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"
32 #define DECL_SUF( func, args )\
33 void func##_mmxext args;\
34 void func##_sse2 args;\
35 void func##_ssse3 args;
37 DECL_SUF( x264_pixel_avg_16x16, ( uint8_t *, int, uint8_t *, int, uint8_t *, int, int ))
38 DECL_SUF( x264_pixel_avg_16x8, ( uint8_t *, int, uint8_t *, int, uint8_t *, int, int ))
39 DECL_SUF( x264_pixel_avg_8x16, ( uint8_t *, int, uint8_t *, int, uint8_t *, int, int ))
40 DECL_SUF( x264_pixel_avg_8x8, ( uint8_t *, int, uint8_t *, int, uint8_t *, int, int ))
41 DECL_SUF( x264_pixel_avg_8x4, ( uint8_t *, int, uint8_t *, int, uint8_t *, int, int ))
42 DECL_SUF( x264_pixel_avg_4x8, ( uint8_t *, int, uint8_t *, int, uint8_t *, int, int ))
43 DECL_SUF( x264_pixel_avg_4x4, ( uint8_t *, int, uint8_t *, int, uint8_t *, int, int ))
44 DECL_SUF( x264_pixel_avg_4x2, ( uint8_t *, int, uint8_t *, int, uint8_t *, int, int ))
45 extern void x264_mc_copy_w4_mmx( uint8_t *, int, uint8_t *, int, int );
46 extern void x264_mc_copy_w8_mmx( uint8_t *, int, uint8_t *, int, int );
47 extern void x264_mc_copy_w16_mmx( uint8_t *, int, uint8_t *, int, int );
48 extern void x264_mc_copy_w16_sse2( uint8_t *, int, uint8_t *, int, int );
49 extern void x264_mc_copy_w16_sse3( uint8_t *, int, uint8_t *, int, int );
50 extern void x264_mc_copy_w16_aligned_sse2( uint8_t *, int, uint8_t *, int, int );
51 extern void x264_prefetch_fenc_mmxext( uint8_t *, int, uint8_t *, int, int );
52 extern void x264_prefetch_ref_mmxext( uint8_t *, int, int );
53 extern void x264_mc_chroma_mmxext( uint8_t *src, int i_src_stride,
54 uint8_t *dst, int i_dst_stride,
55 int dx, int dy, int i_width, int i_height );
56 extern void x264_mc_chroma_sse2( uint8_t *src, int i_src_stride,
57 uint8_t *dst, int i_dst_stride,
58 int dx, int dy, int i_width, int i_height );
59 extern void x264_mc_chroma_ssse3( uint8_t *src, int i_src_stride,
60 uint8_t *dst, int i_dst_stride,
61 int dx, int dy, int i_width, int i_height );
62 extern void x264_mc_chroma_ssse3_cache64( uint8_t *src, int i_src_stride,
63 uint8_t *dst, int i_dst_stride,
64 int dx, int dy, int i_width, int i_height );
65 extern void x264_plane_copy_mmxext( uint8_t *, int, uint8_t *, int, int w, int h);
66 extern void *x264_memcpy_aligned_mmx( void * dst, const void * src, size_t n );
67 extern void *x264_memcpy_aligned_sse2( void * dst, const void * src, size_t n );
68 extern void x264_memzero_aligned_mmx( void * dst, int n );
69 extern void x264_memzero_aligned_sse2( void * dst, int n );
70 extern void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, int stride );
71 extern void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, int stride );
72 extern void x264_integral_init4v_mmx( uint16_t *sum8, uint16_t *sum4, int stride );
73 extern void x264_integral_init4v_sse2( uint16_t *sum8, uint16_t *sum4, int stride );
74 extern void x264_integral_init8v_mmx( uint16_t *sum8, int stride );
75 extern void x264_integral_init8v_sse2( uint16_t *sum8, int stride );
76 extern void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, int stride );
78 extern void x264_frame_init_lowres_core_##cpu( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,\
79 int src_stride, int dst_stride, int width, int height );
81 LOWRES(cache32_mmxext)
85 #define PIXEL_AVG_W(width,cpu)\
86 extern void x264_pixel_avg2_w##width##_##cpu( uint8_t *, int, uint8_t *, int, uint8_t *, int );
87 /* This declares some functions that don't exist, but that isn't a problem. */
88 #define PIXEL_AVG_WALL(cpu)\
89 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);
91 PIXEL_AVG_WALL(mmxext)
92 PIXEL_AVG_WALL(cache32_mmxext)
93 PIXEL_AVG_WALL(cache64_mmxext)
94 PIXEL_AVG_WALL(cache64_sse2)
96 PIXEL_AVG_WALL(sse2_misalign)
98 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
99 static void (* const x264_pixel_avg_wtab_##instr[6])( uint8_t *, int, uint8_t *, int, uint8_t *, int ) =\
102 x264_pixel_avg2_w4_##name1,\
103 x264_pixel_avg2_w8_##name2,\
104 x264_pixel_avg2_w12_##name3,\
105 x264_pixel_avg2_w16_##name4,\
