(xp, tp, fillp); |
12735 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12736 | 0 | status = lstatus; |
12737 | 0 | } |
12738 | |
|
12739 | 0 | *xpp = (void *)xp; |
12740 | 0 | return status; |
12741 | 0 | #endif |
12742 | 0 | } |
12743 | | |
12744 | | int |
12745 | | ncx_putn_int_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
12746 | 0 | { |
12747 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12748 | | |
12749 | | /* basic algorithm is: |
12750 | | * - ensure sane alignment of output data |
12751 | | * - copy (conversion happens automatically) input data |
12752 | | * to output |
12753 | | * - update tp to point at next unconverted input, and xpp to point |
12754 | | * at next location for converted output |
12755 | | */ |
12756 | | long i, j, ni; |
12757 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12758 | | int *xp; |
12759 | | int nrange = 0; /* number of range errors */ |
12760 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12761 | | long cxp = (long) *((char**)xpp); |
12762 | | |
12763 | | realign = (cxp & 7) % SIZEOF_INT; |
12764 | | /* sjl: manually stripmine so we can limit amount of |
12765 | | * vector work space reserved to LOOPCNT elements. Also |
12766 | | * makes vectorisation easy */ |
12767 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12768 | | ni=Min(nelems-j,LOOPCNT); |
12769 | | if (realign) { |
12770 | | xp = tmp; |
12771 | | } else { |
12772 | | xp = (int *) *xpp; |
12773 | | } |
12774 | | /* copy the next block */ |
12775 | | #pragma cdir loopcnt=LOOPCNT |
12776 | | #pragma cdir shortloop |
12777 | | for (i=0; i<ni; i++) { |
12778 | | /* the normal case: */ |
12779 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
12780 | | /* test for range errors (not always needed but do it anyway) */ |
12781 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
12782 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
12783 | | nrange += tp[i] > X_INT_MAX || tp[i] < X_INT_MIN; |
12784 | | } |
12785 | | /* copy workspace back if necessary */ |
12786 | | if (realign) { |
12787 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12788 | | xp = (int *) *xpp; |
12789 | | } |
12790 | | /* update xpp and tp */ |
12791 | | xp += ni; |
12792 | | tp += ni; |
12793 | | *xpp = (void*)xp; |
12794 | | } |
12795 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12796 | | |
12797 | | #else /* not SX */ |
12798 | |
|
12799 | 0 | char *xp = (char *) *xpp; |
12800 | 0 | int status = NC_NOERR; |
12801 | |
|
12802 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12803 | 0 | { |
12804 | 0 | int lstatus = ncx_put_int_double(xp, tp, fillp); |
12805 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12806 | 0 | status = lstatus; |
12807 | 0 | } |
12808 | |
|
12809 | 0 | *xpp = (void *)xp; |
12810 | 0 | return status; |
12811 | 0 | #endif |
12812 | 0 | } |
12813 | | |
12814 | | int |
12815 | | ncx_putn_int_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
12816 | 0 | { |
12817 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12818 | | |
12819 | | /* basic algorithm is: |
12820 | | * - ensure sane alignment of output data |
12821 | | * - copy (conversion happens automatically) input data |
12822 | | * to output |
12823 | | * - update tp to point at next unconverted input, and xpp to point |
12824 | | * at next location for converted output |
12825 | | */ |
12826 | | long i, j, ni; |
12827 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12828 | | int *xp; |
12829 | | int nrange = 0; /* number of range errors */ |
12830 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12831 | | long cxp = (long) *((char**)xpp); |
12832 | | |
12833 | | realign = (cxp & 7) % SIZEOF_INT; |
12834 | | /* sjl: manually stripmine so we can limit amount of |
12835 | | * vector work space reserved to LOOPCNT elements. Also |
12836 | | * makes vectorisation easy */ |
12837 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12838 | | ni=Min(nelems-j,LOOPCNT); |
12839 | | if (realign) { |
12840 | | xp = tmp; |
12841 | | } else { |
12842 | | xp = (int *) *xpp; |
12843 | | } |
12844 | | /* copy the next block */ |
12845 | | #pragma cdir loopcnt=LOOPCNT |
12846 | | #pragma cdir shortloop |
12847 | | for (i=0; i<ni; i++) { |
12848 | | /* the normal case: */ |
12849 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
12850 | | /* test for range errors (not always needed but do it anyway) */ |
12851 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
12852 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
12853 | | nrange += tp[i] > X_INT_MAX || tp[i] < X_INT_MIN; |
12854 | | } |
12855 | | /* copy workspace back if necessary */ |
12856 | | if (realign) { |
12857 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12858 | | xp = (int *) *xpp; |
12859 | | } |
12860 | | /* update xpp and tp */ |
12861 | | xp += ni; |
12862 | | tp += ni; |
12863 | | *xpp = (void*)xp; |
12864 | | } |
12865 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12866 | | |
12867 | | #else /* not SX */ |
12868 | |
|
12869 | 0 | char *xp = (char *) *xpp; |
12870 | 0 | int status = NC_NOERR; |
12871 | |
|
12872 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12873 | 0 | { |
12874 | 0 | int lstatus = ncx_put_int_longlong(xp, tp, fillp); |
12875 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12876 | 0 | status = lstatus; |
12877 | 0 | } |
12878 | |
|
12879 | 0 | *xpp = (void *)xp; |
12880 | 0 | return status; |
12881 | 0 | #endif |
12882 | 0 | } |
12883 | | |
12884 | | int |
12885 | | ncx_putn_int_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
12886 | 0 | { |
12887 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12888 | | |
12889 | | /* basic algorithm is: |
12890 | | * - ensure sane alignment of output data |
12891 | | * - copy (conversion happens automatically) input data |
12892 | | * to output |
12893 | | * - update tp to point at next unconverted input, and xpp to point |
12894 | | * at next location for converted output |
12895 | | */ |
12896 | | long i, j, ni; |
12897 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12898 | | int *xp; |
12899 | | int nrange = 0; /* number of range errors */ |
12900 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12901 | | long cxp = (long) *((char**)xpp); |
12902 | | |
12903 | | realign = (cxp & 7) % SIZEOF_INT; |
12904 | | /* sjl: manually stripmine so we can limit amount of |
12905 | | * vector work space reserved to LOOPCNT elements. Also |
12906 | | * makes vectorisation easy */ |
12907 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12908 | | ni=Min(nelems-j,LOOPCNT); |
12909 | | if (realign) { |
12910 | | xp = tmp; |
12911 | | } else { |
12912 | | xp = (int *) *xpp; |
12913 | | } |
12914 | | /* copy the next block */ |
12915 | | #pragma cdir loopcnt=LOOPCNT |
12916 | | #pragma cdir shortloop |
12917 | | for (i=0; i<ni; i++) { |
12918 | | /* the normal case: */ |
12919 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
12920 | | /* test for range errors (not always needed but do it anyway) */ |
12921 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
12922 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
12923 | | nrange += tp[i] > X_INT_MAX ; |
12924 | | } |
12925 | | /* copy workspace back if necessary */ |
12926 | | if (realign) { |
12927 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12928 | | xp = (int *) *xpp; |
12929 | | } |
12930 | | /* update xpp and tp */ |
12931 | | xp += ni; |
12932 | | tp += ni; |
12933 | | *xpp = (void*)xp; |
12934 | | } |
12935 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
12936 | | |
12937 | | #else /* not SX */ |
12938 | |
|
12939 | 0 | char *xp = (char *) *xpp; |
12940 | 0 | int status = NC_NOERR; |
12941 | |
|
12942 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
12943 | 0 | { |
12944 | 0 | int lstatus = ncx_put_int_uchar(xp, tp, fillp); |
12945 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
12946 | 0 | status = lstatus; |
12947 | 0 | } |
12948 | |
|
12949 | 0 | *xpp = (void *)xp; |
12950 | 0 | return status; |
12951 | 0 | #endif |
12952 | 0 | } |
12953 | | |
12954 | | int |
12955 | | ncx_putn_int_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
12956 | 0 | { |
12957 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
12958 | | |
12959 | | /* basic algorithm is: |
12960 | | * - ensure sane alignment of output data |
12961 | | * - copy (conversion happens automatically) input data |
12962 | | * to output |
12963 | | * - update tp to point at next unconverted input, and xpp to point |
12964 | | * at next location for converted output |
12965 | | */ |
12966 | | long i, j, ni; |
12967 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
12968 | | int *xp; |
12969 | | int nrange = 0; /* number of range errors */ |
12970 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
12971 | | long cxp = (long) *((char**)xpp); |
12972 | | |
12973 | | realign = (cxp & 7) % SIZEOF_INT; |
12974 | | /* sjl: manually stripmine so we can limit amount of |
12975 | | * vector work space reserved to LOOPCNT elements. Also |
12976 | | * makes vectorisation easy */ |
12977 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
12978 | | ni=Min(nelems-j,LOOPCNT); |
12979 | | if (realign) { |
12980 | | xp = tmp; |
12981 | | } else { |
12982 | | xp = (int *) *xpp; |
12983 | | } |
12984 | | /* copy the next block */ |
12985 | | #pragma cdir loopcnt=LOOPCNT |
12986 | | #pragma cdir shortloop |
12987 | | for (i=0; i<ni; i++) { |
12988 | | /* the normal case: */ |
12989 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
12990 | | /* test for range errors (not always needed but do it anyway) */ |
12991 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
12992 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
12993 | | nrange += tp[i] > X_INT_MAX ; |
12994 | | } |
12995 | | /* copy workspace back if necessary */ |
12996 | | if (realign) { |
12997 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
12998 | | xp = (int *) *xpp; |
12999 | | } |
13000 | | /* update xpp and tp */ |
13001 | | xp += ni; |
13002 | | tp += ni; |
13003 | | *xpp = (void*)xp; |
13004 | | } |
13005 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13006 | | |
13007 | | #else /* not SX */ |
13008 | |
|
13009 | 0 | char *xp = (char *) *xpp; |
13010 | 0 | int status = NC_NOERR; |
13011 | |
|
13012 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
13013 | 0 | { |
13014 | 0 | int lstatus = ncx_put_int_ushort(xp, tp, fillp); |
13015 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13016 | 0 | status = lstatus; |
13017 | 0 | } |
13018 | |
|
13019 | 0 | *xpp = (void *)xp; |
13020 | 0 | return status; |
13021 | 0 | #endif |
13022 | 0 | } |
13023 | | |
13024 | | int |
13025 | | ncx_putn_int_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
13026 | 0 | { |
13027 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
13028 | | |
13029 | | /* basic algorithm is: |
13030 | | * - ensure sane alignment of output data |
13031 | | * - copy (conversion happens automatically) input data |
13032 | | * to output |
13033 | | * - update tp to point at next unconverted input, and xpp to point |
13034 | | * at next location for converted output |
13035 | | */ |
13036 | | long i, j, ni; |
13037 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
13038 | | int *xp; |
13039 | | int nrange = 0; /* number of range errors */ |
13040 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13041 | | long cxp = (long) *((char**)xpp); |
13042 | | |
13043 | | realign = (cxp & 7) % SIZEOF_INT; |
13044 | | /* sjl: manually stripmine so we can limit amount of |
13045 | | * vector work space reserved to LOOPCNT elements. Also |
13046 | | * makes vectorisation easy */ |
13047 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13048 | | ni=Min(nelems-j,LOOPCNT); |
13049 | | if (realign) { |
13050 | | xp = tmp; |
13051 | | } else { |
13052 | | xp = (int *) *xpp; |
13053 | | } |
13054 | | /* copy the next block */ |
13055 | | #pragma cdir loopcnt=LOOPCNT |
13056 | | #pragma cdir shortloop |
13057 | | for (i=0; i<ni; i++) { |
13058 | | /* the normal case: */ |
13059 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
13060 | | /* test for range errors (not always needed but do it anyway) */ |
13061 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
13062 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
13063 | | nrange += tp[i] > X_INT_MAX ; |
13064 | | } |
13065 | | /* copy workspace back if necessary */ |
13066 | | if (realign) { |
13067 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
13068 | | xp = (int *) *xpp; |
13069 | | } |
13070 | | /* update xpp and tp */ |
13071 | | xp += ni; |
13072 | | tp += ni; |
13073 | | *xpp = (void*)xp; |
13074 | | } |
13075 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13076 | | |
13077 | | #else /* not SX */ |
13078 | |
|
13079 | 0 | char *xp = (char *) *xpp; |
13080 | 0 | int status = NC_NOERR; |
13081 | |
|
13082 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
13083 | 0 | { |
13084 | 0 | int lstatus = ncx_put_int_uint(xp, tp, fillp); |
13085 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13086 | 0 | status = lstatus; |
13087 | 0 | } |
13088 | |
|
13089 | 0 | *xpp = (void *)xp; |
13090 | 0 | return status; |
13091 | 0 | #endif |
13092 | 0 | } |
13093 | | |
13094 | | int |
13095 | | ncx_putn_int_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
13096 | 0 | { |
13097 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT == SIZEOF_INT |
13098 | | |
13099 | | /* basic algorithm is: |
13100 | | * - ensure sane alignment of output data |
13101 | | * - copy (conversion happens automatically) input data |
13102 | | * to output |
13103 | | * - update tp to point at next unconverted input, and xpp to point |
13104 | | * at next location for converted output |
13105 | | */ |
13106 | | long i, j, ni; |
13107 | | int tmp[LOOPCNT]; /* in case input is misaligned */ |
13108 | | int *xp; |
13109 | | int nrange = 0; /* number of range errors */ |
13110 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13111 | | long cxp = (long) *((char**)xpp); |
13112 | | |
13113 | | realign = (cxp & 7) % SIZEOF_INT; |
13114 | | /* sjl: manually stripmine so we can limit amount of |
13115 | | * vector work space reserved to LOOPCNT elements. Also |
13116 | | * makes vectorisation easy */ |
13117 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13118 | | ni=Min(nelems-j,LOOPCNT); |
13119 | | if (realign) { |
13120 | | xp = tmp; |
13121 | | } else { |
13122 | | xp = (int *) *xpp; |
13123 | | } |
13124 | | /* copy the next block */ |
13125 | | #pragma cdir loopcnt=LOOPCNT |
13126 | | #pragma cdir shortloop |
13127 | | for (i=0; i<ni; i++) { |
13128 | | /* the normal case: */ |
13129 | | xp[i] = (int) Max( X_INT_MIN, Min(X_INT_MAX, (int) tp[i])); |
13130 | | /* test for range errors (not always needed but do it anyway) */ |
13131 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
13132 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
13133 | | nrange += tp[i] > X_INT_MAX ; |
13134 | | } |
13135 | | /* copy workspace back if necessary */ |
13136 | | if (realign) { |
13137 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT); |
13138 | | xp = (int *) *xpp; |
13139 | | } |
13140 | | /* update xpp and tp */ |
13141 | | xp += ni; |
13142 | | tp += ni; |
13143 | | *xpp = (void*)xp; |
13144 | | } |
13145 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13146 | | |
13147 | | #else /* not SX */ |
13148 | |
|
13149 | 0 | char *xp = (char *) *xpp; |
13150 | 0 | int status = NC_NOERR; |
13151 | |
|
13152 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT, tp++) |
13153 | 0 | { |
13154 | 0 | int lstatus = ncx_put_int_ulonglong(xp, tp, fillp); |
13155 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13156 | 0 | status = lstatus; |
13157 | 0 | } |
13158 | |
|
13159 | 0 | *xpp = (void *)xp; |
13160 | 0 | return status; |
13161 | 0 | #endif |
13162 | 0 | } |
13163 | | |
13164 | | |
13165 | | /* uint ----------------------------------------------------------------------*/ |
13166 | | |
13167 | | #if X_SIZEOF_UINT == SIZEOF_UINT |
13168 | | /* optimized version */ |
13169 | | int |
13170 | | ncx_getn_uint_uint(const void **xpp, size_t nelems, unsigned int *tp) |
13171 | 0 | { |
13172 | | #ifdef WORDS_BIGENDIAN |
13173 | | (void) memcpy(tp, *xpp, (size_t)nelems * SIZEOF_UINT); |
13174 | | # else |
13175 | 0 | swapn4b(tp, *xpp, nelems); |
13176 | 0 | # endif |
13177 | 0 | *xpp = (const void *)((const char *)(*xpp) + nelems * X_SIZEOF_UINT); |
13178 | 0 | return NC_NOERR; |
13179 | 0 | } |
13180 | | #else |
13181 | | int |
13182 | | ncx_getn_uint_uint(const void **xpp, size_t nelems, uint *tp) |
13183 | | { |
13184 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13185 | | |
13186 | | /* basic algorithm is: |
13187 | | * - ensure sane alignment of input data |
13188 | | * - copy (conversion happens automatically) input data |
13189 | | * to output |
13190 | | * - update xpp to point at next unconverted input, and tp to point |
13191 | | * at next location for converted output |
13192 | | */ |
13193 | | long i, j, ni; |
13194 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13195 | | uint *xp; |
13196 | | int nrange = 0; /* number of range errors */ |
13197 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13198 | | long cxp = (long) *((char**)xpp); |
13199 | | |
13200 | | realign = (cxp & 7) % SIZEOF_UINT; |
13201 | | /* sjl: manually stripmine so we can limit amount of |
13202 | | * vector work space reserved to LOOPCNT elements. Also |
13203 | | * makes vectorisation easy */ |
13204 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13205 | | ni=Min(nelems-j,LOOPCNT); |
13206 | | if (realign) { |
13207 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13208 | | xp = tmp; |
13209 | | } else { |
13210 | | xp = (uint *) *xpp; |
13211 | | } |
13212 | | /* copy the next block */ |
13213 | | #pragma cdir loopcnt=LOOPCNT |
13214 | | #pragma cdir shortloop |
13215 | | for (i=0; i<ni; i++) { |
13216 | | tp[i] = (uint) Max( UINT_MIN, Min(UINT_MAX, (uint) xp[i])); |
13217 | | /* test for range errors (not always needed but do it anyway) */ |
13218 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13219 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13220 | | nrange += xp[i] > UINT_MAX ; |
13221 | | } |
13222 | | /* update xpp and tp */ |
13223 | | if (realign) xp = (uint *) *xpp; |
13224 | | xp += ni; |
13225 | | tp += ni; |
13226 | | *xpp = (void*)xp; |
13227 | | } |
13228 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13229 | | |
13230 | | #else /* not SX */ |
13231 | | const char *xp = (const char *) *xpp; |
13232 | | int status = NC_NOERR; |
13233 | | |
13234 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13235 | | { |
13236 | | const int lstatus = ncx_get_uint_uint(xp, tp); |
13237 | | if (status == NC_NOERR) /* report the first encountered error */ |
13238 | | status = lstatus; |
13239 | | } |
13240 | | |
13241 | | *xpp = (const void *)xp; |
13242 | | return status; |
13243 | | #endif |
13244 | | } |
13245 | | |
13246 | | #endif |
13247 | | int |
13248 | | ncx_getn_uint_schar(const void **xpp, size_t nelems, schar *tp) |
13249 | 0 | { |
13250 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13251 | | |
13252 | | /* basic algorithm is: |
13253 | | * - ensure sane alignment of input data |
13254 | | * - copy (conversion happens automatically) input data |
13255 | | * to output |
13256 | | * - update xpp to point at next unconverted input, and tp to point |
13257 | | * at next location for converted output |
13258 | | */ |
13259 | | long i, j, ni; |
13260 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13261 | | uint *xp; |
13262 | | int nrange = 0; /* number of range errors */ |
13263 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13264 | | long cxp = (long) *((char**)xpp); |
13265 | | |
13266 | | realign = (cxp & 7) % SIZEOF_UINT; |
13267 | | /* sjl: manually stripmine so we can limit amount of |
13268 | | * vector work space reserved to LOOPCNT elements. Also |
13269 | | * makes vectorisation easy */ |
13270 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13271 | | ni=Min(nelems-j,LOOPCNT); |
13272 | | if (realign) { |
13273 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13274 | | xp = tmp; |
13275 | | } else { |
13276 | | xp = (uint *) *xpp; |
13277 | | } |
13278 | | /* copy the next block */ |
13279 | | #pragma cdir loopcnt=LOOPCNT |
13280 | | #pragma cdir shortloop |
13281 | | for (i=0; i<ni; i++) { |
13282 | | tp[i] = (schar) Max( SCHAR_MIN, Min(SCHAR_MAX, (schar) xp[i])); |
13283 | | /* test for range errors (not always needed but do it anyway) */ |
13284 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13285 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13286 | | nrange += xp[i] > SCHAR_MAX ; |
13287 | | } |
13288 | | /* update xpp and tp */ |
13289 | | if (realign) xp = (uint *) *xpp; |
13290 | | xp += ni; |
13291 | | tp += ni; |
13292 | | *xpp = (void*)xp; |
13293 | | } |
13294 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13295 | | |
13296 | | #else /* not SX */ |
13297 | 0 | const char *xp = (const char *) *xpp; |
13298 | 0 | int status = NC_NOERR; |
13299 | |
|
13300 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13301 | 0 | { |
13302 | 0 | const int lstatus = ncx_get_uint_schar(xp, tp); |
13303 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13304 | 0 | status = lstatus; |
13305 | 0 | } |
13306 | |
|
13307 | 0 | *xpp = (const void *)xp; |
13308 | 0 | return status; |
13309 | 0 | #endif |
13310 | 0 | } |
13311 | | |
13312 | | int |
13313 | | ncx_getn_uint_short(const void **xpp, size_t nelems, short *tp) |
13314 | 0 | { |
13315 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13316 | | |
13317 | | /* basic algorithm is: |
13318 | | * - ensure sane alignment of input data |
13319 | | * - copy (conversion happens automatically) input data |
13320 | | * to output |
13321 | | * - update xpp to point at next unconverted input, and tp to point |
13322 | | * at next location for converted output |
13323 | | */ |
13324 | | long i, j, ni; |
13325 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13326 | | uint *xp; |
13327 | | int nrange = 0; /* number of range errors */ |
13328 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13329 | | long cxp = (long) *((char**)xpp); |
13330 | | |
13331 | | realign = (cxp & 7) % SIZEOF_UINT; |
13332 | | /* sjl: manually stripmine so we can limit amount of |
13333 | | * vector work space reserved to LOOPCNT elements. Also |
13334 | | * makes vectorisation easy */ |
13335 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13336 | | ni=Min(nelems-j,LOOPCNT); |
13337 | | if (realign) { |
13338 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13339 | | xp = tmp; |
13340 | | } else { |
13341 | | xp = (uint *) *xpp; |
13342 | | } |
13343 | | /* copy the next block */ |
13344 | | #pragma cdir loopcnt=LOOPCNT |
13345 | | #pragma cdir shortloop |
13346 | | for (i=0; i<ni; i++) { |
13347 | | tp[i] = (short) Max( SHORT_MIN, Min(SHORT_MAX, (short) xp[i])); |
13348 | | /* test for range errors (not always needed but do it anyway) */ |
13349 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13350 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13351 | | nrange += xp[i] > SHORT_MAX ; |
13352 | | } |
13353 | | /* update xpp and tp */ |
13354 | | if (realign) xp = (uint *) *xpp; |
13355 | | xp += ni; |
13356 | | tp += ni; |
13357 | | *xpp = (void*)xp; |
13358 | | } |
13359 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13360 | | |
13361 | | #else /* not SX */ |
13362 | 0 | const char *xp = (const char *) *xpp; |
13363 | 0 | int status = NC_NOERR; |
13364 | |
|
13365 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13366 | 0 | { |
13367 | 0 | const int lstatus = ncx_get_uint_short(xp, tp); |
13368 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13369 | 0 | status = lstatus; |
13370 | 0 | } |
13371 | |
|
13372 | 0 | *xpp = (const void *)xp; |
13373 | 0 | return status; |
13374 | 0 | #endif |
13375 | 0 | } |
13376 | | |
13377 | | int |
13378 | | ncx_getn_uint_int(const void **xpp, size_t nelems, int *tp) |
13379 | 0 | { |
13380 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13381 | | |
13382 | | /* basic algorithm is: |
13383 | | * - ensure sane alignment of input data |
13384 | | * - copy (conversion happens automatically) input data |
13385 | | * to output |
13386 | | * - update xpp to point at next unconverted input, and tp to point |
13387 | | * at next location for converted output |
13388 | | */ |
13389 | | long i, j, ni; |
13390 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13391 | | uint *xp; |
13392 | | int nrange = 0; /* number of range errors */ |
13393 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13394 | | long cxp = (long) *((char**)xpp); |
13395 | | |
13396 | | realign = (cxp & 7) % SIZEOF_UINT; |
13397 | | /* sjl: manually stripmine so we can limit amount of |
13398 | | * vector work space reserved to LOOPCNT elements. Also |
13399 | | * makes vectorisation easy */ |
13400 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13401 | | ni=Min(nelems-j,LOOPCNT); |
13402 | | if (realign) { |
13403 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13404 | | xp = tmp; |
13405 | | } else { |
13406 | | xp = (uint *) *xpp; |
13407 | | } |
13408 | | /* copy the next block */ |
13409 | | #pragma cdir loopcnt=LOOPCNT |
13410 | | #pragma cdir shortloop |
13411 | | for (i=0; i<ni; i++) { |
13412 | | tp[i] = (int) Max( INT_MIN, Min(INT_MAX, (int) xp[i])); |
13413 | | /* test for range errors (not always needed but do it anyway) */ |
13414 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13415 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13416 | | nrange += xp[i] > INT_MAX ; |
13417 | | } |
13418 | | /* update xpp and tp */ |
13419 | | if (realign) xp = (uint *) *xpp; |
13420 | | xp += ni; |
13421 | | tp += ni; |
13422 | | *xpp = (void*)xp; |
13423 | | } |
13424 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13425 | | |
13426 | | #else /* not SX */ |
13427 | 0 | const char *xp = (const char *) *xpp; |
13428 | 0 | int status = NC_NOERR; |
13429 | |
|
13430 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13431 | 0 | { |
13432 | 0 | const int lstatus = ncx_get_uint_int(xp, tp); |
13433 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13434 | 0 | status = lstatus; |
13435 | 0 | } |
13436 | |
|
13437 | 0 | *xpp = (const void *)xp; |
13438 | 0 | return status; |
13439 | 0 | #endif |
13440 | 0 | } |
13441 | | |
13442 | | int |
13443 | | ncx_getn_uint_long(const void **xpp, size_t nelems, long *tp) |
13444 | 0 | { |
13445 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13446 | | |
13447 | | /* basic algorithm is: |
13448 | | * - ensure sane alignment of input data |
13449 | | * - copy (conversion happens automatically) input data |
13450 | | * to output |
13451 | | * - update xpp to point at next unconverted input, and tp to point |
13452 | | * at next location for converted output |
13453 | | */ |
13454 | | long i, j, ni; |
13455 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13456 | | uint *xp; |
13457 | | int nrange = 0; /* number of range errors */ |
13458 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13459 | | long cxp = (long) *((char**)xpp); |
13460 | | |
13461 | | realign = (cxp & 7) % SIZEOF_UINT; |
13462 | | /* sjl: manually stripmine so we can limit amount of |
13463 | | * vector work space reserved to LOOPCNT elements. Also |
13464 | | * makes vectorisation easy */ |
13465 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13466 | | ni=Min(nelems-j,LOOPCNT); |
13467 | | if (realign) { |
13468 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13469 | | xp = tmp; |
13470 | | } else { |
13471 | | xp = (uint *) *xpp; |
13472 | | } |
13473 | | /* copy the next block */ |
13474 | | #pragma cdir loopcnt=LOOPCNT |
13475 | | #pragma cdir shortloop |
13476 | | for (i=0; i<ni; i++) { |
13477 | | tp[i] = (long) Max( LONG_MIN, Min(LONG_MAX, (long) xp[i])); |
13478 | | /* test for range errors (not always needed but do it anyway) */ |
13479 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13480 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13481 | | nrange += xp[i] > LONG_MAX ; |
13482 | | } |
13483 | | /* update xpp and tp */ |
13484 | | if (realign) xp = (uint *) *xpp; |
13485 | | xp += ni; |
13486 | | tp += ni; |
13487 | | *xpp = (void*)xp; |
13488 | | } |
13489 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13490 | | |
13491 | | #else /* not SX */ |
13492 | 0 | const char *xp = (const char *) *xpp; |
13493 | 0 | int status = NC_NOERR; |
13494 | |
|
13495 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13496 | 0 | { |
13497 | 0 | const int lstatus = ncx_get_uint_long(xp, tp); |
13498 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13499 | 0 | status = lstatus; |
13500 | 0 | } |
13501 | |
|
13502 | 0 | *xpp = (const void *)xp; |
13503 | 0 | return status; |
13504 | 0 | #endif |
13505 | 0 | } |
13506 | | |
13507 | | int |
13508 | | ncx_getn_uint_float(const void **xpp, size_t nelems, float *tp) |
13509 | 0 | { |
13510 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13511 | | |
13512 | | /* basic algorithm is: |
13513 | | * - ensure sane alignment of input data |
13514 | | * - copy (conversion happens automatically) input data |
13515 | | * to output |
13516 | | * - update xpp to point at next unconverted input, and tp to point |
13517 | | * at next location for converted output |
13518 | | */ |
13519 | | long i, j, ni; |
13520 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13521 | | uint *xp; |
13522 | | int nrange = 0; /* number of range errors */ |
13523 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13524 | | long cxp = (long) *((char**)xpp); |
13525 | | |
13526 | | realign = (cxp & 7) % SIZEOF_UINT; |
13527 | | /* sjl: manually stripmine so we can limit amount of |
13528 | | * vector work space reserved to LOOPCNT elements. Also |
13529 | | * makes vectorisation easy */ |
13530 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13531 | | ni=Min(nelems-j,LOOPCNT); |
13532 | | if (realign) { |
13533 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13534 | | xp = tmp; |
13535 | | } else { |
13536 | | xp = (uint *) *xpp; |
13537 | | } |
13538 | | /* copy the next block */ |
13539 | | #pragma cdir loopcnt=LOOPCNT |
13540 | | #pragma cdir shortloop |
13541 | | for (i=0; i<ni; i++) { |
13542 | | tp[i] = (float) Max( FLOAT_MIN, Min(FLOAT_MAX, (float) xp[i])); |
13543 | | /* test for range errors (not always needed but do it anyway) */ |
13544 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13545 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13546 | | nrange += xp[i] > FLOAT_MAX ; |
13547 | | } |
13548 | | /* update xpp and tp */ |
13549 | | if (realign) xp = (uint *) *xpp; |
13550 | | xp += ni; |
13551 | | tp += ni; |
13552 | | *xpp = (void*)xp; |
13553 | | } |
13554 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13555 | | |
13556 | | #else /* not SX */ |
13557 | 0 | const char *xp = (const char *) *xpp; |
13558 | 0 | int status = NC_NOERR; |
13559 | |
|
13560 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13561 | 0 | { |
13562 | 0 | const int lstatus = ncx_get_uint_float(xp, tp); |
13563 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13564 | 0 | status = lstatus; |
13565 | 0 | } |
13566 | |
|
13567 | 0 | *xpp = (const void *)xp; |
13568 | 0 | return status; |
13569 | 0 | #endif |
13570 | 0 | } |
13571 | | |
13572 | | int |
13573 | | ncx_getn_uint_double(const void **xpp, size_t nelems, double *tp) |
13574 | 0 | { |
13575 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13576 | | |
13577 | | /* basic algorithm is: |
13578 | | * - ensure sane alignment of input data |
13579 | | * - copy (conversion happens automatically) input data |
13580 | | * to output |
13581 | | * - update xpp to point at next unconverted input, and tp to point |
13582 | | * at next location for converted output |
13583 | | */ |
13584 | | long i, j, ni; |
13585 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13586 | | uint *xp; |
13587 | | int nrange = 0; /* number of range errors */ |
13588 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13589 | | long cxp = (long) *((char**)xpp); |
13590 | | |
13591 | | realign = (cxp & 7) % SIZEOF_UINT; |
13592 | | /* sjl: manually stripmine so we can limit amount of |
13593 | | * vector work space reserved to LOOPCNT elements. Also |
13594 | | * makes vectorisation easy */ |
13595 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13596 | | ni=Min(nelems-j,LOOPCNT); |
13597 | | if (realign) { |
13598 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13599 | | xp = tmp; |
13600 | | } else { |
13601 | | xp = (uint *) *xpp; |
13602 | | } |
13603 | | /* copy the next block */ |
13604 | | #pragma cdir loopcnt=LOOPCNT |
13605 | | #pragma cdir shortloop |
13606 | | for (i=0; i<ni; i++) { |
13607 | | tp[i] = (double) Max( DOUBLE_MIN, Min(DOUBLE_MAX, (double) xp[i])); |
13608 | | /* test for range errors (not always needed but do it anyway) */ |
13609 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13610 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13611 | | nrange += xp[i] > DOUBLE_MAX ; |
13612 | | } |
13613 | | /* update xpp and tp */ |
13614 | | if (realign) xp = (uint *) *xpp; |
13615 | | xp += ni; |
13616 | | tp += ni; |
13617 | | *xpp = (void*)xp; |
13618 | | } |
13619 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13620 | | |
13621 | | #else /* not SX */ |
13622 | 0 | const char *xp = (const char *) *xpp; |
13623 | 0 | int status = NC_NOERR; |
13624 | |
|
13625 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13626 | 0 | { |
13627 | 0 | const int lstatus = ncx_get_uint_double(xp, tp); |
13628 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13629 | 0 | status = lstatus; |
13630 | 0 | } |
13631 | |
|
13632 | 0 | *xpp = (const void *)xp; |
13633 | 0 | return status; |
13634 | 0 | #endif |
13635 | 0 | } |
13636 | | |
13637 | | int |
13638 | | ncx_getn_uint_longlong(const void **xpp, size_t nelems, longlong *tp) |
13639 | 0 | { |
13640 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13641 | | |
13642 | | /* basic algorithm is: |
13643 | | * - ensure sane alignment of input data |
13644 | | * - copy (conversion happens automatically) input data |
13645 | | * to output |
13646 | | * - update xpp to point at next unconverted input, and tp to point |
13647 | | * at next location for converted output |
13648 | | */ |
13649 | | long i, j, ni; |
13650 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13651 | | uint *xp; |
13652 | | int nrange = 0; /* number of range errors */ |
13653 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13654 | | long cxp = (long) *((char**)xpp); |
13655 | | |
13656 | | realign = (cxp & 7) % SIZEOF_UINT; |
13657 | | /* sjl: manually stripmine so we can limit amount of |
13658 | | * vector work space reserved to LOOPCNT elements. Also |
13659 | | * makes vectorisation easy */ |
13660 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13661 | | ni=Min(nelems-j,LOOPCNT); |
13662 | | if (realign) { |
13663 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13664 | | xp = tmp; |
13665 | | } else { |
13666 | | xp = (uint *) *xpp; |
13667 | | } |
13668 | | /* copy the next block */ |
13669 | | #pragma cdir loopcnt=LOOPCNT |
13670 | | #pragma cdir shortloop |
13671 | | for (i=0; i<ni; i++) { |
13672 | | tp[i] = (longlong) Max( LONGLONG_MIN, Min(LONGLONG_MAX, (longlong) xp[i])); |
13673 | | /* test for range errors (not always needed but do it anyway) */ |
13674 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13675 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13676 | | nrange += xp[i] > LONGLONG_MAX ; |
13677 | | } |
13678 | | /* update xpp and tp */ |
13679 | | if (realign) xp = (uint *) *xpp; |
13680 | | xp += ni; |
13681 | | tp += ni; |
13682 | | *xpp = (void*)xp; |
13683 | | } |
13684 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13685 | | |
13686 | | #else /* not SX */ |
13687 | 0 | const char *xp = (const char *) *xpp; |
13688 | 0 | int status = NC_NOERR; |
13689 | |
|
13690 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13691 | 0 | { |
13692 | 0 | const int lstatus = ncx_get_uint_longlong(xp, tp); |
13693 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13694 | 0 | status = lstatus; |
13695 | 0 | } |
13696 | |
|
13697 | 0 | *xpp = (const void *)xp; |
13698 | 0 | return status; |
13699 | 0 | #endif |
13700 | 0 | } |
13701 | | |
13702 | | int |
13703 | | ncx_getn_uint_uchar(const void **xpp, size_t nelems, uchar *tp) |
13704 | 0 | { |
13705 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13706 | | |
13707 | | /* basic algorithm is: |
13708 | | * - ensure sane alignment of input data |
13709 | | * - copy (conversion happens automatically) input data |
13710 | | * to output |
13711 | | * - update xpp to point at next unconverted input, and tp to point |
13712 | | * at next location for converted output |
13713 | | */ |
13714 | | long i, j, ni; |
13715 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13716 | | uint *xp; |
13717 | | int nrange = 0; /* number of range errors */ |
13718 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13719 | | long cxp = (long) *((char**)xpp); |
13720 | | |
13721 | | realign = (cxp & 7) % SIZEOF_UINT; |
13722 | | /* sjl: manually stripmine so we can limit amount of |
13723 | | * vector work space reserved to LOOPCNT elements. Also |
13724 | | * makes vectorisation easy */ |
13725 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13726 | | ni=Min(nelems-j,LOOPCNT); |
13727 | | if (realign) { |
13728 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13729 | | xp = tmp; |
13730 | | } else { |
13731 | | xp = (uint *) *xpp; |
13732 | | } |
13733 | | /* copy the next block */ |
13734 | | #pragma cdir loopcnt=LOOPCNT |
13735 | | #pragma cdir shortloop |
13736 | | for (i=0; i<ni; i++) { |
13737 | | tp[i] = (uchar) Max( UCHAR_MIN, Min(UCHAR_MAX, (uchar) xp[i])); |
13738 | | /* test for range errors (not always needed but do it anyway) */ |
13739 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13740 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13741 | | nrange += xp[i] > UCHAR_MAX ; |
13742 | | } |
13743 | | /* update xpp and tp */ |
13744 | | if (realign) xp = (uint *) *xpp; |
13745 | | xp += ni; |
13746 | | tp += ni; |
13747 | | *xpp = (void*)xp; |
13748 | | } |
13749 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13750 | | |
13751 | | #else /* not SX */ |
13752 | 0 | const char *xp = (const char *) *xpp; |
13753 | 0 | int status = NC_NOERR; |
13754 | |
|
