/src/qemu/target/i386/tcg/fpu_helper.c
Line | Count | Source |
1 | | /* |
2 | | * x86 FPU, MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI helpers |
3 | | * |
4 | | * Copyright (c) 2003 Fabrice Bellard |
5 | | * |
6 | | * This library is free software; you can redistribute it and/or |
7 | | * modify it under the terms of the GNU Lesser General Public |
8 | | * License as published by the Free Software Foundation; either |
9 | | * version 2.1 of the License, or (at your option) any later version. |
10 | | * |
11 | | * This library is distributed in the hope that it will be useful, |
12 | | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | | * Lesser General Public License for more details. |
15 | | * |
16 | | * You should have received a copy of the GNU Lesser General Public |
17 | | * License along with this library; if not, see <http://www.gnu.org/licenses/>. |
18 | | */ |
19 | | |
20 | | #include "qemu/osdep.h" |
21 | | #include <math.h> |
22 | | #include "cpu.h" |
23 | | #include "tcg-cpu.h" |
24 | | #include "exec/cputlb.h" |
25 | | #include "accel/tcg/cpu-ldst.h" |
26 | | #include "exec/helper-proto.h" |
27 | | #include "fpu/softfloat.h" |
28 | | #include "fpu/softfloat-macros.h" |
29 | | #include "helper-tcg.h" |
30 | | #include "access.h" |
31 | | |
32 | | /* float macros */ |
33 | 0 | #define FT0 (env->ft0) |
34 | 0 | #define ST0 (env->fpregs[env->fpstt].d) |
35 | 0 | #define ST(n) (env->fpregs[(env->fpstt + (n)) & 7].d) |
36 | 0 | #define ST1 ST(1) |
37 | | |
38 | 0 | #define FPU_RC_SHIFT 10 |
39 | 0 | #define FPU_RC_MASK (3 << FPU_RC_SHIFT) |
40 | | #define FPU_RC_NEAR 0x000 |
41 | 0 | #define FPU_RC_DOWN 0x400 |
42 | 0 | #define FPU_RC_UP 0x800 |
43 | 0 | #define FPU_RC_CHOP 0xc00 |
44 | | |
45 | 0 | #define MAXTAN 9223372036854775808.0 |
46 | | |
47 | | /* the following deal with x86 long double-precision numbers */ |
48 | 0 | #define MAXEXPD 0x7fff |
49 | 0 | #define EXPBIAS 16383 |
50 | 0 | #define EXPD(fp) (fp.l.upper & 0x7fff) |
51 | 0 | #define SIGND(fp) ((fp.l.upper) & 0x8000) |
52 | 0 | #define MANTD(fp) (fp.l.lower) |
53 | 0 | #define BIASEXPONENT(fp) fp.l.upper = (fp.l.upper & ~(0x7fff)) | EXPBIAS |
54 | | |
55 | 0 | #define FPUS_IE (1 << 0) |
56 | 0 | #define FPUS_DE (1 << 1) |
57 | 0 | #define FPUS_ZE (1 << 2) |
58 | 0 | #define FPUS_OE (1 << 3) |
59 | 0 | #define FPUS_UE (1 << 4) |
60 | 0 | #define FPUS_PE (1 << 5) |
61 | | #define FPUS_SF (1 << 6) |
62 | 0 | #define FPUS_SE (1 << 7) |
63 | 0 | #define FPUS_B (1 << 15) |
64 | | |
65 | 0 | #define FPUC_EM 0x3f |
66 | | |
67 | 0 | #define floatx80_lg2 make_floatx80(0x3ffd, 0x9a209a84fbcff799LL) |
68 | 0 | #define floatx80_lg2_d make_floatx80(0x3ffd, 0x9a209a84fbcff798LL) |
69 | 0 | #define floatx80_l2e make_floatx80(0x3fff, 0xb8aa3b295c17f0bcLL) |
70 | 0 | #define floatx80_l2e_d make_floatx80(0x3fff, 0xb8aa3b295c17f0bbLL) |
71 | 0 | #define floatx80_l2t make_floatx80(0x4000, 0xd49a784bcd1b8afeLL) |
72 | 0 | #define floatx80_l2t_u make_floatx80(0x4000, 0xd49a784bcd1b8affLL) |
73 | 0 | #define floatx80_ln2_d make_floatx80(0x3ffe, 0xb17217f7d1cf79abLL) |
74 | 0 | #define floatx80_pi_d make_floatx80(0x4000, 0xc90fdaa22168c234LL) |
75 | | |
76 | | static inline void fpush(CPUX86State *env) |
77 | 0 | { |
78 | 0 | env->fpstt = (env->fpstt - 1) & 7; |
79 | 0 | env->fptags[env->fpstt] = 0; /* validate stack entry */ |
80 | 0 | } |
81 | | |
82 | | static inline void fpop(CPUX86State *env) |
83 | 0 | { |
84 | 0 | env->fptags[env->fpstt] = 1; /* invalidate stack entry */ |
85 | 0 | env->fpstt = (env->fpstt + 1) & 7; |
86 | 0 | } |
87 | | |
88 | | static floatx80 do_fldt(X86Access *ac, target_ulong ptr) |
89 | 0 | { |
90 | 0 | CPU_LDoubleU temp; |
91 | |
|
92 | 0 | temp.l.lower = access_ldq(ac, ptr); |
93 | 0 | temp.l.upper = access_ldw(ac, ptr + 8); |
94 | 0 | return temp.d; |
95 | 0 | } |
96 | | |
97 | | static void do_fstt(X86Access *ac, target_ulong ptr, floatx80 f) |
98 | 0 | { |
99 | 0 | CPU_LDoubleU temp; |
100 | |
|
101 | 0 | temp.d = f; |
102 | 0 | access_stq(ac, ptr, temp.l.lower); |
103 | 0 | access_stw(ac, ptr + 8, temp.l.upper); |
104 | 0 | } |
105 | | |
106 | | /* x87 FPU helpers */ |
107 | | |
108 | | static inline double floatx80_to_double(CPUX86State *env, floatx80 a) |
109 | 0 | { |
110 | 0 | union { |
111 | 0 | float64 f64; |
112 | 0 | double d; |
113 | 0 | } u; |
114 | |
|
115 | 0 | u.f64 = floatx80_to_float64(a, &env->fp_status); |
116 | 0 | return u.d; |
117 | 0 | } |
118 | | |
119 | | static inline floatx80 double_to_floatx80(CPUX86State *env, double a) |
120 | 0 | { |
121 | 0 | union { |
122 | 0 | float64 f64; |
123 | 0 | double d; |
124 | 0 | } u; |
125 | |
|
126 | 0 | u.d = a; |
127 | 0 | return float64_to_floatx80(u.f64, &env->fp_status); |
128 | 0 | } |
129 | | |
130 | | static void fpu_set_exception(CPUX86State *env, int mask) |
131 | 0 | { |
132 | 0 | env->fpus |= mask; |
133 | 0 | if (env->fpus & (~env->fpuc & FPUC_EM)) { |
134 | 0 | env->fpus |= FPUS_SE | FPUS_B; |
135 | 0 | } |
136 | 0 | } |
137 | | |
138 | | void cpu_init_fp_statuses(CPUX86State *env) |
139 | 0 | { |
140 | | /* |
141 | | * Initialise the non-runtime-varying fields of the various |
142 | | * float_status words to x86 behaviour. This must be called at |
143 | | * CPU reset because the float_status words are in the |
144 | | * "zeroed on reset" portion of the CPU state struct. |
145 | | * Fields in float_status that vary under guest control are set |
146 | | * via the codepath for setting that register, eg cpu_set_fpuc(). |
147 | | */ |
148 | | /* |
149 | | * Use x87 NaN propagation rules: |
150 | | * SNaN + QNaN => return the QNaN |
151 | | * two SNaNs => return the one with the larger significand, silenced |
152 | | * two QNaNs => return the one with the larger significand |
153 | | * SNaN and a non-NaN => return the SNaN, silenced |
154 | | * QNaN and a non-NaN => return the QNaN |
155 | | * |
156 | | * If we get down to comparing significands and they are the same, |
157 | | * return the NaN with the positive sign bit (if any). |
158 | | */ |
159 | 0 | set_float_2nan_prop_rule(float_2nan_prop_x87, &env->fp_status); |
160 | | /* |
161 | | * TODO: These are incorrect: the x86 Software Developer's Manual vol 1 |
162 | | * section 4.8.3.5 "Operating on SNaNs and QNaNs" says that the |
163 | | * "larger significand" behaviour is only used for x87 FPU operations. |
164 | | * For SSE the required behaviour is to always return the first NaN, |
165 | | * which is float_2nan_prop_ab. |
166 | | * |
167 | | * mmx_status is used only for the AMD 3DNow! instructions, which |
168 | | * are documented in the "3DNow! Technology Manual" as not supporting |
169 | | * NaNs or infinities as inputs. The result of passing two NaNs is |
170 | | * documented as "undefined", so we can do what we choose. |
171 | | * (Strictly there is some behaviour we don't implement correctly |
172 | | * for these "unsupported" NaN and Inf values, like "NaN * 0 == 0".) |
173 | | */ |
174 | 0 | set_float_2nan_prop_rule(float_2nan_prop_x87, &env->mmx_status); |
175 | 0 | set_float_2nan_prop_rule(float_2nan_prop_x87, &env->sse_status); |
176 | | /* |
177 | | * Only SSE has multiply-add instructions. In the SDM Section 14.5.2 |
178 | | * "Fused-Multiply-ADD (FMA) Numeric Behavior" the NaN handling is |
179 | | * specified -- for 0 * inf + NaN the input NaN is selected, and if |
180 | | * there are multiple input NaNs they are selected in the order a, b, c. |
181 | | * We also do not raise Invalid for the 0 * inf + (Q)NaN case. |
182 | | */ |
183 | 0 | set_float_infzeronan_rule(float_infzeronan_dnan_never | |
184 | 0 | float_infzeronan_suppress_invalid, |
185 | 0 | &env->sse_status); |
186 | 0 | set_float_3nan_prop_rule(float_3nan_prop_abc, &env->sse_status); |
187 | | /* Default NaN: sign bit set, most significant frac bit set */ |
188 | 0 | set_float_default_nan_pattern(0b11000000, &env->fp_status); |
189 | 0 | set_float_default_nan_pattern(0b11000000, &env->mmx_status); |
190 | 0 | set_float_default_nan_pattern(0b11000000, &env->sse_status); |
191 | | /* |
192 | | * x86 does flush-to-zero detection after rounding (the SDM |
193 | | * section 10.2.3.3 on the FTZ bit of MXCSR says that we flush |
194 | | * when we detect underflow, which x86 does after rounding). |
195 | | */ |
196 | 0 | set_float_ftz_detection(float_ftz_after_rounding, &env->fp_status); |
197 | 0 | set_float_ftz_detection(float_ftz_after_rounding, &env->mmx_status); |
198 | 0 | set_float_ftz_detection(float_ftz_after_rounding, &env->sse_status); |
199 | 0 | } |
200 | | |
201 | | static inline int save_exception_flags(CPUX86State *env) |
202 | 0 | { |
203 | 0 | int old_flags = get_float_exception_flags(&env->fp_status); |
204 | 0 | set_float_exception_flags(0, &env->fp_status); |
205 | 0 | return old_flags; |
206 | 0 | } |
207 | | |
208 | | static void merge_exception_flags(CPUX86State *env, int old_flags) |
209 | 0 | { |
210 | 0 | int new_flags = get_float_exception_flags(&env->fp_status); |
211 | 0 | float_raise(old_flags, &env->fp_status); |
212 | 0 | fpu_set_exception(env, |
213 | 0 | ((new_flags & float_flag_invalid ? FPUS_IE : 0) | |
214 | 0 | (new_flags & float_flag_divbyzero ? FPUS_ZE : 0) | |
215 | 0 | (new_flags & float_flag_overflow ? FPUS_OE : 0) | |
216 | 0 | (new_flags & float_flag_underflow ? FPUS_UE : 0) | |
217 | 0 | (new_flags & float_flag_inexact ? FPUS_PE : 0) | |
218 | 0 | (new_flags & float_flag_input_denormal_used ? FPUS_DE : 0))); |
219 | 0 | } |
220 | | |
221 | | static inline floatx80 helper_fdiv(CPUX86State *env, floatx80 a, floatx80 b) |
222 | 0 | { |
223 | 0 | int old_flags = save_exception_flags(env); |
224 | 0 | floatx80 ret = floatx80_div(a, b, &env->fp_status); |
225 | 0 | merge_exception_flags(env, old_flags); |
226 | 0 | return ret; |
227 | 0 | } |
228 | | |
229 | | static void fpu_raise_exception(CPUX86State *env, uintptr_t retaddr) |
230 | 0 | { |
231 | 0 | if (env->cr[0] & CR0_NE_MASK) { |
232 | 0 | raise_exception_ra(env, EXCP10_COPR, retaddr); |
233 | 0 | } |
234 | 0 | #if !defined(CONFIG_USER_ONLY) |
235 | 0 | else { |
236 | 0 | fpu_check_raise_ferr_irq(env); |
237 | 0 | } |
238 | 0 | #endif |
239 | 0 | } |
240 | | |
241 | | void HELPER(flds_FT0)(CPUX86State *env, uint32_t val) |
242 | 0 | { |
243 | 0 | int old_flags = save_exception_flags(env); |
244 | 0 | union { |
245 | 0 | float32 f; |
246 | 0 | uint32_t i; |
247 | 0 | } u; |
248 | |
|
249 | 0 | u.i = val; |
250 | 0 | FT0 = float32_to_floatx80(u.f, &env->fp_status); |
251 | 0 | merge_exception_flags(env, old_flags); |
252 | 0 | } |
253 | | |
254 | | void HELPER(fldl_FT0)(CPUX86State *env, uint64_t val) |
255 | 0 | { |
256 | 0 | int old_flags = save_exception_flags(env); |
257 | 0 | union { |
258 | 0 | float64 f; |
259 | 0 | uint64_t i; |
260 | 0 | } u; |
261 | |
|
262 | 0 | u.i = val; |
263 | 0 | FT0 = float64_to_floatx80(u.f, &env->fp_status); |
264 | 0 | merge_exception_flags(env, old_flags); |
265 | 0 | } |
266 | | |
267 | | void HELPER(fildl_FT0)(CPUX86State *env, int32_t val) |
268 | 0 | { |
269 | 0 | FT0 = int32_to_floatx80(val, &env->fp_status); |
270 | 0 | } |
271 | | |
272 | | void HELPER(flds_ST0)(CPUX86State *env, uint32_t val) |
273 | 0 | { |
274 | 0 | int old_flags = save_exception_flags(env); |
275 | 0 | int new_fpstt; |
276 | 0 | union { |
277 | 0 | float32 f; |
278 | 0 | uint32_t i; |
279 | 0 | } u; |
280 | |
|
281 | 0 | new_fpstt = (env->fpstt - 1) & 7; |
282 | 0 | u.i = val; |
283 | 0 | env->fpregs[new_fpstt].d = float32_to_floatx80(u.f, &env->fp_status); |
284 | 0 | env->fpstt = new_fpstt; |
285 | 0 | env->fptags[new_fpstt] = 0; /* validate stack entry */ |
286 | 0 | merge_exception_flags(env, old_flags); |
287 | 0 | } |
288 | | |
289 | | void HELPER(fldl_ST0)(CPUX86State *env, uint64_t val) |
290 | 0 | { |
291 | 0 | int old_flags = save_exception_flags(env); |
292 | 0 | int new_fpstt; |
293 | 0 | union { |
294 | 0 | float64 f; |
295 | 0 | uint64_t i; |
296 | 0 | } u; |
297 | |
|
298 | 0 | new_fpstt = (env->fpstt - 1) & 7; |
299 | 0 | u.i = val; |
300 | 0 | env->fpregs[new_fpstt].d = float64_to_floatx80(u.f, &env->fp_status); |
301 | 0 | env->fpstt = new_fpstt; |
302 | 0 | env->fptags[new_fpstt] = 0; /* validate stack entry */ |
303 | 0 | merge_exception_flags(env, old_flags); |
304 | 0 | } |
305 | | |
306 | | static FloatX80RoundPrec tmp_maximise_precision(float_status *st) |
307 | 0 | { |
308 | 0 | FloatX80RoundPrec old = get_floatx80_rounding_precision(st); |
309 | 0 | set_floatx80_rounding_precision(floatx80_precision_x, st); |
310 | 0 | return old; |
311 | 0 | } |
312 | | |
313 | | void HELPER(fildl_ST0)(CPUX86State *env, int32_t val) |
314 | 0 | { |
315 | 0 | int new_fpstt; |
316 | 0 | FloatX80RoundPrec old = tmp_maximise_precision(&env->fp_status); |
317 | |
|
318 | 0 | new_fpstt = (env->fpstt - 1) & 7; |
319 | 0 | env->fpregs[new_fpstt].d = int32_to_floatx80(val, &env->fp_status); |
320 | 0 | env->fpstt = new_fpstt; |
321 | 0 | env->fptags[new_fpstt] = 0; /* validate stack entry */ |
322 | |
|
323 | 0 | set_floatx80_rounding_precision(old, &env->fp_status); |
324 | 0 | } |
325 | | |
326 | | void HELPER(fildll_ST0)(CPUX86State *env, int64_t val) |
327 | 0 | { |
328 | 0 | int new_fpstt; |
329 | 0 | FloatX80RoundPrec old = tmp_maximise_precision(&env->fp_status); |
330 | |
|
331 | 0 | new_fpstt = (env->fpstt - 1) & 7; |
332 | 0 | env->fpregs[new_fpstt].d = int64_to_floatx80(val, &env->fp_status); |
333 | 0 | env->fpstt = new_fpstt; |
334 | 0 | env->fptags[new_fpstt] = 0; /* validate stack entry */ |
335 | |
|
336 | 0 | set_floatx80_rounding_precision(old, &env->fp_status); |
337 | 0 | } |
338 | | |
339 | | uint32_t HELPER(fsts_ST0)(CPUX86State *env) |
340 | 0 | { |
341 | 0 | int old_flags = save_exception_flags(env); |
342 | 0 | union { |
343 | 0 | float32 f; |
344 | 0 | uint32_t i; |
345 | 0 | } u; |
346 | |
|
347 | 0 | u.f = floatx80_to_float32(ST0, &env->fp_status); |
348 | 0 | merge_exception_flags(env, old_flags); |
349 | 0 | return u.i; |
350 | 0 | } |
351 | | |
352 | | uint64_t HELPER(fstl_ST0)(CPUX86State *env) |
353 | 0 | { |
354 | 0 | int old_flags = save_exception_flags(env); |
355 | 0 | union { |
356 | 0 | float64 f; |
357 | 0 | uint64_t i; |
358 | 0 | } u; |
359 | |
|
360 | 0 | u.f = floatx80_to_float64(ST0, &env->fp_status); |
361 | 0 | merge_exception_flags(env, old_flags); |
362 | 0 | return u.i; |
363 | 0 | } |
364 | | |
365 | | int32_t HELPER(fist_ST0)(CPUX86State *env) |
366 | 0 | { |
367 | 0 | int old_flags = save_exception_flags(env); |
368 | 0 | int32_t val; |
369 | |
|
370 | 0 | val = floatx80_to_int32(ST0, &env->fp_status); |
371 | 0 | if (val != (int16_t)val) { |
372 | 0 | set_float_exception_flags(float_flag_invalid, &env->fp_status); |
373 | 0 | val = -32768; |
374 | 0 | } |
375 | 0 | merge_exception_flags(env, old_flags); |
376 | 0 | return val; |
377 | 0 | } |
378 | | |
379 | | int32_t HELPER(fistl_ST0)(CPUX86State *env) |
380 | 0 | { |
381 | 0 | int old_flags = save_exception_flags(env); |
382 | 0 | int32_t val; |
383 | |
|
384 | 0 | val = floatx80_to_int32(ST0, &env->fp_status); |
385 | 0 | if (get_float_exception_flags(&env->fp_status) & float_flag_invalid) { |
386 | 0 | val = 0x80000000; |
387 | 0 | } |
388 | 0 | merge_exception_flags(env, old_flags); |
389 | 0 | return val; |
390 | 0 | } |
391 | | |
392 | | int64_t HELPER(fistll_ST0)(CPUX86State *env) |
393 | 0 | { |
394 | 0 | int old_flags = save_exception_flags(env); |
395 | 0 | int64_t val; |
396 | |
|
397 | 0 | val = floatx80_to_int64(ST0, &env->fp_status); |
398 | 0 | if (get_float_exception_flags(&env->fp_status) & float_flag_invalid) { |
399 | 0 | val = 0x8000000000000000ULL; |
400 | 0 | } |
401 | 0 | merge_exception_flags(env, old_flags); |
402 | 0 | return val; |
403 | 0 | } |
404 | | |
405 | | int32_t HELPER(fistt_ST0)(CPUX86State *env) |
406 | 0 | { |
407 | 0 | int old_flags = save_exception_flags(env); |
408 | 0 | int32_t val; |
409 | |
|
410 | 0 | val = floatx80_to_int32_round_to_zero(ST0, &env->fp_status); |
411 | 0 | if (val != (int16_t)val) { |
412 | 0 | set_float_exception_flags(float_flag_invalid, &env->fp_status); |
413 | 0 | val = -32768; |
414 | 0 | } |
415 | 0 | merge_exception_flags(env, old_flags); |
416 | 0 | return val; |
417 | 0 | } |
418 | | |
419 | | int32_t HELPER(fisttl_ST0)(CPUX86State *env) |
420 | 0 | { |
421 | 0 | int old_flags = save_exception_flags(env); |
422 | 0 | int32_t val; |
423 | |
|
424 | 0 | val = floatx80_to_int32_round_to_zero(ST0, &env->fp_status); |
425 | 0 | if (get_float_exception_flags(&env->fp_status) & float_flag_invalid) { |
426 | 0 | val = 0x80000000; |
427 | 0 | } |
428 | 0 | merge_exception_flags(env, old_flags); |
429 | 0 | return val; |
430 | 0 | } |
431 | | |
432 | | int64_t HELPER(fisttll_ST0)(CPUX86State *env) |
433 | 0 | { |
434 | 0 | int old_flags = save_exception_flags(env); |
435 | 0 | int64_t val; |
436 | |
|
