| 1 | /* |
| 2 | * RV40 decoder motion compensation functions |
| 3 | * Copyright (c) 2008 Konstantin Shishkov |
| 4 | * |
| 5 | * This file is part of FFmpeg. |
| 6 | * |
| 7 | * FFmpeg 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 | * FFmpeg 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 FFmpeg; 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 | #include "libavutil/avassert.h" |
| 36 | |
| 37 | #define RV40_LOWPASS(OPNAME, OP) \ |
| 38 | static void OPNAME ## rv40_qpel8_h_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\ |
| 39 | const int h, const int C1, const int C2, const int SHIFT){\ |
| 40 | const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;\ |
| 41 | int i;\ |
| 42 | for(i = 0; i < h; i++)\ |
| 43 | {\ |
| 44 | OP(dst[0], (src[-2] + src[ 3] - 5*(src[-1]+src[2]) + src[0]*C1 + src[1]*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 45 | OP(dst[1], (src[-1] + src[ 4] - 5*(src[ 0]+src[3]) + src[1]*C1 + src[2]*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 46 | OP(dst[2], (src[ 0] + src[ 5] - 5*(src[ 1]+src[4]) + src[2]*C1 + src[3]*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 47 | OP(dst[3], (src[ 1] + src[ 6] - 5*(src[ 2]+src[5]) + src[3]*C1 + src[4]*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 48 | OP(dst[4], (src[ 2] + src[ 7] - 5*(src[ 3]+src[6]) + src[4]*C1 + src[5]*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 49 | OP(dst[5], (src[ 3] + src[ 8] - 5*(src[ 4]+src[7]) + src[5]*C1 + src[6]*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 50 | OP(dst[6], (src[ 4] + src[ 9] - 5*(src[ 5]+src[8]) + src[6]*C1 + src[7]*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 51 | OP(dst[7], (src[ 5] + src[10] - 5*(src[ 6]+src[9]) + src[7]*C1 + src[8]*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 52 | dst += dstStride;\ |
| 53 | src += srcStride;\ |
| 54 | }\ |
| 55 | }\ |
| 56 | \ |
| 57 | static void OPNAME ## rv40_qpel8_v_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\ |
| 58 | const int w, const int C1, const int C2, const int SHIFT){\ |
| 59 | const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;\ |
| 60 | int i;\ |
| 61 | for(i = 0; i < w; i++)\ |
| 62 | {\ |
| 63 | const int srcB = src[-2*srcStride];\ |
| 64 | const int srcA = src[-1*srcStride];\ |
| 65 | const int src0 = src[0 *srcStride];\ |
| 66 | const int src1 = src[1 *srcStride];\ |
| 67 | const int src2 = src[2 *srcStride];\ |
| 68 | const int src3 = src[3 *srcStride];\ |
| 69 | const int src4 = src[4 *srcStride];\ |
| 70 | const int src5 = src[5 *srcStride];\ |
| 71 | const int src6 = src[6 *srcStride];\ |
| 72 | const int src7 = src[7 *srcStride];\ |
| 73 | const int src8 = src[8 *srcStride];\ |
| 74 | const int src9 = src[9 *srcStride];\ |
| 75 | const int src10 = src[10*srcStride];\ |
| 76 | OP(dst[0*dstStride], (srcB + src3 - 5*(srcA+src2) + src0*C1 + src1*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 77 | OP(dst[1*dstStride], (srcA + src4 - 5*(src0+src3) + src1*C1 + src2*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 78 | OP(dst[2*dstStride], (src0 + src5 - 5*(src1+src4) + src2*C1 + src3*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 79 | OP(dst[3*dstStride], (src1 + src6 - 5*(src2+src5) + src3*C1 + src4*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 80 | OP(dst[4*dstStride], (src2 + src7 - 5*(src3+src6) + src4*C1 + src5*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 81 | OP(dst[5*dstStride], (src3 + src8 - 5*(src4+src7) + src5*C1 + src6*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 82 | OP(dst[6*dstStride], (src4 + src9 - 5*(src5+src8) + src6*C1 + src7*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 83 | OP(dst[7*dstStride], (src5 + src10 - 5*(src6+src9) + src7*C1 + src8*C2 + (1<<(SHIFT-1))) >> SHIFT);\ |
