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 "libavutil/mem_internal.h"
20
21 #include "libavcodec/sbrdsp.h"
22 #include <float.h>
23
24 #include "checkasm.h"
25
26 #define randomize(buf, len) do { \
27 int i; \
28 for (i = 0; i < len; i++) { \
29 const INTFLOAT f = (INTFLOAT)rnd() / UINT_MAX; \
30 (buf)[i] = f; \
31 } \
32 } while (0)
33
34 #define EPS 0.0001
35
test_sum64x5(void)36 static void test_sum64x5(void)
37 {
38 LOCAL_ALIGNED_16(INTFLOAT, dst0, [64 + 256]);
39 LOCAL_ALIGNED_16(INTFLOAT, dst1, [64 + 256]);
40
41 declare_func(void, INTFLOAT *z);
42
43 randomize((INTFLOAT *)dst0, 64 + 256);
44 memcpy(dst1, dst0, (64 + 256) * sizeof(INTFLOAT));
45 call_ref(dst0);
46 call_new(dst1);
47 if (!float_near_abs_eps_array(dst0, dst1, EPS, 64 + 256))
48 fail();
49 bench_new(dst1);
50 }
51
test_sum_square(void)52 static void test_sum_square(void)
53 {
54 INTFLOAT res0;
55 INTFLOAT res1;
56 LOCAL_ALIGNED_16(INTFLOAT, src, [256], [2]);
57 double t = 4 * 256;
58
59 declare_func_float(INTFLOAT, INTFLOAT (*x)[2], int n);
60
61 randomize((INTFLOAT *)src, 256 * 2);
62 res0 = call_ref(src, 256);
63 res1 = call_new(src, 256);
64 if (!float_near_abs_eps(res0, res1, t * 2 * FLT_EPSILON))
65 fail();
66 bench_new(src, 256);
67 }
68
test_neg_odd_64(void)69 static void test_neg_odd_64(void)
70 {
71 LOCAL_ALIGNED_16(INTFLOAT, dst0, [64]);
72 LOCAL_ALIGNED_16(INTFLOAT, dst1, [64]);
73
74 declare_func(void, INTFLOAT *x);
75
76 randomize((INTFLOAT *)dst0, 64);
77 memcpy(dst1, dst0, (64) * sizeof(INTFLOAT));
78 call_ref(dst0);
79 call_new(dst1);
80 if (!float_near_abs_eps_array(dst0, dst1, EPS, 64))
81 fail();
82 bench_new(dst1);
83 }
84
test_qmf_pre_shuffle(void)85 static void test_qmf_pre_shuffle(void)
86 {
87 LOCAL_ALIGNED_16(INTFLOAT, dst0, [128]);
88 LOCAL_ALIGNED_16(INTFLOAT, dst1, [128]);
89
90 declare_func(void, INTFLOAT *z);
91
92 randomize((INTFLOAT *)dst0, 128);
93 memcpy(dst1, dst0, (128) * sizeof(INTFLOAT));
94 call_ref(dst0);
95 call_new(dst1);
96 if (!float_near_abs_eps_array(dst0, dst1, EPS, 128))
97 fail();
98 bench_new(dst1);
99 }
100
test_qmf_post_shuffle(void)101 static void test_qmf_post_shuffle(void)
102 {
103 LOCAL_ALIGNED_16(INTFLOAT, src, [64]);
104 LOCAL_ALIGNED_16(INTFLOAT, dst0, [32], [2]);
105 LOCAL_ALIGNED_16(INTFLOAT, dst1, [32], [2]);
106
107 declare_func(void, INTFLOAT W[32][2], const INTFLOAT *z);
108
109 randomize((INTFLOAT *)src, 64);
110 call_ref(dst0, src);
111 call_new(dst1, src);
112 if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 64))
113 fail();
114 bench_new(dst1, src);
115 }
116
test_qmf_deint_neg(void)117 static void test_qmf_deint_neg(void)
118 {
119 LOCAL_ALIGNED_16(INTFLOAT, src, [64]);
120 LOCAL_ALIGNED_16(INTFLOAT, dst0, [64]);
121 LOCAL_ALIGNED_16(INTFLOAT, dst1, [64]);
122
123 declare_func(void, INTFLOAT *v, const INTFLOAT *src);
124
125 randomize((INTFLOAT *)src, 64);
