Line | Branch | Exec | Source |
---|---|---|---|
1 | /* | ||
2 | * Copyright (c) 2015 Ronald S. Bultje <rsbultje@gmail.com> | ||
3 | * | ||
4 | * This file is part of FFmpeg. | ||
5 | * | ||
6 | * FFmpeg is free software; you can redistribute it and/or modify | ||
7 | * it under the terms of the GNU General Public License as published by | ||
8 | * the Free Software Foundation; either version 2 of the License, or | ||
9 | * (at your option) any later version. | ||
10 | * | ||
11 | * FFmpeg is distributed in the hope that it will be useful, | ||
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
14 | * GNU General Public License for more details. | ||
15 | * | ||
16 | * You should have received a copy of the GNU General Public License along | ||
17 | * with FFmpeg; if not, write to the Free Software Foundation, Inc., | ||
18 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | ||
19 | */ | ||
20 | |||
21 | #include <math.h> | ||
22 | #include <string.h> | ||
23 | #include "checkasm.h" | ||
24 | #include "libavcodec/vp9data.h" | ||
25 | #include "libavcodec/vp9.h" | ||
26 | #include "libavutil/common.h" | ||
27 | #include "libavutil/emms.h" | ||
28 | #include "libavutil/internal.h" | ||
29 | #include "libavutil/intreadwrite.h" | ||
30 | #include "libavutil/mathematics.h" | ||
31 | #include "libavutil/mem_internal.h" | ||
32 | |||
33 | static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff }; | ||
34 | #define SIZEOF_PIXEL ((bit_depth + 7) / 8) | ||
35 | |||
36 | #define randomize_buffers() \ | ||
37 | do { \ | ||
38 | uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \ | ||
39 | int k; \ | ||
40 | for (k = -4; k < SIZEOF_PIXEL * FFMAX(8, size); k += 4) { \ | ||
41 | uint32_t r = rnd() & mask; \ | ||
42 | AV_WN32A(a + k, r); \ | ||
43 | } \ | ||
44 | for (k = 0; k < size * SIZEOF_PIXEL; k += 4) { \ | ||
45 | uint32_t r = rnd() & mask; \ | ||
46 | AV_WN32A(l + k, r); \ | ||
47 | } \ | ||
48 | } while (0) | ||
49 | |||
50 | 13 | static void check_ipred(void) | |
51 | { | ||
52 | 13 | LOCAL_ALIGNED_32(uint8_t, a_buf, [64 * 2]); | |
53 | 13 | uint8_t *a = &a_buf[32 * 2]; | |
54 | 13 | LOCAL_ALIGNED_32(uint8_t, l, [32 * 2]); | |
55 | 13 | LOCAL_ALIGNED_32(uint8_t, dst0, [32 * 32 * 2]); | |
56 | 13 | LOCAL_ALIGNED_32(uint8_t, dst1, [32 * 32 * 2]); | |
57 | VP9DSPContext dsp; | ||
58 | int tx, mode, bit_depth; | ||
59 |
2/2✓ Branch 1 taken 12 times.
✓ Branch 2 taken 1 times.
|
13 | declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, void, uint8_t *dst, ptrdiff_t stride, |
60 | const uint8_t *left, const uint8_t *top); | ||
61 | static const char *const mode_names[N_INTRA_PRED_MODES] = { | ||
62 | [VERT_PRED] = "vert", | ||
63 | [HOR_PRED] = "hor", | ||
64 | [DC_PRED] = "dc", | ||
65 | [DIAG_DOWN_LEFT_PRED] = "diag_downleft", | ||
66 | [DIAG_DOWN_RIGHT_PRED] = "diag_downright", | ||
67 | [VERT_RIGHT_PRED] = "vert_right", | ||
68 | [HOR_DOWN_PRED] = "hor_down", | ||
69 | [VERT_LEFT_PRED] = "vert_left", | ||
70 | [HOR_UP_PRED] = "hor_up", | ||
71 | [TM_VP8_PRED] = "tm", | ||
72 | [LEFT_DC_PRED] = "dc_left", | ||
73 | [TOP_DC_PRED] = "dc_top", | ||
74 | [DC_128_PRED] = "dc_128", | ||
75 | [DC_127_PRED] = "dc_127", | ||
76 | [DC_129_PRED] = "dc_129", | ||
77 | }; | ||
78 | |||
79 |
2/2✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
|
52 | for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) { |
80 | 39 | ff_vp9dsp_init(&dsp, bit_depth, 0); | |
81 |
2/2✓ Branch 0 taken 156 times.
✓ Branch 1 taken 39 times.
|
195 | for (tx = 0; tx < 4; tx++) { |
82 | 156 | int size = 4 << tx; | |
83 | |||
84 |
2/2✓ Branch 0 taken 2340 times.
✓ Branch 1 taken 156 times.
|
2496 | for (mode = 0; mode < N_INTRA_PRED_MODES; mode++) { |
85 |
2/2✓ Branch 3 taken 503 times.
✓ Branch 4 taken 1837 times.
|
2340 | if (check_func(dsp.intra_pred[tx][mode], "vp9_%s_%dx%d_%dbpp", |
86 | mode_names[mode], size, size, bit_depth)) { | ||
87 |
4/4✓ Branch 1 taken 3889 times.
✓ Branch 2 taken 503 times.
✓ Branch 4 taken 3195 times.
✓ Branch 5 taken 503 times.
|
7587 | randomize_buffers(); |
88 | 503 | call_ref(dst0, size * SIZEOF_PIXEL, l, a); | |
89 | 503 | call_new(dst1, size * SIZEOF_PIXEL, l, a); | |
90 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 503 times.
|
503 | if (memcmp(dst0, dst1, size * size * SIZEOF_PIXEL)) |
91 | ✗ | fail(); | |
92 |
1/8✗ Branch 1 not taken.
