falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
75474 Commits
32 Branches
535MB
Tree: 350e9c6765
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '350e9c6765'
${ noResults }
FFmpeg/tests/checkasm/vp9dsp.c

341 lines
14KB
Raw Blame History 

  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 <string.h>

  22. #include "checkasm.h"

  23. #include "libavcodec/vp9dsp.h"

  24. #include "libavutil/common.h"

  25. #include "libavutil/internal.h"

  26. #include "libavutil/intreadwrite.h"

  27. 

  28. static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff };

  29. #define SIZEOF_PIXEL ((bit_depth + 7) / 8)

  30. 

  31. #define randomize_buffers()                                        \

  32.     do {                                                           \

  33.         uint32_t mask = pixel_mask[(bit_depth - 8) >> 1];          \

  34.         int k;                                                     \

  35.         for (k = -4;  k < SIZEOF_PIXEL * FFMAX(8, size); k += 4) { \

  36.             uint32_t r = rnd() & mask;                             \

  37.             AV_WN32A(a + k, r);                                    \

  38.         }                                                          \

  39.         for (k = 0; k < size * SIZEOF_PIXEL; k += 4) {             \

  40.             uint32_t r = rnd() & mask;                             \

  41.             AV_WN32A(l + k, r);                                    \

  42.         }                                                          \

  43.     } while (0)

  44. 

  45. static void check_ipred(void)

  46. {

  47.     LOCAL_ALIGNED_32(uint8_t, a_buf, [64 * 2]);

  48.     uint8_t *a = &a_buf[32 * 2];

  49.     LOCAL_ALIGNED_32(uint8_t, l, [32 * 2]);

  50.     LOCAL_ALIGNED_32(uint8_t, dst0, [32 * 32 * 2]);

  51.     LOCAL_ALIGNED_32(uint8_t, dst1, [32 * 32 * 2]);

  52.     VP9DSPContext dsp;

  53.     int tx, mode, bit_depth;

  54.     declare_func(void, uint8_t *dst, ptrdiff_t stride,

  55.                  const uint8_t *left, const uint8_t *top);

  56.     static const char *const mode_names[N_INTRA_PRED_MODES] = {

  57.         [VERT_PRED] = "vert",

  58.         [HOR_PRED] = "hor",

  59.         [DC_PRED] = "dc",

  60.         [DIAG_DOWN_LEFT_PRED] = "diag_downleft",

  61.         [DIAG_DOWN_RIGHT_PRED] = "diag_downright",

  62.         [VERT_RIGHT_PRED] = "vert_right",

  63.         [HOR_DOWN_PRED] = "hor_down",

  64.         [VERT_LEFT_PRED] = "vert_left",

  65.         [HOR_UP_PRED] = "hor_up",

  66.         [TM_VP8_PRED] = "tm",

  67.         [LEFT_DC_PRED] = "dc_left",

  68.         [TOP_DC_PRED] = "dc_top",

  69.         [DC_128_PRED] = "dc_128",

  70.         [DC_127_PRED] = "dc_127",

  71.         [DC_129_PRED] = "dc_129",

  72.     };

  73. 

  74.     for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {

  75.         ff_vp9dsp_init(&dsp, bit_depth, 0);

  76.         for (tx = 0; tx < 4; tx++) {

  77.             int size = 4 << tx;

  78. 

  79.             for (mode = 0; mode < N_INTRA_PRED_MODES; mode++) {

  80.                 if (check_func(dsp.intra_pred[tx][mode], "vp9_%s_%dx%d_%dbpp",

  81.                                mode_names[mode], size, size, bit_depth)) {

  82.                     randomize_buffers();

  83.                     call_ref(dst0, size * SIZEOF_PIXEL, l, a);

  84.                     call_new(dst1, size * SIZEOF_PIXEL, l, a);

  85.                     if (memcmp(dst0, dst1, size * size * SIZEOF_PIXEL))

  86.                         fail();

  87.                     bench_new(dst1, size * SIZEOF_PIXEL,l, a);

  88.                 }

  89.             }

  90.         }

  91.     }

  92.     report("ipred");

  93. }

  94. 

  95. #undef randomize_buffers

  96. 

