h264dsp_mmx.c 14.8 KB
Newer Older
1 2 3
/*
 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
 *
4
 * This file is part of Libav.
5
 *
6
 * Libav is free software; you can redistribute it and/or
7 8
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
9
 * version 2.1 of the License, or (at your option) any later version.
10
 *
11
 * Libav is distributed in the hope that it will be useful,
12 13 14 15 16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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
19 20
 */

21
#include "libavutil/cpu.h"
22 23
#include "libavutil/x86_cpu.h"
#include "libavcodec/h264dsp.h"
24
#include "dsputil_mmx.h"
25 26 27

/***********************************/
/* IDCT */
28 29 30
#define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \
void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT (uint8_t *dst, int16_t *block, int stride);

31
IDCT_ADD_FUNC(, 8, mmx)
32
IDCT_ADD_FUNC(, 10, sse2)
33
IDCT_ADD_FUNC(_dc, 8, mmx2)
34
IDCT_ADD_FUNC(_dc, 10, mmx2)
35
IDCT_ADD_FUNC(8_dc, 8, mmx2)
36
IDCT_ADD_FUNC(8_dc, 10, sse2)
37 38
IDCT_ADD_FUNC(8, 8, mmx)
IDCT_ADD_FUNC(8, 8, sse2)
39 40 41 42 43 44 45 46 47 48 49 50 51
IDCT_ADD_FUNC(8, 10, sse2)
#if HAVE_AVX
IDCT_ADD_FUNC(, 10, avx)
IDCT_ADD_FUNC(8_dc, 10, avx)
IDCT_ADD_FUNC(8, 10, avx)
#endif


#define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT) \
void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
                              (uint8_t *dst, const int *block_offset, \
                              DCTELEM *block, int stride, const uint8_t nnzc[6*8]);

52 53 54
IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
IDCT_ADD_REP_FUNC(8, 4, 8, mmx2)
IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
55 56
IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
IDCT_ADD_REP_FUNC(8, 4, 10, avx)
57 58 59
IDCT_ADD_REP_FUNC(, 16, 8, mmx)
IDCT_ADD_REP_FUNC(, 16, 8, mmx2)
IDCT_ADD_REP_FUNC(, 16, 8, sse2)
60
IDCT_ADD_REP_FUNC(, 16, 10, sse2)
61 62 63
IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
IDCT_ADD_REP_FUNC(, 16intra, 8, mmx2)
IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
64 65 66 67 68 69 70 71 72 73 74
IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
#if HAVE_AVX
IDCT_ADD_REP_FUNC(, 16, 10, avx)
IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
#endif


#define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT) \
void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
                              (uint8_t **dst, const int *block_offset, \
                              DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
75 76 77
IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
IDCT_ADD_REP_FUNC2(, 8, 8, mmx2)
IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
78 79 80 81
IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
#if HAVE_AVX
IDCT_ADD_REP_FUNC2(, 8, 10, avx)
#endif
82

83 84
void ff_h264_luma_dc_dequant_idct_mmx (DCTELEM *output, DCTELEM *input, int qmul);
void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul);
85

86 87 88
/***********************************/
/* deblocking */

89 90 91 92
void ff_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],
                                       int bidir, int edges, int step,
                                       int mask_mv0, int mask_mv1, int field);
93

94 95 96 97 98 99 100 101
#define LF_FUNC(DIR, TYPE, DEPTH, OPT) \
void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *pix, int stride, \
                                                                int alpha, int beta, int8_t *tc0);
#define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *pix, int stride, \
                                                                int alpha, int beta);

