/src/htslib/htscodecs/htscodecs/utils.h
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Copyright (c) 2019-2022 Genome Research Ltd. |
3 | | * Author(s): James Bonfield |
4 | | * |
5 | | * Redistribution and use in source and binary forms, with or without |
6 | | * modification, are permitted provided that the following conditions are met: |
7 | | * |
8 | | * 1. Redistributions of source code must retain the above copyright notice, |
9 | | * this list of conditions and the following disclaimer. |
10 | | * |
11 | | * 2. Redistributions in binary form must reproduce the above |
12 | | * copyright notice, this list of conditions and the following |
13 | | * disclaimer in the documentation and/or other materials provided |
14 | | * with the distribution. |
15 | | * |
16 | | * 3. Neither the names Genome Research Ltd and Wellcome Trust Sanger |
17 | | * Institute nor the names of its contributors may be used to endorse |
18 | | * or promote products derived from this software without specific |
19 | | * prior written permission. |
20 | | * |
21 | | * THIS SOFTWARE IS PROVIDED BY GENOME RESEARCH LTD AND CONTRIBUTORS "AS |
22 | | * IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED |
23 | | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A |
24 | | * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL GENOME RESEARCH |
25 | | * LTD OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
26 | | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
27 | | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
28 | | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
29 | | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
30 | | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
31 | | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
32 | | */ |
33 | | |
34 | | #ifndef RANS_UTILS_H |
35 | | #define RANS_UTILS_H |
36 | | |
37 | | #include <string.h> |
38 | | #include <stdlib.h> |
39 | | #include <math.h> |
40 | | |
41 | | #if defined(__GNUC__) || defined(__clang__) |
42 | | # if !defined(__clang__) && __GNUC__ >= 100 |
43 | | // better still on gcc10 for O1 decode of old rans 4x8 |
44 | | // gcc 10=246/205 11=243/205 12=230/197 |
45 | | # define likely(x) __builtin_expect_with_probability((x), 1, 0.99) |
46 | | # else |
47 | | // gcc 10=193/168 11=195/161 12=199/176 |
48 | 0 | # define likely(x) __builtin_expect((x), 1) |
49 | | # endif |
50 | 85 | # define unlikely(x) __builtin_expect((x), 0) |
51 | | #else |
52 | | # define likely(x) (x) |
53 | | # define unlikely(x) (x) |
54 | | #endif |
55 | | |
56 | | /* |
57 | | * Allocates size bytes from the global Thread Local Storage pool. |
58 | | * This is shared by all subsequent calls within this thread. |
59 | | * |
60 | | * Note this is NOT a general purpose allocator and usage outside of this |