✓ Branch 2 taken 503 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
|
503 | bench_new(dst1, size * SIZEOF_PIXEL,l, a); |
93 | } | ||
94 | } | ||
95 | } | ||
96 | } | ||
97 | 13 | report("ipred"); | |
98 | 13 | } | |
99 | |||
100 | #undef randomize_buffers | ||
101 | |||
102 | #define randomize_buffers() \ | ||
103 | do { \ | ||
104 | uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \ | ||
105 | for (y = 0; y < sz; y++) { \ | ||
106 | for (x = 0; x < sz * SIZEOF_PIXEL; x += 4) { \ | ||
107 | uint32_t r = rnd() & mask; \ | ||
108 | AV_WN32A(dst + y * sz * SIZEOF_PIXEL + x, r); \ | ||
109 | AV_WN32A(src + y * sz * SIZEOF_PIXEL + x, rnd() & mask); \ | ||
110 | } \ | ||
111 | for (x = 0; x < sz; x++) { \ | ||
112 | if (bit_depth == 8) { \ | ||
113 | coef[y * sz + x] = src[y * sz + x] - dst[y * sz + x]; \ | ||
114 | } else { \ | ||
115 | ((int32_t *) coef)[y * sz + x] = \ | ||
116 | ((uint16_t *) src)[y * sz + x] - \ | ||
117 | ((uint16_t *) dst)[y * sz + x]; \ | ||
118 | } \ | ||
119 | } \ | ||
120 | } \ | ||
121 | } while(0) | ||
122 | |||
123 | // wht function copied from libvpx | ||
124 | 48 | static void fwht_1d(double *out, const double *in, int sz) | |
125 | { | ||
126 | 48 | double t0 = in[0] + in[1]; | |
127 | 48 | double t3 = in[3] - in[2]; | |
128 | 48 | double t4 = trunc((t0 - t3) * 0.5); | |
129 | 48 | double t1 = t4 - in[1]; | |
130 | 48 | double t2 = t4 - in[2]; | |
131 | |||
132 | 48 | out[0] = t0 - t2; | |
133 | 48 | out[1] = t2; | |
134 | 48 | out[2] = t3 + t1; | |
135 | 48 | out[3] = t1; | |
136 | 48 | } | |
137 | |||
138 | // standard DCT-II | ||
139 | 8672 | static void fdct_1d(double *out, const double *in, int sz) | |
140 | { | ||
141 | int k, n; | ||
142 | |||
143 |
2/2✓ Branch 0 taken 222720 times.
✓ Branch 1 taken 8672 times.
|
231392 | for (k = 0; k < sz; k++) { |
144 | 222720 | out[k] = 0.0; | |
145 |
2/2✓ Branch 0 taken 6459392 times.
✓ Branch 1 taken 222720 times.
|
6682112 | for (n = 0; n < sz; n++) |
146 | 6459392 | out[k] += in[n] * cos(M_PI * (2 * n + 1) * k / (sz * 2.0)); | |
147 | } | ||
148 | 8672 | out[0] *= M_SQRT1_2; | |
149 | 8672 | } | |
150 | |||
151 | // see "Towards jointly optimal spatial prediction and adaptive transform in | ||
152 | // video/image coding", by J. Han, A. Saxena, and K. Rose | ||
153 | // IEEE Proc. ICASSP, pp. 726-729, Mar. 2010. | ||
154 | 128 | static void fadst4_1d(double *out, const double *in, int sz) | |
155 | { | ||
156 | int k, n; | ||
157 | |||
158 |
2/2✓ Branch 0 taken 512 times.
✓ Branch 1 taken 128 times.
|
640 | for (k = 0; k < sz; k++) { |
159 | 512 | out[k] = 0.0; | |
160 |
2/2✓ Branch 0 taken 2048 times.
✓ Branch 1 taken 512 times.
|
2560 | for (n = 0; n < sz; n++) |
161 | 2048 | out[k] += in[n] * sin(M_PI * (n + 1) * (2 * k + 1) / (sz * 2.0 + 1.0)); | |
162 | } | ||
163 | 128 | } | |
164 | |||
165 | // see "A Butterfly Structured Design of The Hybrid Transform Coding Scheme", | ||
166 | // by Jingning Han, Yaowu Xu, and Debargha Mukherjee | ||
167 | // http://static.googleusercontent.com/media/research.google.com/en//pubs/archive/41418.pdf | ||
168 | 832 | static void fadst_1d(double *out, const double *in, int sz) | |
169 | { | ||
170 | int k, n; | ||
171 | |||
172 |
2/2✓ Branch 0 taken 11264 times.
✓ Branch 1 taken 832 times.
|
12096 | for (k = 0; k < sz; k++) { |
173 | 11264 | out[k] = 0.0; | |
174 |
2/2✓ Branch 0 taken 163840 times.