#define LF_FUNCS(type, depth)\
102 103 104 105 106 107
LF_FUNC (h,  chroma,       depth, mmx2)\
LF_IFUNC(h,  chroma_intra, depth, mmx2)\
LF_FUNC (v,  chroma,       depth, mmx2)\
LF_IFUNC(v,  chroma_intra, depth, mmx2)\
LF_FUNC (h,  luma,         depth, mmx2)\
LF_IFUNC(h,  luma_intra,   depth, mmx2)\
108 109 110 111
LF_FUNC (h,  luma,         depth, sse2)\
LF_IFUNC(h,  luma_intra,   depth, sse2)\
LF_FUNC (v,  luma,         depth, sse2)\
LF_IFUNC(v,  luma_intra,   depth, sse2)\
112 113 114 115
LF_FUNC (h,  chroma,       depth, sse2)\
LF_IFUNC(h,  chroma_intra, depth, sse2)\
LF_FUNC (v,  chroma,       depth, sse2)\
LF_IFUNC(v,  chroma_intra, depth, sse2)\
116 117 118
LF_FUNC (h,  luma,         depth,  avx)\
LF_IFUNC(h,  luma_intra,   depth,  avx)\
LF_FUNC (v,  luma,         depth,  avx)\
119 120 121 122 123
LF_IFUNC(v,  luma_intra,   depth,  avx)\
LF_FUNC (h,  chroma,       depth,  avx)\
LF_IFUNC(h,  chroma_intra, depth,  avx)\
LF_FUNC (v,  chroma,       depth,  avx)\
LF_IFUNC(v,  chroma_intra, depth,  avx)
124 125 126 127

LF_FUNCS( uint8_t,  8)
LF_FUNCS(uint16_t, 10)

128
#if ARCH_X86_32
129 130
LF_FUNC (v8, luma,             8, mmx2)
static void ff_deblock_v_luma_8_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
131 132
{
    if((tc0[0] & tc0[1]) >= 0)
133
        ff_deblock_v8_luma_8_mmx2(pix+0, stride, alpha, beta, tc0);
134
    if((tc0[2] & tc0[3]) >= 0)
135
        ff_deblock_v8_luma_8_mmx2(pix+8, stride, alpha, beta, tc0+2);
136
}
137 138
LF_IFUNC(v8, luma_intra,        8, mmx2)
static void ff_deblock_v_luma_intra_8_mmx2(uint8_t *pix, int stride, int alpha, int beta)
139
{
140 141
    ff_deblock_v8_luma_intra_8_mmx2(pix+0, stride, alpha, beta);
    ff_deblock_v8_luma_intra_8_mmx2(pix+8, stride, alpha, beta);
142
}
143
#endif /* ARCH_X86_32 */
144

145 146
LF_FUNC (v,  luma,            10, mmx2)
LF_IFUNC(v,  luma_intra,      10, mmx2)
147

148 149 150
/***********************************/
/* weighted prediction */

151 152 153
#define H264_WEIGHT(W, OPT) \
void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, \
    int stride, int height, int log2_denom, int weight, int offset);
154

155 156 157
#define H264_BIWEIGHT(W, OPT) \
void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, \
    uint8_t *src, int stride, int height, int log2_denom, int weightd, \
158 159
    int weights, int offset);

160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
#define H264_BIWEIGHT_MMX(W) \
H264_WEIGHT  (W, mmx2) \
H264_BIWEIGHT(W, mmx2)

#define H264_BIWEIGHT_MMX_SSE(W) \
H264_BIWEIGHT_MMX(W) \
H264_WEIGHT      (W, sse2) \
H264_BIWEIGHT    (W, sse2) \
H264_BIWEIGHT    (W, ssse3)

H264_BIWEIGHT_MMX_SSE(16)
H264_BIWEIGHT_MMX_SSE( 8)
H264_BIWEIGHT_MMX    ( 4)

#define H264_WEIGHT_10(W, DEPTH, OPT) \
void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
    int stride, int height, int log2_denom, int weight, int offset);

#define H264_BIWEIGHT_10(W, DEPTH, OPT) \
void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT \
    (uint8_t *dst, uint8_t *src, int stride, int height, int log2_denom, \
181 182
     int weightd, int weights, int offset);

