Bump dates to 2014
[x262.git] / common / x86 / mc-c.c
1 /*****************************************************************************
2  * mc-c.c: x86 motion compensation
3  *****************************************************************************
4  * Copyright (C) 2003-2014 x264 project
5  *
6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7  *          Loren Merritt <lorenm@u.washington.edu>
8  *          Jason Garrett-Glaser <darkshikari@gmail.com>
9  *
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.
14  *
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.
19  *
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  *
24  * This program is also available under a commercial proprietary license.
25  * For more information, contact us at licensing@x264.com.
26  *****************************************************************************/
27
28 #include <stdlib.h>
29 #include <stdio.h>
30 #include <string.h>
31
32 #include "common/common.h"
33 #include "mc.h"
34
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;
40
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 ))
50
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 );
53
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 ); \
57     MC_WEIGHT(w,type)
58
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 )
67 #if HIGH_BIT_DEPTH
68 MC_WEIGHT_OFFSET( 8, sse2 )
69 #endif
70 MC_WEIGHT( 8, sse2  )
71 MC_WEIGHT( 4, ssse3 )
72 MC_WEIGHT( 8, ssse3 )
73 MC_WEIGHT( 12, ssse3 )
74 MC_WEIGHT( 16, ssse3 )
75 MC_WEIGHT( 20, ssse3 )
76 MC_WEIGHT( 8, avx2 )
77 MC_WEIGHT( 16, avx2 )
78 MC_WEIGHT( 20, avx2 )
79 #undef MC_OFFSET
80 #undef MC_WEIGHT
81
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_store_interleave_chroma_mmx2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
120 void x264_store_interleave_chroma_sse2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
121 void x264_store_interleave_chroma_avx ( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
122 void x264_load_deinterleave_chroma_fenc_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
123 void x264_load_deinterleave_chroma_fenc_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
124 void x264_load_deinterleave_chroma_fenc_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
125 void x264_load_deinterleave_chroma_fenc_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
126 void x264_load_deinterleave_chroma_fdec_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
127 void x264_load_deinterleave_chroma_fdec_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
128 void x264_load_deinterleave_chroma_fdec_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
129 void x264_load_deinterleave_chroma_fdec_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
130 void *x264_memcpy_aligned_mmx( void *dst, const void *src, size_t n );
131 void *x264_memcpy_aligned_sse( void *dst, const void *src, size_t n );
132 void x264_memzero_aligned_mmx( void *dst, size_t n );
133 void x264_memzero_aligned_sse( void *dst, size_t n );
134 void x264_memzero_aligned_avx( void *dst, size_t n );
135 void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
136 void x264_integral_init4h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
137 void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
138 void x264_integral_init8h_avx ( uint16_t *sum, uint8_t *pix, intptr_t stride );
139 void x264_integral_init8h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
140 void x264_integral_init4v_mmx  ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
141 void x264_integral_init4v_sse2 ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
142 void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
143 void x264_integral_init4v_avx2( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
144 void x264_integral_init8v_mmx ( uint16_t *sum8, intptr_t stride );
145 void x264_integral_init8v_sse2( uint16_t *sum8, intptr_t stride );
146 void x264_integral_init8v_avx2( uint16_t *sum8, intptr_t stride );
147 void x264_mbtree_propagate_cost_sse2( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
148                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
149 void x264_mbtree_propagate_cost_avx ( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
