2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "libavutil/cpu.h"
22 #include "libavutil/x86_cpu.h"
23 #include "libavcodec/h264dsp.h"
24 #include "dsputil_mmx.h"
26 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_3_1 ) = 0x0103010301030103ULL;
28 /***********************************/
31 void ff_h264_idct_add_mmx (uint8_t *dst, int16_t *block, int stride);
32 void ff_h264_idct8_add_mmx (uint8_t *dst, int16_t *block, int stride);
33 void ff_h264_idct8_add_sse2 (uint8_t *dst, int16_t *block, int stride);
34 void ff_h264_idct_dc_add_mmx2 (uint8_t *dst, int16_t *block, int stride);
35 void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride);
37 void ff_h264_idct_add16_mmx (uint8_t *dst, const int *block_offset,
38 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
39 void ff_h264_idct8_add4_mmx (uint8_t *dst, const int *block_offset,
40 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
41 void ff_h264_idct_add16_mmx2 (uint8_t *dst, const int *block_offset,
42 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
43 void ff_h264_idct_add16intra_mmx (uint8_t *dst, const int *block_offset,
44 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
45 void ff_h264_idct_add16intra_mmx2(uint8_t *dst, const int *block_offset,
46 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
47 void ff_h264_idct8_add4_mmx2 (uint8_t *dst, const int *block_offset,
48 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
49 void ff_h264_idct8_add4_sse2 (uint8_t *dst, const int *block_offset,
50 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
51 void ff_h264_idct_add8_mmx (uint8_t **dest, const int *block_offset,
52 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
53 void ff_h264_idct_add8_mmx2 (uint8_t **dest, const int *block_offset,
54 DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
56 void ff_h264_idct_add16_sse2 (uint8_t *dst, const int *block_offset, DCTELEM *block,
57 int stride, const uint8_t nnzc[6*8]);
58 void ff_h264_idct_add16intra_sse2(uint8_t *dst, const int *block_offset, DCTELEM *block,
59 int stride, const uint8_t nnzc[6*8]);
60 void ff_h264_idct_add8_sse2 (uint8_t **dest, const int *block_offset, DCTELEM *block,
61 int stride, const uint8_t nnzc[6*8]);
63 /***********************************/
66 static av_always_inline void h264_loop_filter_strength_iteration_mmx2(int16_t bS[2][4][4], uint8_t nnz[40],
67 int8_t ref[2][40], int16_t mv[2][40][2],
68 int bidir, int edges, int step,
71 const x86_reg d_idx = dir ? -8 : -1;
72 DECLARE_ALIGNED(8, const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
74 for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
79 if(!(mask_mv & edge)) {
82 "movd (%1,%0), %%mm2 \n"
83 "punpckldq 40(%1,%0), %%mm2 \n" // { ref0[bn], ref1[bn] }
84 "pshufw $0x44, (%1), %%mm0 \n" // { ref0[b], ref0[b] }
85 "pshufw $0x44, 40(%1), %%mm1 \n" // { ref1[b], ref1[b] }
86 "pshufw $0x4E, %%mm2, %%mm3 \n"
87 "psubb %%mm2, %%mm0 \n" // { ref0[b]!=ref0[bn], ref0[b]!=ref1[bn] }
88 "psubb %%mm3, %%mm1 \n" // { ref1[b]!=ref1[bn], ref1[b]!=ref0[bn] }
91 "movq (%2,%0,4), %%mm1 \n"
92 "movq 8(%2,%0,4), %%mm2 \n"
93 "movq %%mm1, %%mm3 \n"
94 "movq %%mm2, %%mm4 \n"
95 "psubw (%2), %%mm1 \n"
96 "psubw 8(%2), %%mm2 \n"
97 "psubw 160(%2), %%mm3 \n"
98 "psubw 168(%2), %%mm4 \n"
99 "packsswb %%mm2, %%mm1 \n"
100 "packsswb %%mm4, %%mm3 \n"
101 "paddb %%mm6, %%mm1 \n"
102 "paddb %%mm6, %%mm3 \n"
103 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
104 "psubusb %%mm5, %%mm3 \n"
105 "packsswb %%mm3, %%mm1 \n"
107 "por %%mm1, %%mm0 \n"
108 "movq 160(%2,%0,4), %%mm1 \n"
109 "movq 168(%2,%0,4), %%mm2 \n"
110 "movq %%mm1, %%mm3 \n"
111 "movq %%mm2, %%mm4 \n"
112 "psubw (%2), %%mm1 \n"
113 "psubw 8(%2), %%mm2 \n"
114 "psubw 160(%2), %%mm3 \n"
115 "psubw 168(%2), %%mm4 \n"
116 "packsswb %%mm2, %%mm1 \n"
117 "packsswb %%mm4, %%mm3 \n"
118 "paddb %%mm6, %%mm1 \n"
119 "paddb %%mm6, %%mm3 \n"
120 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
