1 /*****************************************************************************
2 * predict.c: h264 encoder
3 *****************************************************************************
4 * Copyright (C) 2003-2008 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
22 *****************************************************************************/
24 #include "common/common.h"
28 extern void predict_16x16_v_mmx( uint8_t *src );
29 extern void predict_16x16_dc_core_mmxext( uint8_t *src, int i_dc_left );
30 extern void predict_16x16_dc_top_mmxext( uint8_t *src );
31 extern void predict_16x16_p_core_mmxext( uint8_t *src, int i00, int b, int c );
32 extern void predict_8x8c_p_core_mmxext( uint8_t *src, int i00, int b, int c );
33 extern void predict_8x8c_dc_core_mmxext( uint8_t *src, int s2, int s3 );
34 extern void predict_8x8c_v_mmx( uint8_t *src );
35 extern void predict_8x8_v_mmxext( uint8_t *src, uint8_t edge[33] );
36 extern void predict_8x8_dc_mmxext( uint8_t *src, uint8_t edge[33] );
37 extern void predict_8x8_dc_top_mmxext( uint8_t *src, uint8_t edge[33] );
38 extern void predict_8x8_dc_left_mmxext( uint8_t *src, uint8_t edge[33] );
39 extern void predict_8x8_ddl_mmxext( uint8_t *src, uint8_t edge[33] );
40 extern void predict_8x8_ddr_mmxext( uint8_t *src, uint8_t edge[33] );
41 extern void predict_8x8_ddl_sse2( uint8_t *src, uint8_t edge[33] );
42 extern void predict_8x8_ddr_sse2( uint8_t *src, uint8_t edge[33] );
43 extern void predict_8x8_vl_sse2( uint8_t *src, uint8_t edge[33] );
44 extern void predict_8x8_vr_core_mmxext( uint8_t *src, uint8_t edge[33] );
45 extern void predict_4x4_ddl_mmxext( uint8_t *src );
46 extern void predict_4x4_vl_mmxext( uint8_t *src );
47 extern void predict_16x16_dc_top_sse2( uint8_t *src );
48 extern void predict_16x16_dc_core_sse2( uint8_t *src, int i_dc_left );
49 extern void predict_16x16_v_sse2( uint8_t *src );
50 extern void predict_16x16_p_core_sse2( uint8_t *src, int i00, int b, int c );
52 #define PREDICT_16x16_P(name)\
53 static void predict_16x16_p_##name( uint8_t *src )\
59 for( i = 1; i <= 8; i++ )\
61 H += i * ( src[7+i - FDEC_STRIDE ] - src[7-i - FDEC_STRIDE ] );\
62 V += i * ( src[(7+i)*FDEC_STRIDE -1] - src[(7-i)*FDEC_STRIDE -1] );\
64 a = 16 * ( src[15*FDEC_STRIDE -1] + src[15 - FDEC_STRIDE] );\
65 b = ( 5 * H + 32 ) >> 6;\
66 c = ( 5 * V + 32 ) >> 6;\
67 i00 = a - b * 7 - c * 7 + 16;\
68 predict_16x16_p_core_##name( src, i00, b, c );\
71 PREDICT_16x16_P( mmxext )
72 PREDICT_16x16_P( sse2 )
74 static void predict_8x8c_p_mmxext( uint8_t *src )
81 for( i = 1; i <= 4; i++ )
83 H += i * ( src[3+i - FDEC_STRIDE] - src[3-i - FDEC_STRIDE] );
84 V += i * ( src[(3+i)*FDEC_STRIDE -1] - src[(3-i)*FDEC_STRIDE -1] );
87 a = 16 * ( src[7*FDEC_STRIDE -1] + src[7 - FDEC_STRIDE] );
88 b = ( 17 * H + 16 ) >> 5;