✓ Branch 1 taken 11264 times.
|
175104 | for (n = 0; n < sz; n++) |
175 | 163840 | out[k] += in[n] * sin(M_PI * (2 * n + 1) * (2 * k + 1) / (sz * 4.0)); | |
176 | } | ||
177 | 832 | } | |
178 | |||
179 | typedef void (*ftx1d_fn)(double *out, const double *in, int sz); | ||
180 | 281 | static void ftx_2d(double *out, const double *in, enum TxfmMode tx, | |
181 | enum TxfmType txtp, int sz) | ||
182 | { | ||
183 | static const double scaling_factors[5][4] = { | ||
184 | { 4.0, 16.0 * M_SQRT1_2 / 3.0, 16.0 * M_SQRT1_2 / 3.0, 32.0 / 9.0 }, | ||
185 | { 2.0, 2.0, 2.0, 2.0 }, | ||
186 | { 1.0, 1.0, 1.0, 1.0 }, | ||
187 | { 0.25 }, | ||
188 | { 4.0 } | ||
189 | }; | ||
190 | static const ftx1d_fn ftx1d_tbl[5][4][2] = { | ||
191 | { | ||
192 | { fdct_1d, fdct_1d }, | ||
193 | { fadst4_1d, fdct_1d }, | ||
194 | { fdct_1d, fadst4_1d }, | ||
195 | { fadst4_1d, fadst4_1d }, | ||
196 | }, { | ||
197 | { fdct_1d, fdct_1d }, | ||
198 | { fadst_1d, fdct_1d }, | ||
199 | { fdct_1d, fadst_1d }, | ||
200 | { fadst_1d, fadst_1d }, | ||
201 | }, { | ||
202 | { fdct_1d, fdct_1d }, | ||
203 | { fadst_1d, fdct_1d }, | ||
204 | { fdct_1d, fadst_1d }, | ||
205 | { fadst_1d, fadst_1d }, | ||
206 | }, { | ||
207 | { fdct_1d, fdct_1d }, | ||
208 | }, { | ||
209 | { fwht_1d, fwht_1d }, | ||
210 | }, | ||
211 | }; | ||
212 | double temp[1024]; | ||
213 | 281 | double scaling_factor = scaling_factors[tx][txtp]; | |
214 | int i, j; | ||
215 | |||
216 | // cols | ||
217 |
2/2✓ Branch 0 taken 4840 times.
✓ Branch 1 taken 281 times.
|
5121 | for (i = 0; i < sz; ++i) { |
218 | double temp_out[32]; | ||
219 | |||
220 | 4840 | ftx1d_tbl[tx][txtp][0](temp_out, &in[i * sz], sz); | |
221 | // scale and transpose | ||
222 |
2/2✓ Branch 0 taken 117344 times.
✓ Branch 1 taken 4840 times.
|
122184 | for (j = 0; j < sz; ++j) |
223 | 117344 | temp[j * sz + i] = temp_out[j] * scaling_factor; | |
224 | } | ||
225 | |||
226 | // rows | ||
227 |
2/2✓ Branch 0 taken 4840 times.
✓ Branch 1 taken 281 times.
|
5121 | for (i = 0; i < sz; i++) |
228 | 4840 | ftx1d_tbl[tx][txtp][1](&out[i * sz], &temp[i * sz], sz); | |
229 | 281 | } | |
230 | |||
231 | 281 | static void ftx(int16_t *buf, enum TxfmMode tx, | |
232 | enum TxfmType txtp, int sz, int bit_depth) | ||
233 | { | ||
234 | double ind[1024], outd[1024]; | ||
235 | int n; | ||
236 | |||
237 | 281 | emms_c(); | |
238 |
2/2✓ Branch 0 taken 117344 times.
✓ Branch 1 taken 281 times.
|
117625 | for (n = 0; n < sz * sz; n++) { |
239 |
2/2✓ Branch 0 taken 64832 times.
✓ Branch 1 taken 52512 times.
|
117344 | if (bit_depth == 8) |
240 | 64832 | ind[n] = buf[n]; | |
241 | else | ||
242 | 52512 | ind[n] = ((int32_t *) buf)[n]; | |
243 | } | ||
244 | 281 | ftx_2d(outd, ind, tx, txtp, sz); | |
245 |
2/2✓ Branch 0 taken 117344 times.
✓ Branch 1 taken 281 times.
|
117625 | for (n = 0; n < sz * sz; n++) { |
246 |
2/2✓ Branch 0 taken 64832 times.
✓ Branch 1 taken 52512 times.
|
117344 | if (bit_depth == 8) |
247 | 64832 | buf[n] = lrint(outd[n]); | |
248 | else | ||
249 | 52512 | ((int32_t *) buf)[n] = lrint(outd[n]); | |
250 | } | ||
251 | 281 | } | |
252 | |||
253 | 166 | static int copy_subcoefs(int16_t *out, const int16_t *in, enum TxfmMode tx, | |
254 | enum TxfmType txtp, int sz, int sub, int bit_depth) | ||
255 | { | ||
256 | // copy the topleft coefficients such that the return value (being the | ||
257 | // coefficient scantable index for the eob token) guarantees that only | ||
258 | // the topleft $sub out of $sz (where $sz >= $sub) coefficients in both | ||
259 | // dimensions are non-zero. This leads to braching to specific optimized | ||
260 | // simd versions (e.g. dc-only) so that we get full asm coverage in this | ||
261 | // test | ||
262 | |||
263 | int n; | ||
264 | 166 | const int16_t *scan = ff_vp9_scans[tx][txtp]; | |
265 | int eob; | ||
266 | |||
267 |
1/2✓ Branch 0 taken 13087 times.
✗ Branch 1 not taken.
|
13087 | for (n = 0; n < sz * sz; n++) { |
268 | 13087 | int rc = scan[n], rcx = rc % sz, rcy = rc / sz; | |
269 | |||
270 | // find eob for this sub-idct | ||
271 |
3/4✓ Branch 0 taken 12921 times.
✓ Branch 1 taken 166 times.
✓ Branch 2 taken 12921 times.
✗ Branch 3 not taken.
|
13087 | if (rcx >= sub || rcy >= sub) |
272 | break; | ||
273 | |||
274 | // copy coef | ||
275 |
2/2✓ Branch 0 taken 7164 times.