| 84 | dst++;\ |
| 85 | src++;\ |
| 86 | }\ |
| 87 | }\ |
| 88 | \ |
| 89 | static void OPNAME ## rv40_qpel16_v_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\ |
| 90 | const int w, const int C1, const int C2, const int SHIFT){\ |
| 91 | OPNAME ## rv40_qpel8_v_lowpass(dst , src , dstStride, srcStride, 8, C1, C2, SHIFT);\ |
| 92 | OPNAME ## rv40_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride, 8, C1, C2, SHIFT);\ |
| 93 | src += 8*srcStride;\ |
| 94 | dst += 8*dstStride;\ |
| 95 | OPNAME ## rv40_qpel8_v_lowpass(dst , src , dstStride, srcStride, w-8, C1, C2, SHIFT);\ |
| 96 | OPNAME ## rv40_qpel8_v_lowpass(dst+8, src+8, dstStride, srcStride, w-8, C1, C2, SHIFT);\ |
| 97 | }\ |
| 98 | \ |
| 99 | static void OPNAME ## rv40_qpel16_h_lowpass(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride,\ |
| 100 | const int h, const int C1, const int C2, const int SHIFT){\ |
| 101 | OPNAME ## rv40_qpel8_h_lowpass(dst , src , dstStride, srcStride, 8, C1, C2, SHIFT);\ |
| 102 | OPNAME ## rv40_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride, 8, C1, C2, SHIFT);\ |
| 103 | src += 8*srcStride;\ |
| 104 | dst += 8*dstStride;\ |
| 105 | OPNAME ## rv40_qpel8_h_lowpass(dst , src , dstStride, srcStride, h-8, C1, C2, SHIFT);\ |
| 106 | OPNAME ## rv40_qpel8_h_lowpass(dst+8, src+8, dstStride, srcStride, h-8, C1, C2, SHIFT);\ |
| 107 | }\ |
| 108 | \ |
| 109 | |
| 110 | #define RV40_MC(OPNAME, SIZE) \ |
| 111 | static void OPNAME ## rv40_qpel ## SIZE ## _mc10_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 112 | {\ |
| 113 | OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\ |
| 114 | }\ |
| 115 | \ |
| 116 | static void OPNAME ## rv40_qpel ## SIZE ## _mc30_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 117 | {\ |
| 118 | OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\ |
| 119 | }\ |
| 120 | \ |
| 121 | static void OPNAME ## rv40_qpel ## SIZE ## _mc01_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 122 | {\ |
| 123 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\ |
| 124 | }\ |
| 125 | \ |
| 126 | static void OPNAME ## rv40_qpel ## SIZE ## _mc11_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 127 | {\ |
| 128 | uint8_t full[SIZE*(SIZE+5)];\ |
| 129 | uint8_t * const full_mid = full + SIZE*2;\ |
| 130 | put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\ |
| 131 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\ |
| 132 | }\ |
| 133 | \ |
| 134 | static void OPNAME ## rv40_qpel ## SIZE ## _mc21_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 135 | {\ |
| 136 | uint8_t full[SIZE*(SIZE+5)];\ |
| 137 | uint8_t * const full_mid = full + SIZE*2;\ |
| 138 | put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\ |
| 139 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\ |
| 140 | }\ |
| 141 | \ |
| 142 | static void OPNAME ## rv40_qpel ## SIZE ## _mc31_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 143 | {\ |
| 144 | uint8_t full[SIZE*(SIZE+5)];\ |
| 145 | uint8_t * const full_mid = full + SIZE*2;\ |
| 146 | put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 52, 6);\ |
| 147 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 52, 20, 6);\ |
| 148 | }\ |
| 149 | \ |
| 150 | static void OPNAME ## rv40_qpel ## SIZE ## _mc12_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 151 | {\ |
| 152 | uint8_t full[SIZE*(SIZE+5)];\ |
| 153 | uint8_t * const full_mid = full + SIZE*2;\ |
| 154 | put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\ |
| 155 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\ |
| 156 | }\ |
| 157 | \ |