150                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
151 void x264_mbtree_propagate_cost_fma4( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
152                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
153 void x264_mbtree_propagate_cost_avx2_fma3( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
154                                            uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
155
156 #define MC_CHROMA(cpu)\
157 void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, intptr_t i_dst, pixel *src, intptr_t i_src,\
158                            int dx, int dy, int i_width, int i_height );
159 MC_CHROMA(mmx2)
160 MC_CHROMA(sse2)
161 MC_CHROMA(ssse3)
162 MC_CHROMA(ssse3_cache64)
163 MC_CHROMA(avx)
164 MC_CHROMA(avx2)
165
166 #define LOWRES(cpu)\
167 void x264_frame_init_lowres_core_##cpu( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,\
168                                         intptr_t src_stride, intptr_t dst_stride, int width, int height );
169 LOWRES(mmx2)
170 LOWRES(cache32_mmx2)
171 LOWRES(sse2)
172 LOWRES(ssse3)
173 LOWRES(avx)
174 LOWRES(xop)
175 LOWRES(avx2)
176
177 #define PIXEL_AVG_W(width,cpu)\
178 void x264_pixel_avg2_w##width##_##cpu( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t );
179 /* This declares some functions that don't exist, but that isn't a problem. */
180 #define PIXEL_AVG_WALL(cpu)\
181 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);
182
183 PIXEL_AVG_WALL(mmx2)
184 PIXEL_AVG_WALL(cache32_mmx2)
185 PIXEL_AVG_WALL(cache64_mmx2)
186 PIXEL_AVG_WALL(cache64_sse2)
187 PIXEL_AVG_WALL(sse2)
188 PIXEL_AVG_WALL(cache64_ssse3)
189 PIXEL_AVG_WALL(avx2)
190
191 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
192 static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t ) =\
193 {\
194     NULL,\
195     x264_pixel_avg2_w4_##name1,\
196     x264_pixel_avg2_w8_##name2,\
197     x264_pixel_avg2_w12_##name3,\
198     x264_pixel_avg2_w16_##name4,\
199     x264_pixel_avg2_w20_##name5,\
200 };
201
202 #if HIGH_BIT_DEPTH
203 /* we can replace w12/w20 with w10/w18 as only 9/17 pixels in fact are important */
204 #define x264_pixel_avg2_w12_mmx2       x264_pixel_avg2_w10_mmx2
205 #define x264_pixel_avg2_w20_mmx2       x264_pixel_avg2_w18_mmx2
206 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w10_sse2
207 #define x264_pixel_avg2_w20_sse2         x264_pixel_avg2_w18_sse2
208 #define x264_pixel_avg2_w12_avx2         x264_pixel_avg2_w16_avx2
209 #define x264_pixel_avg2_w20_avx2         x264_pixel_avg2_w18_avx2
210 #else
211 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
212 #define x264_pixel_avg2_w12_cache64_ssse3 x264_pixel_avg2_w16_cache64_ssse3
213 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
214 #define x264_pixel_avg2_w12_sse3         x264_pixel_avg2_w16_sse3
215 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w16_sse2
216 #endif // HIGH_BIT_DEPTH
217
218 PIXEL_AVG_WTAB(mmx2, mmx2, mmx2, mmx2, mmx2, mmx2)
219 #if HIGH_BIT_DEPTH
220 PIXEL_AVG_WTAB(sse2, mmx2, sse2, sse2, sse2, sse2)
221 PIXEL_AVG_WTAB(avx2, mmx2, sse2, avx2, avx2, avx2)
222 #else // !HIGH_BIT_DEPTH
223 #if ARCH_X86
224 PIXEL_AVG_WTAB(cache32_mmx2, mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2)
225 PIXEL_AVG_WTAB(cache64_mmx2, mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2)
226 #endif
227 PIXEL_AVG_WTAB(sse2, mmx2, mmx2, sse2, sse2, sse2)
228 PIXEL_AVG_WTAB(cache64_sse2, mmx2, cache64_mmx2, cache64_sse2, cache64_sse2, cache64_sse2)
229 PIXEL_AVG_WTAB(cache64_ssse3, mmx2, cache64_mmx2, cache64_ssse3, cache64_ssse3, cache64_sse2)
230 PIXEL_AVG_WTAB(cache64_ssse3_atom, mmx2, mmx2, cache64_ssse3, cache64_ssse3, sse2)
231 PIXEL_AVG_WTAB(avx2, mmx2, mmx2, sse2, sse2, avx2)
232 #endif // HIGH_BIT_DEPTH
233
234 #define MC_COPY_WTAB(instr, name1, name2, name3)\
235 static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, intptr_t, pixel *, intptr_t, int ) =\
236 {\
237     NULL,\
238     x264_mc_copy_w4_##name1,\
239     x264_mc_copy_w8_##name2,\
240     NULL,\
241     x264_mc_copy_w16_##name3,\
242 };
243
244 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
245 #if HIGH_BIT_DEPTH
246 MC_COPY_WTAB(sse,mmx,sse,sse)
247 MC_COPY_WTAB(avx,mmx,sse,avx)