  97. #define setpx(a,b,c) \

  98.     do { \

  99.         if (SIZEOF_PIXEL == 1) { \

  100.             buf0[(a) + (b) * jstride] = c; \

  101.         } else { \

  102.             ((uint16_t *)buf0)[(a) + (b) * jstride] = c; \

  103.         } \

  104.     } while (0)

  105. #define setdx(a,b,c,d) setpx(a,b,(c)-(d)+(rnd()%((d)*2+1)))

  106. #define setsx(a,b,c,d) setdx(a,b,c,(d) << (bit_depth - 8))

  107. #define randomize_buffers(bidx, lineoff, str) \

  108.     do { \

  109.         uint32_t mask = (1 << bit_depth) - 1; \

  110.         int off = dir ? lineoff : lineoff * 16; \

  111.         int istride = dir ? 1 : 16; \

  112.         int jstride = dir ? str : 1; \

  113.         int i, j; \

  114.         for (i = 0; i < 2; i++) /* flat16 */ { \

  115.             int idx = off + i * istride, p0, q0; \

  116.             setpx(idx,  0, q0 = rnd() & mask); \

  117.             setsx(idx, -1, p0 = q0, E[bidx] >> 2); \

  118.             for (j = 1; j < 8; j++) { \

  119.                 setsx(idx, -1 - j, p0, F[bidx]); \

  120.                 setsx(idx, j, q0, F[bidx]); \

  121.             } \

  122.         } \

  123.         for (i = 2; i < 4; i++) /* flat8 */ { \

  124.             int idx = off + i * istride, p0, q0; \

  125.             setpx(idx,  0, q0 = rnd() & mask); \

  126.             setsx(idx, -1, p0 = q0, E[bidx] >> 2); \

  127.             for (j = 1; j < 4; j++) { \

  128.                 setsx(idx, -1 - j, p0, F[bidx]); \

  129.                 setsx(idx, j, q0, F[bidx]); \

  130.             } \

  131.             for (j = 4; j < 8; j++) { \

  132.                 setpx(idx, -1 - j, rnd() & mask); \

  133.                 setpx(idx, j, rnd() & mask); \

  134.             } \

  135.         } \

  136.         for (i = 4; i < 6; i++) /* regular */ { \

  137.             int idx = off + i * istride, p2, p1, p0, q0, q1, q2; \

  138.             setpx(idx,  0, q0 = rnd() & mask); \

  139.             setsx(idx,  1, q1 = q0, I[bidx]); \

  140.             setsx(idx,  2, q2 = q1, I[bidx]); \

  141.             setsx(idx,  3, q2,      I[bidx]); \

  142.             setsx(idx, -1, p0 = q0, E[bidx] >> 2); \

  143.             setsx(idx, -2, p1 = p0, I[bidx]); \

  144.             setsx(idx, -3, p2 = p1, I[bidx]); \

  145.             setsx(idx, -4, p2,      I[bidx]); \

  146.             for (j = 4; j < 8; j++) { \

  147.                 setpx(idx, -1 - j, rnd() & mask); \

  148.                 setpx(idx, j, rnd() & mask); \

  149.             } \

  150.         } \

  151.         for (i = 6; i < 8; i++) /* off */ { \

  152.             int idx = off + i * istride; \

  153.             for (j = 0; j < 8; j++) { \

  154.                 setpx(idx, -1 - j, rnd() & mask); \

  155.                 setpx(idx, j, rnd() & mask); \

  156.             } \

  157.         } \

  158.     } while (0)

  159. 

  160. static void check_loopfilter(void)

  161. {

  162.     LOCAL_ALIGNED_32(uint8_t, base0, [32 + 16 * 16 * 2]);

  163.     LOCAL_ALIGNED_32(uint8_t, base1, [32 + 16 * 16 * 2]);

  164.     VP9DSPContext dsp;

  165.     int dir, wd, wd2, bit_depth;

  166.     static const char *const dir_name[2] = { "h", "v" };

  167.     static const int E[2] = { 20, 28 }, I[2] = { 10, 16 };

  168.     static const int H[2] = { 7, 11 }, F[2] = { 1, 1 };

  169.     declare_func(void, uint8_t *dst, ptrdiff_t stride, int E, int I, int H);

  170. 

  171.     for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {

  172.         ff_vp9dsp_init(&dsp, bit_depth, 0);

  173. 

