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_top_dc_mmxext (uint8_t *src, int stride);
48 void ff_pred8x8_dc_rv40_mmxext (uint8_t *src, int stride);
49 void ff_pred8x8_dc_mmxext (uint8_t *src, int stride);
50 void ff_pred8x8_vertical_mmx (uint8_t *src, int stride);
51 void ff_pred8x8_horizontal_mmx (uint8_t *src, int stride);
52 void ff_pred8x8_horizontal_mmxext (uint8_t *src, int stride);
53 void ff_pred8x8_horizontal_ssse3 (uint8_t *src, int stride);
54 void ff_pred8x8_plane_mmx (uint8_t *src, int stride);
55 void ff_pred8x8_plane_mmx2 (uint8_t *src, int stride);
56 void ff_pred8x8_plane_sse2 (uint8_t *src, int stride);
57 void ff_pred8x8_plane_ssse3 (uint8_t *src, int stride);
58 void ff_pred8x8_tm_vp8_mmx (uint8_t *src, int stride);
59 void ff_pred8x8_tm_vp8_mmxext (uint8_t *src, int stride);
60 void ff_pred8x8_tm_vp8_sse2 (uint8_t *src, int stride);
61 void ff_pred8x8_tm_vp8_ssse3 (uint8_t *src, int stride);
62 void ff_pred8x8l_top_dc_mmxext (uint8_t *src, int has_topleft, int has_topright, int stride);
63 void ff_pred8x8l_top_dc_ssse3 (uint8_t *src, int has_topleft, int has_topright, int stride);
64 void ff_pred8x8l_dc_mmxext (uint8_t *src, int has_topleft, int has_topright, int stride);
65 void ff_pred8x8l_dc_ssse3 (uint8_t *src, int has_topleft, int has_topright, int stride);
66 void ff_pred8x8l_horizontal_mmxext (uint8_t *src, int has_topleft, int has_topright, int stride);
67 void ff_pred8x8l_horizontal_ssse3 (uint8_t *src, int has_topleft, int has_topright, int stride);
68 void ff_pred8x8l_vertical_mmxext (uint8_t *src, int has_topleft, int has_topright, int stride);
69 void ff_pred8x8l_vertical_ssse3 (uint8_t *src, int has_topleft, int has_topright, int stride);
70 void ff_pred8x8l_down_left_sse2 (uint8_t *src, int has_topleft, int has_topright, int stride);
71 void ff_pred8x8l_down_left_ssse3 (uint8_t *src, int has_topleft, int has_topright, int stride);
72 void ff_pred8x8l_down_right_mmxext (uint8_t *src, int has_topleft, int has_topright, int stride);
73 void ff_pred8x8l_down_right_sse2 (uint8_t *src, int has_topleft, int has_topright, int stride);
74 void ff_pred8x8l_down_right_ssse3 (uint8_t *src, int has_topleft, int has_topright, int stride);
75 void ff_pred8x8l_vertical_right_mmxext(uint8_t *src, int has_topleft, int has_topright, int stride);
76 void ff_pred8x8l_vertical_right_sse2(uint8_t *src, int has_topleft, int has_topright, int stride);
77 void ff_pred8x8l_vertical_right_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
78 void ff_pred8x8l_vertical_left_sse2(uint8_t *src, int has_topleft, int has_topright, int stride);
79 void ff_pred8x8l_vertical_left_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
80 void ff_pred8x8l_horizontal_up_mmxext(uint8_t *src, int has_topleft, int has_topright, int stride);
81 void ff_pred8x8l_horizontal_up_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
82 void ff_pred8x8l_horizontal_down_mmxext(uint8_t *src, int has_topleft, int has_topright, int stride);
83 void ff_pred8x8l_horizontal_down_sse2(uint8_t *src, int has_topleft, int has_topright, int stride);
84 void ff_pred8x8l_horizontal_down_ssse3(uint8_t *src, int has_topleft, int has_topright, int stride);
85 void ff_pred4x4_dc_mmxext (uint8_t *src, const uint8_t *topright, int stride);
86 void ff_pred4x4_down_left_mmxext (uint8_t *src, const uint8_t *topright, int stride);
87 void ff_pred4x4_vertical_left_mmxext(uint8_t *src, const uint8_t *topright, int stride);
88 void ff_pred4x4_tm_vp8_mmx (uint8_t *src, const uint8_t *topright, int stride);
89 void ff_pred4x4_tm_vp8_mmxext (uint8_t *src, const uint8_t *topright, int stride);
90 void ff_pred4x4_tm_vp8_ssse3 (uint8_t *src, const uint8_t *topright, int stride);
91 void ff_pred4x4_vertical_vp8_mmxext(uint8_t *src, const uint8_t *topright, int stride);
93 void ff_h264_pred_init_x86(H264PredContext *h, int codec_id)
95 int mm_flags = av_get_cpu_flags();
98 if (mm_flags & AV_CPU_FLAG_MMX) {
99 h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_mmx;
100 h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_mmx;
101 h->pred8x8 [VERT_PRED8x8] = ff_pred8x8_vertical_mmx;
102 h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_mmx;
103 if (codec_id == CODEC_ID_VP8) {
104 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_mmx;
105 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_mmx;
106 h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_mmx;
108 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_mmx;
109 if (codec_id == CODEC_ID_SVQ3) {
110 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_mmx;
111 } else if (codec_id == CODEC_ID_RV40) {