✓ Branch 1 taken 5757 times.
|
12921 | if (bit_depth == 8) { |
276 | 7164 | out[rc] = in[rc]; | |
277 | } else { | ||
278 | 5757 | AV_COPY32(&out[rc * 2], &in[rc * 2]); | |
279 | } | ||
280 | } | ||
281 | |||
282 | 166 | eob = n; | |
283 | |||
284 |
2/2✓ Branch 0 taken 83335 times.
✓ Branch 1 taken 166 times.
|
83501 | for (; n < sz * sz; n++) { |
285 | 83335 | int rc = scan[n]; | |
286 | |||
287 | // zero | ||
288 |
2/2✓ Branch 0 taken 46180 times.
✓ Branch 1 taken 37155 times.
|
83335 | if (bit_depth == 8) { |
289 | 46180 | out[rc] = 0; | |
290 | } else { | ||
291 | 37155 | AV_ZERO32(&out[rc * 2]); | |
292 | } | ||
293 | } | ||
294 | |||
295 | 166 | return eob; | |
296 | } | ||
297 | |||
298 | 562 | static int is_zero(const int16_t *c, int sz) | |
299 | { | ||
300 | int n; | ||
301 | |||
302 |
2/2✓ Branch 0 taken 169856 times.
✓ Branch 1 taken 562 times.
|
170418 | for (n = 0; n < sz / sizeof(int16_t); n += 2) |
303 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 169856 times.
|
169856 | if (AV_RN32A(&c[n])) |
304 | ✗ | return 0; | |
305 | |||
306 | 562 | return 1; | |
307 | } | ||
308 | |||
309 | #define SIZEOF_COEF (2 * ((bit_depth + 7) / 8)) | ||
310 | |||
311 | 13 | static void check_itxfm(void) | |
312 | { | ||
313 | 13 | LOCAL_ALIGNED_32(uint8_t, src, [32 * 32 * 2]); | |
314 | 13 | LOCAL_ALIGNED_32(uint8_t, dst, [32 * 32 * 2]); | |
315 | 13 | LOCAL_ALIGNED_32(uint8_t, dst0, [32 * 32 * 2]); | |
316 | 13 | LOCAL_ALIGNED_32(uint8_t, dst1, [32 * 32 * 2]); | |
317 | 13 | LOCAL_ALIGNED_32(int16_t, coef, [32 * 32 * 2]); | |
318 | 13 | LOCAL_ALIGNED_32(int16_t, subcoef0, [32 * 32 * 2]); | |
319 | 13 | LOCAL_ALIGNED_32(int16_t, subcoef1, [32 * 32 * 2]); | |
320 |
2/2✓ Branch 1 taken 12 times.
✓ Branch 2 taken 1 times.
|
13 | declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, void, uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob); |
321 | VP9DSPContext dsp; | ||
322 | int y, x, tx, txtp, bit_depth, sub; | ||
323 | static const char *const txtp_types[N_TXFM_TYPES] = { | ||
324 | [DCT_DCT] = "dct_dct", [DCT_ADST] = "adst_dct", | ||
325 | [ADST_DCT] = "dct_adst", [ADST_ADST] = "adst_adst" | ||
326 | }; | ||
327 | |||
328 |
2/2✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
|
52 | for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) { |
329 | 39 | ff_vp9dsp_init(&dsp, bit_depth, 0); | |
330 | |||
331 |
2/2✓ Branch 0 taken 195 times.
✓ Branch 1 taken 39 times.
|
234 | for (tx = TX_4X4; tx <= N_TXFM_SIZES /* 4 = lossless */; tx++) { |
332 | 195 | int sz = 4 << (tx & 3); | |
333 |
2/2✓ Branch 0 taken 117 times.
✓ Branch 1 taken 78 times.
|
195 | int n_txtps = tx < TX_32X32 ? N_TXFM_TYPES : 1; |
334 | |||
335 |
2/2✓ Branch 0 taken 546 times.
✓ Branch 1 taken 195 times.
|
741 | for (txtp = 0; txtp < n_txtps; txtp++) { |
336 | // skip testing sub-IDCTs for WHT or ADST since they don't | ||
337 | // implement it in any of the SIMD functions. If they do, | ||
338 | // consider changing this to ensure we have complete test | ||
339 | // coverage. Test sub=1 for dc-only, then 2, 4, 8, 12, etc, | ||
340 | // since the arm version can distinguish them at that level. | ||
341 |
6/6✓ Branch 0 taken 195 times.
✓ Branch 1 taken 351 times.
✓ Branch 2 taken 39 times.
✓ Branch 3 taken 156 times.
✓ Branch 4 taken 1287 times.
✓ Branch 5 taken 546 times.
|
1833 | for (sub = (txtp == 0 && tx < 4) ? 1 : sz; sub <= sz; |
342 |
2/2✓ Branch 0 taken 312 times.
✓ Branch 1 taken 975 times.
|
1287 | sub < 4 ? (sub <<= 1) : (sub += 4)) { |
343 |
4/4✓ Branch 2 taken 1248 times.
✓ Branch 3 taken 39 times.
✓ Branch 5 taken 281 times.
✓ Branch 6 taken 1006 times.
|
1287 | if (check_func(dsp.itxfm_add[tx][txtp], |
344 | "vp9_inv_%s_%dx%d_sub%d_add_%d", | ||
345 | tx == 4 ? "wht_wht" : txtp_types[txtp], | ||
346 | sz, sz, sub, bit_depth)) { | ||
347 | int eob; | ||
348 | |||
349 |
8/8✓ Branch 2 taken 42464 times.
✓ Branch 3 taken 4840 times.
✓ Branch 4 taken 64832 times.
✓ Branch 5 taken 52512 times.
✓ Branch 6 taken 117344 times.
✓ Branch 7 taken 4840 times.
✓ Branch 8 taken 4840 times.
