1 /*****************************************************************************
2 * mc-c.c: x86 motion compensation
3 *****************************************************************************
4 * Copyright (C) 2003-2014 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.
24 * This program is also available under a commercial proprietary license.
25 * For more information, contact us at licensing@x264.com.
26 *****************************************************************************/
32 #include "common/common.h"
35 #define DECL_SUF( func, args )\
36 void func##_mmx2 args;\
37 void func##_sse2 args;\
38 void func##_ssse3 args;\
39 void func##_avx2 args;
41 DECL_SUF( x264_pixel_avg_16x16, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
42 DECL_SUF( x264_pixel_avg_16x8, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
43 DECL_SUF( x264_pixel_avg_8x16, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
44 DECL_SUF( x264_pixel_avg_8x8, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
45 DECL_SUF( x264_pixel_avg_8x4, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
46 DECL_SUF( x264_pixel_avg_4x16, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
47 DECL_SUF( x264_pixel_avg_4x8, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
48 DECL_SUF( x264_pixel_avg_4x4, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
49 DECL_SUF( x264_pixel_avg_4x2, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
51 #define MC_WEIGHT(w,type) \
52 void x264_mc_weight_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int );
54 #define MC_WEIGHT_OFFSET(w,type) \
55 void x264_mc_offsetadd_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ); \
56 void x264_mc_offsetsub_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ); \
59 MC_WEIGHT_OFFSET( 4, mmx2 )
60 MC_WEIGHT_OFFSET( 8, mmx2 )
61 MC_WEIGHT_OFFSET( 12, mmx2 )
62 MC_WEIGHT_OFFSET( 16, mmx2 )
63 MC_WEIGHT_OFFSET( 20, mmx2 )
64 MC_WEIGHT_OFFSET( 12, sse2 )
65 MC_WEIGHT_OFFSET( 16, sse2 )
66 MC_WEIGHT_OFFSET( 20, sse2 )
68 MC_WEIGHT_OFFSET( 8, sse2 )
73 MC_WEIGHT( 12, ssse3 )
74 MC_WEIGHT( 16, ssse3 )
75 MC_WEIGHT( 20, ssse3 )
82 void x264_mc_copy_w4_mmx ( pixel *, intptr_t, pixel *, intptr_t, int );
83 void x264_mc_copy_w8_mmx ( pixel *, intptr_t, pixel *, intptr_t, int );
84 void x264_mc_copy_w8_sse ( pixel *, intptr_t, pixel *, intptr_t, int );
85 void x264_mc_copy_w16_mmx( pixel *, intptr_t, pixel *, intptr_t, int );
86 void x264_mc_copy_w16_sse( pixel *, intptr_t, pixel *, intptr_t, int );
87 void x264_mc_copy_w16_aligned_sse( pixel *, intptr_t, pixel *, intptr_t, int );
88 void x264_mc_copy_w16_avx( uint16_t *, intptr_t, uint16_t *, intptr_t, int );
89 void x264_mc_copy_w16_aligned_avx( uint16_t *, intptr_t, uint16_t *, intptr_t, int );
90 void x264_prefetch_fenc_420_mmx2( pixel *, intptr_t, pixel *, intptr_t, int );
91 void x264_prefetch_fenc_422_mmx2( pixel *, intptr_t, pixel *, intptr_t, int );
92 void x264_prefetch_ref_mmx2( pixel *, intptr_t, int );
93 void x264_plane_copy_core_mmx2( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
94 void x264_plane_copy_c( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
95 void x264_plane_copy_interleave_core_mmx2( pixel *dst, intptr_t i_dst,
96 pixel *srcu, intptr_t i_srcu,
97 pixel *srcv, intptr_t i_srcv, int w, int h );
98 void x264_plane_copy_interleave_core_sse2( pixel *dst, intptr_t i_dst,
99 pixel *srcu, intptr_t i_srcu,
100 pixel *srcv, intptr_t i_srcv, int w, int h );
101 void x264_plane_copy_interleave_core_avx( pixel *dst, intptr_t i_dst,
102 pixel *srcu, intptr_t i_srcu,
