2 * RV40 decoder motion compensation functions
3 * Copyright (c) 2008 Konstantin Shishkov
5 * This file is part of FFmpeg.
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.
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.
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
24 * RV40 decoder motion compensation functions
27 #include "libavutil/common.h"
28 #include "libavutil/intreadwrite.h"
35 #include "libavutil/avassert.h"
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;\
42 for(i = 0; i < h; i++)\
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);\
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;\
61 for(i = 0; i < w; i++)\
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);\
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);\
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);\
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);\
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);\
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)\
113 OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\
116 static void OPNAME ## rv40_qpel ## SIZE ## _mc30_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
118 OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\
121 static void OPNAME ## rv40_qpel ## SIZE ## _mc01_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
123 OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\
126 static void OPNAME ## rv40_qpel ## SIZE ## _mc11_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
134 static void OPNAME ## rv40_qpel ## SIZE ## _mc21_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
142 static void OPNAME ## rv40_qpel ## SIZE ## _mc31_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
150 static void OPNAME ## rv40_qpel ## SIZE ## _mc12_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
158 static void OPNAME ## rv40_qpel ## SIZE ## _mc22_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
166 static void OPNAME ## rv40_qpel ## SIZE ## _mc32_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
174 static void OPNAME ## rv40_qpel ## SIZE ## _mc03_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
176 OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\
179 static void OPNAME ## rv40_qpel ## SIZE ## _mc13_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
187 static void OPNAME ## rv40_qpel ## SIZE ## _mc23_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
196 #define op_avg(a, b) a = (((a)+cm[b]+1)>>1)
197 #define op_put(a, b) a = cm[b]
199 RV40_LOWPASS(put_ , op_put)
200 RV40_LOWPASS(avg_ , op_avg)
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, \
216 /* FIXME HIGH BIT DEPTH */ \
219 for (j = 0; j < 2; j++) { \
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) + \
226 uint32_t h0 = ((a & 0xFCFCFCFCUL) >> 2) + \
227 ((b & 0xFCFCFCFCUL) >> 2); \
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) + \
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; \
254 pixels += 4 - line_size * (h + 1); \
255 block += 4 - line_size * h; \
259 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_8_c, \
260 OPNAME ## _pixels8_xy2_8_c, \
263 #define op_avg(a, b) a = rnd_avg32(a, b)
264 #define op_put(a, b) a = b
270 static void put_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
272 put_pixels16_xy2_8_c(dst, src, stride, 16);
