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.
44689 Commits
32 Branches
824MB
Tree: c5714b51aa
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c5714b51aa'
${ noResults }
FFmpeg/libavcodec/rv40dsp.c

715 lines
31KB
Raw Blame History 

  1. /*

  2.  * RV40 decoder motion compensation functions

  3.  * Copyright (c) 2008 Konstantin Shishkov

  4.  *

  5.  * This file is part of Libav.

  6.  *

  7.  * Libav is free software; you can redistribute it and/or

  8.  * modify it under the terms of the GNU Lesser General Public

  9.  * License as published by the Free Software Foundation; either

  10.  * version 2.1 of the License, or (at your option) any later version.

  11.  *

  12.  * Libav is distributed in the hope that it will be useful,

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

  14.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  15.  * Lesser General Public License for more details.

  16.  *

  17.  * You should have received a copy of the GNU Lesser General Public

  18.  * License along with Libav; if not, write to the Free Software

  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  20.  */

  21. 

  22. /**

  23.  * @file

  24.  * RV40 decoder motion compensation functions

  25.  */

  26. 

  27. #include "libavutil/common.h"

  28. #include "libavutil/intreadwrite.h"

  29. #include "avcodec.h"

  30. #include "h264qpel.h"

  31. #include "mathops.h"

  32. #include "pixels.h"

  33. #include "rnd_avg.h"

  34. #include "rv34dsp.h"

  35. 

  36. #define RV40_LOWPASS(OPNAME, OP) \

  37. static void OPNAME ## rv40_qpel8_h_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\

  38.                                                      const int h, const int C1, const int C2, const int SHIFT){\

  39.     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;\

  40.     int i;\

  41.     for(i = 0; i < h; i++)\

  42.     {\

  43.         OP(dst[0], (src[-2] + src[ 3] - 5*(src[-1]+src[2]) + src[0]*C1 + src[1]*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  44.         OP(dst[1], (src[-1] + src[ 4] - 5*(src[ 0]+src[3]) + src[1]*C1 + src[2]*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  45.         OP(dst[2], (src[ 0] + src[ 5] - 5*(src[ 1]+src[4]) + src[2]*C1 + src[3]*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  46.         OP(dst[3], (src[ 1] + src[ 6] - 5*(src[ 2]+src[5]) + src[3]*C1 + src[4]*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  47.         OP(dst[4], (src[ 2] + src[ 7] - 5*(src[ 3]+src[6]) + src[4]*C1 + src[5]*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  48.         OP(dst[5], (src[ 3] + src[ 8] - 5*(src[ 4]+src[7]) + src[5]*C1 + src[6]*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  49.         OP(dst[6], (src[ 4] + src[ 9] - 5*(src[ 5]+src[8]) + src[6]*C1 + src[7]*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  50.         OP(dst[7], (src[ 5] + src[10] - 5*(src[ 6]+src[9]) + src[7]*C1 + src[8]*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  51.         dst += dstStride;\

  52.         src += srcStride;\

  53.     }\

  54. }\

  55. \

  56. static void OPNAME ## rv40_qpel8_v_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\

  57.                                            const int w, const int C1, const int C2, const int SHIFT){\

  58.     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;\

  59.     int i;\

  60.     for(i = 0; i < w; i++)\

  61.     {\

  62.         const int srcB  = src[-2*srcStride];\

  63.         const int srcA  = src[-1*srcStride];\

  64.         const int src0  = src[0 *srcStride];\

  65.         const int src1  = src[1 *srcStride];\

  66.         const int src2  = src[2 *srcStride];\

  67.         const int src3  = src[3 *srcStride];\

  68.         const int src4  = src[4 *srcStride];\

  69.         const int src5  = src[5 *srcStride];\

  70.         const int src6  = src[6 *srcStride];\

  71.         const int src7  = src[7 *srcStride];\

  72.         const int src8  = src[8 *srcStride];\

  73.         const int src9  = src[9 *srcStride];\

  74.         const int src10 = src[10*srcStride];\

  75.         OP(dst[0*dstStride], (srcB + src3  - 5*(srcA+src2) + src0*C1 + src1*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  76.         OP(dst[1*dstStride], (srcA + src4  - 5*(src0+src3) + src1*C1 + src2*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  77.         OP(dst[2*dstStride], (src0 + src5  - 5*(src1+src4) + src2*C1 + src3*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  78.         OP(dst[3*dstStride], (src1 + src6  - 5*(src2+src5) + src3*C1 + src4*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  79.         OP(dst[4*dstStride], (src2 + src7  - 5*(src3+src6) + src4*C1 + src5*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  80.         OP(dst[5*dstStride], (src3 + src8  - 5*(src4+src7) + src5*C1 + src6*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  81.         OP(dst[6*dstStride], (src4 + src9  - 5*(src5+src8) + src6*C1 + src7*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  82.         OP(dst[7*dstStride], (src5 + src10 - 5*(src6+src9) + src7*C1 + src8*C2 + (1<<(SHIFT-1))) >> SHIFT);\