  174.         for (dir = 0; dir < 2; dir++) {

  175.             int midoff = (dir ? 8 * 8 : 8) * SIZEOF_PIXEL;

  176.             int midoff_aligned = (dir ? 8 * 8 : 16) * SIZEOF_PIXEL;

  177.             uint8_t *buf0 = base0 + midoff_aligned;

  178.             uint8_t *buf1 = base1 + midoff_aligned;

  179. 

  180.             for (wd = 0; wd < 3; wd++) {

  181.                 // 4/8/16wd_8px

  182.                 if (check_func(dsp.loop_filter_8[wd][dir],

  183.                                "vp9_loop_filter_%s_%d_8_%dbpp",

  184.                                dir_name[dir], 4 << wd, bit_depth)) {

  185.                     randomize_buffers(0, 0, 8);

  186.                     memcpy(buf1 - midoff, buf0 - midoff,

  187.                            16 * 8 * SIZEOF_PIXEL);

  188.                     call_ref(buf0, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]);

  189.                     call_new(buf1, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]);

  190.                     if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 8 * SIZEOF_PIXEL))

  191.                         fail();

  192.                     bench_new(buf1, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]);

  193.                 }

  194.             }

  195. 

  196.             midoff = (dir ? 16 * 8 : 8) * SIZEOF_PIXEL;

  197.             midoff_aligned = (dir ? 16 * 8 : 16) * SIZEOF_PIXEL;

  198. 

  199.             buf0 = base0 + midoff_aligned;

  200.             buf1 = base1 + midoff_aligned;

  201. 

  202.             // 16wd_16px loopfilter

  203.             if (check_func(dsp.loop_filter_16[dir],

  204.                            "vp9_loop_filter_%s_16_16_%dbpp",

  205.                            dir_name[dir], bit_depth)) {

  206.                 randomize_buffers(0, 0, 16);

  207.                 randomize_buffers(0, 8, 16);

  208.                 memcpy(buf1 - midoff, buf0 - midoff, 16 * 16 * SIZEOF_PIXEL);

  209.                 call_ref(buf0, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]);

  210.                 call_new(buf1, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]);

  211.                 if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 16 * SIZEOF_PIXEL))

  212.                     fail();

  213.                 bench_new(buf1, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]);

  214.             }

  215. 

  216.             for (wd = 0; wd < 2; wd++) {

  217.                 for (wd2 = 0; wd2 < 2; wd2++) {

  218.                     // mix2 loopfilter

  219.                     if (check_func(dsp.loop_filter_mix2[wd][wd2][dir],

  220.                                    "vp9_loop_filter_mix2_%s_%d%d_16_%dbpp",

  221.                                    dir_name[dir], 4 << wd, 4 << wd2, bit_depth)) {

  222.                         randomize_buffers(0, 0, 16);

  223.                         randomize_buffers(1, 8, 16);

  224.                         memcpy(buf1 - midoff, buf0 - midoff, 16 * 16 * SIZEOF_PIXEL);

  225. #define M(a) (((a)[1] << 8) | (a)[0])

  226.                         call_ref(buf0, 16 * SIZEOF_PIXEL, M(E), M(I), M(H));

  227.                         call_new(buf1, 16 * SIZEOF_PIXEL, M(E), M(I), M(H));

  228.                         if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 16 * SIZEOF_PIXEL))

  229.                             fail();

  230.                         bench_new(buf1, 16 * SIZEOF_PIXEL, M(E), M(I), M(H));

  231. #undef M

  232.                     }

  233.                 }

  234.             }

  235.         }

  236.     }

  237.     report("loopfilter");

  238. }

  239. 

  240. #undef setsx

  241. #undef setpx

  242. #undef setdx

  243. #undef randomize_buffers

  244. 

  245. #define DST_BUF_SIZE (size * size * SIZEOF_PIXEL)

  246. #define SRC_BUF_STRIDE 72

  247. #define SRC_BUF_SIZE ((size + 7) * SRC_BUF_STRIDE * SIZEOF_PIXEL)

  248. #define src (buf + 3 * SIZEOF_PIXEL * (SRC_BUF_STRIDE + 1))

  249. 