103 pixel *srcv, intptr_t i_srcv, int w, int h );
104 void x264_plane_copy_interleave_c( pixel *dst, intptr_t i_dst,
105 pixel *srcu, intptr_t i_srcu,
106 pixel *srcv, intptr_t i_srcv, int w, int h );
107 void x264_plane_copy_deinterleave_mmx( pixel *dstu, intptr_t i_dstu,
108 pixel *dstv, intptr_t i_dstv,
109 pixel *src, intptr_t i_src, int w, int h );
110 void x264_plane_copy_deinterleave_sse2( pixel *dstu, intptr_t i_dstu,
111 pixel *dstv, intptr_t i_dstv,
112 pixel *src, intptr_t i_src, int w, int h );
113 void x264_plane_copy_deinterleave_ssse3( uint8_t *dstu, intptr_t i_dstu,
114 uint8_t *dstv, intptr_t i_dstv,
115 uint8_t *src, intptr_t i_src, int w, int h );
116 void x264_plane_copy_deinterleave_avx( uint16_t *dstu, intptr_t i_dstu,
117 uint16_t *dstv, intptr_t i_dstv,
118 uint16_t *src, intptr_t i_src, int w, int h );
119 void x264_plane_copy_deinterleave_rgb_sse2 ( pixel *dsta, intptr_t i_dsta,
120 pixel *dstb, intptr_t i_dstb,
121 pixel *dstc, intptr_t i_dstc,
122 pixel *src, intptr_t i_src, int pw, int w, int h );
123 void x264_plane_copy_deinterleave_rgb_ssse3( pixel *dsta, intptr_t i_dsta,
124 pixel *dstb, intptr_t i_dstb,
125 pixel *dstc, intptr_t i_dstc,
126 pixel *src, intptr_t i_src, int pw, int w, int h );
127 void x264_plane_copy_deinterleave_v210_ssse3( uint16_t *dstu, intptr_t i_dstu,
128 uint16_t *dstv, intptr_t i_dstv,
129 uint32_t *src, intptr_t i_src, int w, int h );
130 void x264_plane_copy_deinterleave_v210_avx ( uint16_t *dstu, intptr_t i_dstu,
131 uint16_t *dstv, intptr_t i_dstv,
132 uint32_t *src, intptr_t i_src, int w, int h );
133 void x264_plane_copy_deinterleave_v210_avx2 ( uint16_t *dstu, intptr_t i_dstu,
134 uint16_t *dstv, intptr_t i_dstv,
135 uint32_t *src, intptr_t i_src, int w, int h );
136 void x264_store_interleave_chroma_mmx2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
137 void x264_store_interleave_chroma_sse2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
138 void x264_store_interleave_chroma_avx ( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
139 void x264_load_deinterleave_chroma_fenc_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
140 void x264_load_deinterleave_chroma_fenc_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
141 void x264_load_deinterleave_chroma_fenc_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
142 void x264_load_deinterleave_chroma_fenc_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
143 void x264_load_deinterleave_chroma_fdec_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
144 void x264_load_deinterleave_chroma_fdec_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
145 void x264_load_deinterleave_chroma_fdec_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
146 void x264_load_deinterleave_chroma_fdec_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
147 void *x264_memcpy_aligned_mmx( void *dst, const void *src, size_t n );
148 void *x264_memcpy_aligned_sse( void *dst, const void *src, size_t n );
149 void x264_memzero_aligned_mmx( void *dst, size_t n );
150 void x264_memzero_aligned_sse( void *dst, size_t n );
151 void x264_memzero_aligned_avx( void *dst, size_t n );
152 void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
153 void x264_integral_init4h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
154 void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
155 void x264_integral_init8h_avx ( uint16_t *sum, uint8_t *pix, intptr_t stride );
156 void x264_integral_init8h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
157 void x264_integral_init4v_mmx ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