437 | 0 | val = floatx80_to_int64_round_to_zero(ST0, &env->fp_status); |
438 | 0 | if (get_float_exception_flags(&env->fp_status) & float_flag_invalid) { |
439 | 0 | val = 0x8000000000000000ULL; |
440 | 0 | } |
441 | 0 | merge_exception_flags(env, old_flags); |
442 | 0 | return val; |
443 | 0 | } |
444 | | |
445 | | void HELPER(fldt_ST0)(CPUX86State *env, target_ulong ptr) |
446 | 0 | { |
447 | 0 | int new_fpstt; |
448 | 0 | X86Access ac; |
449 | |
|
450 | 0 | access_prepare(&ac, env, ptr, 10, MMU_DATA_LOAD, GETPC()); |
451 | |
|
452 | 0 | new_fpstt = (env->fpstt - 1) & 7; |
453 | 0 | env->fpregs[new_fpstt].d = do_fldt(&ac, ptr); |
454 | 0 | env->fpstt = new_fpstt; |
455 | 0 | env->fptags[new_fpstt] = 0; /* validate stack entry */ |
456 | 0 | } |
457 | | |
458 | | void HELPER(fstt_ST0)(CPUX86State *env, target_ulong ptr) |
459 | 0 | { |
460 | 0 | X86Access ac; |
461 | |
|
462 | 0 | access_prepare(&ac, env, ptr, 10, MMU_DATA_STORE, GETPC()); |
463 | 0 | do_fstt(&ac, ptr, ST0); |
464 | 0 | } |
465 | | |
466 | | void HELPER(fpush)(CPUX86State *env) |
467 | 0 | { |
468 | 0 | fpush(env); |
469 | 0 | } |
470 | | |
471 | | void HELPER(fpop)(CPUX86State *env) |
472 | 0 | { |
473 | 0 | fpop(env); |
474 | 0 | } |
475 | | |
476 | | void HELPER(fdecstp)(CPUX86State *env) |
477 | 0 | { |
478 | 0 | env->fpstt = (env->fpstt - 1) & 7; |
479 | 0 | env->fpus &= ~0x4700; |
480 | 0 | } |
481 | | |
482 | | void HELPER(fincstp)(CPUX86State *env) |
483 | 0 | { |
484 | 0 | env->fpstt = (env->fpstt + 1) & 7; |
485 | 0 | env->fpus &= ~0x4700; |
486 | 0 | } |
487 | | |
488 | | /* FPU move */ |
489 | | |
490 | | void HELPER(ffree_STN)(CPUX86State *env, int st_index) |
491 | 0 | { |
492 | 0 | env->fptags[(env->fpstt + st_index) & 7] = 1; |
493 | 0 | } |
494 | | |
495 | | void HELPER(fmov_ST0_FT0)(CPUX86State *env) |
496 | 0 | { |
497 | 0 | ST0 = FT0; |
498 | 0 | } |
499 | | |
500 | | void HELPER(fmov_FT0_STN)(CPUX86State *env, int st_index) |
501 | 0 | { |
502 | 0 | FT0 = ST(st_index); |
503 | 0 | } |
504 | | |
505 | | void HELPER(fmov_ST0_STN)(CPUX86State *env, int st_index) |
506 | 0 | { |
507 | 0 | ST0 = ST(st_index); |
508 | 0 | } |
509 | | |
510 | | void HELPER(fmov_STN_ST0)(CPUX86State *env, int st_index) |
511 | 0 | { |
512 | 0 | ST(st_index) = ST0; |
513 | 0 | env->fptags[(env->fpstt + st_index) & 7] = 0; |
514 | 0 | } |
515 | | |
516 | | void HELPER(fxchg_ST0_STN)(CPUX86State *env, int st_index) |
517 | 0 | { |
518 | 0 | floatx80 tmp; |
519 | |
|
520 | 0 | tmp = ST(st_index); |
521 | 0 | ST(st_index) = ST0; |
522 | 0 | ST0 = tmp; |
523 | |
|
524 | 0 | env->fptags[env->fpstt] = 0; |
525 | 0 | env->fptags[(env->fpstt + st_index) & 7] = 0; |
526 | | |
527 | | /* C1 is unconditionally cleared to 0 */ |
528 | 0 | env->fpus &= ~0x0200; |
529 | 0 | } |
530 | | |
531 | | /* FPU operations */ |
532 | | |
533 | | static const int fcom_ccval[4] = {0x0100, 0x4000, 0x0000, 0x4500}; |
534 | | |
535 | | void HELPER(fcom_ST0_FT0)(CPUX86State *env) |
536 | 0 | { |
537 | 0 | int old_flags = save_exception_flags(env); |
538 | 0 | FloatRelation ret; |
539 | |
|
540 | 0 | ret = floatx80_compare(ST0, FT0, &env->fp_status); |
541 | | /* C1 is unconditionally cleared to 0 */ |
542 | 0 | env->fpus = (env->fpus & ~0x4700) | fcom_ccval[ret + 1]; |
543 | 0 | merge_exception_flags(env, old_flags); |
544 | 0 | } |
545 | | |
546 | | void HELPER(fucom_ST0_FT0)(CPUX86State *env) |
547 | 0 | { |
548 | 0 | int old_flags = save_exception_flags(env); |
549 | 0 | FloatRelation ret; |
550 | |
|
551 | 0 | ret = floatx80_compare_quiet(ST0, FT0, &env->fp_status); |
552 | | /* C1 is unconditionally cleared to 0 */ |
553 | 0 | env->fpus = (env->fpus & ~0x4700) | fcom_ccval[ret + 1]; |
554 | 0 | merge_exception_flags(env, old_flags); |
555 | 0 | } |
556 | | |
557 | | static const int fcomi_ccval[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C}; |
558 | | |
559 | | void HELPER(fcomi_ST0_FT0)(CPUX86State *env) |
560 | 0 | { |
561 | 0 | int old_flags = save_exception_flags(env); |
562 | 0 | FloatRelation ret; |
563 | |
|
564 | 0 | ret = floatx80_compare(ST0, FT0, &env->fp_status); |
565 | | /* OF, SF, and AF are unconditionally cleared to 0 */ |
566 | 0 | CC_SRC = fcomi_ccval[ret + 1]; |
567 | 0 | CC_OP = CC_OP_EFLAGS; |
568 | 0 | merge_exception_flags(env, old_flags); |
569 | 0 | } |
570 | | |
571 | | void HELPER(fucomi_ST0_FT0)(CPUX86State *env) |
572 | 0 | { |
573 | 0 | int old_flags = save_exception_flags(env); |
574 | 0 | FloatRelation ret; |
575 | |
|
576 | 0 | ret = floatx80_compare_quiet(ST0, FT0, &env->fp_status); |
577 | | /* OF, SF, and AF are unconditionally cleared to 0 */ |
578 | 0 | CC_SRC = fcomi_ccval[ret + 1]; |
579 | 0 | CC_OP = CC_OP_EFLAGS; |
580 | 0 | merge_exception_flags(env, old_flags); |
581 | 0 | } |
582 | | |
583 | | void HELPER(fadd_ST0_FT0)(CPUX86State *env) |
584 | 0 | { |
585 | 0 | int old_flags = save_exception_flags(env); |
586 | 0 | ST0 = floatx80_add(ST0, FT0, &env->fp_status); |
587 | 0 | merge_exception_flags(env, old_flags); |
588 | 0 | } |
589 | | |
590 | | void HELPER(fmul_ST0_FT0)(CPUX86State *env) |
591 | 0 | { |
592 | 0 | int old_flags = save_exception_flags(env); |
593 | 0 | ST0 = floatx80_mul(ST0, FT0, &env->fp_status); |
594 | 0 | merge_exception_flags(env, old_flags); |
595 | 0 | } |
596 | | |
597 | | void HELPER(fsub_ST0_FT0)(CPUX86State *env) |
598 | 0 | { |
599 | 0 | int old_flags = save_exception_flags(env); |
600 | 0 | ST0 = floatx80_sub(ST0, FT0, &env->fp_status); |
601 | 0 | merge_exception_flags(env, old_flags); |
602 | 0 | } |
603 | | |
604 | | void HELPER(fsubr_ST0_FT0)(CPUX86State *env) |
605 | 0 | { |
606 | 0 | int old_flags = save_exception_flags(env); |
607 | 0 | ST0 = floatx80_sub(FT0, ST0, &env->fp_status); |
608 | 0 | merge_exception_flags(env, old_flags); |
609 | 0 | } |
610 | | |
611 | | void HELPER(fdiv_ST0_FT0)(CPUX86State *env) |
612 | 0 | { |
613 | 0 | ST0 = helper_fdiv(env, ST0, FT0); |
614 | 0 | } |
615 | | |
616 | | void HELPER(fdivr_ST0_FT0)(CPUX86State *env) |
617 | 0 | { |
618 | 0 | ST0 = helper_fdiv(env, FT0, ST0); |
619 | 0 | } |
620 | | |
621 | | /* fp operations between STN and ST0 */ |
622 | | |
623 | | void HELPER(fadd_STN_ST0)(CPUX86State *env, int st_index) |
624 | 0 | { |
625 | 0 | int old_flags = save_exception_flags(env); |
626 | 0 | ST(st_index) = floatx80_add(ST(st_index), ST0, &env->fp_status); |
627 | 0 | merge_exception_flags(env, old_flags); |
628 | 0 | } |
629 | | |
630 | | void HELPER(fmul_STN_ST0)(CPUX86State *env, int st_index) |
631 | 0 | { |
632 | 0 | int old_flags = save_exception_flags(env); |
633 | 0 | ST(st_index) = floatx80_mul(ST(st_index), ST0, &env->fp_status); |
634 | 0 | merge_exception_flags(env, old_flags); |
635 | 0 | } |
636 | | |
637 | | void HELPER(fsub_STN_ST0)(CPUX86State *env, int st_index) |
638 | 0 | { |
639 | 0 | int old_flags = save_exception_flags(env); |
640 | 0 | ST(st_index) = floatx80_sub(ST(st_index), ST0, &env->fp_status); |
641 | 0 | merge_exception_flags(env, old_flags); |
642 | 0 | } |
643 | | |
644 | | void HELPER(fsubr_STN_ST0)(CPUX86State *env, int st_index) |
645 | 0 | { |
646 | 0 | int old_flags = save_exception_flags(env); |
647 | 0 | ST(st_index) = floatx80_sub(ST0, ST(st_index), &env->fp_status); |
648 | 0 | merge_exception_flags(env, old_flags); |
649 | 0 | } |
650 | | |
651 | | void HELPER(fdiv_STN_ST0)(CPUX86State *env, int st_index) |
652 | 0 | { |
653 | 0 | floatx80 *p; |
654 | |
|
655 | 0 | p = &ST(st_index); |
656 | 0 | *p = helper_fdiv(env, *p, ST0); |
657 | 0 | } |
658 | | |
659 | | void HELPER(fdivr_STN_ST0)(CPUX86State *env, int st_index) |
660 | 0 | { |
661 | 0 | floatx80 *p; |
662 | |
|
663 | 0 | p = &ST(st_index); |
664 | 0 | *p = helper_fdiv(env, ST0, *p); |
665 | 0 | } |
666 | | |
667 | | /* misc FPU operations */ |
668 | | void HELPER(fchs_ST0)(CPUX86State *env) |
669 | 0 | { |
670 | 0 | ST0 = floatx80_chs(ST0); |
671 | 0 | } |
672 | | |
673 | | void HELPER(fabs_ST0)(CPUX86State *env) |
674 | 0 | { |
675 | 0 | ST0 = floatx80_abs(ST0); |
676 | 0 | } |
677 | | |
678 | | void HELPER(fld1_ST0)(CPUX86State *env) |
679 | 0 | { |
680 | 0 | ST0 = floatx80_one; |
681 | 0 | } |
682 | | |
683 | | void HELPER(fldl2t_ST0)(CPUX86State *env) |
684 | 0 | { |
685 | 0 | switch (env->fpuc & FPU_RC_MASK) { |
686 | 0 | case FPU_RC_UP: |
687 | 0 | ST0 = floatx80_l2t_u; |
688 | 0 | break; |
689 | 0 | default: |
690 | 0 | ST0 = floatx80_l2t; |
691 | 0 | break; |
692 | 0 | } |
693 | 0 | } |
694 | | |
695 | | void HELPER(fldl2e_ST0)(CPUX86State *env) |
696 | 0 | { |
697 | 0 | switch (env->fpuc & FPU_RC_MASK) { |
698 | 0 | case FPU_RC_DOWN: |
699 | 0 | case FPU_RC_CHOP: |
700 | 0 | ST0 = floatx80_l2e_d; |
701 | 0 | break; |
702 | 0 | default: |
703 | 0 | ST0 = floatx80_l2e; |
704 | 0 | break; |
705 | 0 | } |
706 | 0 | } |
707 | | |
708 | | void HELPER(fldpi_ST0)(CPUX86State *env) |
709 | 0 | { |
710 | 0 | switch (env->fpuc & FPU_RC_MASK) { |
711 | 0 | case FPU_RC_DOWN: |
712 | 0 | case FPU_RC_CHOP: |
713 | 0 | ST0 = floatx80_pi_d; |
714 | 0 | break; |
715 | 0 | default: |
716 | 0 | ST0 = floatx80_pi; |
717 | 0 | break; |
718 | 0 | } |
719 | 0 | } |
720 | | |
721 | | void HELPER(fldlg2_ST0)(CPUX86State *env) |
722 | 0 | { |
723 | 0 | switch (env->fpuc & FPU_RC_MASK) { |
724 | 0 | case FPU_RC_DOWN: |
725 | 0 | case FPU_RC_CHOP: |
726 | 0 | ST0 = floatx80_lg2_d; |
727 | 0 | break; |
728 | 0 | default: |
729 | 0 | ST0 = floatx80_lg2; |
730 | 0 | break; |
731 | 0 | } |
732 | 0 | } |
733 | | |
734 | | void HELPER(fldln2_ST0)(CPUX86State *env) |
735 | 0 | { |
736 | 0 | switch (env->fpuc & FPU_RC_MASK) { |
737 | 0 | case FPU_RC_DOWN: |
738 | 0 | case FPU_RC_CHOP: |
739 | 0 | ST0 = floatx80_ln2_d; |
740 | 0 | break; |
741 | 0 | default: |
742 | 0 | ST0 = floatx80_ln2; |
743 | 0 | break; |
744 | 0 | } |
745 | 0 | } |
746 | | |
747 | | void HELPER(fldz_ST0)(CPUX86State *env) |
748 | 0 | { |
749 | 0 | ST0 = floatx80_zero; |
750 | 0 | } |
751 | | |
752 | | void HELPER(fldz_FT0)(CPUX86State *env) |
753 | 0 | { |
754 | 0 | FT0 = floatx80_zero; |
755 | 0 | } |
756 | | |
757 | | uint32_t HELPER(fnstsw)(CPUX86State *env) |
758 | 0 | { |
759 | 0 | return (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11; |
760 | 0 | } |
761 | | |
762 | | uint32_t HELPER(fnstcw)(CPUX86State *env) |
763 | 0 | { |
764 | 0 | return env->fpuc; |
765 | 0 | } |
766 | | |
767 | | static void set_x86_rounding_mode(unsigned mode, float_status *status) |
768 | 0 | { |
769 | 0 | static FloatRoundMode x86_round_mode[4] = { |
770 | 0 | float_round_nearest_even, |
771 | 0 | float_round_down, |
772 | 0 | float_round_up, |
773 | 0 | float_round_to_zero |
774 | 0 | }; |
775 | 0 | assert(mode < ARRAY_SIZE(x86_round_mode)); |
776 | 0 | set_float_rounding_mode(x86_round_mode[mode], status); |
777 | 0 | } |
778 | | |
779 | | void update_fp_status(CPUX86State *env) |
780 | 0 | { |
781 | 0 | int rnd_mode; |
782 | 0 | FloatX80RoundPrec rnd_prec; |
783 | | |
784 | | /* set rounding mode */ |
785 | 0 | rnd_mode = (env->fpuc & FPU_RC_MASK) >> FPU_RC_SHIFT; |
786 | 0 | set_x86_rounding_mode(rnd_mode, &env->fp_status); |
787 | |
|
788 | 0 | switch ((env->fpuc >> 8) & 3) { |
789 | 0 | case 0: |
790 | 0 | rnd_prec = floatx80_precision_s; |
791 | 0 | break; |
792 | 0 | case 2: |
793 | 0 | rnd_prec = floatx80_precision_d; |
794 | 0 | break; |
795 | 0 | case 3: |
796 | 0 | default: |
797 | 0 | rnd_prec = floatx80_precision_x; |
798 | 0 | break; |
799 | 0 | } |
800 | 0 | set_floatx80_rounding_precision(rnd_prec, &env->fp_status); |
801 | 0 | } |
802 | | |
803 | | void HELPER(fldcw)(CPUX86State *env, uint32_t val) |
804 | 0 | { |
805 | 0 | cpu_set_fpuc(env, val); |
806 | 0 | } |
807 | | |
808 | | void HELPER(fclex)(CPUX86State *env) |
809 | 0 | { |
810 | 0 | env->fpus &= 0x7f00; |
811 | 0 | } |
812 | | |
813 | | void HELPER(fwait)(CPUX86State *env) |
814 | 0 | { |
815 | 0 | if (env->fpus & FPUS_SE) { |
816 | 0 | fpu_raise_exception(env, GETPC()); |
817 | 0 | } |
818 | 0 | } |
819 | | |
820 | | static void do_fninit(CPUX86State *env) |
821 | 0 | { |
822 | 0 | env->fpus = 0; |
823 | 0 | env->fpstt = 0; |
824 | 0 | env->fpcs = 0; |
825 | 0 | env->fpds = 0; |
826 | 0 | env->fpip = 0; |
827 | 0 | env->fpdp = 0; |
828 | 0 | cpu_set_fpuc(env, 0x37f); |
829 | 0 | env->fptags[0] = 1; |
830 | 0 | env->fptags[1] = 1; |
831 | 0 | env->fptags[2] = 1; |
832 | 0 | env->fptags[3] = 1; |
833 | 0 | env->fptags[4] = 1; |
834 | 0 | env->fptags[5] = 1; |
835 | 0 | env->fptags[6] = 1; |
836 | 0 | env->fptags[7] = 1; |
837 | 0 | } |
838 | | |
839 | | void HELPER(fninit)(CPUX86State *env) |
840 | 0 | { |
841 | 0 | do_fninit(env); |
842 | 0 | } |
843 | | |
844 | | /* BCD ops */ |
845 | | |
846 | | void HELPER(fbld_ST0)(CPUX86State *env, target_ulong ptr) |
847 | 0 | { |
848 | 0 | X86Access ac; |
849 | 0 | floatx80 tmp; |
850 | 0 | uint64_t val; |
851 | 0 | unsigned int v; |
852 | 0 | int i; |
853 | |
|
854 | 0 | access_prepare(&ac, env, ptr, 10, MMU_DATA_LOAD, GETPC()); |
855 | |
|
856 | 0 | val = 0; |
857 | 0 | for (i = 8; i >= 0; i--) { |
858 | 0 | v = access_ldb(&ac, ptr + i); |
859 | 0 | val = (val * 100) + ((v >> 4) * 10) + (v & 0xf); |
860 | 0 | } |
861 | 0 | tmp = int64_to_floatx80(val, &env->fp_status); |
862 | 0 | if (access_ldb(&ac, ptr + 9) & 0x80) { |
863 | 0 | tmp = floatx80_chs(tmp); |
864 | 0 | } |
865 | 0 | fpush(env); |
866 | 0 | ST0 = tmp; |
867 | 0 | } |
868 | | |
869 | | void HELPER(fbst_ST0)(CPUX86State *env, target_ulong ptr) |
870 | 0 | { |
871 | 0 | int old_flags = save_exception_flags(env); |
872 | 0 | int v; |
873 | 0 | target_ulong mem_ref, mem_end; |
874 | 0 | int64_t val; |
875 | 0 | CPU_LDoubleU temp; |
876 | 0 | X86Access ac; |
877 | |
|
878 | 0 | access_prepare(&ac, env, ptr, 10, MMU_DATA_STORE, GETPC()); |
879 | 0 | temp.d = ST0; |
880 | |
|
881 | 0 | val = floatx80_to_int64(ST0, &env->fp_status); |
882 | 0 | mem_ref = ptr; |
883 | 0 | if (val >= 1000000000000000000LL || val <= -1000000000000000000LL) { |
884 | 0 | set_float_exception_flags(float_flag_invalid, &env->fp_status); |
885 | 0 | while (mem_ref < ptr + 7) { |
886 | 0 | access_stb(&ac, mem_ref++, 0); |
887 | 0 | } |
888 | 0 | access_stb(&ac, mem_ref++, 0xc0); |
889 | 0 | access_stb(&ac, mem_ref++, 0xff); |
890 | 0 | access_stb(&ac, mem_ref++, 0xff); |
891 | 0 | merge_exception_flags(env, old_flags); |
892 | 0 | return; |
893 | 0 | } |
894 | 0 | mem_end = mem_ref + 9; |
895 | 0 | if (SIGND(temp)) { |
896 | 0 | access_stb(&ac, mem_end, 0x80); |
897 | 0 | val = -val; |
898 | 0 | } else { |
899 | 0 | access_stb(&ac, mem_end, 0x00); |
900 | 0 | } |
901 | 0 | while (mem_ref < mem_end) { |
902 | 0 | if (val == 0) { |
903 | 0 | break; |
904 | 0 | } |
905 | 0 | v = val % 100; |
906 | 0 | val = val / 100; |
907 | 0 | v = ((v / 10) << 4) | (v % 10); |
908 | 0 | access_stb(&ac, mem_ref++, v); |
909 | 0 | } |
910 | 0 | while (mem_ref < mem_end) { |
911 | 0 | access_stb(&ac, mem_ref++, 0); |
912 | 0 | } |
913 | 0 | merge_exception_flags(env, old_flags); |
914 | 0 | } |
915 | | |
916 | | /* 128-bit significand of log(2). */ |
917 | 0 | #define ln2_sig_high 0xb17217f7d1cf79abULL |
918 | 0 | #define ln2_sig_low 0xc9e3b39803f2f6afULL |
919 | | |
920 | | /* |
921 | | * Polynomial coefficients for an approximation to (2^x - 1) / x, on |
922 | | * the interval [-1/64, 1/64]. |
923 | | */ |
924 | 0 | #define f2xm1_coeff_0 make_floatx80(0x3ffe, 0xb17217f7d1cf79acULL) |
925 | 0 | #define f2xm1_coeff_0_low make_floatx80(0xbfbc, 0xd87edabf495b3762ULL) |
926 | 0 | #define f2xm1_coeff_1 make_floatx80(0x3ffc, 0xf5fdeffc162c7543ULL) |
927 | 0 | #define f2xm1_coeff_2 make_floatx80(0x3ffa, 0xe35846b82505fcc7ULL) |
928 | 0 | #define f2xm1_coeff_3 make_floatx80(0x3ff8, 0x9d955b7dd273b899ULL) |
929 | 0 | #define f2xm1_coeff_4 make_floatx80(0x3ff5, 0xaec3ff3c4ef4ac0cULL) |
930 | 0 | #define f2xm1_coeff_5 make_floatx80(0x3ff2, 0xa184897c3a7f0de9ULL) |
931 | 0 | #define f2xm1_coeff_6 make_floatx80(0x3fee, 0xffe634d0ec30d504ULL) |
932 | 0 | #define f2xm1_coeff_7 make_floatx80(0x3feb, 0xb160111d2db515e4ULL) |
933 | | |
934 | | struct f2xm1_data { |
935 | | /* |
936 | | * A value very close to a multiple of 1/32, such that 2^t and 2^t - 1 |
937 | | * are very close to exact floatx80 values. |
938 | | */ |
939 | | floatx80 t; |
940 | | /* The value of 2^t. */ |
941 | | floatx80 exp2; |
942 | | /* The value of 2^t - 1. */ |
943 | | floatx80 exp2m1; |
944 | | }; |
945 | | |
946 | | static const struct f2xm1_data f2xm1_table[65] = { |
947 | | { make_floatx80_init(0xbfff, 0x8000000000000000ULL), |
948 | | make_floatx80_init(0x3ffe, 0x8000000000000000ULL), |
949 | | make_floatx80_init(0xbffe, 0x8000000000000000ULL) }, |
950 | | { make_floatx80_init(0xbffe, 0xf800000000002e7eULL), |
951 | | make_floatx80_init(0x3ffe, 0x82cd8698ac2b9160ULL), |
952 | | make_floatx80_init(0xbffd, 0xfa64f2cea7a8dd40ULL) }, |
953 | | { make_floatx80_init(0xbffe, 0xefffffffffffe960ULL), |