61 | | * library is not advised due to assumptions and limitations in the design. |
62 | | */ |
63 | | void *htscodecs_tls_alloc(size_t size); |
64 | | void *htscodecs_tls_calloc(size_t nmemb, size_t size); |
65 | | void htscodecs_tls_free(void *ptr); |
66 | | |
67 | | |
68 | | /* Fast approximate log base 2 */ |
69 | 0 | static inline double fast_log(double a) { |
70 | 0 | union { double d; long long x; } u = { a }; |
71 | 0 | return (u.x - 4606921278410026770) * 1.539095918623324e-16; /* 1 / 6497320848556798.0; */ |
72 | 0 | } Unexecuted instantiation: arith_dynamic.c:fast_log Unexecuted instantiation: fqzcomp_qual.c:fast_log Unexecuted instantiation: rANS_static4x16pr.c:fast_log Unexecuted instantiation: rANS_static32x16pr_avx2.c:fast_log Unexecuted instantiation: rANS_static32x16pr_avx512.c:fast_log Unexecuted instantiation: rANS_static32x16pr_sse4.c:fast_log Unexecuted instantiation: rANS_static32x16pr.c:fast_log Unexecuted instantiation: rANS_static.c:fast_log Unexecuted instantiation: tokenise_name3.c:fast_log Unexecuted instantiation: utils.c:fast_log |
73 | | |
74 | | /* |
75 | | * Data transpose by N. Common to rANS4x16 and arith_dynamic decoders. |
76 | | * |
77 | | * Tuned for specific common cases of N. |
78 | | */ |
79 | | static inline void unstripe(unsigned char *out, unsigned char *outN, |
80 | | unsigned int ulen, unsigned int N, |
81 | 72 | unsigned int idxN[256]) { |
82 | 72 | int j = 0, k; |
83 | | |
84 | 72 | if (ulen >= N) { |
85 | 52 | switch (N) { |
86 | 32 | case 4: |
87 | 211 | #define LLN 16 |
88 | 32 | if (ulen >= 4*LLN) { |
89 | 25 | while (j < ulen-4*LLN) { |
90 | 7 | int l; |
91 | 119 | for (l = 0; l < LLN; l++) { |
92 | 560 | for (k = 0; k < 4; k++) |
93 | 448 | out[j+k+l*4] = outN[idxN[k]+l]; |
94 | 112 | } |
95 | 35 | for (k = 0; k < 4; k++) |
96 | 28 | idxN[k] += LLN; |
97 | 7 | j += 4*LLN; |
98 | 7 | } |
99 | 18 | } |
100 | 223 | while (j < ulen-4) { |
101 | 955 | for (k = 0; k < 4; k++) |
102 | 764 | out[j++] = outN[idxN[k]++]; |
103 | 191 | } |
104 | 32 | #undef LLN |
105 | 32 | break; |
106 | | |
107 | 0 | case 2: |
108 | 0 | #define LLN 4 |
109 | 0 | if (ulen >= 2*LLN) { |
110 | 0 | while (j < ulen-2*LLN) { |
111 | 0 | int l; |
112 | 0 | for (l = 0; l < LLN; l++) { |
113 | 0 | for (k = 0; k < 2; k++) |
114 | 0 | out[j++] = outN[idxN[k]+l]; |
115 | 0 | } |
116 | 0 | for (k = 0; k < 2; k++) |
117 | 0 | idxN[k] += l; |
118 | 0 | } |
119 | 0 | } |
120 | 0 | while (j < ulen-2) { |
121 | 0 | for (k = 0; k < 2; k++) |
122 | 0 | out[j++] = outN[idxN[k]++]; |
123 | 0 | } |
124 | 0 | #undef LLN |
125 | 0 | break; |
126 | | |
127 | 20 | default: |
128 | | // General case, around 25% slower overall decode |
129 | 91 | while (j < ulen-N) { |
130 | 820 | for (k = 0; k < N; k++) |
131 | 749 | out[j++] = outN[idxN[k]++]; |
132 | 71 | } |