89 c = ( 17 * V + 16 ) >> 5;
90 i00 = a -3*b -3*c + 16;
92 predict_8x8c_p_core_mmxext( src, i00, b, c );
95 #define PREDICT_16x16_DC(name)\
96 static void predict_16x16_dc_##name( uint8_t *src )\
100 for( i = 0; i < 16; i+=2 )\
102 dc += src[-1 + i * FDEC_STRIDE];\
103 dc += src[-1 + (i+1) * FDEC_STRIDE];\
105 predict_16x16_dc_core_##name( src, dc );\
108 PREDICT_16x16_DC( mmxext )
109 PREDICT_16x16_DC( sse2 )
111 static void predict_8x8c_dc_mmxext( uint8_t *src )
114 + src[-1 + 0*FDEC_STRIDE]
115 + src[-1 + 1*FDEC_STRIDE]
116 + src[-1 + 2*FDEC_STRIDE]
117 + src[-1 + 3*FDEC_STRIDE];
120 + src[-1 + 4*FDEC_STRIDE]
121 + src[-1 + 5*FDEC_STRIDE]
122 + src[-1 + 6*FDEC_STRIDE]
123 + src[-1 + 7*FDEC_STRIDE];
125 predict_8x8c_dc_core_mmxext( src, s2, s3 );
129 static void predict_16x16_h( uint8_t *src )
132 for( y = 0; y < 16; y++ )
134 const uint64_t v = 0x0101010101010101ULL * src[-1];
135 uint64_t *p = (uint64_t*)src;
141 static void predict_8x8c_h( uint8_t *src )
144 for( y = 0; y < 8; y++ )
146 *(uint64_t*)src = 0x0101010101010101ULL * src[-1];
151 static void predict_16x16_dc_left( uint8_t *src )
157 for( y = 0; y < 16; y++ )
159 s += src[-1 + y * FDEC_STRIDE];
161 dc = (( s + 8 ) >> 4) * 0x0101010101010101ULL;
163 for( y = 0; y < 16; y++ )
165 uint64_t *p = (uint64_t*)src;
171 static void predict_8x8c_dc_left( uint8_t *src )
174 uint32_t s0 = 0, s1 = 0;
177 for( y = 0; y < 4; y++ )
179 s0 += src[y * FDEC_STRIDE - 1];
180 s1 += src[(y+4) * FDEC_STRIDE - 1];
182 dc0 = (( s0 + 2 ) >> 2) * 0x0101010101010101ULL;
183 dc1 = (( s1 + 2 ) >> 2) * 0x0101010101010101ULL;
185 for( y = 0; y < 4; y++ )
187 *(uint64_t*)src = dc0;
190 for( y = 0; y < 4; y++ )
192 *(uint64_t*)src = dc1;
198 static void predict_8x8c_dc_top( uint8_t *src )
201 uint32_t s0 = 0, s1 = 0;
204 for( x = 0; x < 4; x++ )
206 s0 += src[x - FDEC_STRIDE];
207 s1 += src[x + 4 - FDEC_STRIDE];
209 dc = (( s0 + 2 ) >> 2) * 0x01010101
210 + (( s1 + 2 ) >> 2) * 0x0101010100000000ULL;
212 for( y = 0; y < 8; y++ )
214 *(uint64_t*)src = dc;
222 #define PREDICT_4x4_LOAD_LEFT \
223 const int l0 = src[-1+0*FDEC_STRIDE]; \
224 const int l1 = src[-1+1*FDEC_STRIDE]; \
225 const int l2 = src[-1+2*FDEC_STRIDE]; \
226 UNUSED const int l3 = src[-1+3*FDEC_STRIDE];
228 #define PREDICT_4x4_LOAD_TOP \
229 const int t0 = src[0-1*FDEC_STRIDE]; \
230 const int t1 = src[1-1*FDEC_STRIDE]; \
231 const int t2 = src[2-1*FDEC_STRIDE]; \
232 UNUSED const int t3 = src[3-1*FDEC_STRIDE];
234 #define PREDICT_4x4_LOAD_TOP_RIGHT \
235 const int t4 = src[4-1*FDEC_STRIDE]; \
236 const int t5 = src[5-1*FDEC_STRIDE]; \
237 const int t6 = src[6-1*FDEC_STRIDE]; \
238 UNUSED const int t7 = src[7-1*FDEC_STRIDE];
240 #define F1(a,b) (((a)+(b)+1)>>1)
241 #define F2(a,b,c) (((a)+2*(b)+(c)+2)>>2)
243 #ifdef ARCH_X86_64 // slower on x86