  83.         dst++;\

  84.         src++;\

  85.     }\

  86. }\

  87. \

  88. static void OPNAME ## rv40_qpel16_v_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\

  89.                                             const int w, const int C1, const int C2, const int SHIFT){\

  90.     OPNAME ## rv40_qpel8_v_lowpass(dst  , src  , dstStride, srcStride, 8, C1, C2, SHIFT);\

  91.     OPNAME ## rv40_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride, 8, C1, C2, SHIFT);\

  92.     src += 8*srcStride;\

  93.     dst += 8*dstStride;\

  94.     OPNAME ## rv40_qpel8_v_lowpass(dst  , src  , dstStride, srcStride, w-8, C1, C2, SHIFT);\

  95.     OPNAME ## rv40_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride, w-8, C1, C2, SHIFT);\

  96. }\

  97. \

  98. static void OPNAME ## rv40_qpel16_h_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\

  99.                                             const int h, const int C1, const int C2, const int SHIFT){\

  100.     OPNAME ## rv40_qpel8_h_lowpass(dst  , src  , dstStride, srcStride, 8, C1, C2, SHIFT);\

  101.     OPNAME ## rv40_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride, 8, C1, C2, SHIFT);\

  102.     src += 8*srcStride;\

  103.     dst += 8*dstStride;\

  104.     OPNAME ## rv40_qpel8_h_lowpass(dst  , src  , dstStride, srcStride, h-8, C1, C2, SHIFT);\

  105.     OPNAME ## rv40_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride, h-8, C1, C2, SHIFT);\

  106. }\

  107. \

  108. 

  109. #define RV40_MC(OPNAME, SIZE) \

  110. static void OPNAME ## rv40_qpel ## SIZE ## _mc10_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  111. {\

  112.     OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\

  113. }\

  114. \

  115. static void OPNAME ## rv40_qpel ## SIZE ## _mc30_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  116. {\

  117.     OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\

  118. }\

  119. \

  120. static void OPNAME ## rv40_qpel ## SIZE ## _mc01_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  121. {\

  122.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\

  123. }\

  124. \

  125. static void OPNAME ## rv40_qpel ## SIZE ## _mc11_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  126. {\

  127.     uint8_t full[SIZE*(SIZE+5)];\

  128.     uint8_t * const full_mid = full + SIZE*2;\

  129.     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\

  130.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\

  131. }\

  132. \

  133. static void OPNAME ## rv40_qpel ## SIZE ## _mc21_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  134. {\

  135.     uint8_t full[SIZE*(SIZE+5)];\

  136.     uint8_t * const full_mid = full + SIZE*2;\

  137.     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\

  138.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\

  139. }\

  140. \

  141. static void OPNAME ## rv40_qpel ## SIZE ## _mc31_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  142. {\

  143.     uint8_t full[SIZE*(SIZE+5)];\

  144.     uint8_t * const full_mid = full + SIZE*2;\

  145.     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 52, 6);\

  146.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\

  147. }\

  148. \

  149. static void OPNAME ## rv40_qpel ## SIZE ## _mc12_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  150. {\

  151.     uint8_t full[SIZE*(SIZE+5)];\

  152.     uint8_t * const full_mid = full + SIZE*2;\

  153.     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\

  154.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\

  155. }\

  156. \

  157. static void OPNAME ## rv40_qpel ## SIZE ## _mc22_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  158. {\

  159.     uint8_t full[SIZE*(SIZE+5)];\

  160.     uint8_t * const full_mid = full + SIZE*2;\

  161.     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\

  162.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\

  163. }\

  164. \

  165. static void OPNAME ## rv40_qpel ## SIZE ## _mc32_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  166. {\

  167.     uint8_t full[SIZE*(SIZE+5)];\

  168.     uint8_t * const full_mid = full + SIZE*2;\

  169.     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 52, 6);\

  170.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\

  171. }\

  172. \

  173. static void OPNAME ## rv40_qpel ## SIZE ## _mc03_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  174. {\

  175.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\

  176. }\

  177. \

  178. static void OPNAME ## rv40_qpel ## SIZE ## _mc13_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  179. {\

  180.     uint8_t full[SIZE*(SIZE+5)];\

  181.     uint8_t * const full_mid = full + SIZE*2;\

  182.     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\

  183.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 52, 6);\

  184. }\

  185. \

  186. static void OPNAME ## rv40_qpel ## SIZE ## _mc23_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  187. {\

  188.     uint8_t full[SIZE*(SIZE+5)];\

  189.     uint8_t * const full_mid = full + SIZE*2;\

  190.     put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\

  191.     OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 52, 6);\

  192. }\

  193. \

  194. 