183 184 185 186 187 188 189 190 191
#define H264_BIWEIGHT_10_SSE(W, DEPTH) \
H264_WEIGHT_10  (W, DEPTH, sse2) \
H264_WEIGHT_10  (W, DEPTH, sse4) \
H264_BIWEIGHT_10(W, DEPTH, sse2) \
H264_BIWEIGHT_10(W, DEPTH, sse4)

H264_BIWEIGHT_10_SSE(16, 10)
H264_BIWEIGHT_10_SSE( 8, 10)
H264_BIWEIGHT_10_SSE( 4, 10)
192

193
void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth, const int chroma_format_idc)
194
{
195
#if HAVE_YASM
196
    int mm_flags = av_get_cpu_flags();
197

198
    if (chroma_format_idc == 1 && mm_flags & AV_CPU_FLAG_MMX2) {
199
        c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmx2;
200
    }
201 202

    if (bit_depth == 8) {
203
    if (mm_flags & AV_CPU_FLAG_MMX) {
204 205 206 207 208 209 210
        c->h264_idct_dc_add         =
        c->h264_idct_add            = ff_h264_idct_add_8_mmx;
        c->h264_idct8_dc_add        =
        c->h264_idct8_add           = ff_h264_idct8_add_8_mmx;

        c->h264_idct_add16          = ff_h264_idct_add16_8_mmx;
        c->h264_idct8_add4          = ff_h264_idct8_add4_8_mmx;
211 212
        if (chroma_format_idc == 1)
            c->h264_idct_add8       = ff_h264_idct_add8_8_mmx;
213
        c->h264_idct_add16intra     = ff_h264_idct_add16intra_8_mmx;
214 215
        if (mm_flags & AV_CPU_FLAG_CMOV)
            c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
216

217
        if (mm_flags & AV_CPU_FLAG_MMX2) {
218 219 220 221
            c->h264_idct_dc_add    = ff_h264_idct_dc_add_8_mmx2;
            c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_8_mmx2;
            c->h264_idct_add16     = ff_h264_idct_add16_8_mmx2;
            c->h264_idct8_add4     = ff_h264_idct8_add4_8_mmx2;
222 223
            if (chroma_format_idc == 1)
                c->h264_idct_add8  = ff_h264_idct_add8_8_mmx2;
224
            c->h264_idct_add16intra= ff_h264_idct_add16intra_8_mmx2;
225

226 227
            c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_8_mmx2;
            c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_8_mmx2;
228
            if (chroma_format_idc == 1) {
229 230
                c->h264_h_loop_filter_chroma= ff_deblock_h_chroma_8_mmx2;
                c->h264_h_loop_filter_chroma_intra= ff_deblock_h_chroma_intra_8_mmx2;
231
            }
232
#if ARCH_X86_32
233 234 235 236
            c->h264_v_loop_filter_luma= ff_deblock_v_luma_8_mmx2;
            c->h264_h_loop_filter_luma= ff_deblock_h_luma_8_mmx2;
            c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmx2;
            c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmx2;
237
#endif
238 239 240 241 242 243 244
            c->weight_h264_pixels_tab[0]= ff_h264_weight_16_mmx2;
            c->weight_h264_pixels_tab[1]= ff_h264_weight_8_mmx2;
            c->weight_h264_pixels_tab[2]= ff_h264_weight_4_mmx2;

            c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16_mmx2;
            c->biweight_h264_pixels_tab[1]= ff_h264_biweight_8_mmx2;
            c->biweight_h264_pixels_tab[2]= ff_h264_biweight_4_mmx2;
245

246
            if (mm_flags&AV_CPU_FLAG_SSE2) {
247 248 249 250
                c->h264_idct8_add           = ff_h264_idct8_add_8_sse2;

                c->h264_idct_add16          = ff_h264_idct_add16_8_sse2;
                c->h264_idct8_add4          = ff_h264_idct8_add4_8_sse2;
251 252
                if (chroma_format_idc == 1)
                    c->h264_idct_add8       = ff_h264_idct_add8_8_sse2;
253
                c->h264_idct_add16intra     = ff_h264_idct_add16intra_8_sse2;
254
                c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_sse2;
255