158 void x264_integral_init4v_sse2 ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
159 void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
160 void x264_integral_init4v_avx2( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
161 void x264_integral_init8v_mmx ( uint16_t *sum8, intptr_t stride );
162 void x264_integral_init8v_sse2( uint16_t *sum8, intptr_t stride );
163 void x264_integral_init8v_avx2( uint16_t *sum8, intptr_t stride );
164 void x264_mbtree_propagate_cost_sse2( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
165 uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
166 void x264_mbtree_propagate_cost_avx ( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
167 uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
168 void x264_mbtree_propagate_cost_fma4( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
169 uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
170 void x264_mbtree_propagate_cost_avx2_fma3( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
171 uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
173 #define MC_CHROMA(cpu)\
174 void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, intptr_t i_dst, pixel *src, intptr_t i_src,\
175 int dx, int dy, int i_width, int i_height );
179 MC_CHROMA(ssse3_cache64)
184 void x264_frame_init_lowres_core_##cpu( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,\
185 intptr_t src_stride, intptr_t dst_stride, int width, int height );
194 #define PIXEL_AVG_W(width,cpu)\
195 void x264_pixel_avg2_w##width##_##cpu( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t );
196 /* This declares some functions that don't exist, but that isn't a problem. */
197 #define PIXEL_AVG_WALL(cpu)\
198 PIXEL_AVG_W(4,cpu); PIXEL_AVG_W(8,cpu); PIXEL_AVG_W(10,cpu); PIXEL_AVG_W(12,cpu); PIXEL_AVG_W(16,cpu); PIXEL_AVG_W(18,cpu); PIXEL_AVG_W(20,cpu);
201 PIXEL_AVG_WALL(cache32_mmx2)
202 PIXEL_AVG_WALL(cache64_mmx2)
203 PIXEL_AVG_WALL(cache64_sse2)
205 PIXEL_AVG_WALL(cache64_ssse3)
208 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
209 static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t ) =\
212 x264_pixel_avg2_w4_##name1,\
213 x264_pixel_avg2_w8_##name2,\
214 x264_pixel_avg2_w12_##name3,\
215 x264_pixel_avg2_w16_##name4,\
216 x264_pixel_avg2_w20_##name5,\
220 /* we can replace w12/w20 with w10/w18 as only 9/17 pixels in fact are important */
221 #define x264_pixel_avg2_w12_mmx2 x264_pixel_avg2_w10_mmx2
222 #define x264_pixel_avg2_w20_mmx2 x264_pixel_avg2_w18_mmx2
223 #define x264_pixel_avg2_w12_sse2 x264_pixel_avg2_w10_sse2
224 #define x264_pixel_avg2_w20_sse2 x264_pixel_avg2_w18_sse2
225 #define x264_pixel_avg2_w12_avx2 x264_pixel_avg2_w16_avx2
226 #define x264_pixel_avg2_w20_avx2 x264_pixel_avg2_w18_avx2
228 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
229 #define x264_pixel_avg2_w12_cache64_ssse3 x264_pixel_avg2_w16_cache64_ssse3
230 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
231 #define x264_pixel_avg2_w12_sse3 x264_pixel_avg2_w16_sse3
232 #define x264_pixel_avg2_w12_sse2 x264_pixel_avg2_w16_sse2
233 #endif // HIGH_BIT_DEPTH
235 PIXEL_AVG_WTAB(mmx2, mmx2, mmx2, mmx2, mmx2, mmx2)
237 PIXEL_AVG_WTAB(sse2, mmx2, sse2, sse2, sse2, sse2)
238 PIXEL_AVG_WTAB(avx2, mmx2, sse2, avx2, avx2, avx2)
239 #else // !HIGH_BIT_DEPTH
241 PIXEL_AVG_WTAB(cache32_mmx2, mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2)
242 PIXEL_AVG_WTAB(cache64_mmx2, mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2)
244 PIXEL_AVG_WTAB(sse2, mmx2, mmx2, sse2, sse2, sse2)
