2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
4 * This file is part of Libav.
6 * Libav 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 * Libav 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 Libav; 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"
26 #include "dsputil_x86.h"
28 /***********************************/
30 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \
31 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst, \
35 IDCT_ADD_FUNC(, 8, mmx)
36 IDCT_ADD_FUNC(, 10, sse2)
37 IDCT_ADD_FUNC(_dc, 8, mmxext)
38 IDCT_ADD_FUNC(_dc, 10, mmxext)
39 IDCT_ADD_FUNC(8_dc, 8, mmxext)
40 IDCT_ADD_FUNC(8_dc, 10, sse2)
41 IDCT_ADD_FUNC(8, 8, mmx)
42 IDCT_ADD_FUNC(8, 8, sse2)
43 IDCT_ADD_FUNC(8, 10, sse2)
44 IDCT_ADD_FUNC(, 10, avx)
45 IDCT_ADD_FUNC(8_dc, 10, avx)
46 IDCT_ADD_FUNC(8, 10, avx)
49 #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT) \
50 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
51 (uint8_t *dst, const int *block_offset, \
52 int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
54 IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
55 IDCT_ADD_REP_FUNC(8, 4, 8, mmxext)
56 IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
57 IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
58 IDCT_ADD_REP_FUNC(8, 4, 10, avx)
59 IDCT_ADD_REP_FUNC(, 16, 8, mmx)
60 IDCT_ADD_REP_FUNC(, 16, 8, mmxext)
61 IDCT_ADD_REP_FUNC(, 16, 8, sse2)
62 IDCT_ADD_REP_FUNC(, 16, 10, sse2)
63 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
64 IDCT_ADD_REP_FUNC(, 16intra, 8, mmxext)
65 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
66 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
67 IDCT_ADD_REP_FUNC(, 16, 10, avx)
68 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
71 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT) \
72 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
73 (uint8_t **dst, const int *block_offset, \
74 int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
76 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
77 IDCT_ADD_REP_FUNC2(, 8, 8, mmxext)
78 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
79 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
80 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
82 void ff_h264_luma_dc_dequant_idct_mmx(int16_t *output, int16_t *input, int qmul);
83 void ff_h264_luma_dc_dequant_idct_sse2(int16_t *output, int16_t *input, int qmul);
85 /***********************************/
88 void ff_h264_loop_filter_strength_mmxext(int16_t bS[2][4][4], uint8_t nnz[40],
91 int bidir, int edges, int step,
92 int mask_mv0, int mask_mv1, int field);
94 #define LF_FUNC(DIR, TYPE, DEPTH, OPT) \
95 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \
100 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
101 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \
106 #define LF_FUNCS(type, depth) \
107 LF_FUNC(h, chroma, depth, mmxext) \
108 LF_IFUNC(h, chroma_intra, depth, mmxext) \
109 LF_FUNC(v, chroma, depth, mmxext) \
110 LF_IFUNC(v, chroma_intra, depth, mmxext) \
111 LF_FUNC(h, luma, depth, mmxext) \
112 LF_IFUNC(h, luma_intra, depth, mmxext) \
113 LF_FUNC(h, luma, depth, sse2) \
114 LF_IFUNC(h, luma_intra, depth, sse2) \
115 LF_FUNC(v, luma, depth, sse2) \
116 LF_IFUNC(v, luma_intra, depth, sse2) \
117 LF_FUNC(h, chroma, depth, sse2) \
118 LF_IFUNC(h, chroma_intra, depth, sse2) \
119 LF_FUNC(v, chroma, depth, sse2) \
120 LF_IFUNC(v, chroma_intra, depth, sse2) \
121 LF_FUNC(h, luma, depth, avx) \
122 LF_IFUNC(h, luma_intra, depth, avx) \
123 LF_FUNC(v, luma, depth, avx) \
124 LF_IFUNC(v, luma_intra, depth, avx) \
125 LF_FUNC(h, chroma, depth, avx) \
126 LF_IFUNC(h, chroma_intra, depth, avx) \
127 LF_FUNC(v, chroma, depth, avx) \
128 LF_IFUNC(v, chroma_intra, depth, avx)
131 LF_FUNCS(uint16_t, 10)
133 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
134 LF_FUNC(v8, luma, 8, mmxext)
135 static void deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha,
136 int beta, int8_t *tc0)
138 if ((tc0[0] & tc0[1]) >= 0)
139 ff_deblock_v8_luma_8_mmxext(pix + 0, stride, alpha, beta, tc0);
140 if ((tc0[2] & tc0[3]) >= 0)
141 ff_deblock_v8_luma_8_mmxext(pix + 8, stride, alpha, beta, tc0 + 2);
143 LF_IFUNC(v8, luma_intra, 8, mmxext)
144 static void deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride,
147 ff_deblock_v8_luma_intra_8_mmxext(pix + 0, stride, alpha, beta);
148 ff_deblock_v8_luma_intra_8_mmxext(pix + 8, stride, alpha, beta);
150 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
152 LF_FUNC(v, luma, 10, mmxext)
153 LF_IFUNC(v, luma_intra, 10, mmxext)
155 /***********************************/
156 /* weighted prediction */
158 #define H264_WEIGHT(W, OPT) \
159 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride, \
160 int height, int log2_denom, \
161 int weight, int offset);
163 #define H264_BIWEIGHT(W, OPT) \
164 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src, \
165 int stride, int height, \
166 int log2_denom, int weightd, \
167 int weights, int offset);
169 #define H264_BIWEIGHT_MMX(W) \
170 H264_WEIGHT(W, mmxext) \
171 H264_BIWEIGHT(W, mmxext)
173 #define H264_BIWEIGHT_MMX_SSE(W) \
174 H264_BIWEIGHT_MMX(W) \
175 H264_WEIGHT(W, sse2) \
176 H264_BIWEIGHT(W, sse2) \
177 H264_BIWEIGHT(W, ssse3)
179 H264_BIWEIGHT_MMX_SSE(16)
180 H264_BIWEIGHT_MMX_SSE(8)
183 #define H264_WEIGHT_10(W, DEPTH, OPT) \
184 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
191 #define H264_BIWEIGHT_10(W, DEPTH, OPT) \
192 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
201 #define H264_BIWEIGHT_10_SSE(W, DEPTH) \
202 H264_WEIGHT_10(W, DEPTH, sse2) \
203 H264_WEIGHT_10(W, DEPTH, sse4) \
204 H264_BIWEIGHT_10(W, DEPTH, sse2) \
205 H264_BIWEIGHT_10(W, DEPTH, sse4)
207 H264_BIWEIGHT_10_SSE(16, 10)
208 H264_BIWEIGHT_10_SSE(8, 10)
209 H264_BIWEIGHT_10_SSE(4, 10)
211 av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
212 const int chroma_format_idc)
214 int cpu_flags = av_get_cpu_flags();
216 if (chroma_format_idc == 1 && EXTERNAL_MMXEXT(cpu_flags))
217 c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmxext;
219 if (bit_depth == 8) {
220 if (EXTERNAL_MMX(cpu_flags)) {
221 c->h264_idct_dc_add =
222 c->h264_idct_add = ff_h264_idct_add_8_mmx;
223 c->h264_idct8_dc_add =
224 c->h264_idct8_add = ff_h264_idct8_add_8_mmx;
226 c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
