1 /*****************************************************************************
2 * quant.c: quantization and level-run
3 *****************************************************************************
4 * Copyright (C) 2005-2014 x264 project
5 *
6 * Authors: Loren Merritt <lorenm@u.washington.edu>
7 * Fiona Glaser <fiona@x264.com>
8 * Christian Heine <sennindemokrit@gmx.net>
9 * Henrik Gramner <henrik@gramner.com>
10 *
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2 of the License, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
24 *
25 * This program is also available under a commercial proprietary license.
26 * For more information, contact us at licensing@x264.com.
27 *****************************************************************************/
28
29 #include "common.h"
30
31 #if HAVE_MMX
32 #include "x86/quant.h"
33 #endif
34 #if ARCH_PPC
35 # include "ppc/quant.h"
36 #endif
37 #if ARCH_ARM
38 # include "arm/quant.h"
39 #endif
40 #if ARCH_AARCH64
41 # include "aarch64/quant.h"
42 #endif
43
44 #define QUANT_ONE( coef, mf, f ) \
45 { \
46 if( (coef) > 0 ) \
47 (coef) = (f + (coef)) * (mf) >> 16; \
48 else \
49 (coef) = - ((f - (coef)) * (mf) >> 16); \
50 nz |= (coef); \
51 }
52
quant_8x8(dctcoef dct[64],udctcoef mf[64],udctcoef bias[64])53 static int quant_8x8( dctcoef dct[64], udctcoef mf[64], udctcoef bias[64] )
54 {
55 int nz = 0;
56 int i;
57
58 for( i = 0; i < 64; i++ )
59 QUANT_ONE( dct[i], mf[i], bias[i] );
60 return !!nz;
61 }
62
quant_4x4(dctcoef dct[16],udctcoef mf[16],udctcoef bias[16])63 static int quant_4x4( dctcoef dct[16], udctcoef mf[16], udctcoef bias[16] )
64 {
65 int nz = 0;
66 int i;
67
68 for( i = 0; i < 16; i++ )
69 QUANT_ONE( dct[i], mf[i], bias[i] );
70 return !!nz;
71 }
72
quant_4x4x4(dctcoef dct[4][16],udctcoef mf[16],udctcoef bias[16])73 static int quant_4x4x4( dctcoef dct[4][16], udctcoef mf[16], udctcoef bias[16] )
74 {
75 int nza = 0;
76 int j;
77
78 for( j = 0; j < 4; j++ )
79 {
80 int nz = 0;
81 int i;
82
83 for( i = 0; i < 16; i++ )
84 QUANT_ONE( dct[j][i], mf[i], bias[i] );
85 nza |= (!!nz)<<j;
86 }
87 return nza;
88 }
89
quant_4x4_dc(dctcoef dct[16],int mf,int bias)90 static int quant_4x4_dc( dctcoef dct[16], int mf, int bias )
91 {
92 int nz = 0;
93 int i;
94
95 for( i = 0; i < 16; i++ )
96 QUANT_ONE( dct[i], mf, bias );
97 return !!nz;
98 }
99
quant_2x2_dc(dctcoef dct[4],int mf,int bias)100 static int quant_2x2_dc( dctcoef dct[4], int mf, int bias )
101 {
102 int nz = 0;
103 QUANT_ONE( dct[0], mf, bias );
104 QUANT_ONE( dct[1], mf, bias );
105 QUANT_ONE( dct[2], mf, bias );
106 QUANT_ONE( dct[3], mf, bias );
107 return !!nz;
108 }
109
110 #define DEQUANT_SHL( x ) \
111 dct[x] = ( dct[x] * dequant_mf[i_mf][x] ) << i_qbits
112
113 #define DEQUANT_SHR( x ) \
114 dct[x] = ( dct[x] * dequant_mf[i_mf][x] + f ) >> (-i_qbits)
115