✓ Branch 9 taken 281 times.
|
164929 | randomize_buffers(); |
350 | 281 | ftx(coef, tx, txtp, sz, bit_depth); | |
351 | |||
352 |
2/2✓ Branch 0 taken 166 times.
✓ Branch 1 taken 115 times.
|
281 | if (sub < sz) { |
353 | 166 | eob = copy_subcoefs(subcoef0, coef, tx, txtp, | |
354 | sz, sub, bit_depth); | ||
355 | } else { | ||
356 | 115 | eob = sz * sz; | |
357 | 115 | memcpy(subcoef0, coef, sz * sz * SIZEOF_COEF); | |
358 | } | ||
359 | |||
360 | 281 | memcpy(dst0, dst, sz * sz * SIZEOF_PIXEL); | |
361 | 281 | memcpy(dst1, dst, sz * sz * SIZEOF_PIXEL); | |
362 | 281 | memcpy(subcoef1, subcoef0, sz * sz * SIZEOF_COEF); | |
363 | 281 | call_ref(dst0, sz * SIZEOF_PIXEL, subcoef0, eob); | |
364 | 281 | call_new(dst1, sz * SIZEOF_PIXEL, subcoef1, eob); | |
365 |
1/2✓ Branch 0 taken 281 times.
✗ Branch 1 not taken.
|
281 | if (memcmp(dst0, dst1, sz * sz * SIZEOF_PIXEL) || |
366 |
1/2✓ Branch 1 taken 281 times.
✗ Branch 2 not taken.
|
281 | !is_zero(subcoef0, sz * sz * SIZEOF_COEF) || |
367 |
1/2✗ Branch 1 not taken.
✓ Branch 2 taken 281 times.
|
281 | !is_zero(subcoef1, sz * sz * SIZEOF_COEF)) |
368 | ✗ | fail(); | |
369 | |||
370 |
1/8✗ Branch 1 not taken.
✓ Branch 2 taken 281 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
|
281 | bench_new(dst, sz * SIZEOF_PIXEL, coef, eob); |
371 | } | ||
372 | } | ||
373 | } | ||
374 | } | ||
375 | } | ||
376 | 13 | report("itxfm"); | |
377 | 13 | } | |
378 | |||
379 | #undef randomize_buffers | ||
380 | |||
381 | #define setpx(a,b,c) \ | ||
382 | do { \ | ||
383 | if (SIZEOF_PIXEL == 1) { \ | ||
384 | buf0[(a) + (b) * jstride] = av_clip_uint8(c); \ | ||
385 | } else { \ | ||
386 | ((uint16_t *)buf0)[(a) + (b) * jstride] = av_clip_uintp2(c, bit_depth); \ | ||
387 | } \ | ||
388 | } while (0) | ||
389 | |||
390 | // c can be an assignment and must not be put under () | ||
391 | #define setdx(a,b,c,d) setpx(a,b,c-(d)+(rnd()%((d)*2+1))) | ||
392 | #define setsx(a,b,c,d) setdx(a,b,c,(d) << (bit_depth - 8)) | ||
393 | 298 | static void randomize_loopfilter_buffers(int bidx, int lineoff, int str, | |
394 | int bit_depth, int dir, const int *E, | ||
395 | const int *F, const int *H, const int *I, | ||
396 | uint8_t *buf0, uint8_t *buf1) | ||
397 | { | ||
398 | 298 | uint32_t mask = (1 << bit_depth) - 1; | |
399 |
2/2✓ Branch 0 taken 149 times.
✓ Branch 1 taken 149 times.
|
298 | int off = dir ? lineoff : lineoff * 16; |
400 |
2/2✓ Branch 0 taken 149 times.
✓ Branch 1 taken 149 times.
|
298 | int istride = dir ? 1 : 16; |
401 |
2/2✓ Branch 0 taken 149 times.
✓ Branch 1 taken 149 times.
|
298 | int jstride = dir ? str : 1; |
402 | int i, j; | ||
403 |
2/2✓ Branch 0 taken 596 times.
✓ Branch 1 taken 298 times.
|
894 | for (i = 0; i < 2; i++) /* flat16 */ { |
404 | 596 | int idx = off + i * istride, p0, q0; | |
405 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setpx(idx, 0, q0 = rnd() & mask); |
406 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, -1, p0 = q0, E[bidx] >> 2); |
407 |
2/2✓ Branch 0 taken 4172 times.
✓ Branch 1 taken 596 times.
|
4768 | for (j = 1; j < 8; j++) { |
408 |
2/2✓ Branch 0 taken 1260 times.
✓ Branch 1 taken 2912 times.
|
4172 | setsx(idx, -1 - j, p0, F[bidx]); |
409 |
2/2✓ Branch 0 taken 1260 times.
✓ Branch 1 taken 2912 times.
|
4172 | setsx(idx, j, q0, F[bidx]); |
410 | } | ||
411 | } | ||
412 |
2/2✓ Branch 0 taken 596 times.
✓ Branch 1 taken 298 times.
|
894 | for (i = 2; i < 4; i++) /* flat8 */ { |
413 | 596 | int idx = off + i * istride, p0, q0; | |
414 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setpx(idx, 0, q0 = rnd() & mask); |
415 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, -1, p0 = q0, E[bidx] >> 2); |
416 |
2/2✓ Branch 0 taken 1788 times.
✓ Branch 1 taken 596 times.
|
2384 | for (j = 1; j < 4; j++) { |
417 |
2/2✓ Branch 0 taken 540 times.
✓ Branch 1 taken 1248 times.
|
1788 | setsx(idx, -1 - j, p0, F[bidx]); |
418 |
2/2✓ Branch 0 taken 540 times.