274 static void avg_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
276 avg_pixels16_xy2_8_c(dst, src, stride, 16);
278 static void put_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
280 put_pixels8_xy2_8_c(dst, src, stride, 8);
282 static void avg_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
284 avg_pixels8_xy2_8_c(dst, src, stride, 8);
287 static const int rv40_bias[4][4] = {
294 #define RV40_CHROMA_MC(OPNAME, OP)\
295 static void OPNAME ## rv40_chroma_mc4_c(uint8_t *dst /*align 8*/,\
296 uint8_t *src /*align 1*/,\
297 ptrdiff_t stride, int h, int x, int y)\
299 const int A = (8-x) * (8-y);\
300 const int B = ( x) * (8-y);\
301 const int C = (8-x) * ( y);\
302 const int D = ( x) * ( y);\
304 int bias = rv40_bias[y>>1][x>>1];\
306 av_assert2(x<8 && y<8 && x>=0 && y>=0);\
309 for(i = 0; i < h; i++){\
310 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + bias));\
311 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + bias));\
312 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + bias));\
313 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + bias));\
318 const int E = B + C;\
319 const ptrdiff_t step = C ? stride : 1;\
320 for(i = 0; i < h; i++){\
321 OP(dst[0], (A*src[0] + E*src[step+0] + bias));\
322 OP(dst[1], (A*src[1] + E*src[step+1] + bias));\
323 OP(dst[2], (A*src[2] + E*src[step+2] + bias));\
324 OP(dst[3], (A*src[3] + E*src[step+3] + bias));\
331 static void OPNAME ## rv40_chroma_mc8_c(uint8_t *dst/*align 8*/,\
332 uint8_t *src/*align 1*/,\
333 ptrdiff_t stride, int h, int x, int y)\
335 const int A = (8-x) * (8-y);\
336 const int B = ( x) * (8-y);\
337 const int C = (8-x) * ( y);\
338 const int D = ( x) * ( y);\
340 int bias = rv40_bias[y>>1][x>>1];\
342 av_assert2(x<8 && y<8 && x>=0 && y>=0);\
345 for(i = 0; i < h; i++){\
346 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + bias));\
347 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + bias));\
348 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + bias));\
349 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + bias));\
350 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + bias));\
351 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + bias));\
352 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + bias));\
353 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + bias));\
358 const int E = B + C;\
359 const ptrdiff_t step = C ? stride : 1;\
360 for(i = 0; i < h; i++){\
361 OP(dst[0], (A*src[0] + E*src[step+0] + bias));\
362 OP(dst[1], (A*src[1] + E*src[step+1] + bias));\
363 OP(dst[2], (A*src[2] + E*src[step+2] + bias));\
364 OP(dst[3], (A*src[3] + E*src[step+3] + bias));\
365 OP(dst[4], (A*src[4] + E*src[step+4] + bias));\
366 OP(dst[5], (A*src[5] + E*src[step+5] + bias));\
367 OP(dst[6], (A*src[6] + E*src[step+6] + bias));\
368 OP(dst[7], (A*src[7] + E*src[step+7] + bias));\
375 #define op_avg(a, b) a = (((a)+((b)>>6)+1)>>1)
376 #define op_put(a, b) a = ((b)>>6)
378 RV40_CHROMA_MC(put_, op_put)
379 RV40_CHROMA_MC(avg_, op_avg)
381 #define RV40_WEIGHT_FUNC(size) \
382 static void rv40_weight_func_rnd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\
386 for (j = 0; j < size; j++) {\
387 for (i = 0; i < size; i++)\
388 dst[i] = (((w2 * src1[i]) >> 9) + ((w1 * src2[i]) >> 9) + 0x10) >> 5;\