  195. #define op_avg(a, b)  a = (((a)+cm[b]+1)>>1)

  196. #define op_put(a, b)  a = cm[b]

  197. 

  198. RV40_LOWPASS(put_       , op_put)

  199. RV40_LOWPASS(avg_       , op_avg)

  200. 

  201. #undef op_avg

  202. #undef op_put

  203. 

  204. RV40_MC(put_, 8)

  205. RV40_MC(put_, 16)

  206. RV40_MC(avg_, 8)

  207. RV40_MC(avg_, 16)

  208. 

  209. #define PIXOP2(OPNAME, OP)                                              \

  210. static inline void OPNAME ## _pixels8_xy2_8_c(uint8_t *block,           \

  211.                                               const uint8_t *pixels,    \

  212.                                               ptrdiff_t line_size,      \

  213.                                               int h)                    \

  214. {                                                                       \

  215.     /* FIXME HIGH BIT DEPTH */                                          \

  216.     int j;                                                              \

  217.                                                                         \

  218.     for (j = 0; j < 2; j++) {                                           \

  219.         int i;                                                          \

  220.         const uint32_t a = AV_RN32(pixels);                             \

  221.         const uint32_t b = AV_RN32(pixels + 1);                         \

  222.         uint32_t l0 = (a & 0x03030303UL) +                              \

  223.                       (b & 0x03030303UL) +                              \

  224.                            0x02020202UL;                                \

  225.         uint32_t h0 = ((a & 0xFCFCFCFCUL) >> 2) +                       \

  226.                       ((b & 0xFCFCFCFCUL) >> 2);                        \

  227.         uint32_t l1, h1;                                                \

  228.                                                                         \

  229.         pixels += line_size;                                            \

  230.         for (i = 0; i < h; i += 2) {                                    \

  231.             uint32_t a = AV_RN32(pixels);                               \

  232.             uint32_t b = AV_RN32(pixels + 1);                           \

  233.             l1 = (a & 0x03030303UL) +                                   \

  234.                  (b & 0x03030303UL);                                    \

  235.             h1 = ((a & 0xFCFCFCFCUL) >> 2) +                            \

  236.                  ((b & 0xFCFCFCFCUL) >> 2);                             \

  237.             OP(*((uint32_t *) block),                                   \

  238.                h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL));            \

  239.             pixels += line_size;                                        \

  240.             block  += line_size;                                        \

  241.             a = AV_RN32(pixels);                                        \

  242.             b = AV_RN32(pixels + 1);                                    \

  243.             l0 = (a & 0x03030303UL) +                                   \

  244.                  (b & 0x03030303UL) +                                   \

  245.                       0x02020202UL;                                     \

  246.             h0 = ((a & 0xFCFCFCFCUL) >> 2) +                            \

  247.                  ((b & 0xFCFCFCFCUL) >> 2);                             \

  248.             OP(*((uint32_t *) block),                                   \

  249.                h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL));            \

  250.             pixels += line_size;                                        \

  251.             block  += line_size;                                        \

  252.         }                                                               \

  253.         pixels += 4 - line_size * (h + 1);                              \

  254.         block  += 4 - line_size * h;                                    \

  255.     }                                                                   \

  256. }                                                                       \

  257.                                                                         \

  258. CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_8_c,                             \

  259.                OPNAME ## _pixels8_xy2_8_c,                              \

  260.                8)                                                       \

  261. 

  262. #define op_avg(a, b) a = rnd_avg32(a, b)

  263. #define op_put(a, b) a = b

  264. PIXOP2(avg, op_avg)

  265. PIXOP2(put, op_put)

  266. #undef op_avg

  267. #undef op_put

  268. 

  269. static void put_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)

  270. {

  271.     put_pixels16_xy2_8_c(dst, src, stride, 16);

  272. }

  273. static void avg_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)

  274. {

  275.     avg_pixels16_xy2_8_c(dst, src, stride, 16);

  276. }

  277. static void put_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)

  278. {

  279.     put_pixels8_xy2_8_c(dst, src, stride, 8);

  280. }

  281. static void avg_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)

  282. {

  283.     avg_pixels8_xy2_8_c(dst, src, stride, 8);

  284. }

  285. 

  286. static const int rv40_bias[4][4] = {

  287.     {  0, 16, 32, 16 },

  288.     { 32, 28, 32, 28 },

  289.     {  0, 32, 16, 32 },

  290.     { 32, 28, 32, 28 }

  291. };

  292. 

