1 /*****************************************************************************
2 * mc-c.c: h264 encoder library (Motion Compensation)
3 *****************************************************************************
4 * Copyright (C) 2003-2008 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
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.
23 *****************************************************************************/
29 #include "common/common.h"
33 extern void x264_pixel_avg_16x16_sse2( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
34 extern void x264_pixel_avg_16x8_sse2( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
35 extern void x264_pixel_avg_8x16_sse2( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
36 extern void x264_pixel_avg_8x8_sse2( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
37 extern void x264_pixel_avg_8x4_sse2( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
38 extern void x264_pixel_avg_16x16_mmxext( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
39 extern void x264_pixel_avg_16x8_mmxext( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
40 extern void x264_pixel_avg_8x16_mmxext( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
41 extern void x264_pixel_avg_8x8_mmxext( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
42 extern void x264_pixel_avg_8x4_mmxext( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
43 extern void x264_pixel_avg_4x8_mmxext( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
44 extern void x264_pixel_avg_4x4_mmxext( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
45 extern void x264_pixel_avg_4x2_mmxext( uint8_t *, int, uint8_t *, int, uint8_t *, int, int );
46 extern void x264_mc_copy_w4_mmx( uint8_t *, int, uint8_t *, int, int );
47 extern void x264_mc_copy_w8_mmx( uint8_t *, int, uint8_t *, int, int );
48 extern void x264_mc_copy_w16_mmx( uint8_t *, int, uint8_t *, int, int );
49 extern void x264_mc_copy_w16_sse2( uint8_t *, int, uint8_t *, int, int );
50 extern void x264_mc_copy_w16_sse3( uint8_t *, int, uint8_t *, int, int );
51 extern void x264_mc_copy_w16_aligned_sse2( uint8_t *, int, uint8_t *, int, int );
52 extern void x264_prefetch_fenc_mmxext( uint8_t *, int, uint8_t *, int, int );
53 extern void x264_prefetch_ref_mmxext( uint8_t *, int, int );
54 extern void x264_mc_chroma_mmxext( uint8_t *src, int i_src_stride,
55 uint8_t *dst, int i_dst_stride,
56 int dx, int dy, int i_width, int i_height );
57 extern void x264_mc_chroma_sse2( uint8_t *src, int i_src_stride,
58 uint8_t *dst, int i_dst_stride,
59 int dx, int dy, int i_width, int i_height );
60 extern void x264_mc_chroma_ssse3( uint8_t *src, int i_src_stride,
61 uint8_t *dst, int i_dst_stride,
62 int dx, int dy, int i_width, int i_height );
63 extern void x264_plane_copy_mmxext( uint8_t *, int, uint8_t *, int, int w, int h);
64 extern void *x264_memcpy_aligned_mmx( void * dst, const void * src, size_t n );
65 extern void *x264_memcpy_aligned_sse2( void * dst, const void * src, size_t n );
66 extern void x264_memzero_aligned_mmx( void * dst, int n );
67 extern void x264_memzero_aligned_sse2( void * dst, int n );
69 extern void x264_frame_init_lowres_core_##cpu( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,\
70 int src_stride, int dst_stride, int width, int height );
72 LOWRES(cache32_mmxext)
76 #define PIXEL_AVG_W(width,cpu)\
77 extern void x264_pixel_avg2_w##width##_##cpu( uint8_t *, int, uint8_t *, int, uint8_t *, int );
78 /* This declares some functions that don't exist, but that isn't a problem. */
79 #define PIXEL_AVG_WALL(cpu)\
80 PIXEL_AVG_W(4,cpu); PIXEL_AVG_W(8,cpu); PIXEL_AVG_W(12,cpu); PIXEL_AVG_W(16,cpu); PIXEL_AVG_W(20,cpu);
82 PIXEL_AVG_WALL(mmxext)
83 PIXEL_AVG_WALL(cache32_mmxext)
84 PIXEL_AVG_WALL(cache64_mmxext)
85 PIXEL_AVG_WALL(cache64_sse2)
88 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
89 static void (* const x264_pixel_avg_wtab_##instr[6])( uint8_t *, int, uint8_t *, int, uint8_t *, int ) =\
92 x264_pixel_avg2_w4_##name1,\
93 x264_pixel_avg2_w8_##name2,\
94 x264_pixel_avg2_w12_##name3,\
95 x264_pixel_avg2_w16_##name4,\
96 x264_pixel_avg2_w20_##name5,\
99 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
100 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
101 #define x264_pixel_avg2_w12_sse3 x264_pixel_avg2_w16_sse3
103 PIXEL_AVG_WTAB(mmxext, mmxext, mmxext, mmxext, mmxext, mmxext)
105 PIXEL_AVG_WTAB(cache32_mmxext, mmxext, cache32_mmxext, cache32_mmxext, cache32_mmxext, cache32_mmxext)
107 PIXEL_AVG_WTAB(cache64_mmxext, mmxext, cache64_mmxext, cache64_mmxext, cache64_mmxext, cache64_mmxext)
108 PIXEL_AVG_WTAB(sse2, mmxext, mmxext, mmxext, sse2, sse2)
109 PIXEL_AVG_WTAB(cache64_sse2, mmxext, cache64_mmxext, cache64_sse2, cache64_sse2, cache64_sse2)
111 #define MC_COPY_WTAB(instr, name1, name2, name3)\
112 static void (* const x264_mc_copy_wtab_##instr[5])( uint8_t *, int, uint8_t *, int, int ) =\
115 x264_mc_copy_w4_##name1,\
116 x264_mc_copy_w8_##name2,\
118 x264_mc_copy_w16_##name3,\
121 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
122 MC_COPY_WTAB(sse2,mmx,mmx,sse2)
124 static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
125 static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
127 #define MC_LUMA(name,instr1,instr2)\
128 static void mc_luma_##name( uint8_t *dst, int i_dst_stride,\