13755 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13756 | 0 | { |
13757 | 0 | const int lstatus = ncx_get_uint_uchar(xp, tp); |
13758 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13759 | 0 | status = lstatus; |
13760 | 0 | } |
13761 | |
|
13762 | 0 | *xpp = (const void *)xp; |
13763 | 0 | return status; |
13764 | 0 | #endif |
13765 | 0 | } |
13766 | | |
13767 | | int |
13768 | | ncx_getn_uint_ushort(const void **xpp, size_t nelems, ushort *tp) |
13769 | 0 | { |
13770 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13771 | | |
13772 | | /* basic algorithm is: |
13773 | | * - ensure sane alignment of input data |
13774 | | * - copy (conversion happens automatically) input data |
13775 | | * to output |
13776 | | * - update xpp to point at next unconverted input, and tp to point |
13777 | | * at next location for converted output |
13778 | | */ |
13779 | | long i, j, ni; |
13780 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13781 | | uint *xp; |
13782 | | int nrange = 0; /* number of range errors */ |
13783 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13784 | | long cxp = (long) *((char**)xpp); |
13785 | | |
13786 | | realign = (cxp & 7) % SIZEOF_UINT; |
13787 | | /* sjl: manually stripmine so we can limit amount of |
13788 | | * vector work space reserved to LOOPCNT elements. Also |
13789 | | * makes vectorisation easy */ |
13790 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13791 | | ni=Min(nelems-j,LOOPCNT); |
13792 | | if (realign) { |
13793 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13794 | | xp = tmp; |
13795 | | } else { |
13796 | | xp = (uint *) *xpp; |
13797 | | } |
13798 | | /* copy the next block */ |
13799 | | #pragma cdir loopcnt=LOOPCNT |
13800 | | #pragma cdir shortloop |
13801 | | for (i=0; i<ni; i++) { |
13802 | | tp[i] = (ushort) Max( USHORT_MIN, Min(USHORT_MAX, (ushort) xp[i])); |
13803 | | /* test for range errors (not always needed but do it anyway) */ |
13804 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13805 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13806 | | nrange += xp[i] > USHORT_MAX ; |
13807 | | } |
13808 | | /* update xpp and tp */ |
13809 | | if (realign) xp = (uint *) *xpp; |
13810 | | xp += ni; |
13811 | | tp += ni; |
13812 | | *xpp = (void*)xp; |
13813 | | } |
13814 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13815 | | |
13816 | | #else /* not SX */ |
13817 | 0 | const char *xp = (const char *) *xpp; |
13818 | 0 | int status = NC_NOERR; |
13819 | |
|
13820 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13821 | 0 | { |
13822 | 0 | const int lstatus = ncx_get_uint_ushort(xp, tp); |
13823 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13824 | 0 | status = lstatus; |
13825 | 0 | } |
13826 | |
|
13827 | 0 | *xpp = (const void *)xp; |
13828 | 0 | return status; |
13829 | 0 | #endif |
13830 | 0 | } |
13831 | | |
13832 | | int |
13833 | | ncx_getn_uint_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
13834 | 0 | { |
13835 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13836 | | |
13837 | | /* basic algorithm is: |
13838 | | * - ensure sane alignment of input data |
13839 | | * - copy (conversion happens automatically) input data |
13840 | | * to output |
13841 | | * - update xpp to point at next unconverted input, and tp to point |
13842 | | * at next location for converted output |
13843 | | */ |
13844 | | long i, j, ni; |
13845 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13846 | | uint *xp; |
13847 | | int nrange = 0; /* number of range errors */ |
13848 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13849 | | long cxp = (long) *((char**)xpp); |
13850 | | |
13851 | | realign = (cxp & 7) % SIZEOF_UINT; |
13852 | | /* sjl: manually stripmine so we can limit amount of |
13853 | | * vector work space reserved to LOOPCNT elements. Also |
13854 | | * makes vectorisation easy */ |
13855 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13856 | | ni=Min(nelems-j,LOOPCNT); |
13857 | | if (realign) { |
13858 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT)); |
13859 | | xp = tmp; |
13860 | | } else { |
13861 | | xp = (uint *) *xpp; |
13862 | | } |
13863 | | /* copy the next block */ |
13864 | | #pragma cdir loopcnt=LOOPCNT |
13865 | | #pragma cdir shortloop |
13866 | | for (i=0; i<ni; i++) { |
13867 | | tp[i] = (ulonglong) Max( ULONGLONG_MIN, Min(ULONGLONG_MAX, (ulonglong) xp[i])); |
13868 | | /* test for range errors (not always needed but do it anyway) */ |
13869 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
13870 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
13871 | | nrange += xp[i] > ULONGLONG_MAX ; |
13872 | | } |
13873 | | /* update xpp and tp */ |
13874 | | if (realign) xp = (uint *) *xpp; |
13875 | | xp += ni; |
13876 | | tp += ni; |
13877 | | *xpp = (void*)xp; |
13878 | | } |
13879 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13880 | | |
13881 | | #else /* not SX */ |
13882 | 0 | const char *xp = (const char *) *xpp; |
13883 | 0 | int status = NC_NOERR; |
13884 | |
|
13885 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13886 | 0 | { |
13887 | 0 | const int lstatus = ncx_get_uint_ulonglong(xp, tp); |
13888 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
13889 | 0 | status = lstatus; |
13890 | 0 | } |
13891 | |
|
13892 | 0 | *xpp = (const void *)xp; |
13893 | 0 | return status; |
13894 | 0 | #endif |
13895 | 0 | } |
13896 | | |
13897 | | |
13898 | | #if X_SIZEOF_UINT == SIZEOF_UINT |
13899 | | /* optimized version */ |
13900 | | int |
13901 | | ncx_putn_uint_uint(void **xpp, size_t nelems, const unsigned int *tp, void *fillp) |
13902 | 0 | { |
13903 | | #ifdef WORDS_BIGENDIAN |
13904 | | (void) memcpy(*xpp, tp, (size_t)nelems * X_SIZEOF_UINT); |
13905 | | # else |
13906 | 0 | swapn4b(*xpp, tp, nelems); |
13907 | 0 | # endif |
13908 | 0 | *xpp = (void *)((char *)(*xpp) + nelems * X_SIZEOF_UINT); |
13909 | 0 | return NC_NOERR; |
13910 | 0 | } |
13911 | | #else |
13912 | | int |
13913 | | ncx_putn_uint_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
13914 | | { |
13915 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13916 | | |
13917 | | /* basic algorithm is: |
13918 | | * - ensure sane alignment of output data |
13919 | | * - copy (conversion happens automatically) input data |
13920 | | * to output |
13921 | | * - update tp to point at next unconverted input, and xpp to point |
13922 | | * at next location for converted output |
13923 | | */ |
13924 | | long i, j, ni; |
13925 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13926 | | uint *xp; |
13927 | | int nrange = 0; /* number of range errors */ |
13928 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
13929 | | long cxp = (long) *((char**)xpp); |
13930 | | |
13931 | | realign = (cxp & 7) % SIZEOF_UINT; |
13932 | | /* sjl: manually stripmine so we can limit amount of |
13933 | | * vector work space reserved to LOOPCNT elements. Also |
13934 | | * makes vectorisation easy */ |
13935 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
13936 | | ni=Min(nelems-j,LOOPCNT); |
13937 | | if (realign) { |
13938 | | xp = tmp; |
13939 | | } else { |
13940 | | xp = (uint *) *xpp; |
13941 | | } |
13942 | | /* copy the next block */ |
13943 | | #pragma cdir loopcnt=LOOPCNT |
13944 | | #pragma cdir shortloop |
13945 | | for (i=0; i<ni; i++) { |
13946 | | /* the normal case: */ |
13947 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
13948 | | /* test for range errors (not always needed but do it anyway) */ |
13949 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
13950 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
13951 | | nrange += tp[i] > X_UINT_MAX ; |
13952 | | } |
13953 | | /* copy workspace back if necessary */ |
13954 | | if (realign) { |
13955 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
13956 | | xp = (uint *) *xpp; |
13957 | | } |
13958 | | /* update xpp and tp */ |
13959 | | xp += ni; |
13960 | | tp += ni; |
13961 | | *xpp = (void*)xp; |
13962 | | } |
13963 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
13964 | | |
13965 | | #else /* not SX */ |
13966 | | |
13967 | | char *xp = (char *) *xpp; |
13968 | | int status = NC_NOERR; |
13969 | | |
13970 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
13971 | | { |
13972 | | int lstatus = ncx_put_uint_uint(xp, tp, fillp); |
13973 | | if (status == NC_NOERR) /* report the first encountered error */ |
13974 | | status = lstatus; |
13975 | | } |
13976 | | |
13977 | | *xpp = (void *)xp; |
13978 | | return status; |
13979 | | #endif |
13980 | | } |
13981 | | |
13982 | | #endif |
13983 | | int |
13984 | | ncx_putn_uint_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
13985 | 0 | { |
13986 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
13987 | | |
13988 | | /* basic algorithm is: |
13989 | | * - ensure sane alignment of output data |
13990 | | * - copy (conversion happens automatically) input data |
13991 | | * to output |
13992 | | * - update tp to point at next unconverted input, and xpp to point |
13993 | | * at next location for converted output |
13994 | | */ |
13995 | | long i, j, ni; |
13996 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
13997 | | uint *xp; |
13998 | | int nrange = 0; /* number of range errors */ |
13999 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14000 | | long cxp = (long) *((char**)xpp); |
14001 | | |
14002 | | realign = (cxp & 7) % SIZEOF_UINT; |
14003 | | /* sjl: manually stripmine so we can limit amount of |
14004 | | * vector work space reserved to LOOPCNT elements. Also |
14005 | | * makes vectorisation easy */ |
14006 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14007 | | ni=Min(nelems-j,LOOPCNT); |
14008 | | if (realign) { |
14009 | | xp = tmp; |
14010 | | } else { |
14011 | | xp = (uint *) *xpp; |
14012 | | } |
14013 | | /* copy the next block */ |
14014 | | #pragma cdir loopcnt=LOOPCNT |
14015 | | #pragma cdir shortloop |
14016 | | for (i=0; i<ni; i++) { |
14017 | | /* the normal case: */ |
14018 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14019 | | /* test for range errors (not always needed but do it anyway) */ |
14020 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14021 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14022 | | nrange += tp[i] > X_UINT_MAX || tp[i] < 0; |
14023 | | } |
14024 | | /* copy workspace back if necessary */ |
14025 | | if (realign) { |
14026 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14027 | | xp = (uint *) *xpp; |
14028 | | } |
14029 | | /* update xpp and tp */ |
14030 | | xp += ni; |
14031 | | tp += ni; |
14032 | | *xpp = (void*)xp; |
14033 | | } |
14034 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14035 | | |
14036 | | #else /* not SX */ |
14037 | |
|
14038 | 0 | char *xp = (char *) *xpp; |
14039 | 0 | int status = NC_NOERR; |
14040 | |
|
14041 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14042 | 0 | { |
14043 | 0 | int lstatus = ncx_put_uint_schar(xp, tp, fillp); |
14044 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14045 | 0 | status = lstatus; |
14046 | 0 | } |
14047 | |
|
14048 | 0 | *xpp = (void *)xp; |
14049 | 0 | return status; |
14050 | 0 | #endif |
14051 | 0 | } |
14052 | | |
14053 | | int |
14054 | | ncx_putn_uint_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
14055 | 0 | { |
14056 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14057 | | |
14058 | | /* basic algorithm is: |
14059 | | * - ensure sane alignment of output data |
14060 | | * - copy (conversion happens automatically) input data |
14061 | | * to output |
14062 | | * - update tp to point at next unconverted input, and xpp to point |
14063 | | * at next location for converted output |
14064 | | */ |
14065 | | long i, j, ni; |
14066 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14067 | | uint *xp; |
14068 | | int nrange = 0; /* number of range errors */ |
14069 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14070 | | long cxp = (long) *((char**)xpp); |
14071 | | |
14072 | | realign = (cxp & 7) % SIZEOF_UINT; |
14073 | | /* sjl: manually stripmine so we can limit amount of |
14074 | | * vector work space reserved to LOOPCNT elements. Also |
14075 | | * makes vectorisation easy */ |
14076 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14077 | | ni=Min(nelems-j,LOOPCNT); |
14078 | | if (realign) { |
14079 | | xp = tmp; |
14080 | | } else { |
14081 | | xp = (uint *) *xpp; |
14082 | | } |
14083 | | /* copy the next block */ |
14084 | | #pragma cdir loopcnt=LOOPCNT |
14085 | | #pragma cdir shortloop |
14086 | | for (i=0; i<ni; i++) { |
14087 | | /* the normal case: */ |
14088 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14089 | | /* test for range errors (not always needed but do it anyway) */ |
14090 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14091 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14092 | | nrange += tp[i] > X_UINT_MAX || tp[i] < 0; |
14093 | | } |
14094 | | /* copy workspace back if necessary */ |
14095 | | if (realign) { |
14096 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14097 | | xp = (uint *) *xpp; |
14098 | | } |
14099 | | /* update xpp and tp */ |
14100 | | xp += ni; |
14101 | | tp += ni; |
14102 | | *xpp = (void*)xp; |
14103 | | } |
14104 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14105 | | |
14106 | | #else /* not SX */ |
14107 | |
|
14108 | 0 | char *xp = (char *) *xpp; |
14109 | 0 | int status = NC_NOERR; |
14110 | |
|
14111 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14112 | 0 | { |
14113 | 0 | int lstatus = ncx_put_uint_short(xp, tp, fillp); |
14114 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14115 | 0 | status = lstatus; |
14116 | 0 | } |
14117 | |
|
14118 | 0 | *xpp = (void *)xp; |
14119 | 0 | return status; |
14120 | 0 | #endif |
14121 | 0 | } |
14122 | | |
14123 | | int |
14124 | | ncx_putn_uint_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
14125 | 0 | { |
14126 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14127 | | |
14128 | | /* basic algorithm is: |
14129 | | * - ensure sane alignment of output data |
14130 | | * - copy (conversion happens automatically) input data |
14131 | | * to output |
14132 | | * - update tp to point at next unconverted input, and xpp to point |
14133 | | * at next location for converted output |
14134 | | */ |
14135 | | long i, j, ni; |
14136 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14137 | | uint *xp; |
14138 | | int nrange = 0; /* number of range errors */ |
14139 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14140 | | long cxp = (long) *((char**)xpp); |
14141 | | |
14142 | | realign = (cxp & 7) % SIZEOF_UINT; |
14143 | | /* sjl: manually stripmine so we can limit amount of |
14144 | | * vector work space reserved to LOOPCNT elements. Also |
14145 | | * makes vectorisation easy */ |
14146 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14147 | | ni=Min(nelems-j,LOOPCNT); |
14148 | | if (realign) { |
14149 | | xp = tmp; |
14150 | | } else { |
14151 | | xp = (uint *) *xpp; |
14152 | | } |
14153 | | /* copy the next block */ |
14154 | | #pragma cdir loopcnt=LOOPCNT |
14155 | | #pragma cdir shortloop |
14156 | | for (i=0; i<ni; i++) { |
14157 | | /* the normal case: */ |
14158 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14159 | | /* test for range errors (not always needed but do it anyway) */ |
14160 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14161 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14162 | | nrange += tp[i] > X_UINT_MAX || tp[i] < 0; |
14163 | | } |
14164 | | /* copy workspace back if necessary */ |
14165 | | if (realign) { |
14166 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14167 | | xp = (uint *) *xpp; |
14168 | | } |
14169 | | /* update xpp and tp */ |
14170 | | xp += ni; |
14171 | | tp += ni; |
14172 | | *xpp = (void*)xp; |
14173 | | } |
14174 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14175 | | |
14176 | | #else /* not SX */ |
14177 | |
|
14178 | 0 | char *xp = (char *) *xpp; |
14179 | 0 | int status = NC_NOERR; |
14180 | |
|
14181 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14182 | 0 | { |
14183 | 0 | int lstatus = ncx_put_uint_int(xp, tp, fillp); |
14184 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14185 | 0 | status = lstatus; |
14186 | 0 | } |
14187 | |
|
14188 | 0 | *xpp = (void *)xp; |
14189 | 0 | return status; |
14190 | 0 | #endif |
14191 | 0 | } |
14192 | | |
14193 | | int |
14194 | | ncx_putn_uint_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
14195 | 0 | { |
14196 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14197 | | |
14198 | | /* basic algorithm is: |
14199 | | * - ensure sane alignment of output data |
14200 | | * - copy (conversion happens automatically) input data |
14201 | | * to output |
14202 | | * - update tp to point at next unconverted input, and xpp to point |
14203 | | * at next location for converted output |
14204 | | */ |
14205 | | long i, j, ni; |
14206 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14207 | | uint *xp; |
14208 | | int nrange = 0; /* number of range errors */ |
14209 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14210 | | long cxp = (long) *((char**)xpp); |
14211 | | |
14212 | | realign = (cxp & 7) % SIZEOF_UINT; |
14213 | | /* sjl: manually stripmine so we can limit amount of |
14214 | | * vector work space reserved to LOOPCNT elements. Also |
14215 | | * makes vectorisation easy */ |
14216 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14217 | | ni=Min(nelems-j,LOOPCNT); |
14218 | | if (realign) { |
14219 | | xp = tmp; |
14220 | | } else { |
14221 | | xp = (uint *) *xpp; |
14222 | | } |
14223 | | /* copy the next block */ |
14224 | | #pragma cdir loopcnt=LOOPCNT |
14225 | | #pragma cdir shortloop |
14226 | | for (i=0; i<ni; i++) { |
14227 | | /* the normal case: */ |
14228 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14229 | | /* test for range errors (not always needed but do it anyway) */ |
14230 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14231 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14232 | | nrange += tp[i] > X_UINT_MAX || tp[i] < 0; |
14233 | | } |
14234 | | /* copy workspace back if necessary */ |
14235 | | if (realign) { |
14236 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14237 | | xp = (uint *) *xpp; |
14238 | | } |
14239 | | /* update xpp and tp */ |
14240 | | xp += ni; |
14241 | | tp += ni; |
14242 | | *xpp = (void*)xp; |
14243 | | } |
14244 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14245 | | |
14246 | | #else /* not SX */ |
14247 | |
|
14248 | 0 | char *xp = (char *) *xpp; |
14249 | 0 | int status = NC_NOERR; |
14250 | |
|
14251 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14252 | 0 | { |
14253 | 0 | int lstatus = ncx_put_uint_long(xp, tp, fillp); |
14254 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14255 | 0 | status = lstatus; |
14256 | 0 | } |
14257 | |
|
14258 | 0 | *xpp = (void *)xp; |
14259 | 0 | return status; |
14260 | 0 | #endif |
14261 | 0 | } |
14262 | | |
14263 | | int |
14264 | | ncx_putn_uint_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
14265 | 0 | { |
14266 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14267 | | |
14268 | | /* basic algorithm is: |
14269 | | * - ensure sane alignment of output data |
14270 | | * - copy (conversion happens automatically) input data |
14271 | | * to output |
14272 | | * - update tp to point at next unconverted input, and xpp to point |
14273 | | * at next location for converted output |
14274 | | */ |
14275 | | long i, j, ni; |
14276 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14277 | | uint *xp; |
14278 | | int nrange = 0; /* number of range errors */ |
14279 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14280 | | long cxp = (long) *((char**)xpp); |
14281 | | |
14282 | | realign = (cxp & 7) % SIZEOF_UINT; |
14283 | | /* sjl: manually stripmine so we can limit amount of |
14284 | | * vector work space reserved to LOOPCNT elements. Also |
14285 | | * makes vectorisation easy */ |
14286 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14287 | | ni=Min(nelems-j,LOOPCNT); |
14288 | | if (realign) { |
14289 | | xp = tmp; |
14290 | | } else { |
14291 | | xp = (uint *) *xpp; |
14292 | | } |
14293 | | /* copy the next block */ |
14294 | | #pragma cdir loopcnt=LOOPCNT |
14295 | | #pragma cdir shortloop |
14296 | | for (i=0; i<ni; i++) { |
14297 | | /* the normal case: */ |
14298 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14299 | | /* test for range errors (not always needed but do it anyway) */ |
14300 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14301 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14302 | | nrange += tp[i] > X_UINT_MAX || tp[i] < 0; |
14303 | | } |
14304 | | /* copy workspace back if necessary */ |
14305 | | if (realign) { |
14306 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14307 | | xp = (uint *) *xpp; |
14308 | | } |
14309 | | /* update xpp and tp */ |
14310 | | xp += ni; |
14311 | | tp += ni; |
14312 | | *xpp = (void*)xp; |
14313 | | } |
14314 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14315 | | |
14316 | | #else /* not SX */ |
14317 | |
|
14318 | 0 | char *xp = (char *) *xpp; |
14319 | 0 | int status = NC_NOERR; |
14320 | |
|
14321 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14322 | 0 | { |
14323 | 0 | int lstatus = ncx_put_uint_float(xp, tp, fillp); |
14324 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14325 | 0 | status = lstatus; |
14326 | 0 | } |
14327 | |
|
14328 | 0 | *xpp = (void *)xp; |
14329 | 0 | return status; |
14330 | 0 | #endif |
14331 | 0 | } |
14332 | | |
14333 | | int |
14334 | | ncx_putn_uint_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
14335 | 0 | { |
14336 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14337 | | |
14338 | | /* basic algorithm is: |
14339 | | * - ensure sane alignment of output data |
14340 | | * - copy (conversion happens automatically) input data |
14341 | | * to output |
14342 | | * - update tp to point at next unconverted input, and xpp to point |
14343 | | * at next location for converted output |
14344 | | */ |
14345 | | long i, j, ni; |
14346 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14347 | | uint *xp; |
14348 | | int nrange = 0; /* number of range errors */ |
14349 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14350 | | long cxp = (long) *((char**)xpp); |
14351 | | |
14352 | | realign = (cxp & 7) % SIZEOF_UINT; |
14353 | | /* sjl: manually stripmine so we can limit amount of |
14354 | | * vector work space reserved to LOOPCNT elements. Also |
14355 | | * makes vectorisation easy */ |
14356 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14357 | | ni=Min(nelems-j,LOOPCNT); |
14358 | | if (realign) { |
14359 | | xp = tmp; |
14360 | | } else { |
14361 | | xp = (uint *) *xpp; |
14362 | | } |
14363 | | /* copy the next block */ |
14364 | | #pragma cdir loopcnt=LOOPCNT |
14365 | | #pragma cdir shortloop |
14366 | | for (i=0; i<ni; i++) { |
14367 | | /* the normal case: */ |
14368 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14369 | | /* test for range errors (not always needed but do it anyway) */ |
14370 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14371 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14372 | | nrange += tp[i] > X_UINT_MAX || tp[i] < 0; |
14373 | | } |
14374 | | /* copy workspace back if necessary */ |
14375 | | if (realign) { |
14376 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14377 | | xp = (uint *) *xpp; |
14378 | | } |
14379 | | /* update xpp and tp */ |
14380 | | xp += ni; |
14381 | | tp += ni; |
14382 | | *xpp = (void*)xp; |
14383 | | } |
14384 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14385 | | |
14386 | | #else /* not SX */ |
14387 | |
|
14388 | 0 | char *xp = (char *) *xpp; |
14389 | 0 | int status = NC_NOERR; |
14390 | |
|
14391 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14392 | 0 | { |
14393 | 0 | int lstatus = ncx_put_uint_double(xp, tp, fillp); |
14394 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14395 | 0 | status = lstatus; |
14396 | 0 | } |
14397 | |
|
14398 | 0 | *xpp = (void *)xp; |
14399 | 0 | return status; |
14400 | 0 | #endif |
14401 | 0 | } |
14402 | | |
14403 | | int |
14404 | | ncx_putn_uint_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
14405 | 0 | { |
14406 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14407 | | |
14408 | | /* basic algorithm is: |
14409 | | * - ensure sane alignment of output data |
14410 | | * - copy (conversion happens automatically) input data |
14411 | | * to output |
14412 | | * - update tp to point at next unconverted input, and xpp to point |
14413 | | * at next location for converted output |
14414 | | */ |
14415 | | long i, j, ni; |
14416 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14417 | | uint *xp; |
14418 | | int nrange = 0; /* number of range errors */ |
14419 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14420 | | long cxp = (long) *((char**)xpp); |
14421 | | |
14422 | | realign = (cxp & 7) % SIZEOF_UINT; |
14423 | | /* sjl: manually stripmine so we can limit amount of |
14424 | | * vector work space reserved to LOOPCNT elements. Also |
14425 | | * makes vectorisation easy */ |
14426 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14427 | | ni=Min(nelems-j,LOOPCNT); |
14428 | | if (realign) { |
14429 | | xp = tmp; |
14430 | | } else { |
14431 | | xp = (uint *) *xpp; |
14432 | | } |
14433 | | /* copy the next block */ |
14434 | | #pragma cdir loopcnt=LOOPCNT |
14435 | | #pragma cdir shortloop |
14436 | | for (i=0; i<ni; i++) { |
14437 | | /* the normal case: */ |
14438 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14439 | | /* test for range errors (not always needed but do it anyway) */ |
14440 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14441 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14442 | | nrange += tp[i] > X_UINT_MAX || tp[i] < 0; |
14443 | | } |
14444 | | /* copy workspace back if necessary */ |
14445 | | if (realign) { |
14446 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14447 | | xp = (uint *) *xpp; |
14448 | | } |
14449 | | /* update xpp and tp */ |
14450 | | xp += ni; |
14451 | | tp += ni; |
14452 | | *xpp = (void*)xp; |
14453 | | } |
14454 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14455 | | |
14456 | | #else /* not SX */ |
14457 | |
|
14458 | 0 | char *xp = (char *) *xpp; |
14459 | 0 | int status = NC_NOERR; |
14460 | |
|
14461 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14462 | 0 | { |
14463 | 0 | int lstatus = ncx_put_uint_longlong(xp, tp, fillp); |
14464 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14465 | 0 | status = lstatus; |
14466 | 0 | } |
14467 | |
|
14468 | 0 | *xpp = (void *)xp; |
14469 | 0 | return status; |
14470 | 0 | #endif |
14471 | 0 | } |
14472 | | |
14473 | | int |
14474 | | ncx_putn_uint_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
14475 | 0 | { |
14476 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14477 | | |
14478 | | /* basic algorithm is: |
14479 | | * - ensure sane alignment of output data |
14480 | | * - copy (conversion happens automatically) input data |
14481 | | * to output |
14482 | | * - update tp to point at next unconverted input, and xpp to point |
14483 | | * at next location for converted output |
14484 | | */ |
14485 | | long i, j, ni; |
14486 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14487 | | uint *xp; |
14488 | | int nrange = 0; /* number of range errors */ |
14489 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14490 | | long cxp = (long) *((char**)xpp); |
14491 | | |
14492 | | realign = (cxp & 7) % SIZEOF_UINT; |
14493 | | /* sjl: manually stripmine so we can limit amount of |
14494 | | * vector work space reserved to LOOPCNT elements. Also |
14495 | | * makes vectorisation easy */ |
14496 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14497 | | ni=Min(nelems-j,LOOPCNT); |
14498 | | if (realign) { |
14499 | | xp = tmp; |
14500 | | } else { |
14501 | | xp = (uint *) *xpp; |
14502 | | } |
14503 | | /* copy the next block */ |
14504 | | #pragma cdir loopcnt=LOOPCNT |
14505 | | #pragma cdir shortloop |
14506 | | for (i=0; i<ni; i++) { |
14507 | | /* the normal case: */ |
14508 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14509 | | /* test for range errors (not always needed but do it anyway) */ |
14510 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14511 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14512 | | nrange += tp[i] > X_UINT_MAX ; |
14513 | | } |
14514 | | /* copy workspace back if necessary */ |
14515 | | if (realign) { |
14516 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14517 | | xp = (uint *) *xpp; |
14518 | | } |
14519 | | /* update xpp and tp */ |
14520 | | xp += ni; |
14521 | | tp += ni; |
14522 | | *xpp = (void*)xp; |
14523 | | } |
14524 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14525 | | |
14526 | | #else /* not SX */ |
14527 | |
|
14528 | 0 | char *xp = (char *) *xpp; |
14529 | 0 | int status = NC_NOERR; |
14530 | |
|
14531 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14532 | 0 | { |
14533 | 0 | int lstatus = ncx_put_uint_uchar(xp, tp, fillp); |
14534 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14535 | 0 | status = lstatus; |
14536 | 0 | } |
14537 | |
|
14538 | 0 | *xpp = (void *)xp; |
14539 | 0 | return status; |
14540 | 0 | #endif |
14541 | 0 | } |
14542 | | |
14543 | | int |
14544 | | ncx_putn_uint_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
14545 | 0 | { |
14546 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14547 | | |
14548 | | /* basic algorithm is: |
14549 | | * - ensure sane alignment of output data |
14550 | | * - copy (conversion happens automatically) input data |
14551 | | * to output |
14552 | | * - update tp to point at next unconverted input, and xpp to point |
14553 | | * at next location for converted output |
14554 | | */ |
14555 | | long i, j, ni; |
14556 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14557 | | uint *xp; |
14558 | | int nrange = 0; /* number of range errors */ |
14559 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14560 | | long cxp = (long) *((char**)xpp); |
14561 | | |
14562 | | realign = (cxp & 7) % SIZEOF_UINT; |
14563 | | /* sjl: manually stripmine so we can limit amount of |
14564 | | * vector work space reserved to LOOPCNT elements. Also |
14565 | | * makes vectorisation easy */ |
14566 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14567 | | ni=Min(nelems-j,LOOPCNT); |
14568 | | if (realign) { |
14569 | | xp = tmp; |
14570 | | } else { |
14571 | | xp = (uint *) *xpp; |
14572 | | } |
14573 | | /* copy the next block */ |
14574 | | #pragma cdir loopcnt=LOOPCNT |
14575 | | #pragma cdir shortloop |
14576 | | for (i=0; i<ni; i++) { |
14577 | | /* the normal case: */ |
14578 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14579 | | /* test for range errors (not always needed but do it anyway) */ |
14580 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14581 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14582 | | nrange += tp[i] > X_UINT_MAX ; |
14583 | | } |
14584 | | /* copy workspace back if necessary */ |
14585 | | if (realign) { |
14586 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14587 | | xp = (uint *) *xpp; |
14588 | | } |
14589 | | /* update xpp and tp */ |
14590 | | xp += ni; |
14591 | | tp += ni; |
14592 | | *xpp = (void*)xp; |
14593 | | } |
14594 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14595 | | |
14596 | | #else /* not SX */ |
14597 | |
|
14598 | 0 | char *xp = (char *) *xpp; |
14599 | 0 | int status = NC_NOERR; |
14600 | |
|
14601 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14602 | 0 | { |
14603 | 0 | int lstatus = ncx_put_uint_ushort(xp, tp, fillp); |
14604 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14605 | 0 | status = lstatus; |
14606 | 0 | } |
14607 | |
|
14608 | 0 | *xpp = (void *)xp; |
14609 | 0 | return status; |
14610 | 0 | #endif |
14611 | 0 | } |
14612 | | |
14613 | | int |
14614 | | ncx_putn_uint_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
14615 | 0 | { |
14616 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT == SIZEOF_UINT |
14617 | | |
14618 | | /* basic algorithm is: |
14619 | | * - ensure sane alignment of output data |
14620 | | * - copy (conversion happens automatically) input data |
14621 | | * to output |
14622 | | * - update tp to point at next unconverted input, and xpp to point |
14623 | | * at next location for converted output |
14624 | | */ |
14625 | | long i, j, ni; |
14626 | | uint tmp[LOOPCNT]; /* in case input is misaligned */ |
14627 | | uint *xp; |
14628 | | int nrange = 0; /* number of range errors */ |
14629 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14630 | | long cxp = (long) *((char**)xpp); |
14631 | | |
14632 | | realign = (cxp & 7) % SIZEOF_UINT; |
14633 | | /* sjl: manually stripmine so we can limit amount of |
14634 | | * vector work space reserved to LOOPCNT elements. Also |
14635 | | * makes vectorisation easy */ |
14636 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14637 | | ni=Min(nelems-j,LOOPCNT); |
14638 | | if (realign) { |
14639 | | xp = tmp; |
14640 | | } else { |
14641 | | xp = (uint *) *xpp; |
14642 | | } |
14643 | | /* copy the next block */ |
14644 | | #pragma cdir loopcnt=LOOPCNT |
14645 | | #pragma cdir shortloop |
14646 | | for (i=0; i<ni; i++) { |
14647 | | /* the normal case: */ |
14648 | | xp[i] = (uint) Max( X_UINT_MIN, Min(X_UINT_MAX, (uint) tp[i])); |
14649 | | /* test for range errors (not always needed but do it anyway) */ |
14650 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
14651 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
14652 | | nrange += tp[i] > X_UINT_MAX ; |
14653 | | } |
14654 | | /* copy workspace back if necessary */ |
14655 | | if (realign) { |
14656 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT); |
14657 | | xp = (uint *) *xpp; |
14658 | | } |
14659 | | /* update xpp and tp */ |
14660 | | xp += ni; |
14661 | | tp += ni; |
14662 | | *xpp = (void*)xp; |
14663 | | } |
14664 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14665 | | |
14666 | | #else /* not SX */ |
14667 | |
|
14668 | 0 | char *xp = (char *) *xpp; |
14669 | 0 | int status = NC_NOERR; |
14670 | |
|
14671 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT, tp++) |
14672 | 0 | { |
14673 | 0 | int lstatus = ncx_put_uint_ulonglong(xp, tp, fillp); |
14674 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14675 | 0 | status = lstatus; |
14676 | 0 | } |
14677 | |
|
14678 | 0 | *xpp = (void *)xp; |
14679 | 0 | return status; |
14680 | 0 | #endif |
14681 | 0 | } |
14682 | | |
14683 | | |
14684 | | |
14685 | | /* float ---------------------------------------------------------------------*/ |
14686 | | |
14687 | | #if X_SIZEOF_FLOAT == SIZEOF_FLOAT && !defined(NO_IEEE_FLOAT) |
14688 | | /* optimized version */ |
14689 | | int |
14690 | | ncx_getn_float_float(const void **xpp, size_t nelems, float *tp) |
14691 | 0 | { |
14692 | | #ifdef WORDS_BIGENDIAN |
14693 | | (void) memcpy(tp, *xpp, (size_t)nelems * SIZEOF_FLOAT); |
14694 | | # else |
14695 | 0 | swapn4b(tp, *xpp, nelems); |
14696 | 0 | # endif |
14697 | 0 | *xpp = (const void *)((const char *)(*xpp) + nelems * X_SIZEOF_FLOAT); |
14698 | 0 | return NC_NOERR; |
14699 | 0 | } |
14700 | | #elif defined(vax) && vax != 0 |
14701 | | int |
14702 | | ncx_getn_float_float(const void **xpp, size_t nfloats, float *ip) |
14703 | | { |
14704 | | float *const end = ip + nfloats; |
14705 | | |
14706 | | while (ip < end) |
14707 | | { |
14708 | | struct vax_single *const vsp = (struct vax_single *) ip; |
14709 | | const struct ieee_single *const isp = |
14710 | | (const struct ieee_single *) (*xpp); |
14711 | | unsigned exp = isp->exp_hi << 1 | isp->exp_lo; |
14712 | | |
14713 | | switch(exp) { |
14714 | | case 0 : |
14715 | | /* ieee subnormal */ |
14716 | | if (isp->mant_hi == min.ieee.mant_hi |
14717 | | && isp->mant_lo_hi == min.ieee.mant_lo_hi |
14718 | | && isp->mant_lo_lo == min.ieee.mant_lo_lo) |
14719 | | { |
14720 | | *vsp = min.s; |
14721 | | } |
14722 | | else |
14723 | | { |
14724 | | unsigned mantissa = (isp->mant_hi << 16) |
14725 | | | isp->mant_lo_hi << 8 |
14726 | | | isp->mant_lo_lo; |
14727 | | unsigned tmp = mantissa >> 20; |
14728 | | if (tmp >= 4) { |
14729 | | vsp->exp = 2; |
14730 | | } else if (tmp >= 2) { |
14731 | | vsp->exp = 1; |
14732 | | } else { |
14733 | | *vsp = min.s; |
14734 | | break; |
14735 | | } /* else */ |
14736 | | tmp = mantissa - (1 << (20 + vsp->exp )); |
14737 | | tmp <<= 3 - vsp->exp; |
14738 | | vsp->mantissa2 = tmp; |
14739 | | vsp->mantissa1 = (tmp >> 16); |
14740 | | } |
14741 | | break; |
14742 | | case 0xfe : |
14743 | | case 0xff : |
14744 | | *vsp = max.s; |
14745 | | break; |
14746 | | default : |
14747 | | vsp->exp = exp - IEEE_SNG_BIAS + VAX_SNG_BIAS; |
14748 | | vsp->mantissa2 = isp->mant_lo_hi << 8 | isp->mant_lo_lo; |
14749 | | vsp->mantissa1 = isp->mant_hi; |
14750 | | } |
14751 | | |
14752 | | vsp->sign = isp->sign; |
14753 | | |
14754 | | |
14755 | | ip++; |
14756 | | *xpp = (char *)(*xpp) + X_SIZEOF_FLOAT; |
14757 | | } |
14758 | | return NC_NOERR; |
14759 | | } |
14760 | | #else |
14761 | | int |
14762 | | ncx_getn_float_float(const void **xpp, size_t nelems, float *tp) |
14763 | | { |
14764 | | const char *xp = *xpp; |
14765 | | int status = NC_NOERR; |
14766 | | |
14767 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
14768 | | { |
14769 | | const int lstatus = ncx_get_float_float(xp, tp, fillp); |
14770 | | if (status == NC_NOERR) /* report the first encountered error */ |
14771 | | status = lstatus; |
14772 | | } |
14773 | | |
14774 | | *xpp = (const void *)xp; |
14775 | | return status; |
14776 | | } |
14777 | | |
14778 | | #endif |
14779 | | int |
14780 | | ncx_getn_float_schar(const void **xpp, size_t nelems, schar *tp) |
14781 | 0 | { |
14782 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
14783 | | |
14784 | | /* basic algorithm is: |
14785 | | * - ensure sane alignment of input data |
14786 | | * - copy (conversion happens automatically) input data |
14787 | | * to output |
14788 | | * - update xpp to point at next unconverted input, and tp to point |
14789 | | * at next location for converted output |
14790 | | */ |
14791 | | long i, j, ni; |
14792 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
14793 | | float *xp; |
14794 | | int nrange = 0; /* number of range errors */ |
14795 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14796 | | long cxp = (long) *((char**)xpp); |
14797 | | |
14798 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
14799 | | /* sjl: manually stripmine so we can limit amount of |