112 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_mmx;
114 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_mmx;
119 if (mm_flags & AV_CPU_FLAG_MMX2) {
120 h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_mmxext;
121 h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_mmxext;
122 h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_mmxext;
124 h->pred8x8l [TOP_DC_PRED ] = ff_pred8x8l_top_dc_mmxext;
125 h->pred8x8l [DC_PRED ] = ff_pred8x8l_dc_mmxext;
126 h->pred8x8l [HOR_PRED ] = ff_pred8x8l_horizontal_mmxext;
127 h->pred8x8l [VERT_PRED ] = ff_pred8x8l_vertical_mmxext;
128 h->pred8x8l [DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_mmxext;
129 h->pred8x8l [VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_mmxext;
130 h->pred8x8l [HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_mmxext;
131 h->pred8x8l [HOR_DOWN_PRED ] = ff_pred8x8l_horizontal_down_mmxext;
133 h->pred4x4 [DC_PRED ] = ff_pred4x4_dc_mmxext;
135 if (codec_id == CODEC_ID_VP8 || codec_id == CODEC_ID_H264)
136 h->pred4x4 [DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_mmxext;
137 if (codec_id == CODEC_ID_SVQ3 || codec_id == CODEC_ID_H264)
138 h->pred4x4 [VERT_LEFT_PRED ] = ff_pred4x4_vertical_left_mmxext;
139 if (codec_id == CODEC_ID_SVQ3 || codec_id == CODEC_ID_H264) {
140 h->pred8x8 [TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_mmxext;
141 h->pred8x8 [DC_PRED8x8 ] = ff_pred8x8_dc_mmxext;
144 if (codec_id == CODEC_ID_VP8) {
145 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_mmxext;
146 h->pred8x8 [DC_PRED8x8 ] = ff_pred8x8_dc_rv40_mmxext;
147 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_mmxext;
148 h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_mmxext;
149 h->pred4x4 [VERT_PRED ] = ff_pred4x4_vertical_vp8_mmxext;
151 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_mmx2;
152 if (codec_id == CODEC_ID_SVQ3) {
153 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_mmx2;
154 } else if (codec_id == CODEC_ID_RV40) {
155 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_mmx2;
157 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_mmx2;
162 if (mm_flags & AV_CPU_FLAG_SSE) {
163 h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_sse;
166 if (mm_flags & AV_CPU_FLAG_SSE2) {
167 h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_sse2;
169 h->pred8x8l [DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_sse2;
170 h->pred8x8l [DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_sse2;
171 h->pred8x8l [VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_sse2;
172 h->pred8x8l [VERT_LEFT_PRED ] = ff_pred8x8l_vertical_left_sse2;
173 h->pred8x8l [HOR_DOWN_PRED ] = ff_pred8x8l_horizontal_down_sse2;
175 if (codec_id == CODEC_ID_VP8) {
176 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_sse2;
177 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_sse2;
179 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_sse2;
180 if (codec_id == CODEC_ID_SVQ3) {
181 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_sse2;
182 } else if (codec_id == CODEC_ID_RV40) {
183 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_sse2;
185 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_sse2;
190 if (mm_flags & AV_CPU_FLAG_SSSE3) {
191 h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_ssse3;
192 h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_ssse3;
193 h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_ssse3;
195 h->pred8x8l [TOP_DC_PRED ] = ff_pred8x8l_top_dc_ssse3;
196 h->pred8x8l [DC_PRED ] = ff_pred8x8l_dc_ssse3;
197 h->pred8x8l [HOR_PRED ] = ff_pred8x8l_horizontal_ssse3;
198 h->pred8x8l [VERT_PRED ] = ff_pred8x8l_vertical_ssse3;
199 h->pred8x8l [DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_ssse3;
200 h->pred8x8l [DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_ssse3;
201 h->pred8x8l [VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_ssse3;
202 h->pred8x8l [VERT_LEFT_PRED ] = ff_pred8x8l_vertical_left_ssse3;
203 h->pred8x8l [HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_ssse3;
204 h->pred8x8l [HOR_DOWN_PRED ] = ff_pred8x8l_horizontal_down_ssse3;
206 if (codec_id == CODEC_ID_VP8) {
207 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_ssse3;
208 h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_ssse3;
210 h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_ssse3;
211 if (codec_id == CODEC_ID_SVQ3) {
212 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_ssse3;
213 } else if (codec_id == CODEC_ID_RV40) {
214 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_ssse3;
216 h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_ssse3;