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.
101790 Commits
32 Branches
424MB
Branch: n4.4-jackoutdev
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'n4.4-jackoutdev'
${ noResults }
FFmpeg/tests/checkasm/sw_rgb.c

208 lines
7.7KB
Raw Permalink Blame History 

  1. /*

  2.  *

  3.  * This file is part of FFmpeg.

  4.  *

  5.  * FFmpeg is free software; you can redistribute it and/or modify

  6.  * it under the terms of the GNU General Public License as published by

  7.  * the Free Software Foundation; either version 2 of the License, or

  8.  * (at your option) any later version.

  9.  *

  10.  * FFmpeg is distributed in the hope that it will be useful,

  11.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  12.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

  13.  * GNU General Public License for more details.

  14.  *

  15.  * You should have received a copy of the GNU General Public License along

  16.  * with FFmpeg; if not, write to the Free Software Foundation, Inc.,

  17.  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.

  18.  */

  19. 

  20. #include <string.h>

  21. 

  22. #include "libavutil/common.h"

  23. #include "libavutil/intreadwrite.h"

  24. #include "libavutil/mem.h"

  25. #include "libavutil/mem_internal.h"

  26. 

  27. #include "libswscale/rgb2rgb.h"

  28. 

  29. #include "checkasm.h"

  30. 

  31. #define randomize_buffers(buf, size)      \

  32.     do {                                  \

  33.         int j;                            \

  34.         for (j = 0; j < size; j+=4)       \

  35.             AV_WN32(buf + j, rnd());      \

  36.     } while (0)

  37. 

  38. static const uint8_t width[] = {12, 16, 20, 32, 36, 128};

  39. static const struct {uint8_t w, h, s;} planes[] = {

  40.     {12,16,12}, {16,16,16}, {20,23,25}, {32,18,48}, {8,128,16}, {128,128,128}

  41. };

  42. 

  43. #define MAX_STRIDE 128

  44. #define MAX_HEIGHT 128

  45. 

  46. static void check_shuffle_bytes(void * func, const char * report)

  47. {

  48.     int i;

  49.     LOCAL_ALIGNED_32(uint8_t, src0, [MAX_STRIDE]);

  50.     LOCAL_ALIGNED_32(uint8_t, src1, [MAX_STRIDE]);

  51.     LOCAL_ALIGNED_32(uint8_t, dst0, [MAX_STRIDE]);

  52.     LOCAL_ALIGNED_32(uint8_t, dst1, [MAX_STRIDE]);

  53. 

  54.     declare_func_emms(AV_CPU_FLAG_MMX, void, const uint8_t *src, uint8_t *dst, int src_size);

  55. 

  56.     memset(dst0, 0, MAX_STRIDE);

  57.     memset(dst1, 0, MAX_STRIDE);

  58.     randomize_buffers(src0, MAX_STRIDE);

  59.     memcpy(src1, src0, MAX_STRIDE);

  60. 

  61.     if (check_func(func, "%s", report)) {

  62.         for (i = 0; i < 6; i ++) {

  63.             call_ref(src0, dst0, width[i]);

  64.             call_new(src1, dst1, width[i]);

  65.             if (memcmp(dst0, dst1, MAX_STRIDE))

  66.                 fail();

  67.         }

  68.         bench_new(src0, dst0, width[5]);

  69.     }

  70. }

  71. 

  72. static void check_uyvy_to_422p(void)

  73. {

  74.     int i;

  75. 

  76.     LOCAL_ALIGNED_32(uint8_t, src0, [MAX_STRIDE * MAX_HEIGHT * 2]);

  77.     LOCAL_ALIGNED_32(uint8_t, src1, [MAX_STRIDE * MAX_HEIGHT * 2]);

  78.     LOCAL_ALIGNED_32(uint8_t, dst_y_0, [MAX_STRIDE * MAX_HEIGHT]);

  79.     LOCAL_ALIGNED_32(uint8_t, dst_y_1, [MAX_STRIDE * MAX_HEIGHT]);

  80.     LOCAL_ALIGNED_32(uint8_t, dst_u_0, [(MAX_STRIDE/2) * MAX_HEIGHT]);

  81.     LOCAL_ALIGNED_32(uint8_t, dst_u_1, [(MAX_STRIDE/2) * MAX_HEIGHT]);

  82.     LOCAL_ALIGNED_32(uint8_t, dst_v_0, [(MAX_STRIDE/2) * MAX_HEIGHT]);

  83.     LOCAL_ALIGNED_32(uint8_t, dst_v_1, [(MAX_STRIDE/2) * MAX_HEIGHT]);

  84. 