394 static void rv40_weight_func_nornd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\
398 for (j = 0; j < size; j++) {\
399 for (i = 0; i < size; i++)\
400 dst[i] = (w2 * src1[i] + w1 * src2[i] + 0x10) >> 5;\
411 * dither values for deblocking filter - left/top values
413 static const uint8_t rv40_dither_l[16] = {
414 0x40, 0x50, 0x20, 0x60, 0x30, 0x50, 0x40, 0x30,
415 0x50, 0x40, 0x50, 0x30, 0x60, 0x20, 0x50, 0x40
419 * dither values for deblocking filter - right/bottom values
421 static const uint8_t rv40_dither_r[16] = {
422 0x40, 0x30, 0x60, 0x20, 0x50, 0x30, 0x30, 0x40,
423 0x40, 0x40, 0x50, 0x30, 0x20, 0x60, 0x30, 0x40
426 #define CLIP_SYMM(a, b) av_clip(a, -(b), b)
428 * weaker deblocking very similar to the one described in 4.4.2 of JVT-A003r1
430 static av_always_inline void rv40_weak_loop_filter(uint8_t *src,
432 const ptrdiff_t stride,
441 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
444 for (i = 0; i < 4; i++, src += stride) {
445 int diff_p1p0 = src[-2*step] - src[-1*step];
446 int diff_q1q0 = src[ 1*step] - src[ 0*step];
447 int diff_p1p2 = src[-2*step] - src[-3*step];
448 int diff_q1q2 = src[ 1*step] - src[ 2*step];
450 t = src[0*step] - src[-1*step];
454 u = (alpha * FFABS(t)) >> 7;
455 if (u > 3 - (filter_p1 && filter_q1))
459 if (filter_p1 && filter_q1)
460 t += src[-2*step] - src[1*step];
462 diff = CLIP_SYMM((t + 4) >> 3, lim_p0q0);
463 src[-1*step] = cm[src[-1*step] + diff];
464 src[ 0*step] = cm[src[ 0*step] - diff];
466 if (filter_p1 && FFABS(diff_p1p2) <= beta) {
467 t = (diff_p1p0 + diff_p1p2 - diff) >> 1;
468 src[-2*step] = cm[src[-2*step] - CLIP_SYMM(t, lim_p1)];
471 if (filter_q1 && FFABS(diff_q1q2) <= beta) {
472 t = (diff_q1q0 + diff_q1q2 + diff) >> 1;
473 src[ 1*step] = cm[src[ 1*step] - CLIP_SYMM(t, lim_q1)];
478 static void rv40_h_weak_loop_filter(uint8_t *src, const ptrdiff_t stride,
479 const int filter_p1, const int filter_q1,
480 const int alpha, const int beta,
481 const int lim_p0q0, const int lim_q1,
484 rv40_weak_loop_filter(src, stride, 1, filter_p1, filter_q1,
485 alpha, beta, lim_p0q0, lim_q1, lim_p1);
488 static void rv40_v_weak_loop_filter(uint8_t *src, const ptrdiff_t stride,
489 const int filter_p1, const int filter_q1,
490 const int alpha, const int beta,
491 const int lim_p0q0, const int lim_q1,
494 rv40_weak_loop_filter(src, 1, stride, filter_p1, filter_q1,
495 alpha, beta, lim_p0q0, lim_q1, lim_p1);
498 static av_always_inline void rv40_strong_loop_filter(uint8_t *src,
500 const ptrdiff_t stride,
508 for(i = 0; i < 4; i++, src += stride){
509 int sflag, p0, q0, p1, q1;
510 int t = src[0*step] - src[-1*step];
515 sflag = (alpha * FFABS(t)) >> 7;
519 p0 = (25*src[-3*step] + 26*src[-2*step] + 26*src[-1*step] +
520 26*src[ 0*step] + 25*src[ 1*step] +
521 rv40_dither_l[dmode + i]) >> 7;
523 q0 = (25*src[-2*step] + 26*src[-1*step] + 26*src[ 0*step] +
524 26*src[ 1*step] + 25*src[ 2*step] +
525 rv40_dither_r[dmode + i]) >> 7;
528 p0 = av_clip(p0, src[-1*step] - lims, src[-1*step] + lims);
529 q0 = av_clip(q0, src[ 0*step] - lims, src[ 0*step] + lims);
532 p1 = (25*src[-4*step] + 26*src[-3*step] + 26*src[-2*step] + 26*p0 +
533 25*src[ 0*step] + rv40_dither_l[dmode + i]) >> 7;
534 q1 = (25*src[-1*step] + 26*q0 + 26*src[ 1*step] + 26*src[ 2*step] +
535 25*src[ 3*step] + rv40_dither_r[dmode + i]) >> 7;