248 #else
249 MC_COPY_WTAB(sse,mmx,mmx,sse)
250 #endif
251
252 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
253     static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ) =\
254 {\
255     x264_mc_##function##_w4_##name1,\
256     x264_mc_##function##_w4_##name1,\
257     x264_mc_##function##_w8_##name2,\
258     x264_mc_##function##_w##w12version##_##instr,\
259     x264_mc_##function##_w16_##instr,\
260     x264_mc_##function##_w20_##instr,\
261 };
262
263 #if HIGH_BIT_DEPTH
264 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
265 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
266 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
267 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,12)
268 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,sse2,16)
269 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,sse2,16)
270
271 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
272 {
273     if( w->i_scale == 1<<w->i_denom )
274     {
275         if( w->i_offset < 0 )
276             w->weightfn = h->mc.offsetsub;
277         else
278             w->weightfn = h->mc.offsetadd;
279         for( int i = 0; i < 8; i++ )
280             w->cachea[i] = abs(w->i_offset<<(BIT_DEPTH-8));
281         return;
282     }
283     w->weightfn = h->mc.weight;
284     int den1 = 1<<w->i_denom;
285     int den2 = w->i_scale<<1;
286     int den3 = 1+(w->i_offset<<(BIT_DEPTH-8+1));
287     for( int i = 0; i < 8; i++ )
288     {
289         w->cachea[i] = den1;
290         w->cacheb[i] = i&1 ? den3 : den2;
291     }
292 }
293 #else
294 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
295 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
296 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
297 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,16)
298 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,mmx2,16)
299 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,mmx2,16)
300 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
301 MC_WEIGHT_WTAB(weight,avx2,ssse3,avx2,16)
302
303 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
304 {
305     int i;
306     int16_t den1;
307
308     if( w->i_scale == 1<<w->i_denom )
309     {
310         if( w->i_offset < 0 )
311             w->weightfn = h->mc.offsetsub;
312         else
313             w->weightfn = h->mc.offsetadd;
314         memset( w->cachea, abs(w->i_offset), sizeof(w->cachea) );
315         return;
316     }
317     w->weightfn = h->mc.weight;
318     den1 = 1 << (w->i_denom - 1) | w->i_offset << w->i_denom;
319     for( i = 0; i < 8; i++ )
320     {
321         w->cachea[i] = w->i_scale;
322         w->cacheb[i] = den1;
323     }
324 }
325
326 static void x264_weight_cache_ssse3( x264_t *h, x264_weight_t *w )
327 {
328     int i, den1;
329     if( w->i_scale == 1<<w->i_denom )
330     {
331         if( w->i_offset < 0 )
332             w->weightfn = h->mc.offsetsub;
333         else
334             w->weightfn = h->mc.offsetadd;
335
336         memset( w->cachea, abs( w->i_offset ), sizeof(w->cachea) );
337         return;
338     }
339     w->weightfn = h->mc.weight;
340     den1 = w->i_scale << (8 - w->i_denom);
341     for( i = 0; i < 8; i++ )
342     {
343         w->cachea[i] = den1;
344         w->cacheb[i] = w->i_offset;
345     }
346 }
347 #endif // !HIGH_BIT_DEPTH
348
349 static const uint8_t hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
350 static const uint8_t hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
351
352 #define MC_LUMA(name,instr1,instr2)\
353 static void mc_luma_##name( pixel *dst,    intptr_t i_dst_stride,\
354                             pixel *src[4], intptr_t i_src_stride,\
355                             int mvx, int mvy,\
356                             int i_width, int i_height, const x264_weight_t *weight )\
357 {\
358     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
359     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
360     pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
361     if( qpel_idx & 5 ) /* qpel interpolation needed */\
362     {\
363         pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
364         x264_pixel_avg_wtab_##instr1[i_width>>2](\
