2 * RV40 decoder motion compensation functions
3 * Copyright (c) 2008 Konstantin Shishkov
5 * This file is part of Libav.
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.
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.
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
24 * RV40 decoder motion compensation functions
27 #include "libavutil/common.h"
28 #include "libavutil/intreadwrite.h"
36 #define RV40_LOWPASS(OPNAME, OP) \
37 static av_unused 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;\
41 for(i = 0; i < h; i++)\
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);\
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;\
60 for(i = 0; i < w; i++)\
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);\
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);\
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);\
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);\
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);\
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)\
112 OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\
115 static void OPNAME ## rv40_qpel ## SIZE ## _mc30_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
117 OPNAME ## rv40_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\
120 static void OPNAME ## rv40_qpel ## SIZE ## _mc01_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
122 OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 52, 20, 6);\
125 static void OPNAME ## rv40_qpel ## SIZE ## _mc11_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
133 static void OPNAME ## rv40_qpel ## SIZE ## _mc21_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
141 static void OPNAME ## rv40_qpel ## SIZE ## _mc31_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
149 static void OPNAME ## rv40_qpel ## SIZE ## _mc12_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
157 static void OPNAME ## rv40_qpel ## SIZE ## _mc22_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
165 static void OPNAME ## rv40_qpel ## SIZE ## _mc32_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
173 static void OPNAME ## rv40_qpel ## SIZE ## _mc03_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
175 OPNAME ## rv40_qpel ## SIZE ## _v_lowpass(dst, src, stride, stride, SIZE, 20, 52, 6);\
178 static void OPNAME ## rv40_qpel ## SIZE ## _mc13_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
186 static void OPNAME ## rv40_qpel ## SIZE ## _mc23_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\
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);\
195 #define op_avg(a, b) a = (((a)+cm[b]+1)>>1)
196 #define op_put(a, b) a = cm[b]
198 RV40_LOWPASS(put_ , op_put)
199 RV40_LOWPASS(avg_ , op_avg)
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, \
215 /* FIXME HIGH BIT DEPTH */ \
218 for (j = 0; j < 2; j++) { \
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) + \
225 uint32_t h0 = ((a & 0xFCFCFCFCUL) >> 2) + \
226 ((b & 0xFCFCFCFCUL) >> 2); \
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) + \
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; \
253 pixels += 4 - line_size * (h + 1); \
254 block += 4 - line_size * h; \
258 CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_8_c, \
259 OPNAME ## _pixels8_xy2_8_c, \
