1 ;*****************************************************************************
2 ;* dct-64.asm: h264 encoder library
3 ;*****************************************************************************
4 ;* Copyright (C) 2003-2008 x264 project
6 ;* Authors: Laurent Aimar <fenrir@via.ecp.fr> (initial version)
7 ;* Loren Merritt <lorenm@u.washington.edu> (dct8, misc)
8 ;* Min Chen <chenm001.163.com> (converted to nasm)
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.
23 ;*****************************************************************************
26 %include "x86util.asm"
35 SUMSUB_BA m%8, m%1 ; %8=s07, %1=d07
36 SUMSUB_BA m%7, m%2 ; %7=s16, %2=d16
37 SUMSUB_BA m%6, m%3 ; %6=s25, %3=d25
38 SUMSUB_BA m%5, m%4 ; %5=s34, %4=d34
40 SUMSUB_BA m%5, m%8 ; %5=a0, %8=a2
41 SUMSUB_BA m%6, m%7 ; %6=a1, %7=a3
47 paddw m%9, m%3 ; %9=a4
53 psubw m%10, m%3 ; %10=a7
60 psubw m%1, m%3 ; %1=a5
61 psubw m%4, m%2 ; %4=a6
63 SUMSUB_BA m%6, m%5 ; %6=b0, %5=b4
67 paddw m%2, m%9 ; %2=b1
69 psubw m%9, m%10 ; %9=b7
73 paddw m%3, m%8 ; %3=b2
75 psubw m%8, m%7 ; %8=b6
79 paddw m%7, m%1 ; %7=b3
81 psubw m%4, m%1 ; %4=b5
83 SWAP %1, %6, %4, %7, %8, %9
86 ;-----------------------------------------------------------------------------
87 ; void x264_sub8x8_dct8_sse2( int16_t dct[8][8], uint8_t *pix1, uint8_t *pix2 )
88 ;-----------------------------------------------------------------------------
89 cglobal x264_sub8x8_dct8_sse2
90 LOAD_DIFF m0, m8, m9, [r1+0*FENC_STRIDE], [r2+0*FDEC_STRIDE]
91 LOAD_DIFF m1, m8, m9, [r1+1*FENC_STRIDE], [r2+1*FDEC_STRIDE]
92 LOAD_DIFF m2, m8, m9, [r1+2*FENC_STRIDE], [r2+2*FDEC_STRIDE]
93 LOAD_DIFF m3, m8, m9, [r1+3*FENC_STRIDE], [r2+3*FDEC_STRIDE]
94 LOAD_DIFF m4, m8, m9, [r1+4*FENC_STRIDE], [r2+4*FDEC_STRIDE]
95 LOAD_DIFF m5, m8, m9, [r1+5*FENC_STRIDE], [r2+5*FDEC_STRIDE]
96 LOAD_DIFF m6, m8, m9, [r1+6*FENC_STRIDE], [r2+6*FDEC_STRIDE]
97 LOAD_DIFF m7, m8, m9, [r1+7*FENC_STRIDE], [r2+7*FDEC_STRIDE]
99 DCT8_1D 0,1,2,3,4,5,6,7,8,9
100 TRANSPOSE8x8W 0,1,2,3,4,5,6,7,8
101 DCT8_1D 0,1,2,3,4,5,6,7,8,9
115 SUMSUB_BA m%5, m%1 ; %5=a0, %1=a2
118 psubw m%3, m%7 ; %3=a4
120 paddw m%7, m%10 ; %7=a6
126 paddw m%9, m%6 ; %9=a7
132 psubw m%10, m%2 ; %10=a5
140 psubw m%2, m%4 ; %2=a3
141 psubw m%6, m%8 ; %6=a1
143 SUMSUB_BA m%7, m%5 ; %7=b0, %5=b6
144 SUMSUB_BA m%3, m%1 ; %3=b2, %1=b4
148 paddw m%4, m%6 ; %4=b1
150 psubw m%9, m%6 ; %9=b7
154 paddw m%8, m%2 ; %8=b3
156 psubw m%2, m%10 ; %2=b5
158 SUMSUB_BA m%9, m%7 ; %9=c0, %7=c7
159 SUMSUB_BA m%2, m%3 ; %2=c1, %3=c6
160 SUMSUB_BA m%8, m%1 ; %8=c2, %1=c5
161 SUMSUB_BA m%4, m%5 ; %4=c3, %5=c4
167 ;-----------------------------------------------------------------------------
168 ; void x264_add8x8_idct8_sse2( uint8_t *p_dst, int16_t dct[8][8] )
169 ;-----------------------------------------------------------------------------
170 cglobal x264_add8x8_idct8_sse2
180 IDCT8_1D 0,1,2,3,4,5,6,7,8,9
181 TRANSPOSE8x8W 0,1,2,3,4,5,6,7,8
182 paddw m0, [pw_32 GLOBAL] ; rounding for the >>6 at the end
183 IDCT8_1D 0,1,2,3,4,5,6,7,8,9
186 STORE_DIFF m0, m8, m9, [r0+0*FDEC_STRIDE]
187 STORE_DIFF m1, m8, m9, [r0+1*FDEC_STRIDE]
188 STORE_DIFF m2, m8, m9, [r0+2*FDEC_STRIDE]
189 STORE_DIFF m3, m8, m9, [r0+3*FDEC_STRIDE]
190 STORE_DIFF m4, m8, m9, [r0+4*FDEC_STRIDE]
191 STORE_DIFF m5, m8, m9, [r0+5*FDEC_STRIDE]
192 STORE_DIFF m6, m8, m9, [r0+6*FDEC_STRIDE]
193 STORE_DIFF m7, m8, m9, [r0+7*FDEC_STRIDE]