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/attributes.h"
22 #include "libavutil/cpu.h"
23 #include "libavutil/x86/asm.h"
24 #include "libavutil/x86/cpu.h"
25 #include "libavcodec/h264dsp.h"
27 /***********************************/
29 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \
30 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst, \
34 IDCT_ADD_FUNC(, 8, mmx)
35 IDCT_ADD_FUNC(, 10, sse2)
36 IDCT_ADD_FUNC(_dc, 8, mmxext)
37 IDCT_ADD_FUNC(_dc, 10, mmxext)
38 IDCT_ADD_FUNC(8_dc, 8, mmxext)
39 IDCT_ADD_FUNC(8_dc, 10, sse2)
40 IDCT_ADD_FUNC(8, 8, mmx)
41 IDCT_ADD_FUNC(8, 8, sse2)
42 IDCT_ADD_FUNC(8, 10, sse2)
43 IDCT_ADD_FUNC(, 10, avx)
44 IDCT_ADD_FUNC(8_dc, 10, avx)
45 IDCT_ADD_FUNC(8, 10, avx)
48 #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT) \
49 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
50 (uint8_t *dst, const int *block_offset, \
51 int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
53 IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
54 IDCT_ADD_REP_FUNC(8, 4, 8, mmxext)
55 IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
56 IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
57 IDCT_ADD_REP_FUNC(8, 4, 10, avx)
58 IDCT_ADD_REP_FUNC(, 16, 8, mmx)
59 IDCT_ADD_REP_FUNC(, 16, 8, mmxext)
60 IDCT_ADD_REP_FUNC(, 16, 8, sse2)
61 IDCT_ADD_REP_FUNC(, 16, 10, sse2)
62 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
63 IDCT_ADD_REP_FUNC(, 16intra, 8, mmxext)
64 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
65 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
66 IDCT_ADD_REP_FUNC(, 16, 10, avx)
67 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
70 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT) \
71 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
72 (uint8_t **dst, const int *block_offset, \
73 int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
75 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
76 IDCT_ADD_REP_FUNC2(, 8, 8, mmxext)
77 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
78 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
79 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
81 void ff_h264_luma_dc_dequant_idct_mmx(int16_t *output, int16_t *input, int qmul);
82 void ff_h264_luma_dc_dequant_idct_sse2(int16_t *output, int16_t *input, int qmul);
84 /***********************************/
87 void ff_h264_loop_filter_strength_mmxext(int16_t bS[2][4][4], uint8_t nnz[40],
90 int bidir, int edges, int step,
91 int mask_mv0, int mask_mv1, int field);
93 #define LF_FUNC(DIR, TYPE, DEPTH, OPT) \
94 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \
99 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
100 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \
105 #define LF_FUNCS(type, depth) \
106 LF_FUNC(h, chroma, depth, mmxext) \
107 LF_IFUNC(h, chroma_intra, depth, mmxext) \
108 LF_FUNC(v, chroma, depth, mmxext) \
109 LF_IFUNC(v, chroma_intra, depth, mmxext) \
110 LF_FUNC(h, luma, depth, mmxext) \
111 LF_IFUNC(h, luma_intra, depth, mmxext) \
112 LF_FUNC(h, luma, depth, sse2) \
113 LF_IFUNC(h, luma_intra, depth, sse2) \
114 LF_FUNC(v, luma, depth, sse2) \
115 LF_IFUNC(v, luma_intra, depth, sse2) \
116 LF_FUNC(h, chroma, depth, sse2) \
117 LF_IFUNC(h, chroma_intra, depth, sse2) \
118 LF_FUNC(v, chroma, depth, sse2) \
119 LF_IFUNC(v, chroma_intra, depth, sse2) \
120 LF_FUNC(h, luma, depth, avx) \
121 LF_IFUNC(h, luma_intra, depth, avx) \
122 LF_FUNC(v, luma, depth, avx) \
123 LF_IFUNC(v, luma_intra, depth, avx) \
124 LF_FUNC(h, chroma, depth, avx) \
125 LF_IFUNC(h, chroma_intra, depth, avx) \
126 LF_FUNC(v, chroma, depth, avx) \
127 LF_IFUNC(v, chroma_intra, depth, avx)
130 LF_FUNCS(uint16_t, 10)
132 void ff_deblock_h_chroma422_8_mmxext(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
134 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
135 LF_FUNC(v8, luma, 8, mmxext)
136 static void deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha,
137 int beta, int8_t *tc0)
139 if ((tc0[0] & tc0[1]) >= 0)
140 ff_deblock_v8_luma_8_mmxext(pix + 0, stride, alpha, beta, tc0);
141 if ((tc0[2] & tc0[3]) >= 0)
142 ff_deblock_v8_luma_8_mmxext(pix + 8, stride, alpha, beta, tc0 + 2);
144 LF_IFUNC(v8, luma_intra, 8, mmxext)
145 static void deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride,
148 ff_deblock_v8_luma_intra_8_mmxext(pix + 0, stride, alpha, beta);
149 ff_deblock_v8_luma_intra_8_mmxext(pix + 8, stride, alpha, beta);
151 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
153 LF_FUNC(v, luma, 10, mmxext)
154 LF_IFUNC(v, luma_intra, 10, mmxext)
156 /***********************************/
157 /* weighted prediction */
159 #define H264_WEIGHT(W, OPT) \
160 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride, \
161 int height, int log2_denom, \
162 int weight, int offset);
164 #define H264_BIWEIGHT(W, OPT) \
165 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src, \
166 int stride, int height, \
167 int log2_denom, int weightd, \
168 int weights, int offset);
170 #define H264_BIWEIGHT_MMX(W) \
171 H264_WEIGHT(W, mmxext) \
172 H264_BIWEIGHT(W, mmxext)
174 #define H264_BIWEIGHT_MMX_SSE(W) \
175 H264_BIWEIGHT_MMX(W) \
176 H264_WEIGHT(W, sse2) \
177 H264_BIWEIGHT(W, sse2) \
178 H264_BIWEIGHT(W, ssse3)
180 H264_BIWEIGHT_MMX_SSE(16)
181 H264_BIWEIGHT_MMX_SSE(8)
184 #define H264_WEIGHT_10(W, DEPTH, OPT) \
185 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
192 #define H264_BIWEIGHT_10(W, DEPTH, OPT) \
193 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
202 #define H264_BIWEIGHT_10_SSE(W, DEPTH) \
203 H264_WEIGHT_10(W, DEPTH, sse2) \
204 H264_WEIGHT_10(W, DEPTH, sse4) \
205 H264_BIWEIGHT_10(W, DEPTH, sse2) \
206 H264_BIWEIGHT_10(W, DEPTH, sse4)
208 H264_BIWEIGHT_10_SSE(16, 10)
209 H264_BIWEIGHT_10_SSE(8, 10)
210 H264_BIWEIGHT_10_SSE(4, 10)
212 av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
213 const int chroma_format_idc)
216 int cpu_flags = av_get_cpu_flags();
218 if (EXTERNAL_MMXEXT(cpu_flags) && chroma_format_idc <= 1)
219 c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmxext;
221 if (bit_depth == 8) {
222 if (EXTERNAL_MMX(cpu_flags)) {
223 c->h264_idct_dc_add =
224 c->h264_idct_add = ff_h264_idct_add_8_mmx;
225 c->h264_idct8_dc_add =
226 c->h264_idct8_add = ff_h264_idct8_add_8_mmx;
228 c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