| 158 | static void OPNAME ## rv40_qpel ## SIZE ## _mc22_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 159 | {\ |
| 160 | uint8_t full[SIZE*(SIZE+5)];\ |
| 161 | uint8_t * const full_mid = full + SIZE*2;\ |
| 162 | put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\ |
| 163 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\ |
| 164 | }\ |
| 165 | \ |
| 166 | static void OPNAME ## rv40_qpel ## SIZE ## _mc32_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 167 | {\ |
| 168 | uint8_t full[SIZE*(SIZE+5)];\ |
| 169 | uint8_t * const full_mid = full + SIZE*2;\ |
| 170 | put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 52, 6);\ |
| 171 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 20, 5);\ |
| 172 | }\ |
| 173 | \ |
| 174 | static void OPNAME ## rv40_qpel ## SIZE ## _mc03_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 175 | {\ |
| 176 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\ |
| 177 | }\ |
| 178 | \ |
| 179 | static void OPNAME ## rv40_qpel ## SIZE ## _mc13_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 180 | {\ |
| 181 | uint8_t full[SIZE*(SIZE+5)];\ |
| 182 | uint8_t * const full_mid = full + SIZE*2;\ |
| 183 | put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 52, 20, 6);\ |
| 184 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 52, 6);\ |
| 185 | }\ |
| 186 | \ |
| 187 | static void OPNAME ## rv40_qpel ## SIZE ## _mc23_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\ |
| 188 | {\ |
| 189 | uint8_t full[SIZE*(SIZE+5)];\ |
| 190 | uint8_t * const full_mid = full + SIZE*2;\ |
| 191 | put_rv40_qpel ## SIZE ## _h_lowpass(full, src - 2*stride, SIZE, stride, SIZE+5, 20, 20, 5);\ |
| 192 | OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE, SIZE, 20, 52, 6);\ |
| 193 | }\ |
| 194 | \ |
| 195 | |
| 196 | #define op_avg(a, b) a = (((a)+cm[b]+1)>>1) |
| 197 | #define op_put(a, b) a = cm[b] |
| 198 | |
| 199 | RV40_LOWPASS(put_ , op_put) |
| 200 | RV40_LOWPASS(avg_ , op_avg) |
| 201 | |
| 202 | #undef op_avg |
| 203 | #undef op_put |
| 204 | |
| 205 | RV40_MC(put_, 8) |
| 206 | RV40_MC(put_, 16) |
| 207 | RV40_MC(avg_, 8) |
| 208 | RV40_MC(avg_, 16) |
| 209 | |
| 210 | #define PIXOP2(OPNAME, OP) \ |
| 211 | static inline void OPNAME ## _pixels8_xy2_8_c(uint8_t *block, \ |
| 212 | const uint8_t *pixels, \ |
| 213 | ptrdiff_t line_size, \ |
| 214 | int h) \ |
| 215 | { \ |
| 216 | /* FIXME HIGH BIT DEPTH */ \ |
| 217 | int j; \ |
| 218 | \ |
| 219 | for (j = 0; j < 2; j++) { \ |
| 220 | int i; \ |
| 221 | const uint32_t a = AV_RN32(pixels); \ |
| 222 | const uint32_t b = AV_RN32(pixels + 1); \ |
| 223 | uint32_t l0 = (a & 0x03030303UL) + \ |
| 224 | (b & 0x03030303UL) + \ |
| 225 | 0x02020202UL; \ |
| 226 | uint32_t h0 = ((a & 0xFCFCFCFCUL) >> 2) + \ |
| 227 | ((b & 0xFCFCFCFCUL) >> 2); \ |
| 228 | uint32_t l1, h1; \ |
| 229 | \ |
| 230 | pixels += line_size; \ |
| 231 | for (i = 0; i < h; i += 2) { \ |
| 232 | uint32_t a = AV_RN32(pixels); \ |
| 233 | uint32_t b = AV_RN32(pixels + 1); \ |
| 234 | l1 = (a & 0x03030303UL) + \ |
| 235 | (b & 0x03030303UL); \ |
| 236 | h1 = ((a & 0xFCFCFCFCUL) >> 2) + \ |
| 237 | ((b & 0xFCFCFCFCUL) >> 2); \ |
| 238 | OP(*((uint32_t *) block), \ |
| 239 | h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL)); \ |
| 240 | pixels += line_size; \ |
| 241 | block += line_size; \ |
| 242 | a = AV_RN32(pixels); \ |
| 243 | b = AV_RN32(pixels + 1); \ |
| 244 | l0 = (a & 0x03030303UL) + \ |
| 245 | (b & 0x03030303UL) + \ |
| 246 | 0x02020202UL; \ |
| 247 | h0 = ((a & 0xFCFCFCFCUL) >> 2) + \ |
| 248 | ((b & 0xFCFCFCFCUL) >> 2); \ |
| 249 | OP(*((uint32_t *) block), \ |
| 250 | h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL)); \ |
| 251 | pixels += line_size; \ |
