1 /*****************************************************************************
2 * dct.h: x86 transform and zigzag
3 *****************************************************************************
4 * Copyright (C) 2003-2015 x264 project
6 * Authors: Loren Merritt <lorenm@u.washington.edu>
7 * Laurent Aimar <fenrir@via.ecp.fr>
8 * Fiona Glaser <fiona@x264.com>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
24 * This program is also available under a commercial proprietary license.
25 * For more information, contact us at licensing@x264.com.
26 *****************************************************************************/
28 #ifndef X264_I386_DCT_H
29 #define X264_I386_DCT_H
31 void x264_sub4x4_dct_mmx ( dctcoef dct [16], pixel *pix1, pixel *pix2 );
32 void x264_sub8x8_dct_mmx ( dctcoef dct[ 4][16], pixel *pix1, pixel *pix2 );
33 void x264_sub16x16_dct_mmx ( dctcoef dct[16][16], pixel *pix1, pixel *pix2 );
34 void x264_sub8x8_dct_sse2 ( int16_t dct[ 4][16], uint8_t *pix1, uint8_t *pix2 );
35 void x264_sub16x16_dct_sse2 ( int16_t dct[16][16], uint8_t *pix1, uint8_t *pix2 );
36 void x264_sub4x4_dct_ssse3 ( int16_t dct [16], uint8_t *pix1, uint8_t *pix2 );
37 void x264_sub8x8_dct_ssse3 ( int16_t dct[ 4][16], uint8_t *pix1, uint8_t *pix2 );
38 void x264_sub16x16_dct_ssse3( int16_t dct[16][16], uint8_t *pix1, uint8_t *pix2 );
39 void x264_sub8x8_dct_avx ( int16_t dct[ 4][16], uint8_t *pix1, uint8_t *pix2 );
40 void x264_sub16x16_dct_avx ( int16_t dct[16][16], uint8_t *pix1, uint8_t *pix2 );
41 void x264_sub8x8_dct_xop ( int16_t dct[ 4][16], uint8_t *pix1, uint8_t *pix2 );
42 void x264_sub16x16_dct_xop ( int16_t dct[16][16], uint8_t *pix1, uint8_t *pix2 );
43 void x264_sub8x8_dct_avx2 ( int16_t dct[ 4][16], uint8_t *pix1, uint8_t *pix2 );
44 void x264_sub16x16_dct_avx2 ( int16_t dct[16][16], uint8_t *pix1, uint8_t *pix2 );
45 void x264_sub8x8_dct_dc_mmx2( int16_t dct [ 4], uint8_t *pix1, uint8_t *pix2 );
46 void x264_sub8x8_dct_dc_sse2( dctcoef dct [ 4], pixel *pix1, pixel *pix2 );
47 void x264_sub8x16_dct_dc_sse2 ( dctcoef dct [ 4], pixel *pix1, pixel *pix2 );
48 void x264_sub8x16_dct_dc_ssse3( int16_t dct [ 4], uint8_t *pix1, uint8_t *pix2 );
49 void x264_sub8x16_dct_dc_avx ( dctcoef dct [ 4], pixel *pix1, pixel *pix2 );
51 void x264_add4x4_idct_mmx ( uint8_t *p_dst, int16_t dct [16] );
52 void x264_add4x4_idct_sse2 ( uint16_t *p_dst, int32_t dct [16] );
53 void x264_add4x4_idct_sse4 ( uint8_t *p_dst, int16_t dct [16] );
54 void x264_add4x4_idct_avx ( pixel *p_dst, dctcoef dct [16] );
55 void x264_add8x8_idct_mmx ( uint8_t *p_dst, int16_t dct[ 4][16] );
56 void x264_add8x8_idct_dc_mmx2 ( uint8_t *p_dst, int16_t dct [ 4] );
57 void x264_add16x16_idct_mmx ( uint8_t *p_dst, int16_t dct[16][16] );
58 void x264_add16x16_idct_dc_mmx2 ( uint8_t *p_dst, int16_t dct [16] );
59 void x264_add8x8_idct_sse2 ( pixel *p_dst, dctcoef dct[ 4][16] );
60 void x264_add8x8_idct_avx ( pixel *p_dst, dctcoef dct[ 4][16] );
61 void x264_add8x8_idct_avx2 ( pixel *p_dst, dctcoef dct[ 4][16] );
62 void x264_add16x16_idct_sse2 ( pixel *p_dst, dctcoef dct[16][16] );
63 void x264_add16x16_idct_avx ( pixel *p_dst, dctcoef dct[16][16] );
64 void x264_add16x16_idct_avx2 ( pixel *p_dst, dctcoef dct[16][16] );
65 void x264_add8x8_idct_dc_sse2 ( pixel *p_dst, dctcoef dct [ 4] );
66 void x264_add16x16_idct_dc_sse2 ( pixel *p_dst, dctcoef dct [16] );
67 void x264_add8x8_idct_dc_ssse3 ( uint8_t *p_dst, int16_t dct [ 4] );
68 void x264_add16x16_idct_dc_ssse3( uint8_t *p_dst, int16_t dct [16] );
69 void x264_add8x8_idct_dc_avx ( pixel *p_dst, dctcoef dct [ 4] );
70 void x264_add16x16_idct_dc_avx ( pixel *p_dst, dctcoef dct [16] );
71 void x264_add16x16_idct_dc_avx2 ( uint8_t *p_dst, int16_t dct [16] );
73 void x264_dct4x4dc_mmx2 ( int16_t d[16] );
74 void x264_dct4x4dc_sse2 ( int32_t d[16] );
75 void x264_dct4x4dc_avx ( int32_t d[16] );
76 void x264_idct4x4dc_mmx ( int16_t d[16] );