954 | | make_floatx80_init(0x3ffe, 0x85aac367cc488345ULL), |
955 | | make_floatx80_init(0xbffd, 0xf4aa7930676ef976ULL) }, |
956 | | { make_floatx80_init(0xbffe, 0xe800000000006f10ULL), |
957 | | make_floatx80_init(0x3ffe, 0x88980e8092da5c14ULL), |
958 | | make_floatx80_init(0xbffd, 0xeecfe2feda4b47d8ULL) }, |
959 | | { make_floatx80_init(0xbffe, 0xe000000000008a45ULL), |
960 | | make_floatx80_init(0x3ffe, 0x8b95c1e3ea8ba2a5ULL), |
961 | | make_floatx80_init(0xbffd, 0xe8d47c382ae8bab6ULL) }, |
962 | | { make_floatx80_init(0xbffe, 0xd7ffffffffff8a9eULL), |
963 | | make_floatx80_init(0x3ffe, 0x8ea4398b45cd8116ULL), |
964 | | make_floatx80_init(0xbffd, 0xe2b78ce97464fdd4ULL) }, |
965 | | { make_floatx80_init(0xbffe, 0xd0000000000019a0ULL), |
966 | | make_floatx80_init(0x3ffe, 0x91c3d373ab11b919ULL), |
967 | | make_floatx80_init(0xbffd, 0xdc785918a9dc8dceULL) }, |
968 | | { make_floatx80_init(0xbffe, 0xc7ffffffffff14dfULL), |
969 | | make_floatx80_init(0x3ffe, 0x94f4efa8fef76836ULL), |
970 | | make_floatx80_init(0xbffd, 0xd61620ae02112f94ULL) }, |
971 | | { make_floatx80_init(0xbffe, 0xc000000000006530ULL), |
972 | | make_floatx80_init(0x3ffe, 0x9837f0518db87fbbULL), |
973 | | make_floatx80_init(0xbffd, 0xcf901f5ce48f008aULL) }, |
974 | | { make_floatx80_init(0xbffe, 0xb7ffffffffff1723ULL), |
975 | | make_floatx80_init(0x3ffe, 0x9b8d39b9d54eb74cULL), |
976 | | make_floatx80_init(0xbffd, 0xc8e58c8c55629168ULL) }, |
977 | | { make_floatx80_init(0xbffe, 0xb00000000000b5e1ULL), |
978 | | make_floatx80_init(0x3ffe, 0x9ef5326091a0c366ULL), |
979 | | make_floatx80_init(0xbffd, 0xc2159b3edcbe7934ULL) }, |
980 | | { make_floatx80_init(0xbffe, 0xa800000000006f8aULL), |
981 | | make_floatx80_init(0x3ffe, 0xa27043030c49370aULL), |
982 | | make_floatx80_init(0xbffd, 0xbb1f79f9e76d91ecULL) }, |
983 | | { make_floatx80_init(0xbffe, 0x9fffffffffff816aULL), |
984 | | make_floatx80_init(0x3ffe, 0xa5fed6a9b15171cfULL), |
985 | | make_floatx80_init(0xbffd, 0xb40252ac9d5d1c62ULL) }, |
986 | | { make_floatx80_init(0xbffe, 0x97ffffffffffb621ULL), |
987 | | make_floatx80_init(0x3ffe, 0xa9a15ab4ea7c30e6ULL), |
988 | | make_floatx80_init(0xbffd, 0xacbd4a962b079e34ULL) }, |
989 | | { make_floatx80_init(0xbffe, 0x8fffffffffff162bULL), |
990 | | make_floatx80_init(0x3ffe, 0xad583eea42a1b886ULL), |
991 | | make_floatx80_init(0xbffd, 0xa54f822b7abc8ef4ULL) }, |
992 | | { make_floatx80_init(0xbffe, 0x87ffffffffff4d34ULL), |
993 | | make_floatx80_init(0x3ffe, 0xb123f581d2ac7b51ULL), |
994 | | make_floatx80_init(0xbffd, 0x9db814fc5aa7095eULL) }, |
995 | | { make_floatx80_init(0xbffe, 0x800000000000227dULL), |
996 | | make_floatx80_init(0x3ffe, 0xb504f333f9de539dULL), |
997 | | make_floatx80_init(0xbffd, 0x95f619980c4358c6ULL) }, |
998 | | { make_floatx80_init(0xbffd, 0xefffffffffff3978ULL), |
999 | | make_floatx80_init(0x3ffe, 0xb8fbaf4762fbd0a1ULL), |
1000 | | make_floatx80_init(0xbffd, 0x8e08a1713a085ebeULL) }, |
1001 | | { make_floatx80_init(0xbffd, 0xe00000000000df81ULL), |
1002 | | make_floatx80_init(0x3ffe, 0xbd08a39f580bfd8cULL), |
1003 | | make_floatx80_init(0xbffd, 0x85eeb8c14fe804e8ULL) }, |
1004 | | { make_floatx80_init(0xbffd, 0xd00000000000bccfULL), |
1005 | | make_floatx80_init(0x3ffe, 0xc12c4cca667062f6ULL), |
1006 | | make_floatx80_init(0xbffc, 0xfb4eccd6663e7428ULL) }, |
1007 | | { make_floatx80_init(0xbffd, 0xc00000000000eff0ULL), |
1008 | | make_floatx80_init(0x3ffe, 0xc5672a1155069abeULL), |
1009 | | make_floatx80_init(0xbffc, 0xea6357baabe59508ULL) }, |
1010 | | { make_floatx80_init(0xbffd, 0xb000000000000fe6ULL), |
1011 | | make_floatx80_init(0x3ffe, 0xc9b9bd866e2f234bULL), |
1012 | | make_floatx80_init(0xbffc, 0xd91909e6474372d4ULL) }, |
1013 | | { make_floatx80_init(0xbffd, 0x9fffffffffff2172ULL), |
1014 | | make_floatx80_init(0x3ffe, 0xce248c151f84bf00ULL), |
1015 | | make_floatx80_init(0xbffc, 0xc76dcfab81ed0400ULL) }, |
1016 | | { make_floatx80_init(0xbffd, 0x8fffffffffffafffULL), |
1017 | | make_floatx80_init(0x3ffe, 0xd2a81d91f12afb2bULL), |
1018 | | make_floatx80_init(0xbffc, 0xb55f89b83b541354ULL) }, |
1019 | | { make_floatx80_init(0xbffc, 0xffffffffffff81a3ULL), |
1020 | | make_floatx80_init(0x3ffe, 0xd744fccad69d7d5eULL), |
1021 | | make_floatx80_init(0xbffc, 0xa2ec0cd4a58a0a88ULL) }, |
1022 | | { make_floatx80_init(0xbffc, 0xdfffffffffff1568ULL), |
1023 | | make_floatx80_init(0x3ffe, 0xdbfbb797daf25a44ULL), |
1024 | | make_floatx80_init(0xbffc, 0x901121a0943696f0ULL) }, |
1025 | | { make_floatx80_init(0xbffc, 0xbfffffffffff68daULL), |
1026 | | make_floatx80_init(0x3ffe, 0xe0ccdeec2a94f811ULL), |
1027 | | make_floatx80_init(0xbffb, 0xf999089eab583f78ULL) }, |
1028 | | { make_floatx80_init(0xbffc, 0x9fffffffffff4690ULL), |
1029 | | make_floatx80_init(0x3ffe, 0xe5b906e77c83657eULL), |
1030 | | make_floatx80_init(0xbffb, 0xd237c8c41be4d410ULL) }, |
1031 | | { make_floatx80_init(0xbffb, 0xffffffffffff8aeeULL), |
1032 | | make_floatx80_init(0x3ffe, 0xeac0c6e7dd24427cULL), |
1033 | | make_floatx80_init(0xbffb, 0xa9f9c8c116ddec20ULL) }, |
1034 | | { make_floatx80_init(0xbffb, 0xbfffffffffff2d18ULL), |
1035 | | make_floatx80_init(0x3ffe, 0xefe4b99bdcdb06ebULL), |
1036 | | make_floatx80_init(0xbffb, 0x80da33211927c8a8ULL) }, |
1037 | | { make_floatx80_init(0xbffa, 0xffffffffffff8ccbULL), |
1038 | | make_floatx80_init(0x3ffe, 0xf5257d152486d0f4ULL), |
1039 | | make_floatx80_init(0xbffa, 0xada82eadb792f0c0ULL) }, |
1040 | | { make_floatx80_init(0xbff9, 0xffffffffffff11feULL), |
1041 | | make_floatx80_init(0x3ffe, 0xfa83b2db722a0846ULL), |
1042 | | make_floatx80_init(0xbff9, 0xaf89a491babef740ULL) }, |
1043 | | { floatx80_zero_init, |
1044 | | make_floatx80_init(0x3fff, 0x8000000000000000ULL), |
1045 | | floatx80_zero_init }, |
1046 | | { make_floatx80_init(0x3ff9, 0xffffffffffff2680ULL), |
1047 | | make_floatx80_init(0x3fff, 0x82cd8698ac2b9f6fULL), |
1048 | | make_floatx80_init(0x3ff9, 0xb361a62b0ae7dbc0ULL) }, |
1049 | | { make_floatx80_init(0x3ffb, 0x800000000000b500ULL), |
1050 | | make_floatx80_init(0x3fff, 0x85aac367cc488345ULL), |
1051 | | make_floatx80_init(0x3ffa, 0xb5586cf9891068a0ULL) }, |
1052 | | { make_floatx80_init(0x3ffb, 0xbfffffffffff4b67ULL), |
1053 | | make_floatx80_init(0x3fff, 0x88980e8092da7cceULL), |
1054 | | make_floatx80_init(0x3ffb, 0x8980e8092da7cce0ULL) }, |
1055 | | { make_floatx80_init(0x3ffb, 0xffffffffffffff57ULL), |
1056 | | make_floatx80_init(0x3fff, 0x8b95c1e3ea8bd6dfULL), |
1057 | | make_floatx80_init(0x3ffb, 0xb95c1e3ea8bd6df0ULL) }, |
1058 | | { make_floatx80_init(0x3ffc, 0x9fffffffffff811fULL), |
1059 | | make_floatx80_init(0x3fff, 0x8ea4398b45cd4780ULL), |
1060 | | make_floatx80_init(0x3ffb, 0xea4398b45cd47800ULL) }, |
1061 | | { make_floatx80_init(0x3ffc, 0xbfffffffffff9980ULL), |
1062 | | make_floatx80_init(0x3fff, 0x91c3d373ab11b919ULL), |
1063 | | make_floatx80_init(0x3ffc, 0x8e1e9b9d588dc8c8ULL) }, |
1064 | | { make_floatx80_init(0x3ffc, 0xdffffffffffff631ULL), |
1065 | | make_floatx80_init(0x3fff, 0x94f4efa8fef70864ULL), |
1066 | | make_floatx80_init(0x3ffc, 0xa7a77d47f7b84320ULL) }, |
1067 | | { make_floatx80_init(0x3ffc, 0xffffffffffff2499ULL), |
1068 | | make_floatx80_init(0x3fff, 0x9837f0518db892d4ULL), |
1069 | | make_floatx80_init(0x3ffc, 0xc1bf828c6dc496a0ULL) }, |
1070 | | { make_floatx80_init(0x3ffd, 0x8fffffffffff80fbULL), |
1071 | | make_floatx80_init(0x3fff, 0x9b8d39b9d54e3a79ULL), |
1072 | | make_floatx80_init(0x3ffc, 0xdc69cdceaa71d3c8ULL) }, |
1073 | | { make_floatx80_init(0x3ffd, 0x9fffffffffffbc23ULL), |
1074 | | make_floatx80_init(0x3fff, 0x9ef5326091a10313ULL), |
1075 | | make_floatx80_init(0x3ffc, 0xf7a993048d081898ULL) }, |
1076 | | { make_floatx80_init(0x3ffd, 0xafffffffffff20ecULL), |
1077 | | make_floatx80_init(0x3fff, 0xa27043030c49370aULL), |
1078 | | make_floatx80_init(0x3ffd, 0x89c10c0c3124dc28ULL) }, |
1079 | | { make_floatx80_init(0x3ffd, 0xc00000000000fd2cULL), |
1080 | | make_floatx80_init(0x3fff, 0xa5fed6a9b15171cfULL), |
1081 | | make_floatx80_init(0x3ffd, 0x97fb5aa6c545c73cULL) }, |
1082 | | { make_floatx80_init(0x3ffd, 0xd0000000000093beULL), |
1083 | | make_floatx80_init(0x3fff, 0xa9a15ab4ea7c30e6ULL), |
1084 | | make_floatx80_init(0x3ffd, 0xa6856ad3a9f0c398ULL) }, |
1085 | | { make_floatx80_init(0x3ffd, 0xe00000000000c2aeULL), |
1086 | | make_floatx80_init(0x3fff, 0xad583eea42a17876ULL), |
1087 | | make_floatx80_init(0x3ffd, 0xb560fba90a85e1d8ULL) }, |
1088 | | { make_floatx80_init(0x3ffd, 0xefffffffffff1e3fULL), |
1089 | | make_floatx80_init(0x3fff, 0xb123f581d2abef6cULL), |
1090 | | make_floatx80_init(0x3ffd, 0xc48fd6074aafbdb0ULL) }, |
1091 | | { make_floatx80_init(0x3ffd, 0xffffffffffff1c23ULL), |
1092 | | make_floatx80_init(0x3fff, 0xb504f333f9de2cadULL), |
1093 | | make_floatx80_init(0x3ffd, 0xd413cccfe778b2b4ULL) }, |
1094 | | { make_floatx80_init(0x3ffe, 0x8800000000006344ULL), |
1095 | | make_floatx80_init(0x3fff, 0xb8fbaf4762fbd0a1ULL), |
1096 | | make_floatx80_init(0x3ffd, 0xe3eebd1d8bef4284ULL) }, |
1097 | | { make_floatx80_init(0x3ffe, 0x9000000000005d67ULL), |
1098 | | make_floatx80_init(0x3fff, 0xbd08a39f580c668dULL), |
1099 | | make_floatx80_init(0x3ffd, 0xf4228e7d60319a34ULL) }, |
1100 | | { make_floatx80_init(0x3ffe, 0x9800000000009127ULL), |
1101 | | make_floatx80_init(0x3fff, 0xc12c4cca6670e042ULL), |
1102 | | make_floatx80_init(0x3ffe, 0x82589994cce1c084ULL) }, |
1103 | | { make_floatx80_init(0x3ffe, 0x9fffffffffff06f9ULL), |
1104 | | make_floatx80_init(0x3fff, 0xc5672a11550655c3ULL), |
1105 | | make_floatx80_init(0x3ffe, 0x8ace5422aa0cab86ULL) }, |
1106 | | { make_floatx80_init(0x3ffe, 0xa7fffffffffff80dULL), |
1107 | | make_floatx80_init(0x3fff, 0xc9b9bd866e2f234bULL), |
1108 | | make_floatx80_init(0x3ffe, 0x93737b0cdc5e4696ULL) }, |
1109 | | { make_floatx80_init(0x3ffe, 0xafffffffffff1470ULL), |
1110 | | make_floatx80_init(0x3fff, 0xce248c151f83fd69ULL), |
1111 | | make_floatx80_init(0x3ffe, 0x9c49182a3f07fad2ULL) }, |
1112 | | { make_floatx80_init(0x3ffe, 0xb800000000000e0aULL), |
1113 | | make_floatx80_init(0x3fff, 0xd2a81d91f12aec5cULL), |
1114 | | make_floatx80_init(0x3ffe, 0xa5503b23e255d8b8ULL) }, |
1115 | | { make_floatx80_init(0x3ffe, 0xc00000000000b7faULL), |
1116 | | make_floatx80_init(0x3fff, 0xd744fccad69dd630ULL), |
1117 | | make_floatx80_init(0x3ffe, 0xae89f995ad3bac60ULL) }, |
1118 | | { make_floatx80_init(0x3ffe, 0xc800000000003aa6ULL), |
1119 | | make_floatx80_init(0x3fff, 0xdbfbb797daf25a44ULL), |
1120 | | make_floatx80_init(0x3ffe, 0xb7f76f2fb5e4b488ULL) }, |
1121 | | { make_floatx80_init(0x3ffe, 0xd00000000000a6aeULL), |
1122 | | make_floatx80_init(0x3fff, 0xe0ccdeec2a954685ULL), |
1123 | | make_floatx80_init(0x3ffe, 0xc199bdd8552a8d0aULL) }, |
1124 | | { make_floatx80_init(0x3ffe, 0xd800000000004165ULL), |
1125 | | make_floatx80_init(0x3fff, 0xe5b906e77c837155ULL), |
1126 | | make_floatx80_init(0x3ffe, 0xcb720dcef906e2aaULL) }, |
1127 | | { make_floatx80_init(0x3ffe, 0xe00000000000582cULL), |
1128 | | make_floatx80_init(0x3fff, 0xeac0c6e7dd24713aULL), |
1129 | | make_floatx80_init(0x3ffe, 0xd5818dcfba48e274ULL) }, |
1130 | | { make_floatx80_init(0x3ffe, 0xe800000000001a5dULL), |
1131 | | make_floatx80_init(0x3fff, 0xefe4b99bdcdb06ebULL), |
1132 | | make_floatx80_init(0x3ffe, 0xdfc97337b9b60dd6ULL) }, |
1133 | | { make_floatx80_init(0x3ffe, 0xefffffffffffc1efULL), |
1134 | | make_floatx80_init(0x3fff, 0xf5257d152486a2faULL), |
1135 | | make_floatx80_init(0x3ffe, 0xea4afa2a490d45f4ULL) }, |
1136 | | { make_floatx80_init(0x3ffe, 0xf800000000001069ULL), |
1137 | | make_floatx80_init(0x3fff, 0xfa83b2db722a0e5cULL), |
1138 | | make_floatx80_init(0x3ffe, 0xf50765b6e4541cb8ULL) }, |
1139 | | { make_floatx80_init(0x3fff, 0x8000000000000000ULL), |
1140 | | make_floatx80_init(0x4000, 0x8000000000000000ULL), |
1141 | | make_floatx80_init(0x3fff, 0x8000000000000000ULL) }, |
1142 | | }; |
1143 | | |
1144 | | void HELPER(f2xm1)(CPUX86State *env) |
1145 | 0 | { |
1146 | 0 | int old_flags = save_exception_flags(env); |
1147 | 0 | uint64_t sig = extractFloatx80Frac(ST0); |
1148 | 0 | int32_t exp = extractFloatx80Exp(ST0); |
1149 | 0 | bool sign = extractFloatx80Sign(ST0); |
1150 | |
|
1151 | 0 | if (floatx80_invalid_encoding(ST0, &env->fp_status)) { |
1152 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
1153 | 0 | ST0 = floatx80_default_nan(&env->fp_status); |
1154 | 0 | } else if (floatx80_is_any_nan(ST0)) { |
1155 | 0 | if (floatx80_is_signaling_nan(ST0, &env->fp_status)) { |
1156 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
1157 | 0 | ST0 = floatx80_silence_nan(ST0, &env->fp_status); |
1158 | 0 | } |
1159 | 0 | } else if (exp > 0x3fff || |
1160 | 0 | (exp == 0x3fff && sig != (0x8000000000000000ULL))) { |
1161 | | /* Out of range for the instruction, treat as invalid. */ |
1162 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
1163 | 0 | ST0 = floatx80_default_nan(&env->fp_status); |
1164 | 0 | } else if (exp == 0x3fff) { |
1165 | | /* Argument 1 or -1, exact result 1 or -0.5. */ |
1166 | 0 | if (sign) { |
1167 | 0 | ST0 = make_floatx80(0xbffe, 0x8000000000000000ULL); |
1168 | 0 | } |
1169 | 0 | } else if (exp < 0x3fb0) { |
1170 | 0 | if (!floatx80_is_zero(ST0)) { |
1171 | | /* |
1172 | | * Multiplying the argument by an extra-precision version |
1173 | | * of log(2) is sufficiently precise. Zero arguments are |
1174 | | * returned unchanged. |
1175 | | */ |
1176 | 0 | uint64_t sig0, sig1, sig2; |
1177 | 0 | if (exp == 0) { |
1178 | 0 | normalizeFloatx80Subnormal(sig, &exp, &sig); |
1179 | 0 | } |
1180 | 0 | mul128By64To192(ln2_sig_high, ln2_sig_low, sig, &sig0, &sig1, |
1181 | 0 | &sig2); |
1182 | | /* This result is inexact. */ |
1183 | 0 | sig1 |= 1; |
1184 | 0 | ST0 = normalizeRoundAndPackFloatx80(floatx80_precision_x, |
1185 | 0 | sign, exp, sig0, sig1, |
1186 | 0 | &env->fp_status); |
1187 | 0 | } |
1188 | 0 | } else { |
1189 | 0 | floatx80 tmp, y, accum; |
1190 | 0 | bool asign, bsign; |
1191 | 0 | int32_t n, aexp, bexp; |
1192 | 0 | uint64_t asig0, asig1, asig2, bsig0, bsig1; |
1193 | 0 | FloatRoundMode save_mode = get_float_rounding_mode(&env->fp_status); |
1194 | 0 | FloatX80RoundPrec save_prec = |
1195 | 0 | get_floatx80_rounding_precision(&env->fp_status); |
1196 | 0 | set_float_rounding_mode(float_round_nearest_even, &env->fp_status); |
1197 | 0 | set_floatx80_rounding_precision(floatx80_precision_x, &env->fp_status); |
1198 | | |
1199 | | /* Find the nearest multiple of 1/32 to the argument. */ |
1200 | 0 | tmp = floatx80_scalbn(ST0, 5, &env->fp_status); |
1201 | 0 | n = 32 + floatx80_to_int32(tmp, &env->fp_status); |
1202 | 0 | y = floatx80_sub(ST0, f2xm1_table[n].t, &env->fp_status); |
1203 | |
|
1204 | 0 | if (floatx80_is_zero(y)) { |
1205 | | /* |
1206 | | * Use the value of 2^t - 1 from the table, to avoid |
1207 | | * needing to special-case zero as a result of |
1208 | | * multiplication below. |
1209 | | */ |
1210 | 0 | ST0 = f2xm1_table[n].t; |
1211 | 0 | set_float_exception_flags(float_flag_inexact, &env->fp_status); |
1212 | 0 | set_float_rounding_mode(save_mode, &env->fp_status); |
1213 | 0 | } else { |
1214 | | /* |
1215 | | * Compute the lower parts of a polynomial expansion for |
1216 | | * (2^y - 1) / y. |
1217 | | */ |
1218 | 0 | accum = floatx80_mul(f2xm1_coeff_7, y, &env->fp_status); |
1219 | 0 | accum = floatx80_add(f2xm1_coeff_6, accum, &env->fp_status); |
1220 | 0 | accum = floatx80_mul(accum, y, &env->fp_status); |
1221 | 0 | accum = floatx80_add(f2xm1_coeff_5, accum, &env->fp_status); |
1222 | 0 | accum = floatx80_mul(accum, y, &env->fp_status); |
1223 | 0 | accum = floatx80_add(f2xm1_coeff_4, accum, &env->fp_status); |
1224 | 0 | accum = floatx80_mul(accum, y, &env->fp_status); |
1225 | 0 | accum = floatx80_add(f2xm1_coeff_3, accum, &env->fp_status); |
1226 | 0 | accum = floatx80_mul(accum, y, &env->fp_status); |
1227 | 0 | accum = floatx80_add(f2xm1_coeff_2, accum, &env->fp_status); |
1228 | 0 | accum = floatx80_mul(accum, y, &env->fp_status); |
1229 | 0 | accum = floatx80_add(f2xm1_coeff_1, accum, &env->fp_status); |
1230 | 0 | accum = floatx80_mul(accum, y, &env->fp_status); |
1231 | 0 | accum = floatx80_add(f2xm1_coeff_0_low, accum, &env->fp_status); |
1232 | | |
1233 | | /* |
1234 | | * The full polynomial expansion is f2xm1_coeff_0 + accum |
1235 | | * (where accum has much lower magnitude, and so, in |
1236 | | * particular, carry out of the addition is not possible). |
1237 | | * (This expansion is only accurate to about 70 bits, not |
1238 | | * 128 bits.) |
1239 | | */ |
1240 | 0 | aexp = extractFloatx80Exp(f2xm1_coeff_0); |
1241 | 0 | asign = extractFloatx80Sign(f2xm1_coeff_0); |
1242 | 0 | shift128RightJamming(extractFloatx80Frac(accum), 0, |
1243 | 0 | aexp - extractFloatx80Exp(accum), |
1244 | 0 | &asig0, &asig1); |
1245 | 0 | bsig0 = extractFloatx80Frac(f2xm1_coeff_0); |
1246 | 0 | bsig1 = 0; |