| 252 | block += line_size; \ |
| 253 | } \ |
| 254 | pixels += 4 - line_size * (h + 1); \ |
| 255 | block += 4 - line_size * h; \ |
| 256 | } \ |
| 257 | } \ |
| 258 | \ |
| 259 | CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_8_c, \ |
| 260 | OPNAME ## _pixels8_xy2_8_c, \ |
| 261 | 8) \ |
| 262 | |
| 263 | #define op_avg(a, b) a = rnd_avg32(a, b) |
| 264 | #define op_put(a, b) a = b |
| 265 | PIXOP2(avg, op_avg) |
| 266 | PIXOP2(put, op_put) |
| 267 | #undef op_avg |
| 268 | #undef op_put |
| 269 | |
| 270 | static void put_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride) |
| 271 | { |
| 272 | put_pixels16_xy2_8_c(dst, src, stride, 16); |
| 273 | } |
| 274 | static void avg_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride) |
| 275 | { |
| 276 | avg_pixels16_xy2_8_c(dst, src, stride, 16); |
| 277 | } |
| 278 | static void put_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride) |
| 279 | { |
| 280 | put_pixels8_xy2_8_c(dst, src, stride, 8); |
| 281 | } |
| 282 | static void avg_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride) |
| 283 | { |
| 284 | avg_pixels8_xy2_8_c(dst, src, stride, 8); |
| 285 | } |
| 286 | |
| 287 | static const int rv40_bias[4][4] = { |
| 288 | { 0, 16, 32, 16 }, |
| 289 | { 32, 28, 32, 28 }, |
| 290 | { 0, 32, 16, 32 }, |
| 291 | { 32, 28, 32, 28 } |
| 292 | }; |
| 293 | |
| 294 | #define RV40_CHROMA_MC(OPNAME, OP)\ |
| 295 | static void OPNAME ## rv40_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\ |
| 296 | const int A = (8-x) * (8-y);\ |
| 297 | const int B = ( x) * (8-y);\ |
| 298 | const int C = (8-x) * ( y);\ |
| 299 | const int D = ( x) * ( y);\ |
| 300 | int i;\ |
| 301 | int bias = rv40_bias[y>>1][x>>1];\ |
| 302 | \ |
| 303 | av_assert2(x<8 && y<8 && x>=0 && y>=0);\ |
| 304 | \ |
| 305 | if(D){\ |
| 306 | for(i = 0; i < h; i++){\ |
| 307 | OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + bias));\ |
| 308 | OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + bias));\ |
| 309 | OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + bias));\ |
| 310 | OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + bias));\ |
| 311 | dst += stride;\ |
| 312 | src += stride;\ |
| 313 | }\ |
| 314 | }else{\ |
| 315 | const int E = B + C;\ |
| 316 | const int step = C ? stride : 1;\ |
| 317 | for(i = 0; i < h; i++){\ |
| 318 | OP(dst[0], (A*src[0] + E*src[step+0] + bias));\ |
| 319 | OP(dst[1], (A*src[1] + E*src[step+1] + bias));\ |
| 320 | OP(dst[2], (A*src[2] + E*src[step+2] + bias));\ |
| 321 | OP(dst[3], (A*src[3] + E*src[step+3] + bias));\ |
| 322 | dst += stride;\ |
| 323 | src += stride;\ |
| 324 | }\ |
| 325 | }\ |
| 326 | }\ |
| 327 | \ |
| 328 | static void OPNAME ## rv40_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\ |
| 329 | const int A = (8-x) * (8-y);\ |
| 330 | const int B = ( x) * (8-y);\ |
| 331 | const int C = (8-x) * ( y);\ |
| 332 | const int D = ( x) * ( y);\ |
| 333 | int i;\ |
| 334 | int bias = rv40_bias[y>>1][x>>1];\ |
| 335 | \ |
| 336 | av_assert2(x<8 && y<8 && x>=0 && y>=0);\ |
| 337 | \ |
| 338 | if(D){\ |
| 339 | for(i = 0; i < h; i++){\ |
| 340 | OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + bias));\ |
| 341 | OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + bias));\ |
| 342 | OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + bias));\ |
| 343 | OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + bias));\ |
| 344 | OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + bias));\ |
| 345 | OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + bias));\ |
| 346 | OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + bias));\ |
| 347 | OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + bias));\ |