365                 dst, i_dst_stride, src1, i_src_stride,\
366                 src2, i_height );\
367         if( weight->weightfn )\
368             weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );\
369     }\
370     else if( weight->weightfn )\
371         weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );\
372     else\
373         x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
374 }
375
376 MC_LUMA(mmx2,mmx2,mmx)
377 MC_LUMA(sse2,sse2,sse)
378 #if HIGH_BIT_DEPTH
379 MC_LUMA(avx2,avx2,avx)
380 #else
381 #if ARCH_X86
382 MC_LUMA(cache32_mmx2,cache32_mmx2,mmx)
383 MC_LUMA(cache64_mmx2,cache64_mmx2,mmx)
384 #endif
385 MC_LUMA(cache64_sse2,cache64_sse2,sse)
386 MC_LUMA(cache64_ssse3,cache64_ssse3,sse)
387 MC_LUMA(cache64_ssse3_atom,cache64_ssse3_atom,sse)
388 #endif // !HIGH_BIT_DEPTH
389
390 #define GET_REF(name)\
391 static pixel *get_ref_##name( pixel *dst,   intptr_t *i_dst_stride,\
392                               pixel *src[4], intptr_t i_src_stride,\
393                               int mvx, int mvy,\
394                               int i_width, int i_height, const x264_weight_t *weight )\
395 {\
396     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
397     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
398     pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
399     if( qpel_idx & 5 ) /* qpel interpolation needed */\
400     {\
401         pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
402         x264_pixel_avg_wtab_##name[i_width>>2](\
403                 dst, *i_dst_stride, src1, i_src_stride,\
404                 src2, i_height );\
405         if( weight->weightfn )\
406             weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );\
407         return dst;\
408     }\
409     else if( weight->weightfn )\
410     {\
411         weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );\
412         return dst;\
413     }\
414     else\
415     {\
416         *i_dst_stride = i_src_stride;\
417         return src1;\
418     }\
419 }
420
421 GET_REF(mmx2)
422 GET_REF(sse2)
423 GET_REF(avx2)
424 #if !HIGH_BIT_DEPTH
425 #if ARCH_X86
426 GET_REF(cache32_mmx2)
427 GET_REF(cache64_mmx2)
428 #endif
429 GET_REF(cache64_sse2)
430 GET_REF(cache64_ssse3)
431 GET_REF(cache64_ssse3_atom)
432 #endif // !HIGH_BIT_DEPTH
433
434 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
435 void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, intptr_t stride, intptr_t width);\
436 void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, intptr_t width );\
437 void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, intptr_t width );\
438 static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,\
439                                     intptr_t stride, int width, int height, int16_t *buf )\
440 {\
441     intptr_t realign = (intptr_t)src & (align-1);\
442     src -= realign;\
443     dstv -= realign;\
444     dstc -= realign;\
445     dsth -= realign;\
446     width += realign;\
447     while( height-- )\
448     {\
449         x264_hpel_filter_v_##cpuv( dstv, src, buf+16, stride, width );\
450         x264_hpel_filter_c_##cpuc( dstc, buf+16, width );\
451         x264_hpel_filter_h_##cpuh( dsth, src, width );\
452         dsth += stride;\
453         dstv += stride;\
454         dstc += stride;\
455         src  += stride;\
456     }\
457     x264_sfence();\
458 }
459
460 HPEL(8, mmx2, mmx2, mmx2, mmx2)
461 #if HIGH_BIT_DEPTH
462 HPEL(16, sse2, sse2, sse2, sse2)
463 #else // !HIGH_BIT_DEPTH
464 HPEL(16, sse2_amd, mmx2, mmx2, sse2)
465 #if ARCH_X86_64
466 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 );
467 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 );
468 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 );
469 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 );
470 #else
471 HPEL(16, sse2, sse2, sse2, sse2)
472 HPEL(16, ssse3, ssse3, ssse3, ssse3)
473 HPEL(16, avx, avx, avx, avx)
474 HPEL(32, avx2, avx2, avx2, avx2)
475 #endif
476 #endif // HIGH_BIT_DEPTH
477
478 static void x264_plane_copy_mmx2( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )
479 {
480     int c_w = 16/sizeof(pixel) - 1;
481     if( w < 256 ) { // tiny resolutions don't want non-temporal hints. dunno the exact threshold.