262 #define op_avg(a, b) a = rnd_avg32(a, b)
263 #define op_put(a, b) a = b
269 static void put_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
271 put_pixels16_xy2_8_c(dst, src, stride, 16);
273 static void avg_rv40_qpel16_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
275 avg_pixels16_xy2_8_c(dst, src, stride, 16);
277 static void put_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
279 put_pixels8_xy2_8_c(dst, src, stride, 8);
281 static void avg_rv40_qpel8_mc33_c(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)
283 avg_pixels8_xy2_8_c(dst, src, stride, 8);
286 static const int rv40_bias[4][4] = {
293 #define RV40_CHROMA_MC(OPNAME, OP)\
294 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){\
295 const int A = (8-x) * (8-y);\
296 const int B = ( x) * (8-y);\
297 const int C = (8-x) * ( y);\
298 const int D = ( x) * ( y);\
300 int bias = rv40_bias[y>>1][x>>1];\
302 assert(x<8 && y<8 && x>=0 && y>=0);\
305 for(i = 0; i < h; i++){\
306 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + bias));\
307 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + bias));\
308 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + bias));\
309 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + bias));\
314 const int E = B + C;\
315 const int step = C ? stride : 1;\
316 for(i = 0; i < h; i++){\
317 OP(dst[0], (A*src[0] + E*src[step+0] + bias));\
318 OP(dst[1], (A*src[1] + E*src[step+1] + bias));\
319 OP(dst[2], (A*src[2] + E*src[step+2] + bias));\
320 OP(dst[3], (A*src[3] + E*src[step+3] + bias));\
327 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){\
328 const int A = (8-x) * (8-y);\
329 const int B = ( x) * (8-y);\
330 const int C = (8-x) * ( y);\
331 const int D = ( x) * ( y);\
333 int bias = rv40_bias[y>>1][x>>1];\
335 assert(x<8 && y<8 && x>=0 && y>=0);\
338 for(i = 0; i < h; i++){\
339 OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + bias));\
340 OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + bias));\
341 OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + bias));\
342 OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + bias));\
343 OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + bias));\
344 OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + bias));\
345 OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + bias));\
346 OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + bias));\
351 const int E = B + C;\
352 const int step = C ? stride : 1;\
353 for(i = 0; i < h; i++){\
354 OP(dst[0], (A*src[0] + E*src[step+0] + bias));\
355 OP(dst[1], (A*src[1] + E*src[step+1] + bias));\
356 OP(dst[2], (A*src[2] + E*src[step+2] + bias));\
357 OP(dst[3], (A*src[3] + E*src[step+3] + bias));\
358 OP(dst[4], (A*src[4] + E*src[step+4] + bias));\
359 OP(dst[5], (A*src[5] + E*src[step+5] + bias));\
360 OP(dst[6], (A*src[6] + E*src[step+6] + bias));\
361 OP(dst[7], (A*src[7] + E*src[step+7] + bias));\
368 #define op_avg(a, b) a = (((a)+((b)>>6)+1)>>1)
369 #define op_put(a, b) a = ((b)>>6)
371 RV40_CHROMA_MC(put_, op_put)
372 RV40_CHROMA_MC(avg_, op_avg)
374 #define RV40_WEIGHT_FUNC(size) \
375 static void rv40_weight_func_rnd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\
379 for (j = 0; j < size; j++) {\
380 for (i = 0; i < size; i++)\
381 dst[i] = (((w2 * src1[i]) >> 9) + ((w1 * src2[i]) >> 9) + 0x10) >> 5;\