  293. #define RV40_CHROMA_MC(OPNAME, OP)\

  294. static void OPNAME ## rv40_chroma_mc4_c(uint8_t *dst /*align 8*/,\

  295.                                         uint8_t *src /*align 1*/,\

  296.                                         ptrdiff_t stride, int h, int x, int y)\

  297. {\

  298.     const int A = (8-x) * (8-y);\

  299.     const int B = (  x) * (8-y);\

  300.     const int C = (8-x) * (  y);\

  301.     const int D = (  x) * (  y);\

  302.     int i;\

  303.     int bias = rv40_bias[y>>1][x>>1];\

  304.     \

  305.     assert(x<8 && y<8 && x>=0 && y>=0);\

  306. \

  307.     if(D){\

  308.         for(i = 0; i < h; i++){\

  309.             OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + bias));\

  310.             OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + bias));\

  311.             OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + bias));\

  312.             OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + bias));\

  313.             dst += stride;\

  314.             src += stride;\

  315.         }\

  316.     }else{\

  317.         const int E = B + C;\

  318.         const ptrdiff_t step = C ? stride : 1;\

  319.         for(i = 0; i < h; i++){\

  320.             OP(dst[0], (A*src[0] + E*src[step+0] + bias));\

  321.             OP(dst[1], (A*src[1] + E*src[step+1] + bias));\

  322.             OP(dst[2], (A*src[2] + E*src[step+2] + bias));\

  323.             OP(dst[3], (A*src[3] + E*src[step+3] + bias));\

  324.             dst += stride;\

  325.             src += stride;\

  326.         }\

  327.     }\

  328. }\

  329. \

  330. static void OPNAME ## rv40_chroma_mc8_c(uint8_t *dst/*align 8*/,\

  331.                                         uint8_t *src/*align 1*/,\

  332.                                         ptrdiff_t stride, int h, int x, int y)\

  333. {\

  334.     const int A = (8-x) * (8-y);\

  335.     const int B = (  x) * (8-y);\

  336.     const int C = (8-x) * (  y);\

  337.     const int D = (  x) * (  y);\

  338.     int i;\

  339.     int bias = rv40_bias[y>>1][x>>1];\

  340.     \

  341.     assert(x<8 && y<8 && x>=0 && y>=0);\

  342. \

  343.     if(D){\

  344.         for(i = 0; i < h; i++){\

  345.             OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + bias));\

  346.             OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + bias));\

  347.             OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + bias));\

  348.             OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + bias));\

  349.             OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + bias));\

  350.             OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + bias));\

  351.             OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + bias));\

  352.             OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + bias));\

  353.             dst += stride;\

  354.             src += stride;\

  355.         }\

  356.     }else{\

  357.         const int E = B + C;\

  358.         const ptrdiff_t step = C ? stride : 1;\

  359.         for(i = 0; i < h; i++){\

  360.             OP(dst[0], (A*src[0] + E*src[step+0] + bias));\

  361.             OP(dst[1], (A*src[1] + E*src[step+1] + bias));\

  362.             OP(dst[2], (A*src[2] + E*src[step+2] + bias));\

  363.             OP(dst[3], (A*src[3] + E*src[step+3] + bias));\

  364.             OP(dst[4], (A*src[4] + E*src[step+4] + bias));\

  365.             OP(dst[5], (A*src[5] + E*src[step+5] + bias));\

  366.             OP(dst[6], (A*src[6] + E*src[step+6] + bias));\

  367.             OP(dst[7], (A*src[7] + E*src[step+7] + bias));\

  368.             dst += stride;\

  369.             src += stride;\

  370.         }\

  371.     }\

  372. }

  373. 

  374. #define op_avg(a, b) a = (((a)+((b)>>6)+1)>>1)

  375. #define op_put(a, b) a = ((b)>>6)

  376. 

  377. RV40_CHROMA_MC(put_, op_put)

  378. RV40_CHROMA_MC(avg_, op_avg)

  379. 

  380. #define RV40_WEIGHT_FUNC(size) \

  381. static void rv40_weight_func_rnd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\

  382. {\

  383.     int i, j;\

  384. \

  385.     for (j = 0; j < size; j++) {\

  386.         for (i = 0; i < size; i++)\

  387.             dst[i] = (((w2 * src1[i]) >> 9) + ((w1 * src2[i]) >> 9) + 0x10) >> 5;\

  388.         src1 += stride;\

  389.         src2 += stride;\

  390.         dst  += stride;\

  391.     }\

  392. }\

  393. static void rv40_weight_func_nornd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\

  394. {\

  395.     int i, j;\

  396. \

  397.     for (j = 0; j < size; j++) {\

  398.         for (i = 0; i < size; i++)\

  399.             dst[i] = (w2 * src1[i] + w1 * src2[i] + 0x10) >> 5;\

  400.         src1 += stride;\

  401.         src2 += stride;\

  402.         dst  += stride;\

  403.     }\

  404. }

  405. 