  250. #define randomize_buffers()                               \

  251.     do {                                                  \

  252.         uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \

  253.         int k;                                            \

  254.         for (k = 0; k < SRC_BUF_SIZE; k += 4) {           \

  255.             uint32_t r = rnd() & mask;                    \

  256.             AV_WN32A(buf + k, r);                         \

  257.         }                                                 \

  258.         if (op == 1) {                                    \

  259.             for (k = 0; k < DST_BUF_SIZE; k += 4) {       \

  260.                 uint32_t r = rnd() & mask;                \

  261.                 AV_WN32A(dst0 + k, r);                    \

  262.                 AV_WN32A(dst1 + k, r);                    \

  263.             }                                             \

  264.         }                                                 \

  265.     } while (0)

  266. 

  267. static void check_mc(void)

  268. {

  269.     LOCAL_ALIGNED_32(uint8_t, buf, [72 * 72 * 2]);

  270.     LOCAL_ALIGNED_32(uint8_t, dst0, [64 * 64 * 2]);

  271.     LOCAL_ALIGNED_32(uint8_t, dst1, [64 * 64 * 2]);

  272.     VP9DSPContext dsp;

  273.     int op, hsize, bit_depth, filter, dx, dy;

  274.     declare_func(void, uint8_t *dst, ptrdiff_t dst_stride,

  275.                  const uint8_t *ref, ptrdiff_t ref_stride,

  276.                  int h, int mx, int my);

  277.     static const char *const filter_names[4] = {

  278.         "8tap_smooth", "8tap_regular", "8tap_sharp", "bilin"

  279.     };

  280.     static const char *const subpel_names[2][2] = { { "", "h" }, { "v", "hv" } };

  281.     static const char *const op_names[2] = { "put", "avg" };

  282.     char str[256];

  283. 

  284.     for (op = 0; op < 2; op++) {

  285.         for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {

  286.             ff_vp9dsp_init(&dsp, bit_depth, 0);

  287.             for (hsize = 0; hsize < 5; hsize++) {

  288.                 int size = 64 >> hsize;

  289. 

  290.                 for (filter = 0; filter < 4; filter++) {

  291.                     for (dx = 0; dx < 2; dx++) {

  292.                         for (dy = 0; dy < 2; dy++) {

  293.                             if (dx || dy) {

  294.                                 snprintf(str, sizeof(str),

  295.                                          "%s_%s_%d%s", op_names[op],

  296.                                          filter_names[filter], size,

  297.                                          subpel_names[dy][dx]);

  298.                             } else {

  299.                                 snprintf(str, sizeof(str),

  300.                                          "%s%d", op_names[op], size);

  301.                             }

  302.                             if (check_func(dsp.mc[hsize][filter][op][dx][dy],

  303.                                            "vp9_%s_%dbpp", str, bit_depth)) {

  304.                                 int mx = dx ? 1 + (rnd() % 14) : 0;

  305.                                 int my = dy ? 1 + (rnd() % 14) : 0;

  306.                                 randomize_buffers();

  307.                                 call_ref(dst0, size * SIZEOF_PIXEL,

  308.                                          src, SRC_BUF_STRIDE * SIZEOF_PIXEL,

  309.                                          size, mx, my);

  310.                                 call_new(dst1, size * SIZEOF_PIXEL,

  311.                                          src, SRC_BUF_STRIDE * SIZEOF_PIXEL,

  312.                                          size, mx, my);

  313.                                 if (memcmp(dst0, dst1, DST_BUF_SIZE))

  314.                                     fail();

  315. 

  316.                                 // simd implementations for each filter of subpel

  317.                                 // functions are identical

  318.                                 if (filter >= 1 && filter <= 2) continue;

  319.                                 // 10/12 bpp for bilin are identical

  320.                                 if (bit_depth == 12 && filter == 3) continue;

  321. 

  322.                                 bench_new(dst1, size * SIZEOF_PIXEL,

  323.                                           src, SRC_BUF_STRIDE * SIZEOF_PIXEL,

  324.                                           size, mx, my);

  325.                             }

  326.                         }

  327.                     }

  328.                 }

  329.             }

  330.         }

  331.     }

  332.     report("mc");

  333. }

  334. 

  335. void checkasm_check_vp9dsp(void)

  336. {

  337.     check_ipred();

  338.     check_loopfilter();

  339.     check_mc();

  340. }