256 257
                c->weight_h264_pixels_tab[0]= ff_h264_weight_16_sse2;
                c->weight_h264_pixels_tab[1]= ff_h264_weight_8_sse2;
258

259 260
                c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16_sse2;
                c->biweight_h264_pixels_tab[1]= ff_h264_biweight_8_sse2;
261

262
#if HAVE_ALIGNED_STACK
263 264 265 266
                c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
                c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
267
#endif
268
            }
269
            if (mm_flags&AV_CPU_FLAG_SSSE3) {
270 271
                c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16_ssse3;
                c->biweight_h264_pixels_tab[1]= ff_h264_biweight_8_ssse3;
272
            }
273 274
            if (mm_flags&AV_CPU_FLAG_AVX) {
#if HAVE_ALIGNED_STACK
275 276 277 278 279 280 281 282 283 284 285 286
                c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
                c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
#endif
            }
        }
    }
    } else if (bit_depth == 10) {
    if (mm_flags & AV_CPU_FLAG_MMX) {
        if (mm_flags & AV_CPU_FLAG_MMX2) {
#if ARCH_X86_32
287 288 289 290 291 292
            c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_mmx2;
            c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_mmx2;
            c->h264_v_loop_filter_luma= ff_deblock_v_luma_10_mmx2;
            c->h264_h_loop_filter_luma= ff_deblock_h_luma_10_mmx2;
            c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmx2;
            c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmx2;
293
#endif
294
            c->h264_idct_dc_add= ff_h264_idct_dc_add_10_mmx2;
295
            if (mm_flags&AV_CPU_FLAG_SSE2) {
296 297 298 299
                c->h264_idct_add       = ff_h264_idct_add_10_sse2;
                c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_10_sse2;

                c->h264_idct_add16     = ff_h264_idct_add16_10_sse2;
300 301
                if (chroma_format_idc == 1)
                    c->h264_idct_add8  = ff_h264_idct_add8_10_sse2;
302
                c->h264_idct_add16intra= ff_h264_idct_add16intra_10_sse2;
303 304 305 306
#if HAVE_ALIGNED_STACK
                c->h264_idct8_add      = ff_h264_idct8_add_10_sse2;
                c->h264_idct8_add4     = ff_h264_idct8_add4_10_sse2;
#endif
307

308 309 310 311 312 313 314
                c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
                c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
                c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;

                c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
                c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
                c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
315

316 317
                c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_sse2;
                c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_sse2;
318 319 320 321 322 323 324
#if HAVE_ALIGNED_STACK
                c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
                c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
#endif
            }
325
            if (mm_flags&AV_CPU_FLAG_SSE4) {
326 327 328 329 330 331 332
                c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
                c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
                c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;

                c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
                c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
                c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
333
            }
334
#if HAVE_AVX
335
            if (mm_flags&AV_CPU_FLAG_AVX) {
336 337 338 339 340
                c->h264_idct_dc_add    =
                c->h264_idct_add       = ff_h264_idct_add_10_avx;
                c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_10_avx;

                c->h264_idct_add16     = ff_h264_idct_add16_10_avx;
341 342
                if (chroma_format_idc == 1)
                    c->h264_idct_add8  = ff_h264_idct_add8_10_avx;
343
                c->h264_idct_add16intra= ff_h264_idct_add16intra_10_avx;
344 345 346 347
#if HAVE_ALIGNED_STACK
                c->h264_idct8_add      = ff_h264_idct8_add_10_avx;
                c->h264_idct8_add4     = ff_h264_idct8_add4_10_avx;
#endif
348

349 350
                c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_avx;
                c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_avx;
351 352 353 354 355
#if HAVE_ALIGNED_STACK
                c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
                c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;
                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
356 357
#endif
            }
358
#endif /* HAVE_AVX */
359 360
        }
    }
361
    }
362
#endif
363
}