  406. RV40_WEIGHT_FUNC(16)

  407. RV40_WEIGHT_FUNC(8)

  408. 

  409. /**

  410.  * dither values for deblocking filter - left/top values

  411.  */

  412. static const uint8_t rv40_dither_l[16] = {

  413.     0x40, 0x50, 0x20, 0x60, 0x30, 0x50, 0x40, 0x30,

  414.     0x50, 0x40, 0x50, 0x30, 0x60, 0x20, 0x50, 0x40

  415. };

  416. 

  417. /**

  418.  * dither values for deblocking filter - right/bottom values

  419.  */

  420. static const uint8_t rv40_dither_r[16] = {

  421.     0x40, 0x30, 0x60, 0x20, 0x50, 0x30, 0x30, 0x40,

  422.     0x40, 0x40, 0x50, 0x30, 0x20, 0x60, 0x30, 0x40

  423. };

  424. 

  425. #define CLIP_SYMM(a, b) av_clip(a, -(b), b)

  426. /**

  427.  * weaker deblocking very similar to the one described in 4.4.2 of JVT-A003r1

  428.  */

  429. static av_always_inline void rv40_weak_loop_filter(uint8_t *src,

  430.                                                    const int step,

  431.                                                    const ptrdiff_t stride,

  432.                                                    const int filter_p1,

  433.                                                    const int filter_q1,

  434.                                                    const int alpha,

  435.                                                    const int beta,

  436.                                                    const int lim_p0q0,

  437.                                                    const int lim_q1,

  438.                                                    const int lim_p1)

  439. {

  440.     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;

  441.     int i, t, u, diff;

  442. 

  443.     for (i = 0; i < 4; i++, src += stride) {

  444.         int diff_p1p0 = src[-2*step] - src[-1*step];

  445.         int diff_q1q0 = src[ 1*step] - src[ 0*step];

  446.         int diff_p1p2 = src[-2*step] - src[-3*step];

  447.         int diff_q1q2 = src[ 1*step] - src[ 2*step];

  448. 

  449.         t = src[0*step] - src[-1*step];

  450.         if (!t)

  451.             continue;

  452. 

  453.         u = (alpha * FFABS(t)) >> 7;

  454.         if (u > 3 - (filter_p1 && filter_q1))

  455.             continue;

  456. 

  457.         t <<= 2;

  458.         if (filter_p1 && filter_q1)

  459.             t += src[-2*step] - src[1*step];

  460. 

  461.         diff = CLIP_SYMM((t + 4) >> 3, lim_p0q0);

  462.         src[-1*step] = cm[src[-1*step] + diff];

  463.         src[ 0*step] = cm[src[ 0*step] - diff];

  464. 

  465.         if (filter_p1 && FFABS(diff_p1p2) <= beta) {

  466.             t = (diff_p1p0 + diff_p1p2 - diff) >> 1;

  467.             src[-2*step] = cm[src[-2*step] - CLIP_SYMM(t, lim_p1)];

  468.         }

  469. 

  470.         if (filter_q1 && FFABS(diff_q1q2) <= beta) {

  471.             t = (diff_q1q0 + diff_q1q2 + diff) >> 1;

  472.             src[ 1*step] = cm[src[ 1*step] - CLIP_SYMM(t, lim_q1)];

  473.         }

  474.     }

  475. }

  476. 

  477. static void rv40_h_weak_loop_filter(uint8_t *src, const ptrdiff_t stride,

  478.                                     const int filter_p1, const int filter_q1,

  479.                                     const int alpha, const int beta,

  480.                                     const int lim_p0q0, const int lim_q1,

  481.                                     const int lim_p1)

  482. {

  483.     rv40_weak_loop_filter(src, stride, 1, filter_p1, filter_q1,

  484.                           alpha, beta, lim_p0q0, lim_q1, lim_p1);

  485. }

  486. 

  487. static void rv40_v_weak_loop_filter(uint8_t *src, const ptrdiff_t stride,

  488.                                     const int filter_p1, const int filter_q1,

  489.                                     const int alpha, const int beta,

  490.                                     const int lim_p0q0, const int lim_q1,

  491.                                     const int lim_p1)

  492. {

  493.     rv40_weak_loop_filter(src, 1, stride, filter_p1, filter_q1,

  494.                           alpha, beta, lim_p0q0, lim_q1, lim_p1);

  495. }

  496. 