121 "psubusb %%mm5, %%mm3 \n"
122 "packsswb %%mm3, %%mm1 \n"
124 "pshufw $0x4E, %%mm1, %%mm1 \n"
125 "por %%mm1, %%mm0 \n"
126 "pshufw $0x4E, %%mm0, %%mm1 \n"
127 "pminub %%mm1, %%mm0 \n"
134 "movd (%1), %%mm0 \n"
135 "psubb (%1,%0), %%mm0 \n" // ref[b] != ref[bn]
136 "movq (%2), %%mm1 \n"
137 "movq 8(%2), %%mm2 \n"
138 "psubw (%2,%0,4), %%mm1 \n"
139 "psubw 8(%2,%0,4), %%mm2 \n"
140 "packsswb %%mm2, %%mm1 \n"
141 "paddb %%mm6, %%mm1 \n"
142 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
143 "packsswb %%mm1, %%mm1 \n"
144 "por %%mm1, %%mm0 \n"
153 "por %1, %%mm1 \n" // nnz[b] || nnz[bn]
155 "m"(nnz[b_idx+d_idx])
158 "pminub %%mm7, %%mm1 \n"
159 "pminub %%mm7, %%mm0 \n"
161 "pxor %%mm2, %%mm2 \n"
162 "pmaxub %%mm0, %%mm1 \n"
163 "punpcklbw %%mm2, %%mm1 \n"
165 :"=m"(*bS[dir][edge])
171 static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
172 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
176 ::"m"(ff_pb_1), "m"(ff_pb_3)
184 "movq %%mm6, %%mm5 \n"
185 "paddb %%mm5, %%mm5 \n"
188 // could do a special case for dir==0 && edges==1, but it only reduces the
189 // average filter time by 1.2%
190 h264_loop_filter_strength_iteration_mmx2(bS, nnz, ref, mv, bidir, edges, step, mask_mv1, 1);
191 h264_loop_filter_strength_iteration_mmx2(bS, nnz, ref, mv, bidir, 4, 1, mask_mv0, 0);
194 "movq (%0), %%mm0 \n\t"
195 "movq 8(%0), %%mm1 \n\t"
196 "movq 16(%0), %%mm2 \n\t"
197 "movq 24(%0), %%mm3 \n\t"
198 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
199 "movq %%mm0, (%0) \n\t"
200 "movq %%mm3, 8(%0) \n\t"
201 "movq %%mm4, 16(%0) \n\t"
202 "movq %%mm2, 24(%0) \n\t"
208 #define LF_FUNC(DIR, TYPE, OPT) \
209 void ff_x264_deblock_ ## DIR ## _ ## TYPE ## _ ## OPT (uint8_t *pix, int stride, \
210 int alpha, int beta, int8_t *tc0);
211 #define LF_IFUNC(DIR, TYPE, OPT) \
212 void ff_x264_deblock_ ## DIR ## _ ## TYPE ## _ ## OPT (uint8_t *pix, int stride, \
213 int alpha, int beta);
215 LF_FUNC (h, chroma, mmxext)
216 LF_IFUNC(h, chroma_intra, mmxext)
217 LF_FUNC (v, chroma, mmxext)
218 LF_IFUNC(v, chroma_intra, mmxext)
220 LF_FUNC (h, luma, mmxext)
221 LF_IFUNC(h, luma_intra, mmxext)
222 #if HAVE_YASM && ARCH_X86_32
223 LF_FUNC (v8, luma, mmxext)
224 static void ff_x264_deblock_v_luma_mmxext(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
226 if((tc0[0] & tc0[1]) >= 0)
227 ff_x264_deblock_v8_luma_mmxext(pix+0, stride, alpha, beta, tc0);
228 if((tc0[2] & tc0[3]) >= 0)
229 ff_x264_deblock_v8_luma_mmxext(pix+8, stride, alpha, beta, tc0+2);
231 LF_IFUNC(v8, luma_intra, mmxext)
232 static void ff_x264_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta)
234 ff_x264_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta);
235 ff_x264_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta);
239 LF_FUNC (h, luma, sse2)
240 LF_IFUNC(h, luma_intra, sse2)
241 LF_FUNC (v, luma, sse2)
242 LF_IFUNC(v, luma_intra, sse2)
244 /***********************************/
245 /* weighted prediction */
247 #define H264_WEIGHT(W, H, OPT) \
248 void ff_h264_weight_ ## W ## x ## H ## _ ## OPT(uint8_t *dst, \
249 int stride, int log2_denom, int weight, int offset);
251 #define H264_BIWEIGHT(W, H, OPT) \
252 void ff_h264_biweight_ ## W ## x ## H ## _ ## OPT(uint8_t *dst, \
253 uint8_t *src, int stride, int log2_denom, int weightd, \
254 int weights, int offset);
256 #define H264_BIWEIGHT_MMX(W,H) \
257 H264_WEIGHT (W, H, mmx2) \
258 H264_BIWEIGHT(W, H, mmx2)
260 #define H264_BIWEIGHT_MMX_SSE(W,H) \
261 H264_BIWEIGHT_MMX(W, H) \
262 H264_WEIGHT (W, H, sse2) \
263 H264_BIWEIGHT (W, H, sse2) \
264 H264_BIWEIGHT (W, H, ssse3)
266 H264_BIWEIGHT_MMX_SSE(16, 16)
267 H264_BIWEIGHT_MMX_SSE(16, 8)
268 H264_BIWEIGHT_MMX_SSE( 8, 16)
269 H264_BIWEIGHT_MMX_SSE( 8, 8)
270 H264_BIWEIGHT_MMX_SSE( 8, 4)
271 H264_BIWEIGHT_MMX ( 4, 8)
272 H264_BIWEIGHT_MMX ( 4, 4)
273 H264_BIWEIGHT_MMX ( 4, 2)
275 void ff_h264dsp_init_x86(H264DSPContext *c)
277 int mm_flags = av_get_cpu_flags();