| 348 | dst += stride;\ |
| 349 | src += stride;\ |
| 350 | }\ |
| 351 | }else{\ |
| 352 | const int E = B + C;\ |
| 353 | const int step = C ? stride : 1;\ |
| 354 | for(i = 0; i < h; i++){\ |
| 355 | OP(dst[0], (A*src[0] + E*src[step+0] + bias));\ |
| 356 | OP(dst[1], (A*src[1] + E*src[step+1] + bias));\ |
| 357 | OP(dst[2], (A*src[2] + E*src[step+2] + bias));\ |
| 358 | OP(dst[3], (A*src[3] + E*src[step+3] + bias));\ |
| 359 | OP(dst[4], (A*src[4] + E*src[step+4] + bias));\ |
| 360 | OP(dst[5], (A*src[5] + E*src[step+5] + bias));\ |
| 361 | OP(dst[6], (A*src[6] + E*src[step+6] + bias));\ |
| 362 | OP(dst[7], (A*src[7] + E*src[step+7] + bias));\ |
| 363 | dst += stride;\ |
| 364 | src += stride;\ |
| 365 | }\ |
| 366 | }\ |
| 367 | } |
| 368 | |
| 369 | #define op_avg(a, b) a = (((a)+((b)>>6)+1)>>1) |
| 370 | #define op_put(a, b) a = ((b)>>6) |
| 371 | |
| 372 | RV40_CHROMA_MC(put_, op_put) |
| 373 | RV40_CHROMA_MC(avg_, op_avg) |
| 374 | |
| 375 | #define RV40_WEIGHT_FUNC(size) \ |
| 376 | static void rv40_weight_func_rnd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\ |
| 377 | {\ |
| 378 | int i, j;\ |
| 379 | \ |
| 380 | for (j = 0; j < size; j++) {\ |
| 381 | for (i = 0; i < size; i++)\ |
| 382 | dst[i] = (((w2 * src1[i]) >> 9) + ((w1 * src2[i]) >> 9) + 0x10) >> 5;\ |
| 383 | src1 += stride;\ |
| 384 | src2 += stride;\ |
| 385 | dst += stride;\ |
| 386 | }\ |
| 387 | }\ |
| 388 | static void rv40_weight_func_nornd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\ |
| 389 | {\ |
| 390 | int i, j;\ |
| 391 | \ |
| 392 | for (j = 0; j < size; j++) {\ |
| 393 | for (i = 0; i < size; i++)\ |
| 394 | dst[i] = (w2 * src1[i] + w1 * src2[i] + 0x10) >> 5;\ |
| 395 | src1 += stride;\ |
| 396 | src2 += stride;\ |
| 397 | dst += stride;\ |
| 398 | }\ |
| 399 | } |
| 400 | |
| 401 | RV40_WEIGHT_FUNC(16) |
| 402 | RV40_WEIGHT_FUNC(8) |
| 403 | |
| 404 | /** |
| 405 | * dither values for deblocking filter - left/top values |
| 406 | */ |
| 407 | static const uint8_t rv40_dither_l[16] = { |
| 408 | 0x40, 0x50, 0x20, 0x60, 0x30, 0x50, 0x40, 0x30, |
| 409 | 0x50, 0x40, 0x50, 0x30, 0x60, 0x20, 0x50, 0x40 |
| 410 | }; |
| 411 | |
| 412 | /** |
| 413 | * dither values for deblocking filter - right/bottom values |
| 414 | */ |
| 415 | static const uint8_t rv40_dither_r[16] = { |
| 416 | 0x40, 0x30, 0x60, 0x20, 0x50, 0x30, 0x30, 0x40, |
| 417 | 0x40, 0x40, 0x50, 0x30, 0x20, 0x60, 0x30, 0x40 |
| 418 | }; |
| 419 | |
| 420 | #define CLIP_SYMM(a, b) av_clip(a, -(b), b) |
| 421 | /** |
| 422 | * weaker deblocking very similar to the one described in 4.4.2 of JVT-A003r1 |
| 423 | */ |
| 424 | static av_always_inline void rv40_weak_loop_filter(uint8_t *src, |
| 425 | const int step, |
| 426 | const ptrdiff_t stride, |
| 427 | const int filter_p1, |
| 428 | const int filter_q1, |
| 429 | const int alpha, |
| 430 | const int beta, |
| 431 | const int lim_p0q0, |
| 432 | const int lim_q1, |
| 433 | const int lim_p1) |
| 434 | { |
| 435 | const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP; |
| 436 | int i, t, u, diff; |
| 437 | |
| 438 | for (i = 0; i < 4; i++, src += stride) { |
| 439 | int diff_p1p0 = src[-2*step] - src[-1*step]; |
| 440 | int diff_q1q0 = src[ 1*step] - src[ 0*step]; |
| 441 | int diff_p1p2 = src[-2*step] - src[-3*step]; |
| 442 | int diff_q1q2 = src[ 1*step] - src[ 2*step]; |
| 443 | |
| 444 | t = src[0*step] - src[-1*step]; |
| 445 | if (!t) |
| 446 | continue; |
| 447 | |
| 448 | u = (alpha * FFABS(t)) >> 7; |
| 449 | if (u > 3 - (filter_p1 && filter_q1)) |
| 450 | continue; |
| 451 | |
| 452 | t <<= 2; |
| 453 | if (filter_p1 && filter_q1) |
| 454 | t += src[-2*step] - src[1*step]; |