245 PIXEL_AVG_WTAB(cache64_sse2, mmx2, cache64_mmx2, cache64_sse2, cache64_sse2, cache64_sse2)
246 PIXEL_AVG_WTAB(cache64_ssse3, mmx2, cache64_mmx2, cache64_ssse3, cache64_ssse3, cache64_sse2)
247 PIXEL_AVG_WTAB(cache64_ssse3_atom, mmx2, mmx2, cache64_ssse3, cache64_ssse3, sse2)
248 PIXEL_AVG_WTAB(avx2, mmx2, mmx2, sse2, sse2, avx2)
249 #endif // HIGH_BIT_DEPTH
251 #define MC_COPY_WTAB(instr, name1, name2, name3)\
252 static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, intptr_t, pixel *, intptr_t, int ) =\
255 x264_mc_copy_w4_##name1,\
256 x264_mc_copy_w8_##name2,\
258 x264_mc_copy_w16_##name3,\
261 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
263 MC_COPY_WTAB(sse,mmx,sse,sse)
264 MC_COPY_WTAB(avx,mmx,sse,avx)
266 MC_COPY_WTAB(sse,mmx,mmx,sse)
269 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
270 static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ) =\
272 x264_mc_##function##_w4_##name1,\
273 x264_mc_##function##_w4_##name1,\
274 x264_mc_##function##_w8_##name2,\
275 x264_mc_##function##_w##w12version##_##instr,\
276 x264_mc_##function##_w16_##instr,\
277 x264_mc_##function##_w20_##instr,\
281 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
282 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
283 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
284 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,12)
285 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,sse2,16)
286 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,sse2,16)
288 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
290 if( w->i_scale == 1<<w->i_denom )
292 if( w->i_offset < 0 )
293 w->weightfn = h->mc.offsetsub;
295 w->weightfn = h->mc.offsetadd;
296 for( int i = 0; i < 8; i++ )
297 w->cachea[i] = abs(w->i_offset<<(BIT_DEPTH-8));
300 w->weightfn = h->mc.weight;
301 int den1 = 1<<w->i_denom;
302 int den2 = w->i_scale<<1;
303 int den3 = 1+(w->i_offset<<(BIT_DEPTH-8+1));
304 for( int i = 0; i < 8; i++ )
307 w->cacheb[i] = i&1 ? den3 : den2;
311 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
312 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
313 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
314 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,16)
315 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,mmx2,16)
316 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,mmx2,16)
317 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
318 MC_WEIGHT_WTAB(weight,avx2,ssse3,avx2,16)
320 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
325 if( w->i_scale == 1<<w->i_denom )
327 if( w->i_offset < 0 )
328 w->weightfn = h->mc.offsetsub;
330 w->weightfn = h->mc.offsetadd;
331 memset( w->cachea, abs(w->i_offset), sizeof(w->cachea) );
334 w->weightfn = h->mc.weight;
335 den1 = 1 << (w->i_denom - 1) | w->i_offset << w->i_denom;
336 for( i = 0; i < 8; i++ )
338 w->cachea[i] = w->i_scale;
343 static void x264_weight_cache_ssse3( x264_t *h, x264_weight_t *w )
346 if( w->i_scale == 1<<w->i_denom )
348 if( w->i_offset < 0 )
349 w->weightfn = h->mc.offsetsub;
351 w->weightfn = h->mc.offsetadd;
353 memset( w->cachea, abs( w->i_offset ), sizeof(w->cachea) );
356 w->weightfn = h->mc.weight;
357 den1 = w->i_scale << (8 - w->i_denom);
358 for( i = 0; i < 8; i++ )
361 w->cacheb[i] = w->i_offset;
364 #endif // !HIGH_BIT_DEPTH
366 static const uint8_t hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
367 static const uint8_t hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
369 #define MC_LUMA(name,instr1,instr2)\
370 static void mc_luma_##name( pixel *dst, intptr_t i_dst_stride,\
371 pixel *src[4], intptr_t i_src_stride,\