482         x264_plane_copy_c( dst, i_dst, src, i_src, w, h );
483     } else if( !(w&c_w) ) {
484         x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, w, h );
485     } else if( i_src > 0 ) {
486         // have to use plain memcpy on the last line (in memory order) to avoid overreading src
487         x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, (w+c_w)&~c_w, h-1 );
488         memcpy( dst+i_dst*(h-1), src+i_src*(h-1), w*sizeof(pixel) );
489     } else {
490         memcpy( dst, src, w*sizeof(pixel) );
491         x264_plane_copy_core_mmx2( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h-1 );
492     }
493 }
494
495 #define PLANE_INTERLEAVE(cpu) \
496 static void x264_plane_copy_interleave_##cpu( pixel *dst,  intptr_t i_dst,\
497                                               pixel *srcu, intptr_t i_srcu,\
498                                               pixel *srcv, intptr_t i_srcv, int w, int h )\
499 {\
500     if( !(w&15) ) {\
501         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
502     } else if( w < 16 || (i_srcu ^ i_srcv) ) {\
503         x264_plane_copy_interleave_c( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
504     } else if( i_srcu > 0 ) {\
505         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, (w+15)&~15, h-1 );\
506         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 );\
507     } else {\
508         x264_plane_copy_interleave_c( dst, 0, srcu, 0, srcv, 0, w, 1 );\
509         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 );\
510     }\
511 }
512
513 PLANE_INTERLEAVE(mmx2)
514 PLANE_INTERLEAVE(sse2)
515 #if HIGH_BIT_DEPTH
516 PLANE_INTERLEAVE(avx)
517 #endif
518
519 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
520 {
521     if( !(cpu&X264_CPU_MMX) )
522         return;
523
524     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_mmx;
525     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_mmx;
526
527     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
528
529     pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
530     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
531     pf->copy[PIXEL_8x8]   = x264_mc_copy_w8_mmx;
532     pf->copy[PIXEL_4x4]   = x264_mc_copy_w4_mmx;
533     pf->memcpy_aligned  = x264_memcpy_aligned_mmx;
534     pf->memzero_aligned = x264_memzero_aligned_mmx;
535     pf->integral_init4v = x264_integral_init4v_mmx;
536     pf->integral_init8v = x264_integral_init8v_mmx;
537
538     if( !(cpu&X264_CPU_MMX2) )
539         return;
540
541     pf->prefetch_fenc_420 = x264_prefetch_fenc_420_mmx2;
542     pf->prefetch_fenc_422 = x264_prefetch_fenc_422_mmx2;
543     pf->prefetch_ref  = x264_prefetch_ref_mmx2;
544
545     pf->plane_copy = x264_plane_copy_mmx2;
546     pf->plane_copy_interleave = x264_plane_copy_interleave_mmx2;
547     pf->store_interleave_chroma = x264_store_interleave_chroma_mmx2;
548
549     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmx2;
550     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmx2;
551     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmx2;
552     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmx2;
553     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmx2;
554     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_mmx2;
555     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmx2;
556     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmx2;
557     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmx2;
558
559     pf->mc_luma = mc_luma_mmx2;
560     pf->get_ref = get_ref_mmx2;
561     pf->mc_chroma = x264_mc_chroma_mmx2;
562     pf->hpel_filter = x264_hpel_filter_mmx2;
563     pf->weight = x264_mc_weight_wtab_mmx2;
564     pf->weight_cache = x264_weight_cache_mmx2;
565     pf->offsetadd = x264_mc_offsetadd_wtab_mmx2;
566     pf->offsetsub = x264_mc_offsetsub_wtab_mmx2;
567
568     pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmx2;
569
570     if( cpu&X264_CPU_SSE )
571     {
572         pf->memcpy_aligned  = x264_memcpy_aligned_sse;
573         pf->memzero_aligned = x264_memzero_aligned_sse;
574     }
575
576 #if HIGH_BIT_DEPTH
577 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
578     if( cpu&(X264_CPU_CACHELINE_32|X264_CPU_CACHELINE_64) )
579         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
580 #endif
581
582     if( !(cpu&X264_CPU_SSE2) )
583         return;
584
585     pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
586
587     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
588     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
589