| 455 | |
| 456 | diff = CLIP_SYMM((t + 4) >> 3, lim_p0q0); |
| 457 | src[-1*step] = cm[src[-1*step] + diff]; |
| 458 | src[ 0*step] = cm[src[ 0*step] - diff]; |
| 459 | |
| 460 | if (filter_p1 && FFABS(diff_p1p2) <= beta) { |
| 461 | t = (diff_p1p0 + diff_p1p2 - diff) >> 1; |
| 462 | src[-2*step] = cm[src[-2*step] - CLIP_SYMM(t, lim_p1)]; |
| 463 | } |
| 464 | |
| 465 | if (filter_q1 && FFABS(diff_q1q2) <= beta) { |
| 466 | t = (diff_q1q0 + diff_q1q2 + diff) >> 1; |
| 467 | src[ 1*step] = cm[src[ 1*step] - CLIP_SYMM(t, lim_q1)]; |
| 468 | } |
| 469 | } |
| 470 | } |
| 471 | |
| 472 | static void rv40_h_weak_loop_filter(uint8_t *src, const ptrdiff_t stride, |
| 473 | const int filter_p1, const int filter_q1, |
| 474 | const int alpha, const int beta, |
| 475 | const int lim_p0q0, const int lim_q1, |
| 476 | const int lim_p1) |
| 477 | { |
| 478 | rv40_weak_loop_filter(src, stride, 1, filter_p1, filter_q1, |
| 479 | alpha, beta, lim_p0q0, lim_q1, lim_p1); |
| 480 | } |
| 481 | |
| 482 | static void rv40_v_weak_loop_filter(uint8_t *src, const ptrdiff_t stride, |
| 483 | const int filter_p1, const int filter_q1, |
| 484 | const int alpha, const int beta, |
| 485 | const int lim_p0q0, const int lim_q1, |
| 486 | const int lim_p1) |
| 487 | { |
| 488 | rv40_weak_loop_filter(src, 1, stride, filter_p1, filter_q1, |
| 489 | alpha, beta, lim_p0q0, lim_q1, lim_p1); |
| 490 | } |
| 491 | |
| 492 | static av_always_inline void rv40_strong_loop_filter(uint8_t *src, |
| 493 | const int step, |
| 494 | const ptrdiff_t stride, |
| 495 | const int alpha, |
| 496 | const int lims, |
| 497 | const int dmode, |
| 498 | const int chroma) |
| 499 | { |
| 500 | int i; |
| 501 | |
| 502 | for(i = 0; i < 4; i++, src += stride){ |
| 503 | int sflag, p0, q0, p1, q1; |
| 504 | int t = src[0*step] - src[-1*step]; |
| 505 | |
| 506 | if (!t) |
| 507 | continue; |
| 508 | |
| 509 | sflag = (alpha * FFABS(t)) >> 7; |
| 510 | if (sflag > 1) |
| 511 | continue; |
| 512 | |
| 513 | p0 = (25*src[-3*step] + 26*src[-2*step] + 26*src[-1*step] + |
| 514 | 26*src[ 0*step] + 25*src[ 1*step] + |
| 515 | rv40_dither_l[dmode + i]) >> 7; |
| 516 | |
| 517 | q0 = (25*src[-2*step] + 26*src[-1*step] + 26*src[ 0*step] + |
| 518 | 26*src[ 1*step] + 25*src[ 2*step] + |
| 519 | rv40_dither_r[dmode + i]) >> 7; |
| 520 | |
| 521 | if (sflag) { |
| 522 | p0 = av_clip(p0, src[-1*step] - lims, src[-1*step] + lims); |
| 523 | q0 = av_clip(q0, src[ 0*step] - lims, src[ 0*step] + lims); |
| 524 | } |
| 525 | |
| 526 | p1 = (25*src[-4*step] + 26*src[-3*step] + 26*src[-2*step] + 26*p0 + |
| 527 | 25*src[ 0*step] + rv40_dither_l[dmode + i]) >> 7; |
| 528 | q1 = (25*src[-1*step] + 26*q0 + 26*src[ 1*step] + 26*src[ 2*step] + |
| 529 | 25*src[ 3*step] + rv40_dither_r[dmode + i]) >> 7; |
| 530 | |
| 531 | if (sflag) { |
| 532 | p1 = av_clip(p1, src[-2*step] - lims, src[-2*step] + lims); |
| 533 | q1 = av_clip(q1, src[ 1*step] - lims, src[ 1*step] + lims); |
| 534 | } |
| 535 | |
| 536 | src[-2*step] = p1; |
| 537 | src[-1*step] = p0; |
| 538 | src[ 0*step] = q0; |
| 539 | src[ 1*step] = q1; |
| 540 | |
| 541 | if(!chroma){ |
| 542 | src[-3*step] = (25*src[-1*step] + 26*src[-2*step] + |
| 543 | 51*src[-3*step] + 26*src[-4*step] + 64) >> 7; |
| 544 | src[ 2*step] = (25*src[ 0*step] + 26*src[ 1*step] + |
| 545 | 51*src[ 2*step] + 26*src[ 3*step] + 64) >> 7; |
| 546 | } |
| 547 | } |
| 548 | } |
| 549 | |
| 550 | static void rv40_h_strong_loop_filter(uint8_t *src, const ptrdiff_t stride, |
| 551 | const int alpha, const int lims, |
| 552 | const int dmode, const int chroma) |
| 553 | { |
| 554 | rv40_strong_loop_filter(src, stride, 1, alpha, lims, dmode, chroma); |
| 555 | } |
| 556 | |