✓ Branch 1 taken 1248 times.
|
1788 | setsx(idx, j, q0, F[bidx]); |
419 | } | ||
420 |
2/2✓ Branch 0 taken 2384 times.
✓ Branch 1 taken 596 times.
|
2980 | for (j = 4; j < 8; j++) { |
421 |
2/2✓ Branch 0 taken 720 times.
✓ Branch 1 taken 1664 times.
|
2384 | setpx(idx, -1 - j, rnd() & mask); |
422 |
2/2✓ Branch 0 taken 720 times.
✓ Branch 1 taken 1664 times.
|
2384 | setpx(idx, j, rnd() & mask); |
423 | } | ||
424 | } | ||
425 |
2/2✓ Branch 0 taken 596 times.
✓ Branch 1 taken 298 times.
|
894 | for (i = 4; i < 6; i++) /* regular */ { |
426 | 596 | int idx = off + i * istride, p2, p1, p0, q0, q1, q2; | |
427 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setpx(idx, 0, q0 = rnd() & mask); |
428 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, 1, q1 = q0, I[bidx]); |
429 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, 2, q2 = q1, I[bidx]); |
430 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, 3, q2, I[bidx]); |
431 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, -1, p0 = q0, E[bidx] >> 2); |
432 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, -2, p1 = p0, I[bidx]); |
433 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, -3, p2 = p1, I[bidx]); |
434 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 416 times.
|
596 | setsx(idx, -4, p2, I[bidx]); |
435 |
2/2✓ Branch 0 taken 2384 times.
✓ Branch 1 taken 596 times.
|
2980 | for (j = 4; j < 8; j++) { |
436 |
2/2✓ Branch 0 taken 720 times.
✓ Branch 1 taken 1664 times.
|
2384 | setpx(idx, -1 - j, rnd() & mask); |
437 |
2/2✓ Branch 0 taken 720 times.
✓ Branch 1 taken 1664 times.
|
2384 | setpx(idx, j, rnd() & mask); |
438 | } | ||
439 | } | ||
440 |
2/2✓ Branch 0 taken 596 times.
✓ Branch 1 taken 298 times.
|
894 | for (i = 6; i < 8; i++) /* off */ { |
441 | 596 | int idx = off + i * istride; | |
442 |
2/2✓ Branch 0 taken 4768 times.
✓ Branch 1 taken 596 times.
|
5364 | for (j = 0; j < 8; j++) { |
443 |
2/2✓ Branch 0 taken 1440 times.
✓ Branch 1 taken 3328 times.
|
4768 | setpx(idx, -1 - j, rnd() & mask); |
444 |
2/2✓ Branch 0 taken 1440 times.
✓ Branch 1 taken 3328 times.
|
4768 | setpx(idx, j, rnd() & mask); |
445 | } | ||
446 | } | ||
447 | 298 | } | |
448 | #define randomize_buffers(bidx, lineoff, str) \ | ||
449 | randomize_loopfilter_buffers(bidx, lineoff, str, bit_depth, dir, \ | ||
450 | E, F, H, I, buf0, buf1) | ||
451 | |||
452 | 13 | static void check_loopfilter(void) | |
453 | { | ||
454 | 13 | LOCAL_ALIGNED_32(uint8_t, base0, [32 + 16 * 16 * 2]); | |
455 | 13 | LOCAL_ALIGNED_32(uint8_t, base1, [32 + 16 * 16 * 2]); | |
456 | VP9DSPContext dsp; | ||
457 | int dir, wd, wd2, bit_depth; | ||
458 | static const char *const dir_name[2] = { "h", "v" }; | ||
459 | static const int E[2] = { 20, 28 }, I[2] = { 10, 16 }; | ||
460 | static const int H[2] = { 7, 11 }, F[2] = { 1, 1 }; | ||
461 |
2/2✓ Branch 1 taken 12 times.
✓ Branch 2 taken 1 times.
|
13 | declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, void, uint8_t *dst, ptrdiff_t stride, int E, int I, int H); |
462 | |||
463 |
2/2✓ Branch 0 taken 39 times.
✓ Branch 1 taken 13 times.
|
52 | for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) { |
464 | 39 | ff_vp9dsp_init(&dsp, bit_depth, 0); | |
465 | |||
466 |
2/2✓ Branch 0 taken 78 times.
✓ Branch 1 taken 39 times.
|
117 | for (dir = 0; dir < 2; dir++) { |
467 |
2/2✓ Branch 0 taken 39 times.
✓ Branch 1 taken 39 times.
|
78 | int midoff = (dir ? 8 * 8 : 8) * SIZEOF_PIXEL; |
468 |
2/2✓ Branch 0 taken 39 times.
✓ Branch 1 taken 39 times.
|
78 | int midoff_aligned = (dir ? 8 * 8 : 16) * SIZEOF_PIXEL; |
469 | 78 | uint8_t *buf0 = base0 + midoff_aligned; | |
470 | 78 | uint8_t *buf1 = base1 + midoff_aligned; | |
471 | |||
472 |
2/2✓ Branch 0 taken 234 times.
✓ Branch 1 taken 78 times.
|
312 | for (wd = 0; wd < 3; wd++) { |
473 | // 4/8/16wd_8px | ||
474 |
2/2✓ Branch 3 taken 58 times.
✓ Branch 4 taken 176 times.
|
234 | if (check_func(dsp.loop_filter_8[wd][dir], |
475 | "vp9_loop_filter_%s_%d_8_%dbpp", | ||
476 | dir_name[dir], 4 << wd, bit_depth)) { | ||
477 | 58 | randomize_buffers(0, 0, 8); | |
478 | 58 | memcpy(buf1 - midoff, buf0 - midoff, | |
479 | 58 | 16 * 8 * SIZEOF_PIXEL); | |
480 | 58 | call_ref(buf0, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]); | |
481 | 58 | call_new(buf1, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]); | |
482 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 58 times.
|
58 | if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 8 * SIZEOF_PIXEL)) |
483 | ✗ | fail(); | |
484 |
1/8✗ Branch 1 not taken.
