1 /*
2 * This file is part of FFmpeg.
3 *
4 * FFmpeg is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * FFmpeg is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #include "config.h"
20
21 #include <float.h>
22 #include <stdint.h>
23
24 #include "libavutil/float_dsp.h"
25 #include "libavutil/internal.h"
26 #include "checkasm.h"
27
28 #define LEN 256
29
30 #define randomize_buffer(buf) \
31 do { \
32 int i; \
33 double bmg[2], stddev = 10.0, mean = 0.0; \
34 \
35 for (i = 0; i < LEN; i += 2) { \
36 av_bmg_get(&checkasm_lfg, bmg); \
37 buf[i] = bmg[0] * stddev + mean; \
38 buf[i + 1] = bmg[1] * stddev + mean; \
39 } \
40 } while(0);
41
test_vector_fmul(const float * src0,const float * src1)42 static void test_vector_fmul(const float *src0, const float *src1)
43 {
44 LOCAL_ALIGNED_32(float, cdst, [LEN]);
45 LOCAL_ALIGNED_32(float, odst, [LEN]);
46 int i;
47
48 declare_func(void, float *dst, const float *src0, const float *src1,
49 int len);
50
51 call_ref(cdst, src0, src1, LEN);
52 call_new(odst, src0, src1, LEN);
53 for (i = 0; i < LEN; i++) {
54 double t = fabs(src0[i]) + fabs(src1[i]) + fabs(src0[i] * src1[i]) + 1.0;
55 if (!float_near_abs_eps(cdst[i], odst[i], t * 2 * FLT_EPSILON)) {
56 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
57 i, cdst[i], odst[i], cdst[i] - odst[i]);
58 fail();
59 break;
60 }
61 }
62 bench_new(odst, src0, src1, LEN);
63 }
64
test_vector_dmul(const double * src0,const double * src1)65 static void test_vector_dmul(const double *src0, const double *src1)
66 {
67 LOCAL_ALIGNED_32(double, cdst, [LEN]);
68 LOCAL_ALIGNED_32(double, odst, [LEN]);
69 int i;
70
71 declare_func(void, double *dst, const double *src0, const double *src1,
72 int len);
73
74 call_ref(cdst, src0, src1, LEN);
75 call_new(odst, src0, src1, LEN);
76 for (i = 0; i < LEN; i++) {
77 double t = fabs(src0[i]) + fabs(src1[i]) + fabs(src0[i] * src1[i]) + 1.0;
78 if (!double_near_abs_eps(cdst[i], odst[i], t * 2 * DBL_EPSILON)) {
79 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
80 i, cdst[i], odst[i], cdst[i] - odst[i]);
81 fail();
82 break;
83 }
84 }
85 bench_new(odst, src0, src1, LEN);
86 }
87
88 #define ARBITRARY_FMUL_ADD_CONST 0.005
test_vector_fmul_add(const float * src0,const float * src1,const float * src2)89 static void test_vector_fmul_add(const float *src0, const float *src1, const float *src2)
90 {
91 LOCAL_ALIGNED_32(float, cdst, [LEN]);
92 LOCAL_ALIGNED_32(float, odst, [LEN]);
93 int i;
94
95 declare_func(void, float *dst, const float *src0, const float *src1,
96 const float *src2, int len);
97
98 call_ref(cdst, src0, src1, src2, LEN);
99 call_new(odst, src0, src1, src2, LEN);
100 for (i = 0; i < LEN; i++) {
101 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMUL_ADD_CONST)) {
102 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
103 i, cdst[i], odst[i], cdst[i] - odst[i]);
104 fail();
105 break;
106 }
107 }
108 bench_new(odst, src0, src1, src2, LEN);
109 }
110
test_vector_fmul_scalar(const float * src0,const float * src1)111 static void test_vector_fmul_scalar(const float *src0, const float *src1)
112 {
113 LOCAL_ALIGNED_16(float, cdst, [LEN]);
114 LOCAL_ALIGNED_16(float, odst, [LEN]);
115 int i;
116
117 declare_func(void, float *dst, const float *src, float mul, int len);
118
119 call_ref(cdst, src0, src1[0], LEN);
120 call_new(odst, src0, src1[0], LEN);
121 for (i = 0; i < LEN; i++) {
122 double t = fabs(src0[i]) + fabs(src1[0]) + fabs(src0[i] * src1[0]) + 1.0;
123 if (!float_near_abs_eps(cdst[i], odst[i], t * 2 * FLT_EPSILON)) {
124 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
125 i, cdst[i], odst[i], cdst[i] - odst[i]);
126 fail();
127 break;
128 }
129 }
130 bench_new(odst, src0, src1[0], LEN);
131 }
132
133 #define ARBITRARY_FMUL_WINDOW_CONST 0.008