129 uint8_t *src[4], int i_src_stride,\
131 int i_width, int i_height )\
133 int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
134 int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
135 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
136 if( qpel_idx & 5 ) /* qpel interpolation needed */\
138 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
139 x264_pixel_avg_wtab_##instr1[i_width>>2](\
140 dst, i_dst_stride, src1, i_src_stride,\
145 x264_mc_copy_wtab_##instr2[i_width>>2](\
146 dst, i_dst_stride, src1, i_src_stride, i_height );\
150 MC_LUMA(mmxext,mmxext,mmx)
152 MC_LUMA(cache32_mmxext,cache32_mmxext,mmx)
153 MC_LUMA(cache64_mmxext,cache64_mmxext,mmx)
155 MC_LUMA(sse2,sse2,sse2)
156 MC_LUMA(cache64_sse2,cache64_sse2,sse2)
158 #define GET_REF(name)\
159 static uint8_t *get_ref_##name( uint8_t *dst, int *i_dst_stride,\
160 uint8_t *src[4], int i_src_stride,\
162 int i_width, int i_height )\
164 int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
165 int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
166 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
167 if( qpel_idx & 5 ) /* qpel interpolation needed */\
169 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
170 x264_pixel_avg_wtab_##name[i_width>>2](\
171 dst, *i_dst_stride, src1, i_src_stride,\
177 *i_dst_stride = i_src_stride;\
184 GET_REF(cache32_mmxext)
185 GET_REF(cache64_mmxext)
188 GET_REF(cache64_sse2)
190 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
191 void x264_hpel_filter_v_##cpuv( uint8_t *dst, uint8_t *src, int16_t *buf, int stride, int width);\
192 void x264_hpel_filter_c_##cpuc( uint8_t *dst, int16_t *buf, int width );\
193 void x264_hpel_filter_h_##cpuh( uint8_t *dst, uint8_t *src, int width );\
194 void x264_sfence( void );\
195 static void x264_hpel_filter_##cpu( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,\
196 int stride, int width, int height )\
199 int realign = (long)src & (align-1);\
205 buf = x264_malloc((width+16)*sizeof(int16_t));\
208 x264_hpel_filter_v_##cpuv( dstv, src, buf+8, stride, width );\
209 x264_hpel_filter_c_##cpuc( dstc, buf+8, width );\
210 x264_hpel_filter_h_##cpuh( dsth, src, width );\
220 HPEL(8, mmxext, mmxext, mmxext, mmxext)
221 HPEL(16, sse2_amd, mmxext, mmxext, sse2)
223 void x264_hpel_filter_sse2( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height );
224 void x264_hpel_filter_ssse3( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height );
226 HPEL(16, sse2, sse2, sse2, sse2)
227 HPEL(16, ssse3, sse2, ssse3, ssse3)
230 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
232 if( !(cpu&X264_CPU_MMX) )
235 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
236 pf->copy[PIXEL_8x8] = x264_mc_copy_w8_mmx;
237 pf->copy[PIXEL_4x4] = x264_mc_copy_w4_mmx;
238 pf->memcpy_aligned = x264_memcpy_aligned_mmx;
239 pf->memzero_aligned = x264_memzero_aligned_mmx;
241 if( !(cpu&X264_CPU_MMXEXT) )
244 pf->mc_luma = mc_luma_mmxext;
245 pf->get_ref = get_ref_mmxext;
246 pf->mc_chroma = x264_mc_chroma_mmxext;
248 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmxext;
249 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_mmxext;
250 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_mmxext;
251 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_mmxext;
252 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_mmxext;
253 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_mmxext;
254 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_mmxext;
255 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_mmxext;
257 pf->plane_copy = x264_plane_copy_mmxext;
258 pf->hpel_filter = x264_hpel_filter_mmxext;
259 pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmxext;
261 pf->prefetch_fenc = x264_prefetch_fenc_mmxext;
262 pf->prefetch_ref = x264_prefetch_ref_mmxext;
264 #ifdef ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
265 if( cpu&X264_CPU_CACHELINE_32 )
267 pf->mc_luma = mc_luma_cache32_mmxext;
268 pf->get_ref = get_ref_cache32_mmxext;
269 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
271 else if( cpu&X264_CPU_CACHELINE_64 )
273 pf->mc_luma = mc_luma_cache64_mmxext;
274 pf->get_ref = get_ref_cache64_mmxext;
275 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
279 if( !(cpu&X264_CPU_SSE2) )
282 pf->memcpy_aligned = x264_memcpy_aligned_sse2;
283 pf->memzero_aligned = x264_memzero_aligned_sse2;
284 pf->hpel_filter = x264_hpel_filter_sse2_amd;
286 if( cpu&X264_CPU_SSE2_IS_SLOW )
289 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse2;
290 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
291 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_sse2;
292 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
293 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_sse2;
294 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_sse2;
295 pf->hpel_filter = x264_hpel_filter_sse2;
296 pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
297 pf->mc_chroma = x264_mc_chroma_sse2;
299 if( cpu&X264_CPU_SSE2_IS_FAST )
301 pf->mc_luma = mc_luma_sse2;
302 pf->get_ref = get_ref_sse2;
303 if( cpu&X264_CPU_CACHELINE_64 )
305 pf->mc_luma = mc_luma_cache64_sse2;
306 pf->get_ref = get_ref_cache64_sse2;
310 if( !(cpu&X264_CPU_SSSE3) )
313 pf->hpel_filter = x264_hpel_filter_ssse3;
314 pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
315 pf->mc_chroma = x264_mc_chroma_ssse3;