  497. static av_always_inline void rv40_strong_loop_filter(uint8_t *src,

  498.                                                      const int step,

  499.                                                      const ptrdiff_t stride,

  500.                                                      const int alpha,

  501.                                                      const int lims,

  502.                                                      const int dmode,

  503.                                                      const int chroma)

  504. {

  505.     int i;

  506. 

  507.     for(i = 0; i < 4; i++, src += stride){

  508.         int sflag, p0, q0, p1, q1;

  509.         int t = src[0*step] - src[-1*step];

  510. 

  511.         if (!t)

  512.             continue;

  513. 

  514.         sflag = (alpha * FFABS(t)) >> 7;

  515.         if (sflag > 1)

  516.             continue;

  517. 

  518.         p0 = (25*src[-3*step] + 26*src[-2*step] + 26*src[-1*step] +

  519.               26*src[ 0*step] + 25*src[ 1*step] +

  520.               rv40_dither_l[dmode + i]) >> 7;

  521. 

  522.         q0 = (25*src[-2*step] + 26*src[-1*step] + 26*src[ 0*step] +

  523.               26*src[ 1*step] + 25*src[ 2*step] +

  524.               rv40_dither_r[dmode + i]) >> 7;

  525. 

  526.         if (sflag) {

  527.             p0 = av_clip(p0, src[-1*step] - lims, src[-1*step] + lims);

  528.             q0 = av_clip(q0, src[ 0*step] - lims, src[ 0*step] + lims);

  529.         }

  530. 

  531.         p1 = (25*src[-4*step] + 26*src[-3*step] + 26*src[-2*step] + 26*p0 +

  532.               25*src[ 0*step] + rv40_dither_l[dmode + i]) >> 7;

  533.         q1 = (25*src[-1*step] + 26*q0 + 26*src[ 1*step] + 26*src[ 2*step] +

  534.               25*src[ 3*step] + rv40_dither_r[dmode + i]) >> 7;

  535. 

  536.         if (sflag) {

  537.             p1 = av_clip(p1, src[-2*step] - lims, src[-2*step] + lims);

  538.             q1 = av_clip(q1, src[ 1*step] - lims, src[ 1*step] + lims);

  539.         }

  540. 

  541.         src[-2*step] = p1;

  542.         src[-1*step] = p0;

  543.         src[ 0*step] = q0;

  544.         src[ 1*step] = q1;

  545. 

  546.         if(!chroma){

  547.             src[-3*step] = (25*src[-1*step] + 26*src[-2*step] +

  548.                             51*src[-3*step] + 26*src[-4*step] + 64) >> 7;

  549.             src[ 2*step] = (25*src[ 0*step] + 26*src[ 1*step] +

  550.                             51*src[ 2*step] + 26*src[ 3*step] + 64) >> 7;

  551.         }

  552.     }

  553. }

  554. 

  555. static void rv40_h_strong_loop_filter(uint8_t *src, const ptrdiff_t stride,

  556.                                       const int alpha, const int lims,

  557.                                       const int dmode, const int chroma)

  558. {

  559.     rv40_strong_loop_filter(src, stride, 1, alpha, lims, dmode, chroma);

  560. }

  561. 

  562. static void rv40_v_strong_loop_filter(uint8_t *src, const ptrdiff_t stride,

  563.                                       const int alpha, const int lims,

  564.                                       const int dmode, const int chroma)

  565. {

  566.     rv40_strong_loop_filter(src, 1, stride, alpha, lims, dmode, chroma);

  567. }

  568. 

  569. static av_always_inline int rv40_loop_filter_strength(uint8_t *src,

  570.                                                       int step, ptrdiff_t stride,

  571.                                                       int beta, int beta2,

  572.                                                       int edge,

  573.                                                       int *p1, int *q1)

  574. {

  575.     int sum_p1p0 = 0, sum_q1q0 = 0, sum_p1p2 = 0, sum_q1q2 = 0;

  576.     int strong0 = 0, strong1 = 0;

  577.     uint8_t *ptr;

  578.     int i;

  579. 

  580.     for (i = 0, ptr = src; i < 4; i++, ptr += stride) {

  581.         sum_p1p0 += ptr[-2*step] - ptr[-1*step];

  582.         sum_q1q0 += ptr[ 1*step] - ptr[ 0*step];

  583.     }

  584. 

  585.     *p1 = FFABS(sum_p1p0) < (beta << 2);

  586.     *q1 = FFABS(sum_q1q0) < (beta << 2);

  587. 

  588.     if(!*p1 && !*q1)

  589.         return 0;

  590. 

  591.     if (!edge)

  592.         return 0;

  593. 

  594.     for (i = 0, ptr = src; i < 4; i++, ptr += stride) {

  595.         sum_p1p2 += ptr[-2*step] - ptr[-3*step];

  596.         sum_q1q2 += ptr[ 1*step] - ptr[ 2*step];

  597.     }

  598. 