✓ Branch 2 taken 58 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
|
58 | bench_new(buf1, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]); |
485 | } | ||
486 | } | ||
487 | |||
488 |
2/2✓ Branch 0 taken 39 times.
✓ Branch 1 taken 39 times.
|
78 | midoff = (dir ? 16 * 8 : 8) * SIZEOF_PIXEL; |
489 |
2/2✓ Branch 0 taken 39 times.
✓ Branch 1 taken 39 times.
|
78 | midoff_aligned = (dir ? 16 * 8 : 16) * SIZEOF_PIXEL; |
490 | |||
491 | 78 | buf0 = base0 + midoff_aligned; | |
492 | 78 | buf1 = base1 + midoff_aligned; | |
493 | |||
494 | // 16wd_16px loopfilter | ||
495 |
2/2✓ Branch 3 taken 24 times.
✓ Branch 4 taken 54 times.
|
78 | if (check_func(dsp.loop_filter_16[dir], |
496 | "vp9_loop_filter_%s_16_16_%dbpp", | ||
497 | dir_name[dir], bit_depth)) { | ||
498 | 24 | randomize_buffers(0, 0, 16); | |
499 | 24 | randomize_buffers(0, 8, 16); | |
500 | 24 | memcpy(buf1 - midoff, buf0 - midoff, 16 * 16 * SIZEOF_PIXEL); | |
501 | 24 | call_ref(buf0, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]); | |
502 | 24 | call_new(buf1, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]); | |
503 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 24 times.
|
24 | if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 16 * SIZEOF_PIXEL)) |
504 | ✗ | fail(); | |
505 |
1/8✗ Branch 1 not taken.
✓ Branch 2 taken 24 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
|
24 | bench_new(buf1, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]); |
506 | } | ||
507 | |||
508 |
2/2✓ Branch 0 taken 156 times.
✓ Branch 1 taken 78 times.
|
234 | for (wd = 0; wd < 2; wd++) { |
509 |
2/2✓ Branch 0 taken 312 times.
✓ Branch 1 taken 156 times.
|
468 | for (wd2 = 0; wd2 < 2; wd2++) { |
510 | // mix2 loopfilter | ||
511 |
2/2✓ Branch 3 taken 96 times.
✓ Branch 4 taken 216 times.
|
312 | if (check_func(dsp.loop_filter_mix2[wd][wd2][dir], |
512 | "vp9_loop_filter_mix2_%s_%d%d_16_%dbpp", | ||
513 | dir_name[dir], 4 << wd, 4 << wd2, bit_depth)) { | ||
514 | 96 | randomize_buffers(0, 0, 16); | |
515 | 96 | randomize_buffers(1, 8, 16); | |
516 | 96 | memcpy(buf1 - midoff, buf0 - midoff, 16 * 16 * SIZEOF_PIXEL); | |
517 | #define M(a) (((a)[1] << 8) | (a)[0]) | ||
518 | 96 | call_ref(buf0, 16 * SIZEOF_PIXEL, M(E), M(I), M(H)); | |
519 | 96 | call_new(buf1, 16 * SIZEOF_PIXEL, M(E), M(I), M(H)); | |
520 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 96 times.
|
96 | if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 16 * SIZEOF_PIXEL)) |
521 | ✗ | fail(); | |
522 |
1/8✗ Branch 1 not taken.
✓ Branch 2 taken 96 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
|
96 | bench_new(buf1, 16 * SIZEOF_PIXEL, M(E), M(I), M(H)); |
523 | #undef M | ||
524 | } | ||
525 | } | ||
526 | } | ||
527 | } | ||
528 | } | ||
529 | 13 | report("loopfilter"); | |
530 | 13 | } | |
531 | |||
532 | #undef setsx | ||
533 | #undef setpx | ||
534 | #undef setdx | ||
535 | #undef randomize_buffers | ||
536 | |||
537 | #define DST_BUF_SIZE (size * size * SIZEOF_PIXEL) | ||
538 | #define SRC_BUF_STRIDE 72 | ||
539 | #define SRC_BUF_SIZE ((size + 7) * SRC_BUF_STRIDE * SIZEOF_PIXEL) | ||
540 | #define src (buf + 3 * SIZEOF_PIXEL * (SRC_BUF_STRIDE + 1)) | ||
541 | |||
542 | #define randomize_buffers() \ | ||
543 | do { \ | ||
544 | uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \ | ||
545 | int k; \ | ||
546 | for (k = 0; k < SRC_BUF_SIZE; k += 4) { \ | ||
547 | uint32_t r = rnd() & mask; \ | ||
548 | AV_WN32A(buf + k, r); \ | ||
549 | } \ | ||
550 | if (op == 1) { \ | ||
551 | for (k = 0; k < DST_BUF_SIZE; k += 4) { \ | ||
552 | uint32_t r = rnd() & mask; \ | ||
553 | AV_WN32A(dst0 + k, r); \ | ||
554 | AV_WN32A(dst1 + k, r); \ | ||
555 | } \ | ||
556 | } \ | ||
557 | } while (0) | ||
558 | |||
559 | 13 | static void check_mc(void) | |
560 | { | ||
561 | 13 | LOCAL_ALIGNED_32(uint8_t, buf, [72 * 72 * 2]); | |
562 | 13 | LOCAL_ALIGNED_32(uint8_t, dst0, [64 * 64 * 2]); | |
563 | 13 | LOCAL_ALIGNED_32(uint8_t, dst1, [64 * 64 * 2]); | |
564 | VP9DSPContext dsp; | ||
565 | int op, hsize, bit_depth, filter, dx, dy; | ||
566 |
2/2✓ Branch 1 taken 12 times.