227 c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
228 if (chroma_format_idc == 1)
229 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
230 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
231 if (cpu_flags & AV_CPU_FLAG_CMOV)
232 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
234 if (EXTERNAL_MMXEXT(cpu_flags)) {
235 c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmxext;
236 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmxext;
237 c->h264_idct_add16 = ff_h264_idct_add16_8_mmxext;
238 c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmxext;
239 if (chroma_format_idc == 1)
240 c->h264_idct_add8 = ff_h264_idct_add8_8_mmxext;
241 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmxext;
243 c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_8_mmxext;
244 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmxext;
245 if (chroma_format_idc == 1) {
246 c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_8_mmxext;
247 c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmxext;
249 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
250 c->h264_v_loop_filter_luma = deblock_v_luma_8_mmxext;
251 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_mmxext;
252 c->h264_v_loop_filter_luma_intra = deblock_v_luma_intra_8_mmxext;
253 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
254 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
255 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmxext;
256 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmxext;
257 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmxext;
259 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmxext;
260 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmxext;
261 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmxext;
263 if (EXTERNAL_SSE2(cpu_flags)) {
264 c->h264_idct8_add = ff_h264_idct8_add_8_sse2;
266 c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
267 c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
268 if (chroma_format_idc == 1)
269 c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
270 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_sse2;
271 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
273 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
274 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
276 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
277 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
279 c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
280 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
281 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
282 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
284 if (EXTERNAL_SSSE3(cpu_flags)) {
285 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
286 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
288 if (EXTERNAL_AVX(cpu_flags)) {
289 c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
290 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
291 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
292 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
296 } else if (bit_depth == 10) {
297 if (EXTERNAL_MMX(cpu_flags)) {
298 if (EXTERNAL_MMXEXT(cpu_flags)) {
300 c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_mmxext;
301 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmxext;
302 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_mmxext;
303 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_mmxext;
304 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmxext;
305 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmxext;
306 #endif /* ARCH_X86_32 */
307 c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmxext;
308 if (EXTERNAL_SSE2(cpu_flags)) {
309 c->h264_idct_add = ff_h264_idct_add_10_sse2;
310 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
312 c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
313 if (chroma_format_idc == 1)
314 c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
315 c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
316 #if HAVE_ALIGNED_STACK
317 c->h264_idct8_add = ff_h264_idct8_add_10_sse2;
318 c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
319 #endif /* HAVE_ALIGNED_STACK */
321 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
322 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
323 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
325 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
326 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
327 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
329 c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_sse2;
330 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
331 #if HAVE_ALIGNED_STACK
332 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
333 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
334 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
335 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
336 #endif /* HAVE_ALIGNED_STACK */
338 if (EXTERNAL_SSE4(cpu_flags)) {
339 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
340 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
341 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
343 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
344 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
345 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
347 if (EXTERNAL_AVX(cpu_flags)) {
348 c->h264_idct_dc_add =
349 c->h264_idct_add = ff_h264_idct_add_10_avx;
350 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
352 c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
353 if (chroma_format_idc == 1)
354 c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
355 c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
356 #if HAVE_ALIGNED_STACK
357 c->h264_idct8_add = ff_h264_idct8_add_10_avx;
358 c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
359 #endif /* HAVE_ALIGNED_STACK */
361 c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_avx;
362 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
363 #if HAVE_ALIGNED_STACK
364 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
365 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;
366 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
367 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
368 #endif /* HAVE_ALIGNED_STACK */