229 c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
230 if (chroma_format_idc <= 1)
231 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
232 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
233 if (cpu_flags & AV_CPU_FLAG_CMOV)
234 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
236 if (EXTERNAL_MMXEXT(cpu_flags)) {
237 c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmxext;
238 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmxext;
239 c->h264_idct_add16 = ff_h264_idct_add16_8_mmxext;
240 c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmxext;
241 if (chroma_format_idc <= 1)
242 c->h264_idct_add8 = ff_h264_idct_add8_8_mmxext;
243 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmxext;
245 c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_8_mmxext;
246 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmxext;
247 if (chroma_format_idc <= 1) {
248 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_8_mmxext;
249 c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmxext;
251 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma422_8_mmxext;
253 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
254 c->h264_v_loop_filter_luma = deblock_v_luma_8_mmxext;
255 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_mmxext;
256 c->h264_v_loop_filter_luma_intra = deblock_v_luma_intra_8_mmxext;
257 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
258 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
259 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmxext;
260 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmxext;
261 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmxext;
263 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmxext;
264 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmxext;
265 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmxext;
267 if (EXTERNAL_SSE2(cpu_flags)) {
268 c->h264_idct8_add = ff_h264_idct8_add_8_sse2;
270 c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
271 c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
272 if (chroma_format_idc <= 1)
273 c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
274 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_sse2;
275 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
277 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
278 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
280 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
281 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
283 c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
284 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
285 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
286 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
288 if (EXTERNAL_SSSE3(cpu_flags)) {
289 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
290 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
292 if (EXTERNAL_AVX(cpu_flags)) {
293 c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
294 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
295 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
296 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
298 } else if (bit_depth == 10) {
299 if (EXTERNAL_MMXEXT(cpu_flags)) {
301 c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_mmxext;
302 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmxext;
303 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_mmxext;
304 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_mmxext;
305 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmxext;
306 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmxext;
307 #endif /* ARCH_X86_32 */
308 c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmxext;
310 if (EXTERNAL_SSE2(cpu_flags)) {
311 c->h264_idct_add = ff_h264_idct_add_10_sse2;
312 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
314 c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
315 if (chroma_format_idc <= 1)
316 c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
317 c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
318 #if HAVE_ALIGNED_STACK
319 c->h264_idct8_add = ff_h264_idct8_add_10_sse2;
320 c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
321 #endif /* HAVE_ALIGNED_STACK */
323 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
324 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
325 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
327 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
328 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
329 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
331 c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_sse2;
332 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
333 #if HAVE_ALIGNED_STACK
334 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
335 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
336 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
337 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
338 #endif /* HAVE_ALIGNED_STACK */
340 if (EXTERNAL_SSE4(cpu_flags)) {
341 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
342 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
343 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
345 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
346 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
347 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
349 if (EXTERNAL_AVX(cpu_flags)) {
350 c->h264_idct_dc_add =
351 c->h264_idct_add = ff_h264_idct_add_10_avx;
352 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
354 c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
355 if (chroma_format_idc <= 1)
356 c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
357 c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
358 #if HAVE_ALIGNED_STACK
359 c->h264_idct8_add = ff_h264_idct8_add_10_avx;
360 c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
361 #endif /* HAVE_ALIGNED_STACK */
363 c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_avx;
364 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
365 #if HAVE_ALIGNED_STACK
366 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
367 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;
368 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
369 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
370 #endif /* HAVE_ALIGNED_STACK */