106 x264_pixel_avg2_w20_##name5,\
109 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
110 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
111 #define x264_pixel_avg2_w12_sse3 x264_pixel_avg2_w16_sse3
112 #define x264_pixel_avg2_w12_sse2 x264_pixel_avg2_w16_sse2
114 PIXEL_AVG_WTAB(mmxext, mmxext, mmxext, mmxext, mmxext, mmxext)
116 PIXEL_AVG_WTAB(cache32_mmxext, mmxext, cache32_mmxext, cache32_mmxext, cache32_mmxext, cache32_mmxext)
117 PIXEL_AVG_WTAB(cache64_mmxext, mmxext, cache64_mmxext, cache64_mmxext, cache64_mmxext, cache64_mmxext)
119 PIXEL_AVG_WTAB(sse2, mmxext, mmxext, sse2, sse2, sse2)
120 PIXEL_AVG_WTAB(sse2_misalign, mmxext, mmxext, sse2, sse2, sse2_misalign)
121 PIXEL_AVG_WTAB(cache64_sse2, mmxext, cache64_mmxext, cache64_sse2, cache64_sse2, cache64_sse2)
123 #define MC_COPY_WTAB(instr, name1, name2, name3)\
124 static void (* const x264_mc_copy_wtab_##instr[5])( uint8_t *, int, uint8_t *, int, int ) =\
127 x264_mc_copy_w4_##name1,\
128 x264_mc_copy_w8_##name2,\
130 x264_mc_copy_w16_##name3,\
133 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
134 MC_COPY_WTAB(sse2,mmx,mmx,sse2)
136 static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
137 static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
139 #define MC_LUMA(name,instr1,instr2)\
140 static void mc_luma_##name( uint8_t *dst, int i_dst_stride,\
141 uint8_t *src[4], int i_src_stride,\
143 int i_width, int i_height )\
145 int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
146 int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
147 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
148 if( qpel_idx & 5 ) /* qpel interpolation needed */\
150 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
151 x264_pixel_avg_wtab_##instr1[i_width>>2](\
152 dst, i_dst_stride, src1, i_src_stride,\
157 x264_mc_copy_wtab_##instr2[i_width>>2](\
158 dst, i_dst_stride, src1, i_src_stride, i_height );\
162 MC_LUMA(mmxext,mmxext,mmx)
164 MC_LUMA(cache32_mmxext,cache32_mmxext,mmx)
165 MC_LUMA(cache64_mmxext,cache64_mmxext,mmx)
167 MC_LUMA(sse2,sse2,sse2)
168 MC_LUMA(cache64_sse2,cache64_sse2,sse2)
170 #define GET_REF(name)\
171 static uint8_t *get_ref_##name( uint8_t *dst, int *i_dst_stride,\
172 uint8_t *src[4], int i_src_stride,\
174 int i_width, int i_height )\
176 int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
177 int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
178 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
179 if( qpel_idx & 5 ) /* qpel interpolation needed */\
181 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
182 x264_pixel_avg_wtab_##name[i_width>>2](\
183 dst, *i_dst_stride, src1, i_src_stride,\
189 *i_dst_stride = i_src_stride;\
196 GET_REF(cache32_mmxext)
197 GET_REF(cache64_mmxext)
200 GET_REF(sse2_misalign)
201 GET_REF(cache64_sse2)
203 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
204 void x264_hpel_filter_v_##cpuv( uint8_t *dst, uint8_t *src, int16_t *buf, int stride, int width);\
205 void x264_hpel_filter_c_##cpuc( uint8_t *dst, int16_t *buf, int width );\
206 void x264_hpel_filter_h_##cpuh( uint8_t *dst, uint8_t *src, int width );\
207 void x264_sfence( void );\
208 static void x264_hpel_filter_##cpu( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,\
209 int stride, int width, int height, int16_t *buf )\
211 int realign = (intptr_t)src & (align-1);\
219 x264_hpel_filter_v_##cpuv( dstv, src, buf+8, stride, width );\
220 x264_hpel_filter_c_##cpuc( dstc, buf+8, width );\
221 x264_hpel_filter_h_##cpuh( dsth, src, width );\
230 HPEL(8, mmxext, mmxext, mmxext, mmxext)
231 HPEL(16, sse2_amd, mmxext, mmxext, sse2)
233 void x264_hpel_filter_sse2( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height, int16_t *buf );
234 void x264_hpel_filter_ssse3( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height, int16_t *buf );
236 HPEL(16, sse2, sse2, sse2, sse2)
237 HPEL(16, ssse3, ssse3, ssse3, ssse3)
240 HPEL(16, sse2_misalign, sse2, sse2_misalign, sse2)
242 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
