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 #define ARBITRARY_FMUL_ADD_CONST 0.005
65 static void test_vector_fmul_add(const float *src0, const float *src1, const float *src2)
67 LOCAL_ALIGNED_32(float, cdst, [LEN]);
68 LOCAL_ALIGNED_32(float, odst, [LEN]);
71 declare_func(void, float *dst, const float *src0, const float *src1,
72 const float *src2, int len);
74 call_ref(cdst, src0, src1, src2, LEN);
75 call_new(odst, src0, src1, src2, LEN);
76 for (i = 0; i < LEN; i++) {
77 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMUL_ADD_CONST)) {
78 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
79 i, cdst[i], odst[i], cdst[i] - odst[i]);
84 bench_new(odst, src0, src1, src2, LEN);
87 static void test_vector_fmul_scalar(const float *src0, const float *src1)
89 LOCAL_ALIGNED_16(float, cdst, [LEN]);
90 LOCAL_ALIGNED_16(float, odst, [LEN]);
93 declare_func(void, float *dst, const float *src, float mul, int len);
95 call_ref(cdst, src0, src1[0], LEN);
96 call_new(odst, src0, src1[0], LEN);
97 for (i = 0; i < LEN; i++) {
98 if (!float_near_abs_eps(cdst[i], odst[i], FLT_EPSILON)) {
99 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
100 i, cdst[i], odst[i], cdst[i] - odst[i]);
105 bench_new(odst, src0, src1[0], LEN);
108 #define ARBITRARY_FMUL_WINDOW_CONST 0.008
109 static void test_vector_fmul_window(const float *src0, const float *src1, const float *win)
111 LOCAL_ALIGNED_16(float, cdst, [LEN]);
112 LOCAL_ALIGNED_16(float, odst, [LEN]);
115 declare_func(void, float *dst, const float *src0, const float *src1,
116 const float *win, int len);
118 call_ref(cdst, src0, src1, win, LEN / 2);
119 call_new(odst, src0, src1, win, LEN / 2);
120 for (i = 0; i < LEN; i++) {
121 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMUL_WINDOW_CONST)) {
122 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
123 i, cdst[i], odst[i], cdst[i] - odst[i]);
128 bench_new(odst, src0, src1, win, LEN / 2);
131 #define ARBITRARY_FMAC_SCALAR_CONST 0.005
132 static void test_vector_fmac_scalar(const float *src0, const float *src1, const float *src2)
134 LOCAL_ALIGNED_32(float, cdst, [LEN]);
135 LOCAL_ALIGNED_32(float, odst, [LEN]);
138 declare_func(void, float *dst, const float *src, float mul, int len);
140 memcpy(cdst, src2, LEN * sizeof(*src2));
141 memcpy(odst, src2, LEN * sizeof(*src2));
143 call_ref(cdst, src0, src1[0], LEN);
144 call_new(odst, src0, src1[0], LEN);
145 for (i = 0; i < LEN; i++) {
146 if (!float_near_abs_eps(cdst[i], odst[i], ARBITRARY_FMAC_SCALAR_CONST)) {
147 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
148 i, cdst[i], odst[i], cdst[i] - odst[i]);
153 memcpy(odst, src2, LEN * sizeof(*src2));
154 bench_new(odst, src0, src1[0], LEN);
157 static void test_vector_dmul_scalar(const double *src0, const double *src1)
159 LOCAL_ALIGNED_32(double, cdst, [LEN]);
160 LOCAL_ALIGNED_32(double, odst, [LEN]);
163 declare_func(void, double *dst, const double *src, double mul, int len);
165 call_ref(cdst, src0, src1[0], LEN);
166 call_new(odst, src0, src1[0], LEN);
167 for (i = 0; i < LEN; i++) {
168 double t = fabs(src1[0]) + fabs(src0[i]) + fabs(src1[0] * src0[i]) + 1.0;
169 if (!double_near_abs_eps(cdst[i], odst[i], t * 2 * DBL_EPSILON)) {
170 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n", i,
171 cdst[i], odst[i], cdst[i] - odst[i]);
176 bench_new(odst, src0, src1[0], LEN);
179 #define ARBITRARY_DMAC_SCALAR_CONST 0.005
180 static void test_vector_dmac_scalar(const double *src0, const double *src1, const double *src2)
182 LOCAL_ALIGNED_32(double, cdst, [LEN]);
183 LOCAL_ALIGNED_32(double, odst, [LEN]);
186 declare_func(void, double *dst, const double *src, double mul, int len);
188 memcpy(cdst, src2, LEN * sizeof(*src2));
189 memcpy(odst, src2, LEN * sizeof(*src2));
190 call_ref(cdst, src0, src1[0], LEN);
191 call_new(odst, src0, src1[0], LEN);
192 for (i = 0; i < LEN; i++) {