126 call_ref(dst0, src);
127 call_new(dst1, src);
128 if (!float_near_abs_eps_array(dst0, dst1, EPS, 64))
129 fail();
130 bench_new(dst1, src);
131 }
132
test_qmf_deint_bfly(void)133 static void test_qmf_deint_bfly(void)
134 {
135 LOCAL_ALIGNED_16(INTFLOAT, src0, [64]);
136 LOCAL_ALIGNED_16(INTFLOAT, src1, [64]);
137 LOCAL_ALIGNED_16(INTFLOAT, dst0, [128]);
138 LOCAL_ALIGNED_16(INTFLOAT, dst1, [128]);
139
140 declare_func(void, INTFLOAT *v, const INTFLOAT *src0, const INTFLOAT *src1);
141
142 memset(dst0, 0, 128 * sizeof(INTFLOAT));
143 memset(dst1, 0, 128 * sizeof(INTFLOAT));
144
145 randomize((INTFLOAT *)src0, 64);
146 randomize((INTFLOAT *)src1, 64);
147 call_ref(dst0, src0, src1);
148 call_new(dst1, src0, src1);
149 if (!float_near_abs_eps_array(dst0, dst1, EPS, 128))
150 fail();
151 bench_new(dst1, src0, src1);
152 }
153
test_autocorrelate(void)154 static void test_autocorrelate(void)
155 {
156 LOCAL_ALIGNED_16(INTFLOAT, src, [40], [2]);
157 LOCAL_ALIGNED_16(INTFLOAT, dst0, [3], [2][2]);
158 LOCAL_ALIGNED_16(INTFLOAT, dst1, [3], [2][2]);
159
160 declare_func(void, const INTFLOAT x[40][2], INTFLOAT phi[3][2][2]);
161
162 memset(dst0, 0, 3 * 2 * 2 * sizeof(INTFLOAT));
163 memset(dst1, 0, 3 * 2 * 2 * sizeof(INTFLOAT));
164
165 randomize((INTFLOAT *)src, 80);
166 call_ref(src, dst0);
167 call_new(src, dst1);
168 if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 3 * 2 * 2))
169 fail();
170 bench_new(src, dst1);
171 }
172
test_hf_gen(void)173 static void test_hf_gen(void)
174 {
175 LOCAL_ALIGNED_16(INTFLOAT, low, [128], [2]);
176 LOCAL_ALIGNED_16(INTFLOAT, alpha0, [2]);
177 LOCAL_ALIGNED_16(INTFLOAT, alpha1, [2]);
178 LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
179 LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
180 INTFLOAT bw = (INTFLOAT)rnd() / UINT_MAX;
181 int i;
182
183 declare_func(void, INTFLOAT (*X_high)[2], const INTFLOAT (*X_low)[2],
184 const INTFLOAT alpha0[2], const INTFLOAT alpha1[2],
185 INTFLOAT bw, int start, int end);
186
187 randomize((INTFLOAT *)low, 128 * 2);
188 randomize((INTFLOAT *)alpha0, 2);
189 randomize((INTFLOAT *)alpha1, 2);
190 for (i = 2; i < 64; i += 2) {
191 memset(dst0, 0, 128 * 2 * sizeof(INTFLOAT));
192 memset(dst1, 0, 128 * 2 * sizeof(INTFLOAT));
193 call_ref(dst0, low, alpha0, alpha1, 0.0, i, 128);
194 call_new(dst1, low, alpha0, alpha1, 0.0, i, 128);
195 if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 128 * 2))
196 fail();
197 bench_new(dst1, low, alpha0, alpha1, bw, i, 128);
198 }
199 }
200
test_hf_g_filt(void)201 static void test_hf_g_filt(void)
202 {
203 LOCAL_ALIGNED_16(INTFLOAT, high, [128], [40][2]);
204 LOCAL_ALIGNED_16(INTFLOAT, g_filt, [128]);
205 LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
206 LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
207
208 declare_func(void, INTFLOAT (*Y)[2], const INTFLOAT (*X_high)[40][2],