538 p1 = av_clip(p1, src[-2*step] - lims, src[-2*step] + lims);
539 q1 = av_clip(q1, src[ 1*step] - lims, src[ 1*step] + lims);
548 src[-3*step] = (25*src[-1*step] + 26*src[-2*step] +
549 51*src[-3*step] + 26*src[-4*step] + 64) >> 7;
550 src[ 2*step] = (25*src[ 0*step] + 26*src[ 1*step] +
551 51*src[ 2*step] + 26*src[ 3*step] + 64) >> 7;
556 static void rv40_h_strong_loop_filter(uint8_t *src, const ptrdiff_t stride,
557 const int alpha, const int lims,
558 const int dmode, const int chroma)
560 rv40_strong_loop_filter(src, stride, 1, alpha, lims, dmode, chroma);
563 static void rv40_v_strong_loop_filter(uint8_t *src, const ptrdiff_t stride,
564 const int alpha, const int lims,
565 const int dmode, const int chroma)
567 rv40_strong_loop_filter(src, 1, stride, alpha, lims, dmode, chroma);
570 static av_always_inline int rv40_loop_filter_strength(uint8_t *src,
571 int step, ptrdiff_t stride,
576 int sum_p1p0 = 0, sum_q1q0 = 0, sum_p1p2 = 0, sum_q1q2 = 0;
577 int strong0 = 0, strong1 = 0;
581 for (i = 0, ptr = src; i < 4; i++, ptr += stride) {
582 sum_p1p0 += ptr[-2*step] - ptr[-1*step];
583 sum_q1q0 += ptr[ 1*step] - ptr[ 0*step];
586 *p1 = FFABS(sum_p1p0) < (beta << 2);
587 *q1 = FFABS(sum_q1q0) < (beta << 2);
595 for (i = 0, ptr = src; i < 4; i++, ptr += stride) {
596 sum_p1p2 += ptr[-2*step] - ptr[-3*step];
597 sum_q1q2 += ptr[ 1*step] - ptr[ 2*step];
600 strong0 = *p1 && (FFABS(sum_p1p2) < beta2);
601 strong1 = *q1 && (FFABS(sum_q1q2) < beta2);
603 return strong0 && strong1;
606 static int rv40_h_loop_filter_strength(uint8_t *src, ptrdiff_t stride,
607 int beta, int beta2, int edge,
610 return rv40_loop_filter_strength(src, stride, 1, beta, beta2, edge, p1, q1);
613 static int rv40_v_loop_filter_strength(uint8_t *src, ptrdiff_t stride,
614 int beta, int beta2, int edge,
617 return rv40_loop_filter_strength(src, 1, stride, beta, beta2, edge, p1, q1);
620 av_cold void ff_rv40dsp_init(RV34DSPContext *c)
622 H264QpelContext qpel;
625 ff_h264qpel_init(&qpel, 8);
627 c->put_pixels_tab[0][ 0] = qpel.put_h264_qpel_pixels_tab[0][0];
628 c->put_pixels_tab[0][ 1] = put_rv40_qpel16_mc10_c;
629 c->put_pixels_tab[0][ 2] = qpel.put_h264_qpel_pixels_tab[0][2];
630 c->put_pixels_tab[0][ 3] = put_rv40_qpel16_mc30_c;
631 c->put_pixels_tab[0][ 4] = put_rv40_qpel16_mc01_c;
632 c->put_pixels_tab[0][ 5] = put_rv40_qpel16_mc11_c;
633 c->put_pixels_tab[0][ 6] = put_rv40_qpel16_mc21_c;
634 c->put_pixels_tab[0][ 7] = put_rv40_qpel16_mc31_c;
635 c->put_pixels_tab[0][ 8] = qpel.put_h264_qpel_pixels_tab[0][8];
636 c->put_pixels_tab[0][ 9] = put_rv40_qpel16_mc12_c;
637 c->put_pixels_tab[0][10] = put_rv40_qpel16_mc22_c;
638 c->put_pixels_tab[0][11] = put_rv40_qpel16_mc32_c;
639 c->put_pixels_tab[0][12] = put_rv40_qpel16_mc03_c;
640 c->put_pixels_tab[0][13] = put_rv40_qpel16_mc13_c;
641 c->put_pixels_tab[0][14] = put_rv40_qpel16_mc23_c;
642 c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_c;
643 c->avg_pixels_tab[0][ 0] = qpel.avg_h264_qpel_pixels_tab[0][0];
644 c->avg_pixels_tab[0][ 1] = avg_rv40_qpel16_mc10_c;
645 c->avg_pixels_tab[0][ 2] = qpel.avg_h264_qpel_pixels_tab[0][2];
646 c->avg_pixels_tab[0][ 3] = avg_rv40_qpel16_mc30_c;
647 c->avg_pixels_tab[0][ 4] = avg_rv40_qpel16_mc01_c;
648 c->avg_pixels_tab[0][ 5] = avg_rv40_qpel16_mc11_c;
649 c->avg_pixels_tab[0][ 6] = avg_rv40_qpel16_mc21_c;