193 if (!double_near_abs_eps(cdst[i], odst[i], ARBITRARY_DMAC_SCALAR_CONST)) {
194 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
195 i, cdst[i], odst[i], cdst[i] - odst[i]);
200 memcpy(odst, src2, LEN * sizeof(*src2));
201 bench_new(odst, src0, src1[0], LEN);
204 static void test_butterflies_float(const float *src0, const float *src1)
206 LOCAL_ALIGNED_16(float, cdst, [LEN]);
207 LOCAL_ALIGNED_16(float, odst, [LEN]);
208 LOCAL_ALIGNED_16(float, cdst1, [LEN]);
209 LOCAL_ALIGNED_16(float, odst1, [LEN]);
212 declare_func(void, float *av_restrict src0, float *av_restrict src1,
215 memcpy(cdst, src0, LEN * sizeof(*src0));
216 memcpy(cdst1, src1, LEN * sizeof(*src1));
217 memcpy(odst, src0, LEN * sizeof(*src0));
218 memcpy(odst1, src1, LEN * sizeof(*src1));
220 call_ref(cdst, cdst1, LEN);
221 call_new(odst, odst1, LEN);
222 for (i = 0; i < LEN; i++) {
223 if (!float_near_abs_eps(cdst[i], odst[i], FLT_EPSILON) ||
224 !float_near_abs_eps(cdst1[i], odst1[i], FLT_EPSILON)) {
225 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
226 i, cdst[i], odst[i], cdst[i] - odst[i]);
227 fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
228 i, cdst1[i], odst1[i], cdst1[i] - odst1[i]);
233 memcpy(odst, src0, LEN * sizeof(*src0));
234 memcpy(odst1, src1, LEN * sizeof(*src1));
235 bench_new(odst, odst1, LEN);
238 #define ARBITRARY_SCALARPRODUCT_CONST 0.2
239 static void test_scalarproduct_float(const float *src0, const float *src1)
243 declare_func_float(float, const float *src0, const float *src1, int len);
245 cprod = call_ref(src0, src1, LEN);
246 oprod = call_new(src0, src1, LEN);
247 if (!float_near_abs_eps(cprod, oprod, ARBITRARY_SCALARPRODUCT_CONST)) {
248 fprintf(stderr, "%- .12f - %- .12f = % .12g\n",
249 cprod, oprod, cprod - oprod);
252 bench_new(src0, src1, LEN);
255 void checkasm_check_float_dsp(void)
257 LOCAL_ALIGNED_32(float, src0, [LEN]);
258 LOCAL_ALIGNED_32(float, src1, [LEN]);
259 LOCAL_ALIGNED_32(float, src2, [LEN]);
260 LOCAL_ALIGNED_16(float, src3, [LEN]);
261 LOCAL_ALIGNED_16(float, src4, [LEN]);
262 LOCAL_ALIGNED_16(float, src5, [LEN]);
263 LOCAL_ALIGNED_32(double, dbl_src0, [LEN]);
264 LOCAL_ALIGNED_32(double, dbl_src1, [LEN]);
265 LOCAL_ALIGNED_32(double, dbl_src2, [LEN]);
266 AVFloatDSPContext *fdsp = avpriv_float_dsp_alloc(1);
269 fprintf(stderr, "floatdsp: Out of memory error\n");
273 randomize_buffer(src0);
274 randomize_buffer(src1);
275 randomize_buffer(src2);
276 randomize_buffer(src3);
277 randomize_buffer(src4);
278 randomize_buffer(src5);
279 randomize_buffer(dbl_src0);
280 randomize_buffer(dbl_src1);
281 randomize_buffer(dbl_src2);
283 if (check_func(fdsp->vector_fmul, "vector_fmul"))
284 test_vector_fmul(src0, src1);
285 if (check_func(fdsp->vector_fmul_add, "vector_fmul_add"))
286 test_vector_fmul_add(src0, src1, src2);
287 if (check_func(fdsp->vector_fmul_scalar, "vector_fmul_scalar"))
288 test_vector_fmul_scalar(src3, src4);
289 if (check_func(fdsp->vector_fmul_reverse, "vector_fmul_reverse"))
290 test_vector_fmul(src0, src1);
291 if (check_func(fdsp->vector_fmul_window, "vector_fmul_window"))
292 test_vector_fmul_window(src3, src4, src5);
293 report("vector_fmul");
294 if (check_func(fdsp->vector_fmac_scalar, "vector_fmac_scalar"))
295 test_vector_fmac_scalar(src0, src1, src2);
296 report("vector_fmac");
297 if (check_func(fdsp->vector_dmul_scalar, "vector_dmul_scalar"))
298 test_vector_dmul_scalar(dbl_src0, dbl_src1);
299 report("vector_dmul");
300 if (check_func(fdsp->vector_dmac_scalar, "vector_dmac_scalar"))
301 test_vector_dmac_scalar(dbl_src0, dbl_src1, dbl_src2);
302 report("vector_dmac");
303 if (check_func(fdsp->butterflies_float, "butterflies_float"))
304 test_butterflies_float(src3, src4);
305 report("butterflies_float");
306 if (check_func(fdsp->scalarproduct_float, "scalarproduct_float"))
307 test_scalarproduct_float(src3, src4);
308 report("scalarproduct_float");