  599.     strong0 = *p1 && (FFABS(sum_p1p2) < beta2);

  600.     strong1 = *q1 && (FFABS(sum_q1q2) < beta2);

  601. 

  602.     return strong0 && strong1;

  603. }

  604. 

  605. static int rv40_h_loop_filter_strength(uint8_t *src, ptrdiff_t stride,

  606.                                        int beta, int beta2, int edge,

  607.                                        int *p1, int *q1)

  608. {

  609.     return rv40_loop_filter_strength(src, stride, 1, beta, beta2, edge, p1, q1);

  610. }

  611. 

  612. static int rv40_v_loop_filter_strength(uint8_t *src, ptrdiff_t stride,

  613.                                        int beta, int beta2, int edge,

  614.                                        int *p1, int *q1)

  615. {

  616.     return rv40_loop_filter_strength(src, 1, stride, beta, beta2, edge, p1, q1);

  617. }

  618. 

  619. av_cold void ff_rv40dsp_init(RV34DSPContext *c)

  620. {

  621.     H264QpelContext qpel;

  622. 

  623.     ff_rv34dsp_init(c);

  624.     ff_h264qpel_init(&qpel, 8);

  625. 

  626.     c->put_pixels_tab[0][ 0] = qpel.put_h264_qpel_pixels_tab[0][0];

  627.     c->put_pixels_tab[0][ 1] = put_rv40_qpel16_mc10_c;

  628.     c->put_pixels_tab[0][ 2] = qpel.put_h264_qpel_pixels_tab[0][2];

  629.     c->put_pixels_tab[0][ 3] = put_rv40_qpel16_mc30_c;

  630.     c->put_pixels_tab[0][ 4] = put_rv40_qpel16_mc01_c;

  631.     c->put_pixels_tab[0][ 5] = put_rv40_qpel16_mc11_c;

  632.     c->put_pixels_tab[0][ 6] = put_rv40_qpel16_mc21_c;

  633.     c->put_pixels_tab[0][ 7] = put_rv40_qpel16_mc31_c;

  634.     c->put_pixels_tab[0][ 8] = qpel.put_h264_qpel_pixels_tab[0][8];

  635.     c->put_pixels_tab[0][ 9] = put_rv40_qpel16_mc12_c;

  636.     c->put_pixels_tab[0][10] = put_rv40_qpel16_mc22_c;

  637.     c->put_pixels_tab[0][11] = put_rv40_qpel16_mc32_c;

  638.     c->put_pixels_tab[0][12] = put_rv40_qpel16_mc03_c;

  639.     c->put_pixels_tab[0][13] = put_rv40_qpel16_mc13_c;

  640.     c->put_pixels_tab[0][14] = put_rv40_qpel16_mc23_c;

  641.     c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_c;

  642.     c->avg_pixels_tab[0][ 0] = qpel.avg_h264_qpel_pixels_tab[0][0];

  643.     c->avg_pixels_tab[0][ 1] = avg_rv40_qpel16_mc10_c;

  644.     c->avg_pixels_tab[0][ 2] = qpel.avg_h264_qpel_pixels_tab[0][2];

  645.     c->avg_pixels_tab[0][ 3] = avg_rv40_qpel16_mc30_c;

  646.     c->avg_pixels_tab[0][ 4] = avg_rv40_qpel16_mc01_c;

  647.     c->avg_pixels_tab[0][ 5] = avg_rv40_qpel16_mc11_c;

  648.     c->avg_pixels_tab[0][ 6] = avg_rv40_qpel16_mc21_c;

  649.     c->avg_pixels_tab[0][ 7] = avg_rv40_qpel16_mc31_c;

  650.     c->avg_pixels_tab[0][ 8] = qpel.avg_h264_qpel_pixels_tab[0][8];

  651.     c->avg_pixels_tab[0][ 9] = avg_rv40_qpel16_mc12_c;

  652.     c->avg_pixels_tab[0][10] = avg_rv40_qpel16_mc22_c;

  653.     c->avg_pixels_tab[0][11] = avg_rv40_qpel16_mc32_c;

  654.     c->avg_pixels_tab[0][12] = avg_rv40_qpel16_mc03_c;

  655.     c->avg_pixels_tab[0][13] = avg_rv40_qpel16_mc13_c;

  656.     c->avg_pixels_tab[0][14] = avg_rv40_qpel16_mc23_c;

  657.     c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_c;

  658.     c->put_pixels_tab[1][ 0] = qpel.put_h264_qpel_pixels_tab[1][0];

  659.     c->put_pixels_tab[1][ 1] = put_rv40_qpel8_mc10_c;