373 int i_width, int i_height, const x264_weight_t *weight )\
375 int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
376 int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
377 pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
378 if( qpel_idx & 5 ) /* qpel interpolation needed */\
380 pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
381 x264_pixel_avg_wtab_##instr1[i_width>>2](\
382 dst, i_dst_stride, src1, i_src_stride,\
384 if( weight->weightfn )\
385 weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );\
387 else if( weight->weightfn )\
388 weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );\
390 x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
393 MC_LUMA(mmx2,mmx2,mmx)
394 MC_LUMA(sse2,sse2,sse)
396 MC_LUMA(avx2,avx2,avx)
399 MC_LUMA(cache32_mmx2,cache32_mmx2,mmx)
400 MC_LUMA(cache64_mmx2,cache64_mmx2,mmx)
402 MC_LUMA(cache64_sse2,cache64_sse2,sse)
403 MC_LUMA(cache64_ssse3,cache64_ssse3,sse)
404 MC_LUMA(cache64_ssse3_atom,cache64_ssse3_atom,sse)
405 #endif // !HIGH_BIT_DEPTH
407 #define GET_REF(name)\
408 static pixel *get_ref_##name( pixel *dst, intptr_t *i_dst_stride,\
409 pixel *src[4], intptr_t i_src_stride,\
411 int i_width, int i_height, const x264_weight_t *weight )\
413 int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
414 int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
415 pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
416 if( qpel_idx & 5 ) /* qpel interpolation needed */\
418 pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
419 x264_pixel_avg_wtab_##name[i_width>>2](\
420 dst, *i_dst_stride, src1, i_src_stride,\
422 if( weight->weightfn )\
423 weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );\
426 else if( weight->weightfn )\
428 weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );\
433 *i_dst_stride = i_src_stride;\
443 GET_REF(cache32_mmx2)
444 GET_REF(cache64_mmx2)
446 GET_REF(cache64_sse2)
447 GET_REF(cache64_ssse3)
448 GET_REF(cache64_ssse3_atom)
449 #endif // !HIGH_BIT_DEPTH
451 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
452 void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, intptr_t stride, intptr_t width);\
453 void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, intptr_t width );\
454 void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, intptr_t width );\
455 static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,\
456 intptr_t stride, int width, int height, int16_t *buf )\
458 intptr_t realign = (intptr_t)src & (align-1);\
466 x264_hpel_filter_v_##cpuv( dstv, src, buf+16, stride, width );\
467 x264_hpel_filter_c_##cpuc( dstc, buf+16, width );\
468 x264_hpel_filter_h_##cpuh( dsth, src, width );\
477 HPEL(8, mmx2, mmx2, mmx2, mmx2)
479 HPEL(16, sse2, sse2, sse2, sse2)
480 #else // !HIGH_BIT_DEPTH
481 HPEL(16, sse2_amd, mmx2, mmx2, sse2)
483 void x264_hpel_filter_sse2 ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
484 void x264_hpel_filter_ssse3( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
485 void x264_hpel_filter_avx ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
486 void x264_hpel_filter_avx2 ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
488 HPEL(16, sse2, sse2, sse2, sse2)
489 HPEL(16, ssse3, ssse3, ssse3, ssse3)
490 HPEL(16, avx, avx, avx, avx)
491 HPEL(32, avx2, avx2, avx2, avx2)
493 #endif // HIGH_BIT_DEPTH
495 static void x264_plane_copy_mmx2( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )
497 int c_w = 16/sizeof(pixel) - 1;
498 if( w < 256 ) { // tiny resolutions don't want non-temporal hints. dunno the exact threshold.
