1cabdff1aSopenharmony_ci/*
2cabdff1aSopenharmony_ci * This file is part of FFmpeg.
3cabdff1aSopenharmony_ci *
4cabdff1aSopenharmony_ci * FFmpeg is free software; you can redistribute it and/or
5cabdff1aSopenharmony_ci * modify it under the terms of the GNU Lesser General Public
6cabdff1aSopenharmony_ci * License as published by the Free Software Foundation; either
7cabdff1aSopenharmony_ci * version 2.1 of the License, or (at your option) any later version.
8cabdff1aSopenharmony_ci *
9cabdff1aSopenharmony_ci * FFmpeg is distributed in the hope that it will be useful,
10cabdff1aSopenharmony_ci * but WITHOUT ANY WARRANTY; without even the implied warranty of
11cabdff1aSopenharmony_ci * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12cabdff1aSopenharmony_ci * Lesser General Public License for more details.
13cabdff1aSopenharmony_ci *
14cabdff1aSopenharmony_ci * You should have received a copy of the GNU Lesser General Public
15cabdff1aSopenharmony_ci * License along with FFmpeg; if not, write to the Free Software
16cabdff1aSopenharmony_ci * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17cabdff1aSopenharmony_ci */
18cabdff1aSopenharmony_ci
19cabdff1aSopenharmony_ci#include <stdint.h>
20cabdff1aSopenharmony_ci
21cabdff1aSopenharmony_ci#include "config.h"
22cabdff1aSopenharmony_ci#include "libavutil/attributes.h"
23cabdff1aSopenharmony_ci#include "libavutil/cpu.h"
24cabdff1aSopenharmony_ci#include "libavutil/x86/cpu.h"
25cabdff1aSopenharmony_ci#include "libavcodec/h264chroma.h"
26cabdff1aSopenharmony_ci
27cabdff1aSopenharmony_civoid ff_put_h264_chroma_mc8_rnd_mmx  (uint8_t *dst, uint8_t *src,
28cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
29cabdff1aSopenharmony_civoid ff_avg_h264_chroma_mc8_rnd_mmxext(uint8_t *dst, uint8_t *src,
30cabdff1aSopenharmony_ci                                       ptrdiff_t stride, int h, int x, int y);
31cabdff1aSopenharmony_ci
32cabdff1aSopenharmony_civoid ff_put_h264_chroma_mc4_mmx      (uint8_t *dst, uint8_t *src,
33cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
34cabdff1aSopenharmony_civoid ff_avg_h264_chroma_mc4_mmxext   (uint8_t *dst, uint8_t *src,
35cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
36cabdff1aSopenharmony_ci
37cabdff1aSopenharmony_civoid ff_put_h264_chroma_mc2_mmxext   (uint8_t *dst, uint8_t *src,
38cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
39cabdff1aSopenharmony_civoid ff_avg_h264_chroma_mc2_mmxext   (uint8_t *dst, uint8_t *src,
40cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
41cabdff1aSopenharmony_ci
42cabdff1aSopenharmony_civoid ff_put_h264_chroma_mc8_rnd_ssse3(uint8_t *dst, uint8_t *src,
43cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
44cabdff1aSopenharmony_civoid ff_put_h264_chroma_mc4_ssse3    (uint8_t *dst, uint8_t *src,
45cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
46cabdff1aSopenharmony_ci
47cabdff1aSopenharmony_civoid ff_avg_h264_chroma_mc8_rnd_ssse3(uint8_t *dst, uint8_t *src,
48cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
49cabdff1aSopenharmony_civoid ff_avg_h264_chroma_mc4_ssse3    (uint8_t *dst, uint8_t *src,
50cabdff1aSopenharmony_ci                                      ptrdiff_t stride, int h, int x, int y);
51cabdff1aSopenharmony_ci
52cabdff1aSopenharmony_ci#define CHROMA_MC(OP, NUM, DEPTH, OPT)                                  \
53cabdff1aSopenharmony_civoid ff_ ## OP ## _h264_chroma_mc ## NUM ## _ ## DEPTH ## _ ## OPT      \
54cabdff1aSopenharmony_ci                                      (uint8_t *dst, uint8_t *src,      \
55cabdff1aSopenharmony_ci                                       ptrdiff_t stride, int h, int x, int y);
56cabdff1aSopenharmony_ci
57cabdff1aSopenharmony_ciCHROMA_MC(put, 2, 10, mmxext)
58cabdff1aSopenharmony_ciCHROMA_MC(avg, 2, 10, mmxext)
59cabdff1aSopenharmony_ciCHROMA_MC(put, 4, 10, mmxext)
60cabdff1aSopenharmony_ciCHROMA_MC(avg, 4, 10, mmxext)
61cabdff1aSopenharmony_ciCHROMA_MC(put, 8, 10, sse2)
62cabdff1aSopenharmony_ciCHROMA_MC(avg, 8, 10, sse2)
63cabdff1aSopenharmony_ciCHROMA_MC(put, 8, 10, avx)
64cabdff1aSopenharmony_ciCHROMA_MC(avg, 8, 10, avx)
65cabdff1aSopenharmony_ci
66cabdff1aSopenharmony_ciav_cold void ff_h264chroma_init_x86(H264ChromaContext *c, int bit_depth)
67cabdff1aSopenharmony_ci{
68cabdff1aSopenharmony_ci    int high_bit_depth = bit_depth > 8;
69cabdff1aSopenharmony_ci    int cpu_flags      = av_get_cpu_flags();
70cabdff1aSopenharmony_ci
71cabdff1aSopenharmony_ci    if (EXTERNAL_MMX(cpu_flags) && !high_bit_depth) {
72cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_rnd_mmx;
73cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_mmx;
74cabdff1aSopenharmony_ci    }
75cabdff1aSopenharmony_ci
76cabdff1aSopenharmony_ci    if (EXTERNAL_MMXEXT(cpu_flags) && !high_bit_depth) {
77cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_rnd_mmxext;
78cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_mmxext;
79cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_mmxext;
80cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_mmxext;
81cabdff1aSopenharmony_ci    }
82cabdff1aSopenharmony_ci
83cabdff1aSopenharmony_ci    if (EXTERNAL_MMXEXT(cpu_flags) && bit_depth > 8 && bit_depth <= 10) {
84cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_10_mmxext;
85cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_10_mmxext;
86cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_10_mmxext;
87cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_10_mmxext;
88cabdff1aSopenharmony_ci    }
89cabdff1aSopenharmony_ci
90cabdff1aSopenharmony_ci    if (EXTERNAL_SSE2(cpu_flags) && bit_depth > 8 && bit_depth <= 10) {
91cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_10_sse2;
92cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_10_sse2;
93cabdff1aSopenharmony_ci    }
94cabdff1aSopenharmony_ci
95cabdff1aSopenharmony_ci    if (EXTERNAL_SSSE3(cpu_flags) && !high_bit_depth) {
96cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_rnd_ssse3;
97cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_rnd_ssse3;
98cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_ssse3;
99cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_ssse3;
100cabdff1aSopenharmony_ci    }
101cabdff1aSopenharmony_ci
102cabdff1aSopenharmony_ci    if (EXTERNAL_AVX(cpu_flags) && bit_depth > 8 && bit_depth <= 10) {
103cabdff1aSopenharmony_ci        // AVX implies !cache64.
104cabdff1aSopenharmony_ci        // TODO: Port cache(32|64) detection from x264.
105cabdff1aSopenharmony_ci        c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_10_avx;
106cabdff1aSopenharmony_ci        c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_10_avx;
107cabdff1aSopenharmony_ci    }
108cabdff1aSopenharmony_ci}
109