14800 | | * vector work space reserved to LOOPCNT elements. Also |
14801 | | * makes vectorisation easy */ |
14802 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14803 | | ni=Min(nelems-j,LOOPCNT); |
14804 | | if (realign) { |
14805 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
14806 | | xp = tmp; |
14807 | | } else { |
14808 | | xp = (float *) *xpp; |
14809 | | } |
14810 | | /* copy the next block */ |
14811 | | #pragma cdir loopcnt=LOOPCNT |
14812 | | #pragma cdir shortloop |
14813 | | for (i=0; i<ni; i++) { |
14814 | | tp[i] = (schar) Max( SCHAR_MIN, Min(SCHAR_MAX, (schar) xp[i])); |
14815 | | /* test for range errors (not always needed but do it anyway) */ |
14816 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
14817 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
14818 | | nrange += xp[i] > SCHAR_MAX || xp[i] < SCHAR_MIN; |
14819 | | } |
14820 | | /* update xpp and tp */ |
14821 | | if (realign) xp = (float *) *xpp; |
14822 | | xp += ni; |
14823 | | tp += ni; |
14824 | | *xpp = (void*)xp; |
14825 | | } |
14826 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14827 | | |
14828 | | #else /* not SX */ |
14829 | 0 | const char *xp = (const char *) *xpp; |
14830 | 0 | int status = NC_NOERR; |
14831 | |
|
14832 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
14833 | 0 | { |
14834 | 0 | const int lstatus = ncx_get_float_schar(xp, tp); |
14835 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14836 | 0 | status = lstatus; |
14837 | 0 | } |
14838 | |
|
14839 | 0 | *xpp = (const void *)xp; |
14840 | 0 | return status; |
14841 | 0 | #endif |
14842 | 0 | } |
14843 | | |
14844 | | int |
14845 | | ncx_getn_float_short(const void **xpp, size_t nelems, short *tp) |
14846 | 0 | { |
14847 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
14848 | | |
14849 | | /* basic algorithm is: |
14850 | | * - ensure sane alignment of input data |
14851 | | * - copy (conversion happens automatically) input data |
14852 | | * to output |
14853 | | * - update xpp to point at next unconverted input, and tp to point |
14854 | | * at next location for converted output |
14855 | | */ |
14856 | | long i, j, ni; |
14857 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
14858 | | float *xp; |
14859 | | int nrange = 0; /* number of range errors */ |
14860 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14861 | | long cxp = (long) *((char**)xpp); |
14862 | | |
14863 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
14864 | | /* sjl: manually stripmine so we can limit amount of |
14865 | | * vector work space reserved to LOOPCNT elements. Also |
14866 | | * makes vectorisation easy */ |
14867 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14868 | | ni=Min(nelems-j,LOOPCNT); |
14869 | | if (realign) { |
14870 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
14871 | | xp = tmp; |
14872 | | } else { |
14873 | | xp = (float *) *xpp; |
14874 | | } |
14875 | | /* copy the next block */ |
14876 | | #pragma cdir loopcnt=LOOPCNT |
14877 | | #pragma cdir shortloop |
14878 | | for (i=0; i<ni; i++) { |
14879 | | tp[i] = (short) Max( SHORT_MIN, Min(SHORT_MAX, (short) xp[i])); |
14880 | | /* test for range errors (not always needed but do it anyway) */ |
14881 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
14882 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
14883 | | nrange += xp[i] > SHORT_MAX || xp[i] < SHORT_MIN; |
14884 | | } |
14885 | | /* update xpp and tp */ |
14886 | | if (realign) xp = (float *) *xpp; |
14887 | | xp += ni; |
14888 | | tp += ni; |
14889 | | *xpp = (void*)xp; |
14890 | | } |
14891 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14892 | | |
14893 | | #else /* not SX */ |
14894 | 0 | const char *xp = (const char *) *xpp; |
14895 | 0 | int status = NC_NOERR; |
14896 | |
|
14897 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
14898 | 0 | { |
14899 | 0 | const int lstatus = ncx_get_float_short(xp, tp); |
14900 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14901 | 0 | status = lstatus; |
14902 | 0 | } |
14903 | |
|
14904 | 0 | *xpp = (const void *)xp; |
14905 | 0 | return status; |
14906 | 0 | #endif |
14907 | 0 | } |
14908 | | |
14909 | | int |
14910 | | ncx_getn_float_int(const void **xpp, size_t nelems, int *tp) |
14911 | 0 | { |
14912 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
14913 | | |
14914 | | /* basic algorithm is: |
14915 | | * - ensure sane alignment of input data |
14916 | | * - copy (conversion happens automatically) input data |
14917 | | * to output |
14918 | | * - update xpp to point at next unconverted input, and tp to point |
14919 | | * at next location for converted output |
14920 | | */ |
14921 | | long i, j, ni; |
14922 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
14923 | | float *xp; |
14924 | | int nrange = 0; /* number of range errors */ |
14925 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14926 | | long cxp = (long) *((char**)xpp); |
14927 | | |
14928 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
14929 | | /* sjl: manually stripmine so we can limit amount of |
14930 | | * vector work space reserved to LOOPCNT elements. Also |
14931 | | * makes vectorisation easy */ |
14932 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14933 | | ni=Min(nelems-j,LOOPCNT); |
14934 | | if (realign) { |
14935 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
14936 | | xp = tmp; |
14937 | | } else { |
14938 | | xp = (float *) *xpp; |
14939 | | } |
14940 | | /* copy the next block */ |
14941 | | #pragma cdir loopcnt=LOOPCNT |
14942 | | #pragma cdir shortloop |
14943 | | for (i=0; i<ni; i++) { |
14944 | | tp[i] = (int) Max( INT_MIN, Min(INT_MAX, (int) xp[i])); |
14945 | | /* test for range errors (not always needed but do it anyway) */ |
14946 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
14947 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
14948 | | nrange += xp[i] > INT_MAX || xp[i] < INT_MIN; |
14949 | | } |
14950 | | /* update xpp and tp */ |
14951 | | if (realign) xp = (float *) *xpp; |
14952 | | xp += ni; |
14953 | | tp += ni; |
14954 | | *xpp = (void*)xp; |
14955 | | } |
14956 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
14957 | | |
14958 | | #else /* not SX */ |
14959 | 0 | const char *xp = (const char *) *xpp; |
14960 | 0 | int status = NC_NOERR; |
14961 | |
|
14962 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
14963 | 0 | { |
14964 | 0 | const int lstatus = ncx_get_float_int(xp, tp); |
14965 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
14966 | 0 | status = lstatus; |
14967 | 0 | } |
14968 | |
|
14969 | 0 | *xpp = (const void *)xp; |
14970 | 0 | return status; |
14971 | 0 | #endif |
14972 | 0 | } |
14973 | | |
14974 | | int |
14975 | | ncx_getn_float_long(const void **xpp, size_t nelems, long *tp) |
14976 | 0 | { |
14977 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
14978 | | |
14979 | | /* basic algorithm is: |
14980 | | * - ensure sane alignment of input data |
14981 | | * - copy (conversion happens automatically) input data |
14982 | | * to output |
14983 | | * - update xpp to point at next unconverted input, and tp to point |
14984 | | * at next location for converted output |
14985 | | */ |
14986 | | long i, j, ni; |
14987 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
14988 | | float *xp; |
14989 | | int nrange = 0; /* number of range errors */ |
14990 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
14991 | | long cxp = (long) *((char**)xpp); |
14992 | | |
14993 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
14994 | | /* sjl: manually stripmine so we can limit amount of |
14995 | | * vector work space reserved to LOOPCNT elements. Also |
14996 | | * makes vectorisation easy */ |
14997 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
14998 | | ni=Min(nelems-j,LOOPCNT); |
14999 | | if (realign) { |
15000 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
15001 | | xp = tmp; |
15002 | | } else { |
15003 | | xp = (float *) *xpp; |
15004 | | } |
15005 | | /* copy the next block */ |
15006 | | #pragma cdir loopcnt=LOOPCNT |
15007 | | #pragma cdir shortloop |
15008 | | for (i=0; i<ni; i++) { |
15009 | | tp[i] = (long) Max( LONG_MIN, Min(LONG_MAX, (long) xp[i])); |
15010 | | /* test for range errors (not always needed but do it anyway) */ |
15011 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
15012 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
15013 | | nrange += xp[i] > LONG_MAX || xp[i] < LONG_MIN; |
15014 | | } |
15015 | | /* update xpp and tp */ |
15016 | | if (realign) xp = (float *) *xpp; |
15017 | | xp += ni; |
15018 | | tp += ni; |
15019 | | *xpp = (void*)xp; |
15020 | | } |
15021 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15022 | | |
15023 | | #else /* not SX */ |
15024 | 0 | const char *xp = (const char *) *xpp; |
15025 | 0 | int status = NC_NOERR; |
15026 | |
|
15027 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15028 | 0 | { |
15029 | 0 | const int lstatus = ncx_get_float_long(xp, tp); |
15030 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15031 | 0 | status = lstatus; |
15032 | 0 | } |
15033 | |
|
15034 | 0 | *xpp = (const void *)xp; |
15035 | 0 | return status; |
15036 | 0 | #endif |
15037 | 0 | } |
15038 | | |
15039 | | int |
15040 | | ncx_getn_float_double(const void **xpp, size_t nelems, double *tp) |
15041 | 0 | { |
15042 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15043 | | |
15044 | | /* basic algorithm is: |
15045 | | * - ensure sane alignment of input data |
15046 | | * - copy (conversion happens automatically) input data |
15047 | | * to output |
15048 | | * - update xpp to point at next unconverted input, and tp to point |
15049 | | * at next location for converted output |
15050 | | */ |
15051 | | long i, j, ni; |
15052 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15053 | | float *xp; |
15054 | | int nrange = 0; /* number of range errors */ |
15055 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15056 | | long cxp = (long) *((char**)xpp); |
15057 | | |
15058 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15059 | | /* sjl: manually stripmine so we can limit amount of |
15060 | | * vector work space reserved to LOOPCNT elements. Also |
15061 | | * makes vectorisation easy */ |
15062 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15063 | | ni=Min(nelems-j,LOOPCNT); |
15064 | | if (realign) { |
15065 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
15066 | | xp = tmp; |
15067 | | } else { |
15068 | | xp = (float *) *xpp; |
15069 | | } |
15070 | | /* copy the next block */ |
15071 | | #pragma cdir loopcnt=LOOPCNT |
15072 | | #pragma cdir shortloop |
15073 | | for (i=0; i<ni; i++) { |
15074 | | tp[i] = (double) Max( DOUBLE_MIN, Min(DOUBLE_MAX, (double) xp[i])); |
15075 | | /* test for range errors (not always needed but do it anyway) */ |
15076 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
15077 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
15078 | | nrange += xp[i] > DOUBLE_MAX || xp[i] < DOUBLE_MIN; |
15079 | | } |
15080 | | /* update xpp and tp */ |
15081 | | if (realign) xp = (float *) *xpp; |
15082 | | xp += ni; |
15083 | | tp += ni; |
15084 | | *xpp = (void*)xp; |
15085 | | } |
15086 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15087 | | |
15088 | | #else /* not SX */ |
15089 | 0 | const char *xp = (const char *) *xpp; |
15090 | 0 | int status = NC_NOERR; |
15091 | |
|
15092 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15093 | 0 | { |
15094 | 0 | const int lstatus = ncx_get_float_double(xp, tp); |
15095 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15096 | 0 | status = lstatus; |
15097 | 0 | } |
15098 | |
|
15099 | 0 | *xpp = (const void *)xp; |
15100 | 0 | return status; |
15101 | 0 | #endif |
15102 | 0 | } |
15103 | | |
15104 | | int |
15105 | | ncx_getn_float_longlong(const void **xpp, size_t nelems, longlong *tp) |
15106 | 0 | { |
15107 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15108 | | |
15109 | | /* basic algorithm is: |
15110 | | * - ensure sane alignment of input data |
15111 | | * - copy (conversion happens automatically) input data |
15112 | | * to output |
15113 | | * - update xpp to point at next unconverted input, and tp to point |
15114 | | * at next location for converted output |
15115 | | */ |
15116 | | long i, j, ni; |
15117 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15118 | | float *xp; |
15119 | | int nrange = 0; /* number of range errors */ |
15120 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15121 | | long cxp = (long) *((char**)xpp); |
15122 | | |
15123 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15124 | | /* sjl: manually stripmine so we can limit amount of |
15125 | | * vector work space reserved to LOOPCNT elements. Also |
15126 | | * makes vectorisation easy */ |
15127 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15128 | | ni=Min(nelems-j,LOOPCNT); |
15129 | | if (realign) { |
15130 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
15131 | | xp = tmp; |
15132 | | } else { |
15133 | | xp = (float *) *xpp; |
15134 | | } |
15135 | | /* copy the next block */ |
15136 | | #pragma cdir loopcnt=LOOPCNT |
15137 | | #pragma cdir shortloop |
15138 | | for (i=0; i<ni; i++) { |
15139 | | tp[i] = (longlong) Max( LONGLONG_MIN, Min(LONGLONG_MAX, (longlong) xp[i])); |
15140 | | /* test for range errors (not always needed but do it anyway) */ |
15141 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
15142 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
15143 | | nrange += xp[i] > LONGLONG_MAX || xp[i] < LONGLONG_MIN; |
15144 | | } |
15145 | | /* update xpp and tp */ |
15146 | | if (realign) xp = (float *) *xpp; |
15147 | | xp += ni; |
15148 | | tp += ni; |
15149 | | *xpp = (void*)xp; |
15150 | | } |
15151 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15152 | | |
15153 | | #else /* not SX */ |
15154 | 0 | const char *xp = (const char *) *xpp; |
15155 | 0 | int status = NC_NOERR; |
15156 | |
|
15157 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15158 | 0 | { |
15159 | 0 | const int lstatus = ncx_get_float_longlong(xp, tp); |
15160 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15161 | 0 | status = lstatus; |
15162 | 0 | } |
15163 | |
|
15164 | 0 | *xpp = (const void *)xp; |
15165 | 0 | return status; |
15166 | 0 | #endif |
15167 | 0 | } |
15168 | | |
15169 | | int |
15170 | | ncx_getn_float_ushort(const void **xpp, size_t nelems, ushort *tp) |
15171 | 0 | { |
15172 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15173 | | |
15174 | | /* basic algorithm is: |
15175 | | * - ensure sane alignment of input data |
15176 | | * - copy (conversion happens automatically) input data |
15177 | | * to output |
15178 | | * - update xpp to point at next unconverted input, and tp to point |
15179 | | * at next location for converted output |
15180 | | */ |
15181 | | long i, j, ni; |
15182 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15183 | | float *xp; |
15184 | | int nrange = 0; /* number of range errors */ |
15185 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15186 | | long cxp = (long) *((char**)xpp); |
15187 | | |
15188 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15189 | | /* sjl: manually stripmine so we can limit amount of |
15190 | | * vector work space reserved to LOOPCNT elements. Also |
15191 | | * makes vectorisation easy */ |
15192 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15193 | | ni=Min(nelems-j,LOOPCNT); |
15194 | | if (realign) { |
15195 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
15196 | | xp = tmp; |
15197 | | } else { |
15198 | | xp = (float *) *xpp; |
15199 | | } |
15200 | | /* copy the next block */ |
15201 | | #pragma cdir loopcnt=LOOPCNT |
15202 | | #pragma cdir shortloop |
15203 | | for (i=0; i<ni; i++) { |
15204 | | tp[i] = (ushort) Max( USHORT_MIN, Min(USHORT_MAX, (ushort) xp[i])); |
15205 | | /* test for range errors (not always needed but do it anyway) */ |
15206 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
15207 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
15208 | | nrange += xp[i] > USHORT_MAX || xp[i] < 0; |
15209 | | } |
15210 | | /* update xpp and tp */ |
15211 | | if (realign) xp = (float *) *xpp; |
15212 | | xp += ni; |
15213 | | tp += ni; |
15214 | | *xpp = (void*)xp; |
15215 | | } |
15216 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15217 | | |
15218 | | #else /* not SX */ |
15219 | 0 | const char *xp = (const char *) *xpp; |
15220 | 0 | int status = NC_NOERR; |
15221 | |
|
15222 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15223 | 0 | { |
15224 | 0 | const int lstatus = ncx_get_float_ushort(xp, tp); |
15225 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15226 | 0 | status = lstatus; |
15227 | 0 | } |
15228 | |
|
15229 | 0 | *xpp = (const void *)xp; |
15230 | 0 | return status; |
15231 | 0 | #endif |
15232 | 0 | } |
15233 | | |
15234 | | int |
15235 | | ncx_getn_float_uchar(const void **xpp, size_t nelems, uchar *tp) |
15236 | 0 | { |
15237 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15238 | | |
15239 | | /* basic algorithm is: |
15240 | | * - ensure sane alignment of input data |
15241 | | * - copy (conversion happens automatically) input data |
15242 | | * to output |
15243 | | * - update xpp to point at next unconverted input, and tp to point |
15244 | | * at next location for converted output |
15245 | | */ |
15246 | | long i, j, ni; |
15247 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15248 | | float *xp; |
15249 | | int nrange = 0; /* number of range errors */ |
15250 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15251 | | long cxp = (long) *((char**)xpp); |
15252 | | |
15253 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15254 | | /* sjl: manually stripmine so we can limit amount of |
15255 | | * vector work space reserved to LOOPCNT elements. Also |
15256 | | * makes vectorisation easy */ |
15257 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15258 | | ni=Min(nelems-j,LOOPCNT); |
15259 | | if (realign) { |
15260 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
15261 | | xp = tmp; |
15262 | | } else { |
15263 | | xp = (float *) *xpp; |
15264 | | } |
15265 | | /* copy the next block */ |
15266 | | #pragma cdir loopcnt=LOOPCNT |
15267 | | #pragma cdir shortloop |
15268 | | for (i=0; i<ni; i++) { |
15269 | | tp[i] = (uchar) Max( UCHAR_MIN, Min(UCHAR_MAX, (uchar) xp[i])); |
15270 | | /* test for range errors (not always needed but do it anyway) */ |
15271 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
15272 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
15273 | | nrange += xp[i] > UCHAR_MAX || xp[i] < 0; |
15274 | | } |
15275 | | /* update xpp and tp */ |
15276 | | if (realign) xp = (float *) *xpp; |
15277 | | xp += ni; |
15278 | | tp += ni; |
15279 | | *xpp = (void*)xp; |
15280 | | } |
15281 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15282 | | |
15283 | | #else /* not SX */ |
15284 | 0 | const char *xp = (const char *) *xpp; |
15285 | 0 | int status = NC_NOERR; |
15286 | |
|
15287 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15288 | 0 | { |
15289 | 0 | const int lstatus = ncx_get_float_uchar(xp, tp); |
15290 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15291 | 0 | status = lstatus; |
15292 | 0 | } |
15293 | |
|
15294 | 0 | *xpp = (const void *)xp; |
15295 | 0 | return status; |
15296 | 0 | #endif |
15297 | 0 | } |
15298 | | |
15299 | | int |
15300 | | ncx_getn_float_uint(const void **xpp, size_t nelems, uint *tp) |
15301 | 0 | { |
15302 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15303 | | |
15304 | | /* basic algorithm is: |
15305 | | * - ensure sane alignment of input data |
15306 | | * - copy (conversion happens automatically) input data |
15307 | | * to output |
15308 | | * - update xpp to point at next unconverted input, and tp to point |
15309 | | * at next location for converted output |
15310 | | */ |
15311 | | long i, j, ni; |
15312 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15313 | | float *xp; |
15314 | | int nrange = 0; /* number of range errors */ |
15315 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15316 | | long cxp = (long) *((char**)xpp); |
15317 | | |
15318 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15319 | | /* sjl: manually stripmine so we can limit amount of |
15320 | | * vector work space reserved to LOOPCNT elements. Also |
15321 | | * makes vectorisation easy */ |
15322 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15323 | | ni=Min(nelems-j,LOOPCNT); |
15324 | | if (realign) { |
15325 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
15326 | | xp = tmp; |
15327 | | } else { |
15328 | | xp = (float *) *xpp; |
15329 | | } |
15330 | | /* copy the next block */ |
15331 | | #pragma cdir loopcnt=LOOPCNT |
15332 | | #pragma cdir shortloop |
15333 | | for (i=0; i<ni; i++) { |
15334 | | tp[i] = (uint) Max( UINT_MIN, Min(UINT_MAX, (uint) xp[i])); |
15335 | | /* test for range errors (not always needed but do it anyway) */ |
15336 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
15337 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
15338 | | nrange += xp[i] > UINT_MAX || xp[i] < 0; |
15339 | | } |
15340 | | /* update xpp and tp */ |
15341 | | if (realign) xp = (float *) *xpp; |
15342 | | xp += ni; |
15343 | | tp += ni; |
15344 | | *xpp = (void*)xp; |
15345 | | } |
15346 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15347 | | |
15348 | | #else /* not SX */ |
15349 | 0 | const char *xp = (const char *) *xpp; |
15350 | 0 | int status = NC_NOERR; |
15351 | |
|
15352 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15353 | 0 | { |
15354 | 0 | const int lstatus = ncx_get_float_uint(xp, tp); |
15355 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15356 | 0 | status = lstatus; |
15357 | 0 | } |
15358 | |
|
15359 | 0 | *xpp = (const void *)xp; |
15360 | 0 | return status; |
15361 | 0 | #endif |
15362 | 0 | } |
15363 | | |
15364 | | int |
15365 | | ncx_getn_float_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
15366 | 0 | { |
15367 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15368 | | |
15369 | | /* basic algorithm is: |
15370 | | * - ensure sane alignment of input data |
15371 | | * - copy (conversion happens automatically) input data |
15372 | | * to output |
15373 | | * - update xpp to point at next unconverted input, and tp to point |
15374 | | * at next location for converted output |
15375 | | */ |
15376 | | long i, j, ni; |
15377 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15378 | | float *xp; |
15379 | | int nrange = 0; /* number of range errors */ |
15380 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15381 | | long cxp = (long) *((char**)xpp); |
15382 | | |
15383 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15384 | | /* sjl: manually stripmine so we can limit amount of |
15385 | | * vector work space reserved to LOOPCNT elements. Also |
15386 | | * makes vectorisation easy */ |
15387 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15388 | | ni=Min(nelems-j,LOOPCNT); |
15389 | | if (realign) { |
15390 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_FLOAT)); |
15391 | | xp = tmp; |
15392 | | } else { |
15393 | | xp = (float *) *xpp; |
15394 | | } |
15395 | | /* copy the next block */ |
15396 | | #pragma cdir loopcnt=LOOPCNT |
15397 | | #pragma cdir shortloop |
15398 | | for (i=0; i<ni; i++) { |
15399 | | tp[i] = (ulonglong) Max( ULONGLONG_MIN, Min(ULONGLONG_MAX, (ulonglong) xp[i])); |
15400 | | /* test for range errors (not always needed but do it anyway) */ |
15401 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
15402 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
15403 | | nrange += xp[i] > ULONGLONG_MAX || xp[i] < 0; |
15404 | | } |
15405 | | /* update xpp and tp */ |
15406 | | if (realign) xp = (float *) *xpp; |
15407 | | xp += ni; |
15408 | | tp += ni; |
15409 | | *xpp = (void*)xp; |
15410 | | } |
15411 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15412 | | |
15413 | | #else /* not SX */ |
15414 | 0 | const char *xp = (const char *) *xpp; |
15415 | 0 | int status = NC_NOERR; |
15416 | |
|
15417 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15418 | 0 | { |
15419 | 0 | const int lstatus = ncx_get_float_ulonglong(xp, tp); |
15420 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15421 | 0 | status = lstatus; |
15422 | 0 | } |
15423 | |
|
15424 | 0 | *xpp = (const void *)xp; |
15425 | 0 | return status; |
15426 | 0 | #endif |
15427 | 0 | } |
15428 | | |
15429 | | |
15430 | | int |
15431 | | ncx_putn_float_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
15432 | | #if X_SIZEOF_FLOAT == SIZEOF_FLOAT && !defined(NO_IEEE_FLOAT) |
15433 | | /* optimized version */ |
15434 | 0 | { |
15435 | | #ifdef WORDS_BIGENDIAN |
15436 | | (void) memcpy(*xpp, tp, (size_t)nelems * X_SIZEOF_FLOAT); |
15437 | | # else |
15438 | 0 | swapn4b(*xpp, tp, nelems); |
15439 | 0 | # endif |
15440 | 0 | *xpp = (void *)((char *)(*xpp) + nelems * X_SIZEOF_FLOAT); |
15441 | 0 | return NC_NOERR; |
15442 | 0 | } |
15443 | | #elif defined(vax) && vax != 0 |
15444 | | { |
15445 | | const float *const end = tp + nelems; |
15446 | | |
15447 | | while (tp < end) { |
15448 | | const struct vax_single *const vsp = |
15449 | | (const struct vax_single *)ip; |
15450 | | struct ieee_single *const isp = (struct ieee_single *) (*xpp); |
15451 | | |
15452 | | switch(vsp->exp){ |
15453 | | case 0 : |
15454 | | /* all vax float with zero exponent map to zero */ |
15455 | | *isp = min.ieee; |
15456 | | break; |
15457 | | case 2 : |
15458 | | case 1 : |
15459 | | { |
15460 | | /* These will map to subnormals */ |
15461 | | unsigned mantissa = (vsp->mantissa1 << 16) |
15462 | | | vsp->mantissa2; |
15463 | | mantissa >>= 3 - vsp->exp; |
15464 | | mantissa += (1 << (20 + vsp->exp)); |
15465 | | isp->mant_lo_lo = mantissa; |
15466 | | isp->mant_lo_hi = mantissa >> 8; |
15467 | | isp->mant_hi = mantissa >> 16; |
15468 | | isp->exp_lo = 0; |
15469 | | isp->exp_hi = 0; |
15470 | | } |
15471 | | break; |
15472 | | case 0xff : /* max.s.exp */ |
15473 | | if (vsp->mantissa2 == max.s.mantissa2 && |
15474 | | vsp->mantissa1 == max.s.mantissa1) |
15475 | | { |
15476 | | /* map largest vax float to ieee infinity */ |
15477 | | *isp = max.ieee; |
15478 | | break; |
15479 | | } /* else, fall thru */ |
15480 | | default : |
15481 | | { |
15482 | | unsigned exp = vsp->exp - VAX_SNG_BIAS + IEEE_SNG_BIAS; |
15483 | | isp->exp_hi = exp >> 1; |
15484 | | isp->exp_lo = exp; |
15485 | | isp->mant_lo_lo = vsp->mantissa2; |
15486 | | isp->mant_lo_hi = vsp->mantissa2 >> 8; |
15487 | | isp->mant_hi = vsp->mantissa1; |
15488 | | } |
15489 | | } |
15490 | | |
15491 | | isp->sign = vsp->sign; |
15492 | | |
15493 | | tp++; |
15494 | | *xpp = (char *)(*xpp) + X_SIZEOF_FLOAT; |
15495 | | } |
15496 | | return NC_NOERR; |
15497 | | } |
15498 | | #else |
15499 | | { |
15500 | | char *xp = *xpp; |
15501 | | int status = NC_NOERR; |
15502 | | |
15503 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) { |
15504 | | int lstatus = ncx_put_float_float(xp, tp, fillp); |
15505 | | if (status == NC_NOERR) /* report the first encountered error */ |
15506 | | status = lstatus; |
15507 | | } |
15508 | | |
15509 | | *xpp = (void *)xp; |
15510 | | return status; |
15511 | | } |
15512 | | #endif |
15513 | | int |
15514 | | ncx_putn_float_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
15515 | 0 | { |
15516 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15517 | | |
15518 | | /* basic algorithm is: |
15519 | | * - ensure sane alignment of output data |
15520 | | * - copy (conversion happens automatically) input data |
15521 | | * to output |
15522 | | * - update tp to point at next unconverted input, and xpp to point |
15523 | | * at next location for converted output |
15524 | | */ |
15525 | | long i, j, ni; |
15526 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15527 | | float *xp; |
15528 | | int nrange = 0; /* number of range errors */ |
15529 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15530 | | long cxp = (long) *((char**)xpp); |
15531 | | |
15532 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15533 | | /* sjl: manually stripmine so we can limit amount of |
15534 | | * vector work space reserved to LOOPCNT elements. Also |
15535 | | * makes vectorisation easy */ |
15536 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15537 | | ni=Min(nelems-j,LOOPCNT); |
15538 | | if (realign) { |
15539 | | xp = tmp; |
15540 | | } else { |
15541 | | xp = (float *) *xpp; |
15542 | | } |
15543 | | /* copy the next block */ |
15544 | | #pragma cdir loopcnt=LOOPCNT |
15545 | | #pragma cdir shortloop |
15546 | | for (i=0; i<ni; i++) { |
15547 | | /* the normal case: */ |
15548 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
15549 | | /* test for range errors (not always needed but do it anyway) */ |
15550 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
15551 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
15552 | | nrange += tp[i] > X_FLOAT_MAX || tp[i] < X_FLOAT_MIN; |
15553 | | } |
15554 | | /* copy workspace back if necessary */ |
15555 | | if (realign) { |
15556 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
15557 | | xp = (float *) *xpp; |
15558 | | } |
15559 | | /* update xpp and tp */ |
15560 | | xp += ni; |
15561 | | tp += ni; |
15562 | | *xpp = (void*)xp; |
15563 | | } |
15564 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15565 | | |
15566 | | #else /* not SX */ |
15567 | |
|
15568 | 0 | char *xp = (char *) *xpp; |
15569 | 0 | int status = NC_NOERR; |
15570 | |
|
15571 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15572 | 0 | { |
15573 | 0 | int lstatus = ncx_put_float_schar(xp, tp, fillp); |
15574 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15575 | 0 | status = lstatus; |
15576 | 0 | } |
15577 | |
|
15578 | 0 | *xpp = (void *)xp; |
15579 | 0 | return status; |
15580 | 0 | #endif |
15581 | 0 | } |
15582 | | |
15583 | | int |
15584 | | ncx_putn_float_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
15585 | 0 | { |
15586 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15587 | | |
15588 | | /* basic algorithm is: |
15589 | | * - ensure sane alignment of output data |
15590 | | * - copy (conversion happens automatically) input data |
15591 | | * to output |
15592 | | * - update tp to point at next unconverted input, and xpp to point |
15593 | | * at next location for converted output |
15594 | | */ |
15595 | | long i, j, ni; |
15596 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15597 | | float *xp; |
15598 | | int nrange = 0; /* number of range errors */ |
15599 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15600 | | long cxp = (long) *((char**)xpp); |
15601 | | |
15602 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15603 | | /* sjl: manually stripmine so we can limit amount of |
15604 | | * vector work space reserved to LOOPCNT elements. Also |
15605 | | * makes vectorisation easy */ |
15606 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15607 | | ni=Min(nelems-j,LOOPCNT); |
15608 | | if (realign) { |
15609 | | xp = tmp; |
15610 | | } else { |
15611 | | xp = (float *) *xpp; |
15612 | | } |
15613 | | /* copy the next block */ |
15614 | | #pragma cdir loopcnt=LOOPCNT |
15615 | | #pragma cdir shortloop |
15616 | | for (i=0; i<ni; i++) { |
15617 | | /* the normal case: */ |
15618 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
15619 | | /* test for range errors (not always needed but do it anyway) */ |
15620 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
15621 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
15622 | | nrange += tp[i] > X_FLOAT_MAX || tp[i] < X_FLOAT_MIN; |
15623 | | } |
15624 | | /* copy workspace back if necessary */ |
15625 | | if (realign) { |
15626 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
15627 | | xp = (float *) *xpp; |
15628 | | } |
15629 | | /* update xpp and tp */ |
15630 | | xp += ni; |
15631 | | tp += ni; |
15632 | | *xpp = (void*)xp; |
15633 | | } |
15634 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15635 | | |
15636 | | #else /* not SX */ |
15637 | |
|
15638 | 0 | char *xp = (char *) *xpp; |
15639 | 0 | int status = NC_NOERR; |
15640 | |
|
15641 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15642 | 0 | { |
15643 | 0 | int lstatus = ncx_put_float_short(xp, tp, fillp); |
15644 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15645 | 0 | status = lstatus; |
15646 | 0 | } |
15647 | |
|
15648 | 0 | *xpp = (void *)xp; |
15649 | 0 | return status; |
15650 | 0 | #endif |
15651 | 0 | } |
15652 | | |
15653 | | int |
15654 | | ncx_putn_float_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
15655 | 0 | { |
15656 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15657 | | |
15658 | | /* basic algorithm is: |
15659 | | * - ensure sane alignment of output data |
15660 | | * - copy (conversion happens automatically) input data |
15661 | | * to output |
15662 | | * - update tp to point at next unconverted input, and xpp to point |
15663 | | * at next location for converted output |
15664 | | */ |
15665 | | long i, j, ni; |
15666 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15667 | | float *xp; |
15668 | | int nrange = 0; /* number of range errors */ |
15669 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15670 | | long cxp = (long) *((char**)xpp); |
15671 | | |
15672 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15673 | | /* sjl: manually stripmine so we can limit amount of |
15674 | | * vector work space reserved to LOOPCNT elements. Also |
15675 | | * makes vectorisation easy */ |
15676 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15677 | | ni=Min(nelems-j,LOOPCNT); |
15678 | | if (realign) { |
15679 | | xp = tmp; |
15680 | | } else { |
15681 | | xp = (float *) *xpp; |
15682 | | } |
15683 | | /* copy the next block */ |
15684 | | #pragma cdir loopcnt=LOOPCNT |
15685 | | #pragma cdir shortloop |
15686 | | for (i=0; i<ni; i++) { |
15687 | | /* the normal case: */ |
15688 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
15689 | | /* test for range errors (not always needed but do it anyway) */ |
15690 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
15691 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
15692 | | nrange += tp[i] > X_FLOAT_MAX || tp[i] < X_FLOAT_MIN; |
15693 | | } |
15694 | | /* copy workspace back if necessary */ |
15695 | | if (realign) { |
15696 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
15697 | | xp = (float *) *xpp; |
15698 | | } |
15699 | | /* update xpp and tp */ |
15700 | | xp += ni; |
15701 | | tp += ni; |
15702 | | *xpp = (void*)xp; |
15703 | | } |
15704 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15705 | | |
15706 | | #else /* not SX */ |
15707 | |
|
15708 | 0 | char *xp = (char *) *xpp; |
15709 | 0 | int status = NC_NOERR; |
15710 | |
|
15711 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15712 | 0 | { |
15713 | 0 | int lstatus = ncx_put_float_int(xp, tp, fillp); |
15714 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15715 | 0 | status = lstatus; |
15716 | 0 | } |
15717 | |
|
15718 | 0 | *xpp = (void *)xp; |
15719 | 0 | return status; |
15720 | 0 | #endif |
15721 | 0 | } |
15722 | | |
15723 | | int |
15724 | | ncx_putn_float_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
15725 | 0 | { |
15726 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15727 | | |
15728 | | /* basic algorithm is: |
15729 | | * - ensure sane alignment of output data |
15730 | | * - copy (conversion happens automatically) input data |
15731 | | * to output |
15732 | | * - update tp to point at next unconverted input, and xpp to point |
15733 | | * at next location for converted output |
15734 | | */ |
15735 | | long i, j, ni; |
15736 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15737 | | float *xp; |
15738 | | int nrange = 0; /* number of range errors */ |
15739 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15740 | | long cxp = (long) *((char**)xpp); |
15741 | | |
15742 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15743 | | /* sjl: manually stripmine so we can limit amount of |
15744 | | * vector work space reserved to LOOPCNT elements. Also |
15745 | | * makes vectorisation easy */ |
15746 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15747 | | ni=Min(nelems-j,LOOPCNT); |
15748 | | if (realign) { |
15749 | | xp = tmp; |
15750 | | } else { |
15751 | | xp = (float *) *xpp; |
15752 | | } |
15753 | | /* copy the next block */ |
15754 | | #pragma cdir loopcnt=LOOPCNT |
15755 | | #pragma cdir shortloop |
15756 | | for (i=0; i<ni; i++) { |
15757 | | /* the normal case: */ |
15758 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
15759 | | /* test for range errors (not always needed but do it anyway) */ |
15760 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
15761 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
15762 | | nrange += tp[i] > X_FLOAT_MAX || tp[i] < X_FLOAT_MIN; |
15763 | | } |
15764 | | /* copy workspace back if necessary */ |
15765 | | if (realign) { |
15766 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
15767 | | xp = (float *) *xpp; |
15768 | | } |
15769 | | /* update xpp and tp */ |
15770 | | xp += ni; |
15771 | | tp += ni; |
15772 | | *xpp = (void*)xp; |
15773 | | } |
15774 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15775 | | |
15776 | | #else /* not SX */ |
15777 | |
|
15778 | 0 | char *xp = (char *) *xpp; |
15779 | 0 | int status = NC_NOERR; |
15780 | |
|
15781 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15782 | 0 | { |
15783 | 0 | int lstatus = ncx_put_float_long(xp, tp, fillp); |
15784 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15785 | 0 | status = lstatus; |
15786 | 0 | } |
15787 | |
|
15788 | 0 | *xpp = (void *)xp; |
15789 | 0 | return status; |
15790 | 0 | #endif |
15791 | 0 | } |
15792 | | |
15793 | | int |
15794 | | ncx_putn_float_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
15795 | 0 | { |
15796 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15797 | | |
15798 | | /* basic algorithm is: |
15799 | | * - ensure sane alignment of output data |
15800 | | * - copy (conversion happens automatically) input data |
15801 | | * to output |
15802 | | * - update tp to point at next unconverted input, and xpp to point |
15803 | | * at next location for converted output |
15804 | | */ |
15805 | | long i, j, ni; |
15806 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15807 | | float *xp; |
15808 | | int nrange = 0; /* number of range errors */ |
15809 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15810 | | long cxp = (long) *((char**)xpp); |
15811 | | |
