1 ;*****************************************************************************
2 ;* SSE2-optimized weighted prediction code
3 ;*****************************************************************************
4 ;* Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
5 ;* Copyright (C) 2010 Eli Friedman <eli.friedman@gmail.com>
7 ;* This file is part of FFmpeg.
9 ;* FFmpeg is free software; you can redistribute it and/or
10 ;* modify it under the terms of the GNU Lesser General Public
11 ;* License as published by the Free Software Foundation; either
12 ;* version 2.1 of the License, or (at your option) any later version.
14 ;* FFmpeg is distributed in the hope that it will be useful,
15 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
16 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 ;* Lesser General Public License for more details.
19 ;* You should have received a copy of the GNU Lesser General Public
20 ;* License along with FFmpeg; if not, write to the Free Software
21 ;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 ;******************************************************************************
28 ;-----------------------------------------------------------------------------
31 ; void h264_biweight_16x16_sse2(uint8_t *dst, uint8_t *src, int stride,
32 ; int log2_denom, int weightd, int weights,
35 ; void h264_weight_16x16_sse2(uint8_t *dst, int stride,
36 ; int log2_denom, int weight,
38 ;-----------------------------------------------------------------------------
74 %macro WEIGHT_FUNC_DBL_MM 1
75 cglobal h264_weight_16x%1_mmx2, 5, 5, 0
89 jmp mangle(ff_h264_weight_16x16_mmx2.nextrow)
97 %macro WEIGHT_FUNC_MM 4
98 cglobal h264_weight_%1x%2_%4, 7, 7, %3
103 WEIGHT_OP 0, mmsize/2
110 jmp mangle(ff_h264_weight_%1x16_%4.nextrow)
115 WEIGHT_FUNC_MM 8, 16, 0, mmx2
116 WEIGHT_FUNC_MM 8, 8, 0, mmx2
117 WEIGHT_FUNC_MM 8, 4, 0, mmx2
119 WEIGHT_FUNC_MM 16, 16, 8, sse2
120 WEIGHT_FUNC_MM 16, 8, 8, sse2
122 %macro WEIGHT_FUNC_HALF_MM 5
123 cglobal h264_weight_%1x%2_%5, 5, 5, %4
142 jmp mangle(ff_h264_weight_%1x%3_%5.nextrow)
147 WEIGHT_FUNC_HALF_MM 4, 8, 8, 0, mmx2
148 WEIGHT_FUNC_HALF_MM 4, 4, 8, 0, mmx2
149 WEIGHT_FUNC_HALF_MM 4, 2, 8, 0, mmx2
151 WEIGHT_FUNC_HALF_MM 8, 16, 16, 8, sse2
152 WEIGHT_FUNC_HALF_MM 8, 8, 16, 8, sse2
153 WEIGHT_FUNC_HALF_MM 8, 4, 16, 8, sse2
155 %macro BIWEIGHT_SETUP 0
180 %macro BIWEIGHT_STEPA 3
190 %macro BIWEIGHT_STEPB 0
198 %macro BIWEIGHT_FUNC_DBL_MM 1
199 cglobal h264_biweight_16x%1_mmx2, 7, 7, 0
204 BIWEIGHT_STEPA 0, 1, 0
205 BIWEIGHT_STEPA 1, 2, 4
208 BIWEIGHT_STEPA 0, 1, 8
209 BIWEIGHT_STEPA 1, 2, 12
218 jmp mangle(ff_h264_biweight_16x16_mmx2.nextrow)
223 BIWEIGHT_FUNC_DBL_MM 16
224 BIWEIGHT_FUNC_DBL_MM 8
226 %macro BIWEIGHT_FUNC_MM 4
227 cglobal h264_biweight_%1x%2_%4, 7, 7, %3
232 BIWEIGHT_STEPA 0, 1, 0
233 BIWEIGHT_STEPA 1, 2, mmsize/2
242 jmp mangle(ff_h264_biweight_%1x16_%4.nextrow)
247 BIWEIGHT_FUNC_MM 8, 16, 0, mmx2
248 BIWEIGHT_FUNC_MM 8, 8, 0, mmx2
249 BIWEIGHT_FUNC_MM 8, 4, 0, mmx2
251 BIWEIGHT_FUNC_MM 16, 16, 8, sse2
252 BIWEIGHT_FUNC_MM 16, 8, 8, sse2
254 %macro BIWEIGHT_FUNC_HALF_MM 5
255 cglobal h264_biweight_%1x%2_%5, 7, 7, %4
261 BIWEIGHT_STEPA 0, 1, 0
262 BIWEIGHT_STEPA 1, 2, r2
277 jmp mangle(ff_h264_biweight_%1x%3_%5.nextrow)
282 BIWEIGHT_FUNC_HALF_MM 4, 8, 8, 0, mmx2
283 BIWEIGHT_FUNC_HALF_MM 4, 4, 8, 0, mmx2
284 BIWEIGHT_FUNC_HALF_MM 4, 2, 8, 0, mmx2
286 BIWEIGHT_FUNC_HALF_MM 8, 16, 16, 8, sse2
287 BIWEIGHT_FUNC_HALF_MM 8, 8, 16, 8, sse2
288 BIWEIGHT_FUNC_HALF_MM 8, 4, 16, 8, sse2
290 %macro BIWEIGHT_SSSE3_SETUP 0
307 %macro BIWEIGHT_SSSE3_OP 0
317 %macro BIWEIGHT_SSSE3_16 1
318 cglobal h264_biweight_16x%1_ssse3, 7, 7, 8
337 jmp mangle(ff_h264_biweight_16x16_ssse3.nextrow)
345 %macro BIWEIGHT_SSSE3_8 1
346 cglobal h264_biweight_8x%1_ssse3, 7, 7, 8
368 jmp mangle(ff_h264_biweight_8x16_ssse3.nextrow)