209 const INTFLOAT *g_filt, int m_max, intptr_t ixh);
210
211 randomize((INTFLOAT *)high, 128 * 40 * 2);
212 randomize((INTFLOAT *)g_filt, 128);
213
214 call_ref(dst0, high, g_filt, 128, 20);
215 call_new(dst1, high, g_filt, 128, 20);
216 if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 128 * 2))
217 fail();
218 bench_new(dst1, high, g_filt, 128, 20);
219 }
220
test_hf_apply_noise(const SBRDSPContext * sbrdsp)221 static void test_hf_apply_noise(const SBRDSPContext *sbrdsp)
222 {
223 LOCAL_ALIGNED_16(AAC_FLOAT, s_m, [128]);
224 LOCAL_ALIGNED_16(AAC_FLOAT, q_filt, [128]);
225 LOCAL_ALIGNED_16(INTFLOAT, ref, [128], [2]);
226 LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
227 LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
228 int noise = 0x2a;
229 int i, j;
230
231 declare_func(void, INTFLOAT (*Y)[2], const AAC_FLOAT *s_m,
232 const AAC_FLOAT *q_filt, int noise,
233 int kx, int m_max);
234
235 randomize((INTFLOAT *)ref, 128 * 2);
236 randomize((INTFLOAT *)s_m, 128);
237 randomize((INTFLOAT *)q_filt, 128);
238
239 for (i = 0; i < 4; i++) {
240 if (check_func(sbrdsp->hf_apply_noise[i], "hf_apply_noise_%d", i)) {
241 for (j = 0; j < 2; j++) {
242 memcpy(dst0, ref, 128 * 2 * sizeof(INTFLOAT));
243 memcpy(dst1, ref, 128 * 2 * sizeof(INTFLOAT));
244 call_ref(dst0, s_m, q_filt, noise, j, 128);
245 call_new(dst1, s_m, q_filt, noise, j, 128);
246 if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 128 * 2))
247 fail();
248 bench_new(dst1, s_m, q_filt, noise, j, 128);
249 }
250 }
251 }
252 }
253
checkasm_check_sbrdsp(void)254 void checkasm_check_sbrdsp(void)
255 {
256 SBRDSPContext sbrdsp;
257
258 ff_sbrdsp_init(&sbrdsp);
259
260 if (check_func(sbrdsp.sum64x5, "sum64x5"))
261 test_sum64x5();
262 report("sum64x5");
263
264 if (check_func(sbrdsp.sum_square, "sum_square"))
265 test_sum_square();
266 report("sum_square");
267
268 if (check_func(sbrdsp.neg_odd_64, "neg_odd_64"))
269 test_neg_odd_64();
270 report("neg_odd_64");
271
272 if (check_func(sbrdsp.qmf_pre_shuffle, "qmf_pre_shuffle"))
273 test_qmf_pre_shuffle();
274 report("qmf_pre_shuffle");
275
276 if (check_func(sbrdsp.qmf_post_shuffle, "qmf_post_shuffle"))
277 test_qmf_post_shuffle();
278 report("qmf_post_shuffle");
279
280 if (check_func(sbrdsp.qmf_deint_neg, "qmf_deint_neg"))
281 test_qmf_deint_neg();
282 report("qmf_deint_neg");
283
284 if (check_func(sbrdsp.qmf_deint_bfly, "qmf_deint_bfly"))
285 test_qmf_deint_bfly();
286 report("qmf_deint_bfly");
287
288 if (check_func(sbrdsp.autocorrelate, "autocorrelate"))
289 test_autocorrelate();
290 report("autocorrelate");
291
292 if (check_func(sbrdsp.hf_gen, "hf_gen"))
293 test_hf_gen();
294 report("hf_gen");
295
296 if (check_func(sbrdsp.hf_g_filt, "hf_g_filt"))
297 test_hf_g_filt();
298 report("hf_g_filt");
299
300 test_hf_apply_noise(&sbrdsp);
301 report("hf_apply_noise");
302 }
303