  85.     declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,

  86.                       const uint8_t *src, int width, int height,

  87.                       int lumStride, int chromStride, int srcStride);

  88. 

  89.     randomize_buffers(src0, MAX_STRIDE * MAX_HEIGHT * 2);

  90.     memcpy(src1, src0, MAX_STRIDE * MAX_HEIGHT * 2);

  91. 

  92.     if (check_func(uyvytoyuv422, "uyvytoyuv422")) {

  93.         for (i = 0; i < 6; i ++) {

  94.             memset(dst_y_0, 0, MAX_STRIDE * MAX_HEIGHT);

  95.             memset(dst_y_1, 0, MAX_STRIDE * MAX_HEIGHT);

  96.             memset(dst_u_0, 0, (MAX_STRIDE/2) * MAX_HEIGHT);

  97.             memset(dst_u_1, 0, (MAX_STRIDE/2) * MAX_HEIGHT);

  98.             memset(dst_v_0, 0, (MAX_STRIDE/2) * MAX_HEIGHT);

  99.             memset(dst_v_1, 0, (MAX_STRIDE/2) * MAX_HEIGHT);

  100. 

  101.             call_ref(dst_y_0, dst_u_0, dst_v_0, src0, planes[i].w, planes[i].h,

  102.                      MAX_STRIDE, MAX_STRIDE / 2, planes[i].s);

  103.             call_new(dst_y_1, dst_u_1, dst_v_1, src1, planes[i].w, planes[i].h,

  104.                      MAX_STRIDE, MAX_STRIDE / 2, planes[i].s);

  105.             if (memcmp(dst_y_0, dst_y_1, MAX_STRIDE * MAX_HEIGHT) ||

  106.                 memcmp(dst_u_0, dst_u_1, (MAX_STRIDE/2) * MAX_HEIGHT) ||

  107.                 memcmp(dst_v_0, dst_v_1, (MAX_STRIDE/2) * MAX_HEIGHT))

  108.                 fail();

  109.         }

  110.         bench_new(dst_y_1, dst_u_1, dst_v_1, src1, planes[5].w, planes[5].h,

  111.                   MAX_STRIDE, MAX_STRIDE / 2, planes[5].s);

  112.     }

  113. }

  114. 

  115. static void check_interleave_bytes(void)

  116. {

  117.     LOCAL_ALIGNED_16(uint8_t, src0_buf, [MAX_STRIDE*MAX_HEIGHT+1]);

  118.     LOCAL_ALIGNED_16(uint8_t, src1_buf, [MAX_STRIDE*MAX_HEIGHT+1]);

  119.     LOCAL_ALIGNED_16(uint8_t, dst0_buf, [2*MAX_STRIDE*MAX_HEIGHT+2]);

  120.     LOCAL_ALIGNED_16(uint8_t, dst1_buf, [2*MAX_STRIDE*MAX_HEIGHT+2]);

  121.     // Intentionally using unaligned buffers, as this function doesn't have

  122.     // any alignment requirements.

  123.     uint8_t *src0 = src0_buf + 1;

  124.     uint8_t *src1 = src1_buf + 1;

  125.     uint8_t *dst0 = dst0_buf + 2;

  126.     uint8_t *dst1 = dst1_buf + 2;

  127. 

  128.     declare_func_emms(AV_CPU_FLAG_MMX, void, const uint8_t *, const uint8_t *,

  129.                                        uint8_t *, int, int, int, int, int);

  130. 