| 557 | static void rv40_v_strong_loop_filter(uint8_t *src, const ptrdiff_t stride, |
| 558 | const int alpha, const int lims, |
| 559 | const int dmode, const int chroma) |
| 560 | { |
| 561 | rv40_strong_loop_filter(src, 1, stride, alpha, lims, dmode, chroma); |
| 562 | } |
| 563 | |
| 564 | static av_always_inline int rv40_loop_filter_strength(uint8_t *src, |
| 565 | int step, ptrdiff_t stride, |
| 566 | int beta, int beta2, |
| 567 | int edge, |
| 568 | int *p1, int *q1) |
| 569 | { |
| 570 | int sum_p1p0 = 0, sum_q1q0 = 0, sum_p1p2 = 0, sum_q1q2 = 0; |
| 571 | int strong0 = 0, strong1 = 0; |
| 572 | uint8_t *ptr; |
| 573 | int i; |
| 574 | |
| 575 | for (i = 0, ptr = src; i < 4; i++, ptr += stride) { |
| 576 | sum_p1p0 += ptr[-2*step] - ptr[-1*step]; |
| 577 | sum_q1q0 += ptr[ 1*step] - ptr[ 0*step]; |
| 578 | } |
| 579 | |
| 580 | *p1 = FFABS(sum_p1p0) < (beta << 2); |
| 581 | *q1 = FFABS(sum_q1q0) < (beta << 2); |
| 582 | |
| 583 | if(!*p1 && !*q1) |
| 584 | return 0; |
| 585 | |
| 586 | if (!edge) |
| 587 | return 0; |
| 588 | |
| 589 | for (i = 0, ptr = src; i < 4; i++, ptr += stride) { |
| 590 | sum_p1p2 += ptr[-2*step] - ptr[-3*step]; |
| 591 | sum_q1q2 += ptr[ 1*step] - ptr[ 2*step]; |
| 592 | } |
| 593 | |
| 594 | strong0 = *p1 && (FFABS(sum_p1p2) < beta2); |
| 595 | strong1 = *q1 && (FFABS(sum_q1q2) < beta2); |
| 596 | |
| 597 | return strong0 && strong1; |
| 598 | } |
| 599 | |
| 600 | static int rv40_h_loop_filter_strength(uint8_t *src, ptrdiff_t stride, |
| 601 | int beta, int beta2, int edge, |
| 602 | int *p1, int *q1) |
| 603 | { |
| 604 | return rv40_loop_filter_strength(src, stride, 1, beta, beta2, edge, p1, q1); |
| 605 | } |
| 606 | |
| 607 | static int rv40_v_loop_filter_strength(uint8_t *src, ptrdiff_t stride, |
| 608 | int beta, int beta2, int edge, |
| 609 | int *p1, int *q1) |
| 610 | { |
| 611 | return rv40_loop_filter_strength(src, 1, stride, beta, beta2, edge, p1, q1); |
| 612 | } |
| 613 | |
| 614 | av_cold void ff_rv40dsp_init(RV34DSPContext *c) |
| 615 | { |
| 616 | H264QpelContext qpel; |
| 617 | |
| 618 | ff_rv34dsp_init(c); |
| 619 | ff_h264qpel_init(&qpel, 8); |
| 620 | |
| 621 | c->put_pixels_tab[0][ 0] = qpel.put_h264_qpel_pixels_tab[0][0]; |
| 622 | c->put_pixels_tab[0][ 1] = put_rv40_qpel16_mc10_c; |
| 623 | c->put_pixels_tab[0][ 2] = qpel.put_h264_qpel_pixels_tab[0][2]; |
| 624 | c->put_pixels_tab[0][ 3] = put_rv40_qpel16_mc30_c; |
| 625 | c->put_pixels_tab[0][ 4] = put_rv40_qpel16_mc01_c; |
| 626 | c->put_pixels_tab[0][ 5] = put_rv40_qpel16_mc11_c; |
| 627 | c->put_pixels_tab[0][ 6] = put_rv40_qpel16_mc21_c; |
| 628 | c->put_pixels_tab[0][ 7] = put_rv40_qpel16_mc31_c; |
| 629 | c->put_pixels_tab[0][ 8] = qpel.put_h264_qpel_pixels_tab[0][8]; |
| 630 | c->put_pixels_tab[0][ 9] = put_rv40_qpel16_mc12_c; |
| 631 | c->put_pixels_tab[0][10] = put_rv40_qpel16_mc22_c; |
| 632 | c->put_pixels_tab[0][11] = put_rv40_qpel16_mc32_c; |
| 633 | c->put_pixels_tab[0][12] = put_rv40_qpel16_mc03_c; |
| 634 | c->put_pixels_tab[0][13] = put_rv40_qpel16_mc13_c; |
| 635 | c->put_pixels_tab[0][14] = put_rv40_qpel16_mc23_c; |
| 636 | c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_c; |
| 637 | c->avg_pixels_tab[0][ 0] = qpel.avg_h264_qpel_pixels_tab[0][0]; |
| 638 | c->avg_pixels_tab[0][ 1] = avg_rv40_qpel16_mc10_c; |
| 639 | c->avg_pixels_tab[0][ 2] = qpel.avg_h264_qpel_pixels_tab[0][2]; |
| 640 | c->avg_pixels_tab[0][ 3] = avg_rv40_qpel16_mc30_c; |
| 641 | c->avg_pixels_tab[0][ 4] = avg_rv40_qpel16_mc01_c; |
| 642 | c->avg_pixels_tab[0][ 5] = avg_rv40_qpel16_mc11_c; |
| 643 | c->avg_pixels_tab[0][ 6] = avg_rv40_qpel16_mc21_c; |
| 644 | c->avg_pixels_tab[0][ 7] = avg_rv40_qpel16_mc31_c; |
