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 ))
46 #define MC_WEIGHT(w,type) \
47 extern void x264_mc_weight_w##w##_##type( uint8_t *,int, uint8_t *,int, const x264_weight_t *,int );
49 #define MC_WEIGHT_OFFSET(w,type) \
50 extern void x264_mc_offsetadd_w##w##_##type( uint8_t *,int, uint8_t *,int, const x264_weight_t *,int ); \
51 extern void x264_mc_offsetsub_w##w##_##type( uint8_t *,int, uint8_t *,int, const x264_weight_t *,int ); \
54 MC_WEIGHT_OFFSET( 4, mmxext )
55 MC_WEIGHT_OFFSET( 8, mmxext )
56 MC_WEIGHT_OFFSET( 12, mmxext )
57 MC_WEIGHT_OFFSET( 16, mmxext )
58 MC_WEIGHT_OFFSET( 20, mmxext )
59 MC_WEIGHT_OFFSET( 12, sse2 )
60 MC_WEIGHT_OFFSET( 16, sse2 )
61 MC_WEIGHT_OFFSET( 20, sse2 )
65 MC_WEIGHT( 12, ssse3 )
66 MC_WEIGHT( 16, ssse3 )
67 MC_WEIGHT( 20, ssse3 )
71 extern void x264_mc_copy_w4_mmx( uint8_t *, int, uint8_t *, int, int );
72 extern void x264_mc_copy_w8_mmx( uint8_t *, int, uint8_t *, int, int );
73 extern void x264_mc_copy_w16_mmx( uint8_t *, int, uint8_t *, int, int );
74 extern void x264_mc_copy_w16_sse2( uint8_t *, int, uint8_t *, int, int );
75 extern void x264_mc_copy_w16_sse3( uint8_t *, int, uint8_t *, int, int );
76 extern void x264_mc_copy_w16_aligned_sse2( uint8_t *, int, uint8_t *, int, int );
77 extern void x264_prefetch_fenc_mmxext( uint8_t *, int, uint8_t *, int, int );
78 extern void x264_prefetch_ref_mmxext( uint8_t *, int, int );
79 extern void x264_mc_chroma_mmxext( uint8_t *src, int i_src_stride,
80 uint8_t *dst, int i_dst_stride,
81 int dx, int dy, int i_width, int i_height );
82 extern void x264_mc_chroma_sse2( uint8_t *src, int i_src_stride,
83 uint8_t *dst, int i_dst_stride,
84 int dx, int dy, int i_width, int i_height );
85 extern void x264_mc_chroma_ssse3( uint8_t *src, int i_src_stride,
86 uint8_t *dst, int i_dst_stride,
87 int dx, int dy, int i_width, int i_height );
88 extern void x264_mc_chroma_ssse3_cache64( uint8_t *src, int i_src_stride,
89 uint8_t *dst, int i_dst_stride,
90 int dx, int dy, int i_width, int i_height );
91 extern void x264_plane_copy_mmxext( uint8_t *, int, uint8_t *, int, int w, int h);
92 extern void *x264_memcpy_aligned_mmx( void * dst, const void * src, size_t n );
93 extern void *x264_memcpy_aligned_sse2( void * dst, const void * src, size_t n );
94 extern void x264_memzero_aligned_mmx( void * dst, int n );
95 extern void x264_memzero_aligned_sse2( void * dst, int n );
96 extern void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, int stride );
97 extern void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, int stride );
98 extern void x264_integral_init4v_mmx( uint16_t *sum8, uint16_t *sum4, int stride );
99 extern void x264_integral_init4v_sse2( uint16_t *sum8, uint16_t *sum4, int stride );
100 extern void x264_integral_init8v_mmx( uint16_t *sum8, int stride );
101 extern void x264_integral_init8v_sse2( uint16_t *sum8, int stride );
102 extern void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, int stride );
103 extern void x264_mbtree_propagate_cost_sse2( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
104 uint16_t *inter_costs, uint16_t *inv_qscales, int len );
105 #define LOWRES(cpu) \
106 extern void x264_frame_init_lowres_core_##cpu( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,\
107 int src_stride, int dst_stride, int width, int height );
109 LOWRES(cache32_mmxext)
113 #define PIXEL_AVG_W(width,cpu)\
114 extern void x264_pixel_avg2_w##width##_##cpu( uint8_t *, int, uint8_t *, int, uint8_t *, int );
115 /* This declares some functions that don't exist, but that isn't a problem. */
116 #define PIXEL_AVG_WALL(cpu)\
117 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);
119 PIXEL_AVG_WALL(mmxext)
120 PIXEL_AVG_WALL(cache32_mmxext)