77 void x264_idct4x4dc_sse2 ( int32_t d[16] );
78 void x264_idct4x4dc_avx ( int32_t d[16] );
80 void x264_sub8x8_dct8_mmx ( int16_t dct [64], uint8_t *pix1, uint8_t *pix2 );
81 void x264_sub16x16_dct8_mmx ( int16_t dct[4][64], uint8_t *pix1, uint8_t *pix2 );
82 void x264_sub8x8_dct8_sse2 ( dctcoef dct [64], pixel *pix1, pixel *pix2 );
83 void x264_sub16x16_dct8_sse2 ( dctcoef dct[4][64], pixel *pix1, pixel *pix2 );
84 void x264_sub8x8_dct8_ssse3 ( int16_t dct [64], uint8_t *pix1, uint8_t *pix2 );
85 void x264_sub16x16_dct8_ssse3( int16_t dct[4][64], uint8_t *pix1, uint8_t *pix2 );
86 void x264_sub8x8_dct8_sse4 ( int32_t dct [64], uint16_t *pix1, uint16_t *pix2 );
87 void x264_sub16x16_dct8_sse4 ( int32_t dct[4][64], uint16_t *pix1, uint16_t *pix2 );
88 void x264_sub8x8_dct8_avx ( dctcoef dct [64], pixel *pix1, pixel *pix2 );
89 void x264_sub16x16_dct8_avx ( dctcoef dct[4][64], pixel *pix1, pixel *pix2 );
90 void x264_sub16x16_dct8_avx2 ( dctcoef dct[4][64], pixel *pix1, pixel *pix2 );
93 void x264_add8x8_idct8_mmx ( uint8_t *dst, int16_t dct [64] );
94 void x264_add16x16_idct8_mmx ( uint8_t *dst, int16_t dct[4][64] );
95 void x264_add8x8_idct8_sse2 ( pixel *dst, dctcoef dct [64] );
96 void x264_add16x16_idct8_sse2( pixel *dst, dctcoef dct[4][64] );
97 void x264_add8x8_idct8_avx ( pixel *dst, dctcoef dct [64] );
98 void x264_add16x16_idct8_avx ( pixel *dst, dctcoef dct[4][64] );
100 void x264_zigzag_scan_8x8_frame_xop ( int16_t level[64], int16_t dct[64] );
101 void x264_zigzag_scan_8x8_frame_avx ( dctcoef level[64], dctcoef dct[64] );
102 void x264_zigzag_scan_8x8_frame_ssse3( int16_t level[64], int16_t dct[64] );
103 void x264_zigzag_scan_8x8_frame_sse2 ( dctcoef level[64], dctcoef dct[64] );
104 void x264_zigzag_scan_8x8_frame_mmx2 ( int16_t level[64], int16_t dct[64] );
105 void x264_zigzag_scan_4x4_frame_xop ( dctcoef level[16], dctcoef dct[16] );
106 void x264_zigzag_scan_4x4_frame_avx ( dctcoef level[16], dctcoef dct[16] );
107 void x264_zigzag_scan_4x4_frame_ssse3( int16_t level[16], int16_t dct[16] );
108 void x264_zigzag_scan_4x4_frame_sse2 ( int32_t level[16], int32_t dct[16] );
109 void x264_zigzag_scan_4x4_frame_mmx ( int16_t level[16], int16_t dct[16] );
110 void x264_zigzag_scan_4x4_field_sse2 ( int32_t level[16], int32_t dct[16] );
111 void x264_zigzag_scan_4x4_field_mmx2 ( int16_t level[16], int16_t dct[16] );
112 void x264_zigzag_scan_8x8_field_xop ( int16_t level[64], int16_t dct[64] );
113 void x264_zigzag_scan_8x8_field_avx ( int32_t level[64], int32_t dct[64] );
114 void x264_zigzag_scan_8x8_field_sse4 ( int32_t level[64], int32_t dct[64] );
115 void x264_zigzag_scan_8x8_field_mmx2 ( int16_t level[64], int16_t dct[64] );
116 int x264_zigzag_sub_4x4_frame_avx ( int16_t level[16], const uint8_t *src, uint8_t *dst );
117 int x264_zigzag_sub_4x4_frame_ssse3 ( int16_t level[16], const uint8_t *src, uint8_t *dst );
118 int x264_zigzag_sub_4x4ac_frame_avx ( int16_t level[16], const uint8_t *src, uint8_t *dst, int16_t *dc );
119 int x264_zigzag_sub_4x4ac_frame_ssse3( int16_t level[16], const uint8_t *src, uint8_t *dst, int16_t *dc );
120 int x264_zigzag_sub_4x4_field_avx ( int16_t level[16], const uint8_t *src, uint8_t *dst );
121 int x264_zigzag_sub_4x4_field_ssse3 ( int16_t level[16], const uint8_t *src, uint8_t *dst );
122 int x264_zigzag_sub_4x4ac_field_avx ( int16_t level[16], const uint8_t *src, uint8_t *dst, int16_t *dc );
123 int x264_zigzag_sub_4x4ac_field_ssse3( int16_t level[16], const uint8_t *src, uint8_t *dst, int16_t *dc );
124 void x264_zigzag_interleave_8x8_cavlc_mmx ( int16_t *dst, int16_t *src, uint8_t *nnz );
125 void x264_zigzag_interleave_8x8_cavlc_sse2( dctcoef *dst, dctcoef *src, uint8_t *nnz );
126 void x264_zigzag_interleave_8x8_cavlc_avx ( dctcoef *dst, dctcoef *src, uint8_t *nnz );
127 void x264_zigzag_interleave_8x8_cavlc_avx2( int16_t *dst, int16_t *src, uint8_t *nnz );