15812 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15813 | | /* sjl: manually stripmine so we can limit amount of |
15814 | | * vector work space reserved to LOOPCNT elements. Also |
15815 | | * makes vectorisation easy */ |
15816 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15817 | | ni=Min(nelems-j,LOOPCNT); |
15818 | | if (realign) { |
15819 | | xp = tmp; |
15820 | | } else { |
15821 | | xp = (float *) *xpp; |
15822 | | } |
15823 | | /* copy the next block */ |
15824 | | #pragma cdir loopcnt=LOOPCNT |
15825 | | #pragma cdir shortloop |
15826 | | for (i=0; i<ni; i++) { |
15827 | | /* the normal case: */ |
15828 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
15829 | | /* test for range errors (not always needed but do it anyway) */ |
15830 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
15831 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
15832 | | nrange += tp[i] > X_FLOAT_MAX || tp[i] < X_FLOAT_MIN; |
15833 | | } |
15834 | | /* copy workspace back if necessary */ |
15835 | | if (realign) { |
15836 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
15837 | | xp = (float *) *xpp; |
15838 | | } |
15839 | | /* update xpp and tp */ |
15840 | | xp += ni; |
15841 | | tp += ni; |
15842 | | *xpp = (void*)xp; |
15843 | | } |
15844 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15845 | | |
15846 | | #else /* not SX */ |
15847 | |
|
15848 | 0 | char *xp = (char *) *xpp; |
15849 | 0 | int status = NC_NOERR; |
15850 | |
|
15851 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15852 | 0 | { |
15853 | 0 | int lstatus = ncx_put_float_double(xp, tp, fillp); |
15854 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15855 | 0 | status = lstatus; |
15856 | 0 | } |
15857 | |
|
15858 | 0 | *xpp = (void *)xp; |
15859 | 0 | return status; |
15860 | 0 | #endif |
15861 | 0 | } |
15862 | | |
15863 | | int |
15864 | | ncx_putn_float_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
15865 | 0 | { |
15866 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15867 | | |
15868 | | /* basic algorithm is: |
15869 | | * - ensure sane alignment of output data |
15870 | | * - copy (conversion happens automatically) input data |
15871 | | * to output |
15872 | | * - update tp to point at next unconverted input, and xpp to point |
15873 | | * at next location for converted output |
15874 | | */ |
15875 | | long i, j, ni; |
15876 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15877 | | float *xp; |
15878 | | int nrange = 0; /* number of range errors */ |
15879 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15880 | | long cxp = (long) *((char**)xpp); |
15881 | | |
15882 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15883 | | /* sjl: manually stripmine so we can limit amount of |
15884 | | * vector work space reserved to LOOPCNT elements. Also |
15885 | | * makes vectorisation easy */ |
15886 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15887 | | ni=Min(nelems-j,LOOPCNT); |
15888 | | if (realign) { |
15889 | | xp = tmp; |
15890 | | } else { |
15891 | | xp = (float *) *xpp; |
15892 | | } |
15893 | | /* copy the next block */ |
15894 | | #pragma cdir loopcnt=LOOPCNT |
15895 | | #pragma cdir shortloop |
15896 | | for (i=0; i<ni; i++) { |
15897 | | /* the normal case: */ |
15898 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
15899 | | /* test for range errors (not always needed but do it anyway) */ |
15900 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
15901 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
15902 | | nrange += tp[i] > X_FLOAT_MAX || tp[i] < X_FLOAT_MIN; |
15903 | | } |
15904 | | /* copy workspace back if necessary */ |
15905 | | if (realign) { |
15906 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
15907 | | xp = (float *) *xpp; |
15908 | | } |
15909 | | /* update xpp and tp */ |
15910 | | xp += ni; |
15911 | | tp += ni; |
15912 | | *xpp = (void*)xp; |
15913 | | } |
15914 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15915 | | |
15916 | | #else /* not SX */ |
15917 | |
|
15918 | 0 | char *xp = (char *) *xpp; |
15919 | 0 | int status = NC_NOERR; |
15920 | |
|
15921 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15922 | 0 | { |
15923 | 0 | int lstatus = ncx_put_float_longlong(xp, tp, fillp); |
15924 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15925 | 0 | status = lstatus; |
15926 | 0 | } |
15927 | |
|
15928 | 0 | *xpp = (void *)xp; |
15929 | 0 | return status; |
15930 | 0 | #endif |
15931 | 0 | } |
15932 | | |
15933 | | int |
15934 | | ncx_putn_float_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
15935 | 0 | { |
15936 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
15937 | | |
15938 | | /* basic algorithm is: |
15939 | | * - ensure sane alignment of output data |
15940 | | * - copy (conversion happens automatically) input data |
15941 | | * to output |
15942 | | * - update tp to point at next unconverted input, and xpp to point |
15943 | | * at next location for converted output |
15944 | | */ |
15945 | | long i, j, ni; |
15946 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
15947 | | float *xp; |
15948 | | int nrange = 0; /* number of range errors */ |
15949 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
15950 | | long cxp = (long) *((char**)xpp); |
15951 | | |
15952 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
15953 | | /* sjl: manually stripmine so we can limit amount of |
15954 | | * vector work space reserved to LOOPCNT elements. Also |
15955 | | * makes vectorisation easy */ |
15956 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
15957 | | ni=Min(nelems-j,LOOPCNT); |
15958 | | if (realign) { |
15959 | | xp = tmp; |
15960 | | } else { |
15961 | | xp = (float *) *xpp; |
15962 | | } |
15963 | | /* copy the next block */ |
15964 | | #pragma cdir loopcnt=LOOPCNT |
15965 | | #pragma cdir shortloop |
15966 | | for (i=0; i<ni; i++) { |
15967 | | /* the normal case: */ |
15968 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
15969 | | /* test for range errors (not always needed but do it anyway) */ |
15970 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
15971 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
15972 | | nrange += tp[i] > X_FLOAT_MAX ; |
15973 | | } |
15974 | | /* copy workspace back if necessary */ |
15975 | | if (realign) { |
15976 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
15977 | | xp = (float *) *xpp; |
15978 | | } |
15979 | | /* update xpp and tp */ |
15980 | | xp += ni; |
15981 | | tp += ni; |
15982 | | *xpp = (void*)xp; |
15983 | | } |
15984 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
15985 | | |
15986 | | #else /* not SX */ |
15987 | |
|
15988 | 0 | char *xp = (char *) *xpp; |
15989 | 0 | int status = NC_NOERR; |
15990 | |
|
15991 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
15992 | 0 | { |
15993 | 0 | int lstatus = ncx_put_float_uchar(xp, tp, fillp); |
15994 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
15995 | 0 | status = lstatus; |
15996 | 0 | } |
15997 | |
|
15998 | 0 | *xpp = (void *)xp; |
15999 | 0 | return status; |
16000 | 0 | #endif |
16001 | 0 | } |
16002 | | |
16003 | | int |
16004 | | ncx_putn_float_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
16005 | 0 | { |
16006 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
16007 | | |
16008 | | /* basic algorithm is: |
16009 | | * - ensure sane alignment of output data |
16010 | | * - copy (conversion happens automatically) input data |
16011 | | * to output |
16012 | | * - update tp to point at next unconverted input, and xpp to point |
16013 | | * at next location for converted output |
16014 | | */ |
16015 | | long i, j, ni; |
16016 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
16017 | | float *xp; |
16018 | | int nrange = 0; /* number of range errors */ |
16019 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16020 | | long cxp = (long) *((char**)xpp); |
16021 | | |
16022 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
16023 | | /* sjl: manually stripmine so we can limit amount of |
16024 | | * vector work space reserved to LOOPCNT elements. Also |
16025 | | * makes vectorisation easy */ |
16026 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16027 | | ni=Min(nelems-j,LOOPCNT); |
16028 | | if (realign) { |
16029 | | xp = tmp; |
16030 | | } else { |
16031 | | xp = (float *) *xpp; |
16032 | | } |
16033 | | /* copy the next block */ |
16034 | | #pragma cdir loopcnt=LOOPCNT |
16035 | | #pragma cdir shortloop |
16036 | | for (i=0; i<ni; i++) { |
16037 | | /* the normal case: */ |
16038 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
16039 | | /* test for range errors (not always needed but do it anyway) */ |
16040 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
16041 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
16042 | | nrange += tp[i] > X_FLOAT_MAX ; |
16043 | | } |
16044 | | /* copy workspace back if necessary */ |
16045 | | if (realign) { |
16046 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
16047 | | xp = (float *) *xpp; |
16048 | | } |
16049 | | /* update xpp and tp */ |
16050 | | xp += ni; |
16051 | | tp += ni; |
16052 | | *xpp = (void*)xp; |
16053 | | } |
16054 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16055 | | |
16056 | | #else /* not SX */ |
16057 | |
|
16058 | 0 | char *xp = (char *) *xpp; |
16059 | 0 | int status = NC_NOERR; |
16060 | |
|
16061 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
16062 | 0 | { |
16063 | 0 | int lstatus = ncx_put_float_ushort(xp, tp, fillp); |
16064 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16065 | 0 | status = lstatus; |
16066 | 0 | } |
16067 | |
|
16068 | 0 | *xpp = (void *)xp; |
16069 | 0 | return status; |
16070 | 0 | #endif |
16071 | 0 | } |
16072 | | |
16073 | | int |
16074 | | ncx_putn_float_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
16075 | 0 | { |
16076 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
16077 | | |
16078 | | /* basic algorithm is: |
16079 | | * - ensure sane alignment of output data |
16080 | | * - copy (conversion happens automatically) input data |
16081 | | * to output |
16082 | | * - update tp to point at next unconverted input, and xpp to point |
16083 | | * at next location for converted output |
16084 | | */ |
16085 | | long i, j, ni; |
16086 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
16087 | | float *xp; |
16088 | | int nrange = 0; /* number of range errors */ |
16089 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16090 | | long cxp = (long) *((char**)xpp); |
16091 | | |
16092 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
16093 | | /* sjl: manually stripmine so we can limit amount of |
16094 | | * vector work space reserved to LOOPCNT elements. Also |
16095 | | * makes vectorisation easy */ |
16096 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16097 | | ni=Min(nelems-j,LOOPCNT); |
16098 | | if (realign) { |
16099 | | xp = tmp; |
16100 | | } else { |
16101 | | xp = (float *) *xpp; |
16102 | | } |
16103 | | /* copy the next block */ |
16104 | | #pragma cdir loopcnt=LOOPCNT |
16105 | | #pragma cdir shortloop |
16106 | | for (i=0; i<ni; i++) { |
16107 | | /* the normal case: */ |
16108 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
16109 | | /* test for range errors (not always needed but do it anyway) */ |
16110 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
16111 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
16112 | | nrange += tp[i] > X_FLOAT_MAX ; |
16113 | | } |
16114 | | /* copy workspace back if necessary */ |
16115 | | if (realign) { |
16116 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
16117 | | xp = (float *) *xpp; |
16118 | | } |
16119 | | /* update xpp and tp */ |
16120 | | xp += ni; |
16121 | | tp += ni; |
16122 | | *xpp = (void*)xp; |
16123 | | } |
16124 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16125 | | |
16126 | | #else /* not SX */ |
16127 | |
|
16128 | 0 | char *xp = (char *) *xpp; |
16129 | 0 | int status = NC_NOERR; |
16130 | |
|
16131 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
16132 | 0 | { |
16133 | 0 | int lstatus = ncx_put_float_uint(xp, tp, fillp); |
16134 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16135 | 0 | status = lstatus; |
16136 | 0 | } |
16137 | |
|
16138 | 0 | *xpp = (void *)xp; |
16139 | 0 | return status; |
16140 | 0 | #endif |
16141 | 0 | } |
16142 | | |
16143 | | int |
16144 | | ncx_putn_float_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
16145 | 0 | { |
16146 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_FLOAT == SIZEOF_FLOAT |
16147 | | |
16148 | | /* basic algorithm is: |
16149 | | * - ensure sane alignment of output data |
16150 | | * - copy (conversion happens automatically) input data |
16151 | | * to output |
16152 | | * - update tp to point at next unconverted input, and xpp to point |
16153 | | * at next location for converted output |
16154 | | */ |
16155 | | long i, j, ni; |
16156 | | float tmp[LOOPCNT]; /* in case input is misaligned */ |
16157 | | float *xp; |
16158 | | int nrange = 0; /* number of range errors */ |
16159 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16160 | | long cxp = (long) *((char**)xpp); |
16161 | | |
16162 | | realign = (cxp & 7) % SIZEOF_FLOAT; |
16163 | | /* sjl: manually stripmine so we can limit amount of |
16164 | | * vector work space reserved to LOOPCNT elements. Also |
16165 | | * makes vectorisation easy */ |
16166 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16167 | | ni=Min(nelems-j,LOOPCNT); |
16168 | | if (realign) { |
16169 | | xp = tmp; |
16170 | | } else { |
16171 | | xp = (float *) *xpp; |
16172 | | } |
16173 | | /* copy the next block */ |
16174 | | #pragma cdir loopcnt=LOOPCNT |
16175 | | #pragma cdir shortloop |
16176 | | for (i=0; i<ni; i++) { |
16177 | | /* the normal case: */ |
16178 | | xp[i] = (float) Max( X_FLOAT_MIN, Min(X_FLOAT_MAX, (float) tp[i])); |
16179 | | /* test for range errors (not always needed but do it anyway) */ |
16180 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
16181 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
16182 | | nrange += tp[i] > X_FLOAT_MAX ; |
16183 | | } |
16184 | | /* copy workspace back if necessary */ |
16185 | | if (realign) { |
16186 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_FLOAT); |
16187 | | xp = (float *) *xpp; |
16188 | | } |
16189 | | /* update xpp and tp */ |
16190 | | xp += ni; |
16191 | | tp += ni; |
16192 | | *xpp = (void*)xp; |
16193 | | } |
16194 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16195 | | |
16196 | | #else /* not SX */ |
16197 | |
|
16198 | 0 | char *xp = (char *) *xpp; |
16199 | 0 | int status = NC_NOERR; |
16200 | |
|
16201 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_FLOAT, tp++) |
16202 | 0 | { |
16203 | 0 | int lstatus = ncx_put_float_ulonglong(xp, tp, fillp); |
16204 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16205 | 0 | status = lstatus; |
16206 | 0 | } |
16207 | |
|
16208 | 0 | *xpp = (void *)xp; |
16209 | 0 | return status; |
16210 | 0 | #endif |
16211 | 0 | } |
16212 | | |
16213 | | |
16214 | | /* double --------------------------------------------------------------------*/ |
16215 | | |
16216 | | #if X_SIZEOF_DOUBLE == SIZEOF_DOUBLE && !defined(NO_IEEE_FLOAT) |
16217 | | /* optimized version */ |
16218 | | int |
16219 | | ncx_getn_double_double(const void **xpp, size_t nelems, double *tp) |
16220 | 0 | { |
16221 | | #ifdef WORDS_BIGENDIAN |
16222 | | (void) memcpy(tp, *xpp, (size_t)nelems * SIZEOF_DOUBLE); |
16223 | | # else |
16224 | 0 | swapn8b(tp, *xpp, nelems); |
16225 | 0 | # endif |
16226 | 0 | *xpp = (const void *)((const char *)(*xpp) + nelems * X_SIZEOF_DOUBLE); |
16227 | 0 | return NC_NOERR; |
16228 | 0 | } |
16229 | | #elif defined(vax) && vax != 0 |
16230 | | int |
16231 | | ncx_getn_double_double(const void **xpp, size_t ndoubles, double *ip) |
16232 | | { |
16233 | | double *const end = ip + ndoubles; |
16234 | | |
16235 | | while (ip < end) |
16236 | | { |
16237 | | struct vax_double *const vdp = |
16238 | | (struct vax_double *)ip; |
16239 | | const struct ieee_double *const idp = |
16240 | | (const struct ieee_double *) (*xpp); |
16241 | | { |
16242 | | const struct dbl_limits *lim; |
16243 | | int ii; |
16244 | | for (ii = 0, lim = dbl_limits; |
16245 | | ii < sizeof(dbl_limits)/sizeof(struct dbl_limits); |
16246 | | ii++, lim++) |
16247 | | { |
16248 | | if ((idp->mant_lo == lim->ieee.mant_lo) |
16249 | | && (idp->mant_4 == lim->ieee.mant_4) |
16250 | | && (idp->mant_5 == lim->ieee.mant_5) |
16251 | | && (idp->mant_6 == lim->ieee.mant_6) |
16252 | | && (idp->exp_lo == lim->ieee.exp_lo) |
16253 | | && (idp->exp_hi == lim->ieee.exp_hi) |
16254 | | ) |
16255 | | { |
16256 | | *vdp = lim->d; |
16257 | | goto doneit; |
16258 | | } |
16259 | | } |
16260 | | } |
16261 | | { |
16262 | | unsigned exp = idp->exp_hi << 4 | idp->exp_lo; |
16263 | | vdp->exp = exp - IEEE_DBL_BIAS + VAX_DBL_BIAS; |
16264 | | } |
16265 | | { |
16266 | | unsigned mant_hi = ((idp->mant_6 << 16) |
16267 | | | (idp->mant_5 << 8) |
16268 | | | idp->mant_4); |
16269 | | unsigned mant_lo = SWAP4(idp->mant_lo); |
16270 | | vdp->mantissa1 = (mant_hi >> 13); |
16271 | | vdp->mantissa2 = ((mant_hi & MASK(13)) << 3) |
16272 | | | (mant_lo >> 29); |
16273 | | vdp->mantissa3 = (mant_lo >> 13); |
16274 | | vdp->mantissa4 = (mant_lo << 3); |
16275 | | } |
16276 | | doneit: |
16277 | | vdp->sign = idp->sign; |
16278 | | |
16279 | | ip++; |
16280 | | *xpp = (char *)(*xpp) + X_SIZEOF_DOUBLE; |
16281 | | } |
16282 | | return NC_NOERR; |
16283 | | } |
16284 | | /* vax */ |
16285 | | #else |
16286 | | int |
16287 | | ncx_getn_double_double(const void **xpp, size_t nelems, double *tp) |
16288 | | { |
16289 | | const char *xp = *xpp; |
16290 | | int status = NC_NOERR; |
16291 | | |
16292 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16293 | | { |
16294 | | const int lstatus = ncx_get_double_double(xp, tp, fillp); |
16295 | | if (status == NC_NOERR) /* report the first encountered error */ |
16296 | | status = lstatus; |
16297 | | } |
16298 | | |
16299 | | *xpp = (const void *)xp; |
16300 | | return status; |
16301 | | } |
16302 | | #endif |
16303 | | int |
16304 | | ncx_getn_double_schar(const void **xpp, size_t nelems, schar *tp) |
16305 | 0 | { |
16306 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16307 | | |
16308 | | /* basic algorithm is: |
16309 | | * - ensure sane alignment of input data |
16310 | | * - copy (conversion happens automatically) input data |
16311 | | * to output |
16312 | | * - update xpp to point at next unconverted input, and tp to point |
16313 | | * at next location for converted output |
16314 | | */ |
16315 | | long i, j, ni; |
16316 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16317 | | double *xp; |
16318 | | int nrange = 0; /* number of range errors */ |
16319 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16320 | | long cxp = (long) *((char**)xpp); |
16321 | | |
16322 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16323 | | /* sjl: manually stripmine so we can limit amount of |
16324 | | * vector work space reserved to LOOPCNT elements. Also |
16325 | | * makes vectorisation easy */ |
16326 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16327 | | ni=Min(nelems-j,LOOPCNT); |
16328 | | if (realign) { |
16329 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16330 | | xp = tmp; |
16331 | | } else { |
16332 | | xp = (double *) *xpp; |
16333 | | } |
16334 | | /* copy the next block */ |
16335 | | #pragma cdir loopcnt=LOOPCNT |
16336 | | #pragma cdir shortloop |
16337 | | for (i=0; i<ni; i++) { |
16338 | | tp[i] = (schar) Max( SCHAR_MIN, Min(SCHAR_MAX, (schar) xp[i])); |
16339 | | /* test for range errors (not always needed but do it anyway) */ |
16340 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16341 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16342 | | nrange += xp[i] > SCHAR_MAX || xp[i] < SCHAR_MIN; |
16343 | | } |
16344 | | /* update xpp and tp */ |
16345 | | if (realign) xp = (double *) *xpp; |
16346 | | xp += ni; |
16347 | | tp += ni; |
16348 | | *xpp = (void*)xp; |
16349 | | } |
16350 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16351 | | |
16352 | | #else /* not SX */ |
16353 | 0 | const char *xp = (const char *) *xpp; |
16354 | 0 | int status = NC_NOERR; |
16355 | |
|
16356 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16357 | 0 | { |
16358 | 0 | const int lstatus = ncx_get_double_schar(xp, tp); |
16359 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16360 | 0 | status = lstatus; |
16361 | 0 | } |
16362 | |
|
16363 | 0 | *xpp = (const void *)xp; |
16364 | 0 | return status; |
16365 | 0 | #endif |
16366 | 0 | } |
16367 | | |
16368 | | int |
16369 | | ncx_getn_double_short(const void **xpp, size_t nelems, short *tp) |
16370 | 0 | { |
16371 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16372 | | |
16373 | | /* basic algorithm is: |
16374 | | * - ensure sane alignment of input data |
16375 | | * - copy (conversion happens automatically) input data |
16376 | | * to output |
16377 | | * - update xpp to point at next unconverted input, and tp to point |
16378 | | * at next location for converted output |
16379 | | */ |
16380 | | long i, j, ni; |
16381 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16382 | | double *xp; |
16383 | | int nrange = 0; /* number of range errors */ |
16384 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16385 | | long cxp = (long) *((char**)xpp); |
16386 | | |
16387 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16388 | | /* sjl: manually stripmine so we can limit amount of |
16389 | | * vector work space reserved to LOOPCNT elements. Also |
16390 | | * makes vectorisation easy */ |
16391 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16392 | | ni=Min(nelems-j,LOOPCNT); |
16393 | | if (realign) { |
16394 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16395 | | xp = tmp; |
16396 | | } else { |
16397 | | xp = (double *) *xpp; |
16398 | | } |
16399 | | /* copy the next block */ |
16400 | | #pragma cdir loopcnt=LOOPCNT |
16401 | | #pragma cdir shortloop |
16402 | | for (i=0; i<ni; i++) { |
16403 | | tp[i] = (short) Max( SHORT_MIN, Min(SHORT_MAX, (short) xp[i])); |
16404 | | /* test for range errors (not always needed but do it anyway) */ |
16405 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16406 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16407 | | nrange += xp[i] > SHORT_MAX || xp[i] < SHORT_MIN; |
16408 | | } |
16409 | | /* update xpp and tp */ |
16410 | | if (realign) xp = (double *) *xpp; |
16411 | | xp += ni; |
16412 | | tp += ni; |
16413 | | *xpp = (void*)xp; |
16414 | | } |
16415 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16416 | | |
16417 | | #else /* not SX */ |
16418 | 0 | const char *xp = (const char *) *xpp; |
16419 | 0 | int status = NC_NOERR; |
16420 | |
|
16421 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16422 | 0 | { |
16423 | 0 | const int lstatus = ncx_get_double_short(xp, tp); |
16424 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16425 | 0 | status = lstatus; |
16426 | 0 | } |
16427 | |
|
16428 | 0 | *xpp = (const void *)xp; |
16429 | 0 | return status; |
16430 | 0 | #endif |
16431 | 0 | } |
16432 | | |
16433 | | int |
16434 | | ncx_getn_double_int(const void **xpp, size_t nelems, int *tp) |
16435 | 0 | { |
16436 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16437 | | |
16438 | | /* basic algorithm is: |
16439 | | * - ensure sane alignment of input data |
16440 | | * - copy (conversion happens automatically) input data |
16441 | | * to output |
16442 | | * - update xpp to point at next unconverted input, and tp to point |
16443 | | * at next location for converted output |
16444 | | */ |
16445 | | long i, j, ni; |
16446 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16447 | | double *xp; |
16448 | | int nrange = 0; /* number of range errors */ |
16449 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16450 | | long cxp = (long) *((char**)xpp); |
16451 | | |
16452 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16453 | | /* sjl: manually stripmine so we can limit amount of |
16454 | | * vector work space reserved to LOOPCNT elements. Also |
16455 | | * makes vectorisation easy */ |
16456 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16457 | | ni=Min(nelems-j,LOOPCNT); |
16458 | | if (realign) { |
16459 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16460 | | xp = tmp; |
16461 | | } else { |
16462 | | xp = (double *) *xpp; |
16463 | | } |
16464 | | /* copy the next block */ |
16465 | | #pragma cdir loopcnt=LOOPCNT |
16466 | | #pragma cdir shortloop |
16467 | | for (i=0; i<ni; i++) { |
16468 | | tp[i] = (int) Max( INT_MIN, Min(INT_MAX, (int) xp[i])); |
16469 | | /* test for range errors (not always needed but do it anyway) */ |
16470 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16471 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16472 | | nrange += xp[i] > INT_MAX || xp[i] < INT_MIN; |
16473 | | } |
16474 | | /* update xpp and tp */ |
16475 | | if (realign) xp = (double *) *xpp; |
16476 | | xp += ni; |
16477 | | tp += ni; |
16478 | | *xpp = (void*)xp; |
16479 | | } |
16480 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16481 | | |
16482 | | #else /* not SX */ |
16483 | 0 | const char *xp = (const char *) *xpp; |
16484 | 0 | int status = NC_NOERR; |
16485 | |
|
16486 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16487 | 0 | { |
16488 | 0 | const int lstatus = ncx_get_double_int(xp, tp); |
16489 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16490 | 0 | status = lstatus; |
16491 | 0 | } |
16492 | |
|
16493 | 0 | *xpp = (const void *)xp; |
16494 | 0 | return status; |
16495 | 0 | #endif |
16496 | 0 | } |
16497 | | |
16498 | | int |
16499 | | ncx_getn_double_long(const void **xpp, size_t nelems, long *tp) |
16500 | 0 | { |
16501 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16502 | | |
16503 | | /* basic algorithm is: |
16504 | | * - ensure sane alignment of input data |
16505 | | * - copy (conversion happens automatically) input data |
16506 | | * to output |
16507 | | * - update xpp to point at next unconverted input, and tp to point |
16508 | | * at next location for converted output |
16509 | | */ |
16510 | | long i, j, ni; |
16511 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16512 | | double *xp; |
16513 | | int nrange = 0; /* number of range errors */ |
16514 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16515 | | long cxp = (long) *((char**)xpp); |
16516 | | |
16517 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16518 | | /* sjl: manually stripmine so we can limit amount of |
16519 | | * vector work space reserved to LOOPCNT elements. Also |
16520 | | * makes vectorisation easy */ |
16521 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16522 | | ni=Min(nelems-j,LOOPCNT); |
16523 | | if (realign) { |
16524 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16525 | | xp = tmp; |
16526 | | } else { |
16527 | | xp = (double *) *xpp; |
16528 | | } |
16529 | | /* copy the next block */ |
16530 | | #pragma cdir loopcnt=LOOPCNT |
16531 | | #pragma cdir shortloop |
16532 | | for (i=0; i<ni; i++) { |
16533 | | tp[i] = (long) Max( LONG_MIN, Min(LONG_MAX, (long) xp[i])); |
16534 | | /* test for range errors (not always needed but do it anyway) */ |
16535 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16536 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16537 | | nrange += xp[i] > LONG_MAX || xp[i] < LONG_MIN; |
16538 | | } |
16539 | | /* update xpp and tp */ |
16540 | | if (realign) xp = (double *) *xpp; |
16541 | | xp += ni; |
16542 | | tp += ni; |
16543 | | *xpp = (void*)xp; |
16544 | | } |
16545 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16546 | | |
16547 | | #else /* not SX */ |
16548 | 0 | const char *xp = (const char *) *xpp; |
16549 | 0 | int status = NC_NOERR; |
16550 | |
|
16551 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16552 | 0 | { |
16553 | 0 | const int lstatus = ncx_get_double_long(xp, tp); |
16554 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16555 | 0 | status = lstatus; |
16556 | 0 | } |
16557 | |
|
16558 | 0 | *xpp = (const void *)xp; |
16559 | 0 | return status; |
16560 | 0 | #endif |
16561 | 0 | } |
16562 | | |
16563 | | int |
16564 | | ncx_getn_double_float(const void **xpp, size_t nelems, float *tp) |
16565 | 0 | { |
16566 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16567 | | |
16568 | | /* basic algorithm is: |
16569 | | * - ensure sane alignment of input data |
16570 | | * - copy (conversion happens automatically) input data |
16571 | | * to output |
16572 | | * - update xpp to point at next unconverted input, and tp to point |
16573 | | * at next location for converted output |
16574 | | */ |
16575 | | long i, j, ni; |
16576 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16577 | | double *xp; |
16578 | | int nrange = 0; /* number of range errors */ |
16579 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16580 | | long cxp = (long) *((char**)xpp); |
16581 | | |
16582 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16583 | | /* sjl: manually stripmine so we can limit amount of |
16584 | | * vector work space reserved to LOOPCNT elements. Also |
16585 | | * makes vectorisation easy */ |
16586 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16587 | | ni=Min(nelems-j,LOOPCNT); |
16588 | | if (realign) { |
16589 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16590 | | xp = tmp; |
16591 | | } else { |
16592 | | xp = (double *) *xpp; |
16593 | | } |
16594 | | /* copy the next block */ |
16595 | | #pragma cdir loopcnt=LOOPCNT |
16596 | | #pragma cdir shortloop |
16597 | | for (i=0; i<ni; i++) { |
16598 | | tp[i] = (float) Max( FLOAT_MIN, Min(FLOAT_MAX, (float) xp[i])); |
16599 | | /* test for range errors (not always needed but do it anyway) */ |
16600 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16601 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16602 | | nrange += xp[i] > FLOAT_MAX || xp[i] < FLOAT_MIN; |
16603 | | } |
16604 | | /* update xpp and tp */ |
16605 | | if (realign) xp = (double *) *xpp; |
16606 | | xp += ni; |
16607 | | tp += ni; |
16608 | | *xpp = (void*)xp; |
16609 | | } |
16610 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16611 | | |
16612 | | #else /* not SX */ |
16613 | 0 | const char *xp = (const char *) *xpp; |
16614 | 0 | int status = NC_NOERR; |
16615 | |
|
16616 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16617 | 0 | { |
16618 | 0 | const int lstatus = ncx_get_double_float(xp, tp); |
16619 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16620 | 0 | status = lstatus; |
16621 | 0 | } |
16622 | |
|
16623 | 0 | *xpp = (const void *)xp; |
16624 | 0 | return status; |
16625 | 0 | #endif |
16626 | 0 | } |
16627 | | |
16628 | | int |
16629 | | ncx_getn_double_longlong(const void **xpp, size_t nelems, longlong *tp) |
16630 | 0 | { |
16631 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16632 | | |
16633 | | /* basic algorithm is: |
16634 | | * - ensure sane alignment of input data |
16635 | | * - copy (conversion happens automatically) input data |
16636 | | * to output |
16637 | | * - update xpp to point at next unconverted input, and tp to point |
16638 | | * at next location for converted output |
16639 | | */ |
16640 | | long i, j, ni; |
16641 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16642 | | double *xp; |
16643 | | int nrange = 0; /* number of range errors */ |
16644 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16645 | | long cxp = (long) *((char**)xpp); |
16646 | | |
16647 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16648 | | /* sjl: manually stripmine so we can limit amount of |
16649 | | * vector work space reserved to LOOPCNT elements. Also |
16650 | | * makes vectorisation easy */ |
16651 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16652 | | ni=Min(nelems-j,LOOPCNT); |
16653 | | if (realign) { |
16654 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16655 | | xp = tmp; |
16656 | | } else { |
16657 | | xp = (double *) *xpp; |
16658 | | } |
16659 | | /* copy the next block */ |
16660 | | #pragma cdir loopcnt=LOOPCNT |
16661 | | #pragma cdir shortloop |
16662 | | for (i=0; i<ni; i++) { |
16663 | | tp[i] = (longlong) Max( LONGLONG_MIN, Min(LONGLONG_MAX, (longlong) xp[i])); |
16664 | | /* test for range errors (not always needed but do it anyway) */ |
16665 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16666 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16667 | | nrange += xp[i] > LONGLONG_MAX || xp[i] < LONGLONG_MIN; |
16668 | | } |
16669 | | /* update xpp and tp */ |
16670 | | if (realign) xp = (double *) *xpp; |
16671 | | xp += ni; |
16672 | | tp += ni; |
16673 | | *xpp = (void*)xp; |
16674 | | } |
16675 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16676 | | |
16677 | | #else /* not SX */ |
16678 | 0 | const char *xp = (const char *) *xpp; |
16679 | 0 | int status = NC_NOERR; |
16680 | |
|
16681 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16682 | 0 | { |
16683 | 0 | const int lstatus = ncx_get_double_longlong(xp, tp); |
16684 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16685 | 0 | status = lstatus; |
16686 | 0 | } |
16687 | |
|
16688 | 0 | *xpp = (const void *)xp; |
16689 | 0 | return status; |
16690 | 0 | #endif |
16691 | 0 | } |
16692 | | |
16693 | | int |
16694 | | ncx_getn_double_uchar(const void **xpp, size_t nelems, uchar *tp) |
16695 | 0 | { |
16696 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16697 | | |
16698 | | /* basic algorithm is: |
16699 | | * - ensure sane alignment of input data |
16700 | | * - copy (conversion happens automatically) input data |
16701 | | * to output |
16702 | | * - update xpp to point at next unconverted input, and tp to point |
16703 | | * at next location for converted output |
16704 | | */ |
16705 | | long i, j, ni; |
16706 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16707 | | double *xp; |
16708 | | int nrange = 0; /* number of range errors */ |
16709 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16710 | | long cxp = (long) *((char**)xpp); |
16711 | | |
16712 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16713 | | /* sjl: manually stripmine so we can limit amount of |
16714 | | * vector work space reserved to LOOPCNT elements. Also |
16715 | | * makes vectorisation easy */ |
16716 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16717 | | ni=Min(nelems-j,LOOPCNT); |
16718 | | if (realign) { |
16719 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16720 | | xp = tmp; |
16721 | | } else { |
16722 | | xp = (double *) *xpp; |
16723 | | } |
16724 | | /* copy the next block */ |
16725 | | #pragma cdir loopcnt=LOOPCNT |
16726 | | #pragma cdir shortloop |
16727 | | for (i=0; i<ni; i++) { |
16728 | | tp[i] = (uchar) Max( UCHAR_MIN, Min(UCHAR_MAX, (uchar) xp[i])); |
16729 | | /* test for range errors (not always needed but do it anyway) */ |
16730 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16731 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16732 | | nrange += xp[i] > UCHAR_MAX || xp[i] < 0; |
16733 | | } |
16734 | | /* update xpp and tp */ |
16735 | | if (realign) xp = (double *) *xpp; |
16736 | | xp += ni; |
16737 | | tp += ni; |
16738 | | *xpp = (void*)xp; |
16739 | | } |
16740 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16741 | | |
16742 | | #else /* not SX */ |
16743 | 0 | const char *xp = (const char *) *xpp; |
16744 | 0 | int status = NC_NOERR; |
16745 | |
|
16746 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16747 | 0 | { |
16748 | 0 | const int lstatus = ncx_get_double_uchar(xp, tp); |
16749 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16750 | 0 | status = lstatus; |
16751 | 0 | } |
16752 | |
|
16753 | 0 | *xpp = (const void *)xp; |
16754 | 0 | return status; |
16755 | 0 | #endif |
16756 | 0 | } |
16757 | | |
16758 | | int |
16759 | | ncx_getn_double_ushort(const void **xpp, size_t nelems, ushort *tp) |
16760 | 0 | { |
16761 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16762 | | |
16763 | | /* basic algorithm is: |
16764 | | * - ensure sane alignment of input data |
16765 | | * - copy (conversion happens automatically) input data |
16766 | | * to output |
16767 | | * - update xpp to point at next unconverted input, and tp to point |
16768 | | * at next location for converted output |
16769 | | */ |
16770 | | long i, j, ni; |
16771 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16772 | | double *xp; |
16773 | | int nrange = 0; /* number of range errors */ |
16774 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16775 | | long cxp = (long) *((char**)xpp); |
16776 | | |
16777 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16778 | | /* sjl: manually stripmine so we can limit amount of |
16779 | | * vector work space reserved to LOOPCNT elements. Also |
16780 | | * makes vectorisation easy */ |
16781 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16782 | | ni=Min(nelems-j,LOOPCNT); |
16783 | | if (realign) { |
16784 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16785 | | xp = tmp; |
16786 | | } else { |
16787 | | xp = (double *) *xpp; |
16788 | | } |
16789 | | /* copy the next block */ |
16790 | | #pragma cdir loopcnt=LOOPCNT |
16791 | | #pragma cdir shortloop |
16792 | | for (i=0; i<ni; i++) { |
16793 | | tp[i] = (ushort) Max( USHORT_MIN, Min(USHORT_MAX, (ushort) xp[i])); |
16794 | | /* test for range errors (not always needed but do it anyway) */ |
16795 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16796 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16797 | | nrange += xp[i] > USHORT_MAX || xp[i] < 0; |
16798 | | } |
16799 | | /* update xpp and tp */ |
16800 | | if (realign) xp = (double *) *xpp; |
16801 | | xp += ni; |
16802 | | tp += ni; |
16803 | | *xpp = (void*)xp; |
16804 | | } |
16805 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16806 | | |
16807 | | #else /* not SX */ |
16808 | 0 | const char *xp = (const char *) *xpp; |
16809 | 0 | int status = NC_NOERR; |
16810 | |
|
16811 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16812 | 0 | { |
16813 | 0 | const int lstatus = ncx_get_double_ushort(xp, tp); |
16814 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16815 | 0 | status = lstatus; |
16816 | 0 | } |
16817 | |
|
16818 | 0 | *xpp = (const void *)xp; |
16819 | 0 | return status; |
16820 | 0 | #endif |
16821 | 0 | } |
16822 | | |
16823 | | int |
16824 | | ncx_getn_double_uint(const void **xpp, size_t nelems, uint *tp) |
16825 | 0 | { |
16826 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16827 | | |
16828 | | /* basic algorithm is: |
16829 | | * - ensure sane alignment of input data |