245 static void predict_4x4_ddl( uint8_t *src )
248 PREDICT_4x4_LOAD_TOP_RIGHT
249 uint32_t vec = (F2(t3,t4,t5)<< 0)
252 + (F2(t6,t7,t7)<<24);
253 *(uint32_t*)&src[3*FDEC_STRIDE] = vec;
254 *(uint32_t*)&src[2*FDEC_STRIDE] = vec = (vec<<8) + F2(t2,t3,t4);
255 *(uint32_t*)&src[1*FDEC_STRIDE] = vec = (vec<<8) + F2(t1,t2,t3);
256 *(uint32_t*)&src[0*FDEC_STRIDE] = vec = (vec<<8) + F2(t0,t1,t2);
260 static void predict_4x4_ddr( uint8_t *src )
262 const int lt = src[-1-FDEC_STRIDE];
263 PREDICT_4x4_LOAD_LEFT
265 uint32_t vec = (F2(l0,lt,t0)<< 0)
268 + (F2(t1,t2,t3)<<24);
269 *(uint32_t*)&src[0*FDEC_STRIDE] = vec;
270 *(uint32_t*)&src[1*FDEC_STRIDE] = vec = (vec<<8) + F2(l1,l0,lt);
271 *(uint32_t*)&src[2*FDEC_STRIDE] = vec = (vec<<8) + F2(l2,l1,l0);
272 *(uint32_t*)&src[3*FDEC_STRIDE] = vec = (vec<<8) + F2(l3,l2,l1);
275 static void predict_4x4_vr( uint8_t *src )
277 const int lt = src[-1-FDEC_STRIDE];
278 PREDICT_4x4_LOAD_LEFT
280 const int ltt0 = lt + t0 + 1;
281 const int t0t1 = t0 + t1 + 1;
282 const int t1t2 = t1 + t2 + 1;
283 const int t2t3 = t2 + t3 + 1;
284 const int l0lt = l0 + lt + 1;
285 const int l1l0 = l1 + l0 + 1;
286 const int l2l1 = l2 + l1 + 1;
288 src[0*FDEC_STRIDE+0]=
289 src[2*FDEC_STRIDE+1]= ltt0 >> 1;
291 src[0*FDEC_STRIDE+1]=
292 src[2*FDEC_STRIDE+2]= t0t1 >> 1;
294 src[0*FDEC_STRIDE+2]=
295 src[2*FDEC_STRIDE+3]= t1t2 >> 1;
297 src[0*FDEC_STRIDE+3]= t2t3 >> 1;
299 src[1*FDEC_STRIDE+0]=
300 src[3*FDEC_STRIDE+1]= (l0lt + ltt0) >> 2;
302 src[1*FDEC_STRIDE+1]=
303 src[3*FDEC_STRIDE+2]= (ltt0 + t0t1) >> 2;
305 src[1*FDEC_STRIDE+2]=
306 src[3*FDEC_STRIDE+3]= (t0t1 + t1t2) >> 2;
308 src[1*FDEC_STRIDE+3]= (t1t2 + t2t3) >> 2;
309 src[2*FDEC_STRIDE+0]= (l1l0 + l0lt) >> 2;
310 src[3*FDEC_STRIDE+0]= (l2l1 + l1l0) >> 2;
313 static void predict_4x4_hd( uint8_t *src )
315 const int lt= src[-1-1*FDEC_STRIDE];
316 PREDICT_4x4_LOAD_LEFT
318 const int ltt0 = lt + t0 + 1;
319 const int t0t1 = t0 + t1 + 1;
320 const int t1t2 = t1 + t2 + 1;
321 const int l0lt = l0 + lt + 1;
322 const int l1l0 = l1 + l0 + 1;
323 const int l2l1 = l2 + l1 + 1;
324 const int l3l2 = l3 + l2 + 1;
326 src[0*FDEC_STRIDE+0]=
327 src[1*FDEC_STRIDE+2]= l0lt >> 1;
328 src[0*FDEC_STRIDE+1]=
329 src[1*FDEC_STRIDE+3]= (l0lt + ltt0) >> 2;
330 src[0*FDEC_STRIDE+2]= (ltt0 + t0t1) >> 2;
331 src[0*FDEC_STRIDE+3]= (t0t1 + t1t2) >> 2;
332 src[1*FDEC_STRIDE+0]=
333 src[2*FDEC_STRIDE+2]= l1l0 >> 1;
334 src[1*FDEC_STRIDE+1]=
335 src[2*FDEC_STRIDE+3]= (l0lt + l1l0) >> 2;
336 src[2*FDEC_STRIDE+0]=
337 src[3*FDEC_STRIDE+2]= l2l1 >> 1;
338 src[2*FDEC_STRIDE+1]=
339 src[3*FDEC_STRIDE+3]= (l1l0 + l2l1) >> 2;
340 src[3*FDEC_STRIDE+0]= l3l2 >> 1;
341 src[3*FDEC_STRIDE+1]= (l2l1 + l3l2) >> 2;
345 static void predict_4x4_vl( uint8_t *src )
348 PREDICT_4x4_LOAD_TOP_RIGHT
349 const int t0t1 = t0 + t1 + 1;