387 static void rv40_weight_func_nornd_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, ptrdiff_t stride)\
391 for (j = 0; j < size; j++) {\
392 for (i = 0; i < size; i++)\
393 dst[i] = (w2 * src1[i] + w1 * src2[i] + 0x10) >> 5;\
404 * dither values for deblocking filter - left/top values
406 static const uint8_t rv40_dither_l[16] = {
407 0x40, 0x50, 0x20, 0x60, 0x30, 0x50, 0x40, 0x30,
408 0x50, 0x40, 0x50, 0x30, 0x60, 0x20, 0x50, 0x40
412 * dither values for deblocking filter - right/bottom values
414 static const uint8_t rv40_dither_r[16] = {
415 0x40, 0x30, 0x60, 0x20, 0x50, 0x30, 0x30, 0x40,
416 0x40, 0x40, 0x50, 0x30, 0x20, 0x60, 0x30, 0x40
419 #define CLIP_SYMM(a, b) av_clip(a, -(b), b)
421 * weaker deblocking very similar to the one described in 4.4.2 of JVT-A003r1
423 static av_always_inline void rv40_weak_loop_filter(uint8_t *src,
425 const ptrdiff_t stride,
434 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
437 for (i = 0; i < 4; i++, src += stride) {
438 int diff_p1p0 = src[-2*step] - src[-1*step];
439 int diff_q1q0 = src[ 1*step] - src[ 0*step];
440 int diff_p1p2 = src[-2*step] - src[-3*step];
441 int diff_q1q2 = src[ 1*step] - src[ 2*step];
443 t = src[0*step] - src[-1*step];
447 u = (alpha * FFABS(t)) >> 7;
448 if (u > 3 - (filter_p1 && filter_q1))
452 if (filter_p1 && filter_q1)
453 t += src[-2*step] - src[1*step];
455 diff = CLIP_SYMM((t + 4) >> 3, lim_p0q0);
456 src[-1*step] = cm[src[-1*step] + diff];
457 src[ 0*step] = cm[src[ 0*step] - diff];
459 if (filter_p1 && FFABS(diff_p1p2) <= beta) {
460 t = (diff_p1p0 + diff_p1p2 - diff) >> 1;
461 src[-2*step] = cm[src[-2*step] - CLIP_SYMM(t, lim_p1)];
464 if (filter_q1 && FFABS(diff_q1q2) <= beta) {
465 t = (diff_q1q0 + diff_q1q2 + diff) >> 1;
466 src[ 1*step] = cm[src[ 1*step] - CLIP_SYMM(t, lim_q1)];
471 static void rv40_h_weak_loop_filter(uint8_t *src, const ptrdiff_t stride,
472 const int filter_p1, const int filter_q1,
473 const int alpha, const int beta,
474 const int lim_p0q0, const int lim_q1,
477 rv40_weak_loop_filter(src, stride, 1, filter_p1, filter_q1,
478 alpha, beta, lim_p0q0, lim_q1, lim_p1);
481 static void rv40_v_weak_loop_filter(uint8_t *src, const ptrdiff_t stride,
482 const int filter_p1, const int filter_q1,
483 const int alpha, const int beta,
484 const int lim_p0q0, const int lim_q1,
487 rv40_weak_loop_filter(src, 1, stride, filter_p1, filter_q1,
488 alpha, beta, lim_p0q0, lim_q1, lim_p1);
491 static av_always_inline void rv40_strong_loop_filter(uint8_t *src,
493 const ptrdiff_t stride,
501 for(i = 0; i < 4; i++, src += stride){
502 int sflag, p0, q0, p1, q1;
503 int t = src[0*step] - src[-1*step];
508 sflag = (alpha * FFABS(t)) >> 7;
512 p0 = (25*src[-3*step] + 26*src[-2*step] + 26*src[-1*step] +
513 26*src[ 0*step] + 25*src[ 1*step] +
514 rv40_dither_l[dmode + i]) >> 7;
516 q0 = (25*src[-2*step] + 26*src[-1*step] + 26*src[ 0*step] +
517 26*src[ 1*step] + 25*src[ 2*step] +
518 rv40_dither_r[dmode + i]) >> 7;
521 p0 = av_clip(p0, src[-1*step] - lims, src[-1*step] + lims);
522 q0 = av_clip(q0, src[ 0*step] - lims, src[ 0*step] + lims);
525 p1 = (25*src[-4*step] + 26*src[-3*step] + 26*src[-2*step] + 26*p0 +
526 25*src[ 0*step] + rv40_dither_l[dmode + i]) >> 7;
527 q1 = (25*src[-1*step] + 26*q0 + 26*src[ 1*step] + 26*src[ 2*step] +
