2 * This file is part of FFmpeg.
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.
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.
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.
24 #include "libavutil/float_dsp.h"
25 #include "libavutil/internal.h"
30 #define randomize_buffer(buf) \
33 double bmg[2], stddev = 10.0, mean = 0.0; \
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; \
42 static void test_vector_fmul(const float *src0, const float *src1)
44 LOCAL_ALIGNED_32(float, cdst, [LEN]);
45 LOCAL_ALIGNED_32(float, odst, [LEN]);
48 declare_func(void, float *dst, const float *src0, const float *src1,
51 call_ref(cdst, src0, src1, LEN);
52 call_new(odst, src0, src1, LEN);
53 for (i = 0; i < LEN; i++) {
54 if (!float_near_abs_eps(cdst[i], odst[i], FLT_EPSILON)) {
55 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
56 i, cdst[i], odst[i], cdst[i] - odst[i]);
61 bench_new(odst, src0, src1, LEN);
64 static void test_vector_dmul(const double *src0, const double *src1)
66 LOCAL_ALIGNED_32(double, cdst, [LEN]);
67 LOCAL_ALIGNED_32(double, odst, [LEN]);
70 declare_func(void, double *dst, const double *src0, const double *src1,
73 call_ref(cdst, src0, src1, LEN);
74 call_new(odst, src0, src1, LEN);
75 for (i = 0; i < LEN; i++) {
76 if (!double_near_abs_eps(cdst[i], odst[i], DBL_EPSILON)) {
77 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
78 i, cdst[i], odst[i], cdst[i] - odst[i]);
83 bench_new(odst, src0, src1, LEN);
86 #define ARBITRARY_FMUL_ADD_CONST 0.005
87 static void test_vector_fmul_add(const float *src0, const float *src1, const float *src2)
89 LOCAL_ALIGNED_32(float, cdst, [LEN]);
90 LOCAL_ALIGNED_32(float, odst, [LEN]);
93 declare_func(void, float *dst, const float *src0, const float *src1,
94 const float *src2, int len);
96 call_ref(cdst, src0, src1, src2, LEN);
97 call_new(odst, src0, src1, src2, LEN);
98 for (i = 0; i < LEN; i++) {
99 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMUL_ADD_CONST)) {
100 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
101 i, cdst[i], odst[i], cdst[i] - odst[i]);
106 bench_new(odst, src0, src1, src2, LEN);
109 static void test_vector_fmul_scalar(const float *src0, const float *src1)
111 LOCAL_ALIGNED_16(float, cdst, [LEN]);
112 LOCAL_ALIGNED_16(float, odst, [LEN]);
115 declare_func(void, float *dst, const float *src, float mul, int len);
117 call_ref(cdst, src0, src1[0], LEN);
118 call_new(odst, src0, src1[0], LEN);
119 for (i = 0; i < LEN; i++) {
120 if (!float_near_abs_eps(cdst[i], odst[i], FLT_EPSILON)) {
121 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
122 i, cdst[i], odst[i], cdst[i] - odst[i]);
127 bench_new(odst, src0, src1[0], LEN);
130 #define ARBITRARY_FMUL_WINDOW_CONST 0.008
131 static void test_vector_fmul_window(const float *src0, const float *src1, const float *win)
133 LOCAL_ALIGNED_16(float, cdst, [LEN]);
134 LOCAL_ALIGNED_16(float, odst, [LEN]);
137 declare_func(void, float *dst, const float *src0, const float *src1,
138 const float *win, int len);
140 call_ref(cdst, src0, src1, win, LEN / 2);
141 call_new(odst, src0, src1, win, LEN / 2);