350 const int t1t2 = t1 + t2 + 1;
351 const int t2t3 = t2 + t3 + 1;
352 const int t3t4 = t3 + t4 + 1;
353 const int t4t5 = t4 + t5 + 1;
354 const int t5t6 = t5 + t6 + 1;
356 src[0*FDEC_STRIDE+0]= t0t1 >> 1;
357 src[0*FDEC_STRIDE+1]=
358 src[2*FDEC_STRIDE+0]= t1t2 >> 1;
359 src[0*FDEC_STRIDE+2]=
360 src[2*FDEC_STRIDE+1]= t2t3 >> 1;
361 src[0*FDEC_STRIDE+3]=
362 src[2*FDEC_STRIDE+2]= t3t4 >> 1;
363 src[2*FDEC_STRIDE+3]= t4t5 >> 1;
364 src[1*FDEC_STRIDE+0]= (t0t1 + t1t2) >> 2;
365 src[1*FDEC_STRIDE+1]=
366 src[3*FDEC_STRIDE+0]= (t1t2 + t2t3) >> 2;
367 src[1*FDEC_STRIDE+2]=
368 src[3*FDEC_STRIDE+1]= (t2t3 + t3t4) >> 2;
369 src[1*FDEC_STRIDE+3]=
370 src[3*FDEC_STRIDE+2]= (t3t4 + t4t5) >> 2;
371 src[3*FDEC_STRIDE+3]= (t4t5 + t5t6) >> 2;
375 static void predict_4x4_hu( uint8_t *src )
377 PREDICT_4x4_LOAD_LEFT
378 const int l1l0 = l1 + l0 + 1;
379 const int l2l1 = l2 + l1 + 1;
380 const int l3l2 = l3 + l2 + 1;
382 src[0*FDEC_STRIDE+0]= l1l0 >> 1;
383 src[0*FDEC_STRIDE+1]= (l1l0 + l2l1) >> 2;
385 src[0*FDEC_STRIDE+2]=
386 src[1*FDEC_STRIDE+0]= l2l1 >> 1;
388 src[0*FDEC_STRIDE+3]=
389 src[1*FDEC_STRIDE+1]= (l2l1 + l3l2) >> 2;
391 src[1*FDEC_STRIDE+2]=
392 src[2*FDEC_STRIDE+0]= l3l2 >> 1;
394 src[1*FDEC_STRIDE+3]=
395 src[2*FDEC_STRIDE+1]= (l2 + 3*l3 + 2) >> 2;
397 src[2*FDEC_STRIDE+3]=
398 src[3*FDEC_STRIDE+1]=
399 src[3*FDEC_STRIDE+0]=
400 src[2*FDEC_STRIDE+2]=
401 src[3*FDEC_STRIDE+2]=
402 src[3*FDEC_STRIDE+3]= l3;
406 /****************************************************************************
407 * 8x8 prediction for intra luma block
408 ****************************************************************************/
411 UNUSED int l##y = edge[14-y];
413 UNUSED int t##x = edge[16+x];
414 #define PREDICT_8x8_LOAD_TOPLEFT \
416 #define PREDICT_8x8_LOAD_LEFT \
417 PL(0) PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) PL(7)
418 #define PREDICT_8x8_LOAD_TOP \
419 PT(0) PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) PT(7)
421 #define PREDICT_8x8_DC(v) \
423 for( y = 0; y < 8; y++ ) { \
424 ((uint32_t*)src)[0] = \
425 ((uint32_t*)src)[1] = v; \
426 src += FDEC_STRIDE; \
429 #define SRC(x,y) src[(x)+(y)*FDEC_STRIDE]
431 static void predict_8x8_vr_mmxext( uint8_t *src, uint8_t edge[33] )
433 predict_8x8_vr_core_mmxext( src, edge );
435 PREDICT_8x8_LOAD_TOPLEFT
436 PREDICT_8x8_LOAD_LEFT
437 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
438 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
439 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
440 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
441 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
442 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
446 #define SUMSUB(a,b,c,d,e,f,g,h)\
452 #define INTRA_SA8D_X3(cpu) \