650 c->avg_pixels_tab[0][ 7] = avg_rv40_qpel16_mc31_c;
651 c->avg_pixels_tab[0][ 8] = qpel.avg_h264_qpel_pixels_tab[0][8];
652 c->avg_pixels_tab[0][ 9] = avg_rv40_qpel16_mc12_c;
653 c->avg_pixels_tab[0][10] = avg_rv40_qpel16_mc22_c;
654 c->avg_pixels_tab[0][11] = avg_rv40_qpel16_mc32_c;
655 c->avg_pixels_tab[0][12] = avg_rv40_qpel16_mc03_c;
656 c->avg_pixels_tab[0][13] = avg_rv40_qpel16_mc13_c;
657 c->avg_pixels_tab[0][14] = avg_rv40_qpel16_mc23_c;
658 c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_c;
659 c->put_pixels_tab[1][ 0] = qpel.put_h264_qpel_pixels_tab[1][0];
660 c->put_pixels_tab[1][ 1] = put_rv40_qpel8_mc10_c;
661 c->put_pixels_tab[1][ 2] = qpel.put_h264_qpel_pixels_tab[1][2];
662 c->put_pixels_tab[1][ 3] = put_rv40_qpel8_mc30_c;
663 c->put_pixels_tab[1][ 4] = put_rv40_qpel8_mc01_c;
664 c->put_pixels_tab[1][ 5] = put_rv40_qpel8_mc11_c;
665 c->put_pixels_tab[1][ 6] = put_rv40_qpel8_mc21_c;
666 c->put_pixels_tab[1][ 7] = put_rv40_qpel8_mc31_c;
667 c->put_pixels_tab[1][ 8] = qpel.put_h264_qpel_pixels_tab[1][8];
668 c->put_pixels_tab[1][ 9] = put_rv40_qpel8_mc12_c;
669 c->put_pixels_tab[1][10] = put_rv40_qpel8_mc22_c;
670 c->put_pixels_tab[1][11] = put_rv40_qpel8_mc32_c;
671 c->put_pixels_tab[1][12] = put_rv40_qpel8_mc03_c;
672 c->put_pixels_tab[1][13] = put_rv40_qpel8_mc13_c;
673 c->put_pixels_tab[1][14] = put_rv40_qpel8_mc23_c;
674 c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_c;
675 c->avg_pixels_tab[1][ 0] = qpel.avg_h264_qpel_pixels_tab[1][0];
676 c->avg_pixels_tab[1][ 1] = avg_rv40_qpel8_mc10_c;
677 c->avg_pixels_tab[1][ 2] = qpel.avg_h264_qpel_pixels_tab[1][2];
678 c->avg_pixels_tab[1][ 3] = avg_rv40_qpel8_mc30_c;
679 c->avg_pixels_tab[1][ 4] = avg_rv40_qpel8_mc01_c;
680 c->avg_pixels_tab[1][ 5] = avg_rv40_qpel8_mc11_c;
681 c->avg_pixels_tab[1][ 6] = avg_rv40_qpel8_mc21_c;
682 c->avg_pixels_tab[1][ 7] = avg_rv40_qpel8_mc31_c;
683 c->avg_pixels_tab[1][ 8] = qpel.avg_h264_qpel_pixels_tab[1][8];
684 c->avg_pixels_tab[1][ 9] = avg_rv40_qpel8_mc12_c;
685 c->avg_pixels_tab[1][10] = avg_rv40_qpel8_mc22_c;
686 c->avg_pixels_tab[1][11] = avg_rv40_qpel8_mc32_c;
687 c->avg_pixels_tab[1][12] = avg_rv40_qpel8_mc03_c;
688 c->avg_pixels_tab[1][13] = avg_rv40_qpel8_mc13_c;
689 c->avg_pixels_tab[1][14] = avg_rv40_qpel8_mc23_c;
690 c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_c;
692 c->put_chroma_pixels_tab[0] = put_rv40_chroma_mc8_c;
693 c->put_chroma_pixels_tab[1] = put_rv40_chroma_mc4_c;
694 c->avg_chroma_pixels_tab[0] = avg_rv40_chroma_mc8_c;
695 c->avg_chroma_pixels_tab[1] = avg_rv40_chroma_mc4_c;
697 c->rv40_weight_pixels_tab[0][0] = rv40_weight_func_rnd_16;
698 c->rv40_weight_pixels_tab[0][1] = rv40_weight_func_rnd_8;
699 c->rv40_weight_pixels_tab[1][0] = rv40_weight_func_nornd_16;
700 c->rv40_weight_pixels_tab[1][1] = rv40_weight_func_nornd_8;
702 c->rv40_weak_loop_filter[0] = rv40_h_weak_loop_filter;
703 c->rv40_weak_loop_filter[1] = rv40_v_weak_loop_filter;
704 c->rv40_strong_loop_filter[0] = rv40_h_strong_loop_filter;
705 c->rv40_strong_loop_filter[1] = rv40_v_strong_loop_filter;
706 c->rv40_loop_filter_strength[0] = rv40_h_loop_filter_strength;
707 c->rv40_loop_filter_strength[1] = rv40_v_loop_filter_strength;
710 ff_rv40dsp_init_aarch64(c);
712 ff_rv40dsp_init_arm(c);
714 ff_rv40dsp_init_x86(c);