244 if( !(cpu&X264_CPU_MMX) )
247 pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
248 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
249 pf->copy[PIXEL_8x8] = x264_mc_copy_w8_mmx;
250 pf->copy[PIXEL_4x4] = x264_mc_copy_w4_mmx;
251 pf->memcpy_aligned = x264_memcpy_aligned_mmx;
252 pf->memzero_aligned = x264_memzero_aligned_mmx;
253 pf->integral_init4v = x264_integral_init4v_mmx;
254 pf->integral_init8v = x264_integral_init8v_mmx;
256 if( !(cpu&X264_CPU_MMXEXT) )
259 pf->mc_luma = mc_luma_mmxext;
260 pf->get_ref = get_ref_mmxext;
261 pf->mc_chroma = x264_mc_chroma_mmxext;
263 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmxext;
264 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_mmxext;
265 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_mmxext;
266 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_mmxext;
267 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_mmxext;
268 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_mmxext;
269 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_mmxext;
270 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_mmxext;
272 pf->plane_copy = x264_plane_copy_mmxext;
273 pf->hpel_filter = x264_hpel_filter_mmxext;
274 pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmxext;
276 pf->prefetch_fenc = x264_prefetch_fenc_mmxext;
277 pf->prefetch_ref = x264_prefetch_ref_mmxext;
279 #ifdef ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
280 if( cpu&X264_CPU_CACHELINE_32 )
282 pf->mc_luma = mc_luma_cache32_mmxext;
283 pf->get_ref = get_ref_cache32_mmxext;
284 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
286 else if( cpu&X264_CPU_CACHELINE_64 )
288 pf->mc_luma = mc_luma_cache64_mmxext;
289 pf->get_ref = get_ref_cache64_mmxext;
290 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
294 if( !(cpu&X264_CPU_SSE2) )
297 pf->memcpy_aligned = x264_memcpy_aligned_sse2;
298 pf->memzero_aligned = x264_memzero_aligned_sse2;
299 pf->integral_init4v = x264_integral_init4v_sse2;
300 pf->integral_init8v = x264_integral_init8v_sse2;
301 pf->hpel_filter = x264_hpel_filter_sse2_amd;
303 if( cpu&X264_CPU_SSE2_IS_SLOW )
306 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse2;
307 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
308 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_sse2;
309 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
310 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_sse2;
311 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_sse2;
312 pf->hpel_filter = x264_hpel_filter_sse2;
313 if( cpu&X264_CPU_SSE_MISALIGN )
314 pf->hpel_filter = x264_hpel_filter_sse2_misalign;
315 pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
316 pf->mc_chroma = x264_mc_chroma_sse2;
318 if( cpu&X264_CPU_SSE2_IS_FAST )
320 pf->mc_luma = mc_luma_sse2;
321 pf->get_ref = get_ref_sse2;
322 if( cpu&X264_CPU_CACHELINE_64 )
324 pf->mc_luma = mc_luma_cache64_sse2;
325 pf->get_ref = get_ref_cache64_sse2;
327 if( cpu&X264_CPU_SSE_MISALIGN )
328 pf->get_ref = get_ref_sse2_misalign;
331 if( !(cpu&X264_CPU_SSSE3) )
334 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
335 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_ssse3;
336 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_ssse3;
337 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_ssse3;
338 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_ssse3;
339 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_ssse3;
340 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_ssse3;
341 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_ssse3;
343 pf->hpel_filter = x264_hpel_filter_ssse3;
344 pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
345 pf->mc_chroma = x264_mc_chroma_ssse3;
346 if( cpu&X264_CPU_CACHELINE_64 )
347 pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
349 if( cpu&X264_CPU_SHUFFLE_IS_FAST )
350 pf->integral_init4v = x264_integral_init4v_ssse3;
352 if( !(cpu&X264_CPU_SSE4) )
355 pf->integral_init4h = x264_integral_init4h_sse4;
356 pf->integral_init8h = x264_integral_init8h_sse4;