133 | 20 | break; |
134 | 52 | } |
135 | 52 | } |
136 | 499 | for (k = 0; j < ulen; k++) |
137 | 427 | out[j++] = outN[idxN[k]++]; |
138 | 72 | } Line | Count | Source | 81 | 41 | unsigned int idxN[256]) { | 82 | 41 | int j = 0, k; | 83 | | | 84 | 41 | if (ulen >= N) { | 85 | 21 | switch (N) { | 86 | 21 | case 4: | 87 | 21 | #define LLN 16 | 88 | 21 | if (ulen >= 4*LLN) { | 89 | 20 | while (j < ulen-4*LLN) { | 90 | 7 | int l; | 91 | 119 | for (l = 0; l < LLN; l++) { | 92 | 560 | for (k = 0; k < 4; k++) | 93 | 448 | out[j+k+l*4] = outN[idxN[k]+l]; | 94 | 112 | } | 95 | 35 | for (k = 0; k < 4; k++) | 96 | 28 | idxN[k] += LLN; | 97 | 7 | j += 4*LLN; | 98 | 7 | } | 99 | 13 | } | 100 | 137 | while (j < ulen-4) { | 101 | 580 | for (k = 0; k < 4; k++) | 102 | 464 | out[j++] = outN[idxN[k]++]; | 103 | 116 | } | 104 | 21 | #undef LLN | 105 | 21 | break; | 106 | | | 107 | 0 | case 2: | 108 | 0 | #define LLN 4 | 109 | 0 | if (ulen >= 2*LLN) { | 110 | 0 | while (j < ulen-2*LLN) { | 111 | 0 | int l; | 112 | 0 | for (l = 0; l < LLN; l++) { | 113 | 0 | for (k = 0; k < 2; k++) | 114 | 0 | out[j++] = outN[idxN[k]+l]; | 115 | 0 | } | 116 | 0 | for (k = 0; k < 2; k++) | 117 | 0 | idxN[k] += l; | 118 | 0 | } | 119 | 0 | } | 120 | 0 | while (j < ulen-2) { | 121 | 0 | for (k = 0; k < 2; k++) | 122 | 0 | out[j++] = outN[idxN[k]++]; | 123 | 0 | } | 124 | 0 | #undef LLN | 125 | 0 | break; | 126 | | | 127 | 0 | default: | 128 | | // General case, around 25% slower overall decode | 129 | 0 | while (j < ulen-N) { | 130 | 0 | for (k = 0; k < N; k++) | 131 | 0 | out[j++] = outN[idxN[k]++]; | 132 | 0 | } | 133 | 0 | break; | 134 | 21 | } | 135 | 21 | } | 136 | 116 | for (k = 0; j < ulen; k++) | 137 | 75 | out[j++] = outN[idxN[k]++]; | 138 | 41 | } |
Unexecuted instantiation: fqzcomp_qual.c:unstripe rANS_static4x16pr.c:unstripe Line | Count | Source | 81 | 31 | unsigned int idxN[256]) { | 82 | 31 | int j = 0, k; | 83 | | | 84 | 31 | if (ulen >= N) { | 85 | 31 | switch (N) { | 86 | 11 | case 4: | 87 | 11 | #define LLN 16 | 88 | 11 | if (ulen >= 4*LLN) { | 89 | 5 | while (j < ulen-4*LLN) { | 90 | 0 | int l; | 91 | 0 | for (l = 0; l < LLN; l++) { | 92 | 0 | for (k = 0; k < 4; k++) | 93 | 0 | out[j+k+l*4] = outN[idxN[k]+l]; | 94 | 0 | } | 95 | 0 | for (k = 0; k < 4; k++) | 96 | 0 | idxN[k] += LLN; | 97 | 0 | j += 4*LLN; | 98 | 0 | } | 99 | 5 | } | 100 | 86 | while (j < ulen-4) { | 101 | 375 | for (k = 0; k < 4; k++) | 102 | 300 | out[j++] = outN[idxN[k]++]; | 103 | 75 | } | 104 | 11 | #undef LLN | 105 | 11 | break; | 106 | | | 107 | 0 | case 2: | 108 | 0 | #define LLN 4 | 109 | 0 | if (ulen >= 2*LLN) { | 110 | 0 | while (j < ulen-2*LLN) { | 111 | 0 | int l; | 112 | 0 | for (l = 0; l < LLN; l++) { | 113 | 0 | for (k = 0; k < 2; k++) | 114 | 0 | out[j++] = outN[idxN[k]+l]; | 115 | 0 | } | 116 | 0 | for (k = 0; k < 2; k++) | 117 | 0 | idxN[k] += l; | 118 | 0 | } | 119 | 0 | } | 120 | 0 | while (j < ulen-2) { | 121 | 0 | for (k = 0; k < 2; k++) | 122 | 0 | out[j++] = outN[idxN[k]++]; | 123 | 0 | } | 124 | 0 | #undef LLN | 125 | 0 | break; | 126 | | | 127 | 20 | default: | 128 | | // General case, around 25% slower overall decode | 129 | 91 | while (j < ulen-N) { | 130 | 820 | for (k = 0; k < N; k++) | 131 | 749 | out[j++] = outN[idxN[k]++]; | 132 | 71 | } | 133 | 20 | break; | 134 | 31 | } | 135 | 31 | } | 136 | 383 | for (k = 0; j < ulen; k++) | 137 | 352 | out[j++] = outN[idxN[k]++]; | 138 | 31 | } |
Unexecuted instantiation: rANS_static32x16pr_avx2.c:unstripe Unexecuted instantiation: rANS_static32x16pr_avx512.c:unstripe Unexecuted instantiation: rANS_static32x16pr_sse4.c:unstripe Unexecuted instantiation: rANS_static32x16pr.c:unstripe Unexecuted instantiation: rANS_static.c:unstripe Unexecuted instantiation: tokenise_name3.c:unstripe Unexecuted instantiation: utils.c:unstripe |
139 | | |
140 | | #define MAGIC 8 |
141 | | |
142 | | /* |
143 | | * Order 0 histogram construction. 8-way unrolled to avoid cache collisions. |
144 | | */ |
145 | | static inline |
146 | 0 | void hist8(unsigned char *in, unsigned int in_size, uint32_t F0[256]) { |
147 | 0 | if (in_size > 500000) { |
148 | 0 | uint32_t *f0 = htscodecs_tls_calloc((65536+37)*3, sizeof(*f0)); |
149 | 0 | uint32_t *f1 = f0 + 65536+37; |
150 | 0 | uint32_t *f2 = f1 + 65536+37; |
151 | |
|
152 | 0 | uint32_t i, i8 = in_size & ~15; |
153 | |
|
154 | 0 | for (i = 0; i < i8; i+=16) { |
155 | 0 | uint16_t i16a[4], i16b[4]; |
156 | 0 | memcpy(i16a, in+i, 8); |
157 | 0 | f0[i16a[0]]++; |
158 | 0 | f1[i16a[1]]++; |
159 | 0 | f2[i16a[2]]++; |
160 | 0 | f0[i16a[3]]++; |
161 | |
|
162 | 0 | memcpy(i16b, in+i+8, 8); |
163 | 0 | f1[i16b[0]]++; |
164 | 0 | f0[i16b[1]]++; |
165 | 0 | f1[i16b[2]]++; |
166 | 0 | f2[i16b[3]]++; |
167 | 0 | } |
168 | |
|
169 | 0 | while (i < in_size) |
170 | 0 | F0[in[i++]]++; |
171 | |
|
172 | 0 | for (i = 0; i < 65536; i++) { |
173 | 0 | F0[i & 0xff] += f0[i] + f1[i] + f2[i]; |
174 | 0 | F0[i >> 8 ] += f0[i] + f1[i] + f2[i]; |
175 | 0 | } |
176 | 0 | htscodecs_tls_free(f0); |
177 | 0 | } else { |
178 | 0 | uint32_t F1[256+MAGIC] = {0}, F2[256+MAGIC] = {0}, F3[256+MAGIC] = {0}; |
179 | 0 | uint32_t i, i8 = in_size & ~7; |
180 | |
|
181 | 0 | for (i = 0; i < i8; i+=8) { |
182 | 0 | F0[in[i+0]]++; |
183 | 0 | F1[in[i+1]]++; |
184 | 0 | F2[in[i+2]]++; |
185 | 0 | F3[in[i+3]]++; |
186 | 0 | F0[in[i+4]]++; |
187 | 0 | F1[in[i+5]]++; |
188 | 0 | F2[in[i+6]]++; |
189 | 0 | F3[in[i+7]]++; |
190 | 0 | } |
191 | |
|
192 | 0 | while (i < in_size) |
193 | 0 | F0[in[i++]]++; |
194 | |
|
195 | 0 | for (i = 0; i < 256; i++) |
196 | 0 | F0[i] += F1[i] + F2[i] + F3[i]; |
197 | 0 | } |