test_vector_fmul_window(const float * src0,const float * src1,const float * win)134 static void test_vector_fmul_window(const float *src0, const float *src1, const float *win)
135 {
136 LOCAL_ALIGNED_16(float, cdst, [LEN]);
137 LOCAL_ALIGNED_16(float, odst, [LEN]);
138 int i;
139
140 declare_func(void, float *dst, const float *src0, const float *src1,
141 const float *win, int len);
142
143 call_ref(cdst, src0, src1, win, LEN / 2);
144 call_new(odst, src0, src1, win, LEN / 2);
145 for (i = 0; i < LEN; i++) {
146 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMUL_WINDOW_CONST)) {
147 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
148 i, cdst[i], odst[i], cdst[i] - odst[i]);
149 fail();
150 break;
151 }
152 }
153 bench_new(odst, src0, src1, win, LEN / 2);
154 }
155
156 #define ARBITRARY_FMAC_SCALAR_CONST 0.005
test_vector_fmac_scalar(const float * src0,const float * src1,const float * src2)157 static void test_vector_fmac_scalar(const float *src0, const float *src1, const float *src2)
158 {
159 LOCAL_ALIGNED_32(float, cdst, [LEN]);
160 LOCAL_ALIGNED_32(float, odst, [LEN]);
161 int i;
162
163 declare_func(void, float *dst, const float *src, float mul, int len);
164
165 memcpy(cdst, src2, LEN * sizeof(*src2));
166 memcpy(odst, src2, LEN * sizeof(*src2));
167
168 call_ref(cdst, src0, src1[0], LEN);
169 call_new(odst, src0, src1[0], LEN);
170 for (i = 0; i < LEN; i++) {
171 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMAC_SCALAR_CONST)) {
172 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
173 i, cdst[i], odst[i], cdst[i] - odst[i]);
174 fail();
175 break;
176 }
177 }
178 memcpy(odst, src2, LEN * sizeof(*src2));
179 bench_new(odst, src0, src1[0], LEN);
180 }
181
test_vector_dmul_scalar(const double * src0,const double * src1)182 static void test_vector_dmul_scalar(const double *src0, const double *src1)
183 {
184 LOCAL_ALIGNED_32(double, cdst, [LEN]);
185 LOCAL_ALIGNED_32(double, odst, [LEN]);
186 int i;
187
188 declare_func(void, double *dst, const double *src, double mul, int len);
189
190 call_ref(cdst, src0, src1[0], LEN);
191 call_new(odst, src0, src1[0], LEN);
192 for (i = 0; i < LEN; i++) {
193 double t = fabs(src1[0]) + fabs(src0[i]) + fabs(src1[0] * src0[i]) + 1.0;
194 if (!double_near_abs_eps(cdst[i], odst[i], t * 2 * DBL_EPSILON)) {
195 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n", i,
196 cdst[i], odst[i], cdst[i] - odst[i]);
197 fail();
198 break;
199 }
200 }
201 bench_new(odst, src0, src1[0], LEN);
202 }
203
204 #define ARBITRARY_DMAC_SCALAR_CONST 0.005
test_vector_dmac_scalar(const double * src0,const double * src1,const double * src2)205 static void test_vector_dmac_scalar(const double *src0, const double *src1, const double *src2)
206 {
207 LOCAL_ALIGNED_32(double, cdst, [LEN]);
208 LOCAL_ALIGNED_32(double, odst, [LEN]);
209 int i;
210
211 declare_func(void, double *dst, const double *src, double mul, int len);
212
213 memcpy(cdst, src2, LEN * sizeof(*src2));
214 memcpy(odst, src2, LEN * sizeof(*src2));
215 call_ref(cdst, src0, src1[0], LEN);
216 call_new(odst, src0, src1[0], LEN);
217 for (i = 0; i < LEN; i++) {
218 if (!double_near_abs_eps(cdst[i], odst[i], ARBITRARY_DMAC_SCALAR_CONST)) {
219 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
220 i, cdst[i], odst[i], cdst[i] - odst[i]);
221 fail();
222 break;
223 }
224 }
225 memcpy(odst, src2, LEN * sizeof(*src2));
226 bench_new(odst, src0, src1[0], LEN);
227 }
228
test_butterflies_float(const float * src0,const float * src1)229 static void test_butterflies_float(const float *src0, const float *src1)
230 {
231 LOCAL_ALIGNED_16(float, cdst, [LEN]);
232 LOCAL_ALIGNED_16(float, odst, [LEN]);
233 LOCAL_ALIGNED_16(float, cdst1, [LEN]);
234 LOCAL_ALIGNED_16(float, odst1, [LEN]);
235 int i;
236