121 PIXEL_AVG_WALL(cache64_mmxext)
122 PIXEL_AVG_WALL(cache64_sse2)
124 PIXEL_AVG_WALL(sse2_misalign)
125 PIXEL_AVG_WALL(cache64_ssse3)
127 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
128 static void (* const x264_pixel_avg_wtab_##instr[6])( uint8_t *, int, uint8_t *, int, uint8_t *, int ) =\
131 x264_pixel_avg2_w4_##name1,\
132 x264_pixel_avg2_w8_##name2,\
133 x264_pixel_avg2_w12_##name3,\
134 x264_pixel_avg2_w16_##name4,\
135 x264_pixel_avg2_w20_##name5,\
138 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
139 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
140 #define x264_pixel_avg2_w12_sse3 x264_pixel_avg2_w16_sse3
141 #define x264_pixel_avg2_w12_sse2 x264_pixel_avg2_w16_sse2
143 PIXEL_AVG_WTAB(mmxext, mmxext, mmxext, mmxext, mmxext, mmxext)
145 PIXEL_AVG_WTAB(cache32_mmxext, mmxext, cache32_mmxext, cache32_mmxext, cache32_mmxext, cache32_mmxext)
146 PIXEL_AVG_WTAB(cache64_mmxext, mmxext, cache64_mmxext, cache64_mmxext, cache64_mmxext, cache64_mmxext)
148 PIXEL_AVG_WTAB(sse2, mmxext, mmxext, sse2, sse2, sse2)
149 PIXEL_AVG_WTAB(sse2_misalign, mmxext, mmxext, sse2, sse2, sse2_misalign)
150 PIXEL_AVG_WTAB(cache64_sse2, mmxext, cache64_mmxext, cache64_sse2, cache64_sse2, cache64_sse2)
151 PIXEL_AVG_WTAB(cache64_ssse3, mmxext, cache64_mmxext, cache64_sse2, cache64_ssse3, cache64_sse2)
153 #define MC_COPY_WTAB(instr, name1, name2, name3)\
154 static void (* const x264_mc_copy_wtab_##instr[5])( uint8_t *, int, uint8_t *, int, int ) =\
157 x264_mc_copy_w4_##name1,\
158 x264_mc_copy_w8_##name2,\
160 x264_mc_copy_w16_##name3,\
163 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
164 MC_COPY_WTAB(sse2,mmx,mmx,sse2)
166 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
167 static void (* x264_mc_##function##_wtab_##instr[6])( uint8_t *, int, uint8_t *, int, const x264_weight_t *, int ) =\
169 x264_mc_##function##_w4_##name1,\
170 x264_mc_##function##_w4_##name1,\
171 x264_mc_##function##_w8_##name2,\
172 x264_mc_##function##_w##w12version##_##instr,\
173 x264_mc_##function##_w16_##instr,\
174 x264_mc_##function##_w20_##instr,\
177 MC_WEIGHT_WTAB(weight,mmxext,mmxext,mmxext,12)
178 MC_WEIGHT_WTAB(offsetadd,mmxext,mmxext,mmxext,12)
179 MC_WEIGHT_WTAB(offsetsub,mmxext,mmxext,mmxext,12)
180 MC_WEIGHT_WTAB(weight,sse2,mmxext,sse2,16)
181 MC_WEIGHT_WTAB(offsetadd,sse2,mmxext,mmxext,16)
182 MC_WEIGHT_WTAB(offsetsub,sse2,mmxext,mmxext,16)
183 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
185 static void x264_weight_cache_mmxext( x264_t *h, x264_weight_t *w )
190 if( w->i_scale == 1<<w->i_denom )
192 if( w->i_offset < 0 )
193 w->weightfn = h->mc.offsetsub;
195 w->weightfn = h->mc.offsetadd;
196 memset( w->cachea, abs(w->i_offset), sizeof(w->cachea) );
199 w->weightfn = h->mc.weight;
200 den1 = 1 << (w->i_denom - 1) | w->i_offset << w->i_denom;
201 for( i = 0; i < 8; i++ )
203 w->cachea[i] = w->i_scale;
208 static void x264_weight_cache_ssse3( x264_t *h, x264_weight_t *w )
211 if( w->i_scale == 1<<w->i_denom )
213 if( w->i_offset < 0 )
214 w->weightfn = h->mc.offsetsub;
216 w->weightfn = h->mc.offsetadd;
218 memset( w->cachea, abs( w->i_offset ), sizeof(w->cachea) );
221 w->weightfn = h->mc.weight;
222 den1 = w->i_scale << (8 - w->i_denom);
226 w->cacheb[i] = w->i_offset;
230 static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
231 static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
233 #define MC_LUMA(name,instr1,instr2)\
234 static void mc_luma_##name( uint8_t *dst, int i_dst_stride,\
235 uint8_t *src[4], int i_src_stride,\
237 int i_width, int i_height, const x264_weight_t *weight )\
239 int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
240 int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
241 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