198 | 0 | } Unexecuted instantiation: arith_dynamic.c:hist8 Unexecuted instantiation: fqzcomp_qual.c:hist8 Unexecuted instantiation: rANS_static4x16pr.c:hist8 Unexecuted instantiation: rANS_static32x16pr_avx2.c:hist8 Unexecuted instantiation: rANS_static32x16pr_avx512.c:hist8 Unexecuted instantiation: rANS_static32x16pr_sse4.c:hist8 Unexecuted instantiation: rANS_static32x16pr.c:hist8 Unexecuted instantiation: rANS_static.c:hist8 Unexecuted instantiation: tokenise_name3.c:hist8 Unexecuted instantiation: utils.c:hist8 |
199 | | |
200 | | // Hist8 with a crude entropy (bits / byte) estimator. |
201 | | static inline |
202 | 0 | double hist8e(unsigned char *in, unsigned int in_size, uint32_t F0[256]) { |
203 | 0 | uint32_t F1[256+MAGIC] = {0}, F2[256+MAGIC] = {0}, F3[256+MAGIC] = {0}; |
204 | 0 | uint32_t F4[256+MAGIC] = {0}, F5[256+MAGIC] = {0}, F6[256+MAGIC] = {0}; |
205 | 0 | uint32_t F7[256+MAGIC] = {0}; |
206 | |
|
207 | 0 | #ifdef __GNUC__ |
208 | 0 | double e = 0, in_size_r2 = log(1.0/in_size)/log(2); |
209 | | #else |
210 | | double e = 0, in_size_r2 = log(1.0/in_size); |
211 | | #endif |
212 | |
|
213 | 0 | unsigned int i, i8 = in_size & ~7; |
214 | 0 | for (i = 0; i < i8; i+=8) { |
215 | 0 | F0[in[i+0]]++; |
216 | 0 | F1[in[i+1]]++; |
217 | 0 | F2[in[i+2]]++; |
218 | 0 | F3[in[i+3]]++; |
219 | 0 | F4[in[i+4]]++; |
220 | 0 | F5[in[i+5]]++; |
221 | 0 | F6[in[i+6]]++; |
222 | 0 | F7[in[i+7]]++; |
223 | 0 | } |
224 | 0 | while (i < in_size) |
225 | 0 | F0[in[i++]]++; |
226 | |
|
227 | 0 | for (i = 0; i < 256; i++) { |
228 | 0 | F0[i] += F1[i] + F2[i] + F3[i] + F4[i] + F5[i] + F6[i] + F7[i]; |
229 | 0 | #ifdef __GNUC__ |
230 | 0 | e -= F0[i] * (32 - __builtin_clz(F0[i]|1) + in_size_r2); |
231 | | #else |
232 | | e -= F0[i] * (fast_log(F0[i]) + in_size_r2); |
233 | | #endif |
234 | 0 | } |
235 | |
|
236 | | #ifndef __GNUC__ |
237 | | e /= log(2); |
238 | | #endif |
239 | 0 | return e/in_size; |
240 | 0 | } Unexecuted instantiation: arith_dynamic.c:hist8e Unexecuted instantiation: fqzcomp_qual.c:hist8e Unexecuted instantiation: rANS_static4x16pr.c:hist8e Unexecuted instantiation: rANS_static32x16pr_avx2.c:hist8e Unexecuted instantiation: rANS_static32x16pr_avx512.c:hist8e Unexecuted instantiation: rANS_static32x16pr_sse4.c:hist8e Unexecuted instantiation: rANS_static32x16pr.c:hist8e Unexecuted instantiation: rANS_static.c:hist8e Unexecuted instantiation: tokenise_name3.c:hist8e Unexecuted instantiation: utils.c:hist8e |
241 | | |
242 | | /* |
243 | | * A variant of hist8 that simply marks the presence of a symbol rather |
244 | | * than its frequency. |
245 | | */ |
246 | | static inline |
247 | | void present8(unsigned char *in, unsigned int in_size, |
248 | 0 | uint32_t F0[256]) { |
249 | 0 | uint32_t F1[256+MAGIC] = {0}, F2[256+MAGIC] = {0}, F3[256+MAGIC] = {0}; |