237 declare_func(void, float *av_restrict src0, float *av_restrict src1,
238 int len);
239
240 memcpy(cdst, src0, LEN * sizeof(*src0));
241 memcpy(cdst1, src1, LEN * sizeof(*src1));
242 memcpy(odst, src0, LEN * sizeof(*src0));
243 memcpy(odst1, src1, LEN * sizeof(*src1));
244
245 call_ref(cdst, cdst1, LEN);
246 call_new(odst, odst1, LEN);
247 for (i = 0; i < LEN; i++) {
248 if (!float_near_abs_eps(cdst[i], odst[i], FLT_EPSILON) ||
249 !float_near_abs_eps(cdst1[i], odst1[i], FLT_EPSILON)) {
250 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
251 i, cdst[i], odst[i], cdst[i] - odst[i]);
252 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
253 i, cdst1[i], odst1[i], cdst1[i] - odst1[i]);
254 fail();
255 break;
256 }
257 }
258 memcpy(odst, src0, LEN * sizeof(*src0));
259 memcpy(odst1, src1, LEN * sizeof(*src1));
260 bench_new(odst, odst1, LEN);
261 }
262
263 #define ARBITRARY_SCALARPRODUCT_CONST 0.2
test_scalarproduct_float(const float * src0,const float * src1)264 static void test_scalarproduct_float(const float *src0, const float *src1)
265 {
266 float cprod, oprod;
267
268 declare_func_float(float, const float *src0, const float *src1, int len);
269
270 cprod = call_ref(src0, src1, LEN);
271 oprod = call_new(src0, src1, LEN);
272 if (!float_near_abs_eps(cprod, oprod, ARBITRARY_SCALARPRODUCT_CONST)) {
273 fprintf(stderr, "%- .12f - %- .12f = % .12g\n",
274 cprod, oprod, cprod - oprod);
275 fail();
276 }
277 bench_new(src0, src1, LEN);
278 }
279
checkasm_check_float_dsp(void)280 void checkasm_check_float_dsp(void)
281 {
282 LOCAL_ALIGNED_32(float, src0, [LEN]);
283 LOCAL_ALIGNED_32(float, src1, [LEN]);
284 LOCAL_ALIGNED_32(float, src2, [LEN]);
285 LOCAL_ALIGNED_16(float, src3, [LEN]);
286 LOCAL_ALIGNED_16(float, src4, [LEN]);
287 LOCAL_ALIGNED_16(float, src5, [LEN]);
288 LOCAL_ALIGNED_32(double, dbl_src0, [LEN]);
289 LOCAL_ALIGNED_32(double, dbl_src1, [LEN]);
290 LOCAL_ALIGNED_32(double, dbl_src2, [LEN]);
291 AVFloatDSPContext *fdsp = avpriv_float_dsp_alloc(1);
292
293 if (!fdsp) {
294 fprintf(stderr, "floatdsp: Out of memory error\n");
295 return;
296 }
297
298 randomize_buffer(src0);
299 randomize_buffer(src1);
300 randomize_buffer(src2);
301 randomize_buffer(src3);
302 randomize_buffer(src4);
303 randomize_buffer(src5);
304 randomize_buffer(dbl_src0);
305 randomize_buffer(dbl_src1);
306 randomize_buffer(dbl_src2);
307
308 if (check_func(fdsp->vector_fmul, "vector_fmul"))
309 test_vector_fmul(src0, src1);
310 if (check_func(fdsp->vector_fmul_add, "vector_fmul_add"))
311 test_vector_fmul_add(src0, src1, src2);
312 if (check_func(fdsp->vector_fmul_scalar, "vector_fmul_scalar"))
313 test_vector_fmul_scalar(src3, src4);
314 if (check_func(fdsp->vector_fmul_reverse, "vector_fmul_reverse"))
315 test_vector_fmul(src0, src1);
316 if (check_func(fdsp->vector_fmul_window, "vector_fmul_window"))
317 test_vector_fmul_window(src3, src4, src5);
318 report("vector_fmul");
319 if (check_func(fdsp->vector_fmac_scalar, "vector_fmac_scalar"))
320 test_vector_fmac_scalar(src0, src1, src2);
321 report("vector_fmac");
322 if (check_func(fdsp->vector_dmul, "vector_dmul"))
323 test_vector_dmul(dbl_src0, dbl_src1);
324 if (check_func(fdsp->vector_dmul_scalar, "vector_dmul_scalar"))
325 test_vector_dmul_scalar(dbl_src0, dbl_src1);
326 report("vector_dmul");
327 if (check_func(fdsp->vector_dmac_scalar, "vector_dmac_scalar"))
328 test_vector_dmac_scalar(dbl_src0, dbl_src1, dbl_src2);
329 report("vector_dmac");
330 if (check_func(fdsp->butterflies_float, "butterflies_float"))
331 test_butterflies_float(src3, src4);
332 report("butterflies_float");
333 if (check_func(fdsp->scalarproduct_float, "scalarproduct_float"))
334 test_scalarproduct_float(src3, src4);
335 report("scalarproduct_float");
336
337 av_freep(&fdsp);
338 }
339