453 void x264_intra_sa8d_x3_8x8_##cpu( uint8_t *fenc, uint8_t edge[33], int res[3] )\
455 PREDICT_8x8_LOAD_TOP\
456 PREDICT_8x8_LOAD_LEFT\
458 DECLARE_ALIGNED_16( int16_t sa8d_1d[2][8] );\
459 SUMSUB(l0,l4,l1,l5,l2,l6,l3,l7);\
460 SUMSUB(l0,l2,l1,l3,l4,l6,l5,l7);\
461 SUMSUB(l0,l1,l2,l3,l4,l5,l6,l7);\
470 SUMSUB(t0,t4,t1,t5,t2,t6,t3,t7);\
471 SUMSUB(t0,t2,t1,t3,t4,t6,t5,t7);\
472 SUMSUB(t0,t1,t2,t3,t4,t5,t6,t7);\
481 x264_intra_sa8d_x3_8x8_core_##cpu( fenc, sa8d_1d, res );\
488 INTRA_SA8D_X3(mmxext)
491 /****************************************************************************
492 * Exported functions:
493 ****************************************************************************/
494 void x264_predict_16x16_init_mmx( int cpu, x264_predict_t pf[7] )
496 if( !(cpu&X264_CPU_MMX) )
499 pf[I_PRED_16x16_H] = predict_16x16_h;
500 pf[I_PRED_16x16_DC_LEFT] = predict_16x16_dc_left;
502 pf[I_PRED_16x16_V] = predict_16x16_v_mmx;
503 if( !(cpu&X264_CPU_MMXEXT) )
505 pf[I_PRED_16x16_DC] = predict_16x16_dc_mmxext;
506 pf[I_PRED_16x16_DC_TOP] = predict_16x16_dc_top_mmxext;
507 pf[I_PRED_16x16_P] = predict_16x16_p_mmxext;
508 if( !(cpu&X264_CPU_SSE2) )
510 pf[I_PRED_16x16_DC] = predict_16x16_dc_sse2;
511 pf[I_PRED_16x16_V] = predict_16x16_v_sse2;
512 if( cpu&X264_CPU_SSE2_IS_SLOW )
514 pf[I_PRED_16x16_DC_TOP] = predict_16x16_dc_top_sse2;
515 pf[I_PRED_16x16_P] = predict_16x16_p_sse2;
518 void x264_predict_8x8c_init_mmx( int cpu, x264_predict_t pf[7] )
520 if( !(cpu&X264_CPU_MMX) )
523 pf[I_PRED_CHROMA_H] = predict_8x8c_h;
524 pf[I_PRED_CHROMA_DC_LEFT] = predict_8x8c_dc_left;
525 pf[I_PRED_CHROMA_DC_TOP] = predict_8x8c_dc_top;
527 pf[I_PRED_CHROMA_V] = predict_8x8c_v_mmx;
528 if( !(cpu&X264_CPU_MMXEXT) )
530 pf[I_PRED_CHROMA_P] = predict_8x8c_p_mmxext;
531 pf[I_PRED_CHROMA_DC] = predict_8x8c_dc_mmxext;
534 void x264_predict_8x8_init_mmx( int cpu, x264_predict8x8_t pf[12] )
536 if( !(cpu&X264_CPU_MMXEXT) )
538 pf[I_PRED_8x8_V] = predict_8x8_v_mmxext;
539 pf[I_PRED_8x8_DC] = predict_8x8_dc_mmxext;
540 pf[I_PRED_8x8_DC_TOP] = predict_8x8_dc_top_mmxext;
541 pf[I_PRED_8x8_DC_LEFT]= predict_8x8_dc_left_mmxext;
542 pf[I_PRED_8x8_VR] = predict_8x8_vr_mmxext;
544 pf[I_PRED_8x8_DDL] = predict_8x8_ddl_mmxext;
545 pf[I_PRED_8x8_DDR] = predict_8x8_ddr_mmxext;
547 if( !(cpu&X264_CPU_SSE2) )
549 pf[I_PRED_8x8_DDL] = predict_8x8_ddl_sse2;
550 pf[I_PRED_8x8_VL] = predict_8x8_vl_sse2;
551 pf[I_PRED_8x8_DDR] = predict_8x8_ddr_sse2;
554 void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
556 if( !(cpu&X264_CPU_MMX) )
559 pf[I_PRED_4x4_DDR] = predict_4x4_ddr;
560 pf[I_PRED_4x4_VR] = predict_4x4_vr;
561 pf[I_PRED_4x4_HD] = predict_4x4_hd;
562 pf[I_PRED_4x4_HU] = predict_4x4_hu;
564 if( !(cpu&X264_CPU_MMXEXT) )
566 pf[I_PRED_4x4_DDL] = predict_4x4_ddl_mmxext;
567 pf[I_PRED_4x4_VL] = predict_4x4_vl_mmxext;