250 | 0 | uint32_t F4[256+MAGIC] = {0}, F5[256+MAGIC] = {0}, F6[256+MAGIC] = {0}; |
251 | 0 | uint32_t F7[256+MAGIC] = {0}; |
252 | 0 |
|
253 | 0 | unsigned int i, i8 = in_size & ~7; |
254 | 0 | for (i = 0; i < i8; i+=8) { |
255 | 0 | F0[in[i+0]]=1; |
256 | 0 | F1[in[i+1]]=1; |
257 | 0 | F2[in[i+2]]=1; |
258 | 0 | F3[in[i+3]]=1; |
259 | 0 | F4[in[i+4]]=1; |
260 | 0 | F5[in[i+5]]=1; |
261 | 0 | F6[in[i+6]]=1; |
262 | 0 | F7[in[i+7]]=1; |
263 | 0 | } |
264 | 0 | while (i < in_size) |
265 | 0 | F0[in[i++]]=1; |
266 | 0 |
|
267 | 0 | for (i = 0; i < 256; i++) |
268 | 0 | F0[i] += F1[i] + F2[i] + F3[i] + F4[i] + F5[i] + F6[i] + F7[i]; |
269 | 0 | } Unexecuted instantiation: arith_dynamic.c:present8 Unexecuted instantiation: fqzcomp_qual.c:present8 Unexecuted instantiation: rANS_static4x16pr.c:present8 Unexecuted instantiation: rANS_static32x16pr_avx2.c:present8 Unexecuted instantiation: rANS_static32x16pr_avx512.c:present8 Unexecuted instantiation: rANS_static32x16pr_sse4.c:present8 Unexecuted instantiation: rANS_static32x16pr.c:present8 Unexecuted instantiation: rANS_static.c:present8 Unexecuted instantiation: tokenise_name3.c:present8 Unexecuted instantiation: utils.c:present8 |
270 | | |
271 | | /* |
272 | | * Order 1 histogram construction. 4-way unrolled to avoid cache collisions. |
273 | | */ |
274 | | #if 1 |
275 | | static inline |
276 | | void hist1_4(unsigned char *in, unsigned int in_size, |
277 | 0 | uint32_t F0[256][256], uint32_t *T0) { |
278 | 0 | unsigned char l = 0, c; |
279 | 0 | unsigned char *in_end = in + in_size; |
280 | |
|
281 | 0 | unsigned char cc[5] = {0}; |
282 | 0 | if (in_size > 500000) { |
283 | 0 | uint32_t (*F1)[259] = htscodecs_tls_calloc(256, sizeof(*F1)); |
284 | 0 | while (in < in_end-8) { |
285 | 0 | memcpy(cc, in, 4); in += 4; |
286 | 0 | F0[cc[4]][cc[0]]++; |
287 | 0 | F1[cc[0]][cc[1]]++; |
288 | 0 | F0[cc[1]][cc[2]]++; |
289 | 0 | F1[cc[2]][cc[3]]++; |
290 | 0 | cc[4] = cc[3]; |
291 | |
|
292 | 0 | memcpy(cc, in, 4); in += 4; |
293 | 0 | F0[cc[4]][cc[0]]++; |
294 | 0 | F1[cc[0]][cc[1]]++; |
295 | 0 | F0[cc[1]][cc[2]]++; |
296 | 0 | F1[cc[2]][cc[3]]++; |
297 | 0 | cc[4] = cc[3]; |
298 | 0 | } |
299 | 0 | l = cc[3]; |
300 | |
|
301 | 0 | while (in < in_end) { |
302 | 0 | F0[l][c = *in++]++; |
303 | 0 | l = c; |
304 | 0 | } |
305 | 0 | T0[l]++; |
306 | |
|
307 | 0 | int i, j; |
308 | 0 | for (i = 0; i < 256; i++) { |
309 | 0 | int tt = 0; |
310 | 0 | for (j = 0; j < 256; j++) { |
311 | 0 | F0[i][j] += F1[i][j]; |
312 | 0 | tt += F0[i][j]; |
313 | 0 | } |
314 | 0 | T0[i]+=tt; |
315 | 0 | } |
316 | 0 | htscodecs_tls_free(F1); |
317 | 0 | } else { |
318 | 0 | while (in < in_end-8) { |
319 | 0 | memcpy(cc, in, 4); in += 4; |
320 | 0 | F0[cc[4]][cc[0]]++; |
321 | 0 | F0[cc[0]][cc[1]]++; |
322 | 0 | F0[cc[1]][cc[2]]++; |
323 | 0 | F0[cc[2]][cc[3]]++; |