1247 | 0 | if (asign == extractFloatx80Sign(accum)) { |
1248 | 0 | add128(bsig0, bsig1, asig0, asig1, &asig0, &asig1); |
1249 | 0 | } else { |
1250 | 0 | sub128(bsig0, bsig1, asig0, asig1, &asig0, &asig1); |
1251 | 0 | } |
1252 | | /* And thus compute an approximation to 2^y - 1. */ |
1253 | 0 | mul128By64To192(asig0, asig1, extractFloatx80Frac(y), |
1254 | 0 | &asig0, &asig1, &asig2); |
1255 | 0 | aexp += extractFloatx80Exp(y) - 0x3ffe; |
1256 | 0 | asign ^= extractFloatx80Sign(y); |
1257 | 0 | if (n != 32) { |
1258 | | /* |
1259 | | * Multiply this by the precomputed value of 2^t and |
1260 | | * add that of 2^t - 1. |
1261 | | */ |
1262 | 0 | mul128By64To192(asig0, asig1, |
1263 | 0 | extractFloatx80Frac(f2xm1_table[n].exp2), |
1264 | 0 | &asig0, &asig1, &asig2); |
1265 | 0 | aexp += extractFloatx80Exp(f2xm1_table[n].exp2) - 0x3ffe; |
1266 | 0 | bexp = extractFloatx80Exp(f2xm1_table[n].exp2m1); |
1267 | 0 | bsig0 = extractFloatx80Frac(f2xm1_table[n].exp2m1); |
1268 | 0 | bsig1 = 0; |
1269 | 0 | if (bexp < aexp) { |
1270 | 0 | shift128RightJamming(bsig0, bsig1, aexp - bexp, |
1271 | 0 | &bsig0, &bsig1); |
1272 | 0 | } else if (aexp < bexp) { |
1273 | 0 | shift128RightJamming(asig0, asig1, bexp - aexp, |
1274 | 0 | &asig0, &asig1); |
1275 | 0 | aexp = bexp; |
1276 | 0 | } |
1277 | | /* The sign of 2^t - 1 is always that of the result. */ |
1278 | 0 | bsign = extractFloatx80Sign(f2xm1_table[n].exp2m1); |
1279 | 0 | if (asign == bsign) { |
1280 | | /* Avoid possible carry out of the addition. */ |
1281 | 0 | shift128RightJamming(asig0, asig1, 1, |
1282 | 0 | &asig0, &asig1); |
1283 | 0 | shift128RightJamming(bsig0, bsig1, 1, |
1284 | 0 | &bsig0, &bsig1); |
1285 | 0 | ++aexp; |
1286 | 0 | add128(asig0, asig1, bsig0, bsig1, &asig0, &asig1); |
1287 | 0 | } else { |
1288 | 0 | sub128(bsig0, bsig1, asig0, asig1, &asig0, &asig1); |
1289 | 0 | asign = bsign; |
1290 | 0 | } |
1291 | 0 | } |
1292 | 0 | set_float_rounding_mode(save_mode, &env->fp_status); |
1293 | | /* This result is inexact. */ |
1294 | 0 | asig1 |= 1; |
1295 | 0 | ST0 = normalizeRoundAndPackFloatx80(floatx80_precision_x, |
1296 | 0 | asign, aexp, asig0, asig1, |
1297 | 0 | &env->fp_status); |
1298 | 0 | } |
1299 | |
|
1300 | 0 | set_floatx80_rounding_precision(save_prec, &env->fp_status); |
1301 | 0 | } |
1302 | 0 | merge_exception_flags(env, old_flags); |
1303 | 0 | } |
1304 | | |
1305 | | void HELPER(fptan)(CPUX86State *env) |
1306 | 0 | { |
1307 | 0 | double fptemp = floatx80_to_double(env, ST0); |
1308 | |
|
1309 | 0 | if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) { |
1310 | 0 | env->fpus |= 0x400; |
1311 | 0 | } else { |
1312 | 0 | fptemp = tan(fptemp); |
1313 | 0 | ST0 = double_to_floatx80(env, fptemp); |
1314 | 0 | fpush(env); |
1315 | 0 | ST0 = floatx80_one; |
1316 | 0 | env->fpus &= ~0x400; /* C2 <-- 0 */ |
1317 | | /* the above code is for |arg| < 2**52 only */ |
1318 | 0 | } |
1319 | 0 | } |
1320 | | |
1321 | | /* Values of pi/4, pi/2, 3pi/4 and pi, with 128-bit precision. */ |
1322 | 0 | #define pi_4_exp 0x3ffe |
1323 | 0 | #define pi_4_sig_high 0xc90fdaa22168c234ULL |
1324 | 0 | #define pi_4_sig_low 0xc4c6628b80dc1cd1ULL |
1325 | 0 | #define pi_2_exp 0x3fff |
1326 | 0 | #define pi_2_sig_high 0xc90fdaa22168c234ULL |
1327 | 0 | #define pi_2_sig_low 0xc4c6628b80dc1cd1ULL |
1328 | 0 | #define pi_34_exp 0x4000 |
1329 | 0 | #define pi_34_sig_high 0x96cbe3f9990e91a7ULL |
1330 | 0 | #define pi_34_sig_low 0x9394c9e8a0a5159dULL |
1331 | 0 | #define pi_exp 0x4000 |
1332 | 0 | #define pi_sig_high 0xc90fdaa22168c234ULL |
1333 | 0 | #define pi_sig_low 0xc4c6628b80dc1cd1ULL |
1334 | | |
1335 | | /* |
1336 | | * Polynomial coefficients for an approximation to atan(x), with only |
1337 | | * odd powers of x used, for x in the interval [-1/16, 1/16]. (Unlike |
1338 | | * for some other approximations, no low part is needed for the first |
1339 | | * coefficient here to achieve a sufficiently accurate result, because |
1340 | | * the coefficient in this minimax approximation is very close to |
1341 | | * exactly 1.) |
1342 | | */ |
1343 | 0 | #define fpatan_coeff_0 make_floatx80(0x3fff, 0x8000000000000000ULL) |
1344 | 0 | #define fpatan_coeff_1 make_floatx80(0xbffd, 0xaaaaaaaaaaaaaa43ULL) |
1345 | 0 | #define fpatan_coeff_2 make_floatx80(0x3ffc, 0xccccccccccbfe4f8ULL) |
1346 | 0 | #define fpatan_coeff_3 make_floatx80(0xbffc, 0x92492491fbab2e66ULL) |
1347 | 0 | #define fpatan_coeff_4 make_floatx80(0x3ffb, 0xe38e372881ea1e0bULL) |
1348 | 0 | #define fpatan_coeff_5 make_floatx80(0xbffb, 0xba2c0104bbdd0615ULL) |
1349 | 0 | #define fpatan_coeff_6 make_floatx80(0x3ffb, 0x9baf7ebf898b42efULL) |
1350 | | |
1351 | | struct fpatan_data { |
1352 | | /* High and low parts of atan(x). */ |
1353 | | floatx80 atan_high, atan_low; |
1354 | | }; |
1355 | | |
1356 | | static const struct fpatan_data fpatan_table[9] = { |
1357 | | { floatx80_zero_init, |
1358 | | floatx80_zero_init }, |
1359 | | { make_floatx80_init(0x3ffb, 0xfeadd4d5617b6e33ULL), |
1360 | | make_floatx80_init(0xbfb9, 0xdda19d8305ddc420ULL) }, |
1361 | | { make_floatx80_init(0x3ffc, 0xfadbafc96406eb15ULL), |
1362 | | make_floatx80_init(0x3fbb, 0xdb8f3debef442fccULL) }, |
1363 | | { make_floatx80_init(0x3ffd, 0xb7b0ca0f26f78474ULL), |
1364 | | make_floatx80_init(0xbfbc, 0xeab9bdba460376faULL) }, |
1365 | | { make_floatx80_init(0x3ffd, 0xed63382b0dda7b45ULL), |
1366 | | make_floatx80_init(0x3fbc, 0xdfc88bd978751a06ULL) }, |
1367 | | { make_floatx80_init(0x3ffe, 0x8f005d5ef7f59f9bULL), |
1368 | | make_floatx80_init(0x3fbd, 0xb906bc2ccb886e90ULL) }, |
1369 | | { make_floatx80_init(0x3ffe, 0xa4bc7d1934f70924ULL), |
1370 | | make_floatx80_init(0x3fbb, 0xcd43f9522bed64f8ULL) }, |
1371 | | { make_floatx80_init(0x3ffe, 0xb8053e2bc2319e74ULL), |
1372 | | make_floatx80_init(0xbfbc, 0xd3496ab7bd6eef0cULL) }, |
1373 | | { make_floatx80_init(0x3ffe, 0xc90fdaa22168c235ULL), |
1374 | | make_floatx80_init(0xbfbc, 0xece675d1fc8f8cbcULL) }, |
1375 | | }; |
1376 | | |
1377 | | void HELPER(fpatan)(CPUX86State *env) |
1378 | 0 | { |
1379 | 0 | int old_flags = save_exception_flags(env); |
1380 | 0 | uint64_t arg0_sig = extractFloatx80Frac(ST0); |
1381 | 0 | int32_t arg0_exp = extractFloatx80Exp(ST0); |
1382 | 0 | bool arg0_sign = extractFloatx80Sign(ST0); |
1383 | 0 | uint64_t arg1_sig = extractFloatx80Frac(ST1); |
1384 | 0 | int32_t arg1_exp = extractFloatx80Exp(ST1); |
1385 | 0 | bool arg1_sign = extractFloatx80Sign(ST1); |
1386 | |
|
1387 | 0 | if (floatx80_invalid_encoding(ST0, &env->fp_status) || |
1388 | 0 | floatx80_invalid_encoding(ST1, &env->fp_status)) { |
1389 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
1390 | 0 | ST1 = floatx80_default_nan(&env->fp_status); |
1391 | 0 | } else if (floatx80_is_signaling_nan(ST0, &env->fp_status)) { |
1392 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
1393 | 0 | ST1 = floatx80_silence_nan(ST0, &env->fp_status); |
1394 | 0 | } else if (floatx80_is_signaling_nan(ST1, &env->fp_status)) { |
1395 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
1396 | 0 | ST1 = floatx80_silence_nan(ST1, &env->fp_status); |
1397 | 0 | } else if (floatx80_is_any_nan(ST0)) { |
1398 | 0 | ST1 = ST0; |
1399 | 0 | } else if (floatx80_is_any_nan(ST1)) { |
1400 | | /* Pass this NaN through. */ |
1401 | 0 | } else if (floatx80_is_zero(ST1) && !arg0_sign) { |
1402 | | /* Pass this zero through. */ |
1403 | 0 | } else if (((floatx80_is_infinity(ST0, &env->fp_status) && |
1404 | 0 | !floatx80_is_infinity(ST1, &env->fp_status)) || |
1405 | 0 | arg0_exp - arg1_exp >= 80) && |
1406 | 0 | !arg0_sign) { |
1407 | | /* |
1408 | | * Dividing ST1 by ST0 gives the correct result up to |
1409 | | * rounding, and avoids spurious underflow exceptions that |
1410 | | * might result from passing some small values through the |
1411 | | * polynomial approximation, but if a finite nonzero result of |
1412 | | * division is exact, the result of fpatan is still inexact |
1413 | | * (and underflowing where appropriate). |
1414 | | */ |
1415 | 0 | FloatX80RoundPrec save_prec = |
1416 | 0 | get_floatx80_rounding_precision(&env->fp_status); |
1417 | 0 | set_floatx80_rounding_precision(floatx80_precision_x, &env->fp_status); |
1418 | 0 | ST1 = floatx80_div(ST1, ST0, &env->fp_status); |
1419 | 0 | set_floatx80_rounding_precision(save_prec, &env->fp_status); |
1420 | 0 | if (!floatx80_is_zero(ST1) && |
1421 | 0 | !(get_float_exception_flags(&env->fp_status) & |
1422 | 0 | float_flag_inexact)) { |
1423 | | /* |
1424 | | * The mathematical result is very slightly closer to zero |
1425 | | * than this exact result. Round a value with the |
1426 | | * significand adjusted accordingly to get the correct |
1427 | | * exceptions, and possibly an adjusted result depending |
1428 | | * on the rounding mode. |
1429 | | */ |
1430 | 0 | uint64_t sig = extractFloatx80Frac(ST1); |
1431 | 0 | int32_t exp = extractFloatx80Exp(ST1); |
1432 | 0 | bool sign = extractFloatx80Sign(ST1); |
1433 | 0 | if (exp == 0) { |
1434 | 0 | normalizeFloatx80Subnormal(sig, &exp, &sig); |
1435 | 0 | } |
1436 | 0 | ST1 = normalizeRoundAndPackFloatx80(floatx80_precision_x, |
1437 | 0 | sign, exp, sig - 1, |
1438 | 0 | -1, &env->fp_status); |
1439 | 0 | } |
1440 | 0 | } else { |
1441 | | /* The result is inexact. */ |
1442 | 0 | bool rsign = arg1_sign; |
1443 | 0 | int32_t rexp; |
1444 | 0 | uint64_t rsig0, rsig1; |
1445 | 0 | if (floatx80_is_zero(ST1)) { |
1446 | | /* |
1447 | | * ST0 is negative. The result is pi with the sign of |
1448 | | * ST1. |
1449 | | */ |
1450 | 0 | rexp = pi_exp; |
1451 | 0 | rsig0 = pi_sig_high; |
1452 | 0 | rsig1 = pi_sig_low; |
1453 | 0 | } else if (floatx80_is_infinity(ST1, &env->fp_status)) { |
1454 | 0 | if (floatx80_is_infinity(ST0, &env->fp_status)) { |
1455 | 0 | if (arg0_sign) { |
1456 | 0 | rexp = pi_34_exp; |
1457 | 0 | rsig0 = pi_34_sig_high; |
1458 | 0 | rsig1 = pi_34_sig_low; |
1459 | 0 | } else { |
1460 | 0 | rexp = pi_4_exp; |
1461 | 0 | rsig0 = pi_4_sig_high; |
1462 | 0 | rsig1 = pi_4_sig_low; |
1463 | 0 | } |
1464 | 0 | } else { |
1465 | 0 | rexp = pi_2_exp; |
1466 | 0 | rsig0 = pi_2_sig_high; |
1467 | 0 | rsig1 = pi_2_sig_low; |
1468 | 0 | } |
1469 | 0 | } else if (floatx80_is_zero(ST0) || arg1_exp - arg0_exp >= 80) { |
1470 | 0 | rexp = pi_2_exp; |
1471 | 0 | rsig0 = pi_2_sig_high; |
1472 | 0 | rsig1 = pi_2_sig_low; |
1473 | 0 | } else if (floatx80_is_infinity(ST0, &env->fp_status) || |
1474 | 0 | arg0_exp - arg1_exp >= 80) { |
1475 | | /* ST0 is negative. */ |
1476 | 0 | rexp = pi_exp; |
1477 | 0 | rsig0 = pi_sig_high; |
1478 | 0 | rsig1 = pi_sig_low; |
1479 | 0 | } else { |
1480 | | /* |
1481 | | * ST0 and ST1 are finite, nonzero and with exponents not |
1482 | | * too far apart. |
1483 | | */ |
1484 | 0 | int32_t adj_exp, num_exp, den_exp, xexp, yexp, n, texp, zexp, aexp; |
1485 | 0 | int32_t azexp, axexp; |
1486 | 0 | bool adj_sub, ysign, zsign; |
1487 | 0 | uint64_t adj_sig0, adj_sig1, num_sig, den_sig, xsig0, xsig1; |
1488 | 0 | uint64_t msig0, msig1, msig2, remsig0, remsig1, remsig2; |
1489 | 0 | uint64_t ysig0, ysig1, tsig, zsig0, zsig1, asig0, asig1; |
1490 | 0 | uint64_t azsig0, azsig1; |
1491 | 0 | uint64_t azsig2, azsig3, axsig0, axsig1; |
1492 | 0 | floatx80 x8; |
1493 | 0 | FloatRoundMode save_mode = get_float_rounding_mode(&env->fp_status); |
1494 | 0 | FloatX80RoundPrec save_prec = |
1495 | 0 | get_floatx80_rounding_precision(&env->fp_status); |
1496 | 0 | set_float_rounding_mode(float_round_nearest_even, &env->fp_status); |
1497 | 0 | set_floatx80_rounding_precision(floatx80_precision_x, |
1498 | 0 | &env->fp_status); |
1499 | |
|
1500 | 0 | if (arg0_exp == 0) { |
1501 | 0 | normalizeFloatx80Subnormal(arg0_sig, &arg0_exp, &arg0_sig); |
1502 | 0 | } |
1503 | 0 | if (arg1_exp == 0) { |
1504 | 0 | normalizeFloatx80Subnormal(arg1_sig, &arg1_exp, &arg1_sig); |
1505 | 0 | } |
1506 | 0 | if (arg0_exp > arg1_exp || |
1507 | 0 | (arg0_exp == arg1_exp && arg0_sig >= arg1_sig)) { |
1508 | | /* Work with abs(ST1) / abs(ST0). */ |
1509 | 0 | num_exp = arg1_exp; |
1510 | 0 | num_sig = arg1_sig; |
1511 | 0 | den_exp = arg0_exp; |
1512 | 0 | den_sig = arg0_sig; |
1513 | 0 | if (arg0_sign) { |
1514 | | /* The result is subtracted from pi. */ |
1515 | 0 | adj_exp = pi_exp; |
1516 | 0 | adj_sig0 = pi_sig_high; |
1517 | 0 | adj_sig1 = pi_sig_low; |
1518 | 0 | adj_sub = true; |
1519 | 0 | } else { |
1520 | | /* The result is used as-is. */ |
1521 | 0 | adj_exp = 0; |
1522 | 0 | adj_sig0 = 0; |
1523 | 0 | adj_sig1 = 0; |
1524 | 0 | adj_sub = false; |
1525 | 0 | } |
1526 | 0 | } else { |
1527 | | /* Work with abs(ST0) / abs(ST1). */ |
1528 | 0 | num_exp = arg0_exp; |
1529 | 0 | num_sig = arg0_sig; |
1530 | 0 | den_exp = arg1_exp; |
1531 | 0 | den_sig = arg1_sig; |
1532 | | /* The result is added to or subtracted from pi/2. */ |
1533 | 0 | adj_exp = pi_2_exp; |
1534 | 0 | adj_sig0 = pi_2_sig_high; |
1535 | 0 | adj_sig1 = pi_2_sig_low; |
1536 | 0 | adj_sub = !arg0_sign; |
1537 | 0 | } |
1538 | | |
1539 | | /* |
1540 | | * Compute x = num/den, where 0 < x <= 1 and x is not too |
1541 | | * small. |
1542 | | */ |
1543 | 0 | xexp = num_exp - den_exp + 0x3ffe; |
1544 | 0 | remsig0 = num_sig; |
1545 | 0 | remsig1 = 0; |
1546 | 0 | if (den_sig <= remsig0) { |
1547 | 0 | shift128Right(remsig0, remsig1, 1, &remsig0, &remsig1); |
1548 | 0 | ++xexp; |
1549 | 0 | } |
1550 | 0 | xsig0 = estimateDiv128To64(remsig0, remsig1, den_sig); |
1551 | 0 | mul64To128(den_sig, xsig0, &msig0, &msig1); |
1552 | 0 | sub128(remsig0, remsig1, msig0, msig1, &remsig0, &remsig1); |
1553 | 0 | while ((int64_t) remsig0 < 0) { |
1554 | 0 | --xsig0; |
1555 | 0 | add128(remsig0, remsig1, 0, den_sig, &remsig0, &remsig1); |
1556 | 0 | } |
1557 | 0 | xsig1 = estimateDiv128To64(remsig1, 0, den_sig); |
1558 | | /* |
1559 | | * No need to correct any estimation error in xsig1; even |
1560 | | * with such error, it is accurate enough. |
1561 | | */ |
1562 | | |
1563 | | /* |
1564 | | * Split x as x = t + y, where t = n/8 is the nearest |
1565 | | * multiple of 1/8 to x. |
1566 | | */ |
1567 | 0 | x8 = normalizeRoundAndPackFloatx80(floatx80_precision_x, |
1568 | 0 | false, xexp + 3, xsig0, |
1569 | 0 | xsig1, &env->fp_status); |
1570 | 0 | n = floatx80_to_int32(x8, &env->fp_status); |
1571 | 0 | if (n == 0) { |
1572 | 0 | ysign = false; |
1573 | 0 | yexp = xexp; |
1574 | 0 | ysig0 = xsig0; |
1575 | 0 | ysig1 = xsig1; |
1576 | 0 | texp = 0; |
1577 | 0 | tsig = 0; |
1578 | 0 | } else { |
1579 | 0 | int shift = clz32(n) + 32; |
1580 | 0 | texp = 0x403b - shift; |
1581 | 0 | tsig = n; |
1582 | 0 | tsig <<= shift; |
1583 | 0 | if (texp == xexp) { |
1584 | 0 | sub128(xsig0, xsig1, tsig, 0, &ysig0, &ysig1); |
1585 | 0 | if ((int64_t) ysig0 >= 0) { |
1586 | 0 | ysign = false; |
1587 | 0 | if (ysig0 == 0) { |
1588 | 0 | if (ysig1 == 0) { |
1589 | 0 | yexp = 0; |
1590 | 0 | } else { |
1591 | 0 | shift = clz64(ysig1) + 64; |
1592 | 0 | yexp = xexp - shift; |
1593 | 0 | shift128Left(ysig0, ysig1, shift, |
1594 | 0 | &ysig0, &ysig1); |
1595 | 0 | } |
1596 | 0 | } else { |
1597 | 0 | shift = clz64(ysig0); |
1598 | 0 | yexp = xexp - shift; |
1599 | 0 | shift128Left(ysig0, ysig1, shift, &ysig0, &ysig1); |
1600 | 0 | } |
1601 | 0 | } else { |
1602 | 0 | ysign = true; |
1603 | 0 | sub128(0, 0, ysig0, ysig1, &ysig0, &ysig1); |
1604 | 0 | if (ysig0 == 0) { |
1605 | 0 | shift = clz64(ysig1) + 64; |
1606 | 0 | } else { |
1607 | 0 | shift = clz64(ysig0); |
1608 | 0 | } |
1609 | 0 | yexp = xexp - shift; |
1610 | 0 | shift128Left(ysig0, ysig1, shift, &ysig0, &ysig1); |
1611 | 0 | } |
1612 | 0 | } else { |
1613 | | /* |
1614 | | * t's exponent must be greater than x's because t |
1615 | | * is positive and the nearest multiple of 1/8 to |
1616 | | * x, and if x has a greater exponent, the power |
1617 | | * of 2 with that exponent is also a multiple of |
1618 | | * 1/8. |
1619 | | */ |
1620 | 0 | uint64_t usig0, usig1; |
1621 | 0 | shift128RightJamming(xsig0, xsig1, texp - xexp, |
1622 | 0 | &usig0, &usig1); |
1623 | 0 | ysign = true; |
1624 | 0 | sub128(tsig, 0, usig0, usig1, &ysig0, &ysig1); |
1625 | 0 | if (ysig0 == 0) { |
1626 | 0 | shift = clz64(ysig1) + 64; |
1627 | 0 | } else { |
1628 | 0 | shift = clz64(ysig0); |
1629 | 0 | } |
1630 | 0 | yexp = texp - shift; |
1631 | 0 | shift128Left(ysig0, ysig1, shift, &ysig0, &ysig1); |
1632 | 0 | } |
1633 | 0 | } |
1634 | | |
1635 | | /* |
1636 | | * Compute z = y/(1+tx), so arctan(x) = arctan(t) + |
1637 | | * arctan(z). |
1638 | | */ |
1639 | 0 | zsign = ysign; |
1640 | 0 | if (texp == 0 || yexp == 0) { |
1641 | 0 | zexp = yexp; |
1642 | 0 | zsig0 = ysig0; |
1643 | 0 | zsig1 = ysig1; |
1644 | 0 | } else { |
1645 | | /* |
1646 | | * t <= 1, x <= 1 and if both are 1 then y is 0, so tx < 1. |
1647 | | */ |
1648 | 0 | int32_t dexp = texp + xexp - 0x3ffe; |
1649 | 0 | uint64_t dsig0, dsig1, dsig2; |