242 if( qpel_idx & 5 ) /* qpel interpolation needed */\
244 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
245 x264_pixel_avg_wtab_##instr1[i_width>>2](\
246 dst, i_dst_stride, src1, i_src_stride,\
248 if( weight->weightfn )\
249 weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );\
251 else if( weight->weightfn )\
252 weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );\
254 x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
257 MC_LUMA(mmxext,mmxext,mmx)
259 MC_LUMA(cache32_mmxext,cache32_mmxext,mmx)
260 MC_LUMA(cache64_mmxext,cache64_mmxext,mmx)
262 MC_LUMA(sse2,sse2,sse2)
263 MC_LUMA(cache64_sse2,cache64_sse2,sse2)
264 MC_LUMA(cache64_ssse3,cache64_ssse3,sse2)
266 #define GET_REF(name)\
267 static uint8_t *get_ref_##name( uint8_t *dst, int *i_dst_stride,\
268 uint8_t *src[4], int i_src_stride,\
270 int i_width, int i_height, const x264_weight_t *weight )\
272 int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
273 int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
274 uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
275 if( qpel_idx & 5 ) /* qpel interpolation needed */\
277 uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
278 x264_pixel_avg_wtab_##name[i_width>>2](\
279 dst, *i_dst_stride, src1, i_src_stride,\
281 if( weight->weightfn )\
282 weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );\
285 else if( weight->weightfn )\
287 weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );\
292 *i_dst_stride = i_src_stride;\
299 GET_REF(cache32_mmxext)
300 GET_REF(cache64_mmxext)
303 GET_REF(sse2_misalign)
304 GET_REF(cache64_sse2)
305 GET_REF(cache64_ssse3)
307 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
308 void x264_hpel_filter_v_##cpuv( uint8_t *dst, uint8_t *src, int16_t *buf, int stride, int width);\
309 void x264_hpel_filter_c_##cpuc( uint8_t *dst, int16_t *buf, int width );\
310 void x264_hpel_filter_h_##cpuh( uint8_t *dst, uint8_t *src, int width );\
311 void x264_sfence( void );\
312 static void x264_hpel_filter_##cpu( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,\
313 int stride, int width, int height, int16_t *buf )\
315 int realign = (intptr_t)src & (align-1);\
323 x264_hpel_filter_v_##cpuv( dstv, src, buf+8, stride, width );\
324 x264_hpel_filter_c_##cpuc( dstc, buf+8, width );\
325 x264_hpel_filter_h_##cpuh( dsth, src, width );\
334 HPEL(8, mmxext, mmxext, mmxext, mmxext)
335 HPEL(16, sse2_amd, mmxext, mmxext, sse2)
337 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 );
338 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 );
340 HPEL(16, sse2, sse2, sse2, sse2)
341 HPEL(16, ssse3, ssse3, ssse3, ssse3)
344 HPEL(16, sse2_misalign, sse2, sse2_misalign, sse2)
346 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
348 if( !(cpu&X264_CPU_MMX) )
351 pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
352 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
353 pf->copy[PIXEL_8x8] = x264_mc_copy_w8_mmx;
354 pf->copy[PIXEL_4x4] = x264_mc_copy_w4_mmx;
355 pf->memcpy_aligned = x264_memcpy_aligned_mmx;
356 pf->memzero_aligned = x264_memzero_aligned_mmx;
357 pf->integral_init4v = x264_integral_init4v_mmx;
358 pf->integral_init8v = x264_integral_init8v_mmx;
360 if( !(cpu&X264_CPU_MMXEXT) )
363 pf->mc_luma = mc_luma_mmxext;
364 pf->get_ref = get_ref_mmxext;
365 pf->mc_chroma = x264_mc_chroma_mmxext;
367 pf->weight = x264_mc_weight_wtab_mmxext;
368 pf->offsetadd = x264_mc_offsetadd_wtab_mmxext;
369 pf->offsetsub = x264_mc_offsetsub_wtab_mmxext;
370 pf->weight_cache = x264_weight_cache_mmxext;
372 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmxext;
373 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_mmxext;
374 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_mmxext;