| 645 | c->avg_pixels_tab[0][ 8] = qpel.avg_h264_qpel_pixels_tab[0][8]; |
| 646 | c->avg_pixels_tab[0][ 9] = avg_rv40_qpel16_mc12_c; |
| 647 | c->avg_pixels_tab[0][10] = avg_rv40_qpel16_mc22_c; |
| 648 | c->avg_pixels_tab[0][11] = avg_rv40_qpel16_mc32_c; |
| 649 | c->avg_pixels_tab[0][12] = avg_rv40_qpel16_mc03_c; |
| 650 | c->avg_pixels_tab[0][13] = avg_rv40_qpel16_mc13_c; |
| 651 | c->avg_pixels_tab[0][14] = avg_rv40_qpel16_mc23_c; |
| 652 | c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_c; |
| 653 | c->put_pixels_tab[1][ 0] = qpel.put_h264_qpel_pixels_tab[1][0]; |
| 654 | c->put_pixels_tab[1][ 1] = put_rv40_qpel8_mc10_c; |
| 655 | c->put_pixels_tab[1][ 2] = qpel.put_h264_qpel_pixels_tab[1][2]; |
| 656 | c->put_pixels_tab[1][ 3] = put_rv40_qpel8_mc30_c; |
| 657 | c->put_pixels_tab[1][ 4] = put_rv40_qpel8_mc01_c; |
| 658 | c->put_pixels_tab[1][ 5] = put_rv40_qpel8_mc11_c; |
| 659 | c->put_pixels_tab[1][ 6] = put_rv40_qpel8_mc21_c; |
| 660 | c->put_pixels_tab[1][ 7] = put_rv40_qpel8_mc31_c; |
| 661 | c->put_pixels_tab[1][ 8] = qpel.put_h264_qpel_pixels_tab[1][8]; |
| 662 | c->put_pixels_tab[1][ 9] = put_rv40_qpel8_mc12_c; |
| 663 | c->put_pixels_tab[1][10] = put_rv40_qpel8_mc22_c; |
| 664 | c->put_pixels_tab[1][11] = put_rv40_qpel8_mc32_c; |
| 665 | c->put_pixels_tab[1][12] = put_rv40_qpel8_mc03_c; |
| 666 | c->put_pixels_tab[1][13] = put_rv40_qpel8_mc13_c; |
| 667 | c->put_pixels_tab[1][14] = put_rv40_qpel8_mc23_c; |
| 668 | c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_c; |
| 669 | c->avg_pixels_tab[1][ 0] = qpel.avg_h264_qpel_pixels_tab[1][0]; |
| 670 | c->avg_pixels_tab[1][ 1] = avg_rv40_qpel8_mc10_c; |
| 671 | c->avg_pixels_tab[1][ 2] = qpel.avg_h264_qpel_pixels_tab[1][2]; |
| 672 | c->avg_pixels_tab[1][ 3] = avg_rv40_qpel8_mc30_c; |
| 673 | c->avg_pixels_tab[1][ 4] = avg_rv40_qpel8_mc01_c; |
| 674 | c->avg_pixels_tab[1][ 5] = avg_rv40_qpel8_mc11_c; |
| 675 | c->avg_pixels_tab[1][ 6] = avg_rv40_qpel8_mc21_c; |
| 676 | c->avg_pixels_tab[1][ 7] = avg_rv40_qpel8_mc31_c; |
| 677 | c->avg_pixels_tab[1][ 8] = qpel.avg_h264_qpel_pixels_tab[1][8]; |
| 678 | c->avg_pixels_tab[1][ 9] = avg_rv40_qpel8_mc12_c; |
| 679 | c->avg_pixels_tab[1][10] = avg_rv40_qpel8_mc22_c; |
| 680 | c->avg_pixels_tab[1][11] = avg_rv40_qpel8_mc32_c; |
| 681 | c->avg_pixels_tab[1][12] = avg_rv40_qpel8_mc03_c; |
| 682 | c->avg_pixels_tab[1][13] = avg_rv40_qpel8_mc13_c; |
| 683 | c->avg_pixels_tab[1][14] = avg_rv40_qpel8_mc23_c; |
| 684 | c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_c; |
| 685 | |
| 686 | c->put_chroma_pixels_tab[0] = put_rv40_chroma_mc8_c; |
| 687 | c->put_chroma_pixels_tab[1] = put_rv40_chroma_mc4_c; |
| 688 | c->avg_chroma_pixels_tab[0] = avg_rv40_chroma_mc8_c; |
| 689 | c->avg_chroma_pixels_tab[1] = avg_rv40_chroma_mc4_c; |
| 690 | |
| 691 | c->rv40_weight_pixels_tab[0][0] = rv40_weight_func_rnd_16; |
| 692 | c->rv40_weight_pixels_tab[0][1] = rv40_weight_func_rnd_8; |
| 693 | c->rv40_weight_pixels_tab[1][0] = rv40_weight_func_nornd_16; |
| 694 | c->rv40_weight_pixels_tab[1][1] = rv40_weight_func_nornd_8; |
| 695 | |
| 696 | c->rv40_weak_loop_filter[0] = rv40_h_weak_loop_filter; |
| 697 | c->rv40_weak_loop_filter[1] = rv40_v_weak_loop_filter; |
| 698 | c->rv40_strong_loop_filter[0] = rv40_h_strong_loop_filter; |
| 699 | c->rv40_strong_loop_filter[1] = rv40_v_strong_loop_filter; |
| 700 | c->rv40_loop_filter_strength[0] = rv40_h_loop_filter_strength; |
| 701 | c->rv40_loop_filter_strength[1] = rv40_v_loop_filter_strength; |
| 702 | |
| 703 | if (ARCH_AARCH64) |
| 704 | ff_rv40dsp_init_aarch64(c); |
| 705 | if (ARCH_ARM) |
| 706 | ff_rv40dsp_init_arm(c); |
| 707 | if (ARCH_X86) |
| 708 | ff_rv40dsp_init_x86(c); |
| 709 | } |