279 if (mm_flags & AV_CPU_FLAG_MMX2) {
280 c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;
283 if (mm_flags & AV_CPU_FLAG_MMX) {
285 c->h264_idct_add= ff_h264_idct_add_mmx;
286 c->h264_idct8_dc_add=
287 c->h264_idct8_add= ff_h264_idct8_add_mmx;
289 c->h264_idct_add16 = ff_h264_idct_add16_mmx;
290 c->h264_idct8_add4 = ff_h264_idct8_add4_mmx;
291 c->h264_idct_add8 = ff_h264_idct_add8_mmx;
292 c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx;
294 if (mm_flags & AV_CPU_FLAG_MMX2) {
295 c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2;
296 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2;
297 c->h264_idct_add16 = ff_h264_idct_add16_mmx2;
298 c->h264_idct8_add4 = ff_h264_idct8_add4_mmx2;
299 c->h264_idct_add8 = ff_h264_idct_add8_mmx2;
300 c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx2;
302 c->h264_v_loop_filter_chroma= ff_x264_deblock_v_chroma_mmxext;
303 c->h264_h_loop_filter_chroma= ff_x264_deblock_h_chroma_mmxext;
304 c->h264_v_loop_filter_chroma_intra= ff_x264_deblock_v_chroma_intra_mmxext;
305 c->h264_h_loop_filter_chroma_intra= ff_x264_deblock_h_chroma_intra_mmxext;
307 c->h264_v_loop_filter_luma= ff_x264_deblock_v_luma_mmxext;
308 c->h264_h_loop_filter_luma= ff_x264_deblock_h_luma_mmxext;
309 c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_mmxext;
310 c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_mmxext;
312 c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
313 c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
314 c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;
315 c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;
316 c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;
317 c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;
318 c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;
319 c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;
321 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;
322 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;
323 c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;
324 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;
325 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;
326 c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;
327 c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;
328 c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;
330 if (mm_flags&AV_CPU_FLAG_SSE2) {
331 c->h264_idct8_add = ff_h264_idct8_add_sse2;
332 c->h264_idct8_add4= ff_h264_idct8_add4_sse2;
334 c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_sse2;
335 c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_sse2;
336 c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_sse2;
337 c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_sse2;
338 c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_sse2;
340 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_sse2;
341 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_sse2;
342 c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_sse2;
343 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_sse2;
344 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_sse2;
346 #if HAVE_ALIGNED_STACK
347 c->h264_v_loop_filter_luma = ff_x264_deblock_v_luma_sse2;
348 c->h264_h_loop_filter_luma = ff_x264_deblock_h_luma_sse2;
349 c->h264_v_loop_filter_luma_intra = ff_x264_deblock_v_luma_intra_sse2;
350 c->h264_h_loop_filter_luma_intra = ff_x264_deblock_h_luma_intra_sse2;
353 c->h264_idct_add16 = ff_h264_idct_add16_sse2;
354 c->h264_idct_add8 = ff_h264_idct_add8_sse2;
355 c->h264_idct_add16intra = ff_h264_idct_add16intra_sse2;
357 if (mm_flags&AV_CPU_FLAG_SSSE3) {
358 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_ssse3;
359 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_ssse3;
360 c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_ssse3;
361 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_ssse3;
362 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_ssse3;