142 for (i = 0; i < LEN; i++) {
143 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMUL_WINDOW_CONST)) {
144 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
145 i, cdst[i], odst[i], cdst[i] - odst[i]);
150 bench_new(odst, src0, src1, win, LEN / 2);
153 #define ARBITRARY_FMAC_SCALAR_CONST 0.005
154 static void test_vector_fmac_scalar(const float *src0, const float *src1, const float *src2)
156 LOCAL_ALIGNED_32(float, cdst, [LEN]);
157 LOCAL_ALIGNED_32(float, odst, [LEN]);
160 declare_func(void, float *dst, const float *src, float mul, int len);
162 memcpy(cdst, src2, LEN * sizeof(*src2));
163 memcpy(odst, src2, LEN * sizeof(*src2));
165 call_ref(cdst, src0, src1[0], LEN);
166 call_new(odst, src0, src1[0], LEN);
167 for (i = 0; i < LEN; i++) {
168 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMAC_SCALAR_CONST)) {
169 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
170 i, cdst[i], odst[i], cdst[i] - odst[i]);
175 memcpy(odst, src2, LEN * sizeof(*src2));
176 bench_new(odst, src0, src1[0], LEN);
179 static void test_vector_dmul_scalar(const double *src0, const double *src1)
181 LOCAL_ALIGNED_32(double, cdst, [LEN]);
182 LOCAL_ALIGNED_32(double, odst, [LEN]);
185 declare_func(void, double *dst, const double *src, double mul, int len);
187 call_ref(cdst, src0, src1[0], LEN);
188 call_new(odst, src0, src1[0], LEN);
189 for (i = 0; i < LEN; i++) {
190 double t = fabs(src1[0]) + fabs(src0[i]) + fabs(src1[0] * src0[i]) + 1.0;
191 if (!double_near_abs_eps(cdst[i], odst[i], t * 2 * DBL_EPSILON)) {
192 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n", i,
193 cdst[i], odst[i], cdst[i] - odst[i]);
198 bench_new(odst, src0, src1[0], LEN);
201 #define ARBITRARY_DMAC_SCALAR_CONST 0.005
202 static void test_vector_dmac_scalar(const double *src0, const double *src1, const double *src2)
204 LOCAL_ALIGNED_32(double, cdst, [LEN]);
205 LOCAL_ALIGNED_32(double, odst, [LEN]);
208 declare_func(void, double *dst, const double *src, double mul, int len);
210 memcpy(cdst, src2, LEN * sizeof(*src2));
211 memcpy(odst, src2, LEN * sizeof(*src2));
212 call_ref(cdst, src0, src1[0], LEN);
213 call_new(odst, src0, src1[0], LEN);
214 for (i = 0; i < LEN; i++) {
215 if (!double_near_abs_eps(cdst[i], odst[i], ARBITRARY_DMAC_SCALAR_CONST)) {
216 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
217 i, cdst[i], odst[i], cdst[i] - odst[i]);
222 memcpy(odst, src2, LEN * sizeof(*src2));
223 bench_new(odst, src0, src1[0], LEN);
226 static void test_butterflies_float(const float *src0, const float *src1)
228 LOCAL_ALIGNED_16(float, cdst, [LEN]);
229 LOCAL_ALIGNED_16(float, odst, [LEN]);
230 LOCAL_ALIGNED_16(float, cdst1, [LEN]);
231 LOCAL_ALIGNED_16(float, odst1, [LEN]);
234 declare_func(void, float *av_restrict src0, float *av_restrict src1,
237 memcpy(cdst, src0, LEN * sizeof(*src0));
238 memcpy(cdst1, src1, LEN * sizeof(*src1));
239 memcpy(odst, src0, LEN * sizeof(*src0));
240 memcpy(odst1, src1, LEN * sizeof(*src1));
242 call_ref(cdst, cdst1, LEN);
243 call_new(odst, odst1, LEN);
244 for (i = 0; i < LEN; i++) {