16830 | | * - copy (conversion happens automatically) input data |
16831 | | * to output |
16832 | | * - update xpp to point at next unconverted input, and tp to point |
16833 | | * at next location for converted output |
16834 | | */ |
16835 | | long i, j, ni; |
16836 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16837 | | double *xp; |
16838 | | int nrange = 0; /* number of range errors */ |
16839 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16840 | | long cxp = (long) *((char**)xpp); |
16841 | | |
16842 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16843 | | /* sjl: manually stripmine so we can limit amount of |
16844 | | * vector work space reserved to LOOPCNT elements. Also |
16845 | | * makes vectorisation easy */ |
16846 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16847 | | ni=Min(nelems-j,LOOPCNT); |
16848 | | if (realign) { |
16849 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16850 | | xp = tmp; |
16851 | | } else { |
16852 | | xp = (double *) *xpp; |
16853 | | } |
16854 | | /* copy the next block */ |
16855 | | #pragma cdir loopcnt=LOOPCNT |
16856 | | #pragma cdir shortloop |
16857 | | for (i=0; i<ni; i++) { |
16858 | | tp[i] = (uint) Max( UINT_MIN, Min(UINT_MAX, (uint) xp[i])); |
16859 | | /* test for range errors (not always needed but do it anyway) */ |
16860 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16861 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16862 | | nrange += xp[i] > UINT_MAX || xp[i] < 0; |
16863 | | } |
16864 | | /* update xpp and tp */ |
16865 | | if (realign) xp = (double *) *xpp; |
16866 | | xp += ni; |
16867 | | tp += ni; |
16868 | | *xpp = (void*)xp; |
16869 | | } |
16870 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16871 | | |
16872 | | #else /* not SX */ |
16873 | 0 | const char *xp = (const char *) *xpp; |
16874 | 0 | int status = NC_NOERR; |
16875 | |
|
16876 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16877 | 0 | { |
16878 | 0 | const int lstatus = ncx_get_double_uint(xp, tp); |
16879 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16880 | 0 | status = lstatus; |
16881 | 0 | } |
16882 | |
|
16883 | 0 | *xpp = (const void *)xp; |
16884 | 0 | return status; |
16885 | 0 | #endif |
16886 | 0 | } |
16887 | | |
16888 | | int |
16889 | | ncx_getn_double_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
16890 | 0 | { |
16891 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
16892 | | |
16893 | | /* basic algorithm is: |
16894 | | * - ensure sane alignment of input data |
16895 | | * - copy (conversion happens automatically) input data |
16896 | | * to output |
16897 | | * - update xpp to point at next unconverted input, and tp to point |
16898 | | * at next location for converted output |
16899 | | */ |
16900 | | long i, j, ni; |
16901 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
16902 | | double *xp; |
16903 | | int nrange = 0; /* number of range errors */ |
16904 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
16905 | | long cxp = (long) *((char**)xpp); |
16906 | | |
16907 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
16908 | | /* sjl: manually stripmine so we can limit amount of |
16909 | | * vector work space reserved to LOOPCNT elements. Also |
16910 | | * makes vectorisation easy */ |
16911 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
16912 | | ni=Min(nelems-j,LOOPCNT); |
16913 | | if (realign) { |
16914 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_DOUBLE)); |
16915 | | xp = tmp; |
16916 | | } else { |
16917 | | xp = (double *) *xpp; |
16918 | | } |
16919 | | /* copy the next block */ |
16920 | | #pragma cdir loopcnt=LOOPCNT |
16921 | | #pragma cdir shortloop |
16922 | | for (i=0; i<ni; i++) { |
16923 | | tp[i] = (ulonglong) Max( ULONGLONG_MIN, Min(ULONGLONG_MAX, (ulonglong) xp[i])); |
16924 | | /* test for range errors (not always needed but do it anyway) */ |
16925 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
16926 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
16927 | | nrange += xp[i] > ULONGLONG_MAX || xp[i] < 0; |
16928 | | } |
16929 | | /* update xpp and tp */ |
16930 | | if (realign) xp = (double *) *xpp; |
16931 | | xp += ni; |
16932 | | tp += ni; |
16933 | | *xpp = (void*)xp; |
16934 | | } |
16935 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
16936 | | |
16937 | | #else /* not SX */ |
16938 | 0 | const char *xp = (const char *) *xpp; |
16939 | 0 | int status = NC_NOERR; |
16940 | |
|
16941 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
16942 | 0 | { |
16943 | 0 | const int lstatus = ncx_get_double_ulonglong(xp, tp); |
16944 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
16945 | 0 | status = lstatus; |
16946 | 0 | } |
16947 | |
|
16948 | 0 | *xpp = (const void *)xp; |
16949 | 0 | return status; |
16950 | 0 | #endif |
16951 | 0 | } |
16952 | | |
16953 | | |
16954 | | #if X_SIZEOF_DOUBLE == SIZEOF_DOUBLE && !defined(NO_IEEE_FLOAT) |
16955 | | /* optimized version */ |
16956 | | int |
16957 | | ncx_putn_double_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
16958 | 0 | { |
16959 | | #ifdef WORDS_BIGENDIAN |
16960 | | (void) memcpy(*xpp, tp, (size_t)nelems * X_SIZEOF_DOUBLE); |
16961 | | # else |
16962 | 0 | swapn8b(*xpp, tp, nelems); |
16963 | 0 | # endif |
16964 | 0 | *xpp = (void *)((char *)(*xpp) + nelems * X_SIZEOF_DOUBLE); |
16965 | 0 | return NC_NOERR; |
16966 | 0 | } |
16967 | | #elif defined(vax) && vax != 0 |
16968 | | int |
16969 | | ncx_putn_double_double(void **xpp, size_t ndoubles, const double *ip, void *fillp) |
16970 | | { |
16971 | | const double *const end = ip + ndoubles; |
16972 | | |
16973 | | while (ip < end) |
16974 | | { |
16975 | | const struct vax_double *const vdp = |
16976 | | (const struct vax_double *)ip; |
16977 | | struct ieee_double *const idp = |
16978 | | (struct ieee_double *) (*xpp); |
16979 | | |
16980 | | if ((vdp->mantissa4 > (dbl_limits[0].d.mantissa4 - 3)) && |
16981 | | (vdp->mantissa3 == dbl_limits[0].d.mantissa3) && |
16982 | | (vdp->mantissa2 == dbl_limits[0].d.mantissa2) && |
16983 | | (vdp->mantissa1 == dbl_limits[0].d.mantissa1) && |
16984 | | (vdp->exp == dbl_limits[0].d.exp)) |
16985 | | { |
16986 | | *idp = dbl_limits[0].ieee; |
16987 | | goto shipit; |
16988 | | } |
16989 | | if ((vdp->mantissa4 == dbl_limits[1].d.mantissa4) && |
16990 | | (vdp->mantissa3 == dbl_limits[1].d.mantissa3) && |
16991 | | (vdp->mantissa2 == dbl_limits[1].d.mantissa2) && |
16992 | | (vdp->mantissa1 == dbl_limits[1].d.mantissa1) && |
16993 | | (vdp->exp == dbl_limits[1].d.exp)) |
16994 | | { |
16995 | | *idp = dbl_limits[1].ieee; |
16996 | | goto shipit; |
16997 | | } |
16998 | | |
16999 | | { |
17000 | | unsigned exp = vdp->exp - VAX_DBL_BIAS + IEEE_DBL_BIAS; |
17001 | | |
17002 | | unsigned mant_lo = ((vdp->mantissa2 & MASK(3)) << 29) | |
17003 | | (vdp->mantissa3 << 13) | |
17004 | | ((vdp->mantissa4 >> 3) & MASK(13)); |
17005 | | |
17006 | | unsigned mant_hi = (vdp->mantissa1 << 13) |
17007 | | | (vdp->mantissa2 >> 3); |
17008 | | |
17009 | | if ((vdp->mantissa4 & 7) > 4) |
17010 | | { |
17011 | | /* round up */ |
17012 | | mant_lo++; |
17013 | | if (mant_lo == 0) |
17014 | | { |
17015 | | mant_hi++; |
17016 | | if (mant_hi > 0xffffff) |
17017 | | { |
17018 | | mant_hi = 0; |
17019 | | exp++; |
17020 | | } |
17021 | | } |
17022 | | } |
17023 | | |
17024 | | idp->mant_lo = SWAP4(mant_lo); |
17025 | | idp->mant_6 = mant_hi >> 16; |
17026 | | idp->mant_5 = (mant_hi & 0xff00) >> 8; |
17027 | | idp->mant_4 = mant_hi; |
17028 | | idp->exp_hi = exp >> 4; |
17029 | | idp->exp_lo = exp; |
17030 | | } |
17031 | | |
17032 | | shipit: |
17033 | | idp->sign = vdp->sign; |
17034 | | |
17035 | | ip++; |
17036 | | *xpp = (char *)(*xpp) + X_SIZEOF_DOUBLE; |
17037 | | } |
17038 | | return NC_NOERR; |
17039 | | } |
17040 | | /* vax */ |
17041 | | #else |
17042 | | int |
17043 | | ncx_putn_double_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
17044 | | { |
17045 | | char *xp = *xpp; |
17046 | | int status = NC_NOERR; |
17047 | | |
17048 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17049 | | { |
17050 | | int lstatus = ncx_put_double_double(xp, tp, fillp); |
17051 | | if (status == NC_NOERR) /* report the first encountered error */ |
17052 | | status = lstatus; |
17053 | | } |
17054 | | |
17055 | | *xpp = (void *)xp; |
17056 | | return status; |
17057 | | } |
17058 | | #endif |
17059 | | int |
17060 | | ncx_putn_double_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
17061 | 0 | { |
17062 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17063 | | |
17064 | | /* basic algorithm is: |
17065 | | * - ensure sane alignment of output data |
17066 | | * - copy (conversion happens automatically) input data |
17067 | | * to output |
17068 | | * - update tp to point at next unconverted input, and xpp to point |
17069 | | * at next location for converted output |
17070 | | */ |
17071 | | long i, j, ni; |
17072 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17073 | | double *xp; |
17074 | | int nrange = 0; /* number of range errors */ |
17075 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17076 | | long cxp = (long) *((char**)xpp); |
17077 | | |
17078 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17079 | | /* sjl: manually stripmine so we can limit amount of |
17080 | | * vector work space reserved to LOOPCNT elements. Also |
17081 | | * makes vectorisation easy */ |
17082 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17083 | | ni=Min(nelems-j,LOOPCNT); |
17084 | | if (realign) { |
17085 | | xp = tmp; |
17086 | | } else { |
17087 | | xp = (double *) *xpp; |
17088 | | } |
17089 | | /* copy the next block */ |
17090 | | #pragma cdir loopcnt=LOOPCNT |
17091 | | #pragma cdir shortloop |
17092 | | for (i=0; i<ni; i++) { |
17093 | | /* the normal case: */ |
17094 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17095 | | /* test for range errors (not always needed but do it anyway) */ |
17096 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17097 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17098 | | nrange += tp[i] > X_DOUBLE_MAX || tp[i] < X_DOUBLE_MIN; |
17099 | | } |
17100 | | /* copy workspace back if necessary */ |
17101 | | if (realign) { |
17102 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17103 | | xp = (double *) *xpp; |
17104 | | } |
17105 | | /* update xpp and tp */ |
17106 | | xp += ni; |
17107 | | tp += ni; |
17108 | | *xpp = (void*)xp; |
17109 | | } |
17110 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17111 | | |
17112 | | #else /* not SX */ |
17113 | |
|
17114 | 0 | char *xp = (char *) *xpp; |
17115 | 0 | int status = NC_NOERR; |
17116 | |
|
17117 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17118 | 0 | { |
17119 | 0 | int lstatus = ncx_put_double_schar(xp, tp, fillp); |
17120 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17121 | 0 | status = lstatus; |
17122 | 0 | } |
17123 | |
|
17124 | 0 | *xpp = (void *)xp; |
17125 | 0 | return status; |
17126 | 0 | #endif |
17127 | 0 | } |
17128 | | |
17129 | | int |
17130 | | ncx_putn_double_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
17131 | 0 | { |
17132 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17133 | | |
17134 | | /* basic algorithm is: |
17135 | | * - ensure sane alignment of output data |
17136 | | * - copy (conversion happens automatically) input data |
17137 | | * to output |
17138 | | * - update tp to point at next unconverted input, and xpp to point |
17139 | | * at next location for converted output |
17140 | | */ |
17141 | | long i, j, ni; |
17142 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17143 | | double *xp; |
17144 | | int nrange = 0; /* number of range errors */ |
17145 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17146 | | long cxp = (long) *((char**)xpp); |
17147 | | |
17148 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17149 | | /* sjl: manually stripmine so we can limit amount of |
17150 | | * vector work space reserved to LOOPCNT elements. Also |
17151 | | * makes vectorisation easy */ |
17152 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17153 | | ni=Min(nelems-j,LOOPCNT); |
17154 | | if (realign) { |
17155 | | xp = tmp; |
17156 | | } else { |
17157 | | xp = (double *) *xpp; |
17158 | | } |
17159 | | /* copy the next block */ |
17160 | | #pragma cdir loopcnt=LOOPCNT |
17161 | | #pragma cdir shortloop |
17162 | | for (i=0; i<ni; i++) { |
17163 | | /* the normal case: */ |
17164 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17165 | | /* test for range errors (not always needed but do it anyway) */ |
17166 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17167 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17168 | | nrange += tp[i] > X_DOUBLE_MAX || tp[i] < X_DOUBLE_MIN; |
17169 | | } |
17170 | | /* copy workspace back if necessary */ |
17171 | | if (realign) { |
17172 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17173 | | xp = (double *) *xpp; |
17174 | | } |
17175 | | /* update xpp and tp */ |
17176 | | xp += ni; |
17177 | | tp += ni; |
17178 | | *xpp = (void*)xp; |
17179 | | } |
17180 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17181 | | |
17182 | | #else /* not SX */ |
17183 | |
|
17184 | 0 | char *xp = (char *) *xpp; |
17185 | 0 | int status = NC_NOERR; |
17186 | |
|
17187 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17188 | 0 | { |
17189 | 0 | int lstatus = ncx_put_double_short(xp, tp, fillp); |
17190 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17191 | 0 | status = lstatus; |
17192 | 0 | } |
17193 | |
|
17194 | 0 | *xpp = (void *)xp; |
17195 | 0 | return status; |
17196 | 0 | #endif |
17197 | 0 | } |
17198 | | |
17199 | | int |
17200 | | ncx_putn_double_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
17201 | 0 | { |
17202 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17203 | | |
17204 | | /* basic algorithm is: |
17205 | | * - ensure sane alignment of output data |
17206 | | * - copy (conversion happens automatically) input data |
17207 | | * to output |
17208 | | * - update tp to point at next unconverted input, and xpp to point |
17209 | | * at next location for converted output |
17210 | | */ |
17211 | | long i, j, ni; |
17212 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17213 | | double *xp; |
17214 | | int nrange = 0; /* number of range errors */ |
17215 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17216 | | long cxp = (long) *((char**)xpp); |
17217 | | |
17218 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17219 | | /* sjl: manually stripmine so we can limit amount of |
17220 | | * vector work space reserved to LOOPCNT elements. Also |
17221 | | * makes vectorisation easy */ |
17222 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17223 | | ni=Min(nelems-j,LOOPCNT); |
17224 | | if (realign) { |
17225 | | xp = tmp; |
17226 | | } else { |
17227 | | xp = (double *) *xpp; |
17228 | | } |
17229 | | /* copy the next block */ |
17230 | | #pragma cdir loopcnt=LOOPCNT |
17231 | | #pragma cdir shortloop |
17232 | | for (i=0; i<ni; i++) { |
17233 | | /* the normal case: */ |
17234 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17235 | | /* test for range errors (not always needed but do it anyway) */ |
17236 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17237 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17238 | | nrange += tp[i] > X_DOUBLE_MAX || tp[i] < X_DOUBLE_MIN; |
17239 | | } |
17240 | | /* copy workspace back if necessary */ |
17241 | | if (realign) { |
17242 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17243 | | xp = (double *) *xpp; |
17244 | | } |
17245 | | /* update xpp and tp */ |
17246 | | xp += ni; |
17247 | | tp += ni; |
17248 | | *xpp = (void*)xp; |
17249 | | } |
17250 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17251 | | |
17252 | | #else /* not SX */ |
17253 | |
|
17254 | 0 | char *xp = (char *) *xpp; |
17255 | 0 | int status = NC_NOERR; |
17256 | |
|
17257 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17258 | 0 | { |
17259 | 0 | int lstatus = ncx_put_double_int(xp, tp, fillp); |
17260 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17261 | 0 | status = lstatus; |
17262 | 0 | } |
17263 | |
|
17264 | 0 | *xpp = (void *)xp; |
17265 | 0 | return status; |
17266 | 0 | #endif |
17267 | 0 | } |
17268 | | |
17269 | | int |
17270 | | ncx_putn_double_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
17271 | 0 | { |
17272 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17273 | | |
17274 | | /* basic algorithm is: |
17275 | | * - ensure sane alignment of output data |
17276 | | * - copy (conversion happens automatically) input data |
17277 | | * to output |
17278 | | * - update tp to point at next unconverted input, and xpp to point |
17279 | | * at next location for converted output |
17280 | | */ |
17281 | | long i, j, ni; |
17282 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17283 | | double *xp; |
17284 | | int nrange = 0; /* number of range errors */ |
17285 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17286 | | long cxp = (long) *((char**)xpp); |
17287 | | |
17288 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17289 | | /* sjl: manually stripmine so we can limit amount of |
17290 | | * vector work space reserved to LOOPCNT elements. Also |
17291 | | * makes vectorisation easy */ |
17292 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17293 | | ni=Min(nelems-j,LOOPCNT); |
17294 | | if (realign) { |
17295 | | xp = tmp; |
17296 | | } else { |
17297 | | xp = (double *) *xpp; |
17298 | | } |
17299 | | /* copy the next block */ |
17300 | | #pragma cdir loopcnt=LOOPCNT |
17301 | | #pragma cdir shortloop |
17302 | | for (i=0; i<ni; i++) { |
17303 | | /* the normal case: */ |
17304 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17305 | | /* test for range errors (not always needed but do it anyway) */ |
17306 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17307 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17308 | | nrange += tp[i] > X_DOUBLE_MAX || tp[i] < X_DOUBLE_MIN; |
17309 | | } |
17310 | | /* copy workspace back if necessary */ |
17311 | | if (realign) { |
17312 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17313 | | xp = (double *) *xpp; |
17314 | | } |
17315 | | /* update xpp and tp */ |
17316 | | xp += ni; |
17317 | | tp += ni; |
17318 | | *xpp = (void*)xp; |
17319 | | } |
17320 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17321 | | |
17322 | | #else /* not SX */ |
17323 | |
|
17324 | 0 | char *xp = (char *) *xpp; |
17325 | 0 | int status = NC_NOERR; |
17326 | |
|
17327 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17328 | 0 | { |
17329 | 0 | int lstatus = ncx_put_double_long(xp, tp, fillp); |
17330 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17331 | 0 | status = lstatus; |
17332 | 0 | } |
17333 | |
|
17334 | 0 | *xpp = (void *)xp; |
17335 | 0 | return status; |
17336 | 0 | #endif |
17337 | 0 | } |
17338 | | |
17339 | | int |
17340 | | ncx_putn_double_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
17341 | 0 | { |
17342 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17343 | | |
17344 | | /* basic algorithm is: |
17345 | | * - ensure sane alignment of output data |
17346 | | * - copy (conversion happens automatically) input data |
17347 | | * to output |
17348 | | * - update tp to point at next unconverted input, and xpp to point |
17349 | | * at next location for converted output |
17350 | | */ |
17351 | | long i, j, ni; |
17352 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17353 | | double *xp; |
17354 | | int nrange = 0; /* number of range errors */ |
17355 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17356 | | long cxp = (long) *((char**)xpp); |
17357 | | |
17358 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17359 | | /* sjl: manually stripmine so we can limit amount of |
17360 | | * vector work space reserved to LOOPCNT elements. Also |
17361 | | * makes vectorisation easy */ |
17362 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17363 | | ni=Min(nelems-j,LOOPCNT); |
17364 | | if (realign) { |
17365 | | xp = tmp; |
17366 | | } else { |
17367 | | xp = (double *) *xpp; |
17368 | | } |
17369 | | /* copy the next block */ |
17370 | | #pragma cdir loopcnt=LOOPCNT |
17371 | | #pragma cdir shortloop |
17372 | | for (i=0; i<ni; i++) { |
17373 | | /* the normal case: */ |
17374 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17375 | | /* test for range errors (not always needed but do it anyway) */ |
17376 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17377 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17378 | | nrange += tp[i] > X_DOUBLE_MAX || tp[i] < X_DOUBLE_MIN; |
17379 | | } |
17380 | | /* copy workspace back if necessary */ |
17381 | | if (realign) { |
17382 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17383 | | xp = (double *) *xpp; |
17384 | | } |
17385 | | /* update xpp and tp */ |
17386 | | xp += ni; |
17387 | | tp += ni; |
17388 | | *xpp = (void*)xp; |
17389 | | } |
17390 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17391 | | |
17392 | | #else /* not SX */ |
17393 | |
|
17394 | 0 | char *xp = (char *) *xpp; |
17395 | 0 | int status = NC_NOERR; |
17396 | |
|
17397 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17398 | 0 | { |
17399 | 0 | int lstatus = ncx_put_double_float(xp, tp, fillp); |
17400 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17401 | 0 | status = lstatus; |
17402 | 0 | } |
17403 | |
|
17404 | 0 | *xpp = (void *)xp; |
17405 | 0 | return status; |
17406 | 0 | #endif |
17407 | 0 | } |
17408 | | |
17409 | | int |
17410 | | ncx_putn_double_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
17411 | 0 | { |
17412 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17413 | | |
17414 | | /* basic algorithm is: |
17415 | | * - ensure sane alignment of output data |
17416 | | * - copy (conversion happens automatically) input data |
17417 | | * to output |
17418 | | * - update tp to point at next unconverted input, and xpp to point |
17419 | | * at next location for converted output |
17420 | | */ |
17421 | | long i, j, ni; |
17422 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17423 | | double *xp; |
17424 | | int nrange = 0; /* number of range errors */ |
17425 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17426 | | long cxp = (long) *((char**)xpp); |
17427 | | |
17428 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17429 | | /* sjl: manually stripmine so we can limit amount of |
17430 | | * vector work space reserved to LOOPCNT elements. Also |
17431 | | * makes vectorisation easy */ |
17432 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17433 | | ni=Min(nelems-j,LOOPCNT); |
17434 | | if (realign) { |
17435 | | xp = tmp; |
17436 | | } else { |
17437 | | xp = (double *) *xpp; |
17438 | | } |
17439 | | /* copy the next block */ |
17440 | | #pragma cdir loopcnt=LOOPCNT |
17441 | | #pragma cdir shortloop |
17442 | | for (i=0; i<ni; i++) { |
17443 | | /* the normal case: */ |
17444 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17445 | | /* test for range errors (not always needed but do it anyway) */ |
17446 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17447 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17448 | | nrange += tp[i] > X_DOUBLE_MAX || tp[i] < X_DOUBLE_MIN; |
17449 | | } |
17450 | | /* copy workspace back if necessary */ |
17451 | | if (realign) { |
17452 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17453 | | xp = (double *) *xpp; |
17454 | | } |
17455 | | /* update xpp and tp */ |
17456 | | xp += ni; |
17457 | | tp += ni; |
17458 | | *xpp = (void*)xp; |
17459 | | } |
17460 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17461 | | |
17462 | | #else /* not SX */ |
17463 | |
|
17464 | 0 | char *xp = (char *) *xpp; |
17465 | 0 | int status = NC_NOERR; |
17466 | |
|
17467 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17468 | 0 | { |
17469 | 0 | int lstatus = ncx_put_double_longlong(xp, tp, fillp); |
17470 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17471 | 0 | status = lstatus; |
17472 | 0 | } |
17473 | |
|
17474 | 0 | *xpp = (void *)xp; |
17475 | 0 | return status; |
17476 | 0 | #endif |
17477 | 0 | } |
17478 | | |
17479 | | int |
17480 | | ncx_putn_double_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
17481 | 0 | { |
17482 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17483 | | |
17484 | | /* basic algorithm is: |
17485 | | * - ensure sane alignment of output data |
17486 | | * - copy (conversion happens automatically) input data |
17487 | | * to output |
17488 | | * - update tp to point at next unconverted input, and xpp to point |
17489 | | * at next location for converted output |
17490 | | */ |
17491 | | long i, j, ni; |
17492 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17493 | | double *xp; |
17494 | | int nrange = 0; /* number of range errors */ |
17495 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17496 | | long cxp = (long) *((char**)xpp); |
17497 | | |
17498 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17499 | | /* sjl: manually stripmine so we can limit amount of |
17500 | | * vector work space reserved to LOOPCNT elements. Also |
17501 | | * makes vectorisation easy */ |
17502 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17503 | | ni=Min(nelems-j,LOOPCNT); |
17504 | | if (realign) { |
17505 | | xp = tmp; |
17506 | | } else { |
17507 | | xp = (double *) *xpp; |
17508 | | } |
17509 | | /* copy the next block */ |
17510 | | #pragma cdir loopcnt=LOOPCNT |
17511 | | #pragma cdir shortloop |
17512 | | for (i=0; i<ni; i++) { |
17513 | | /* the normal case: */ |
17514 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17515 | | /* test for range errors (not always needed but do it anyway) */ |
17516 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17517 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17518 | | nrange += tp[i] > X_DOUBLE_MAX ; |
17519 | | } |
17520 | | /* copy workspace back if necessary */ |
17521 | | if (realign) { |
17522 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17523 | | xp = (double *) *xpp; |
17524 | | } |
17525 | | /* update xpp and tp */ |
17526 | | xp += ni; |
17527 | | tp += ni; |
17528 | | *xpp = (void*)xp; |
17529 | | } |
17530 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17531 | | |
17532 | | #else /* not SX */ |
17533 | |
|
17534 | 0 | char *xp = (char *) *xpp; |
17535 | 0 | int status = NC_NOERR; |
17536 | |
|
17537 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17538 | 0 | { |
17539 | 0 | int lstatus = ncx_put_double_uchar(xp, tp, fillp); |
17540 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17541 | 0 | status = lstatus; |
17542 | 0 | } |
17543 | |
|
17544 | 0 | *xpp = (void *)xp; |
17545 | 0 | return status; |
17546 | 0 | #endif |
17547 | 0 | } |
17548 | | |
17549 | | int |
17550 | | ncx_putn_double_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
17551 | 0 | { |
17552 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17553 | | |
17554 | | /* basic algorithm is: |
17555 | | * - ensure sane alignment of output data |
17556 | | * - copy (conversion happens automatically) input data |
17557 | | * to output |
17558 | | * - update tp to point at next unconverted input, and xpp to point |
17559 | | * at next location for converted output |
17560 | | */ |
17561 | | long i, j, ni; |
17562 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17563 | | double *xp; |
17564 | | int nrange = 0; /* number of range errors */ |
17565 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17566 | | long cxp = (long) *((char**)xpp); |
17567 | | |
17568 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17569 | | /* sjl: manually stripmine so we can limit amount of |
17570 | | * vector work space reserved to LOOPCNT elements. Also |
17571 | | * makes vectorisation easy */ |
17572 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17573 | | ni=Min(nelems-j,LOOPCNT); |
17574 | | if (realign) { |
17575 | | xp = tmp; |
17576 | | } else { |
17577 | | xp = (double *) *xpp; |
17578 | | } |
17579 | | /* copy the next block */ |
17580 | | #pragma cdir loopcnt=LOOPCNT |
17581 | | #pragma cdir shortloop |
17582 | | for (i=0; i<ni; i++) { |
17583 | | /* the normal case: */ |
17584 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17585 | | /* test for range errors (not always needed but do it anyway) */ |
17586 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17587 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17588 | | nrange += tp[i] > X_DOUBLE_MAX ; |
17589 | | } |
17590 | | /* copy workspace back if necessary */ |
17591 | | if (realign) { |
17592 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17593 | | xp = (double *) *xpp; |
17594 | | } |
17595 | | /* update xpp and tp */ |
17596 | | xp += ni; |
17597 | | tp += ni; |
17598 | | *xpp = (void*)xp; |
17599 | | } |
17600 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17601 | | |
17602 | | #else /* not SX */ |
17603 | |
|
17604 | 0 | char *xp = (char *) *xpp; |
17605 | 0 | int status = NC_NOERR; |
17606 | |
|
17607 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17608 | 0 | { |
17609 | 0 | int lstatus = ncx_put_double_ushort(xp, tp, fillp); |
17610 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17611 | 0 | status = lstatus; |
17612 | 0 | } |
17613 | |
|
17614 | 0 | *xpp = (void *)xp; |
17615 | 0 | return status; |
17616 | 0 | #endif |
17617 | 0 | } |
17618 | | |
17619 | | int |
17620 | | ncx_putn_double_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
17621 | 0 | { |
17622 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17623 | | |
17624 | | /* basic algorithm is: |
17625 | | * - ensure sane alignment of output data |
17626 | | * - copy (conversion happens automatically) input data |
17627 | | * to output |
17628 | | * - update tp to point at next unconverted input, and xpp to point |
17629 | | * at next location for converted output |
17630 | | */ |
17631 | | long i, j, ni; |
17632 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17633 | | double *xp; |
17634 | | int nrange = 0; /* number of range errors */ |
17635 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17636 | | long cxp = (long) *((char**)xpp); |
17637 | | |
17638 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17639 | | /* sjl: manually stripmine so we can limit amount of |
17640 | | * vector work space reserved to LOOPCNT elements. Also |
17641 | | * makes vectorisation easy */ |
17642 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17643 | | ni=Min(nelems-j,LOOPCNT); |
17644 | | if (realign) { |
17645 | | xp = tmp; |
17646 | | } else { |
17647 | | xp = (double *) *xpp; |
17648 | | } |
17649 | | /* copy the next block */ |
17650 | | #pragma cdir loopcnt=LOOPCNT |
17651 | | #pragma cdir shortloop |
17652 | | for (i=0; i<ni; i++) { |
17653 | | /* the normal case: */ |
17654 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17655 | | /* test for range errors (not always needed but do it anyway) */ |
17656 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17657 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17658 | | nrange += tp[i] > X_DOUBLE_MAX ; |
17659 | | } |
17660 | | /* copy workspace back if necessary */ |
17661 | | if (realign) { |
17662 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17663 | | xp = (double *) *xpp; |
17664 | | } |
17665 | | /* update xpp and tp */ |
17666 | | xp += ni; |
17667 | | tp += ni; |
17668 | | *xpp = (void*)xp; |
17669 | | } |
17670 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17671 | | |
17672 | | #else /* not SX */ |
17673 | |
|
17674 | 0 | char *xp = (char *) *xpp; |
17675 | 0 | int status = NC_NOERR; |
17676 | |
|
17677 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17678 | 0 | { |
17679 | 0 | int lstatus = ncx_put_double_uint(xp, tp, fillp); |
17680 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17681 | 0 | status = lstatus; |
17682 | 0 | } |
17683 | |
|
17684 | 0 | *xpp = (void *)xp; |
17685 | 0 | return status; |
17686 | 0 | #endif |
17687 | 0 | } |
17688 | | |
17689 | | int |
17690 | | ncx_putn_double_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
17691 | 0 | { |
17692 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_DOUBLE == SIZEOF_DOUBLE |
17693 | | |
17694 | | /* basic algorithm is: |
17695 | | * - ensure sane alignment of output data |
17696 | | * - copy (conversion happens automatically) input data |
17697 | | * to output |
17698 | | * - update tp to point at next unconverted input, and xpp to point |
17699 | | * at next location for converted output |
17700 | | */ |
17701 | | long i, j, ni; |
17702 | | double tmp[LOOPCNT]; /* in case input is misaligned */ |
17703 | | double *xp; |
17704 | | int nrange = 0; /* number of range errors */ |
17705 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17706 | | long cxp = (long) *((char**)xpp); |
17707 | | |
17708 | | realign = (cxp & 7) % SIZEOF_DOUBLE; |
17709 | | /* sjl: manually stripmine so we can limit amount of |
17710 | | * vector work space reserved to LOOPCNT elements. Also |
17711 | | * makes vectorisation easy */ |
17712 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17713 | | ni=Min(nelems-j,LOOPCNT); |
17714 | | if (realign) { |
17715 | | xp = tmp; |
17716 | | } else { |
17717 | | xp = (double *) *xpp; |
17718 | | } |
17719 | | /* copy the next block */ |
17720 | | #pragma cdir loopcnt=LOOPCNT |
17721 | | #pragma cdir shortloop |
17722 | | for (i=0; i<ni; i++) { |
17723 | | /* the normal case: */ |
17724 | | xp[i] = (double) Max( X_DOUBLE_MIN, Min(X_DOUBLE_MAX, (double) tp[i])); |
17725 | | /* test for range errors (not always needed but do it anyway) */ |
17726 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
17727 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
17728 | | nrange += tp[i] > X_DOUBLE_MAX ; |
17729 | | } |
17730 | | /* copy workspace back if necessary */ |
17731 | | if (realign) { |
17732 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_DOUBLE); |
17733 | | xp = (double *) *xpp; |
17734 | | } |
17735 | | /* update xpp and tp */ |
17736 | | xp += ni; |
17737 | | tp += ni; |
17738 | | *xpp = (void*)xp; |
17739 | | } |
17740 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17741 | | |
17742 | | #else /* not SX */ |
17743 | |
|
17744 | 0 | char *xp = (char *) *xpp; |
17745 | 0 | int status = NC_NOERR; |
17746 | |
|
17747 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_DOUBLE, tp++) |
17748 | 0 | { |
17749 | 0 | int lstatus = ncx_put_double_ulonglong(xp, tp, fillp); |
17750 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17751 | 0 | status = lstatus; |
17752 | 0 | } |
17753 | |
|
17754 | 0 | *xpp = (void *)xp; |
17755 | 0 | return status; |
17756 | 0 | #endif |
17757 | 0 | } |
17758 | | |
17759 | | |
17760 | | |
17761 | | /* longlong ------------------------------------------------------------------*/ |
17762 | | |
17763 | | #if X_SIZEOF_INT64 == SIZEOF_LONGLONG |
17764 | | /* optimized version */ |
17765 | | int |
17766 | | ncx_getn_longlong_longlong(const void **xpp, size_t nelems, long long *tp) |
17767 | 0 | { |
17768 | | #ifdef WORDS_BIGENDIAN |
17769 | | (void) memcpy(tp, *xpp, (size_t)nelems * SIZEOF_LONG_LONG); |
17770 | | # else |
17771 | 0 | swapn8b(tp, *xpp, nelems); |
17772 | 0 | # endif |
17773 | 0 | *xpp = (const void *)((const char *)(*xpp) + nelems * X_SIZEOF_INT64); |
17774 | 0 | return NC_NOERR; |
17775 | 0 | } |
17776 | | #else |
17777 | | int |
17778 | | ncx_getn_longlong_longlong(const void **xpp, size_t nelems, longlong *tp) |
17779 | | { |
17780 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
17781 | | |
17782 | | /* basic algorithm is: |
17783 | | * - ensure sane alignment of input data |
17784 | | * - copy (conversion happens automatically) input data |
17785 | | * to output |
17786 | | * - update xpp to point at next unconverted input, and tp to point |
17787 | | * at next location for converted output |
17788 | | */ |
17789 | | long i, j, ni; |
17790 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
17791 | | int64 *xp; |
17792 | | int nrange = 0; /* number of range errors */ |
17793 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17794 | | long cxp = (long) *((char**)xpp); |
17795 | | |
17796 | | realign = (cxp & 7) % SIZEOF_INT64; |
17797 | | /* sjl: manually stripmine so we can limit amount of |
17798 | | * vector work space reserved to LOOPCNT elements. Also |
17799 | | * makes vectorisation easy */ |
17800 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17801 | | ni=Min(nelems-j,LOOPCNT); |
17802 | | if (realign) { |
17803 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
17804 | | xp = tmp; |
17805 | | } else { |
17806 | | xp = (int64 *) *xpp; |
17807 | | } |
17808 | | /* copy the next block */ |
17809 | | #pragma cdir loopcnt=LOOPCNT |
17810 | | #pragma cdir shortloop |
17811 | | for (i=0; i<ni; i++) { |
17812 | | tp[i] = (longlong) Max( LONGLONG_MIN, Min(LONGLONG_MAX, (longlong) xp[i])); |
17813 | | /* test for range errors (not always needed but do it anyway) */ |
17814 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
17815 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
17816 | | nrange += xp[i] > LONGLONG_MAX || xp[i] < LONGLONG_MIN; |
17817 | | } |
17818 | | /* update xpp and tp */ |
17819 | | if (realign) xp = (int64 *) *xpp; |
17820 | | xp += ni; |
17821 | | tp += ni; |
17822 | | *xpp = (void*)xp; |
17823 | | } |
17824 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17825 | | |
17826 | | #else /* not SX */ |
17827 | | const char *xp = (const char *) *xpp; |
17828 | | int status = NC_NOERR; |
17829 | | |
17830 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
17831 | | { |
17832 | | const int lstatus = ncx_get_longlong_longlong(xp, tp); |
17833 | | if (status == NC_NOERR) /* report the first encountered error */ |
17834 | | status = lstatus; |
17835 | | } |
17836 | | |
17837 | | *xpp = (const void *)xp; |
17838 | | return status; |
17839 | | #endif |
17840 | | } |
17841 | | |
17842 | | #endif |
17843 | | int |
17844 | | ncx_getn_longlong_schar(const void **xpp, size_t nelems, schar *tp) |
17845 | 0 | { |
17846 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
17847 | | |
17848 | | /* basic algorithm is: |
17849 | | * - ensure sane alignment of input data |
17850 | | * - copy (conversion happens automatically) input data |
17851 | | * to output |
17852 | | * - update xpp to point at next unconverted input, and tp to point |
17853 | | * at next location for converted output |
17854 | | */ |
17855 | | long i, j, ni; |
17856 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
17857 | | int64 *xp; |
17858 | | int nrange = 0; /* number of range errors */ |