324 | 0 | cc[4] = cc[3]; |
325 | |
|
326 | 0 | memcpy(cc, in, 4); in += 4; |
327 | 0 | F0[cc[4]][cc[0]]++; |
328 | 0 | F0[cc[0]][cc[1]]++; |
329 | 0 | F0[cc[1]][cc[2]]++; |
330 | 0 | F0[cc[2]][cc[3]]++; |
331 | 0 | cc[4] = cc[3]; |
332 | 0 | } |
333 | 0 | l = cc[3]; |
334 | |
|
335 | 0 | while (in < in_end) { |
336 | 0 | F0[l][c = *in++]++; |
337 | 0 | l = c; |
338 | 0 | } |
339 | 0 | T0[l]++; |
340 | |
|
341 | 0 | int i, j; |
342 | 0 | for (i = 0; i < 256; i++) { |
343 | 0 | int tt = 0; |
344 | 0 | for (j = 0; j < 256; j++) |
345 | 0 | tt += F0[i][j]; |
346 | 0 | T0[i]+=tt; |
347 | 0 | } |
348 | 0 | } |
349 | 0 | } Unexecuted instantiation: arith_dynamic.c:hist1_4 Unexecuted instantiation: fqzcomp_qual.c:hist1_4 Unexecuted instantiation: rANS_static4x16pr.c:hist1_4 Unexecuted instantiation: rANS_static32x16pr_avx2.c:hist1_4 Unexecuted instantiation: rANS_static32x16pr_avx512.c:hist1_4 Unexecuted instantiation: rANS_static32x16pr_sse4.c:hist1_4 Unexecuted instantiation: rANS_static32x16pr.c:hist1_4 Unexecuted instantiation: rANS_static.c:hist1_4 Unexecuted instantiation: tokenise_name3.c:hist1_4 Unexecuted instantiation: utils.c:hist1_4 |
350 | | |
351 | | #else |
352 | | // 16 bit mode, similar to O0 freq. |
353 | | // This is better on some low entropy data, but generally we prefer to do |
354 | | // bit-packing and/or RLE to turn it into higher-entropy data first. |
355 | | // |
356 | | // Kept here for posterity incase we need it again, as it's quick tricky. |
357 | | static inline |
358 | | void hist1_4(unsigned char *in, unsigned int in_size, |
359 | | uint32_t F0[256][256], uint32_t *T0) { |
360 | | uint32_t f0[65536+MAGIC] = {0}; |
361 | | uint32_t f1[65536+MAGIC] = {0}; |
362 | | |
363 | | uint32_t i, i8 = (in_size-1) & ~15; |
364 | | |
365 | | T0[0]++; f0[in[0]<<8]++; |
366 | | for (i = 0; i < i8; i+=16) { |
367 | | uint16_t i16a[16]; |
368 | | memcpy(i16a, in+i, 16); // faster in 2 as gcc recognises this |
369 | | memcpy(i16a+8, in+i+1, 16); // faster in 2 as gcc recognises this |
370 | | |
371 | | f0[i16a[0]]++; |
372 | | f1[i16a[1]]++; |
373 | | f0[i16a[2]]++; |
374 | | f1[i16a[3]]++; |
375 | | f0[i16a[4]]++; |
376 | | f1[i16a[5]]++; |
377 | | f0[i16a[6]]++; |
378 | | f1[i16a[7]]++; |
379 | | f0[i16a[8]]++; |
380 | | f1[i16a[9]]++; |
381 | | f0[i16a[10]]++; |
382 | | f1[i16a[11]]++; |
383 | | f0[i16a[12]]++; |
384 | | f1[i16a[13]]++; |
385 | | f0[i16a[14]]++; |
386 | | f1[i16a[15]]++; |
387 | | } |
388 | | |
389 | | while (i < in_size-1) { |
390 | | F0[in[i]][in[i+1]]++; |
391 | | T0[in[i+1]]++; |
392 | | i++; |
393 | | } |
394 | | |
395 | | for (i = 0; i < 65536; i++) { |
396 | | F0[i&0xff][i>>8] += f0[i] + f1[i]; |
397 | | T0[i>>8] += f0[i] + f1[i]; |
398 | | } |
399 | | } |
400 | | #endif |
401 | | |
402 | | #endif /* RANS_UTILS_H */ |