245 if (!float_near_abs_eps(cdst[i], odst[i], FLT_EPSILON) ||
246 !float_near_abs_eps(cdst1[i], odst1[i], FLT_EPSILON)) {
247 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
248 i, cdst[i], odst[i], cdst[i] - odst[i]);
249 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
250 i, cdst1[i], odst1[i], cdst1[i] - odst1[i]);
255 memcpy(odst, src0, LEN * sizeof(*src0));
256 memcpy(odst1, src1, LEN * sizeof(*src1));
257 bench_new(odst, odst1, LEN);
260 #define ARBITRARY_SCALARPRODUCT_CONST 0.2
261 static void test_scalarproduct_float(const float *src0, const float *src1)
265 declare_func_float(float, const float *src0, const float *src1, int len);
267 cprod = call_ref(src0, src1, LEN);
268 oprod = call_new(src0, src1, LEN);
269 if (!float_near_abs_eps(cprod, oprod, ARBITRARY_SCALARPRODUCT_CONST)) {
270 fprintf(stderr, "%- .12f - %- .12f = % .12g\n",
271 cprod, oprod, cprod - oprod);
274 bench_new(src0, src1, LEN);
277 void checkasm_check_float_dsp(void)
279 LOCAL_ALIGNED_32(float, src0, [LEN]);
280 LOCAL_ALIGNED_32(float, src1, [LEN]);
281 LOCAL_ALIGNED_32(float, src2, [LEN]);
282 LOCAL_ALIGNED_16(float, src3, [LEN]);
283 LOCAL_ALIGNED_16(float, src4, [LEN]);
284 LOCAL_ALIGNED_16(float, src5, [LEN]);
285 LOCAL_ALIGNED_32(double, dbl_src0, [LEN]);
286 LOCAL_ALIGNED_32(double, dbl_src1, [LEN]);
287 LOCAL_ALIGNED_32(double, dbl_src2, [LEN]);
288 AVFloatDSPContext *fdsp = avpriv_float_dsp_alloc(1);
291 fprintf(stderr, "floatdsp: Out of memory error\n");
295 randomize_buffer(src0);
296 randomize_buffer(src1);
297 randomize_buffer(src2);
298 randomize_buffer(src3);
299 randomize_buffer(src4);
300 randomize_buffer(src5);
301 randomize_buffer(dbl_src0);
302 randomize_buffer(dbl_src1);
303 randomize_buffer(dbl_src2);
305 if (check_func(fdsp->vector_fmul, "vector_fmul"))
306 test_vector_fmul(src0, src1);
307 if (check_func(fdsp->vector_fmul_add, "vector_fmul_add"))
308 test_vector_fmul_add(src0, src1, src2);
309 if (check_func(fdsp->vector_fmul_scalar, "vector_fmul_scalar"))
310 test_vector_fmul_scalar(src3, src4);
311 if (check_func(fdsp->vector_fmul_reverse, "vector_fmul_reverse"))
312 test_vector_fmul(src0, src1);
313 if (check_func(fdsp->vector_fmul_window, "vector_fmul_window"))
314 test_vector_fmul_window(src3, src4, src5);
315 report("vector_fmul");
316 if (check_func(fdsp->vector_fmac_scalar, "vector_fmac_scalar"))
317 test_vector_fmac_scalar(src0, src1, src2);
318 report("vector_fmac");
319 if (check_func(fdsp->vector_dmul, "vector_dmul"))
320 test_vector_dmul(dbl_src0, dbl_src1);
321 if (check_func(fdsp->vector_dmul_scalar, "vector_dmul_scalar"))
322 test_vector_dmul_scalar(dbl_src0, dbl_src1);
323 report("vector_dmul");
324 if (check_func(fdsp->vector_dmac_scalar, "vector_dmac_scalar"))
325 test_vector_dmac_scalar(dbl_src0, dbl_src1, dbl_src2);
326 report("vector_dmac");
327 if (check_func(fdsp->butterflies_float, "butterflies_float"))
328 test_butterflies_float(src3, src4);
329 report("butterflies_float");
330 if (check_func(fdsp->scalarproduct_float, "scalarproduct_float"))
331 test_scalarproduct_float(src3, src4);
332 report("scalarproduct_float");