499 x264_plane_copy_c( dst, i_dst, src, i_src, w, h );
500 } else if( !(w&c_w) ) {
501 x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, w, h );
502 } else if( i_src > 0 ) {
503 // have to use plain memcpy on the last line (in memory order) to avoid overreading src
504 x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, (w+c_w)&~c_w, h-1 );
505 memcpy( dst+i_dst*(h-1), src+i_src*(h-1), w*sizeof(pixel) );
507 memcpy( dst, src, w*sizeof(pixel) );
508 x264_plane_copy_core_mmx2( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h-1 );
512 #define PLANE_INTERLEAVE(cpu) \
513 static void x264_plane_copy_interleave_##cpu( pixel *dst, intptr_t i_dst,\
514 pixel *srcu, intptr_t i_srcu,\
515 pixel *srcv, intptr_t i_srcv, int w, int h )\
518 x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
519 } else if( w < 16 || (i_srcu ^ i_srcv) ) {\
520 x264_plane_copy_interleave_c( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
521 } else if( i_srcu > 0 ) {\
522 x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, (w+15)&~15, h-1 );\
523 x264_plane_copy_interleave_c( dst+i_dst*(h-1), 0, srcu+i_srcu*(h-1), 0, srcv+i_srcv*(h-1), 0, w, 1 );\
525 x264_plane_copy_interleave_c( dst, 0, srcu, 0, srcv, 0, w, 1 );\
526 x264_plane_copy_interleave_core_##cpu( dst+i_dst, i_dst, srcu+i_srcu, i_srcu, srcv+i_srcv, i_srcv, (w+15)&~15, h-1 );\
530 PLANE_INTERLEAVE(mmx2)
531 PLANE_INTERLEAVE(sse2)
533 PLANE_INTERLEAVE(avx)
536 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
538 if( !(cpu&X264_CPU_MMX) )
541 pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_mmx;
542 pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_mmx;
544 pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
546 pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
547 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
548 pf->copy[PIXEL_8x8] = x264_mc_copy_w8_mmx;
549 pf->copy[PIXEL_4x4] = x264_mc_copy_w4_mmx;
550 pf->memcpy_aligned = x264_memcpy_aligned_mmx;
551 pf->memzero_aligned = x264_memzero_aligned_mmx;
552 pf->integral_init4v = x264_integral_init4v_mmx;
553 pf->integral_init8v = x264_integral_init8v_mmx;
555 if( !(cpu&X264_CPU_MMX2) )
558 pf->prefetch_fenc_420 = x264_prefetch_fenc_420_mmx2;
559 pf->prefetch_fenc_422 = x264_prefetch_fenc_422_mmx2;
560 pf->prefetch_ref = x264_prefetch_ref_mmx2;
562 pf->plane_copy = x264_plane_copy_mmx2;
563 pf->plane_copy_interleave = x264_plane_copy_interleave_mmx2;
564 pf->store_interleave_chroma = x264_store_interleave_chroma_mmx2;
566 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmx2;
567 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_mmx2;
568 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_mmx2;
569 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_mmx2;
570 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_mmx2;
571 pf->avg[PIXEL_4x16] = x264_pixel_avg_4x16_mmx2;
572 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_mmx2;
573 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_mmx2;
574 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_mmx2;
576 pf->mc_luma = mc_luma_mmx2;
577 pf->get_ref = get_ref_mmx2;
578 pf->mc_chroma = x264_mc_chroma_mmx2;
579 pf->hpel_filter = x264_hpel_filter_mmx2;
580 pf->weight = x264_mc_weight_wtab_mmx2;
581 pf->weight_cache = x264_weight_cache_mmx2;
582 pf->offsetadd = x264_mc_offsetadd_wtab_mmx2;
583 pf->offsetsub = x264_mc_offsetsub_wtab_mmx2;
585 pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmx2;
587 if( cpu&X264_CPU_SSE )
589 pf->memcpy_aligned = x264_memcpy_aligned_sse;
590 pf->memzero_aligned = x264_memzero_aligned_sse;
594 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
595 if( cpu&(X264_CPU_CACHELINE_32|X264_CPU_CACHELINE_64) )
596 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
599 if( !(cpu&X264_CPU_SSE2) )
602 pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
604 pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
605 pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
607 pf->plane_copy_interleave = x264_plane_copy_interleave_sse2;