528 25*src[ 3*step] + rv40_dither_r[dmode + i]) >> 7;
531 p1 = av_clip(p1, src[-2*step] - lims, src[-2*step] + lims);
532 q1 = av_clip(q1, src[ 1*step] - lims, src[ 1*step] + lims);
541 src[-3*step] = (25*src[-1*step] + 26*src[-2*step] +
542 51*src[-3*step] + 26*src[-4*step] + 64) >> 7;
543 src[ 2*step] = (25*src[ 0*step] + 26*src[ 1*step] +
544 51*src[ 2*step] + 26*src[ 3*step] + 64) >> 7;
549 static void rv40_h_strong_loop_filter(uint8_t *src, const ptrdiff_t stride,
550 const int alpha, const int lims,
551 const int dmode, const int chroma)
553 rv40_strong_loop_filter(src, stride, 1, alpha, lims, dmode, chroma);
556 static void rv40_v_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, 1, stride, alpha, lims, dmode, chroma);
563 static av_always_inline int rv40_loop_filter_strength(uint8_t *src,
564 int step, ptrdiff_t stride,
569 int sum_p1p0 = 0, sum_q1q0 = 0, sum_p1p2 = 0, sum_q1q2 = 0;
570 int strong0 = 0, strong1 = 0;
574 for (i = 0, ptr = src; i < 4; i++, ptr += stride) {
575 sum_p1p0 += ptr[-2*step] - ptr[-1*step];
576 sum_q1q0 += ptr[ 1*step] - ptr[ 0*step];
579 *p1 = FFABS(sum_p1p0) < (beta << 2);
580 *q1 = FFABS(sum_q1q0) < (beta << 2);
588 for (i = 0, ptr = src; i < 4; i++, ptr += stride) {
589 sum_p1p2 += ptr[-2*step] - ptr[-3*step];
590 sum_q1q2 += ptr[ 1*step] - ptr[ 2*step];
593 strong0 = *p1 && (FFABS(sum_p1p2) < beta2);
594 strong1 = *q1 && (FFABS(sum_q1q2) < beta2);
596 return strong0 && strong1;
599 static int rv40_h_loop_filter_strength(uint8_t *src, ptrdiff_t stride,
600 int beta, int beta2, int edge,
603 return rv40_loop_filter_strength(src, stride, 1, beta, beta2, edge, p1, q1);
606 static int rv40_v_loop_filter_strength(uint8_t *src, ptrdiff_t stride,
607 int beta, int beta2, int edge,
610 return rv40_loop_filter_strength(src, 1, stride, beta, beta2, edge, p1, q1);
613 av_cold void ff_rv40dsp_init(RV34DSPContext *c)
615 H264QpelContext qpel;
618 ff_h264qpel_init(&qpel, 8);
620 c->put_pixels_tab[0][ 0] = qpel.put_h264_qpel_pixels_tab[0][0];
621 c->put_pixels_tab[0][ 1] = put_rv40_qpel16_mc10_c;
622 c->put_pixels_tab[0][ 2] = qpel.put_h264_qpel_pixels_tab[0][2];
623 c->put_pixels_tab[0][ 3] = put_rv40_qpel16_mc30_c;
624 c->put_pixels_tab[0][ 4] = put_rv40_qpel16_mc01_c;
625 c->put_pixels_tab[0][ 5] = put_rv40_qpel16_mc11_c;
626 c->put_pixels_tab[0][ 6] = put_rv40_qpel16_mc21_c;
627 c->put_pixels_tab[0][ 7] = put_rv40_qpel16_mc31_c;
628 c->put_pixels_tab[0][ 8] = qpel.put_h264_qpel_pixels_tab[0][8];
629 c->put_pixels_tab[0][ 9] = put_rv40_qpel16_mc12_c;
630 c->put_pixels_tab[0][10] = put_rv40_qpel16_mc22_c;
631 c->put_pixels_tab[0][11] = put_rv40_qpel16_mc32_c;
632 c->put_pixels_tab[0][12] = put_rv40_qpel16_mc03_c;
633 c->put_pixels_tab[0][13] = put_rv40_qpel16_mc13_c;
634 c->put_pixels_tab[0][14] = put_rv40_qpel16_mc23_c;
635 c->put_pixels_tab[0][15] = put_rv40_qpel16_mc33_c;
636 c->avg_pixels_tab[0][ 0] = qpel.avg_h264_qpel_pixels_tab[0][0];
637 c->avg_pixels_tab[0][ 1] = avg_rv40_qpel16_mc10_c;
638 c->avg_pixels_tab[0][ 2] = qpel.avg_h264_qpel_pixels_tab[0][2];
639 c->avg_pixels_tab[0][ 3] = avg_rv40_qpel16_mc30_c;
640 c->avg_pixels_tab[0][ 4] = avg_rv40_qpel16_mc01_c;
641 c->avg_pixels_tab[0][ 5] = avg_rv40_qpel16_mc11_c;
642 c->avg_pixels_tab[0][ 6] = avg_rv40_qpel16_mc21_c;