375 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_mmxext;
376 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_mmxext;
377 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_mmxext;
378 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_mmxext;
379 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_mmxext;
381 pf->plane_copy = x264_plane_copy_mmxext;
382 pf->hpel_filter = x264_hpel_filter_mmxext;
383 pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmxext;
385 pf->prefetch_fenc = x264_prefetch_fenc_mmxext;
386 pf->prefetch_ref = x264_prefetch_ref_mmxext;
388 #ifdef ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
389 if( cpu&X264_CPU_CACHELINE_32 )
391 pf->mc_luma = mc_luma_cache32_mmxext;
392 pf->get_ref = get_ref_cache32_mmxext;
393 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
395 else if( cpu&X264_CPU_CACHELINE_64 )
397 pf->mc_luma = mc_luma_cache64_mmxext;
398 pf->get_ref = get_ref_cache64_mmxext;
399 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
403 if( !(cpu&X264_CPU_SSE2) )
406 pf->memcpy_aligned = x264_memcpy_aligned_sse2;
407 pf->memzero_aligned = x264_memzero_aligned_sse2;
408 pf->integral_init4v = x264_integral_init4v_sse2;
409 pf->integral_init8v = x264_integral_init8v_sse2;
410 pf->hpel_filter = x264_hpel_filter_sse2_amd;
411 pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
413 if( cpu&X264_CPU_SSE2_IS_SLOW )
416 pf->weight = x264_mc_weight_wtab_sse2;
417 pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
418 pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
420 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse2;
421 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
422 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_sse2;
423 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
424 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_sse2;
425 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_sse2;
426 pf->hpel_filter = x264_hpel_filter_sse2;
427 if( cpu&X264_CPU_SSE_MISALIGN )
428 pf->hpel_filter = x264_hpel_filter_sse2_misalign;
429 pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
430 pf->mc_chroma = x264_mc_chroma_sse2;
432 if( cpu&X264_CPU_SSE2_IS_FAST )
434 pf->mc_luma = mc_luma_sse2;
435 pf->get_ref = get_ref_sse2;
436 if( cpu&X264_CPU_CACHELINE_64 )
438 pf->mc_luma = mc_luma_cache64_sse2;
439 pf->get_ref = get_ref_cache64_sse2;
441 if( cpu&X264_CPU_SSE_MISALIGN )
442 pf->get_ref = get_ref_sse2_misalign;
445 if( !(cpu&X264_CPU_SSSE3) )
448 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
449 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_ssse3;
450 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_ssse3;
451 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_ssse3;
452 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_ssse3;
453 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_ssse3;
454 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_ssse3;
455 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_ssse3;
457 pf->hpel_filter = x264_hpel_filter_ssse3;
458 pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
459 pf->mc_chroma = x264_mc_chroma_ssse3;
460 if( cpu&X264_CPU_CACHELINE_64 )
462 pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
463 pf->mc_luma = mc_luma_cache64_ssse3;
464 pf->get_ref = get_ref_cache64_ssse3;
466 /* ssse3 weight is slower on Nehalem, so only assign here. */
467 pf->weight_cache = x264_weight_cache_ssse3;
468 pf->weight = x264_mc_weight_wtab_ssse3;
471 if( cpu&X264_CPU_SHUFFLE_IS_FAST )
472 pf->integral_init4v = x264_integral_init4v_ssse3;
474 if( !(cpu&X264_CPU_SSE4) )
477 pf->integral_init4h = x264_integral_init4h_sse4;
478 pf->integral_init8h = x264_integral_init8h_sse4;