590     pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
591     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
592
593     if( cpu&X264_CPU_SSE2_IS_FAST )
594     {
595         pf->get_ref = get_ref_sse2;
596         pf->mc_luma = mc_luma_sse2;
597         pf->hpel_filter = x264_hpel_filter_sse2;
598     }
599
600     pf->integral_init4v = x264_integral_init4v_sse2;
601     pf->integral_init8v = x264_integral_init8v_sse2;
602     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
603     pf->store_interleave_chroma = x264_store_interleave_chroma_sse2;
604     pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
605     pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
606
607     if( cpu&X264_CPU_SSE2_IS_SLOW )
608         return;
609
610     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
611     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
612     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_sse2;
613     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_sse2;
614     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_sse2;
615     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_sse2;
616     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_sse2;
617     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_sse2;
618     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_sse2;
619
620     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
621     pf->weight = x264_mc_weight_wtab_sse2;
622
623     if( !(cpu&X264_CPU_STACK_MOD4) )
624         pf->mc_chroma = x264_mc_chroma_sse2;
625
626     if( !(cpu&X264_CPU_SSSE3) )
627         return;
628
629     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
630
631     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
632         pf->integral_init4v = x264_integral_init4v_ssse3;
633
634     if( !(cpu&X264_CPU_AVX) )
635         return;
636
637     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
638     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_avx;
639     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_avx;
640     pf->plane_copy_interleave        = x264_plane_copy_interleave_avx;
641     pf->plane_copy_deinterleave      = x264_plane_copy_deinterleave_avx;
642     pf->store_interleave_chroma      = x264_store_interleave_chroma_avx;
643     pf->copy[PIXEL_16x16]            = x264_mc_copy_w16_aligned_avx;
644
645     if( !(cpu&X264_CPU_STACK_MOD4) )
646         pf->mc_chroma = x264_mc_chroma_avx;
647
648     if( cpu&X264_CPU_XOP )
649         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
650
651     if( cpu&X264_CPU_AVX2 )
652         pf->mc_luma = mc_luma_avx2;
653 #else // !HIGH_BIT_DEPTH
654
655 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
656     if( cpu&X264_CPU_CACHELINE_32 )
657     {
658         pf->mc_luma = mc_luma_cache32_mmx2;
659         pf->get_ref = get_ref_cache32_mmx2;
660         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
661     }
662     else if( cpu&X264_CPU_CACHELINE_64 )
663     {
664         pf->mc_luma = mc_luma_cache64_mmx2;
665         pf->get_ref = get_ref_cache64_mmx2;
666         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
667     }
668 #endif
669
670     if( !(cpu&X264_CPU_SSE2) )
671         return;
672
673     pf->integral_init4v = x264_integral_init4v_sse2;
674     pf->integral_init8v = x264_integral_init8v_sse2;
675     pf->hpel_filter = x264_hpel_filter_sse2_amd;
676     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
677
678     if( !(cpu&X264_CPU_SSE2_IS_SLOW) )
679     {
680         pf->weight = x264_mc_weight_wtab_sse2;
681         if( !(cpu&X264_CPU_SLOW_ATOM) )
682         {
683             pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
684             pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
685         }
686
687         pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
688         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
689         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
690         pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
691         pf->avg[PIXEL_8x8]  = x264_pixel_avg_8x8_sse2;
692         pf->avg[PIXEL_8x4]  = x264_pixel_avg_8x4_sse2;
693         pf->hpel_filter = x264_hpel_filter_sse2;
694         pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
695         if( !(cpu&X264_CPU_STACK_MOD4) )
696             pf->mc_chroma = x264_mc_chroma_sse2;
697
698         if( cpu&X264_CPU_SSE2_IS_FAST )
699         {
700             pf->store_interleave_chroma = x264_store_interleave_chroma_sse2; // FIXME sse2fast? sse2medium?