17859 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17860 | | long cxp = (long) *((char**)xpp); |
17861 | | |
17862 | | realign = (cxp & 7) % SIZEOF_INT64; |
17863 | | /* sjl: manually stripmine so we can limit amount of |
17864 | | * vector work space reserved to LOOPCNT elements. Also |
17865 | | * makes vectorisation easy */ |
17866 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17867 | | ni=Min(nelems-j,LOOPCNT); |
17868 | | if (realign) { |
17869 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
17870 | | xp = tmp; |
17871 | | } else { |
17872 | | xp = (int64 *) *xpp; |
17873 | | } |
17874 | | /* copy the next block */ |
17875 | | #pragma cdir loopcnt=LOOPCNT |
17876 | | #pragma cdir shortloop |
17877 | | for (i=0; i<ni; i++) { |
17878 | | tp[i] = (schar) Max( SCHAR_MIN, Min(SCHAR_MAX, (schar) xp[i])); |
17879 | | /* test for range errors (not always needed but do it anyway) */ |
17880 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
17881 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
17882 | | nrange += xp[i] > SCHAR_MAX || xp[i] < SCHAR_MIN; |
17883 | | } |
17884 | | /* update xpp and tp */ |
17885 | | if (realign) xp = (int64 *) *xpp; |
17886 | | xp += ni; |
17887 | | tp += ni; |
17888 | | *xpp = (void*)xp; |
17889 | | } |
17890 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17891 | | |
17892 | | #else /* not SX */ |
17893 | 0 | const char *xp = (const char *) *xpp; |
17894 | 0 | int status = NC_NOERR; |
17895 | |
|
17896 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
17897 | 0 | { |
17898 | 0 | const int lstatus = ncx_get_longlong_schar(xp, tp); |
17899 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17900 | 0 | status = lstatus; |
17901 | 0 | } |
17902 | |
|
17903 | 0 | *xpp = (const void *)xp; |
17904 | 0 | return status; |
17905 | 0 | #endif |
17906 | 0 | } |
17907 | | |
17908 | | int |
17909 | | ncx_getn_longlong_short(const void **xpp, size_t nelems, short *tp) |
17910 | 0 | { |
17911 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
17912 | | |
17913 | | /* basic algorithm is: |
17914 | | * - ensure sane alignment of input data |
17915 | | * - copy (conversion happens automatically) input data |
17916 | | * to output |
17917 | | * - update xpp to point at next unconverted input, and tp to point |
17918 | | * at next location for converted output |
17919 | | */ |
17920 | | long i, j, ni; |
17921 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
17922 | | int64 *xp; |
17923 | | int nrange = 0; /* number of range errors */ |
17924 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17925 | | long cxp = (long) *((char**)xpp); |
17926 | | |
17927 | | realign = (cxp & 7) % SIZEOF_INT64; |
17928 | | /* sjl: manually stripmine so we can limit amount of |
17929 | | * vector work space reserved to LOOPCNT elements. Also |
17930 | | * makes vectorisation easy */ |
17931 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17932 | | ni=Min(nelems-j,LOOPCNT); |
17933 | | if (realign) { |
17934 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
17935 | | xp = tmp; |
17936 | | } else { |
17937 | | xp = (int64 *) *xpp; |
17938 | | } |
17939 | | /* copy the next block */ |
17940 | | #pragma cdir loopcnt=LOOPCNT |
17941 | | #pragma cdir shortloop |
17942 | | for (i=0; i<ni; i++) { |
17943 | | tp[i] = (short) Max( SHORT_MIN, Min(SHORT_MAX, (short) xp[i])); |
17944 | | /* test for range errors (not always needed but do it anyway) */ |
17945 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
17946 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
17947 | | nrange += xp[i] > SHORT_MAX || xp[i] < SHORT_MIN; |
17948 | | } |
17949 | | /* update xpp and tp */ |
17950 | | if (realign) xp = (int64 *) *xpp; |
17951 | | xp += ni; |
17952 | | tp += ni; |
17953 | | *xpp = (void*)xp; |
17954 | | } |
17955 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
17956 | | |
17957 | | #else /* not SX */ |
17958 | 0 | const char *xp = (const char *) *xpp; |
17959 | 0 | int status = NC_NOERR; |
17960 | |
|
17961 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
17962 | 0 | { |
17963 | 0 | const int lstatus = ncx_get_longlong_short(xp, tp); |
17964 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
17965 | 0 | status = lstatus; |
17966 | 0 | } |
17967 | |
|
17968 | 0 | *xpp = (const void *)xp; |
17969 | 0 | return status; |
17970 | 0 | #endif |
17971 | 0 | } |
17972 | | |
17973 | | int |
17974 | | ncx_getn_longlong_int(const void **xpp, size_t nelems, int *tp) |
17975 | 0 | { |
17976 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
17977 | | |
17978 | | /* basic algorithm is: |
17979 | | * - ensure sane alignment of input data |
17980 | | * - copy (conversion happens automatically) input data |
17981 | | * to output |
17982 | | * - update xpp to point at next unconverted input, and tp to point |
17983 | | * at next location for converted output |
17984 | | */ |
17985 | | long i, j, ni; |
17986 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
17987 | | int64 *xp; |
17988 | | int nrange = 0; /* number of range errors */ |
17989 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
17990 | | long cxp = (long) *((char**)xpp); |
17991 | | |
17992 | | realign = (cxp & 7) % SIZEOF_INT64; |
17993 | | /* sjl: manually stripmine so we can limit amount of |
17994 | | * vector work space reserved to LOOPCNT elements. Also |
17995 | | * makes vectorisation easy */ |
17996 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
17997 | | ni=Min(nelems-j,LOOPCNT); |
17998 | | if (realign) { |
17999 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
18000 | | xp = tmp; |
18001 | | } else { |
18002 | | xp = (int64 *) *xpp; |
18003 | | } |
18004 | | /* copy the next block */ |
18005 | | #pragma cdir loopcnt=LOOPCNT |
18006 | | #pragma cdir shortloop |
18007 | | for (i=0; i<ni; i++) { |
18008 | | tp[i] = (int) Max( INT_MIN, Min(INT_MAX, (int) xp[i])); |
18009 | | /* test for range errors (not always needed but do it anyway) */ |
18010 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
18011 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
18012 | | nrange += xp[i] > INT_MAX || xp[i] < INT_MIN; |
18013 | | } |
18014 | | /* update xpp and tp */ |
18015 | | if (realign) xp = (int64 *) *xpp; |
18016 | | xp += ni; |
18017 | | tp += ni; |
18018 | | *xpp = (void*)xp; |
18019 | | } |
18020 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18021 | | |
18022 | | #else /* not SX */ |
18023 | 0 | const char *xp = (const char *) *xpp; |
18024 | 0 | int status = NC_NOERR; |
18025 | |
|
18026 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18027 | 0 | { |
18028 | 0 | const int lstatus = ncx_get_longlong_int(xp, tp); |
18029 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18030 | 0 | status = lstatus; |
18031 | 0 | } |
18032 | |
|
18033 | 0 | *xpp = (const void *)xp; |
18034 | 0 | return status; |
18035 | 0 | #endif |
18036 | 0 | } |
18037 | | |
18038 | | int |
18039 | | ncx_getn_longlong_long(const void **xpp, size_t nelems, long *tp) |
18040 | 0 | { |
18041 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18042 | | |
18043 | | /* basic algorithm is: |
18044 | | * - ensure sane alignment of input data |
18045 | | * - copy (conversion happens automatically) input data |
18046 | | * to output |
18047 | | * - update xpp to point at next unconverted input, and tp to point |
18048 | | * at next location for converted output |
18049 | | */ |
18050 | | long i, j, ni; |
18051 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18052 | | int64 *xp; |
18053 | | int nrange = 0; /* number of range errors */ |
18054 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18055 | | long cxp = (long) *((char**)xpp); |
18056 | | |
18057 | | realign = (cxp & 7) % SIZEOF_INT64; |
18058 | | /* sjl: manually stripmine so we can limit amount of |
18059 | | * vector work space reserved to LOOPCNT elements. Also |
18060 | | * makes vectorisation easy */ |
18061 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18062 | | ni=Min(nelems-j,LOOPCNT); |
18063 | | if (realign) { |
18064 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
18065 | | xp = tmp; |
18066 | | } else { |
18067 | | xp = (int64 *) *xpp; |
18068 | | } |
18069 | | /* copy the next block */ |
18070 | | #pragma cdir loopcnt=LOOPCNT |
18071 | | #pragma cdir shortloop |
18072 | | for (i=0; i<ni; i++) { |
18073 | | tp[i] = (long) Max( LONG_MIN, Min(LONG_MAX, (long) xp[i])); |
18074 | | /* test for range errors (not always needed but do it anyway) */ |
18075 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
18076 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
18077 | | nrange += xp[i] > LONG_MAX || xp[i] < LONG_MIN; |
18078 | | } |
18079 | | /* update xpp and tp */ |
18080 | | if (realign) xp = (int64 *) *xpp; |
18081 | | xp += ni; |
18082 | | tp += ni; |
18083 | | *xpp = (void*)xp; |
18084 | | } |
18085 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18086 | | |
18087 | | #else /* not SX */ |
18088 | 0 | const char *xp = (const char *) *xpp; |
18089 | 0 | int status = NC_NOERR; |
18090 | |
|
18091 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18092 | 0 | { |
18093 | 0 | const int lstatus = ncx_get_longlong_long(xp, tp); |
18094 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18095 | 0 | status = lstatus; |
18096 | 0 | } |
18097 | |
|
18098 | 0 | *xpp = (const void *)xp; |
18099 | 0 | return status; |
18100 | 0 | #endif |
18101 | 0 | } |
18102 | | |
18103 | | int |
18104 | | ncx_getn_longlong_float(const void **xpp, size_t nelems, float *tp) |
18105 | 0 | { |
18106 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18107 | | |
18108 | | /* basic algorithm is: |
18109 | | * - ensure sane alignment of input data |
18110 | | * - copy (conversion happens automatically) input data |
18111 | | * to output |
18112 | | * - update xpp to point at next unconverted input, and tp to point |
18113 | | * at next location for converted output |
18114 | | */ |
18115 | | long i, j, ni; |
18116 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18117 | | int64 *xp; |
18118 | | int nrange = 0; /* number of range errors */ |
18119 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18120 | | long cxp = (long) *((char**)xpp); |
18121 | | |
18122 | | realign = (cxp & 7) % SIZEOF_INT64; |
18123 | | /* sjl: manually stripmine so we can limit amount of |
18124 | | * vector work space reserved to LOOPCNT elements. Also |
18125 | | * makes vectorisation easy */ |
18126 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18127 | | ni=Min(nelems-j,LOOPCNT); |
18128 | | if (realign) { |
18129 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
18130 | | xp = tmp; |
18131 | | } else { |
18132 | | xp = (int64 *) *xpp; |
18133 | | } |
18134 | | /* copy the next block */ |
18135 | | #pragma cdir loopcnt=LOOPCNT |
18136 | | #pragma cdir shortloop |
18137 | | for (i=0; i<ni; i++) { |
18138 | | tp[i] = (float) Max( FLOAT_MIN, Min(FLOAT_MAX, (float) xp[i])); |
18139 | | /* test for range errors (not always needed but do it anyway) */ |
18140 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
18141 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
18142 | | nrange += xp[i] > FLOAT_MAX || xp[i] < FLOAT_MIN; |
18143 | | } |
18144 | | /* update xpp and tp */ |
18145 | | if (realign) xp = (int64 *) *xpp; |
18146 | | xp += ni; |
18147 | | tp += ni; |
18148 | | *xpp = (void*)xp; |
18149 | | } |
18150 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18151 | | |
18152 | | #else /* not SX */ |
18153 | 0 | const char *xp = (const char *) *xpp; |
18154 | 0 | int status = NC_NOERR; |
18155 | |
|
18156 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18157 | 0 | { |
18158 | 0 | const int lstatus = ncx_get_longlong_float(xp, tp); |
18159 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18160 | 0 | status = lstatus; |
18161 | 0 | } |
18162 | |
|
18163 | 0 | *xpp = (const void *)xp; |
18164 | 0 | return status; |
18165 | 0 | #endif |
18166 | 0 | } |
18167 | | |
18168 | | int |
18169 | | ncx_getn_longlong_double(const void **xpp, size_t nelems, double *tp) |
18170 | 0 | { |
18171 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18172 | | |
18173 | | /* basic algorithm is: |
18174 | | * - ensure sane alignment of input data |
18175 | | * - copy (conversion happens automatically) input data |
18176 | | * to output |
18177 | | * - update xpp to point at next unconverted input, and tp to point |
18178 | | * at next location for converted output |
18179 | | */ |
18180 | | long i, j, ni; |
18181 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18182 | | int64 *xp; |
18183 | | int nrange = 0; /* number of range errors */ |
18184 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18185 | | long cxp = (long) *((char**)xpp); |
18186 | | |
18187 | | realign = (cxp & 7) % SIZEOF_INT64; |
18188 | | /* sjl: manually stripmine so we can limit amount of |
18189 | | * vector work space reserved to LOOPCNT elements. Also |
18190 | | * makes vectorisation easy */ |
18191 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18192 | | ni=Min(nelems-j,LOOPCNT); |
18193 | | if (realign) { |
18194 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
18195 | | xp = tmp; |
18196 | | } else { |
18197 | | xp = (int64 *) *xpp; |
18198 | | } |
18199 | | /* copy the next block */ |
18200 | | #pragma cdir loopcnt=LOOPCNT |
18201 | | #pragma cdir shortloop |
18202 | | for (i=0; i<ni; i++) { |
18203 | | tp[i] = (double) Max( DOUBLE_MIN, Min(DOUBLE_MAX, (double) xp[i])); |
18204 | | /* test for range errors (not always needed but do it anyway) */ |
18205 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
18206 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
18207 | | nrange += xp[i] > DOUBLE_MAX || xp[i] < DOUBLE_MIN; |
18208 | | } |
18209 | | /* update xpp and tp */ |
18210 | | if (realign) xp = (int64 *) *xpp; |
18211 | | xp += ni; |
18212 | | tp += ni; |
18213 | | *xpp = (void*)xp; |
18214 | | } |
18215 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18216 | | |
18217 | | #else /* not SX */ |
18218 | 0 | const char *xp = (const char *) *xpp; |
18219 | 0 | int status = NC_NOERR; |
18220 | |
|
18221 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18222 | 0 | { |
18223 | 0 | const int lstatus = ncx_get_longlong_double(xp, tp); |
18224 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18225 | 0 | status = lstatus; |
18226 | 0 | } |
18227 | |
|
18228 | 0 | *xpp = (const void *)xp; |
18229 | 0 | return status; |
18230 | 0 | #endif |
18231 | 0 | } |
18232 | | |
18233 | | int |
18234 | | ncx_getn_longlong_uchar(const void **xpp, size_t nelems, uchar *tp) |
18235 | 0 | { |
18236 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18237 | | |
18238 | | /* basic algorithm is: |
18239 | | * - ensure sane alignment of input data |
18240 | | * - copy (conversion happens automatically) input data |
18241 | | * to output |
18242 | | * - update xpp to point at next unconverted input, and tp to point |
18243 | | * at next location for converted output |
18244 | | */ |
18245 | | long i, j, ni; |
18246 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18247 | | int64 *xp; |
18248 | | int nrange = 0; /* number of range errors */ |
18249 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18250 | | long cxp = (long) *((char**)xpp); |
18251 | | |
18252 | | realign = (cxp & 7) % SIZEOF_INT64; |
18253 | | /* sjl: manually stripmine so we can limit amount of |
18254 | | * vector work space reserved to LOOPCNT elements. Also |
18255 | | * makes vectorisation easy */ |
18256 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18257 | | ni=Min(nelems-j,LOOPCNT); |
18258 | | if (realign) { |
18259 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
18260 | | xp = tmp; |
18261 | | } else { |
18262 | | xp = (int64 *) *xpp; |
18263 | | } |
18264 | | /* copy the next block */ |
18265 | | #pragma cdir loopcnt=LOOPCNT |
18266 | | #pragma cdir shortloop |
18267 | | for (i=0; i<ni; i++) { |
18268 | | tp[i] = (uchar) Max( UCHAR_MIN, Min(UCHAR_MAX, (uchar) xp[i])); |
18269 | | /* test for range errors (not always needed but do it anyway) */ |
18270 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
18271 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
18272 | | nrange += xp[i] > UCHAR_MAX || xp[i] < 0; |
18273 | | } |
18274 | | /* update xpp and tp */ |
18275 | | if (realign) xp = (int64 *) *xpp; |
18276 | | xp += ni; |
18277 | | tp += ni; |
18278 | | *xpp = (void*)xp; |
18279 | | } |
18280 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18281 | | |
18282 | | #else /* not SX */ |
18283 | 0 | const char *xp = (const char *) *xpp; |
18284 | 0 | int status = NC_NOERR; |
18285 | |
|
18286 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18287 | 0 | { |
18288 | 0 | const int lstatus = ncx_get_longlong_uchar(xp, tp); |
18289 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18290 | 0 | status = lstatus; |
18291 | 0 | } |
18292 | |
|
18293 | 0 | *xpp = (const void *)xp; |
18294 | 0 | return status; |
18295 | 0 | #endif |
18296 | 0 | } |
18297 | | |
18298 | | int |
18299 | | ncx_getn_longlong_ushort(const void **xpp, size_t nelems, ushort *tp) |
18300 | 0 | { |
18301 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18302 | | |
18303 | | /* basic algorithm is: |
18304 | | * - ensure sane alignment of input data |
18305 | | * - copy (conversion happens automatically) input data |
18306 | | * to output |
18307 | | * - update xpp to point at next unconverted input, and tp to point |
18308 | | * at next location for converted output |
18309 | | */ |
18310 | | long i, j, ni; |
18311 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18312 | | int64 *xp; |
18313 | | int nrange = 0; /* number of range errors */ |
18314 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18315 | | long cxp = (long) *((char**)xpp); |
18316 | | |
18317 | | realign = (cxp & 7) % SIZEOF_INT64; |
18318 | | /* sjl: manually stripmine so we can limit amount of |
18319 | | * vector work space reserved to LOOPCNT elements. Also |
18320 | | * makes vectorisation easy */ |
18321 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18322 | | ni=Min(nelems-j,LOOPCNT); |
18323 | | if (realign) { |
18324 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
18325 | | xp = tmp; |
18326 | | } else { |
18327 | | xp = (int64 *) *xpp; |
18328 | | } |
18329 | | /* copy the next block */ |
18330 | | #pragma cdir loopcnt=LOOPCNT |
18331 | | #pragma cdir shortloop |
18332 | | for (i=0; i<ni; i++) { |
18333 | | tp[i] = (ushort) Max( USHORT_MIN, Min(USHORT_MAX, (ushort) xp[i])); |
18334 | | /* test for range errors (not always needed but do it anyway) */ |
18335 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
18336 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
18337 | | nrange += xp[i] > USHORT_MAX || xp[i] < 0; |
18338 | | } |
18339 | | /* update xpp and tp */ |
18340 | | if (realign) xp = (int64 *) *xpp; |
18341 | | xp += ni; |
18342 | | tp += ni; |
18343 | | *xpp = (void*)xp; |
18344 | | } |
18345 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18346 | | |
18347 | | #else /* not SX */ |
18348 | 0 | const char *xp = (const char *) *xpp; |
18349 | 0 | int status = NC_NOERR; |
18350 | |
|
18351 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18352 | 0 | { |
18353 | 0 | const int lstatus = ncx_get_longlong_ushort(xp, tp); |
18354 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18355 | 0 | status = lstatus; |
18356 | 0 | } |
18357 | |
|
18358 | 0 | *xpp = (const void *)xp; |
18359 | 0 | return status; |
18360 | 0 | #endif |
18361 | 0 | } |
18362 | | |
18363 | | int |
18364 | | ncx_getn_longlong_uint(const void **xpp, size_t nelems, uint *tp) |
18365 | 0 | { |
18366 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18367 | | |
18368 | | /* basic algorithm is: |
18369 | | * - ensure sane alignment of input data |
18370 | | * - copy (conversion happens automatically) input data |
18371 | | * to output |
18372 | | * - update xpp to point at next unconverted input, and tp to point |
18373 | | * at next location for converted output |
18374 | | */ |
18375 | | long i, j, ni; |
18376 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18377 | | int64 *xp; |
18378 | | int nrange = 0; /* number of range errors */ |
18379 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18380 | | long cxp = (long) *((char**)xpp); |
18381 | | |
18382 | | realign = (cxp & 7) % SIZEOF_INT64; |
18383 | | /* sjl: manually stripmine so we can limit amount of |
18384 | | * vector work space reserved to LOOPCNT elements. Also |
18385 | | * makes vectorisation easy */ |
18386 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18387 | | ni=Min(nelems-j,LOOPCNT); |
18388 | | if (realign) { |
18389 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
18390 | | xp = tmp; |
18391 | | } else { |
18392 | | xp = (int64 *) *xpp; |
18393 | | } |
18394 | | /* copy the next block */ |
18395 | | #pragma cdir loopcnt=LOOPCNT |
18396 | | #pragma cdir shortloop |
18397 | | for (i=0; i<ni; i++) { |
18398 | | tp[i] = (uint) Max( UINT_MIN, Min(UINT_MAX, (uint) xp[i])); |
18399 | | /* test for range errors (not always needed but do it anyway) */ |
18400 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
18401 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
18402 | | nrange += xp[i] > UINT_MAX || xp[i] < 0; |
18403 | | } |
18404 | | /* update xpp and tp */ |
18405 | | if (realign) xp = (int64 *) *xpp; |
18406 | | xp += ni; |
18407 | | tp += ni; |
18408 | | *xpp = (void*)xp; |
18409 | | } |
18410 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18411 | | |
18412 | | #else /* not SX */ |
18413 | 0 | const char *xp = (const char *) *xpp; |
18414 | 0 | int status = NC_NOERR; |
18415 | |
|
18416 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18417 | 0 | { |
18418 | 0 | const int lstatus = ncx_get_longlong_uint(xp, tp); |
18419 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18420 | 0 | status = lstatus; |
18421 | 0 | } |
18422 | |
|
18423 | 0 | *xpp = (const void *)xp; |
18424 | 0 | return status; |
18425 | 0 | #endif |
18426 | 0 | } |
18427 | | |
18428 | | int |
18429 | | ncx_getn_longlong_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
18430 | 0 | { |
18431 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18432 | | |
18433 | | /* basic algorithm is: |
18434 | | * - ensure sane alignment of input data |
18435 | | * - copy (conversion happens automatically) input data |
18436 | | * to output |
18437 | | * - update xpp to point at next unconverted input, and tp to point |
18438 | | * at next location for converted output |
18439 | | */ |
18440 | | long i, j, ni; |
18441 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18442 | | int64 *xp; |
18443 | | int nrange = 0; /* number of range errors */ |
18444 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18445 | | long cxp = (long) *((char**)xpp); |
18446 | | |
18447 | | realign = (cxp & 7) % SIZEOF_INT64; |
18448 | | /* sjl: manually stripmine so we can limit amount of |
18449 | | * vector work space reserved to LOOPCNT elements. Also |
18450 | | * makes vectorisation easy */ |
18451 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18452 | | ni=Min(nelems-j,LOOPCNT); |
18453 | | if (realign) { |
18454 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_INT64)); |
18455 | | xp = tmp; |
18456 | | } else { |
18457 | | xp = (int64 *) *xpp; |
18458 | | } |
18459 | | /* copy the next block */ |
18460 | | #pragma cdir loopcnt=LOOPCNT |
18461 | | #pragma cdir shortloop |
18462 | | for (i=0; i<ni; i++) { |
18463 | | tp[i] = (ulonglong) Max( ULONGLONG_MIN, Min(ULONGLONG_MAX, (ulonglong) xp[i])); |
18464 | | /* test for range errors (not always needed but do it anyway) */ |
18465 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
18466 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
18467 | | nrange += xp[i] > ULONGLONG_MAX || xp[i] < 0; |
18468 | | } |
18469 | | /* update xpp and tp */ |
18470 | | if (realign) xp = (int64 *) *xpp; |
18471 | | xp += ni; |
18472 | | tp += ni; |
18473 | | *xpp = (void*)xp; |
18474 | | } |
18475 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18476 | | |
18477 | | #else /* not SX */ |
18478 | 0 | const char *xp = (const char *) *xpp; |
18479 | 0 | int status = NC_NOERR; |
18480 | |
|
18481 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18482 | 0 | { |
18483 | 0 | const int lstatus = ncx_get_longlong_ulonglong(xp, tp); |
18484 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18485 | 0 | status = lstatus; |
18486 | 0 | } |
18487 | |
|
18488 | 0 | *xpp = (const void *)xp; |
18489 | 0 | return status; |
18490 | 0 | #endif |
18491 | 0 | } |
18492 | | |
18493 | | |
18494 | | #if X_SIZEOF_INT64 == SIZEOF_LONGLONG |
18495 | | /* optimized version */ |
18496 | | int |
18497 | | ncx_putn_longlong_longlong(void **xpp, size_t nelems, const long long *tp, void *fillp) |
18498 | 0 | { |
18499 | | #ifdef WORDS_BIGENDIAN |
18500 | | (void) memcpy(*xpp, tp, (size_t)nelems * X_SIZEOF_INT64); |
18501 | | # else |
18502 | 0 | swapn8b(*xpp, tp, nelems); |
18503 | 0 | # endif |
18504 | 0 | *xpp = (void *)((char *)(*xpp) + nelems * X_SIZEOF_INT64); |
18505 | 0 | return NC_NOERR; |
18506 | 0 | } |
18507 | | #else |
18508 | | int |
18509 | | ncx_putn_longlong_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
18510 | | { |
18511 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18512 | | |
18513 | | /* basic algorithm is: |
18514 | | * - ensure sane alignment of output data |
18515 | | * - copy (conversion happens automatically) input data |
18516 | | * to output |
18517 | | * - update tp to point at next unconverted input, and xpp to point |
18518 | | * at next location for converted output |
18519 | | */ |
18520 | | long i, j, ni; |
18521 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18522 | | int64 *xp; |
18523 | | int nrange = 0; /* number of range errors */ |
18524 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18525 | | long cxp = (long) *((char**)xpp); |
18526 | | |
18527 | | realign = (cxp & 7) % SIZEOF_INT64; |
18528 | | /* sjl: manually stripmine so we can limit amount of |
18529 | | * vector work space reserved to LOOPCNT elements. Also |
18530 | | * makes vectorisation easy */ |
18531 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18532 | | ni=Min(nelems-j,LOOPCNT); |
18533 | | if (realign) { |
18534 | | xp = tmp; |
18535 | | } else { |
18536 | | xp = (int64 *) *xpp; |
18537 | | } |
18538 | | /* copy the next block */ |
18539 | | #pragma cdir loopcnt=LOOPCNT |
18540 | | #pragma cdir shortloop |
18541 | | for (i=0; i<ni; i++) { |
18542 | | /* the normal case: */ |
18543 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
18544 | | /* test for range errors (not always needed but do it anyway) */ |
18545 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
18546 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
18547 | | nrange += tp[i] > X_INT64_MAX || tp[i] < X_INT64_MIN; |
18548 | | } |
18549 | | /* copy workspace back if necessary */ |
18550 | | if (realign) { |
18551 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
18552 | | xp = (int64 *) *xpp; |
18553 | | } |
18554 | | /* update xpp and tp */ |
18555 | | xp += ni; |
18556 | | tp += ni; |
18557 | | *xpp = (void*)xp; |
18558 | | } |
18559 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18560 | | |
18561 | | #else /* not SX */ |
18562 | | |
18563 | | char *xp = (char *) *xpp; |
18564 | | int status = NC_NOERR; |
18565 | | |
18566 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18567 | | { |
18568 | | int lstatus = ncx_put_longlong_longlong(xp, tp, fillp); |
18569 | | if (status == NC_NOERR) /* report the first encountered error */ |
18570 | | status = lstatus; |
18571 | | } |
18572 | | |
18573 | | *xpp = (void *)xp; |
18574 | | return status; |
18575 | | #endif |
18576 | | } |
18577 | | |
18578 | | #endif |
18579 | | int |
18580 | | ncx_putn_longlong_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
18581 | 0 | { |
18582 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18583 | | |
18584 | | /* basic algorithm is: |
18585 | | * - ensure sane alignment of output data |
18586 | | * - copy (conversion happens automatically) input data |
18587 | | * to output |
18588 | | * - update tp to point at next unconverted input, and xpp to point |
18589 | | * at next location for converted output |
18590 | | */ |
18591 | | long i, j, ni; |
18592 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18593 | | int64 *xp; |
18594 | | int nrange = 0; /* number of range errors */ |
18595 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18596 | | long cxp = (long) *((char**)xpp); |
18597 | | |
18598 | | realign = (cxp & 7) % SIZEOF_INT64; |
18599 | | /* sjl: manually stripmine so we can limit amount of |
18600 | | * vector work space reserved to LOOPCNT elements. Also |
18601 | | * makes vectorisation easy */ |
18602 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18603 | | ni=Min(nelems-j,LOOPCNT); |
18604 | | if (realign) { |
18605 | | xp = tmp; |
18606 | | } else { |
18607 | | xp = (int64 *) *xpp; |
18608 | | } |
18609 | | /* copy the next block */ |
18610 | | #pragma cdir loopcnt=LOOPCNT |
18611 | | #pragma cdir shortloop |
18612 | | for (i=0; i<ni; i++) { |
18613 | | /* the normal case: */ |
18614 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
18615 | | /* test for range errors (not always needed but do it anyway) */ |
18616 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
18617 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
18618 | | nrange += tp[i] > X_INT64_MAX || tp[i] < X_INT64_MIN; |
18619 | | } |
18620 | | /* copy workspace back if necessary */ |
18621 | | if (realign) { |
18622 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
18623 | | xp = (int64 *) *xpp; |
18624 | | } |
18625 | | /* update xpp and tp */ |
18626 | | xp += ni; |
18627 | | tp += ni; |
18628 | | *xpp = (void*)xp; |
18629 | | } |
18630 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18631 | | |
18632 | | #else /* not SX */ |
18633 | |
|
18634 | 0 | char *xp = (char *) *xpp; |
18635 | 0 | int status = NC_NOERR; |
18636 | |
|
18637 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18638 | 0 | { |
18639 | 0 | int lstatus = ncx_put_longlong_schar(xp, tp, fillp); |
18640 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18641 | 0 | status = lstatus; |
18642 | 0 | } |
18643 | |
|
18644 | 0 | *xpp = (void *)xp; |
18645 | 0 | return status; |
18646 | 0 | #endif |
18647 | 0 | } |
18648 | | |
18649 | | int |
18650 | | ncx_putn_longlong_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
18651 | 0 | { |
18652 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18653 | | |
18654 | | /* basic algorithm is: |
18655 | | * - ensure sane alignment of output data |
18656 | | * - copy (conversion happens automatically) input data |
18657 | | * to output |
18658 | | * - update tp to point at next unconverted input, and xpp to point |
18659 | | * at next location for converted output |
18660 | | */ |
18661 | | long i, j, ni; |
18662 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18663 | | int64 *xp; |
18664 | | int nrange = 0; /* number of range errors */ |
18665 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18666 | | long cxp = (long) *((char**)xpp); |
18667 | | |
18668 | | realign = (cxp & 7) % SIZEOF_INT64; |
18669 | | /* sjl: manually stripmine so we can limit amount of |
18670 | | * vector work space reserved to LOOPCNT elements. Also |
18671 | | * makes vectorisation easy */ |
18672 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18673 | | ni=Min(nelems-j,LOOPCNT); |
18674 | | if (realign) { |
18675 | | xp = tmp; |
18676 | | } else { |
18677 | | xp = (int64 *) *xpp; |
18678 | | } |
18679 | | /* copy the next block */ |
18680 | | #pragma cdir loopcnt=LOOPCNT |
18681 | | #pragma cdir shortloop |
18682 | | for (i=0; i<ni; i++) { |
18683 | | /* the normal case: */ |
18684 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
18685 | | /* test for range errors (not always needed but do it anyway) */ |
18686 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
18687 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
18688 | | nrange += tp[i] > X_INT64_MAX || tp[i] < X_INT64_MIN; |
18689 | | } |
18690 | | /* copy workspace back if necessary */ |
18691 | | if (realign) { |
18692 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
18693 | | xp = (int64 *) *xpp; |
18694 | | } |
18695 | | /* update xpp and tp */ |
18696 | | xp += ni; |
18697 | | tp += ni; |
18698 | | *xpp = (void*)xp; |
18699 | | } |
18700 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18701 | | |
18702 | | #else /* not SX */ |
18703 | |
|
18704 | 0 | char *xp = (char *) *xpp; |
18705 | 0 | int status = NC_NOERR; |
18706 | |
|
18707 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18708 | 0 | { |
18709 | 0 | int lstatus = ncx_put_longlong_short(xp, tp, fillp); |
18710 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18711 | 0 | status = lstatus; |
18712 | 0 | } |
18713 | |
|
18714 | 0 | *xpp = (void *)xp; |
18715 | 0 | return status; |
18716 | 0 | #endif |
18717 | 0 | } |
18718 | | |
18719 | | int |
18720 | | ncx_putn_longlong_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
18721 | 0 | { |
18722 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18723 | | |
18724 | | /* basic algorithm is: |
18725 | | * - ensure sane alignment of output data |
18726 | | * - copy (conversion happens automatically) input data |
18727 | | * to output |
18728 | | * - update tp to point at next unconverted input, and xpp to point |
18729 | | * at next location for converted output |
18730 | | */ |
18731 | | long i, j, ni; |
18732 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18733 | | int64 *xp; |
18734 | | int nrange = 0; /* number of range errors */ |
18735 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18736 | | long cxp = (long) *((char**)xpp); |
18737 | | |
18738 | | realign = (cxp & 7) % SIZEOF_INT64; |
18739 | | /* sjl: manually stripmine so we can limit amount of |
18740 | | * vector work space reserved to LOOPCNT elements. Also |
18741 | | * makes vectorisation easy */ |
18742 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18743 | | ni=Min(nelems-j,LOOPCNT); |
18744 | | if (realign) { |
18745 | | xp = tmp; |
18746 | | } else { |
18747 | | xp = (int64 *) *xpp; |
18748 | | } |
18749 | | /* copy the next block */ |
18750 | | #pragma cdir loopcnt=LOOPCNT |
18751 | | #pragma cdir shortloop |
18752 | | for (i=0; i<ni; i++) { |
18753 | | /* the normal case: */ |
18754 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
18755 | | /* test for range errors (not always needed but do it anyway) */ |
18756 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
18757 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
18758 | | nrange += tp[i] > X_INT64_MAX || tp[i] < X_INT64_MIN; |
18759 | | } |
18760 | | /* copy workspace back if necessary */ |
18761 | | if (realign) { |
18762 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
18763 | | xp = (int64 *) *xpp; |
18764 | | } |
18765 | | /* update xpp and tp */ |
18766 | | xp += ni; |
18767 | | tp += ni; |
18768 | | *xpp = (void*)xp; |
18769 | | } |
18770 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18771 | | |
18772 | | #else /* not SX */ |
18773 | |
|
18774 | 0 | char *xp = (char *) *xpp; |
18775 | 0 | int status = NC_NOERR; |
18776 | |
|
18777 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18778 | 0 | { |
18779 | 0 | int lstatus = ncx_put_longlong_int(xp, tp, fillp); |
18780 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18781 | 0 | status = lstatus; |
18782 | 0 | } |
18783 | |
|
18784 | 0 | *xpp = (void *)xp; |
18785 | 0 | return status; |
18786 | 0 | #endif |
18787 | 0 | } |
18788 | | |
18789 | | int |
18790 | | ncx_putn_longlong_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
18791 | 0 | { |
18792 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18793 | | |
18794 | | /* basic algorithm is: |
18795 | | * - ensure sane alignment of output data |
18796 | | * - copy (conversion happens automatically) input data |
18797 | | * to output |
18798 | | * - update tp to point at next unconverted input, and xpp to point |
18799 | | * at next location for converted output |
18800 | | */ |
18801 | | long i, j, ni; |
18802 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18803 | | int64 *xp; |
18804 | | int nrange = 0; /* number of range errors */ |
18805 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18806 | | long cxp = (long) *((char**)xpp); |
18807 | | |
18808 | | realign = (cxp & 7) % SIZEOF_INT64; |
18809 | | /* sjl: manually stripmine so we can limit amount of |
18810 | | * vector work space reserved to LOOPCNT elements. Also |
18811 | | * makes vectorisation easy */ |
18812 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18813 | | ni=Min(nelems-j,LOOPCNT); |
18814 | | if (realign) { |
18815 | | xp = tmp; |
18816 | | } else { |
18817 | | xp = (int64 *) *xpp; |
18818 | | } |
18819 | | /* copy the next block */ |
18820 | | #pragma cdir loopcnt=LOOPCNT |
18821 | | #pragma cdir shortloop |
18822 | | for (i=0; i<ni; i++) { |
18823 | | /* the normal case: */ |
18824 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
18825 | | /* test for range errors (not always needed but do it anyway) */ |
18826 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
18827 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
18828 | | nrange += tp[i] > X_INT64_MAX || tp[i] < X_INT64_MIN; |
18829 | | } |
18830 | | /* copy workspace back if necessary */ |
18831 | | if (realign) { |
18832 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
18833 | | xp = (int64 *) *xpp; |
18834 | | } |
18835 | | /* update xpp and tp */ |
18836 | | xp += ni; |
18837 | | tp += ni; |
18838 | | *xpp = (void*)xp; |
18839 | | } |
18840 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18841 | | |
18842 | | #else /* not SX */ |
18843 | |
|
18844 | 0 | char *xp = (char *) *xpp; |
18845 | 0 | int status = NC_NOERR; |
18846 | |
|
18847 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18848 | 0 | { |
18849 | 0 | int lstatus = ncx_put_longlong_long(xp, tp, fillp); |
18850 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18851 | 0 | status = lstatus; |
18852 | 0 | } |
18853 | |
|
18854 | 0 | *xpp = (void *)xp; |
18855 | 0 | return status; |
18856 | 0 | #endif |
18857 | 0 | } |
18858 | | |
18859 | | int |
18860 | | ncx_putn_longlong_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
18861 | 0 | { |
18862 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18863 | | |
18864 | | /* basic algorithm is: |
18865 | | * - ensure sane alignment of output data |