1650 | 0 | mul128By64To192(xsig0, xsig1, tsig, &dsig0, &dsig1, &dsig2); |
1651 | | /* |
1652 | | * dexp <= 0x3fff (and if equal, dsig0 has a leading 0 |
1653 | | * bit). Add 1 to produce the denominator 1+tx. |
1654 | | */ |
1655 | 0 | shift128RightJamming(dsig0, dsig1, 0x3fff - dexp, |
1656 | 0 | &dsig0, &dsig1); |
1657 | 0 | dsig0 |= 0x8000000000000000ULL; |
1658 | 0 | zexp = yexp - 1; |
1659 | 0 | remsig0 = ysig0; |
1660 | 0 | remsig1 = ysig1; |
1661 | 0 | remsig2 = 0; |
1662 | 0 | if (dsig0 <= remsig0) { |
1663 | 0 | shift128Right(remsig0, remsig1, 1, &remsig0, &remsig1); |
1664 | 0 | ++zexp; |
1665 | 0 | } |
1666 | 0 | zsig0 = estimateDiv128To64(remsig0, remsig1, dsig0); |
1667 | 0 | mul128By64To192(dsig0, dsig1, zsig0, &msig0, &msig1, &msig2); |
1668 | 0 | sub192(remsig0, remsig1, remsig2, msig0, msig1, msig2, |
1669 | 0 | &remsig0, &remsig1, &remsig2); |
1670 | 0 | while ((int64_t) remsig0 < 0) { |
1671 | 0 | --zsig0; |
1672 | 0 | add192(remsig0, remsig1, remsig2, 0, dsig0, dsig1, |
1673 | 0 | &remsig0, &remsig1, &remsig2); |
1674 | 0 | } |
1675 | 0 | zsig1 = estimateDiv128To64(remsig1, remsig2, dsig0); |
1676 | | /* No need to correct any estimation error in zsig1. */ |
1677 | 0 | } |
1678 | |
|
1679 | 0 | if (zexp == 0) { |
1680 | 0 | azexp = 0; |
1681 | 0 | azsig0 = 0; |
1682 | 0 | azsig1 = 0; |
1683 | 0 | } else { |
1684 | 0 | floatx80 z2, accum; |
1685 | 0 | uint64_t z2sig0, z2sig1, z2sig2, z2sig3; |
1686 | | /* Compute z^2. */ |
1687 | 0 | mul128To256(zsig0, zsig1, zsig0, zsig1, |
1688 | 0 | &z2sig0, &z2sig1, &z2sig2, &z2sig3); |
1689 | 0 | z2 = normalizeRoundAndPackFloatx80(floatx80_precision_x, false, |
1690 | 0 | zexp + zexp - 0x3ffe, |
1691 | 0 | z2sig0, z2sig1, |
1692 | 0 | &env->fp_status); |
1693 | | |
1694 | | /* Compute the lower parts of the polynomial expansion. */ |
1695 | 0 | accum = floatx80_mul(fpatan_coeff_6, z2, &env->fp_status); |
1696 | 0 | accum = floatx80_add(fpatan_coeff_5, accum, &env->fp_status); |
1697 | 0 | accum = floatx80_mul(accum, z2, &env->fp_status); |
1698 | 0 | accum = floatx80_add(fpatan_coeff_4, accum, &env->fp_status); |
1699 | 0 | accum = floatx80_mul(accum, z2, &env->fp_status); |
1700 | 0 | accum = floatx80_add(fpatan_coeff_3, accum, &env->fp_status); |
1701 | 0 | accum = floatx80_mul(accum, z2, &env->fp_status); |
1702 | 0 | accum = floatx80_add(fpatan_coeff_2, accum, &env->fp_status); |
1703 | 0 | accum = floatx80_mul(accum, z2, &env->fp_status); |
1704 | 0 | accum = floatx80_add(fpatan_coeff_1, accum, &env->fp_status); |
1705 | 0 | accum = floatx80_mul(accum, z2, &env->fp_status); |
1706 | | |
1707 | | /* |
1708 | | * The full polynomial expansion is z*(fpatan_coeff_0 + accum). |
1709 | | * fpatan_coeff_0 is 1, and accum is negative and much smaller. |
1710 | | */ |
1711 | 0 | aexp = extractFloatx80Exp(fpatan_coeff_0); |
1712 | 0 | shift128RightJamming(extractFloatx80Frac(accum), 0, |
1713 | 0 | aexp - extractFloatx80Exp(accum), |
1714 | 0 | &asig0, &asig1); |
1715 | 0 | sub128(extractFloatx80Frac(fpatan_coeff_0), 0, asig0, asig1, |
1716 | 0 | &asig0, &asig1); |
1717 | | /* Multiply by z to compute arctan(z). */ |
1718 | 0 | azexp = aexp + zexp - 0x3ffe; |
1719 | 0 | mul128To256(asig0, asig1, zsig0, zsig1, &azsig0, &azsig1, |
1720 | 0 | &azsig2, &azsig3); |
1721 | 0 | } |
1722 | | |
1723 | | /* Add arctan(t) (positive or zero) and arctan(z) (sign zsign). */ |
1724 | 0 | if (texp == 0) { |
1725 | | /* z is positive. */ |
1726 | 0 | axexp = azexp; |
1727 | 0 | axsig0 = azsig0; |
1728 | 0 | axsig1 = azsig1; |
1729 | 0 | } else { |
1730 | 0 | bool low_sign = extractFloatx80Sign(fpatan_table[n].atan_low); |
1731 | 0 | int32_t low_exp = extractFloatx80Exp(fpatan_table[n].atan_low); |
1732 | 0 | uint64_t low_sig0 = |
1733 | 0 | extractFloatx80Frac(fpatan_table[n].atan_low); |
1734 | 0 | uint64_t low_sig1 = 0; |
1735 | 0 | axexp = extractFloatx80Exp(fpatan_table[n].atan_high); |
1736 | 0 | axsig0 = extractFloatx80Frac(fpatan_table[n].atan_high); |
1737 | 0 | axsig1 = 0; |
1738 | 0 | shift128RightJamming(low_sig0, low_sig1, axexp - low_exp, |
1739 | 0 | &low_sig0, &low_sig1); |
1740 | 0 | if (low_sign) { |
1741 | 0 | sub128(axsig0, axsig1, low_sig0, low_sig1, |
1742 | 0 | &axsig0, &axsig1); |
1743 | 0 | } else { |
1744 | 0 | add128(axsig0, axsig1, low_sig0, low_sig1, |
1745 | 0 | &axsig0, &axsig1); |
1746 | 0 | } |
1747 | 0 | if (azexp >= axexp) { |
1748 | 0 | shift128RightJamming(axsig0, axsig1, azexp - axexp + 1, |
1749 | 0 | &axsig0, &axsig1); |
1750 | 0 | axexp = azexp + 1; |
1751 | 0 | shift128RightJamming(azsig0, azsig1, 1, |
1752 | 0 | &azsig0, &azsig1); |
1753 | 0 | } else { |
1754 | 0 | shift128RightJamming(axsig0, axsig1, 1, |
1755 | 0 | &axsig0, &axsig1); |
1756 | 0 | shift128RightJamming(azsig0, azsig1, axexp - azexp + 1, |
1757 | 0 | &azsig0, &azsig1); |
1758 | 0 | ++axexp; |
1759 | 0 | } |
1760 | 0 | if (zsign) { |
1761 | 0 | sub128(axsig0, axsig1, azsig0, azsig1, |
1762 | 0 | &axsig0, &axsig1); |
1763 | 0 | } else { |
1764 | 0 | add128(axsig0, axsig1, azsig0, azsig1, |
1765 | 0 | &axsig0, &axsig1); |
1766 | 0 | } |
1767 | 0 | } |
1768 | |
|
1769 | 0 | if (adj_exp == 0) { |
1770 | 0 | rexp = axexp; |
1771 | 0 | rsig0 = axsig0; |
1772 | 0 | rsig1 = axsig1; |
1773 | 0 | } else { |
1774 | | /* |
1775 | | * Add or subtract arctan(x) (exponent axexp, |
1776 | | * significand axsig0 and axsig1, positive, not |
1777 | | * necessarily normalized) to the number given by |
1778 | | * adj_exp, adj_sig0 and adj_sig1, according to |
1779 | | * adj_sub. |
1780 | | */ |
1781 | 0 | if (adj_exp >= axexp) { |
1782 | 0 | shift128RightJamming(axsig0, axsig1, adj_exp - axexp + 1, |
1783 | 0 | &axsig0, &axsig1); |
1784 | 0 | rexp = adj_exp + 1; |
1785 | 0 | shift128RightJamming(adj_sig0, adj_sig1, 1, |
1786 | 0 | &adj_sig0, &adj_sig1); |
1787 | 0 | } else { |
1788 | 0 | shift128RightJamming(axsig0, axsig1, 1, |
1789 | 0 | &axsig0, &axsig1); |
1790 | 0 | shift128RightJamming(adj_sig0, adj_sig1, |
1791 | 0 | axexp - adj_exp + 1, |
1792 | 0 | &adj_sig0, &adj_sig1); |
1793 | 0 | rexp = axexp + 1; |
1794 | 0 | } |
1795 | 0 | if (adj_sub) { |
1796 | 0 | sub128(adj_sig0, adj_sig1, axsig0, axsig1, |
1797 | 0 | &rsig0, &rsig1); |
1798 | 0 | } else { |
1799 | 0 | add128(adj_sig0, adj_sig1, axsig0, axsig1, |
1800 | 0 | &rsig0, &rsig1); |
1801 | 0 | } |
1802 | 0 | } |
1803 | |
|
1804 | 0 | set_float_rounding_mode(save_mode, &env->fp_status); |
1805 | 0 | set_floatx80_rounding_precision(save_prec, &env->fp_status); |
1806 | 0 | } |
1807 | | /* This result is inexact. */ |
1808 | 0 | rsig1 |= 1; |
1809 | 0 | ST1 = normalizeRoundAndPackFloatx80(floatx80_precision_x, rsign, rexp, |
1810 | 0 | rsig0, rsig1, &env->fp_status); |
1811 | 0 | } |
1812 | |
|
1813 | 0 | fpop(env); |
1814 | 0 | merge_exception_flags(env, old_flags); |
1815 | 0 | } |
1816 | | |
1817 | | /* fpush() only validates the new top. FXTRACT also needs ST(1) validated. */ |
1818 | | static inline void fpush_fxtract(CPUX86State *env) |
1819 | 0 | { |
1820 | 0 | fpush(env); |
1821 | 0 | env->fptags[(env->fpstt + 1) & 7] = 0; |
1822 | 0 | } |
1823 | | |
1824 | | void HELPER(fxtract)(CPUX86State *env) |
1825 | 0 | { |
1826 | 0 | int old_flags = save_exception_flags(env); |
1827 | 0 | CPU_LDoubleU temp; |
1828 | |
|
1829 | 0 | temp.d = ST0; |
1830 | |
|
1831 | 0 | if (floatx80_is_zero(ST0)) { |
1832 | | /* Easy way to generate -inf and raising division by 0 exception */ |
1833 | 0 | ST0 = floatx80_div(floatx80_chs(floatx80_one), floatx80_zero, |
1834 | 0 | &env->fp_status); |
1835 | 0 | fpush_fxtract(env); |
1836 | 0 | ST0 = temp.d; |
1837 | 0 | } else if (floatx80_invalid_encoding(ST0, &env->fp_status)) { |
1838 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
1839 | 0 | ST0 = floatx80_default_nan(&env->fp_status); |
1840 | 0 | fpush_fxtract(env); |
1841 | 0 | ST0 = ST1; |
1842 | 0 | } else if (floatx80_is_any_nan(ST0)) { |
1843 | 0 | if (floatx80_is_signaling_nan(ST0, &env->fp_status)) { |
1844 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
1845 | 0 | ST0 = floatx80_silence_nan(ST0, &env->fp_status); |
1846 | 0 | } |
1847 | 0 | fpush_fxtract(env); |
1848 | 0 | ST0 = ST1; |
1849 | 0 | } else if (floatx80_is_infinity(ST0, &env->fp_status)) { |
1850 | 0 | fpush_fxtract(env); |
1851 | 0 | ST0 = ST1; |
1852 | 0 | ST1 = floatx80_default_inf(0, &env->fp_status); |
1853 | 0 | } else { |
1854 | 0 | int expdif; |
1855 | |
|
1856 | 0 | if (EXPD(temp) == 0) { |
1857 | 0 | int shift = clz64(temp.l.lower); |
1858 | 0 | temp.l.lower <<= shift; |
1859 | 0 | expdif = 1 - EXPBIAS - shift; |
1860 | 0 | float_raise(float_flag_input_denormal_flushed, &env->fp_status); |
1861 | 0 | } else { |
1862 | 0 | expdif = EXPD(temp) - EXPBIAS; |
1863 | 0 | } |
1864 | | /* DP exponent bias */ |
1865 | 0 | ST0 = int32_to_floatx80(expdif, &env->fp_status); |
1866 | 0 | fpush_fxtract(env); |
1867 | 0 | BIASEXPONENT(temp); |
1868 | 0 | ST0 = temp.d; |
1869 | 0 | } |
1870 | 0 | merge_exception_flags(env, old_flags); |
1871 | 0 | } |
1872 | | |
1873 | | static void helper_fprem_common(CPUX86State *env, bool mod) |
1874 | 0 | { |
1875 | 0 | int old_flags = save_exception_flags(env); |
1876 | 0 | uint64_t quotient; |
1877 | 0 | CPU_LDoubleU temp0, temp1; |
1878 | 0 | int exp0, exp1, expdiff; |
1879 | |
|
1880 | 0 | temp0.d = ST0; |
1881 | 0 | temp1.d = ST1; |
1882 | 0 | exp0 = EXPD(temp0); |
1883 | 0 | exp1 = EXPD(temp1); |
1884 | |
|
1885 | 0 | env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */ |
1886 | 0 | if (floatx80_is_zero(ST0) || floatx80_is_zero(ST1) || |
1887 | 0 | exp0 == 0x7fff || exp1 == 0x7fff || |
1888 | 0 | floatx80_invalid_encoding(ST0, &env->fp_status) || |
1889 | 0 | floatx80_invalid_encoding(ST1, &env->fp_status)) { |
1890 | 0 | ST0 = floatx80_modrem(ST0, ST1, mod, "ient, &env->fp_status); |
1891 | 0 | } else { |
1892 | 0 | if (exp0 == 0) { |
1893 | 0 | exp0 = 1 - clz64(temp0.l.lower); |
1894 | 0 | } |
1895 | 0 | if (exp1 == 0) { |
1896 | 0 | exp1 = 1 - clz64(temp1.l.lower); |
1897 | 0 | } |
1898 | 0 | expdiff = exp0 - exp1; |
1899 | 0 | if (expdiff < 64) { |
1900 | 0 | ST0 = floatx80_modrem(ST0, ST1, mod, "ient, &env->fp_status); |
1901 | 0 | env->fpus |= (quotient & 0x4) << (8 - 2); /* (C0) <-- q2 */ |
1902 | 0 | env->fpus |= (quotient & 0x2) << (14 - 1); /* (C3) <-- q1 */ |
1903 | 0 | env->fpus |= (quotient & 0x1) << (9 - 0); /* (C1) <-- q0 */ |
1904 | 0 | } else { |
1905 | | /* |
1906 | | * Partial remainder. This choice of how many bits to |
1907 | | * process at once is specified in AMD instruction set |
1908 | | * manuals, and empirically is followed by Intel |
1909 | | * processors as well; it ensures that the final remainder |
1910 | | * operation in a loop does produce the correct low three |
1911 | | * bits of the quotient. AMD manuals specify that the |
1912 | | * flags other than C2 are cleared, and empirically Intel |
1913 | | * processors clear them as well. |
1914 | | */ |
1915 | 0 | int n = 32 + (expdiff % 32); |
1916 | 0 | temp1.d = floatx80_scalbn(temp1.d, expdiff - n, &env->fp_status); |
1917 | 0 | ST0 = floatx80_mod(ST0, temp1.d, &env->fp_status); |
1918 | 0 | env->fpus |= 0x400; /* C2 <-- 1 */ |
1919 | 0 | } |
1920 | 0 | } |
1921 | 0 | merge_exception_flags(env, old_flags); |
1922 | 0 | } |
1923 | | |
1924 | | void HELPER(fprem1)(CPUX86State *env) |
1925 | 0 | { |
1926 | 0 | helper_fprem_common(env, false); |
1927 | 0 | } |
1928 | | |
1929 | | void HELPER(fprem)(CPUX86State *env) |
1930 | 0 | { |
1931 | 0 | helper_fprem_common(env, true); |
1932 | 0 | } |
1933 | | |
1934 | | /* 128-bit significand of log2(e). */ |
1935 | 0 | #define log2_e_sig_high 0xb8aa3b295c17f0bbULL |
1936 | 0 | #define log2_e_sig_low 0xbe87fed0691d3e89ULL |
1937 | | |
1938 | | /* |
1939 | | * Polynomial coefficients for an approximation to log2((1+x)/(1-x)), |
1940 | | * with only odd powers of x used, for x in the interval [2*sqrt(2)-3, |
1941 | | * 3-2*sqrt(2)], which corresponds to logarithms of numbers in the |
1942 | | * interval [sqrt(2)/2, sqrt(2)]. |
1943 | | */ |
1944 | 0 | #define fyl2x_coeff_0 make_floatx80(0x4000, 0xb8aa3b295c17f0bcULL) |
1945 | 0 | #define fyl2x_coeff_0_low make_floatx80(0xbfbf, 0x834972fe2d7bab1bULL) |
1946 | 0 | #define fyl2x_coeff_1 make_floatx80(0x3ffe, 0xf6384ee1d01febb8ULL) |
1947 | 0 | #define fyl2x_coeff_2 make_floatx80(0x3ffe, 0x93bb62877cdfa2e3ULL) |
1948 | 0 | #define fyl2x_coeff_3 make_floatx80(0x3ffd, 0xd30bb153d808f269ULL) |
1949 | 0 | #define fyl2x_coeff_4 make_floatx80(0x3ffd, 0xa42589eaf451499eULL) |
1950 | 0 | #define fyl2x_coeff_5 make_floatx80(0x3ffd, 0x864d42c0f8f17517ULL) |
1951 | 0 | #define fyl2x_coeff_6 make_floatx80(0x3ffc, 0xe3476578adf26272ULL) |
1952 | 0 | #define fyl2x_coeff_7 make_floatx80(0x3ffc, 0xc506c5f874e6d80fULL) |
1953 | 0 | #define fyl2x_coeff_8 make_floatx80(0x3ffc, 0xac5cf50cc57d6372ULL) |
1954 | 0 | #define fyl2x_coeff_9 make_floatx80(0x3ffc, 0xb1ed0066d971a103ULL) |
1955 | | |
1956 | | /* |
1957 | | * Compute an approximation of log2(1+arg), where 1+arg is in the |
1958 | | * interval [sqrt(2)/2, sqrt(2)]. It is assumed that when this |
1959 | | * function is called, rounding precision is set to 80 and the |
1960 | | * round-to-nearest mode is in effect. arg must not be exactly zero, |
1961 | | * and must not be so close to zero that underflow might occur. |
1962 | | */ |
1963 | | static void helper_fyl2x_common(CPUX86State *env, floatx80 arg, int32_t *exp, |
1964 | | uint64_t *sig0, uint64_t *sig1) |
1965 | 0 | { |
1966 | 0 | uint64_t arg0_sig = extractFloatx80Frac(arg); |
1967 | 0 | int32_t arg0_exp = extractFloatx80Exp(arg); |
1968 | 0 | bool arg0_sign = extractFloatx80Sign(arg); |
1969 | 0 | bool asign; |
1970 | 0 | int32_t dexp, texp, aexp; |
1971 | 0 | uint64_t dsig0, dsig1, tsig0, tsig1, rsig0, rsig1, rsig2; |
1972 | 0 | uint64_t msig0, msig1, msig2, t2sig0, t2sig1, t2sig2, t2sig3; |
1973 | 0 | uint64_t asig0, asig1, asig2, asig3, bsig0, bsig1; |
1974 | 0 | floatx80 t2, accum; |
1975 | | |
1976 | | /* |
1977 | | * Compute an approximation of arg/(2+arg), with extra precision, |
1978 | | * as the argument to a polynomial approximation. The extra |
1979 | | * precision is only needed for the first term of the |
1980 | | * approximation, with subsequent terms being significantly |
1981 | | * smaller; the approximation only uses odd exponents, and the |
1982 | | * square of arg/(2+arg) is at most 17-12*sqrt(2) = 0.029.... |
1983 | | */ |
1984 | 0 | if (arg0_sign) { |
1985 | 0 | dexp = 0x3fff; |
1986 | 0 | shift128RightJamming(arg0_sig, 0, dexp - arg0_exp, &dsig0, &dsig1); |
1987 | 0 | sub128(0, 0, dsig0, dsig1, &dsig0, &dsig1); |
1988 | 0 | } else { |
1989 | 0 | dexp = 0x4000; |
1990 | 0 | shift128RightJamming(arg0_sig, 0, dexp - arg0_exp, &dsig0, &dsig1); |
1991 | 0 | dsig0 |= 0x8000000000000000ULL; |
1992 | 0 | } |
1993 | 0 | texp = arg0_exp - dexp + 0x3ffe; |
1994 | 0 | rsig0 = arg0_sig; |
1995 | 0 | rsig1 = 0; |
1996 | 0 | rsig2 = 0; |
1997 | 0 | if (dsig0 <= rsig0) { |
1998 | 0 | shift128Right(rsig0, rsig1, 1, &rsig0, &rsig1); |
1999 | 0 | ++texp; |
2000 | 0 | } |
2001 | 0 | tsig0 = estimateDiv128To64(rsig0, rsig1, dsig0); |
2002 | 0 | mul128By64To192(dsig0, dsig1, tsig0, &msig0, &msig1, &msig2); |
2003 | 0 | sub192(rsig0, rsig1, rsig2, msig0, msig1, msig2, |
2004 | 0 | &rsig0, &rsig1, &rsig2); |
2005 | 0 | while ((int64_t) rsig0 < 0) { |
2006 | 0 | --tsig0; |
2007 | 0 | add192(rsig0, rsig1, rsig2, 0, dsig0, dsig1, |
2008 | 0 | &rsig0, &rsig1, &rsig2); |
2009 | 0 | } |
2010 | 0 | tsig1 = estimateDiv128To64(rsig1, rsig2, dsig0); |
2011 | | /* |
2012 | | * No need to correct any estimation error in tsig1; even with |
2013 | | * such error, it is accurate enough. Now compute the square of |
2014 | | * that approximation. |
2015 | | */ |
2016 | 0 | mul128To256(tsig0, tsig1, tsig0, tsig1, |
2017 | 0 | &t2sig0, &t2sig1, &t2sig2, &t2sig3); |
2018 | 0 | t2 = normalizeRoundAndPackFloatx80(floatx80_precision_x, false, |
2019 | 0 | texp + texp - 0x3ffe, |
2020 | 0 | t2sig0, t2sig1, &env->fp_status); |
2021 | | |
2022 | | /* Compute the lower parts of the polynomial expansion. */ |
2023 | 0 | accum = floatx80_mul(fyl2x_coeff_9, t2, &env->fp_status); |
2024 | 0 | accum = floatx80_add(fyl2x_coeff_8, accum, &env->fp_status); |
2025 | 0 | accum = floatx80_mul(accum, t2, &env->fp_status); |
2026 | 0 | accum = floatx80_add(fyl2x_coeff_7, accum, &env->fp_status); |
2027 | 0 | accum = floatx80_mul(accum, t2, &env->fp_status); |
2028 | 0 | accum = floatx80_add(fyl2x_coeff_6, accum, &env->fp_status); |
2029 | 0 | accum = floatx80_mul(accum, t2, &env->fp_status); |
2030 | 0 | accum = floatx80_add(fyl2x_coeff_5, accum, &env->fp_status); |
2031 | 0 | accum = floatx80_mul(accum, t2, &env->fp_status); |
2032 | 0 | accum = floatx80_add(fyl2x_coeff_4, accum, &env->fp_status); |
2033 | 0 | accum = floatx80_mul(accum, t2, &env->fp_status); |