✓ Branch 2 taken 1 times.
|
13 | declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, void, uint8_t *dst, ptrdiff_t dst_stride, |
567 | const uint8_t *ref, ptrdiff_t ref_stride, | ||
568 | int h, int mx, int my); | ||
569 | static const char *const filter_names[4] = { | ||
570 | "8tap_smooth", "8tap_regular", "8tap_sharp", "bilin" | ||
571 | }; | ||
572 | static const char *const subpel_names[2][2] = { { "", "h" }, { "v", "hv" } }; | ||
573 | static const char *const op_names[2] = { "put", "avg" }; | ||
574 | char str[256]; | ||
575 | |||
576 |
2/2✓ Branch 0 taken 26 times.
✓ Branch 1 taken 13 times.
|
39 | for (op = 0; op < 2; op++) { |
577 |
2/2✓ Branch 0 taken 78 times.
✓ Branch 1 taken 26 times.
|
104 | for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) { |
578 | 78 | ff_vp9dsp_init(&dsp, bit_depth, 0); | |
579 |
2/2✓ Branch 0 taken 390 times.
✓ Branch 1 taken 78 times.
|
468 | for (hsize = 0; hsize < 5; hsize++) { |
580 | 390 | int size = 64 >> hsize; | |
581 | |||
582 |
2/2✓ Branch 0 taken 1560 times.
✓ Branch 1 taken 390 times.
|
1950 | for (filter = 0; filter < 4; filter++) { |
583 |
2/2✓ Branch 0 taken 3120 times.
✓ Branch 1 taken 1560 times.
|
4680 | for (dx = 0; dx < 2; dx++) { |
584 |
2/2✓ Branch 0 taken 6240 times.
✓ Branch 1 taken 3120 times.
|
9360 | for (dy = 0; dy < 2; dy++) { |
585 |
4/4✓ Branch 0 taken 3120 times.
✓ Branch 1 taken 3120 times.
✓ Branch 2 taken 1560 times.
✓ Branch 3 taken 1560 times.
|
6240 | if (dx || dy) { |
586 | 4680 | snprintf(str, sizeof(str), | |
587 | 4680 | "%s_%s_%d%s", op_names[op], | |
588 | 4680 | filter_names[filter], size, | |
589 | 4680 | subpel_names[dy][dx]); | |
590 | } else { | ||
591 | 1560 | snprintf(str, sizeof(str), | |
592 | 1560 | "%s%d", op_names[op], size); | |
593 | } | ||
594 |
2/2✓ Branch 3 taken 940 times.
✓ Branch 4 taken 5300 times.
|
6240 | if (check_func(dsp.mc[hsize][filter][op][dx][dy], |
595 | "vp9_%s_%dbpp", str, bit_depth)) { | ||
596 |
2/2✓ Branch 0 taken 576 times.
✓ Branch 1 taken 364 times.
|
940 | int mx = dx ? 1 + (rnd() % 14) : 0; |
597 |
2/2✓ Branch 0 taken 576 times.
✓ Branch 1 taken 364 times.
|
940 | int my = dy ? 1 + (rnd() % 14) : 0; |
598 |
6/6✓ Branch 1 taken 940896 times.
✓ Branch 2 taken 940 times.
✓ Branch 3 taken 470 times.
✓ Branch 4 taken 470 times.
✓ Branch 6 taken 235696 times.
✓ Branch 7 taken 470 times.
|
1177532 | randomize_buffers(); |
599 | 940 | call_ref(dst0, size * SIZEOF_PIXEL, | |
600 | src, SRC_BUF_STRIDE * SIZEOF_PIXEL, | ||
601 | size, mx, my); | ||
602 | 940 | call_new(dst1, size * SIZEOF_PIXEL, | |
603 | src, SRC_BUF_STRIDE * SIZEOF_PIXEL, | ||
604 | size, mx, my); | ||
605 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 940 times.
|
940 | if (memcmp(dst0, dst1, DST_BUF_SIZE)) |
606 | ✗ | fail(); | |
607 | |||
608 | // simd implementations for each filter of subpel | ||
609 | // functions are identical | ||
610 |
4/4✓ Branch 0 taken 606 times.
✓ Branch 1 taken 334 times.
✓ Branch 2 taken 516 times.
✓ Branch 3 taken 90 times.
|
940 | if (filter >= 1 && filter <= 2) continue; |
611 | // 10/12 bpp for bilin are identical | ||
612 |
4/4✓ Branch 0 taken 134 times.
✓ Branch 1 taken 290 times.
✓ Branch 2 taken 30 times.
✓ Branch 3 taken 104 times.
|
424 | if (bit_depth == 12 && filter == 3) continue; |
613 | |||
614 |
1/8✗ Branch 1 not taken.
✓ Branch 2 taken 394 times.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
|
394 | bench_new(dst1, size * SIZEOF_PIXEL, |
615 | src, SRC_BUF_STRIDE * SIZEOF_PIXEL, | ||
616 | size, mx, my); | ||
617 | } | ||
618 | } | ||
619 | } | ||
620 | } | ||
621 | } | ||
622 | } | ||
623 | } | ||
624 | 13 | report("mc"); | |
625 | 13 | } | |
626 | |||
627 | 13 | void checkasm_check_vp9dsp(void) | |
628 | { | ||
629 | 13 | check_ipred(); | |
630 | 13 | check_itxfm(); | |
631 | 13 | check_loopfilter(); | |
632 | 13 | check_mc(); | |
633 | 13 | } | |
634 |