608 pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
610 if( cpu&X264_CPU_SSE2_IS_FAST )
612 pf->get_ref = get_ref_sse2;
613 pf->mc_luma = mc_luma_sse2;
614 pf->hpel_filter = x264_hpel_filter_sse2;
617 pf->integral_init4v = x264_integral_init4v_sse2;
618 pf->integral_init8v = x264_integral_init8v_sse2;
619 pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
620 pf->store_interleave_chroma = x264_store_interleave_chroma_sse2;
621 pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
622 pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
624 if( cpu&X264_CPU_SSE2_IS_SLOW )
627 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
628 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_sse2;
629 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
630 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_sse2;
631 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_sse2;
632 pf->avg[PIXEL_4x16] = x264_pixel_avg_4x16_sse2;
633 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_sse2;
634 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_sse2;
635 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_sse2;
637 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
638 pf->weight = x264_mc_weight_wtab_sse2;
640 if( !(cpu&X264_CPU_STACK_MOD4) )
641 pf->mc_chroma = x264_mc_chroma_sse2;
643 if( !(cpu&X264_CPU_SSSE3) )
646 pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
647 pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_ssse3;
649 if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
650 pf->integral_init4v = x264_integral_init4v_ssse3;
652 if( !(cpu&X264_CPU_AVX) )
655 pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
656 pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_avx;
657 pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_avx;
658 pf->plane_copy_interleave = x264_plane_copy_interleave_avx;
659 pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_avx;
660 pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx;
661 pf->store_interleave_chroma = x264_store_interleave_chroma_avx;
662 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_avx;
664 if( !(cpu&X264_CPU_STACK_MOD4) )
665 pf->mc_chroma = x264_mc_chroma_avx;
667 if( cpu&X264_CPU_XOP )
668 pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
670 if( cpu&X264_CPU_AVX2 )
672 pf->mc_luma = mc_luma_avx2;
673 pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx2;
675 #else // !HIGH_BIT_DEPTH
677 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
678 if( cpu&X264_CPU_CACHELINE_32 )
680 pf->mc_luma = mc_luma_cache32_mmx2;
681 pf->get_ref = get_ref_cache32_mmx2;
682 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
684 else if( cpu&X264_CPU_CACHELINE_64 )
686 pf->mc_luma = mc_luma_cache64_mmx2;
687 pf->get_ref = get_ref_cache64_mmx2;
688 pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
692 if( !(cpu&X264_CPU_SSE2) )
695 pf->integral_init4v = x264_integral_init4v_sse2;
696 pf->integral_init8v = x264_integral_init8v_sse2;
697 pf->hpel_filter = x264_hpel_filter_sse2_amd;
698 pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
699 pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_sse2;
701 if( !(cpu&X264_CPU_SSE2_IS_SLOW) )
703 pf->weight = x264_mc_weight_wtab_sse2;
704 if( !(cpu&X264_CPU_SLOW_ATOM) )
706 pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
707 pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
710 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
711 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
712 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_sse2;
713 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
714 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_sse2;
715 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_sse2;
716 pf->hpel_filter = x264_hpel_filter_sse2;
717 pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
718 if( !(cpu&X264_CPU_STACK_MOD4) )
719 pf->mc_chroma = x264_mc_chroma_sse2;
721 if( cpu&X264_CPU_SSE2_IS_FAST )
723 pf->store_interleave_chroma = x264_store_interleave_chroma_sse2; // FIXME sse2fast? sse2medium?