2034 | 0 | accum = floatx80_add(fyl2x_coeff_3, accum, &env->fp_status); |
2035 | 0 | accum = floatx80_mul(accum, t2, &env->fp_status); |
2036 | 0 | accum = floatx80_add(fyl2x_coeff_2, accum, &env->fp_status); |
2037 | 0 | accum = floatx80_mul(accum, t2, &env->fp_status); |
2038 | 0 | accum = floatx80_add(fyl2x_coeff_1, accum, &env->fp_status); |
2039 | 0 | accum = floatx80_mul(accum, t2, &env->fp_status); |
2040 | 0 | accum = floatx80_add(fyl2x_coeff_0_low, accum, &env->fp_status); |
2041 | | |
2042 | | /* |
2043 | | * The full polynomial expansion is fyl2x_coeff_0 + accum (where |
2044 | | * accum has much lower magnitude, and so, in particular, carry |
2045 | | * out of the addition is not possible), multiplied by t. (This |
2046 | | * expansion is only accurate to about 70 bits, not 128 bits.) |
2047 | | */ |
2048 | 0 | aexp = extractFloatx80Exp(fyl2x_coeff_0); |
2049 | 0 | asign = extractFloatx80Sign(fyl2x_coeff_0); |
2050 | 0 | shift128RightJamming(extractFloatx80Frac(accum), 0, |
2051 | 0 | aexp - extractFloatx80Exp(accum), |
2052 | 0 | &asig0, &asig1); |
2053 | 0 | bsig0 = extractFloatx80Frac(fyl2x_coeff_0); |
2054 | 0 | bsig1 = 0; |
2055 | 0 | if (asign == extractFloatx80Sign(accum)) { |
2056 | 0 | add128(bsig0, bsig1, asig0, asig1, &asig0, &asig1); |
2057 | 0 | } else { |
2058 | 0 | sub128(bsig0, bsig1, asig0, asig1, &asig0, &asig1); |
2059 | 0 | } |
2060 | | /* Multiply by t to compute the required result. */ |
2061 | 0 | mul128To256(asig0, asig1, tsig0, tsig1, |
2062 | 0 | &asig0, &asig1, &asig2, &asig3); |
2063 | 0 | aexp += texp - 0x3ffe; |
2064 | 0 | *exp = aexp; |
2065 | 0 | *sig0 = asig0; |
2066 | 0 | *sig1 = asig1; |
2067 | 0 | } |
2068 | | |
2069 | | void HELPER(fyl2xp1)(CPUX86State *env) |
2070 | 0 | { |
2071 | 0 | int old_flags = save_exception_flags(env); |
2072 | 0 | uint64_t arg0_sig = extractFloatx80Frac(ST0); |
2073 | 0 | int32_t arg0_exp = extractFloatx80Exp(ST0); |
2074 | 0 | bool arg0_sign = extractFloatx80Sign(ST0); |
2075 | 0 | uint64_t arg1_sig = extractFloatx80Frac(ST1); |
2076 | 0 | int32_t arg1_exp = extractFloatx80Exp(ST1); |
2077 | 0 | bool arg1_sign = extractFloatx80Sign(ST1); |
2078 | |
|
2079 | 0 | if (floatx80_invalid_encoding(ST0, &env->fp_status) || |
2080 | 0 | floatx80_invalid_encoding(ST1, &env->fp_status)) { |
2081 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2082 | 0 | ST1 = floatx80_default_nan(&env->fp_status); |
2083 | 0 | } else if (floatx80_is_signaling_nan(ST0, &env->fp_status)) { |
2084 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2085 | 0 | ST1 = floatx80_silence_nan(ST0, &env->fp_status); |
2086 | 0 | } else if (floatx80_is_signaling_nan(ST1, &env->fp_status)) { |
2087 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2088 | 0 | ST1 = floatx80_silence_nan(ST1, &env->fp_status); |
2089 | 0 | } else if (floatx80_is_any_nan(ST0)) { |
2090 | 0 | ST1 = ST0; |
2091 | 0 | } else if (floatx80_is_any_nan(ST1)) { |
2092 | | /* Pass this NaN through. */ |
2093 | 0 | } else if (arg0_exp > 0x3ffd || |
2094 | 0 | (arg0_exp == 0x3ffd && arg0_sig > (arg0_sign ? |
2095 | 0 | 0x95f619980c4336f7ULL : |
2096 | 0 | 0xd413cccfe7799211ULL))) { |
2097 | | /* |
2098 | | * Out of range for the instruction (ST0 must have absolute |
2099 | | * value less than 1 - sqrt(2)/2 = 0.292..., according to |
2100 | | * Intel manuals; AMD manuals allow a range from sqrt(2)/2 - 1 |
2101 | | * to sqrt(2) - 1, which we allow here), treat as invalid. |
2102 | | */ |
2103 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2104 | 0 | ST1 = floatx80_default_nan(&env->fp_status); |
2105 | 0 | } else if (floatx80_is_zero(ST0) || floatx80_is_zero(ST1) || |
2106 | 0 | arg1_exp == 0x7fff) { |
2107 | | /* |
2108 | | * One argument is zero, or multiplying by infinity; correct |
2109 | | * result is exact and can be obtained by multiplying the |
2110 | | * arguments. |
2111 | | */ |
2112 | 0 | ST1 = floatx80_mul(ST0, ST1, &env->fp_status); |
2113 | 0 | } else if (arg0_exp < 0x3fb0) { |
2114 | | /* |
2115 | | * Multiplying both arguments and an extra-precision version |
2116 | | * of log2(e) is sufficiently precise. |
2117 | | */ |
2118 | 0 | uint64_t sig0, sig1, sig2; |
2119 | 0 | int32_t exp; |
2120 | 0 | if (arg0_exp == 0) { |
2121 | 0 | normalizeFloatx80Subnormal(arg0_sig, &arg0_exp, &arg0_sig); |
2122 | 0 | } |
2123 | 0 | if (arg1_exp == 0) { |
2124 | 0 | normalizeFloatx80Subnormal(arg1_sig, &arg1_exp, &arg1_sig); |
2125 | 0 | } |
2126 | 0 | mul128By64To192(log2_e_sig_high, log2_e_sig_low, arg0_sig, |
2127 | 0 | &sig0, &sig1, &sig2); |
2128 | 0 | exp = arg0_exp + 1; |
2129 | 0 | mul128By64To192(sig0, sig1, arg1_sig, &sig0, &sig1, &sig2); |
2130 | 0 | exp += arg1_exp - 0x3ffe; |
2131 | | /* This result is inexact. */ |
2132 | 0 | sig1 |= 1; |
2133 | 0 | ST1 = normalizeRoundAndPackFloatx80(floatx80_precision_x, |
2134 | 0 | arg0_sign ^ arg1_sign, exp, |
2135 | 0 | sig0, sig1, &env->fp_status); |
2136 | 0 | } else { |
2137 | 0 | int32_t aexp; |
2138 | 0 | uint64_t asig0, asig1, asig2; |
2139 | 0 | FloatRoundMode save_mode = get_float_rounding_mode(&env->fp_status); |
2140 | 0 | FloatX80RoundPrec save_prec = |
2141 | 0 | get_floatx80_rounding_precision(&env->fp_status); |
2142 | 0 | set_float_rounding_mode(float_round_nearest_even, &env->fp_status); |
2143 | 0 | set_floatx80_rounding_precision(floatx80_precision_x, &env->fp_status); |
2144 | |
|
2145 | 0 | helper_fyl2x_common(env, ST0, &aexp, &asig0, &asig1); |
2146 | | /* |
2147 | | * Multiply by the second argument to compute the required |
2148 | | * result. |
2149 | | */ |
2150 | 0 | if (arg1_exp == 0) { |
2151 | 0 | normalizeFloatx80Subnormal(arg1_sig, &arg1_exp, &arg1_sig); |
2152 | 0 | } |
2153 | 0 | mul128By64To192(asig0, asig1, arg1_sig, &asig0, &asig1, &asig2); |
2154 | 0 | aexp += arg1_exp - 0x3ffe; |
2155 | | /* This result is inexact. */ |
2156 | 0 | asig1 |= 1; |
2157 | 0 | set_float_rounding_mode(save_mode, &env->fp_status); |
2158 | 0 | ST1 = normalizeRoundAndPackFloatx80(floatx80_precision_x, |
2159 | 0 | arg0_sign ^ arg1_sign, aexp, |
2160 | 0 | asig0, asig1, &env->fp_status); |
2161 | 0 | set_floatx80_rounding_precision(save_prec, &env->fp_status); |
2162 | 0 | } |
2163 | 0 | fpop(env); |
2164 | 0 | merge_exception_flags(env, old_flags); |
2165 | 0 | } |
2166 | | |
2167 | | void HELPER(fyl2x)(CPUX86State *env) |
2168 | 0 | { |
2169 | 0 | int old_flags = save_exception_flags(env); |
2170 | 0 | uint64_t arg0_sig = extractFloatx80Frac(ST0); |
2171 | 0 | int32_t arg0_exp = extractFloatx80Exp(ST0); |
2172 | 0 | bool arg0_sign = extractFloatx80Sign(ST0); |
2173 | 0 | uint64_t arg1_sig = extractFloatx80Frac(ST1); |
2174 | 0 | int32_t arg1_exp = extractFloatx80Exp(ST1); |
2175 | 0 | bool arg1_sign = extractFloatx80Sign(ST1); |
2176 | |
|
2177 | 0 | if (floatx80_invalid_encoding(ST0, &env->fp_status) || |
2178 | 0 | floatx80_invalid_encoding(ST1, &env->fp_status)) { |
2179 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2180 | 0 | ST1 = floatx80_default_nan(&env->fp_status); |
2181 | 0 | } else if (floatx80_is_signaling_nan(ST0, &env->fp_status)) { |
2182 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2183 | 0 | ST1 = floatx80_silence_nan(ST0, &env->fp_status); |
2184 | 0 | } else if (floatx80_is_signaling_nan(ST1, &env->fp_status)) { |
2185 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2186 | 0 | ST1 = floatx80_silence_nan(ST1, &env->fp_status); |
2187 | 0 | } else if (floatx80_is_any_nan(ST0)) { |
2188 | 0 | ST1 = ST0; |
2189 | 0 | } else if (floatx80_is_any_nan(ST1)) { |
2190 | | /* Pass this NaN through. */ |
2191 | 0 | } else if (arg0_sign && !floatx80_is_zero(ST0)) { |
2192 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2193 | 0 | ST1 = floatx80_default_nan(&env->fp_status); |
2194 | 0 | } else if (floatx80_is_infinity(ST1, &env->fp_status)) { |
2195 | 0 | FloatRelation cmp = floatx80_compare(ST0, floatx80_one, |
2196 | 0 | &env->fp_status); |
2197 | 0 | switch (cmp) { |
2198 | 0 | case float_relation_less: |
2199 | 0 | ST1 = floatx80_chs(ST1); |
2200 | 0 | break; |
2201 | 0 | case float_relation_greater: |
2202 | | /* Result is infinity of the same sign as ST1. */ |
2203 | 0 | break; |
2204 | 0 | default: |
2205 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2206 | 0 | ST1 = floatx80_default_nan(&env->fp_status); |
2207 | 0 | break; |
2208 | 0 | } |
2209 | 0 | } else if (floatx80_is_infinity(ST0, &env->fp_status)) { |
2210 | 0 | if (floatx80_is_zero(ST1)) { |
2211 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2212 | 0 | ST1 = floatx80_default_nan(&env->fp_status); |
2213 | 0 | } else if (arg1_sign) { |
2214 | 0 | ST1 = floatx80_chs(ST0); |
2215 | 0 | } else { |
2216 | 0 | ST1 = ST0; |
2217 | 0 | } |
2218 | 0 | } else if (floatx80_is_zero(ST0)) { |
2219 | 0 | if (floatx80_is_zero(ST1)) { |
2220 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2221 | 0 | ST1 = floatx80_default_nan(&env->fp_status); |
2222 | 0 | } else { |
2223 | | /* Result is infinity with opposite sign to ST1. */ |
2224 | 0 | float_raise(float_flag_divbyzero, &env->fp_status); |
2225 | 0 | ST1 = make_floatx80(arg1_sign ? 0x7fff : 0xffff, |
2226 | 0 | 0x8000000000000000ULL); |
2227 | 0 | } |
2228 | 0 | } else if (floatx80_is_zero(ST1)) { |
2229 | 0 | if (floatx80_lt(ST0, floatx80_one, &env->fp_status)) { |
2230 | 0 | ST1 = floatx80_chs(ST1); |
2231 | 0 | } |
2232 | | /* Otherwise, ST1 is already the correct result. */ |
2233 | 0 | } else if (floatx80_eq(ST0, floatx80_one, &env->fp_status)) { |
2234 | 0 | if (arg1_sign) { |
2235 | 0 | ST1 = floatx80_chs(floatx80_zero); |
2236 | 0 | } else { |
2237 | 0 | ST1 = floatx80_zero; |
2238 | 0 | } |
2239 | 0 | } else { |
2240 | 0 | int32_t int_exp; |
2241 | 0 | floatx80 arg0_m1; |
2242 | 0 | FloatRoundMode save_mode = get_float_rounding_mode(&env->fp_status); |
2243 | 0 | FloatX80RoundPrec save_prec = |
2244 | 0 | get_floatx80_rounding_precision(&env->fp_status); |
2245 | 0 | set_float_rounding_mode(float_round_nearest_even, &env->fp_status); |
2246 | 0 | set_floatx80_rounding_precision(floatx80_precision_x, &env->fp_status); |
2247 | |
|
2248 | 0 | if (arg0_exp == 0) { |
2249 | 0 | normalizeFloatx80Subnormal(arg0_sig, &arg0_exp, &arg0_sig); |
2250 | 0 | } |
2251 | 0 | if (arg1_exp == 0) { |
2252 | 0 | normalizeFloatx80Subnormal(arg1_sig, &arg1_exp, &arg1_sig); |
2253 | 0 | } |
2254 | 0 | int_exp = arg0_exp - 0x3fff; |
2255 | 0 | if (arg0_sig > 0xb504f333f9de6484ULL) { |
2256 | 0 | ++int_exp; |
2257 | 0 | } |
2258 | 0 | arg0_m1 = floatx80_sub(floatx80_scalbn(ST0, -int_exp, |
2259 | 0 | &env->fp_status), |
2260 | 0 | floatx80_one, &env->fp_status); |
2261 | 0 | if (floatx80_is_zero(arg0_m1)) { |
2262 | | /* Exact power of 2; multiply by ST1. */ |
2263 | 0 | set_float_rounding_mode(save_mode, &env->fp_status); |
2264 | 0 | ST1 = floatx80_mul(int32_to_floatx80(int_exp, &env->fp_status), |
2265 | 0 | ST1, &env->fp_status); |
2266 | 0 | } else { |
2267 | 0 | bool asign = extractFloatx80Sign(arg0_m1); |
2268 | 0 | int32_t aexp; |
2269 | 0 | uint64_t asig0, asig1, asig2; |
2270 | 0 | helper_fyl2x_common(env, arg0_m1, &aexp, &asig0, &asig1); |
2271 | 0 | if (int_exp != 0) { |
2272 | 0 | bool isign = (int_exp < 0); |
2273 | 0 | int32_t iexp; |
2274 | 0 | uint64_t isig; |
2275 | 0 | int shift; |
2276 | 0 | int_exp = isign ? -int_exp : int_exp; |
2277 | 0 | shift = clz32(int_exp) + 32; |
2278 | 0 | isig = int_exp; |
2279 | 0 | isig <<= shift; |
2280 | 0 | iexp = 0x403e - shift; |
2281 | 0 | shift128RightJamming(asig0, asig1, iexp - aexp, |
2282 | 0 | &asig0, &asig1); |
2283 | 0 | if (asign == isign) { |
2284 | 0 | add128(isig, 0, asig0, asig1, &asig0, &asig1); |
2285 | 0 | } else { |
2286 | 0 | sub128(isig, 0, asig0, asig1, &asig0, &asig1); |
2287 | 0 | } |
2288 | 0 | aexp = iexp; |
2289 | 0 | asign = isign; |
2290 | 0 | } |
2291 | | /* |
2292 | | * Multiply by the second argument to compute the required |
2293 | | * result. |
2294 | | */ |
2295 | 0 | if (arg1_exp == 0) { |
2296 | 0 | normalizeFloatx80Subnormal(arg1_sig, &arg1_exp, &arg1_sig); |
2297 | 0 | } |
2298 | 0 | mul128By64To192(asig0, asig1, arg1_sig, &asig0, &asig1, &asig2); |
2299 | 0 | aexp += arg1_exp - 0x3ffe; |
2300 | | /* This result is inexact. */ |
2301 | 0 | asig1 |= 1; |
2302 | 0 | set_float_rounding_mode(save_mode, &env->fp_status); |
2303 | 0 | ST1 = normalizeRoundAndPackFloatx80(floatx80_precision_x, |
2304 | 0 | asign ^ arg1_sign, aexp, |
2305 | 0 | asig0, asig1, &env->fp_status); |
2306 | 0 | } |
2307 | |
|
2308 | 0 | set_floatx80_rounding_precision(save_prec, &env->fp_status); |
2309 | 0 | } |
2310 | 0 | fpop(env); |
2311 | 0 | merge_exception_flags(env, old_flags); |
2312 | 0 | } |
2313 | | |
2314 | | void HELPER(fsqrt)(CPUX86State *env) |
2315 | 0 | { |
2316 | 0 | int old_flags = save_exception_flags(env); |
2317 | 0 | if (floatx80_is_neg(ST0)) { |
2318 | 0 | env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */ |
2319 | 0 | env->fpus |= 0x400; |
2320 | 0 | } |
2321 | 0 | ST0 = floatx80_sqrt(ST0, &env->fp_status); |
2322 | 0 | merge_exception_flags(env, old_flags); |
2323 | 0 | } |
2324 | | |
2325 | | void HELPER(fsincos)(CPUX86State *env) |
2326 | 0 | { |
2327 | 0 | double fptemp = floatx80_to_double(env, ST0); |
2328 | |
|
2329 | 0 | if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) { |
2330 | 0 | env->fpus |= 0x400; |
2331 | 0 | } else { |
2332 | 0 | ST0 = double_to_floatx80(env, sin(fptemp)); |
2333 | 0 | fpush(env); |
2334 | 0 | ST0 = double_to_floatx80(env, cos(fptemp)); |
2335 | 0 | env->fpus &= ~0x400; /* C2 <-- 0 */ |
2336 | | /* the above code is for |arg| < 2**63 only */ |
2337 | 0 | } |
2338 | 0 | } |
2339 | | |
2340 | | void HELPER(frndint)(CPUX86State *env) |
2341 | 0 | { |
2342 | 0 | int old_flags = save_exception_flags(env); |
2343 | 0 | ST0 = floatx80_round_to_int(ST0, &env->fp_status); |
2344 | 0 | merge_exception_flags(env, old_flags); |
2345 | 0 | } |
2346 | | |
2347 | | void HELPER(fscale)(CPUX86State *env) |
2348 | 0 | { |
2349 | 0 | int old_flags = save_exception_flags(env); |
2350 | 0 | if (floatx80_invalid_encoding(ST1, &env->fp_status) || |
2351 | 0 | floatx80_invalid_encoding(ST0, &env->fp_status)) { |
2352 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2353 | 0 | ST0 = floatx80_default_nan(&env->fp_status); |
2354 | 0 | } else if (floatx80_is_any_nan(ST1)) { |
2355 | 0 | if (floatx80_is_signaling_nan(ST0, &env->fp_status)) { |
2356 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2357 | 0 | } |
2358 | 0 | ST0 = ST1; |
2359 | 0 | if (floatx80_is_signaling_nan(ST0, &env->fp_status)) { |
2360 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2361 | 0 | ST0 = floatx80_silence_nan(ST0, &env->fp_status); |
2362 | 0 | } |
2363 | 0 | } else if (floatx80_is_infinity(ST1, &env->fp_status) && |
2364 | 0 | !floatx80_invalid_encoding(ST0, &env->fp_status) && |
2365 | 0 | !floatx80_is_any_nan(ST0)) { |
2366 | 0 | if (floatx80_is_neg(ST1)) { |
2367 | 0 | if (floatx80_is_infinity(ST0, &env->fp_status)) { |
2368 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2369 | 0 | ST0 = floatx80_default_nan(&env->fp_status); |
2370 | 0 | } else { |
2371 | 0 | ST0 = (floatx80_is_neg(ST0) ? |
2372 | 0 | floatx80_chs(floatx80_zero) : |
2373 | 0 | floatx80_zero); |
2374 | 0 | } |
2375 | 0 | } else { |
2376 | 0 | if (floatx80_is_zero(ST0)) { |
2377 | 0 | float_raise(float_flag_invalid, &env->fp_status); |
2378 | 0 | ST0 = floatx80_default_nan(&env->fp_status); |
2379 | 0 | } else { |
2380 | 0 | ST0 = floatx80_default_inf(floatx80_is_neg(ST0), |
2381 | 0 | &env->fp_status); |
2382 | 0 | } |
2383 | 0 | } |
2384 | 0 | } else { |
2385 | 0 | int n; |
2386 | 0 | FloatX80RoundPrec save = get_floatx80_rounding_precision(&env->fp_status); |
2387 | 0 | int save_flags = get_float_exception_flags(&env->fp_status); |
2388 | 0 | set_float_exception_flags(0, &env->fp_status); |
2389 | 0 | n = floatx80_to_int32_round_to_zero(ST1, &env->fp_status); |
2390 | 0 | set_float_exception_flags(save_flags, &env->fp_status); |
2391 | 0 | set_floatx80_rounding_precision(floatx80_precision_x, &env->fp_status); |
2392 | 0 | ST0 = floatx80_scalbn(ST0, n, &env->fp_status); |
2393 | 0 | set_floatx80_rounding_precision(save, &env->fp_status); |
2394 | 0 | } |
2395 | 0 | merge_exception_flags(env, old_flags); |
2396 | 0 | } |
2397 | | |
2398 | | void HELPER(fsin)(CPUX86State *env) |
2399 | 0 | { |
2400 | 0 | double fptemp = floatx80_to_double(env, ST0); |
2401 | |
|
2402 | 0 | if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) { |
2403 | 0 | env->fpus |= 0x400; |
2404 | 0 | } else { |
2405 | 0 | ST0 = double_to_floatx80(env, sin(fptemp)); |
2406 | 0 | env->fpus &= ~0x400; /* C2 <-- 0 */ |
2407 | | /* the above code is for |arg| < 2**53 only */ |
2408 | 0 | } |
2409 | 0 | } |
2410 | | |
2411 | | void HELPER(fcos)(CPUX86State *env) |
2412 | 0 | { |
2413 | 0 | double fptemp = floatx80_to_double(env, ST0); |