18866 | | * - copy (conversion happens automatically) input data |
18867 | | * to output |
18868 | | * - update tp to point at next unconverted input, and xpp to point |
18869 | | * at next location for converted output |
18870 | | */ |
18871 | | long i, j, ni; |
18872 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18873 | | int64 *xp; |
18874 | | int nrange = 0; /* number of range errors */ |
18875 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18876 | | long cxp = (long) *((char**)xpp); |
18877 | | |
18878 | | realign = (cxp & 7) % SIZEOF_INT64; |
18879 | | /* sjl: manually stripmine so we can limit amount of |
18880 | | * vector work space reserved to LOOPCNT elements. Also |
18881 | | * makes vectorisation easy */ |
18882 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18883 | | ni=Min(nelems-j,LOOPCNT); |
18884 | | if (realign) { |
18885 | | xp = tmp; |
18886 | | } else { |
18887 | | xp = (int64 *) *xpp; |
18888 | | } |
18889 | | /* copy the next block */ |
18890 | | #pragma cdir loopcnt=LOOPCNT |
18891 | | #pragma cdir shortloop |
18892 | | for (i=0; i<ni; i++) { |
18893 | | /* the normal case: */ |
18894 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
18895 | | /* test for range errors (not always needed but do it anyway) */ |
18896 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
18897 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
18898 | | nrange += tp[i] > X_INT64_MAX || tp[i] < X_INT64_MIN; |
18899 | | } |
18900 | | /* copy workspace back if necessary */ |
18901 | | if (realign) { |
18902 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
18903 | | xp = (int64 *) *xpp; |
18904 | | } |
18905 | | /* update xpp and tp */ |
18906 | | xp += ni; |
18907 | | tp += ni; |
18908 | | *xpp = (void*)xp; |
18909 | | } |
18910 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18911 | | |
18912 | | #else /* not SX */ |
18913 | |
|
18914 | 0 | char *xp = (char *) *xpp; |
18915 | 0 | int status = NC_NOERR; |
18916 | |
|
18917 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18918 | 0 | { |
18919 | 0 | int lstatus = ncx_put_longlong_float(xp, tp, fillp); |
18920 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18921 | 0 | status = lstatus; |
18922 | 0 | } |
18923 | |
|
18924 | 0 | *xpp = (void *)xp; |
18925 | 0 | return status; |
18926 | 0 | #endif |
18927 | 0 | } |
18928 | | |
18929 | | int |
18930 | | ncx_putn_longlong_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
18931 | 0 | { |
18932 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
18933 | | |
18934 | | /* basic algorithm is: |
18935 | | * - ensure sane alignment of output data |
18936 | | * - copy (conversion happens automatically) input data |
18937 | | * to output |
18938 | | * - update tp to point at next unconverted input, and xpp to point |
18939 | | * at next location for converted output |
18940 | | */ |
18941 | | long i, j, ni; |
18942 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
18943 | | int64 *xp; |
18944 | | int nrange = 0; /* number of range errors */ |
18945 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
18946 | | long cxp = (long) *((char**)xpp); |
18947 | | |
18948 | | realign = (cxp & 7) % SIZEOF_INT64; |
18949 | | /* sjl: manually stripmine so we can limit amount of |
18950 | | * vector work space reserved to LOOPCNT elements. Also |
18951 | | * makes vectorisation easy */ |
18952 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
18953 | | ni=Min(nelems-j,LOOPCNT); |
18954 | | if (realign) { |
18955 | | xp = tmp; |
18956 | | } else { |
18957 | | xp = (int64 *) *xpp; |
18958 | | } |
18959 | | /* copy the next block */ |
18960 | | #pragma cdir loopcnt=LOOPCNT |
18961 | | #pragma cdir shortloop |
18962 | | for (i=0; i<ni; i++) { |
18963 | | /* the normal case: */ |
18964 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
18965 | | /* test for range errors (not always needed but do it anyway) */ |
18966 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
18967 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
18968 | | nrange += tp[i] > X_INT64_MAX || tp[i] < X_INT64_MIN; |
18969 | | } |
18970 | | /* copy workspace back if necessary */ |
18971 | | if (realign) { |
18972 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
18973 | | xp = (int64 *) *xpp; |
18974 | | } |
18975 | | /* update xpp and tp */ |
18976 | | xp += ni; |
18977 | | tp += ni; |
18978 | | *xpp = (void*)xp; |
18979 | | } |
18980 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
18981 | | |
18982 | | #else /* not SX */ |
18983 | |
|
18984 | 0 | char *xp = (char *) *xpp; |
18985 | 0 | int status = NC_NOERR; |
18986 | |
|
18987 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
18988 | 0 | { |
18989 | 0 | int lstatus = ncx_put_longlong_double(xp, tp, fillp); |
18990 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
18991 | 0 | status = lstatus; |
18992 | 0 | } |
18993 | |
|
18994 | 0 | *xpp = (void *)xp; |
18995 | 0 | return status; |
18996 | 0 | #endif |
18997 | 0 | } |
18998 | | |
18999 | | int |
19000 | | ncx_putn_longlong_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
19001 | 0 | { |
19002 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
19003 | | |
19004 | | /* basic algorithm is: |
19005 | | * - ensure sane alignment of output data |
19006 | | * - copy (conversion happens automatically) input data |
19007 | | * to output |
19008 | | * - update tp to point at next unconverted input, and xpp to point |
19009 | | * at next location for converted output |
19010 | | */ |
19011 | | long i, j, ni; |
19012 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19013 | | int64 *xp; |
19014 | | int nrange = 0; /* number of range errors */ |
19015 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19016 | | long cxp = (long) *((char**)xpp); |
19017 | | |
19018 | | realign = (cxp & 7) % SIZEOF_INT64; |
19019 | | /* sjl: manually stripmine so we can limit amount of |
19020 | | * vector work space reserved to LOOPCNT elements. Also |
19021 | | * makes vectorisation easy */ |
19022 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19023 | | ni=Min(nelems-j,LOOPCNT); |
19024 | | if (realign) { |
19025 | | xp = tmp; |
19026 | | } else { |
19027 | | xp = (int64 *) *xpp; |
19028 | | } |
19029 | | /* copy the next block */ |
19030 | | #pragma cdir loopcnt=LOOPCNT |
19031 | | #pragma cdir shortloop |
19032 | | for (i=0; i<ni; i++) { |
19033 | | /* the normal case: */ |
19034 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
19035 | | /* test for range errors (not always needed but do it anyway) */ |
19036 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
19037 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
19038 | | nrange += tp[i] > X_INT64_MAX ; |
19039 | | } |
19040 | | /* copy workspace back if necessary */ |
19041 | | if (realign) { |
19042 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
19043 | | xp = (int64 *) *xpp; |
19044 | | } |
19045 | | /* update xpp and tp */ |
19046 | | xp += ni; |
19047 | | tp += ni; |
19048 | | *xpp = (void*)xp; |
19049 | | } |
19050 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19051 | | |
19052 | | #else /* not SX */ |
19053 | |
|
19054 | 0 | char *xp = (char *) *xpp; |
19055 | 0 | int status = NC_NOERR; |
19056 | |
|
19057 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
19058 | 0 | { |
19059 | 0 | int lstatus = ncx_put_longlong_uchar(xp, tp, fillp); |
19060 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19061 | 0 | status = lstatus; |
19062 | 0 | } |
19063 | |
|
19064 | 0 | *xpp = (void *)xp; |
19065 | 0 | return status; |
19066 | 0 | #endif |
19067 | 0 | } |
19068 | | |
19069 | | int |
19070 | | ncx_putn_longlong_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
19071 | 0 | { |
19072 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
19073 | | |
19074 | | /* basic algorithm is: |
19075 | | * - ensure sane alignment of output data |
19076 | | * - copy (conversion happens automatically) input data |
19077 | | * to output |
19078 | | * - update tp to point at next unconverted input, and xpp to point |
19079 | | * at next location for converted output |
19080 | | */ |
19081 | | long i, j, ni; |
19082 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19083 | | int64 *xp; |
19084 | | int nrange = 0; /* number of range errors */ |
19085 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19086 | | long cxp = (long) *((char**)xpp); |
19087 | | |
19088 | | realign = (cxp & 7) % SIZEOF_INT64; |
19089 | | /* sjl: manually stripmine so we can limit amount of |
19090 | | * vector work space reserved to LOOPCNT elements. Also |
19091 | | * makes vectorisation easy */ |
19092 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19093 | | ni=Min(nelems-j,LOOPCNT); |
19094 | | if (realign) { |
19095 | | xp = tmp; |
19096 | | } else { |
19097 | | xp = (int64 *) *xpp; |
19098 | | } |
19099 | | /* copy the next block */ |
19100 | | #pragma cdir loopcnt=LOOPCNT |
19101 | | #pragma cdir shortloop |
19102 | | for (i=0; i<ni; i++) { |
19103 | | /* the normal case: */ |
19104 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
19105 | | /* test for range errors (not always needed but do it anyway) */ |
19106 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
19107 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
19108 | | nrange += tp[i] > X_INT64_MAX ; |
19109 | | } |
19110 | | /* copy workspace back if necessary */ |
19111 | | if (realign) { |
19112 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
19113 | | xp = (int64 *) *xpp; |
19114 | | } |
19115 | | /* update xpp and tp */ |
19116 | | xp += ni; |
19117 | | tp += ni; |
19118 | | *xpp = (void*)xp; |
19119 | | } |
19120 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19121 | | |
19122 | | #else /* not SX */ |
19123 | |
|
19124 | 0 | char *xp = (char *) *xpp; |
19125 | 0 | int status = NC_NOERR; |
19126 | |
|
19127 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
19128 | 0 | { |
19129 | 0 | int lstatus = ncx_put_longlong_ushort(xp, tp, fillp); |
19130 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19131 | 0 | status = lstatus; |
19132 | 0 | } |
19133 | |
|
19134 | 0 | *xpp = (void *)xp; |
19135 | 0 | return status; |
19136 | 0 | #endif |
19137 | 0 | } |
19138 | | |
19139 | | int |
19140 | | ncx_putn_longlong_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
19141 | 0 | { |
19142 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
19143 | | |
19144 | | /* basic algorithm is: |
19145 | | * - ensure sane alignment of output data |
19146 | | * - copy (conversion happens automatically) input data |
19147 | | * to output |
19148 | | * - update tp to point at next unconverted input, and xpp to point |
19149 | | * at next location for converted output |
19150 | | */ |
19151 | | long i, j, ni; |
19152 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19153 | | int64 *xp; |
19154 | | int nrange = 0; /* number of range errors */ |
19155 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19156 | | long cxp = (long) *((char**)xpp); |
19157 | | |
19158 | | realign = (cxp & 7) % SIZEOF_INT64; |
19159 | | /* sjl: manually stripmine so we can limit amount of |
19160 | | * vector work space reserved to LOOPCNT elements. Also |
19161 | | * makes vectorisation easy */ |
19162 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19163 | | ni=Min(nelems-j,LOOPCNT); |
19164 | | if (realign) { |
19165 | | xp = tmp; |
19166 | | } else { |
19167 | | xp = (int64 *) *xpp; |
19168 | | } |
19169 | | /* copy the next block */ |
19170 | | #pragma cdir loopcnt=LOOPCNT |
19171 | | #pragma cdir shortloop |
19172 | | for (i=0; i<ni; i++) { |
19173 | | /* the normal case: */ |
19174 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
19175 | | /* test for range errors (not always needed but do it anyway) */ |
19176 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
19177 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
19178 | | nrange += tp[i] > X_INT64_MAX ; |
19179 | | } |
19180 | | /* copy workspace back if necessary */ |
19181 | | if (realign) { |
19182 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
19183 | | xp = (int64 *) *xpp; |
19184 | | } |
19185 | | /* update xpp and tp */ |
19186 | | xp += ni; |
19187 | | tp += ni; |
19188 | | *xpp = (void*)xp; |
19189 | | } |
19190 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19191 | | |
19192 | | #else /* not SX */ |
19193 | |
|
19194 | 0 | char *xp = (char *) *xpp; |
19195 | 0 | int status = NC_NOERR; |
19196 | |
|
19197 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
19198 | 0 | { |
19199 | 0 | int lstatus = ncx_put_longlong_uint(xp, tp, fillp); |
19200 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19201 | 0 | status = lstatus; |
19202 | 0 | } |
19203 | |
|
19204 | 0 | *xpp = (void *)xp; |
19205 | 0 | return status; |
19206 | 0 | #endif |
19207 | 0 | } |
19208 | | |
19209 | | int |
19210 | | ncx_putn_longlong_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
19211 | 0 | { |
19212 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_INT64 == SIZEOF_INT64 |
19213 | | |
19214 | | /* basic algorithm is: |
19215 | | * - ensure sane alignment of output data |
19216 | | * - copy (conversion happens automatically) input data |
19217 | | * to output |
19218 | | * - update tp to point at next unconverted input, and xpp to point |
19219 | | * at next location for converted output |
19220 | | */ |
19221 | | long i, j, ni; |
19222 | | int64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19223 | | int64 *xp; |
19224 | | int nrange = 0; /* number of range errors */ |
19225 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19226 | | long cxp = (long) *((char**)xpp); |
19227 | | |
19228 | | realign = (cxp & 7) % SIZEOF_INT64; |
19229 | | /* sjl: manually stripmine so we can limit amount of |
19230 | | * vector work space reserved to LOOPCNT elements. Also |
19231 | | * makes vectorisation easy */ |
19232 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19233 | | ni=Min(nelems-j,LOOPCNT); |
19234 | | if (realign) { |
19235 | | xp = tmp; |
19236 | | } else { |
19237 | | xp = (int64 *) *xpp; |
19238 | | } |
19239 | | /* copy the next block */ |
19240 | | #pragma cdir loopcnt=LOOPCNT |
19241 | | #pragma cdir shortloop |
19242 | | for (i=0; i<ni; i++) { |
19243 | | /* the normal case: */ |
19244 | | xp[i] = (int64) Max( X_INT64_MIN, Min(X_INT64_MAX, (int64) tp[i])); |
19245 | | /* test for range errors (not always needed but do it anyway) */ |
19246 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
19247 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
19248 | | nrange += tp[i] > X_INT64_MAX ; |
19249 | | } |
19250 | | /* copy workspace back if necessary */ |
19251 | | if (realign) { |
19252 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_INT64); |
19253 | | xp = (int64 *) *xpp; |
19254 | | } |
19255 | | /* update xpp and tp */ |
19256 | | xp += ni; |
19257 | | tp += ni; |
19258 | | *xpp = (void*)xp; |
19259 | | } |
19260 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19261 | | |
19262 | | #else /* not SX */ |
19263 | |
|
19264 | 0 | char *xp = (char *) *xpp; |
19265 | 0 | int status = NC_NOERR; |
19266 | |
|
19267 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_INT64, tp++) |
19268 | 0 | { |
19269 | 0 | int lstatus = ncx_put_longlong_ulonglong(xp, tp, fillp); |
19270 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19271 | 0 | status = lstatus; |
19272 | 0 | } |
19273 | |
|
19274 | 0 | *xpp = (void *)xp; |
19275 | 0 | return status; |
19276 | 0 | #endif |
19277 | 0 | } |
19278 | | |
19279 | | |
19280 | | /* uint64 --------------------------------------------------------------------*/ |
19281 | | |
19282 | | #if X_SIZEOF_UINT64 == SIZEOF_ULONGLONG |
19283 | | /* optimized version */ |
19284 | | int |
19285 | | ncx_getn_ulonglong_ulonglong(const void **xpp, size_t nelems, unsigned long long *tp) |
19286 | 0 | { |
19287 | | #ifdef WORDS_BIGENDIAN |
19288 | | (void) memcpy(tp, *xpp, (size_t)nelems * SIZEOF_UNSIGNED_LONG_LONG); |
19289 | | # else |
19290 | 0 | swapn8b(tp, *xpp, nelems); |
19291 | 0 | # endif |
19292 | 0 | *xpp = (const void *)((const char *)(*xpp) + nelems * X_SIZEOF_UINT64); |
19293 | 0 | return NC_NOERR; |
19294 | 0 | } |
19295 | | #else |
19296 | | int |
19297 | | ncx_getn_ulonglong_ulonglong(const void **xpp, size_t nelems, ulonglong *tp) |
19298 | | { |
19299 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19300 | | |
19301 | | /* basic algorithm is: |
19302 | | * - ensure sane alignment of input data |
19303 | | * - copy (conversion happens automatically) input data |
19304 | | * to output |
19305 | | * - update xpp to point at next unconverted input, and tp to point |
19306 | | * at next location for converted output |
19307 | | */ |
19308 | | long i, j, ni; |
19309 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19310 | | uint64 *xp; |
19311 | | int nrange = 0; /* number of range errors */ |
19312 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19313 | | long cxp = (long) *((char**)xpp); |
19314 | | |
19315 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19316 | | /* sjl: manually stripmine so we can limit amount of |
19317 | | * vector work space reserved to LOOPCNT elements. Also |
19318 | | * makes vectorisation easy */ |
19319 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19320 | | ni=Min(nelems-j,LOOPCNT); |
19321 | | if (realign) { |
19322 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19323 | | xp = tmp; |
19324 | | } else { |
19325 | | xp = (uint64 *) *xpp; |
19326 | | } |
19327 | | /* copy the next block */ |
19328 | | #pragma cdir loopcnt=LOOPCNT |
19329 | | #pragma cdir shortloop |
19330 | | for (i=0; i<ni; i++) { |
19331 | | tp[i] = (ulonglong) Max( ULONGLONG_MIN, Min(ULONGLONG_MAX, (ulonglong) xp[i])); |
19332 | | /* test for range errors (not always needed but do it anyway) */ |
19333 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19334 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19335 | | nrange += xp[i] > ULONGLONG_MAX ; |
19336 | | } |
19337 | | /* update xpp and tp */ |
19338 | | if (realign) xp = (uint64 *) *xpp; |
19339 | | xp += ni; |
19340 | | tp += ni; |
19341 | | *xpp = (void*)xp; |
19342 | | } |
19343 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19344 | | |
19345 | | #else /* not SX */ |
19346 | | const char *xp = (const char *) *xpp; |
19347 | | int status = NC_NOERR; |
19348 | | |
19349 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19350 | | { |
19351 | | const int lstatus = ncx_get_ulonglong_ulonglong(xp, tp); |
19352 | | if (status == NC_NOERR) /* report the first encountered error */ |
19353 | | status = lstatus; |
19354 | | } |
19355 | | |
19356 | | *xpp = (const void *)xp; |
19357 | | return status; |
19358 | | #endif |
19359 | | } |
19360 | | |
19361 | | #endif |
19362 | | int |
19363 | | ncx_getn_ulonglong_schar(const void **xpp, size_t nelems, schar *tp) |
19364 | 0 | { |
19365 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19366 | | |
19367 | | /* basic algorithm is: |
19368 | | * - ensure sane alignment of input data |
19369 | | * - copy (conversion happens automatically) input data |
19370 | | * to output |
19371 | | * - update xpp to point at next unconverted input, and tp to point |
19372 | | * at next location for converted output |
19373 | | */ |
19374 | | long i, j, ni; |
19375 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19376 | | uint64 *xp; |
19377 | | int nrange = 0; /* number of range errors */ |
19378 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19379 | | long cxp = (long) *((char**)xpp); |
19380 | | |
19381 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19382 | | /* sjl: manually stripmine so we can limit amount of |
19383 | | * vector work space reserved to LOOPCNT elements. Also |
19384 | | * makes vectorisation easy */ |
19385 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19386 | | ni=Min(nelems-j,LOOPCNT); |
19387 | | if (realign) { |
19388 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19389 | | xp = tmp; |
19390 | | } else { |
19391 | | xp = (uint64 *) *xpp; |
19392 | | } |
19393 | | /* copy the next block */ |
19394 | | #pragma cdir loopcnt=LOOPCNT |
19395 | | #pragma cdir shortloop |
19396 | | for (i=0; i<ni; i++) { |
19397 | | tp[i] = (schar) Max( SCHAR_MIN, Min(SCHAR_MAX, (schar) xp[i])); |
19398 | | /* test for range errors (not always needed but do it anyway) */ |
19399 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19400 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19401 | | nrange += xp[i] > SCHAR_MAX ; |
19402 | | } |
19403 | | /* update xpp and tp */ |
19404 | | if (realign) xp = (uint64 *) *xpp; |
19405 | | xp += ni; |
19406 | | tp += ni; |
19407 | | *xpp = (void*)xp; |
19408 | | } |
19409 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19410 | | |
19411 | | #else /* not SX */ |
19412 | 0 | const char *xp = (const char *) *xpp; |
19413 | 0 | int status = NC_NOERR; |
19414 | |
|
19415 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19416 | 0 | { |
19417 | 0 | const int lstatus = ncx_get_ulonglong_schar(xp, tp); |
19418 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19419 | 0 | status = lstatus; |
19420 | 0 | } |
19421 | |
|
19422 | 0 | *xpp = (const void *)xp; |
19423 | 0 | return status; |
19424 | 0 | #endif |
19425 | 0 | } |
19426 | | |
19427 | | int |
19428 | | ncx_getn_ulonglong_short(const void **xpp, size_t nelems, short *tp) |
19429 | 0 | { |
19430 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19431 | | |
19432 | | /* basic algorithm is: |
19433 | | * - ensure sane alignment of input data |
19434 | | * - copy (conversion happens automatically) input data |
19435 | | * to output |
19436 | | * - update xpp to point at next unconverted input, and tp to point |
19437 | | * at next location for converted output |
19438 | | */ |
19439 | | long i, j, ni; |
19440 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19441 | | uint64 *xp; |
19442 | | int nrange = 0; /* number of range errors */ |
19443 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19444 | | long cxp = (long) *((char**)xpp); |
19445 | | |
19446 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19447 | | /* sjl: manually stripmine so we can limit amount of |
19448 | | * vector work space reserved to LOOPCNT elements. Also |
19449 | | * makes vectorisation easy */ |
19450 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19451 | | ni=Min(nelems-j,LOOPCNT); |
19452 | | if (realign) { |
19453 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19454 | | xp = tmp; |
19455 | | } else { |
19456 | | xp = (uint64 *) *xpp; |
19457 | | } |
19458 | | /* copy the next block */ |
19459 | | #pragma cdir loopcnt=LOOPCNT |
19460 | | #pragma cdir shortloop |
19461 | | for (i=0; i<ni; i++) { |
19462 | | tp[i] = (short) Max( SHORT_MIN, Min(SHORT_MAX, (short) xp[i])); |
19463 | | /* test for range errors (not always needed but do it anyway) */ |
19464 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19465 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19466 | | nrange += xp[i] > SHORT_MAX ; |
19467 | | } |
19468 | | /* update xpp and tp */ |
19469 | | if (realign) xp = (uint64 *) *xpp; |
19470 | | xp += ni; |
19471 | | tp += ni; |
19472 | | *xpp = (void*)xp; |
19473 | | } |
19474 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19475 | | |
19476 | | #else /* not SX */ |
19477 | 0 | const char *xp = (const char *) *xpp; |
19478 | 0 | int status = NC_NOERR; |
19479 | |
|
19480 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19481 | 0 | { |
19482 | 0 | const int lstatus = ncx_get_ulonglong_short(xp, tp); |
19483 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19484 | 0 | status = lstatus; |
19485 | 0 | } |
19486 | |
|
19487 | 0 | *xpp = (const void *)xp; |
19488 | 0 | return status; |
19489 | 0 | #endif |
19490 | 0 | } |
19491 | | |
19492 | | int |
19493 | | ncx_getn_ulonglong_int(const void **xpp, size_t nelems, int *tp) |
19494 | 0 | { |
19495 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19496 | | |
19497 | | /* basic algorithm is: |
19498 | | * - ensure sane alignment of input data |
19499 | | * - copy (conversion happens automatically) input data |
19500 | | * to output |
19501 | | * - update xpp to point at next unconverted input, and tp to point |
19502 | | * at next location for converted output |
19503 | | */ |
19504 | | long i, j, ni; |
19505 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19506 | | uint64 *xp; |
19507 | | int nrange = 0; /* number of range errors */ |
19508 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19509 | | long cxp = (long) *((char**)xpp); |
19510 | | |
19511 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19512 | | /* sjl: manually stripmine so we can limit amount of |
19513 | | * vector work space reserved to LOOPCNT elements. Also |
19514 | | * makes vectorisation easy */ |
19515 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19516 | | ni=Min(nelems-j,LOOPCNT); |
19517 | | if (realign) { |
19518 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19519 | | xp = tmp; |
19520 | | } else { |
19521 | | xp = (uint64 *) *xpp; |
19522 | | } |
19523 | | /* copy the next block */ |
19524 | | #pragma cdir loopcnt=LOOPCNT |
19525 | | #pragma cdir shortloop |
19526 | | for (i=0; i<ni; i++) { |
19527 | | tp[i] = (int) Max( INT_MIN, Min(INT_MAX, (int) xp[i])); |
19528 | | /* test for range errors (not always needed but do it anyway) */ |
19529 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19530 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19531 | | nrange += xp[i] > INT_MAX ; |
19532 | | } |
19533 | | /* update xpp and tp */ |
19534 | | if (realign) xp = (uint64 *) *xpp; |
19535 | | xp += ni; |
19536 | | tp += ni; |
19537 | | *xpp = (void*)xp; |
19538 | | } |
19539 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19540 | | |
19541 | | #else /* not SX */ |
19542 | 0 | const char *xp = (const char *) *xpp; |
19543 | 0 | int status = NC_NOERR; |
19544 | |
|
19545 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19546 | 0 | { |
19547 | 0 | const int lstatus = ncx_get_ulonglong_int(xp, tp); |
19548 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19549 | 0 | status = lstatus; |
19550 | 0 | } |
19551 | |
|
19552 | 0 | *xpp = (const void *)xp; |
19553 | 0 | return status; |
19554 | 0 | #endif |
19555 | 0 | } |
19556 | | |
19557 | | int |
19558 | | ncx_getn_ulonglong_long(const void **xpp, size_t nelems, long *tp) |
19559 | 0 | { |
19560 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19561 | | |
19562 | | /* basic algorithm is: |
19563 | | * - ensure sane alignment of input data |
19564 | | * - copy (conversion happens automatically) input data |
19565 | | * to output |
19566 | | * - update xpp to point at next unconverted input, and tp to point |
19567 | | * at next location for converted output |
19568 | | */ |
19569 | | long i, j, ni; |
19570 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19571 | | uint64 *xp; |
19572 | | int nrange = 0; /* number of range errors */ |
19573 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19574 | | long cxp = (long) *((char**)xpp); |
19575 | | |
19576 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19577 | | /* sjl: manually stripmine so we can limit amount of |
19578 | | * vector work space reserved to LOOPCNT elements. Also |
19579 | | * makes vectorisation easy */ |
19580 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19581 | | ni=Min(nelems-j,LOOPCNT); |
19582 | | if (realign) { |
19583 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19584 | | xp = tmp; |
19585 | | } else { |
19586 | | xp = (uint64 *) *xpp; |
19587 | | } |
19588 | | /* copy the next block */ |
19589 | | #pragma cdir loopcnt=LOOPCNT |
19590 | | #pragma cdir shortloop |
19591 | | for (i=0; i<ni; i++) { |
19592 | | tp[i] = (long) Max( LONG_MIN, Min(LONG_MAX, (long) xp[i])); |
19593 | | /* test for range errors (not always needed but do it anyway) */ |
19594 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19595 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19596 | | nrange += xp[i] > LONG_MAX ; |
19597 | | } |
19598 | | /* update xpp and tp */ |
19599 | | if (realign) xp = (uint64 *) *xpp; |
19600 | | xp += ni; |
19601 | | tp += ni; |
19602 | | *xpp = (void*)xp; |
19603 | | } |
19604 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19605 | | |
19606 | | #else /* not SX */ |
19607 | 0 | const char *xp = (const char *) *xpp; |
19608 | 0 | int status = NC_NOERR; |
19609 | |
|
19610 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19611 | 0 | { |
19612 | 0 | const int lstatus = ncx_get_ulonglong_long(xp, tp); |
19613 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19614 | 0 | status = lstatus; |
19615 | 0 | } |
19616 | |
|
19617 | 0 | *xpp = (const void *)xp; |
19618 | 0 | return status; |
19619 | 0 | #endif |
19620 | 0 | } |
19621 | | |
19622 | | int |
19623 | | ncx_getn_ulonglong_float(const void **xpp, size_t nelems, float *tp) |
19624 | 0 | { |
19625 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19626 | | |
19627 | | /* basic algorithm is: |
19628 | | * - ensure sane alignment of input data |
19629 | | * - copy (conversion happens automatically) input data |
19630 | | * to output |
19631 | | * - update xpp to point at next unconverted input, and tp to point |
19632 | | * at next location for converted output |
19633 | | */ |
19634 | | long i, j, ni; |
19635 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19636 | | uint64 *xp; |
19637 | | int nrange = 0; /* number of range errors */ |
19638 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19639 | | long cxp = (long) *((char**)xpp); |
19640 | | |
19641 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19642 | | /* sjl: manually stripmine so we can limit amount of |
19643 | | * vector work space reserved to LOOPCNT elements. Also |
19644 | | * makes vectorisation easy */ |
19645 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19646 | | ni=Min(nelems-j,LOOPCNT); |
19647 | | if (realign) { |
19648 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19649 | | xp = tmp; |
19650 | | } else { |
19651 | | xp = (uint64 *) *xpp; |
19652 | | } |
19653 | | /* copy the next block */ |
19654 | | #pragma cdir loopcnt=LOOPCNT |
19655 | | #pragma cdir shortloop |
19656 | | for (i=0; i<ni; i++) { |
19657 | | tp[i] = (float) Max( FLOAT_MIN, Min(FLOAT_MAX, (float) xp[i])); |
19658 | | /* test for range errors (not always needed but do it anyway) */ |
19659 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19660 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19661 | | nrange += xp[i] > FLOAT_MAX ; |
19662 | | } |
19663 | | /* update xpp and tp */ |
19664 | | if (realign) xp = (uint64 *) *xpp; |
19665 | | xp += ni; |
19666 | | tp += ni; |
19667 | | *xpp = (void*)xp; |
19668 | | } |
19669 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19670 | | |
19671 | | #else /* not SX */ |
19672 | 0 | const char *xp = (const char *) *xpp; |
19673 | 0 | int status = NC_NOERR; |
19674 | |
|
19675 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19676 | 0 | { |
19677 | 0 | const int lstatus = ncx_get_ulonglong_float(xp, tp); |
19678 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19679 | 0 | status = lstatus; |
19680 | 0 | } |
19681 | |
|
19682 | 0 | *xpp = (const void *)xp; |
19683 | 0 | return status; |
19684 | 0 | #endif |
19685 | 0 | } |
19686 | | |
19687 | | int |
19688 | | ncx_getn_ulonglong_double(const void **xpp, size_t nelems, double *tp) |
19689 | 0 | { |
19690 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19691 | | |
19692 | | /* basic algorithm is: |
19693 | | * - ensure sane alignment of input data |
19694 | | * - copy (conversion happens automatically) input data |
19695 | | * to output |
19696 | | * - update xpp to point at next unconverted input, and tp to point |
19697 | | * at next location for converted output |
19698 | | */ |
19699 | | long i, j, ni; |
19700 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19701 | | uint64 *xp; |
19702 | | int nrange = 0; /* number of range errors */ |
19703 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19704 | | long cxp = (long) *((char**)xpp); |
19705 | | |
19706 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19707 | | /* sjl: manually stripmine so we can limit amount of |
19708 | | * vector work space reserved to LOOPCNT elements. Also |
19709 | | * makes vectorisation easy */ |
19710 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19711 | | ni=Min(nelems-j,LOOPCNT); |
19712 | | if (realign) { |
19713 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19714 | | xp = tmp; |
19715 | | } else { |
19716 | | xp = (uint64 *) *xpp; |
19717 | | } |
19718 | | /* copy the next block */ |
19719 | | #pragma cdir loopcnt=LOOPCNT |
19720 | | #pragma cdir shortloop |
19721 | | for (i=0; i<ni; i++) { |
19722 | | tp[i] = (double) Max( DOUBLE_MIN, Min(DOUBLE_MAX, (double) xp[i])); |
19723 | | /* test for range errors (not always needed but do it anyway) */ |
19724 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19725 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19726 | | nrange += xp[i] > DOUBLE_MAX ; |
19727 | | } |
19728 | | /* update xpp and tp */ |
19729 | | if (realign) xp = (uint64 *) *xpp; |
19730 | | xp += ni; |
19731 | | tp += ni; |
19732 | | *xpp = (void*)xp; |
19733 | | } |
19734 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19735 | | |
19736 | | #else /* not SX */ |
19737 | 0 | const char *xp = (const char *) *xpp; |
19738 | 0 | int status = NC_NOERR; |
19739 | |
|
19740 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19741 | 0 | { |
19742 | 0 | const int lstatus = ncx_get_ulonglong_double(xp, tp); |
19743 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19744 | 0 | status = lstatus; |
19745 | 0 | } |
19746 | |
|
19747 | 0 | *xpp = (const void *)xp; |
19748 | 0 | return status; |
19749 | 0 | #endif |
19750 | 0 | } |
19751 | | |
19752 | | int |
19753 | | ncx_getn_ulonglong_longlong(const void **xpp, size_t nelems, longlong *tp) |
19754 | 0 | { |
19755 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19756 | | |
19757 | | /* basic algorithm is: |
19758 | | * - ensure sane alignment of input data |
19759 | | * - copy (conversion happens automatically) input data |
19760 | | * to output |
19761 | | * - update xpp to point at next unconverted input, and tp to point |
19762 | | * at next location for converted output |
19763 | | */ |
19764 | | long i, j, ni; |
19765 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19766 | | uint64 *xp; |
19767 | | int nrange = 0; /* number of range errors */ |
19768 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19769 | | long cxp = (long) *((char**)xpp); |
19770 | | |
19771 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19772 | | /* sjl: manually stripmine so we can limit amount of |
19773 | | * vector work space reserved to LOOPCNT elements. Also |
19774 | | * makes vectorisation easy */ |
19775 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19776 | | ni=Min(nelems-j,LOOPCNT); |
19777 | | if (realign) { |
19778 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19779 | | xp = tmp; |
19780 | | } else { |
19781 | | xp = (uint64 *) *xpp; |
19782 | | } |
19783 | | /* copy the next block */ |
19784 | | #pragma cdir loopcnt=LOOPCNT |
19785 | | #pragma cdir shortloop |
19786 | | for (i=0; i<ni; i++) { |
19787 | | tp[i] = (longlong) Max( LONGLONG_MIN, Min(LONGLONG_MAX, (longlong) xp[i])); |
19788 | | /* test for range errors (not always needed but do it anyway) */ |
19789 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19790 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19791 | | nrange += xp[i] > LONGLONG_MAX ; |
19792 | | } |
19793 | | /* update xpp and tp */ |
19794 | | if (realign) xp = (uint64 *) *xpp; |
19795 | | xp += ni; |
19796 | | tp += ni; |
19797 | | *xpp = (void*)xp; |
19798 | | } |
19799 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19800 | | |
19801 | | #else /* not SX */ |
19802 | 0 | const char *xp = (const char *) *xpp; |
19803 | 0 | int status = NC_NOERR; |
19804 | |
|
19805 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19806 | 0 | { |
19807 | 0 | const int lstatus = ncx_get_ulonglong_longlong(xp, tp); |
19808 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19809 | 0 | status = lstatus; |
19810 | 0 | } |
19811 | |
|
19812 | 0 | *xpp = (const void *)xp; |
19813 | 0 | return status; |
19814 | 0 | #endif |
19815 | 0 | } |
19816 | | |
19817 | | int |
19818 | | ncx_getn_ulonglong_uchar(const void **xpp, size_t nelems, uchar *tp) |
19819 | 0 | { |
19820 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19821 | | |
19822 | | /* basic algorithm is: |
19823 | | * - ensure sane alignment of input data |
19824 | | * - copy (conversion happens automatically) input data |
19825 | | * to output |
19826 | | * - update xpp to point at next unconverted input, and tp to point |
19827 | | * at next location for converted output |
19828 | | */ |
19829 | | long i, j, ni; |
19830 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19831 | | uint64 *xp; |
19832 | | int nrange = 0; /* number of range errors */ |
19833 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19834 | | long cxp = (long) *((char**)xpp); |
19835 | | |
19836 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19837 | | /* sjl: manually stripmine so we can limit amount of |
19838 | | * vector work space reserved to LOOPCNT elements. Also |
19839 | | * makes vectorisation easy */ |
19840 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19841 | | ni=Min(nelems-j,LOOPCNT); |
19842 | | if (realign) { |
19843 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19844 | | xp = tmp; |
19845 | | } else { |
19846 | | xp = (uint64 *) *xpp; |
19847 | | } |
19848 | | /* copy the next block */ |
19849 | | #pragma cdir loopcnt=LOOPCNT |
19850 | | #pragma cdir shortloop |
19851 | | for (i=0; i<ni; i++) { |
19852 | | tp[i] = (uchar) Max( UCHAR_MIN, Min(UCHAR_MAX, (uchar) xp[i])); |
19853 | | /* test for range errors (not always needed but do it anyway) */ |
19854 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19855 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19856 | | nrange += xp[i] > UCHAR_MAX ; |
19857 | | } |
19858 | | /* update xpp and tp */ |
19859 | | if (realign) xp = (uint64 *) *xpp; |
19860 | | xp += ni; |
19861 | | tp += ni; |
19862 | | *xpp = (void*)xp; |
19863 | | } |
19864 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19865 | | |
19866 | | #else /* not SX */ |
19867 | 0 | const char *xp = (const char *) *xpp; |
19868 | 0 | int status = NC_NOERR; |
19869 | |
|
19870 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19871 | 0 | { |