  131.     randomize_buffers(src0, MAX_STRIDE * MAX_HEIGHT);

  132.     randomize_buffers(src1, MAX_STRIDE * MAX_HEIGHT);

  133. 

  134.     if (check_func(interleaveBytes, "interleave_bytes")) {

  135.         for (int i = 0; i <= 16; i++) {

  136.             // Try all widths [1,16], and try one random width.

  137. 

  138.             int w = i > 0 ? i : (1 + (rnd() % (MAX_STRIDE-2)));

  139.             int h = 1 + (rnd() % (MAX_HEIGHT-2));

  140. 

  141.             int src0_offset = 0, src0_stride = MAX_STRIDE;

  142.             int src1_offset = 0, src1_stride = MAX_STRIDE;

  143.             int dst_offset  = 0, dst_stride  = 2 * MAX_STRIDE;

  144. 

  145.             memset(dst0, 0, 2 * MAX_STRIDE * MAX_HEIGHT);

  146.             memset(dst1, 0, 2 * MAX_STRIDE * MAX_HEIGHT);

  147. 

  148.             // Try different combinations of negative strides

  149.             if (i & 1) {

  150.                 src0_offset = (h-1)*src0_stride;

  151.                 src0_stride = -src0_stride;

  152.             }

  153.             if (i & 2) {

  154.                 src1_offset = (h-1)*src1_stride;

  155.                 src1_stride = -src1_stride;

  156.             }

  157.             if (i & 4) {

  158.                 dst_offset = (h-1)*dst_stride;

  159.                 dst_stride = -dst_stride;

  160.             }

  161. 

  162.             call_ref(src0 + src0_offset, src1 + src1_offset, dst0 + dst_offset,

  163.                      w, h, src0_stride, src1_stride, dst_stride);

  164.             call_new(src0 + src0_offset, src1 + src1_offset, dst1 + dst_offset,

  165.                      w, h, src0_stride, src1_stride, dst_stride);

  166.             // Check a one pixel-pair edge around the destination area,

  167.             // to catch overwrites past the end.

  168.             checkasm_check(uint8_t, dst0, 2*MAX_STRIDE, dst1, 2*MAX_STRIDE,

  169.                            2 * w + 2, h + 1, "dst");

  170.         }

  171. 

  172.         bench_new(src0, src1, dst1, 127, MAX_HEIGHT,

  173.                   MAX_STRIDE, MAX_STRIDE, 2*MAX_STRIDE);

  174.     }

  175.     if (check_func(interleaveBytes, "interleave_bytes_aligned")) {

  176.         // Bench the function in a more typical case, with aligned

  177.         // buffers and widths.

  178.         bench_new(src0_buf, src1_buf, dst1_buf, 128, MAX_HEIGHT,

  179.                   MAX_STRIDE, MAX_STRIDE, 2*MAX_STRIDE);

  180.     }

  181. }

  182. 

  183. void checkasm_check_sw_rgb(void)

  184. {

  185.     ff_sws_rgb2rgb_init();

  186. 

  187.     check_shuffle_bytes(shuffle_bytes_2103, "shuffle_bytes_2103");

  188.     report("shuffle_bytes_2103");

  189. 

  190.     check_shuffle_bytes(shuffle_bytes_0321, "shuffle_bytes_0321");

  191.     report("shuffle_bytes_0321");

  192. 

  193.     check_shuffle_bytes(shuffle_bytes_1230, "shuffle_bytes_1230");

  194.     report("shuffle_bytes_1230");

  195. 

  196.     check_shuffle_bytes(shuffle_bytes_3012, "shuffle_bytes_3012");

  197.     report("shuffle_bytes_3012");

  198. 

  199.     check_shuffle_bytes(shuffle_bytes_3210, "shuffle_bytes_3210");

  200.     report("shuffle_bytes_3210");

  201. 

  202.     check_uyvy_to_422p();

  203.     report("uyvytoyuv422");

  204. 

  205.     check_interleave_bytes();

  206.     report("interleave_bytes");

  207. }