2414 | |
|
2415 | 0 | if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) { |
2416 | 0 | env->fpus |= 0x400; |
2417 | 0 | } else { |
2418 | 0 | ST0 = double_to_floatx80(env, cos(fptemp)); |
2419 | 0 | env->fpus &= ~0x400; /* C2 <-- 0 */ |
2420 | | /* the above code is for |arg| < 2**63 only */ |
2421 | 0 | } |
2422 | 0 | } |
2423 | | |
2424 | | void HELPER(fxam_ST0)(CPUX86State *env) |
2425 | 0 | { |
2426 | 0 | CPU_LDoubleU temp; |
2427 | 0 | int expdif; |
2428 | |
|
2429 | 0 | temp.d = ST0; |
2430 | |
|
2431 | 0 | env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */ |
2432 | 0 | if (SIGND(temp)) { |
2433 | 0 | env->fpus |= 0x200; /* C1 <-- 1 */ |
2434 | 0 | } |
2435 | |
|
2436 | 0 | if (env->fptags[env->fpstt]) { |
2437 | 0 | env->fpus |= 0x4100; /* Empty */ |
2438 | 0 | return; |
2439 | 0 | } |
2440 | | |
2441 | 0 | expdif = EXPD(temp); |
2442 | 0 | if (expdif == MAXEXPD) { |
2443 | 0 | if (MANTD(temp) == 0x8000000000000000ULL) { |
2444 | 0 | env->fpus |= 0x500; /* Infinity */ |
2445 | 0 | } else if (MANTD(temp) & 0x8000000000000000ULL) { |
2446 | 0 | env->fpus |= 0x100; /* NaN */ |
2447 | 0 | } |
2448 | 0 | } else if (expdif == 0) { |
2449 | 0 | if (MANTD(temp) == 0) { |
2450 | 0 | env->fpus |= 0x4000; /* Zero */ |
2451 | 0 | } else { |
2452 | 0 | env->fpus |= 0x4400; /* Denormal */ |
2453 | 0 | } |
2454 | 0 | } else if (MANTD(temp) & 0x8000000000000000ULL) { |
2455 | 0 | env->fpus |= 0x400; |
2456 | 0 | } |
2457 | 0 | } |
2458 | | |
2459 | | static void do_fstenv(X86Access *ac, target_ulong ptr, int data32) |
2460 | 0 | { |
2461 | 0 | CPUX86State *env = ac->env; |
2462 | 0 | int fpus, fptag, exp, i; |
2463 | 0 | uint64_t mant; |
2464 | 0 | CPU_LDoubleU tmp; |
2465 | |
|
2466 | 0 | fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11; |
2467 | 0 | fptag = 0; |
2468 | 0 | for (i = 7; i >= 0; i--) { |
2469 | 0 | fptag <<= 2; |
2470 | 0 | if (env->fptags[i]) { |
2471 | 0 | fptag |= 3; |
2472 | 0 | } else { |
2473 | 0 | tmp.d = env->fpregs[i].d; |
2474 | 0 | exp = EXPD(tmp); |
2475 | 0 | mant = MANTD(tmp); |
2476 | 0 | if (exp == 0 && mant == 0) { |
2477 | | /* zero */ |
2478 | 0 | fptag |= 1; |
2479 | 0 | } else if (exp == 0 || exp == MAXEXPD |
2480 | 0 | || (mant & (1LL << 63)) == 0) { |
2481 | | /* NaNs, infinity, denormal */ |
2482 | 0 | fptag |= 2; |
2483 | 0 | } |
2484 | 0 | } |
2485 | 0 | } |
2486 | 0 | if (data32) { |
2487 | | /* 32 bit */ |
2488 | 0 | access_stl(ac, ptr, env->fpuc); |
2489 | 0 | access_stl(ac, ptr + 4, fpus); |
2490 | 0 | access_stl(ac, ptr + 8, fptag); |
2491 | 0 | access_stl(ac, ptr + 12, env->fpip); /* fpip */ |
2492 | 0 | access_stl(ac, ptr + 16, env->fpcs); /* fpcs */ |
2493 | 0 | access_stl(ac, ptr + 20, env->fpdp); /* fpoo */ |
2494 | 0 | access_stl(ac, ptr + 24, env->fpds); /* fpos */ |
2495 | 0 | } else { |
2496 | | /* 16 bit */ |
2497 | 0 | access_stw(ac, ptr, env->fpuc); |
2498 | 0 | access_stw(ac, ptr + 2, fpus); |
2499 | 0 | access_stw(ac, ptr + 4, fptag); |
2500 | 0 | access_stw(ac, ptr + 6, env->fpip); |
2501 | 0 | access_stw(ac, ptr + 8, env->fpcs); |
2502 | 0 | access_stw(ac, ptr + 10, env->fpdp); |
2503 | 0 | access_stw(ac, ptr + 12, env->fpds); |
2504 | 0 | } |
2505 | 0 | } |
2506 | | |
2507 | | void HELPER(fstenv)(CPUX86State *env, target_ulong ptr, int data32) |
2508 | 0 | { |
2509 | 0 | X86Access ac; |
2510 | |
|
2511 | 0 | access_prepare(&ac, env, ptr, 14 << data32, MMU_DATA_STORE, GETPC()); |
2512 | 0 | do_fstenv(&ac, ptr, data32); |
2513 | 0 | } |
2514 | | |
2515 | | static void cpu_set_fpus(CPUX86State *env, uint16_t fpus) |
2516 | 0 | { |
2517 | 0 | env->fpstt = (fpus >> 11) & 7; |
2518 | 0 | env->fpus = fpus & ~0x3800 & ~FPUS_B; |
2519 | 0 | env->fpus |= env->fpus & FPUS_SE ? FPUS_B : 0; |
2520 | 0 | #if !defined(CONFIG_USER_ONLY) |
2521 | 0 | if (!(env->fpus & FPUS_SE)) { |
2522 | | /* |
2523 | | * Here the processor deasserts FERR#; in response, the chipset deasserts |
2524 | | * IGNNE#. |
2525 | | */ |
2526 | 0 | cpu_clear_ignne(); |
2527 | 0 | } |
2528 | 0 | #endif |
2529 | 0 | } |
2530 | | |
2531 | | static void do_fldenv(X86Access *ac, target_ulong ptr, int data32) |
2532 | 0 | { |
2533 | 0 | int i, fpus, fptag; |
2534 | 0 | CPUX86State *env = ac->env; |
2535 | |
|
2536 | 0 | cpu_set_fpuc(env, access_ldw(ac, ptr)); |
2537 | 0 | fpus = access_ldw(ac, ptr + (2 << data32)); |
2538 | 0 | fptag = access_ldw(ac, ptr + (4 << data32)); |
2539 | |
|
2540 | 0 | cpu_set_fpus(env, fpus); |
2541 | 0 | for (i = 0; i < 8; i++) { |
2542 | 0 | env->fptags[i] = ((fptag & 3) == 3); |
2543 | 0 | fptag >>= 2; |
2544 | 0 | } |
2545 | 0 | } |
2546 | | |
2547 | | void HELPER(fldenv)(CPUX86State *env, target_ulong ptr, int data32) |
2548 | 0 | { |
2549 | 0 | X86Access ac; |
2550 | |
|
2551 | 0 | access_prepare(&ac, env, ptr, 14 << data32, MMU_DATA_STORE, GETPC()); |
2552 | 0 | do_fldenv(&ac, ptr, data32); |
2553 | 0 | } |
2554 | | |
2555 | | /* |
2556 | | * Store the environment and the register stack, as FSAVE does, but |
2557 | | * without the FNINIT that FSAVE performs afterward. |
2558 | | */ |
2559 | | static void do_fsave(X86Access *ac, target_ulong ptr, int data32) |
2560 | 0 | { |
2561 | 0 | CPUX86State *env = ac->env; |
2562 | |
|
2563 | 0 | do_fstenv(ac, ptr, data32); |
2564 | 0 | ptr += 14 << data32; |
2565 | |
|
2566 | 0 | for (int i = 0; i < 8; i++) { |
2567 | 0 | floatx80 tmp = ST(i); |
2568 | 0 | do_fstt(ac, ptr, tmp); |
2569 | 0 | ptr += 10; |
2570 | 0 | } |
2571 | 0 | } |
2572 | | |
2573 | | void HELPER(fsave)(CPUX86State *env, target_ulong ptr, int data32) |
2574 | 0 | { |
2575 | 0 | int size = (14 << data32) + 80; |
2576 | 0 | X86Access ac; |
2577 | |
|
2578 | 0 | access_prepare(&ac, env, ptr, size, MMU_DATA_STORE, GETPC()); |
2579 | 0 | do_fsave(&ac, ptr, data32); |
2580 | 0 | do_fninit(env); |
2581 | 0 | } |
2582 | | |
2583 | | static void do_frstor(X86Access *ac, target_ulong ptr, int data32) |
2584 | 0 | { |
2585 | 0 | CPUX86State *env = ac->env; |
2586 | |
|
2587 | 0 | do_fldenv(ac, ptr, data32); |
2588 | 0 | ptr += 14 << data32; |
2589 | |
|
2590 | 0 | for (int i = 0; i < 8; i++) { |
2591 | 0 | floatx80 tmp = do_fldt(ac, ptr); |
2592 | 0 | ST(i) = tmp; |
2593 | 0 | ptr += 10; |
2594 | 0 | } |
2595 | 0 | } |
2596 | | |
2597 | | void HELPER(frstor)(CPUX86State *env, target_ulong ptr, int data32) |
2598 | 0 | { |
2599 | 0 | int size = (14 << data32) + 80; |
2600 | 0 | X86Access ac; |
2601 | |
|
2602 | 0 | access_prepare(&ac, env, ptr, size, MMU_DATA_LOAD, GETPC()); |
2603 | 0 | do_frstor(&ac, ptr, data32); |
2604 | 0 | } |
2605 | | |
2606 | 0 | #define XO(X) offsetof(X86XSaveArea, X) |
2607 | | |
2608 | | static void do_xsave_fpu(X86Access *ac, target_ulong ptr) |
2609 | 0 | { |
2610 | 0 | CPUX86State *env = ac->env; |
2611 | 0 | int fpus, fptag, i; |
2612 | 0 | target_ulong addr; |
2613 | |
|
2614 | 0 | fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11; |
2615 | 0 | fptag = 0; |
2616 | 0 | for (i = 0; i < 8; i++) { |
2617 | 0 | fptag |= (env->fptags[i] << i); |
2618 | 0 | } |
2619 | |
|
2620 | 0 | access_stw(ac, ptr + XO(legacy.fcw), env->fpuc); |
2621 | 0 | access_stw(ac, ptr + XO(legacy.fsw), fpus); |
2622 | 0 | access_stw(ac, ptr + XO(legacy.ftw), fptag ^ 0xff); |
2623 | 0 | access_stw(ac, ptr + XO(legacy.fpop), 0); |
2624 | | |
2625 | | /* In 32-bit mode this is eip, sel, dp, sel. |
2626 | | In 64-bit mode this is rip, rdp. |
2627 | | But in either case we don't write actual data, just zeros. */ |
2628 | 0 | access_stq(ac, ptr + XO(legacy.fpip), 0); /* eip+sel; rip */ |
2629 | 0 | access_stq(ac, ptr + XO(legacy.fpdp), 0); /* edp+sel; rdp */ |
2630 | |
|
2631 | 0 | addr = ptr + XO(legacy.fpregs); |
2632 | |
|
2633 | 0 | for (i = 0; i < 8; i++) { |
2634 | 0 | floatx80 tmp = ST(i); |
2635 | 0 | do_fstt(ac, addr, tmp); |
2636 | 0 | access_stw(ac, addr + 10, 0); |
2637 | 0 | access_stl(ac, addr + 12, 0); |
2638 | 0 | addr += 16; |
2639 | 0 | } |
2640 | 0 | } |
2641 | | |
2642 | | static void do_xsave_mxcsr(X86Access *ac, target_ulong ptr) |
2643 | 0 | { |
2644 | 0 | CPUX86State *env = ac->env; |
2645 | |
|
2646 | 0 | update_mxcsr_from_sse_status(env); |
2647 | 0 | access_stl(ac, ptr + XO(legacy.mxcsr), env->mxcsr); |
2648 | 0 | access_stl(ac, ptr + XO(legacy.mxcsr_mask), 0x0000ffff); |
2649 | 0 | } |
2650 | | |
2651 | | static void do_xsave_sse(X86Access *ac, target_ulong ptr) |
2652 | 0 | { |
2653 | 0 | CPUX86State *env = ac->env; |
2654 | 0 | int i, nb_xmm_regs; |
2655 | 0 | target_ulong addr; |
2656 | |
|
2657 | 0 | if (env->hflags & HF_CS64_MASK) { |
2658 | 0 | nb_xmm_regs = 16; |
2659 | 0 | } else { |
2660 | 0 | nb_xmm_regs = 8; |
2661 | 0 | } |
2662 | |
|
2663 | 0 | addr = ptr + XO(legacy.xmm_regs); |
2664 | 0 | for (i = 0; i < nb_xmm_regs; i++) { |
2665 | 0 | access_stq(ac, addr, env->xmm_regs[i].ZMM_Q(0)); |
2666 | 0 | access_stq(ac, addr + 8, env->xmm_regs[i].ZMM_Q(1)); |
2667 | 0 | addr += 16; |
2668 | 0 | } |
2669 | 0 | } |
2670 | | |
2671 | | static void do_xsave_ymmh(X86Access *ac, target_ulong ptr) |
2672 | 0 | { |
2673 | 0 | CPUX86State *env = ac->env; |
2674 | 0 | int i, nb_xmm_regs; |
2675 | |
|
2676 | 0 | if (env->hflags & HF_CS64_MASK) { |
2677 | 0 | nb_xmm_regs = 16; |
2678 | 0 | } else { |
2679 | 0 | nb_xmm_regs = 8; |
2680 | 0 | } |
2681 | |
|
2682 | 0 | for (i = 0; i < nb_xmm_regs; i++, ptr += 16) { |
2683 | 0 | access_stq(ac, ptr, env->xmm_regs[i].ZMM_Q(2)); |
2684 | 0 | access_stq(ac, ptr + 8, env->xmm_regs[i].ZMM_Q(3)); |
2685 | 0 | } |
2686 | 0 | } |
2687 | | |
2688 | | static void do_xsave_bndregs(X86Access *ac, target_ulong ptr) |
2689 | 0 | { |
2690 | 0 | CPUX86State *env = ac->env; |
2691 | 0 | target_ulong addr = ptr + offsetof(XSaveBNDREG, bnd_regs); |
2692 | 0 | int i; |
2693 | |
|
2694 | 0 | for (i = 0; i < 4; i++, addr += 16) { |
2695 | 0 | access_stq(ac, addr, env->bnd_regs[i].lb); |
2696 | 0 | access_stq(ac, addr + 8, env->bnd_regs[i].ub); |
2697 | 0 | } |
2698 | 0 | } |
2699 | | |
2700 | | static void do_xsave_bndcsr(X86Access *ac, target_ulong ptr) |
2701 | 0 | { |
2702 | 0 | CPUX86State *env = ac->env; |
2703 | |
|
2704 | 0 | access_stq(ac, ptr + offsetof(XSaveBNDCSR, bndcsr.cfgu), |
2705 | 0 | env->bndcs_regs.cfgu); |
2706 | 0 | access_stq(ac, ptr + offsetof(XSaveBNDCSR, bndcsr.sts), |
2707 | 0 | env->bndcs_regs.sts); |
2708 | 0 | } |
2709 | | |
2710 | | static void do_xsave_pkru(X86Access *ac, target_ulong ptr) |
2711 | 0 | { |
2712 | 0 | access_stq(ac, ptr, ac->env->pkru); |
2713 | 0 | } |
2714 | | |
2715 | | #ifdef TARGET_X86_64 |
2716 | | static void do_xsave_apx(X86Access *ac, target_ulong ptr) |
2717 | | { |
2718 | | CPUX86State *env = ac->env; |
2719 | | int i; |
2720 | | |
2721 | | for (i = 16; i < 32; i++) { |
2722 | | access_stq(ac, ptr, env->regs[i]); |
2723 | | ptr += 8; |
2724 | | } |
2725 | | } |
2726 | | #endif |
2727 | | |
2728 | | static void do_fxsave(X86Access *ac, target_ulong ptr) |
2729 | 0 | { |
2730 | 0 | CPUX86State *env = ac->env; |
2731 | |
|
2732 | 0 | do_xsave_fpu(ac, ptr); |
2733 | 0 | if (env->cr[4] & CR4_OSFXSR_MASK) { |
2734 | 0 | do_xsave_mxcsr(ac, ptr); |
2735 | | /* Fast FXSAVE leaves out the XMM registers */ |
2736 | 0 | if (!(env->efer & MSR_EFER_FFXSR) |
2737 | 0 | || (env->hflags & HF_CPL_MASK) |
2738 | 0 | || !(env->hflags & HF_LMA_MASK)) { |
2739 | 0 | do_xsave_sse(ac, ptr); |
2740 | 0 | } |
2741 | 0 | } |
2742 | 0 | } |
2743 | | |
2744 | | void HELPER(fxsave)(CPUX86State *env, target_ulong ptr) |
2745 | 0 | { |
2746 | 0 | uintptr_t ra = GETPC(); |
2747 | 0 | X86Access ac; |
2748 | | |
2749 | | /* The operand must be 16 byte aligned */ |
2750 | 0 | if (ptr & 0xf) { |
2751 | 0 | raise_exception_ra(env, EXCP0D_GPF, ra); |
2752 | 0 | } |
2753 | | |
2754 | 0 | access_prepare(&ac, env, ptr, sizeof(X86LegacyXSaveArea), |
2755 | 0 | MMU_DATA_STORE, ra); |
2756 | 0 | do_fxsave(&ac, ptr); |
2757 | 0 | } |
2758 | | |
2759 | | static uint64_t get_xinuse(CPUX86State *env) |
2760 | 0 | { |
2761 | 0 | uint64_t inuse = -1; |
2762 | | |
2763 | | /* For the most part, we don't track XINUSE. We could calculate it |
2764 | | here for all components, but it's probably less work to simply |
2765 | | indicate in use. That said, the state of BNDREGS is important |
2766 | | enough to track in HFLAGS, so we might as well use that here. */ |
2767 | 0 | if ((env->hflags & HF_MPX_IU_MASK) == 0) { |
2768 | 0 | inuse &= ~XSTATE_BNDREGS_MASK; |
2769 | 0 | } |
2770 | 0 | return inuse; |
2771 | 0 | } |
2772 | | |
2773 | | static void do_xsave_access(X86Access *ac, target_ulong ptr, uint64_t rfbm, |
2774 | | uint64_t inuse, uint64_t opt) |
2775 | 0 | { |
2776 | 0 | uint64_t old_bv, new_bv; |
2777 | |
|
2778 | 0 | if (opt & XSTATE_FP_MASK) { |
2779 | 0 | do_xsave_fpu(ac, ptr); |
2780 | 0 | } |
2781 | 0 | if (rfbm & XSTATE_SSE_MASK) { |
2782 | | /* Note that saving MXCSR is not suppressed by XSAVEOPT. */ |
2783 | 0 | do_xsave_mxcsr(ac, ptr); |
2784 | 0 | } |
2785 | 0 | if (opt & XSTATE_SSE_MASK) { |
2786 | 0 | do_xsave_sse(ac, ptr); |
2787 | 0 | } |
2788 | 0 | if (opt & XSTATE_YMM_MASK) { |
2789 | 0 | do_xsave_ymmh(ac, ptr + XO(avx_state)); |
2790 | 0 | } |
2791 | 0 | if (opt & XSTATE_BNDREGS_MASK) { |
2792 | 0 | do_xsave_bndregs(ac, ptr + XO(mpx_state.bndreg)); |
2793 | 0 | } |
2794 | 0 | if (opt & XSTATE_BNDCSR_MASK) { |
2795 | 0 | do_xsave_bndcsr(ac, ptr + XO(mpx_state.bndcsr)); |
2796 | 0 | } |
2797 | 0 | if (opt & XSTATE_PKRU_MASK) { |
2798 | 0 | do_xsave_pkru(ac, ptr + XO(pkru_state)); |
2799 | 0 | } |
2800 | | #ifdef TARGET_X86_64 |
2801 | | if (opt & XSTATE_APX_MASK) { |
2802 | | do_xsave_apx(ac, ptr + XO(apx_state)); |
2803 | | } |
2804 | | #endif |
2805 | | |
2806 | | /* Update the XSTATE_BV field. */ |
2807 | 0 | old_bv = access_ldq(ac, ptr + XO(header.xstate_bv)); |
2808 | 0 | new_bv = (old_bv & ~rfbm) | (inuse & rfbm); |
2809 | 0 | access_stq(ac, ptr + XO(header.xstate_bv), new_bv); |
2810 | 0 | } |
2811 | | |
2812 | | static void do_xsave_chk(CPUX86State *env, target_ulong ptr, uintptr_t ra) |
2813 | 0 | { |
2814 | | /* The OS must have enabled XSAVE. */ |
2815 | 0 | if (!(env->cr[4] & CR4_OSXSAVE_MASK)) { |
2816 | 0 | raise_exception_ra(env, EXCP06_ILLOP, ra); |
2817 | 0 | } |
2818 | | |
2819 | | /* The operand must be 64 byte aligned. */ |
2820 | 0 | if (ptr & 63) { |
2821 | 0 | raise_exception_ra(env, EXCP0D_GPF, ra); |
2822 | 0 | } |
2823 | 0 | } |
2824 | | |
2825 | | static void do_xsave(CPUX86State *env, target_ulong ptr, uint64_t rfbm, |
2826 | | uint64_t inuse, uint64_t opt, uintptr_t ra) |
2827 | 0 | { |
2828 | 0 | X86Access ac; |
2829 | 0 | unsigned size; |
2830 | |
|
2831 | 0 | do_xsave_chk(env, ptr, ra); |
2832 | | |
2833 | | /* Never save anything not enabled by XCR0. */ |
2834 | 0 | rfbm &= env->xcr0; |
2835 | 0 | opt &= rfbm; |
2836 | 0 | size = xsave_area_size(opt, false); |
2837 | |
|
2838 | 0 | access_prepare(&ac, env, ptr, size, MMU_DATA_STORE, ra); |
2839 | 0 | do_xsave_access(&ac, ptr, rfbm, inuse, opt); |
2840 | 0 | } |
2841 | | |
2842 | | void HELPER(xsave)(CPUX86State *env, target_ulong ptr, uint64_t rfbm) |
2843 | 0 | { |
2844 | 0 | do_xsave(env, ptr, rfbm, get_xinuse(env), rfbm, GETPC()); |
2845 | 0 | } |
2846 | | |
2847 | | void HELPER(xsaveopt)(CPUX86State *env, target_ulong ptr, uint64_t rfbm) |
2848 | 0 | { |
2849 | 0 | uint64_t inuse = get_xinuse(env); |
2850 | 0 | do_xsave(env, ptr, rfbm, inuse, inuse, GETPC()); |
2851 | 0 | } |
2852 | | |
2853 | | static void do_xrstor_fpu(X86Access *ac, target_ulong ptr) |
2854 | 0 | { |
2855 | 0 | CPUX86State *env = ac->env; |
2856 | 0 | int i, fpuc, fpus, fptag; |
2857 | 0 | target_ulong addr; |
2858 | |
|
2859 | 0 | fpuc = access_ldw(ac, ptr + XO(legacy.fcw)); |
2860 | 0 | fpus = access_ldw(ac, ptr + XO(legacy.fsw)); |
2861 | 0 | fptag = access_ldw(ac, ptr + XO(legacy.ftw)); |
2862 | 0 | cpu_set_fpuc(env, fpuc); |
2863 | 0 | cpu_set_fpus(env, fpus); |
2864 | |
|
2865 | 0 | fptag ^= 0xff; |
2866 | 0 | for (i = 0; i < 8; i++) { |
2867 | 0 | env->fptags[i] = ((fptag >> i) & 1); |
2868 | 0 | } |
2869 | |
|
2870 | 0 | addr = ptr + XO(legacy.fpregs); |
2871 | |
|
2872 | 0 | for (i = 0; i < 8; i++) { |
2873 | 0 | floatx80 tmp = do_fldt(ac, addr); |
2874 | 0 | ST(i) = tmp; |
2875 | 0 | addr += 16; |
2876 | 0 | } |
2877 | 0 | } |
2878 | | |
2879 | | static void do_xrstor_mxcsr(X86Access *ac, target_ulong ptr) |
2880 | 0 | { |
2881 | 0 | CPUX86State *env = ac->env; |
2882 | 0 | cpu_set_mxcsr(env, access_ldl(ac, ptr + XO(legacy.mxcsr))); |
2883 | 0 | } |
2884 | | |
2885 | | static void do_xrstor_sse(X86Access *ac, target_ulong ptr) |
2886 | 0 | { |
2887 | 0 | CPUX86State *env = ac->env; |
2888 | 0 | int i, nb_xmm_regs; |
2889 | 0 | target_ulong addr; |
2890 | |
|
2891 | 0 | if (env->hflags & HF_CS64_MASK) { |
2892 | 0 | nb_xmm_regs = 16; |
2893 | 0 | } else { |
2894 | 0 | nb_xmm_regs = 8; |
2895 | 0 | } |
2896 | |
|
2897 | 0 | addr = ptr + XO(legacy.xmm_regs); |