  660.     c->put_pixels_tab[1][ 2] = qpel.put_h264_qpel_pixels_tab[1][2];

  661.     c->put_pixels_tab[1][ 3] = put_rv40_qpel8_mc30_c;

  662.     c->put_pixels_tab[1][ 4] = put_rv40_qpel8_mc01_c;

  663.     c->put_pixels_tab[1][ 5] = put_rv40_qpel8_mc11_c;

  664.     c->put_pixels_tab[1][ 6] = put_rv40_qpel8_mc21_c;

  665.     c->put_pixels_tab[1][ 7] = put_rv40_qpel8_mc31_c;

  666.     c->put_pixels_tab[1][ 8] = qpel.put_h264_qpel_pixels_tab[1][8];

  667.     c->put_pixels_tab[1][ 9] = put_rv40_qpel8_mc12_c;

  668.     c->put_pixels_tab[1][10] = put_rv40_qpel8_mc22_c;

  669.     c->put_pixels_tab[1][11] = put_rv40_qpel8_mc32_c;

  670.     c->put_pixels_tab[1][12] = put_rv40_qpel8_mc03_c;

  671.     c->put_pixels_tab[1][13] = put_rv40_qpel8_mc13_c;

  672.     c->put_pixels_tab[1][14] = put_rv40_qpel8_mc23_c;

  673.     c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_c;

  674.     c->avg_pixels_tab[1][ 0] = qpel.avg_h264_qpel_pixels_tab[1][0];

  675.     c->avg_pixels_tab[1][ 1] = avg_rv40_qpel8_mc10_c;

  676.     c->avg_pixels_tab[1][ 2] = qpel.avg_h264_qpel_pixels_tab[1][2];

  677.     c->avg_pixels_tab[1][ 3] = avg_rv40_qpel8_mc30_c;

  678.     c->avg_pixels_tab[1][ 4] = avg_rv40_qpel8_mc01_c;

  679.     c->avg_pixels_tab[1][ 5] = avg_rv40_qpel8_mc11_c;

  680.     c->avg_pixels_tab[1][ 6] = avg_rv40_qpel8_mc21_c;

  681.     c->avg_pixels_tab[1][ 7] = avg_rv40_qpel8_mc31_c;

  682.     c->avg_pixels_tab[1][ 8] = qpel.avg_h264_qpel_pixels_tab[1][8];

  683.     c->avg_pixels_tab[1][ 9] = avg_rv40_qpel8_mc12_c;

  684.     c->avg_pixels_tab[1][10] = avg_rv40_qpel8_mc22_c;

  685.     c->avg_pixels_tab[1][11] = avg_rv40_qpel8_mc32_c;

  686.     c->avg_pixels_tab[1][12] = avg_rv40_qpel8_mc03_c;

  687.     c->avg_pixels_tab[1][13] = avg_rv40_qpel8_mc13_c;

  688.     c->avg_pixels_tab[1][14] = avg_rv40_qpel8_mc23_c;

  689.     c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_c;

  690. 

  691.     c->put_chroma_pixels_tab[0] = put_rv40_chroma_mc8_c;

  692.     c->put_chroma_pixels_tab[1] = put_rv40_chroma_mc4_c;

  693.     c->avg_chroma_pixels_tab[0] = avg_rv40_chroma_mc8_c;

  694.     c->avg_chroma_pixels_tab[1] = avg_rv40_chroma_mc4_c;

  695. 

  696.     c->rv40_weight_pixels_tab[0][0] = rv40_weight_func_rnd_16;

  697.     c->rv40_weight_pixels_tab[0][1] = rv40_weight_func_rnd_8;

  698.     c->rv40_weight_pixels_tab[1][0] = rv40_weight_func_nornd_16;

  699.     c->rv40_weight_pixels_tab[1][1] = rv40_weight_func_nornd_8;

  700. 

  701.     c->rv40_weak_loop_filter[0]     = rv40_h_weak_loop_filter;

  702.     c->rv40_weak_loop_filter[1]     = rv40_v_weak_loop_filter;

  703.     c->rv40_strong_loop_filter[0]   = rv40_h_strong_loop_filter;

  704.     c->rv40_strong_loop_filter[1]   = rv40_v_strong_loop_filter;

  705.     c->rv40_loop_filter_strength[0] = rv40_h_loop_filter_strength;

  706.     c->rv40_loop_filter_strength[1] = rv40_v_loop_filter_strength;

  707. 

  708.     if (ARCH_AARCH64)

  709.         ff_rv40dsp_init_aarch64(c);

  710.     if (ARCH_ARM)

  711.         ff_rv40dsp_init_arm(c);

  712.     if (ARCH_X86)

  713.         ff_rv40dsp_init_x86(c);

  714. }