19872 | 0 | const int lstatus = ncx_get_ulonglong_uchar(xp, tp); |
19873 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19874 | 0 | status = lstatus; |
19875 | 0 | } |
19876 | |
|
19877 | 0 | *xpp = (const void *)xp; |
19878 | 0 | return status; |
19879 | 0 | #endif |
19880 | 0 | } |
19881 | | |
19882 | | int |
19883 | | ncx_getn_ulonglong_ushort(const void **xpp, size_t nelems, ushort *tp) |
19884 | 0 | { |
19885 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19886 | | |
19887 | | /* basic algorithm is: |
19888 | | * - ensure sane alignment of input data |
19889 | | * - copy (conversion happens automatically) input data |
19890 | | * to output |
19891 | | * - update xpp to point at next unconverted input, and tp to point |
19892 | | * at next location for converted output |
19893 | | */ |
19894 | | long i, j, ni; |
19895 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19896 | | uint64 *xp; |
19897 | | int nrange = 0; /* number of range errors */ |
19898 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19899 | | long cxp = (long) *((char**)xpp); |
19900 | | |
19901 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19902 | | /* sjl: manually stripmine so we can limit amount of |
19903 | | * vector work space reserved to LOOPCNT elements. Also |
19904 | | * makes vectorisation easy */ |
19905 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19906 | | ni=Min(nelems-j,LOOPCNT); |
19907 | | if (realign) { |
19908 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19909 | | xp = tmp; |
19910 | | } else { |
19911 | | xp = (uint64 *) *xpp; |
19912 | | } |
19913 | | /* copy the next block */ |
19914 | | #pragma cdir loopcnt=LOOPCNT |
19915 | | #pragma cdir shortloop |
19916 | | for (i=0; i<ni; i++) { |
19917 | | tp[i] = (ushort) Max( USHORT_MIN, Min(USHORT_MAX, (ushort) xp[i])); |
19918 | | /* test for range errors (not always needed but do it anyway) */ |
19919 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19920 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19921 | | nrange += xp[i] > USHORT_MAX ; |
19922 | | } |
19923 | | /* update xpp and tp */ |
19924 | | if (realign) xp = (uint64 *) *xpp; |
19925 | | xp += ni; |
19926 | | tp += ni; |
19927 | | *xpp = (void*)xp; |
19928 | | } |
19929 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19930 | | |
19931 | | #else /* not SX */ |
19932 | 0 | const char *xp = (const char *) *xpp; |
19933 | 0 | int status = NC_NOERR; |
19934 | |
|
19935 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
19936 | 0 | { |
19937 | 0 | const int lstatus = ncx_get_ulonglong_ushort(xp, tp); |
19938 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
19939 | 0 | status = lstatus; |
19940 | 0 | } |
19941 | |
|
19942 | 0 | *xpp = (const void *)xp; |
19943 | 0 | return status; |
19944 | 0 | #endif |
19945 | 0 | } |
19946 | | |
19947 | | int |
19948 | | ncx_getn_ulonglong_uint(const void **xpp, size_t nelems, uint *tp) |
19949 | 0 | { |
19950 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
19951 | | |
19952 | | /* basic algorithm is: |
19953 | | * - ensure sane alignment of input data |
19954 | | * - copy (conversion happens automatically) input data |
19955 | | * to output |
19956 | | * - update xpp to point at next unconverted input, and tp to point |
19957 | | * at next location for converted output |
19958 | | */ |
19959 | | long i, j, ni; |
19960 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
19961 | | uint64 *xp; |
19962 | | int nrange = 0; /* number of range errors */ |
19963 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
19964 | | long cxp = (long) *((char**)xpp); |
19965 | | |
19966 | | realign = (cxp & 7) % SIZEOF_UINT64; |
19967 | | /* sjl: manually stripmine so we can limit amount of |
19968 | | * vector work space reserved to LOOPCNT elements. Also |
19969 | | * makes vectorisation easy */ |
19970 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
19971 | | ni=Min(nelems-j,LOOPCNT); |
19972 | | if (realign) { |
19973 | | memcpy(tmp, *xpp, (size_t)(ni*SIZEOF_UINT64)); |
19974 | | xp = tmp; |
19975 | | } else { |
19976 | | xp = (uint64 *) *xpp; |
19977 | | } |
19978 | | /* copy the next block */ |
19979 | | #pragma cdir loopcnt=LOOPCNT |
19980 | | #pragma cdir shortloop |
19981 | | for (i=0; i<ni; i++) { |
19982 | | tp[i] = (uint) Max( UINT_MIN, Min(UINT_MAX, (uint) xp[i])); |
19983 | | /* test for range errors (not always needed but do it anyway) */ |
19984 | | /* if xpp is unsigned, we need not check if xp[i] < _MIN */ |
19985 | | /* if xpp is signed && tp is unsigned, we need check if xp[i] >= 0 */ |
19986 | | nrange += xp[i] > UINT_MAX ; |
19987 | | } |
19988 | | /* update xpp and tp */ |
19989 | | if (realign) xp = (uint64 *) *xpp; |
19990 | | xp += ni; |
19991 | | tp += ni; |
19992 | | *xpp = (void*)xp; |
19993 | | } |
19994 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
19995 | | |
19996 | | #else /* not SX */ |
19997 | 0 | const char *xp = (const char *) *xpp; |
19998 | 0 | int status = NC_NOERR; |
19999 | |
|
20000 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20001 | 0 | { |
20002 | 0 | const int lstatus = ncx_get_ulonglong_uint(xp, tp); |
20003 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20004 | 0 | status = lstatus; |
20005 | 0 | } |
20006 | |
|
20007 | 0 | *xpp = (const void *)xp; |
20008 | 0 | return status; |
20009 | 0 | #endif |
20010 | 0 | } |
20011 | | |
20012 | | |
20013 | | #if X_SIZEOF_UINT64 == SIZEOF_ULONGLONG |
20014 | | /* optimized version */ |
20015 | | int |
20016 | | ncx_putn_ulonglong_ulonglong(void **xpp, size_t nelems, const unsigned long long *tp, void *fillp) |
20017 | 0 | { |
20018 | | #ifdef WORDS_BIGENDIAN |
20019 | | (void) memcpy(*xpp, tp, (size_t)nelems * X_SIZEOF_UINT64); |
20020 | | # else |
20021 | 0 | swapn8b(*xpp, tp, nelems); |
20022 | 0 | # endif |
20023 | 0 | *xpp = (void *)((char *)(*xpp) + nelems * X_SIZEOF_UINT64); |
20024 | 0 | return NC_NOERR; |
20025 | 0 | } |
20026 | | #else |
20027 | | int |
20028 | | ncx_putn_ulonglong_ulonglong(void **xpp, size_t nelems, const ulonglong *tp, void *fillp) |
20029 | | { |
20030 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20031 | | |
20032 | | /* basic algorithm is: |
20033 | | * - ensure sane alignment of output data |
20034 | | * - copy (conversion happens automatically) input data |
20035 | | * to output |
20036 | | * - update tp to point at next unconverted input, and xpp to point |
20037 | | * at next location for converted output |
20038 | | */ |
20039 | | long i, j, ni; |
20040 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20041 | | uint64 *xp; |
20042 | | int nrange = 0; /* number of range errors */ |
20043 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20044 | | long cxp = (long) *((char**)xpp); |
20045 | | |
20046 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20047 | | /* sjl: manually stripmine so we can limit amount of |
20048 | | * vector work space reserved to LOOPCNT elements. Also |
20049 | | * makes vectorisation easy */ |
20050 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20051 | | ni=Min(nelems-j,LOOPCNT); |
20052 | | if (realign) { |
20053 | | xp = tmp; |
20054 | | } else { |
20055 | | xp = (uint64 *) *xpp; |
20056 | | } |
20057 | | /* copy the next block */ |
20058 | | #pragma cdir loopcnt=LOOPCNT |
20059 | | #pragma cdir shortloop |
20060 | | for (i=0; i<ni; i++) { |
20061 | | /* the normal case: */ |
20062 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20063 | | /* test for range errors (not always needed but do it anyway) */ |
20064 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20065 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20066 | | nrange += tp[i] > X_UINT64_MAX ; |
20067 | | } |
20068 | | /* copy workspace back if necessary */ |
20069 | | if (realign) { |
20070 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20071 | | xp = (uint64 *) *xpp; |
20072 | | } |
20073 | | /* update xpp and tp */ |
20074 | | xp += ni; |
20075 | | tp += ni; |
20076 | | *xpp = (void*)xp; |
20077 | | } |
20078 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20079 | | |
20080 | | #else /* not SX */ |
20081 | | |
20082 | | char *xp = (char *) *xpp; |
20083 | | int status = NC_NOERR; |
20084 | | |
20085 | | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20086 | | { |
20087 | | int lstatus = ncx_put_ulonglong_ulonglong(xp, tp, fillp); |
20088 | | if (status == NC_NOERR) /* report the first encountered error */ |
20089 | | status = lstatus; |
20090 | | } |
20091 | | |
20092 | | *xpp = (void *)xp; |
20093 | | return status; |
20094 | | #endif |
20095 | | } |
20096 | | |
20097 | | #endif |
20098 | | int |
20099 | | ncx_putn_ulonglong_schar(void **xpp, size_t nelems, const schar *tp, void *fillp) |
20100 | 0 | { |
20101 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20102 | | |
20103 | | /* basic algorithm is: |
20104 | | * - ensure sane alignment of output data |
20105 | | * - copy (conversion happens automatically) input data |
20106 | | * to output |
20107 | | * - update tp to point at next unconverted input, and xpp to point |
20108 | | * at next location for converted output |
20109 | | */ |
20110 | | long i, j, ni; |
20111 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20112 | | uint64 *xp; |
20113 | | int nrange = 0; /* number of range errors */ |
20114 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20115 | | long cxp = (long) *((char**)xpp); |
20116 | | |
20117 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20118 | | /* sjl: manually stripmine so we can limit amount of |
20119 | | * vector work space reserved to LOOPCNT elements. Also |
20120 | | * makes vectorisation easy */ |
20121 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20122 | | ni=Min(nelems-j,LOOPCNT); |
20123 | | if (realign) { |
20124 | | xp = tmp; |
20125 | | } else { |
20126 | | xp = (uint64 *) *xpp; |
20127 | | } |
20128 | | /* copy the next block */ |
20129 | | #pragma cdir loopcnt=LOOPCNT |
20130 | | #pragma cdir shortloop |
20131 | | for (i=0; i<ni; i++) { |
20132 | | /* the normal case: */ |
20133 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20134 | | /* test for range errors (not always needed but do it anyway) */ |
20135 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20136 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20137 | | nrange += tp[i] > X_UINT64_MAX || tp[i] < 0; |
20138 | | } |
20139 | | /* copy workspace back if necessary */ |
20140 | | if (realign) { |
20141 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20142 | | xp = (uint64 *) *xpp; |
20143 | | } |
20144 | | /* update xpp and tp */ |
20145 | | xp += ni; |
20146 | | tp += ni; |
20147 | | *xpp = (void*)xp; |
20148 | | } |
20149 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20150 | | |
20151 | | #else /* not SX */ |
20152 | |
|
20153 | 0 | char *xp = (char *) *xpp; |
20154 | 0 | int status = NC_NOERR; |
20155 | |
|
20156 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20157 | 0 | { |
20158 | 0 | int lstatus = ncx_put_ulonglong_schar(xp, tp, fillp); |
20159 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20160 | 0 | status = lstatus; |
20161 | 0 | } |
20162 | |
|
20163 | 0 | *xpp = (void *)xp; |
20164 | 0 | return status; |
20165 | 0 | #endif |
20166 | 0 | } |
20167 | | |
20168 | | int |
20169 | | ncx_putn_ulonglong_short(void **xpp, size_t nelems, const short *tp, void *fillp) |
20170 | 0 | { |
20171 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20172 | | |
20173 | | /* basic algorithm is: |
20174 | | * - ensure sane alignment of output data |
20175 | | * - copy (conversion happens automatically) input data |
20176 | | * to output |
20177 | | * - update tp to point at next unconverted input, and xpp to point |
20178 | | * at next location for converted output |
20179 | | */ |
20180 | | long i, j, ni; |
20181 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20182 | | uint64 *xp; |
20183 | | int nrange = 0; /* number of range errors */ |
20184 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20185 | | long cxp = (long) *((char**)xpp); |
20186 | | |
20187 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20188 | | /* sjl: manually stripmine so we can limit amount of |
20189 | | * vector work space reserved to LOOPCNT elements. Also |
20190 | | * makes vectorisation easy */ |
20191 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20192 | | ni=Min(nelems-j,LOOPCNT); |
20193 | | if (realign) { |
20194 | | xp = tmp; |
20195 | | } else { |
20196 | | xp = (uint64 *) *xpp; |
20197 | | } |
20198 | | /* copy the next block */ |
20199 | | #pragma cdir loopcnt=LOOPCNT |
20200 | | #pragma cdir shortloop |
20201 | | for (i=0; i<ni; i++) { |
20202 | | /* the normal case: */ |
20203 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20204 | | /* test for range errors (not always needed but do it anyway) */ |
20205 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20206 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20207 | | nrange += tp[i] > X_UINT64_MAX || tp[i] < 0; |
20208 | | } |
20209 | | /* copy workspace back if necessary */ |
20210 | | if (realign) { |
20211 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20212 | | xp = (uint64 *) *xpp; |
20213 | | } |
20214 | | /* update xpp and tp */ |
20215 | | xp += ni; |
20216 | | tp += ni; |
20217 | | *xpp = (void*)xp; |
20218 | | } |
20219 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20220 | | |
20221 | | #else /* not SX */ |
20222 | |
|
20223 | 0 | char *xp = (char *) *xpp; |
20224 | 0 | int status = NC_NOERR; |
20225 | |
|
20226 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20227 | 0 | { |
20228 | 0 | int lstatus = ncx_put_ulonglong_short(xp, tp, fillp); |
20229 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20230 | 0 | status = lstatus; |
20231 | 0 | } |
20232 | |
|
20233 | 0 | *xpp = (void *)xp; |
20234 | 0 | return status; |
20235 | 0 | #endif |
20236 | 0 | } |
20237 | | |
20238 | | int |
20239 | | ncx_putn_ulonglong_int(void **xpp, size_t nelems, const int *tp, void *fillp) |
20240 | 0 | { |
20241 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20242 | | |
20243 | | /* basic algorithm is: |
20244 | | * - ensure sane alignment of output data |
20245 | | * - copy (conversion happens automatically) input data |
20246 | | * to output |
20247 | | * - update tp to point at next unconverted input, and xpp to point |
20248 | | * at next location for converted output |
20249 | | */ |
20250 | | long i, j, ni; |
20251 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20252 | | uint64 *xp; |
20253 | | int nrange = 0; /* number of range errors */ |
20254 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20255 | | long cxp = (long) *((char**)xpp); |
20256 | | |
20257 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20258 | | /* sjl: manually stripmine so we can limit amount of |
20259 | | * vector work space reserved to LOOPCNT elements. Also |
20260 | | * makes vectorisation easy */ |
20261 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20262 | | ni=Min(nelems-j,LOOPCNT); |
20263 | | if (realign) { |
20264 | | xp = tmp; |
20265 | | } else { |
20266 | | xp = (uint64 *) *xpp; |
20267 | | } |
20268 | | /* copy the next block */ |
20269 | | #pragma cdir loopcnt=LOOPCNT |
20270 | | #pragma cdir shortloop |
20271 | | for (i=0; i<ni; i++) { |
20272 | | /* the normal case: */ |
20273 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20274 | | /* test for range errors (not always needed but do it anyway) */ |
20275 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20276 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20277 | | nrange += tp[i] > X_UINT64_MAX || tp[i] < 0; |
20278 | | } |
20279 | | /* copy workspace back if necessary */ |
20280 | | if (realign) { |
20281 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20282 | | xp = (uint64 *) *xpp; |
20283 | | } |
20284 | | /* update xpp and tp */ |
20285 | | xp += ni; |
20286 | | tp += ni; |
20287 | | *xpp = (void*)xp; |
20288 | | } |
20289 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20290 | | |
20291 | | #else /* not SX */ |
20292 | |
|
20293 | 0 | char *xp = (char *) *xpp; |
20294 | 0 | int status = NC_NOERR; |
20295 | |
|
20296 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20297 | 0 | { |
20298 | 0 | int lstatus = ncx_put_ulonglong_int(xp, tp, fillp); |
20299 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20300 | 0 | status = lstatus; |
20301 | 0 | } |
20302 | |
|
20303 | 0 | *xpp = (void *)xp; |
20304 | 0 | return status; |
20305 | 0 | #endif |
20306 | 0 | } |
20307 | | |
20308 | | int |
20309 | | ncx_putn_ulonglong_long(void **xpp, size_t nelems, const long *tp, void *fillp) |
20310 | 0 | { |
20311 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20312 | | |
20313 | | /* basic algorithm is: |
20314 | | * - ensure sane alignment of output data |
20315 | | * - copy (conversion happens automatically) input data |
20316 | | * to output |
20317 | | * - update tp to point at next unconverted input, and xpp to point |
20318 | | * at next location for converted output |
20319 | | */ |
20320 | | long i, j, ni; |
20321 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20322 | | uint64 *xp; |
20323 | | int nrange = 0; /* number of range errors */ |
20324 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20325 | | long cxp = (long) *((char**)xpp); |
20326 | | |
20327 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20328 | | /* sjl: manually stripmine so we can limit amount of |
20329 | | * vector work space reserved to LOOPCNT elements. Also |
20330 | | * makes vectorisation easy */ |
20331 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20332 | | ni=Min(nelems-j,LOOPCNT); |
20333 | | if (realign) { |
20334 | | xp = tmp; |
20335 | | } else { |
20336 | | xp = (uint64 *) *xpp; |
20337 | | } |
20338 | | /* copy the next block */ |
20339 | | #pragma cdir loopcnt=LOOPCNT |
20340 | | #pragma cdir shortloop |
20341 | | for (i=0; i<ni; i++) { |
20342 | | /* the normal case: */ |
20343 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20344 | | /* test for range errors (not always needed but do it anyway) */ |
20345 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20346 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20347 | | nrange += tp[i] > X_UINT64_MAX || tp[i] < 0; |
20348 | | } |
20349 | | /* copy workspace back if necessary */ |
20350 | | if (realign) { |
20351 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20352 | | xp = (uint64 *) *xpp; |
20353 | | } |
20354 | | /* update xpp and tp */ |
20355 | | xp += ni; |
20356 | | tp += ni; |
20357 | | *xpp = (void*)xp; |
20358 | | } |
20359 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20360 | | |
20361 | | #else /* not SX */ |
20362 | |
|
20363 | 0 | char *xp = (char *) *xpp; |
20364 | 0 | int status = NC_NOERR; |
20365 | |
|
20366 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20367 | 0 | { |
20368 | 0 | int lstatus = ncx_put_ulonglong_long(xp, tp, fillp); |
20369 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20370 | 0 | status = lstatus; |
20371 | 0 | } |
20372 | |
|
20373 | 0 | *xpp = (void *)xp; |
20374 | 0 | return status; |
20375 | 0 | #endif |
20376 | 0 | } |
20377 | | |
20378 | | int |
20379 | | ncx_putn_ulonglong_float(void **xpp, size_t nelems, const float *tp, void *fillp) |
20380 | 0 | { |
20381 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20382 | | |
20383 | | /* basic algorithm is: |
20384 | | * - ensure sane alignment of output data |
20385 | | * - copy (conversion happens automatically) input data |
20386 | | * to output |
20387 | | * - update tp to point at next unconverted input, and xpp to point |
20388 | | * at next location for converted output |
20389 | | */ |
20390 | | long i, j, ni; |
20391 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20392 | | uint64 *xp; |
20393 | | int nrange = 0; /* number of range errors */ |
20394 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20395 | | long cxp = (long) *((char**)xpp); |
20396 | | |
20397 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20398 | | /* sjl: manually stripmine so we can limit amount of |
20399 | | * vector work space reserved to LOOPCNT elements. Also |
20400 | | * makes vectorisation easy */ |
20401 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20402 | | ni=Min(nelems-j,LOOPCNT); |
20403 | | if (realign) { |
20404 | | xp = tmp; |
20405 | | } else { |
20406 | | xp = (uint64 *) *xpp; |
20407 | | } |
20408 | | /* copy the next block */ |
20409 | | #pragma cdir loopcnt=LOOPCNT |
20410 | | #pragma cdir shortloop |
20411 | | for (i=0; i<ni; i++) { |
20412 | | /* the normal case: */ |
20413 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20414 | | /* test for range errors (not always needed but do it anyway) */ |
20415 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20416 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20417 | | nrange += tp[i] > X_UINT64_MAX || tp[i] < 0; |
20418 | | } |
20419 | | /* copy workspace back if necessary */ |
20420 | | if (realign) { |
20421 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20422 | | xp = (uint64 *) *xpp; |
20423 | | } |
20424 | | /* update xpp and tp */ |
20425 | | xp += ni; |
20426 | | tp += ni; |
20427 | | *xpp = (void*)xp; |
20428 | | } |
20429 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20430 | | |
20431 | | #else /* not SX */ |
20432 | |
|
20433 | 0 | char *xp = (char *) *xpp; |
20434 | 0 | int status = NC_NOERR; |
20435 | |
|
20436 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20437 | 0 | { |
20438 | 0 | int lstatus = ncx_put_ulonglong_float(xp, tp, fillp); |
20439 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20440 | 0 | status = lstatus; |
20441 | 0 | } |
20442 | |
|
20443 | 0 | *xpp = (void *)xp; |
20444 | 0 | return status; |
20445 | 0 | #endif |
20446 | 0 | } |
20447 | | |
20448 | | int |
20449 | | ncx_putn_ulonglong_double(void **xpp, size_t nelems, const double *tp, void *fillp) |
20450 | 0 | { |
20451 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20452 | | |
20453 | | /* basic algorithm is: |
20454 | | * - ensure sane alignment of output data |
20455 | | * - copy (conversion happens automatically) input data |
20456 | | * to output |
20457 | | * - update tp to point at next unconverted input, and xpp to point |
20458 | | * at next location for converted output |
20459 | | */ |
20460 | | long i, j, ni; |
20461 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20462 | | uint64 *xp; |
20463 | | int nrange = 0; /* number of range errors */ |
20464 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20465 | | long cxp = (long) *((char**)xpp); |
20466 | | |
20467 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20468 | | /* sjl: manually stripmine so we can limit amount of |
20469 | | * vector work space reserved to LOOPCNT elements. Also |
20470 | | * makes vectorisation easy */ |
20471 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20472 | | ni=Min(nelems-j,LOOPCNT); |
20473 | | if (realign) { |
20474 | | xp = tmp; |
20475 | | } else { |
20476 | | xp = (uint64 *) *xpp; |
20477 | | } |
20478 | | /* copy the next block */ |
20479 | | #pragma cdir loopcnt=LOOPCNT |
20480 | | #pragma cdir shortloop |
20481 | | for (i=0; i<ni; i++) { |
20482 | | /* the normal case: */ |
20483 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20484 | | /* test for range errors (not always needed but do it anyway) */ |
20485 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20486 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20487 | | nrange += tp[i] > X_UINT64_MAX || tp[i] < 0; |
20488 | | } |
20489 | | /* copy workspace back if necessary */ |
20490 | | if (realign) { |
20491 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20492 | | xp = (uint64 *) *xpp; |
20493 | | } |
20494 | | /* update xpp and tp */ |
20495 | | xp += ni; |
20496 | | tp += ni; |
20497 | | *xpp = (void*)xp; |
20498 | | } |
20499 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20500 | | |
20501 | | #else /* not SX */ |
20502 | |
|
20503 | 0 | char *xp = (char *) *xpp; |
20504 | 0 | int status = NC_NOERR; |
20505 | |
|
20506 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20507 | 0 | { |
20508 | 0 | int lstatus = ncx_put_ulonglong_double(xp, tp, fillp); |
20509 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20510 | 0 | status = lstatus; |
20511 | 0 | } |
20512 | |
|
20513 | 0 | *xpp = (void *)xp; |
20514 | 0 | return status; |
20515 | 0 | #endif |
20516 | 0 | } |
20517 | | |
20518 | | int |
20519 | | ncx_putn_ulonglong_longlong(void **xpp, size_t nelems, const longlong *tp, void *fillp) |
20520 | 0 | { |
20521 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20522 | | |
20523 | | /* basic algorithm is: |
20524 | | * - ensure sane alignment of output data |
20525 | | * - copy (conversion happens automatically) input data |
20526 | | * to output |
20527 | | * - update tp to point at next unconverted input, and xpp to point |
20528 | | * at next location for converted output |
20529 | | */ |
20530 | | long i, j, ni; |
20531 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20532 | | uint64 *xp; |
20533 | | int nrange = 0; /* number of range errors */ |
20534 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20535 | | long cxp = (long) *((char**)xpp); |
20536 | | |
20537 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20538 | | /* sjl: manually stripmine so we can limit amount of |
20539 | | * vector work space reserved to LOOPCNT elements. Also |
20540 | | * makes vectorisation easy */ |
20541 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20542 | | ni=Min(nelems-j,LOOPCNT); |
20543 | | if (realign) { |
20544 | | xp = tmp; |
20545 | | } else { |
20546 | | xp = (uint64 *) *xpp; |
20547 | | } |
20548 | | /* copy the next block */ |
20549 | | #pragma cdir loopcnt=LOOPCNT |
20550 | | #pragma cdir shortloop |
20551 | | for (i=0; i<ni; i++) { |
20552 | | /* the normal case: */ |
20553 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20554 | | /* test for range errors (not always needed but do it anyway) */ |
20555 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20556 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20557 | | nrange += tp[i] > X_UINT64_MAX || tp[i] < 0; |
20558 | | } |
20559 | | /* copy workspace back if necessary */ |
20560 | | if (realign) { |
20561 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20562 | | xp = (uint64 *) *xpp; |
20563 | | } |
20564 | | /* update xpp and tp */ |
20565 | | xp += ni; |
20566 | | tp += ni; |
20567 | | *xpp = (void*)xp; |
20568 | | } |
20569 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20570 | | |
20571 | | #else /* not SX */ |
20572 | |
|
20573 | 0 | char *xp = (char *) *xpp; |
20574 | 0 | int status = NC_NOERR; |
20575 | |
|
20576 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20577 | 0 | { |
20578 | 0 | int lstatus = ncx_put_ulonglong_longlong(xp, tp, fillp); |
20579 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20580 | 0 | status = lstatus; |
20581 | 0 | } |
20582 | |
|
20583 | 0 | *xpp = (void *)xp; |
20584 | 0 | return status; |
20585 | 0 | #endif |
20586 | 0 | } |
20587 | | |
20588 | | int |
20589 | | ncx_putn_ulonglong_uchar(void **xpp, size_t nelems, const uchar *tp, void *fillp) |
20590 | 0 | { |
20591 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20592 | | |
20593 | | /* basic algorithm is: |
20594 | | * - ensure sane alignment of output data |
20595 | | * - copy (conversion happens automatically) input data |
20596 | | * to output |
20597 | | * - update tp to point at next unconverted input, and xpp to point |
20598 | | * at next location for converted output |
20599 | | */ |
20600 | | long i, j, ni; |
20601 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20602 | | uint64 *xp; |
20603 | | int nrange = 0; /* number of range errors */ |
20604 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20605 | | long cxp = (long) *((char**)xpp); |
20606 | | |
20607 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20608 | | /* sjl: manually stripmine so we can limit amount of |
20609 | | * vector work space reserved to LOOPCNT elements. Also |
20610 | | * makes vectorisation easy */ |
20611 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20612 | | ni=Min(nelems-j,LOOPCNT); |
20613 | | if (realign) { |
20614 | | xp = tmp; |
20615 | | } else { |
20616 | | xp = (uint64 *) *xpp; |
20617 | | } |
20618 | | /* copy the next block */ |
20619 | | #pragma cdir loopcnt=LOOPCNT |
20620 | | #pragma cdir shortloop |
20621 | | for (i=0; i<ni; i++) { |
20622 | | /* the normal case: */ |
20623 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20624 | | /* test for range errors (not always needed but do it anyway) */ |
20625 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20626 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20627 | | nrange += tp[i] > X_UINT64_MAX ; |
20628 | | } |
20629 | | /* copy workspace back if necessary */ |
20630 | | if (realign) { |
20631 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20632 | | xp = (uint64 *) *xpp; |
20633 | | } |
20634 | | /* update xpp and tp */ |
20635 | | xp += ni; |
20636 | | tp += ni; |
20637 | | *xpp = (void*)xp; |
20638 | | } |
20639 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20640 | | |
20641 | | #else /* not SX */ |
20642 | |
|
20643 | 0 | char *xp = (char *) *xpp; |
20644 | 0 | int status = NC_NOERR; |
20645 | |
|
20646 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20647 | 0 | { |
20648 | 0 | int lstatus = ncx_put_ulonglong_uchar(xp, tp, fillp); |
20649 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20650 | 0 | status = lstatus; |
20651 | 0 | } |
20652 | |
|
20653 | 0 | *xpp = (void *)xp; |
20654 | 0 | return status; |
20655 | 0 | #endif |
20656 | 0 | } |
20657 | | |
20658 | | int |
20659 | | ncx_putn_ulonglong_ushort(void **xpp, size_t nelems, const ushort *tp, void *fillp) |
20660 | 0 | { |
20661 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20662 | | |
20663 | | /* basic algorithm is: |
20664 | | * - ensure sane alignment of output data |
20665 | | * - copy (conversion happens automatically) input data |
20666 | | * to output |
20667 | | * - update tp to point at next unconverted input, and xpp to point |
20668 | | * at next location for converted output |
20669 | | */ |
20670 | | long i, j, ni; |
20671 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20672 | | uint64 *xp; |
20673 | | int nrange = 0; /* number of range errors */ |
20674 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20675 | | long cxp = (long) *((char**)xpp); |
20676 | | |
20677 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20678 | | /* sjl: manually stripmine so we can limit amount of |
20679 | | * vector work space reserved to LOOPCNT elements. Also |
20680 | | * makes vectorisation easy */ |
20681 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20682 | | ni=Min(nelems-j,LOOPCNT); |
20683 | | if (realign) { |
20684 | | xp = tmp; |
20685 | | } else { |
20686 | | xp = (uint64 *) *xpp; |
20687 | | } |
20688 | | /* copy the next block */ |
20689 | | #pragma cdir loopcnt=LOOPCNT |
20690 | | #pragma cdir shortloop |
20691 | | for (i=0; i<ni; i++) { |
20692 | | /* the normal case: */ |
20693 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20694 | | /* test for range errors (not always needed but do it anyway) */ |
20695 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20696 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20697 | | nrange += tp[i] > X_UINT64_MAX ; |
20698 | | } |
20699 | | /* copy workspace back if necessary */ |
20700 | | if (realign) { |
20701 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20702 | | xp = (uint64 *) *xpp; |
20703 | | } |
20704 | | /* update xpp and tp */ |
20705 | | xp += ni; |
20706 | | tp += ni; |
20707 | | *xpp = (void*)xp; |
20708 | | } |
20709 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20710 | | |
20711 | | #else /* not SX */ |
20712 | |
|
20713 | 0 | char *xp = (char *) *xpp; |
20714 | 0 | int status = NC_NOERR; |
20715 | |
|
20716 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20717 | 0 | { |
20718 | 0 | int lstatus = ncx_put_ulonglong_ushort(xp, tp, fillp); |
20719 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20720 | 0 | status = lstatus; |
20721 | 0 | } |
20722 | |
|
20723 | 0 | *xpp = (void *)xp; |
20724 | 0 | return status; |
20725 | 0 | #endif |
20726 | 0 | } |
20727 | | |
20728 | | int |
20729 | | ncx_putn_ulonglong_uint(void **xpp, size_t nelems, const uint *tp, void *fillp) |
20730 | 0 | { |
20731 | | #if defined(_SX) && _SX != 0 && X_SIZEOF_UINT64 == SIZEOF_UINT64 |
20732 | | |
20733 | | /* basic algorithm is: |
20734 | | * - ensure sane alignment of output data |
20735 | | * - copy (conversion happens automatically) input data |
20736 | | * to output |
20737 | | * - update tp to point at next unconverted input, and xpp to point |
20738 | | * at next location for converted output |
20739 | | */ |
20740 | | long i, j, ni; |
20741 | | uint64 tmp[LOOPCNT]; /* in case input is misaligned */ |
20742 | | uint64 *xp; |
20743 | | int nrange = 0; /* number of range errors */ |
20744 | | int realign = 0; /* "do we need to fix input data alignment?" */ |
20745 | | long cxp = (long) *((char**)xpp); |
20746 | | |
20747 | | realign = (cxp & 7) % SIZEOF_UINT64; |
20748 | | /* sjl: manually stripmine so we can limit amount of |
20749 | | * vector work space reserved to LOOPCNT elements. Also |
20750 | | * makes vectorisation easy */ |
20751 | | for (j=0; j<nelems && nrange==0; j+=LOOPCNT) { |
20752 | | ni=Min(nelems-j,LOOPCNT); |
20753 | | if (realign) { |
20754 | | xp = tmp; |
20755 | | } else { |
20756 | | xp = (uint64 *) *xpp; |
20757 | | } |
20758 | | /* copy the next block */ |
20759 | | #pragma cdir loopcnt=LOOPCNT |
20760 | | #pragma cdir shortloop |
20761 | | for (i=0; i<ni; i++) { |
20762 | | /* the normal case: */ |
20763 | | xp[i] = (uint64) Max( X_UINT64_MIN, Min(X_UINT64_MAX, (uint64) tp[i])); |
20764 | | /* test for range errors (not always needed but do it anyway) */ |
20765 | | /* if xpp is unsigned && tp is signed, we need check if tp[i] >= 0 */ |
20766 | | /* if tp is unsigned, we need not check if tp[i] < X__MIN */ |
20767 | | nrange += tp[i] > X_UINT64_MAX ; |
20768 | | } |
20769 | | /* copy workspace back if necessary */ |
20770 | | if (realign) { |
20771 | | memcpy(*xpp, tmp, (size_t)*ni*X_SIZEOF_UINT64); |
20772 | | xp = (uint64 *) *xpp; |
20773 | | } |
20774 | | /* update xpp and tp */ |
20775 | | xp += ni; |
20776 | | tp += ni; |
20777 | | *xpp = (void*)xp; |
20778 | | } |
20779 | | return nrange == 0 ? NC_NOERR : NC_ERANGE; |
20780 | | |
20781 | | #else /* not SX */ |
20782 | |
|
20783 | 0 | char *xp = (char *) *xpp; |
20784 | 0 | int status = NC_NOERR; |
20785 | |
|
20786 | 0 | for( ; nelems != 0; nelems--, xp += X_SIZEOF_UINT64, tp++) |
20787 | 0 | { |
20788 | 0 | int lstatus = ncx_put_ulonglong_uint(xp, tp, fillp); |
20789 | 0 | if (status == NC_NOERR) /* report the first encountered error */ |
20790 | 0 | status = lstatus; |
20791 | 0 | } |
20792 | |
|
20793 | 0 | *xpp = (void *)xp; |
20794 | 0 | return status; |
20795 | 0 | #endif |
20796 | 0 | } |
20797 | | |
20798 | | |
20799 | | |
20800 | | /* |
20801 | | * Other aggregate conversion functions. |
20802 | | */ |
20803 | | |
20804 | | /* text */ |
20805 | | |
20806 | | int |
20807 | | ncx_getn_text(const void **xpp, size_t nelems, char *tp) |
20808 | 0 | { |
20809 | 0 | (void) memcpy(tp, *xpp, (size_t)nelems); |
20810 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
20811 | 0 | return NC_NOERR; |
20812 | |
|
20813 | 0 | } |
20814 | | |
20815 | | int |
20816 | | ncx_pad_getn_text(const void **xpp, size_t nelems, char *tp) |
20817 | 602k | { |
20818 | 602k | size_t rndup = nelems % X_ALIGN; |
20819 | | |
20820 | 602k | if (rndup) |
20821 | 4.87k | rndup = X_ALIGN - rndup; |
20822 | | |
20823 | 602k | (void) memcpy(tp, *xpp, (size_t)nelems); |
20824 | 602k | *xpp = (void *)((char *)(*xpp) + nelems + rndup); |
20825 | | |
20826 | 602k | return NC_NOERR; |
20827 | | |
20828 | 602k | } |
20829 | | |
20830 | | int |
20831 | | ncx_putn_text(void **xpp, size_t nelems, const char *tp) |
20832 | 0 | { |
20833 | 0 | (void) memcpy(*xpp, tp, (size_t)nelems); |
20834 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
20835 | |
|
20836 | 0 | return NC_NOERR; |
20837 | |
|
20838 | 0 | } |
20839 | | |
20840 | | int |
20841 | | ncx_pad_putn_text(void **xpp, size_t nelems, const char *tp) |
20842 | 0 | { |
20843 | 0 | size_t rndup = nelems % X_ALIGN; |
20844 | |
|
20845 | 0 | if (rndup) |
20846 | 0 | rndup = X_ALIGN - rndup; |
20847 | |
|
20848 | 0 | (void) memcpy(*xpp, tp, (size_t)nelems); |
20849 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
20850 | |
|
20851 | 0 | if (rndup) |
20852 | 0 | { |
20853 | 0 | (void) memcpy(*xpp, nada, (size_t)rndup); |
20854 | 0 | *xpp = (void *)((char *)(*xpp) + rndup); |
20855 | 0 | } |
20856 | |
|
20857 | 0 | return NC_NOERR; |
20858 | |
|
20859 | 0 | } |
20860 | | |
20861 | | |
20862 | | /* opaque */ |
20863 | | |
20864 | | int |
20865 | | ncx_getn_void(const void **xpp, size_t nelems, void *tp) |
20866 | 0 | { |
20867 | 0 | (void) memcpy(tp, *xpp, (size_t)nelems); |
20868 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
20869 | 0 | return NC_NOERR; |
20870 | |
|
20871 | 0 | } |
20872 | | |
20873 | | int |
20874 | | ncx_pad_getn_void(const void **xpp, size_t nelems, void *tp) |
20875 | 0 | { |
20876 | 0 | size_t rndup = nelems % X_ALIGN; |
20877 | |
|
20878 | 0 | if (rndup) |
20879 | 0 | rndup = X_ALIGN - rndup; |
20880 | |
|
20881 | 0 | (void) memcpy(tp, *xpp, (size_t)nelems); |
20882 | 0 | *xpp = (void *)((char *)(*xpp) + nelems + rndup); |
20883 | |
|
20884 | 0 | return NC_NOERR; |
20885 | |
|
20886 | 0 | } |
20887 | | |
20888 | | int |
20889 | | ncx_putn_void(void **xpp, size_t nelems, const void *tp) |
20890 | 0 | { |
20891 | 0 | (void) memcpy(*xpp, tp, (size_t)nelems); |
20892 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
20893 | |
|
20894 | 0 | return NC_NOERR; |
20895 | |
|
20896 | 0 | } |
20897 | | |
20898 | | int |
20899 | | ncx_pad_putn_void(void **xpp, size_t nelems, const void *tp) |
20900 | 0 | { |
20901 | 0 | size_t rndup = nelems % X_ALIGN; |
20902 | |
|
20903 | 0 | if (rndup) |
20904 | 0 | rndup = X_ALIGN - rndup; |
20905 | |
|
20906 | 0 | (void) memcpy(*xpp, tp, (size_t)nelems); |
20907 | 0 | *xpp = (void *)((char *)(*xpp) + nelems); |
20908 | |
|
20909 | 0 | if (rndup) |
20910 | 0 | { |
20911 | 0 | (void) memcpy(*xpp, nada, (size_t)rndup); |
20912 | 0 | *xpp = (void *)((char *)(*xpp) + rndup); |
20913 | 0 | } |
20914 | |
|
20915 | 0 | return NC_NOERR; |
20916 | |
|
20917 | 0 | } |