2898 | 0 | for (i = 0; i < nb_xmm_regs; i++) { |
2899 | 0 | env->xmm_regs[i].ZMM_Q(0) = access_ldq(ac, addr); |
2900 | 0 | env->xmm_regs[i].ZMM_Q(1) = access_ldq(ac, addr + 8); |
2901 | 0 | addr += 16; |
2902 | 0 | } |
2903 | 0 | } |
2904 | | |
2905 | | static void do_clear_sse(CPUX86State *env) |
2906 | 0 | { |
2907 | 0 | int i, nb_xmm_regs; |
2908 | |
|
2909 | 0 | if (env->hflags & HF_CS64_MASK) { |
2910 | 0 | nb_xmm_regs = 16; |
2911 | 0 | } else { |
2912 | 0 | nb_xmm_regs = 8; |
2913 | 0 | } |
2914 | |
|
2915 | 0 | for (i = 0; i < nb_xmm_regs; i++) { |
2916 | 0 | env->xmm_regs[i].ZMM_Q(0) = 0; |
2917 | 0 | env->xmm_regs[i].ZMM_Q(1) = 0; |
2918 | 0 | } |
2919 | 0 | } |
2920 | | |
2921 | | static void do_xrstor_ymmh(X86Access *ac, target_ulong ptr) |
2922 | 0 | { |
2923 | 0 | CPUX86State *env = ac->env; |
2924 | 0 | int i, nb_xmm_regs; |
2925 | |
|
2926 | 0 | if (env->hflags & HF_CS64_MASK) { |
2927 | 0 | nb_xmm_regs = 16; |
2928 | 0 | } else { |
2929 | 0 | nb_xmm_regs = 8; |
2930 | 0 | } |
2931 | |
|
2932 | 0 | for (i = 0; i < nb_xmm_regs; i++, ptr += 16) { |
2933 | 0 | env->xmm_regs[i].ZMM_Q(2) = access_ldq(ac, ptr); |
2934 | 0 | env->xmm_regs[i].ZMM_Q(3) = access_ldq(ac, ptr + 8); |
2935 | 0 | } |
2936 | 0 | } |
2937 | | |
2938 | | static void do_clear_ymmh(CPUX86State *env) |
2939 | 0 | { |
2940 | 0 | int i, nb_xmm_regs; |
2941 | |
|
2942 | 0 | if (env->hflags & HF_CS64_MASK) { |
2943 | 0 | nb_xmm_regs = 16; |
2944 | 0 | } else { |
2945 | 0 | nb_xmm_regs = 8; |
2946 | 0 | } |
2947 | |
|
2948 | 0 | for (i = 0; i < nb_xmm_regs; i++) { |
2949 | 0 | env->xmm_regs[i].ZMM_Q(2) = 0; |
2950 | 0 | env->xmm_regs[i].ZMM_Q(3) = 0; |
2951 | 0 | } |
2952 | 0 | } |
2953 | | |
2954 | | static void do_xrstor_bndregs(X86Access *ac, target_ulong ptr) |
2955 | 0 | { |
2956 | 0 | CPUX86State *env = ac->env; |
2957 | 0 | target_ulong addr = ptr + offsetof(XSaveBNDREG, bnd_regs); |
2958 | 0 | int i; |
2959 | |
|
2960 | 0 | for (i = 0; i < 4; i++, addr += 16) { |
2961 | 0 | env->bnd_regs[i].lb = access_ldq(ac, addr); |
2962 | 0 | env->bnd_regs[i].ub = access_ldq(ac, addr + 8); |
2963 | 0 | } |
2964 | 0 | } |
2965 | | |
2966 | | static void do_xrstor_bndcsr(X86Access *ac, target_ulong ptr) |
2967 | 0 | { |
2968 | 0 | CPUX86State *env = ac->env; |
2969 | | |
2970 | | /* FIXME: Extend highest implemented bit of linear address. */ |
2971 | 0 | env->bndcs_regs.cfgu |
2972 | 0 | = access_ldq(ac, ptr + offsetof(XSaveBNDCSR, bndcsr.cfgu)); |
2973 | 0 | env->bndcs_regs.sts |
2974 | 0 | = access_ldq(ac, ptr + offsetof(XSaveBNDCSR, bndcsr.sts)); |
2975 | 0 | } |
2976 | | |
2977 | | static void do_xrstor_pkru(X86Access *ac, target_ulong ptr) |
2978 | 0 | { |
2979 | 0 | ac->env->pkru = access_ldq(ac, ptr); |
2980 | 0 | } |
2981 | | |
2982 | | static void do_fxrstor(X86Access *ac, target_ulong ptr) |
2983 | 0 | { |
2984 | 0 | CPUX86State *env = ac->env; |
2985 | |
|
2986 | 0 | do_xrstor_fpu(ac, ptr); |
2987 | 0 | if (env->cr[4] & CR4_OSFXSR_MASK) { |
2988 | 0 | do_xrstor_mxcsr(ac, ptr); |
2989 | | /* Fast FXRSTOR leaves out the XMM registers */ |
2990 | 0 | if (!(env->efer & MSR_EFER_FFXSR) |
2991 | 0 | || (env->hflags & HF_CPL_MASK) |
2992 | 0 | || !(env->hflags & HF_LMA_MASK)) { |
2993 | 0 | do_xrstor_sse(ac, ptr); |
2994 | 0 | } |
2995 | 0 | } |
2996 | 0 | } |
2997 | | |
2998 | | #ifdef TARGET_X86_64 |
2999 | | static void do_xrstor_apx(X86Access *ac, target_ulong ptr) |
3000 | | { |
3001 | | CPUX86State *env = ac->env; |
3002 | | int i; |
3003 | | |
3004 | | for (i = 16; i < 32; i++) { |
3005 | | env->regs[i] = access_ldq(ac, ptr); |
3006 | | ptr += 8; |
3007 | | } |
3008 | | } |
3009 | | |
3010 | | static void do_clear_apx(X86Access *ac) |
3011 | | { |
3012 | | CPUX86State *env = ac->env; |
3013 | | int i; |
3014 | | |
3015 | | for (i = 16; i < 32; i++) { |
3016 | | env->regs[i] = 0; |
3017 | | } |
3018 | | } |
3019 | | #endif |
3020 | | |
3021 | | void HELPER(fxrstor)(CPUX86State *env, target_ulong ptr) |
3022 | 0 | { |
3023 | 0 | uintptr_t ra = GETPC(); |
3024 | 0 | X86Access ac; |
3025 | | |
3026 | | /* The operand must be 16 byte aligned */ |
3027 | 0 | if (ptr & 0xf) { |
3028 | 0 | raise_exception_ra(env, EXCP0D_GPF, ra); |
3029 | 0 | } |
3030 | | |
3031 | 0 | access_prepare(&ac, env, ptr, sizeof(X86LegacyXSaveArea), |
3032 | 0 | MMU_DATA_LOAD, ra); |
3033 | 0 | do_fxrstor(&ac, ptr); |
3034 | 0 | } |
3035 | | |
3036 | | static bool valid_xrstor_header(X86Access *ac, uint64_t *pxsbv, |
3037 | | target_ulong ptr) |
3038 | 0 | { |
3039 | 0 | uint64_t xstate_bv, xcomp_bv, reserve0; |
3040 | |
|
3041 | 0 | xstate_bv = access_ldq(ac, ptr + XO(header.xstate_bv)); |
3042 | 0 | xcomp_bv = access_ldq(ac, ptr + XO(header.xcomp_bv)); |
3043 | 0 | reserve0 = access_ldq(ac, ptr + XO(header.reserve0)); |
3044 | 0 | *pxsbv = xstate_bv; |
3045 | | |
3046 | | /* |
3047 | | * XCOMP_BV bit 63 indicates compact form, which we do not support, |
3048 | | * and thus must raise #GP. That leaves us in standard form. |
3049 | | * In standard form, bytes 23:8 must be zero -- which is both |
3050 | | * XCOMP_BV and the following 64-bit field. |
3051 | | */ |
3052 | 0 | if (xcomp_bv || reserve0) { |
3053 | 0 | return false; |
3054 | 0 | } |
3055 | | |
3056 | | /* The XSTATE_BV field must not set bits not present in XCR0. */ |
3057 | 0 | return (xstate_bv & ~ac->env->xcr0) == 0; |
3058 | 0 | } |
3059 | | |
3060 | | static void do_xrstor(X86Access *ac, target_ulong ptr, |
3061 | | uint64_t rfbm, uint64_t xstate_bv) |
3062 | 0 | { |
3063 | 0 | CPUX86State *env = ac->env; |
3064 | |
|
3065 | 0 | if (rfbm & XSTATE_FP_MASK) { |
3066 | 0 | if (xstate_bv & XSTATE_FP_MASK) { |
3067 | 0 | do_xrstor_fpu(ac, ptr); |
3068 | 0 | } else { |
3069 | 0 | do_fninit(env); |
3070 | 0 | memset(env->fpregs, 0, sizeof(env->fpregs)); |
3071 | 0 | } |
3072 | 0 | } |
3073 | 0 | if (rfbm & XSTATE_SSE_MASK) { |
3074 | | /* Note that the standard form of XRSTOR loads MXCSR from memory |
3075 | | whether or not the XSTATE_BV bit is set. */ |
3076 | 0 | do_xrstor_mxcsr(ac, ptr); |
3077 | 0 | if (xstate_bv & XSTATE_SSE_MASK) { |
3078 | 0 | do_xrstor_sse(ac, ptr); |
3079 | 0 | } else { |
3080 | 0 | do_clear_sse(env); |
3081 | 0 | } |
3082 | 0 | } |
3083 | 0 | if (rfbm & XSTATE_YMM_MASK) { |
3084 | 0 | if (xstate_bv & XSTATE_YMM_MASK) { |
3085 | 0 | do_xrstor_ymmh(ac, ptr + XO(avx_state)); |
3086 | 0 | } else { |
3087 | 0 | do_clear_ymmh(env); |
3088 | 0 | } |
3089 | 0 | } |
3090 | 0 | if (rfbm & XSTATE_BNDREGS_MASK) { |
3091 | 0 | if (xstate_bv & XSTATE_BNDREGS_MASK) { |
3092 | 0 | do_xrstor_bndregs(ac, ptr + XO(mpx_state.bndreg)); |
3093 | 0 | env->hflags |= HF_MPX_IU_MASK; |
3094 | 0 | } else { |
3095 | 0 | memset(env->bnd_regs, 0, sizeof(env->bnd_regs)); |
3096 | 0 | env->hflags &= ~HF_MPX_IU_MASK; |
3097 | 0 | } |
3098 | 0 | } |
3099 | 0 | if (rfbm & XSTATE_BNDCSR_MASK) { |
3100 | 0 | if (xstate_bv & XSTATE_BNDCSR_MASK) { |
3101 | 0 | do_xrstor_bndcsr(ac, ptr + XO(mpx_state.bndcsr)); |
3102 | 0 | } else { |
3103 | 0 | memset(&env->bndcs_regs, 0, sizeof(env->bndcs_regs)); |
3104 | 0 | } |
3105 | 0 | cpu_sync_bndcs_hflags(env); |
3106 | 0 | } |
3107 | 0 | if (rfbm & XSTATE_PKRU_MASK) { |
3108 | 0 | uint64_t old_pkru = env->pkru; |
3109 | 0 | if (xstate_bv & XSTATE_PKRU_MASK) { |
3110 | 0 | do_xrstor_pkru(ac, ptr + XO(pkru_state)); |
3111 | 0 | } else { |
3112 | 0 | env->pkru = 0; |
3113 | 0 | } |
3114 | 0 | if (env->pkru != old_pkru) { |
3115 | 0 | CPUState *cs = env_cpu(env); |
3116 | 0 | tlb_flush(cs); |
3117 | 0 | } |
3118 | 0 | } |
3119 | | #ifdef TARGET_X86_64 |
3120 | | if (rfbm & XSTATE_APX_MASK) { |
3121 | | if (xstate_bv & XSTATE_APX_MASK) { |
3122 | | do_xrstor_apx(ac, ptr + XO(apx_state)); |
3123 | | } else { |
3124 | | do_clear_apx(ac); |
3125 | | } |
3126 | | } |
3127 | | #endif |
3128 | 0 | } |
3129 | | |
3130 | | #undef XO |
3131 | | |
3132 | | void HELPER(xrstor)(CPUX86State *env, target_ulong ptr, uint64_t rfbm) |
3133 | 0 | { |
3134 | 0 | uintptr_t ra = GETPC(); |
3135 | 0 | X86Access ac; |
3136 | 0 | uint64_t xstate_bv; |
3137 | 0 | unsigned size, size_ext; |
3138 | |
|
3139 | 0 | do_xsave_chk(env, ptr, ra); |
3140 | | |
3141 | | /* Begin with just the minimum size to validate the header. */ |
3142 | 0 | size = sizeof(X86LegacyXSaveArea) + sizeof(X86XSaveHeader); |
3143 | 0 | access_prepare(&ac, env, ptr, size, MMU_DATA_LOAD, ra); |
3144 | 0 | if (!valid_xrstor_header(&ac, &xstate_bv, ptr)) { |
3145 | 0 | raise_exception_ra(env, EXCP0D_GPF, ra); |
3146 | 0 | } |
3147 | | |
3148 | 0 | rfbm &= env->xcr0; |
3149 | 0 | size_ext = xsave_area_size(rfbm & xstate_bv, false); |
3150 | 0 | if (size < size_ext) { |
3151 | | /* TODO: See if existing page probe has covered extra size. */ |
3152 | 0 | access_prepare(&ac, env, ptr, size_ext, MMU_DATA_LOAD, ra); |
3153 | 0 | } |
3154 | |
|
3155 | 0 | do_xrstor(&ac, ptr, rfbm, xstate_bv); |
3156 | 0 | } |
3157 | | |
3158 | | #if defined(CONFIG_USER_ONLY) |
3159 | | void cpu_x86_fsave(CPUX86State *env, void *host, size_t len) |
3160 | | { |
3161 | | cpu_x86_fsave_noinit(env, host, len); |
3162 | | do_fninit(env); |
3163 | | } |
3164 | | |
3165 | | void cpu_x86_fsave_noinit(CPUX86State *env, void *host, size_t len) |
3166 | | { |
3167 | | X86Access ac = { |
3168 | | .haddr1 = host, |
3169 | | .size = 4 * 7 + 8 * 10, |
3170 | | .env = env, |
3171 | | }; |
3172 | | |
3173 | | assert(ac.size <= len); |
3174 | | do_fsave(&ac, 0, true); |
3175 | | } |
3176 | | |
3177 | | void cpu_x86_frstor(CPUX86State *env, void *host, size_t len) |
3178 | | { |
3179 | | X86Access ac = { |
3180 | | .haddr1 = host, |
3181 | | .size = 4 * 7 + 8 * 10, |
3182 | | .env = env, |
3183 | | }; |
3184 | | |
3185 | | assert(ac.size <= len); |
3186 | | do_frstor(&ac, 0, true); |
3187 | | } |
3188 | | |
3189 | | void cpu_x86_fxsave(CPUX86State *env, void *host, size_t len) |
3190 | | { |
3191 | | X86Access ac = { |
3192 | | .haddr1 = host, |
3193 | | .size = sizeof(X86LegacyXSaveArea), |
3194 | | .env = env, |
3195 | | }; |
3196 | | |
3197 | | assert(ac.size <= len); |
3198 | | do_fxsave(&ac, 0); |
3199 | | } |
3200 | | |
3201 | | void cpu_x86_fxrstor(CPUX86State *env, void *host, size_t len) |
3202 | | { |
3203 | | X86Access ac = { |
3204 | | .haddr1 = host, |
3205 | | .size = sizeof(X86LegacyXSaveArea), |
3206 | | .env = env, |
3207 | | }; |
3208 | | |
3209 | | assert(ac.size <= len); |
3210 | | do_fxrstor(&ac, 0); |
3211 | | } |
3212 | | |
3213 | | void cpu_x86_xsave(CPUX86State *env, void *host, size_t len, uint64_t rfbm) |
3214 | | { |
3215 | | X86Access ac = { |
3216 | | .haddr1 = host, |
3217 | | .env = env, |
3218 | | }; |
3219 | | |
3220 | | /* |
3221 | | * Since this is only called from user-level signal handling, |
3222 | | * we should have done the job correctly there. |
3223 | | */ |
3224 | | assert((rfbm & ~env->xcr0) == 0); |
3225 | | ac.size = xsave_area_size(rfbm, false); |
3226 | | assert(ac.size <= len); |
3227 | | do_xsave_access(&ac, 0, rfbm, get_xinuse(env), rfbm); |
3228 | | } |
3229 | | |
3230 | | bool cpu_x86_xrstor(CPUX86State *env, void *host, size_t len, uint64_t rfbm) |
3231 | | { |
3232 | | X86Access ac = { |
3233 | | .haddr1 = host, |
3234 | | .env = env, |
3235 | | }; |
3236 | | uint64_t xstate_bv; |
3237 | | |
3238 | | /* |
3239 | | * Since this is only called from user-level signal handling, |
3240 | | * we should have done the job correctly there. |
3241 | | */ |
3242 | | assert((rfbm & ~env->xcr0) == 0); |
3243 | | ac.size = xsave_area_size(rfbm, false); |
3244 | | assert(ac.size <= len); |
3245 | | |
3246 | | if (!valid_xrstor_header(&ac, &xstate_bv, 0)) { |
3247 | | return false; |
3248 | | } |
3249 | | do_xrstor(&ac, 0, rfbm, xstate_bv); |
3250 | | return true; |
3251 | | } |
3252 | | #endif |
3253 | | |
3254 | | uint64_t HELPER(xgetbv)(CPUX86State *env, uint32_t ecx) |
3255 | 0 | { |
3256 | | /* The OS must have enabled XSAVE. */ |
3257 | 0 | if (!(env->cr[4] & CR4_OSXSAVE_MASK)) { |
3258 | 0 | raise_exception_ra(env, EXCP06_ILLOP, GETPC()); |
3259 | 0 | } |
3260 | | |
3261 | 0 | switch (ecx) { |
3262 | 0 | case 0: |
3263 | 0 | return env->xcr0; |
3264 | 0 | case 1: |
3265 | 0 | if (env->features[FEAT_XSAVE] & CPUID_XSAVE_XGETBV1) { |
3266 | 0 | return env->xcr0 & get_xinuse(env); |
3267 | 0 | } |
3268 | 0 | break; |
3269 | 0 | } |
3270 | 0 | raise_exception_ra(env, EXCP0D_GPF, GETPC()); |
3271 | 0 | } |
3272 | | |
3273 | | void HELPER(xsetbv)(CPUX86State *env, uint32_t ecx, uint64_t mask) |
3274 | 0 | { |
3275 | 0 | uint32_t dummy, ena_lo, ena_hi; |
3276 | 0 | uint64_t ena; |
3277 | | |
3278 | | /* The OS must have enabled XSAVE. */ |
3279 | 0 | if (!(env->cr[4] & CR4_OSXSAVE_MASK)) { |
3280 | 0 | raise_exception_ra(env, EXCP06_ILLOP, GETPC()); |
3281 | 0 | } |
3282 | | |
3283 | | /* Only XCR0 is defined at present; the FPU may not be disabled. */ |
3284 | 0 | if (ecx != 0 || (mask & XSTATE_FP_MASK) == 0) { |
3285 | 0 | goto do_gpf; |
3286 | 0 | } |
3287 | | |
3288 | | /* SSE can be disabled, but only if AVX is disabled too. */ |
3289 | 0 | if ((mask & (XSTATE_SSE_MASK | XSTATE_YMM_MASK)) == XSTATE_YMM_MASK) { |
3290 | 0 | goto do_gpf; |
3291 | 0 | } |
3292 | | |
3293 | | /* Disallow enabling unimplemented features. */ |
3294 | 0 | cpu_x86_cpuid(env, 0x0d, 0, &ena_lo, &dummy, &dummy, &ena_hi); |
3295 | 0 | ena = ((uint64_t)ena_hi << 32) | ena_lo; |
3296 | 0 | if (mask & ~ena) { |
3297 | 0 | goto do_gpf; |
3298 | 0 | } |
3299 | | |
3300 | | /* Disallow enabling only half of MPX. */ |
3301 | 0 | if ((mask ^ (mask * (XSTATE_BNDCSR_MASK / XSTATE_BNDREGS_MASK))) |
3302 | 0 | & XSTATE_BNDCSR_MASK) { |
3303 | 0 | goto do_gpf; |
3304 | 0 | } |
3305 | | |
3306 | 0 | env->xcr0 = mask; |
3307 | 0 | cpu_sync_bndcs_hflags(env); |
3308 | 0 | cpu_sync_avx_hflag(env); |
3309 | 0 | cpu_sync_apx_hflag(env); |
3310 | 0 | return; |
3311 | | |
3312 | 0 | do_gpf: |
3313 | 0 | raise_exception_ra(env, EXCP0D_GPF, GETPC()); |
3314 | 0 | } |
3315 | | |
3316 | | /* MMX/SSE */ |
3317 | | /* XXX: optimize by storing fptt and fptags in the static cpu state */ |
3318 | | |
3319 | 0 | #define SSE_DAZ 0x0040 |
3320 | 0 | #define SSE_RC_SHIFT 13 |
3321 | 0 | #define SSE_RC_MASK (3 << SSE_RC_SHIFT) |
3322 | 0 | #define SSE_FZ 0x8000 |
3323 | | |
3324 | | void update_mxcsr_status(CPUX86State *env) |
3325 | 0 | { |
3326 | 0 | uint32_t mxcsr = env->mxcsr; |
3327 | 0 | int rnd_type; |
3328 | | |
3329 | | /* set rounding mode */ |
3330 | 0 | rnd_type = (mxcsr & SSE_RC_MASK) >> SSE_RC_SHIFT; |
3331 | 0 | set_x86_rounding_mode(rnd_type, &env->sse_status); |
3332 | | |
3333 | | /* Set exception flags. */ |
3334 | 0 | set_float_exception_flags((mxcsr & FPUS_IE ? float_flag_invalid : 0) | |
3335 | 0 | (mxcsr & FPUS_DE ? float_flag_input_denormal_used : 0) | |
3336 | 0 | (mxcsr & FPUS_ZE ? float_flag_divbyzero : 0) | |
3337 | 0 | (mxcsr & FPUS_OE ? float_flag_overflow : 0) | |
3338 | 0 | (mxcsr & FPUS_UE ? float_flag_underflow : 0) | |
3339 | 0 | (mxcsr & FPUS_PE ? float_flag_inexact : 0), |
3340 | 0 | &env->sse_status); |
3341 | | |
3342 | | /* set denormals are zero */ |
3343 | 0 | set_flush_inputs_to_zero((mxcsr & SSE_DAZ) ? 1 : 0, &env->sse_status); |
3344 | | |
3345 | | /* set flush to zero */ |
3346 | 0 | set_flush_to_zero((mxcsr & SSE_FZ) ? 1 : 0, &env->sse_status); |
3347 | 0 | } |
3348 | | |
3349 | | void update_mxcsr_from_sse_status(CPUX86State *env) |
3350 | 0 | { |
3351 | 0 | int flags = get_float_exception_flags(&env->sse_status); |
3352 | 0 | env->mxcsr |= ((flags & float_flag_invalid ? FPUS_IE : 0) | |
3353 | 0 | (flags & float_flag_input_denormal_used ? FPUS_DE : 0) | |
3354 | 0 | (flags & float_flag_divbyzero ? FPUS_ZE : 0) | |
3355 | 0 | (flags & float_flag_overflow ? FPUS_OE : 0) | |
3356 | 0 | (flags & float_flag_underflow ? FPUS_UE : 0) | |
3357 | 0 | (flags & float_flag_inexact ? FPUS_PE : 0) | |
3358 | 0 | (flags & float_flag_output_denormal_flushed ? FPUS_UE | FPUS_PE : |
3359 | 0 | 0)); |
3360 | 0 | } |
3361 | | |
3362 | | void HELPER(update_mxcsr)(CPUX86State *env) |
3363 | 0 | { |
3364 | 0 | update_mxcsr_from_sse_status(env); |
3365 | 0 | } |
3366 | | |
3367 | | void HELPER(ldmxcsr)(CPUX86State *env, uint32_t val) |
3368 | 0 | { |
3369 | 0 | cpu_set_mxcsr(env, val); |
3370 | 0 | } |
3371 | | |
3372 | | void HELPER(enter_mmx)(CPUX86State *env) |
3373 | 0 | { |
3374 | 0 | env->fpstt = 0; |
3375 | 0 | *(uint32_t *)(env->fptags) = 0; |
3376 | 0 | *(uint32_t *)(env->fptags + 4) = 0; |
3377 | 0 | } |
3378 | | |
3379 | | void HELPER(emms)(CPUX86State *env) |
3380 | 0 | { |
3381 | | /* set to empty state */ |
3382 | 0 | *(uint32_t *)(env->fptags) = 0x01010101; |
3383 | 0 | *(uint32_t *)(env->fptags + 4) = 0x01010101; |
3384 | 0 | } |
3385 | | |
3386 | 0 | #define SHIFT 0 |
3387 | | #include "ops_sse.h" |
3388 | | |
3389 | 0 | #define SHIFT 1 |
3390 | | #include "ops_sse.h" |
3391 | | |
3392 | 0 | #define SHIFT 2 |
3393 | | #include "ops_sse.h" |