724 pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
725 pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
726 pf->plane_copy_interleave = x264_plane_copy_interleave_sse2;
727 pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
728 pf->mc_luma = mc_luma_sse2;
729 pf->get_ref = get_ref_sse2;
730 if( cpu&X264_CPU_CACHELINE_64 )
732 pf->mc_luma = mc_luma_cache64_sse2;
733 pf->get_ref = get_ref_cache64_sse2;
738 if( !(cpu&X264_CPU_SSSE3) )
741 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
742 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_ssse3;
743 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_ssse3;
744 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_ssse3;
745 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_ssse3;
746 pf->avg[PIXEL_4x16] = x264_pixel_avg_4x16_ssse3;
747 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_ssse3;
748 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_ssse3;
749 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_ssse3;
750 pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_ssse3;
752 if( !(cpu&X264_CPU_SLOW_PSHUFB) )
754 pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_ssse3;
755 pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_ssse3;
756 pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
759 if( !(cpu&X264_CPU_SLOW_PALIGNR) )
762 if( !(cpu&X264_CPU_SLOW_ATOM) ) /* The 64-bit version is slower, but the 32-bit version is faster? */
764 pf->hpel_filter = x264_hpel_filter_ssse3;
765 pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
767 if( !(cpu&X264_CPU_STACK_MOD4) )
768 pf->mc_chroma = x264_mc_chroma_ssse3;
770 if( cpu&X264_CPU_CACHELINE_64 )
772 if( !(cpu&X264_CPU_STACK_MOD4) )
773 pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
774 pf->mc_luma = mc_luma_cache64_ssse3;
775 pf->get_ref = get_ref_cache64_ssse3;
776 if( cpu&X264_CPU_SLOW_ATOM )
778 pf->mc_luma = mc_luma_cache64_ssse3_atom;
779 pf->get_ref = get_ref_cache64_ssse3_atom;
783 pf->weight_cache = x264_weight_cache_ssse3;
784 pf->weight = x264_mc_weight_wtab_ssse3;
786 if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
787 pf->integral_init4v = x264_integral_init4v_ssse3;
789 if( !(cpu&X264_CPU_SSE4) )
792 pf->integral_init4h = x264_integral_init4h_sse4;
793 pf->integral_init8h = x264_integral_init8h_sse4;
795 if( !(cpu&X264_CPU_AVX) )
798 pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
799 pf->integral_init8h = x264_integral_init8h_avx;
800 pf->hpel_filter = x264_hpel_filter_avx;
802 if( !(cpu&X264_CPU_STACK_MOD4) )
803 pf->mc_chroma = x264_mc_chroma_avx;
805 if( cpu&X264_CPU_XOP )
806 pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
808 if( cpu&X264_CPU_AVX2 )
810 pf->hpel_filter = x264_hpel_filter_avx2;
811 pf->mc_chroma = x264_mc_chroma_avx2;
812 pf->weight = x264_mc_weight_wtab_avx2;
813 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_avx2;
814 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_avx2;
815 pf->integral_init8v = x264_integral_init8v_avx2;
816 pf->integral_init4v = x264_integral_init4v_avx2;
817 pf->integral_init8h = x264_integral_init8h_avx2;
818 pf->integral_init4h = x264_integral_init4h_avx2;
819 pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx2;
821 #endif // HIGH_BIT_DEPTH
823 if( !(cpu&X264_CPU_AVX) )
825 pf->memzero_aligned = x264_memzero_aligned_avx;
826 pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
828 if( cpu&X264_CPU_FMA4 )
829 pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_fma4;
831 if( !(cpu&X264_CPU_AVX2) )
833 pf->get_ref = get_ref_avx2;
835 if( cpu&X264_CPU_FMA3 )
836 pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx2_fma3;