643 c->avg_pixels_tab[0][ 7] = avg_rv40_qpel16_mc31_c;
644 c->avg_pixels_tab[0][ 8] = qpel.avg_h264_qpel_pixels_tab[0][8];
645 c->avg_pixels_tab[0][ 9] = avg_rv40_qpel16_mc12_c;
646 c->avg_pixels_tab[0][10] = avg_rv40_qpel16_mc22_c;
647 c->avg_pixels_tab[0][11] = avg_rv40_qpel16_mc32_c;
648 c->avg_pixels_tab[0][12] = avg_rv40_qpel16_mc03_c;
649 c->avg_pixels_tab[0][13] = avg_rv40_qpel16_mc13_c;
650 c->avg_pixels_tab[0][14] = avg_rv40_qpel16_mc23_c;
651 c->avg_pixels_tab[0][15] = avg_rv40_qpel16_mc33_c;
652 c->put_pixels_tab[1][ 0] = qpel.put_h264_qpel_pixels_tab[1][0];
653 c->put_pixels_tab[1][ 1] = put_rv40_qpel8_mc10_c;
654 c->put_pixels_tab[1][ 2] = qpel.put_h264_qpel_pixels_tab[1][2];
655 c->put_pixels_tab[1][ 3] = put_rv40_qpel8_mc30_c;
656 c->put_pixels_tab[1][ 4] = put_rv40_qpel8_mc01_c;
657 c->put_pixels_tab[1][ 5] = put_rv40_qpel8_mc11_c;
658 c->put_pixels_tab[1][ 6] = put_rv40_qpel8_mc21_c;
659 c->put_pixels_tab[1][ 7] = put_rv40_qpel8_mc31_c;
660 c->put_pixels_tab[1][ 8] = qpel.put_h264_qpel_pixels_tab[1][8];
661 c->put_pixels_tab[1][ 9] = put_rv40_qpel8_mc12_c;
662 c->put_pixels_tab[1][10] = put_rv40_qpel8_mc22_c;
663 c->put_pixels_tab[1][11] = put_rv40_qpel8_mc32_c;
664 c->put_pixels_tab[1][12] = put_rv40_qpel8_mc03_c;
665 c->put_pixels_tab[1][13] = put_rv40_qpel8_mc13_c;
666 c->put_pixels_tab[1][14] = put_rv40_qpel8_mc23_c;
667 c->put_pixels_tab[1][15] = put_rv40_qpel8_mc33_c;
668 c->avg_pixels_tab[1][ 0] = qpel.avg_h264_qpel_pixels_tab[1][0];
669 c->avg_pixels_tab[1][ 1] = avg_rv40_qpel8_mc10_c;
670 c->avg_pixels_tab[1][ 2] = qpel.avg_h264_qpel_pixels_tab[1][2];
671 c->avg_pixels_tab[1][ 3] = avg_rv40_qpel8_mc30_c;
672 c->avg_pixels_tab[1][ 4] = avg_rv40_qpel8_mc01_c;
673 c->avg_pixels_tab[1][ 5] = avg_rv40_qpel8_mc11_c;
674 c->avg_pixels_tab[1][ 6] = avg_rv40_qpel8_mc21_c;
675 c->avg_pixels_tab[1][ 7] = avg_rv40_qpel8_mc31_c;
676 c->avg_pixels_tab[1][ 8] = qpel.avg_h264_qpel_pixels_tab[1][8];
677 c->avg_pixels_tab[1][ 9] = avg_rv40_qpel8_mc12_c;
678 c->avg_pixels_tab[1][10] = avg_rv40_qpel8_mc22_c;
679 c->avg_pixels_tab[1][11] = avg_rv40_qpel8_mc32_c;
680 c->avg_pixels_tab[1][12] = avg_rv40_qpel8_mc03_c;
681 c->avg_pixels_tab[1][13] = avg_rv40_qpel8_mc13_c;
682 c->avg_pixels_tab[1][14] = avg_rv40_qpel8_mc23_c;
683 c->avg_pixels_tab[1][15] = avg_rv40_qpel8_mc33_c;
685 c->put_chroma_pixels_tab[0] = put_rv40_chroma_mc8_c;
686 c->put_chroma_pixels_tab[1] = put_rv40_chroma_mc4_c;
687 c->avg_chroma_pixels_tab[0] = avg_rv40_chroma_mc8_c;
688 c->avg_chroma_pixels_tab[1] = avg_rv40_chroma_mc4_c;
690 c->rv40_weight_pixels_tab[0][0] = rv40_weight_func_rnd_16;
691 c->rv40_weight_pixels_tab[0][1] = rv40_weight_func_rnd_8;
692 c->rv40_weight_pixels_tab[1][0] = rv40_weight_func_nornd_16;
693 c->rv40_weight_pixels_tab[1][1] = rv40_weight_func_nornd_8;
695 c->rv40_weak_loop_filter[0] = rv40_h_weak_loop_filter;
696 c->rv40_weak_loop_filter[1] = rv40_v_weak_loop_filter;
697 c->rv40_strong_loop_filter[0] = rv40_h_strong_loop_filter;
698 c->rv40_strong_loop_filter[1] = rv40_v_strong_loop_filter;
699 c->rv40_loop_filter_strength[0] = rv40_h_loop_filter_strength;
700 c->rv40_loop_filter_strength[1] = rv40_v_loop_filter_strength;
703 ff_rv40dsp_init_aarch64(c);
705 ff_rv40dsp_init_arm(c);
707 ff_rv40dsp_init_x86(c);