2 * Copyright (c) 2010 Jason Garrett-Glaser
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 "libavcodec/h264pred.h"
24 void ff_pred16x16_vertical_mmx (uint8_t *src, int stride);
25 void ff_pred16x16_vertical_sse (uint8_t *src, int stride);
26 void ff_pred16x16_horizontal_mmx (uint8_t *src, int stride);
27 void ff_pred16x16_horizontal_mmxext(uint8_t *src, int stride);
28 void ff_pred16x16_horizontal_ssse3 (uint8_t *src, int stride);
29 void ff_pred16x16_dc_mmxext (uint8_t *src, int stride);
30 void ff_pred16x16_dc_sse2 (uint8_t *src, int stride);
31 void ff_pred16x16_dc_ssse3 (uint8_t *src, int stride);
32 void ff_pred16x16_plane_h264_mmx (uint8_t *src, int stride);
33 void ff_pred16x16_plane_h264_mmx2 (uint8_t *src, int stride);
34 void ff_pred16x16_plane_h264_sse2 (uint8_t *src, int stride);
35 void ff_pred16x16_plane_h264_ssse3 (uint8_t *src, int stride);
36 void ff_pred16x16_plane_rv40_mmx (uint8_t *src, int stride);
37 void ff_pred16x16_plane_rv40_mmx2 (uint8_t *src, int stride);
38 void ff_pred16x16_plane_rv40_sse2 (uint8_t *src, int stride);
39 void ff_pred16x16_plane_rv40_ssse3 (uint8_t *src, int stride);
40 void ff_pred16x16_plane_svq3_mmx (uint8_t *src, int stride);
41 void ff_pred16x16_plane_svq3_mmx2 (uint8_t *src, int stride);
42 void ff_pred16x16_plane_svq3_sse2 (uint8_t *src, int stride);
43 void ff_pred16x16_plane_svq3_ssse3 (uint8_t *src, int stride);
44 void ff_pred16x16_tm_vp8_mmx (uint8_t *src, int stride);
45 void ff_pred16x16_tm_vp8_mmxext (uint8_t *src, int stride);
46 void ff_pred16x16_tm_vp8_sse2 (uint8_t *src, int stride);
47 void ff_pred8x8_dc_rv40_mmxext (uint8_t *src, int stride);
48 void ff_pred8x8_vertical_mmx (uint8_t *src, int stride);
49 void ff_pred8x8_horizontal_mmx (uint8_t *src, int stride);
50 void ff_pred8x8_horizontal_mmxext (uint8_t *src, int stride);
51 void ff_pred8x8_horizontal_ssse3 (uint8_t *src, int stride);
52 void ff_pred8x8_plane_mmx (uint8_t *src, int stride);
53 void ff_pred8x8_plane_mmx2 (uint8_t *src, int stride);
54 void ff_pred8x8_plane_sse2 (uint8_t *src, int stride);
55 void ff_pred8x8_plane_ssse3 (uint8_t *src, int stride);
56 void ff_pred8x8_tm_vp8_mmx (uint8_t *src, int stride);
57 void ff_pred8x8_tm_vp8_mmxext (uint8_t *src, int stride);
58 void ff_pred8x8_tm_vp8_sse2 (uint8_t *src, int stride);
59 void ff_pred8x8_tm_vp8_ssse3 (uint8_t *src, int stride);
60 void ff_pred4x4_dc_mmxext (uint8_t *src, const uint8_t *topright, int stride);
61 void ff_pred4x4_down_left_mmxext (uint8_t *src, const uint8_t *topright, int stride);
62 void ff_pred4x4_tm_vp8_mmx (uint8_t *src, const uint8_t *topright, int stride);
63 void ff_pred4x4_tm_vp8_mmxext (uint8_t *src, const uint8_t *topright, int stride);
64 void ff_pred4x4_tm_vp8_ssse3 (uint8_t *src, const uint8_t *topright, int stride);
65 void ff_pred4x4_vertical_vp8_mmxext(uint8_t *src, const uint8_t *topright, int stride);
67 void ff_h264_pred_init_x86(H264PredContext *h, int codec_id)
69 int mm_flags = av_get_cpu_flags();
72 if (mm_flags & AV_CPU_FLAG_MMX) {
73 h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_mmx;
74 h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_mmx;
75 h->pred8x8 [VERT_PRED8x8] = ff_pred8x8_vertical_mmx;
76 h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_mmx;
77 if (codec_id == CODEC_ID_VP8) {
78 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_mmx;
79 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_mmx;
80 h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_mmx;
82 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_mmx;
83 if (codec_id == CODEC_ID_SVQ3) {
84 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_mmx;
85 } else if (codec_id == CODEC_ID_RV40) {
86 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_mmx;
88 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_mmx;
93 if (mm_flags & AV_CPU_FLAG_MMX2) {
94 h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_mmxext;
95 h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_mmxext;
96 h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_mmxext;
97 h->pred4x4 [DC_PRED ] = ff_pred4x4_dc_mmxext;
99 if (codec_id == CODEC_ID_VP8 || codec_id == CODEC_ID_H264)
100 h->pred4x4 [DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_mmxext;
102 if (codec_id == CODEC_ID_VP8) {
103 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_mmxext;
104 h->pred8x8 [DC_PRED8x8 ] = ff_pred8x8_dc_rv40_mmxext;
105 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_mmxext;
106 h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_mmxext;
107 h->pred4x4 [VERT_PRED ] = ff_pred4x4_vertical_vp8_mmxext;
109 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_mmx2;
110 if (codec_id == CODEC_ID_SVQ3) {
111 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_mmx2;
112 } else if (codec_id == CODEC_ID_RV40) {
113 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_mmx2;
115 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_mmx2;
120 if (mm_flags & AV_CPU_FLAG_SSE) {
121 h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_sse;
124 if (mm_flags & AV_CPU_FLAG_SSE2) {
125 h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_sse2;
126 if (codec_id == CODEC_ID_VP8) {
127 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_sse2;
128 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_sse2;
130 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_sse2;
131 if (codec_id == CODEC_ID_SVQ3) {
132 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_sse2;
133 } else if (codec_id == CODEC_ID_RV40) {
134 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_sse2;
136 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_sse2;
141 if (mm_flags & AV_CPU_FLAG_SSSE3) {
142 h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_ssse3;
143 h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_ssse3;
144 h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_ssse3;
145 if (codec_id == CODEC_ID_VP8) {
146 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_ssse3;
147 h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_ssse3;
149 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_ssse3;
150 if (codec_id == CODEC_ID_SVQ3) {
151 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_ssse3;
152 } else if (codec_id == CODEC_ID_RV40) {
153 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_ssse3;
155 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_ssse3;