701             pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
702             pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
703             pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
704             pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
705             pf->mc_luma = mc_luma_sse2;
706             pf->get_ref = get_ref_sse2;
707             if( cpu&X264_CPU_CACHELINE_64 )
708             {
709                 pf->mc_luma = mc_luma_cache64_sse2;
710                 pf->get_ref = get_ref_cache64_sse2;
711             }
712         }
713     }
714
715     if( !(cpu&X264_CPU_SSSE3) )
716         return;
717
718     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
719     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_ssse3;
720     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_ssse3;
721     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_ssse3;
722     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_ssse3;
723     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_ssse3;
724     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_ssse3;
725     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_ssse3;
726     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_ssse3;
727
728     if( !(cpu&X264_CPU_SLOW_PSHUFB) )
729     {
730         pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_ssse3;
731         pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_ssse3;
732         pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
733     }
734
735     if( !(cpu&X264_CPU_SLOW_PALIGNR) )
736     {
737 #if ARCH_X86_64
738         if( !(cpu&X264_CPU_SLOW_ATOM) ) /* The 64-bit version is slower, but the 32-bit version is faster? */
739 #endif
740             pf->hpel_filter = x264_hpel_filter_ssse3;
741         pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
742     }
743     if( !(cpu&X264_CPU_STACK_MOD4) )
744         pf->mc_chroma = x264_mc_chroma_ssse3;
745
746     if( cpu&X264_CPU_CACHELINE_64 )
747     {
748         if( !(cpu&X264_CPU_STACK_MOD4) )
749             pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
750         pf->mc_luma = mc_luma_cache64_ssse3;
751         pf->get_ref = get_ref_cache64_ssse3;
752         if( cpu&X264_CPU_SLOW_ATOM )
753         {
754             pf->mc_luma = mc_luma_cache64_ssse3_atom;
755             pf->get_ref = get_ref_cache64_ssse3_atom;
756         }
757     }
758
759     pf->weight_cache = x264_weight_cache_ssse3;
760     pf->weight = x264_mc_weight_wtab_ssse3;
761
762     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
763         pf->integral_init4v = x264_integral_init4v_ssse3;
764
765     if( !(cpu&X264_CPU_SSE4) )
766         return;
767
768     pf->integral_init4h = x264_integral_init4h_sse4;
769     pf->integral_init8h = x264_integral_init8h_sse4;
770
771     if( !(cpu&X264_CPU_AVX) )
772         return;
773
774     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
775     pf->integral_init8h = x264_integral_init8h_avx;
776     pf->hpel_filter = x264_hpel_filter_avx;
777
778     if( !(cpu&X264_CPU_STACK_MOD4) )
779         pf->mc_chroma = x264_mc_chroma_avx;
780
781     if( cpu&X264_CPU_XOP )
782         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
783
784     if( cpu&X264_CPU_AVX2 )
785     {
786         pf->hpel_filter = x264_hpel_filter_avx2;
787         pf->mc_chroma = x264_mc_chroma_avx2;
788         pf->weight = x264_mc_weight_wtab_avx2;
789         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_avx2;
790         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_avx2;
791         pf->integral_init8v = x264_integral_init8v_avx2;
792         pf->integral_init4v = x264_integral_init4v_avx2;
793         pf->integral_init8h = x264_integral_init8h_avx2;
794         pf->integral_init4h = x264_integral_init4h_avx2;
795         pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx2;
796     }
797 #endif // HIGH_BIT_DEPTH
798
799     if( !(cpu&X264_CPU_AVX) )
800         return;
801     pf->memzero_aligned = x264_memzero_aligned_avx;
802     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
803
804     if( cpu&X264_CPU_FMA4 )
805         pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_fma4;
806
807     if( !(cpu&X264_CPU_AVX2) )
808         return;
809     pf->get_ref = get_ref_avx2;
810
811     if( cpu&X264_CPU_FMA3 )
812         pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx2_fma3;
813 }