dequant_4x4(dctcoef dct[16],int dequant_mf[6][16],int i_qp)116 static void dequant_4x4( dctcoef dct[16], int dequant_mf[6][16], int i_qp )
117 {
118 const int i_mf = i_qp%6;
119 const int i_qbits = i_qp/6 - 4;
120
121 if( i_qbits >= 0 )
122 {
123 int i;
124 for( i = 0; i < 16; i++ )
125 DEQUANT_SHL( i );
126 }
127 else
128 {
129 const int f = 1 << (-i_qbits-1);
130 int i;
131
132 for( i = 0; i < 16; i++ )
133 DEQUANT_SHR( i );
134 }
135 }
136
dequant_8x8(dctcoef dct[64],int dequant_mf[6][64],int i_qp)137 static void dequant_8x8( dctcoef dct[64], int dequant_mf[6][64], int i_qp )
138 {
139 const int i_mf = i_qp%6;
140 const int i_qbits = i_qp/6 - 6;
141
142 if( i_qbits >= 0 )
143 {
144 int i;
145
146 for( i = 0; i < 64; i++ )
147 DEQUANT_SHL( i );
148 }
149 else
150 {
151 const int f = 1 << (-i_qbits-1);
152 int i;
153
154 for( i = 0; i < 64; i++ )
155 DEQUANT_SHR( i );
156 }
157 }
158
dequant_4x4_dc(dctcoef dct[16],int dequant_mf[6][16],int i_qp)159 static void dequant_4x4_dc( dctcoef dct[16], int dequant_mf[6][16], int i_qp )
160 {
161 const int i_qbits = i_qp/6 - 6;
162
163 if( i_qbits >= 0 )
164 {
165 const int i_dmf = dequant_mf[i_qp%6][0] << i_qbits;
166 int i;
167
168 for( i = 0; i < 16; i++ )
169 dct[i] *= i_dmf;
170 }
171 else
172 {
173 const int i_dmf = dequant_mf[i_qp%6][0];
174 const int f = 1 << (-i_qbits-1);
175 int i;
176
177 for( i = 0; i < 16; i++ )
178 dct[i] = ( dct[i] * i_dmf + f ) >> (-i_qbits);
179 }
180 }
181
182 #define IDCT_DEQUANT_2X4_START \
183 int a0 = dct[0] + dct[1]; \
184 int a1 = dct[2] + dct[3]; \
185 int a2 = dct[4] + dct[5]; \
186 int a3 = dct[6] + dct[7]; \
187 int a4 = dct[0] - dct[1]; \
188 int a5 = dct[2] - dct[3]; \
189 int a6 = dct[4] - dct[5]; \
190 int a7 = dct[6] - dct[7]; \
191 int b0 = a0 + a1; \
192 int b1 = a2 + a3; \
193 int b2 = a4 + a5; \
194 int b3 = a6 + a7; \
195 int b4 = a0 - a1; \
196 int b5 = a2 - a3; \
197 int b6 = a4 - a5; \
198 int b7 = a6 - a7;
199
idct_dequant_2x4_dc(dctcoef dct[8],dctcoef dct4x4[8][16],int dequant_mf[6][16],int i_qp)200 static void idct_dequant_2x4_dc( dctcoef dct[8], dctcoef dct4x4[8][16], int dequant_mf[6][16], int i_qp )
201 {
202 IDCT_DEQUANT_2X4_START
203 int dmf = dequant_mf[i_qp%6][0] << i_qp/6;
204 dct4x4[0][0] = ((b0 + b1) * dmf + 32) >> 6;
205 dct4x4[1][0] = ((b2 + b3) * dmf + 32) >> 6;
206 dct4x4[2][0] = ((b0 - b1) * dmf + 32) >> 6;
207 dct4x4[3][0] = ((b2 - b3) * dmf + 32) >> 6;
208 dct4x4[4][0] = ((b4 - b5) * dmf + 32) >> 6;
209 dct4x4[5][0] = ((b6 - b7) * dmf + 32) >> 6;
210 dct4x4[6][0] = ((b4 + b5) * dmf + 32) >> 6;
211 dct4x4[7][0] = ((b6 + b7) * dmf + 32) >> 6;
212 }
213
idct_dequant_2x4_dconly(dctcoef dct[8],int dequant_mf[6][16],int i_qp)214 static void idct_dequant_2x4_dconly( dctcoef dct[8], int dequant_mf[6][16], int i_qp )
215 {
216 IDCT_DEQUANT_2X4_START
217 int dmf = dequant_mf[i_qp%6][0] << i_qp/6;
218 dct[0] = ((b0 + b1) * dmf + 32) >> 6;
219 dct[1] = ((b2 + b3) * dmf + 32) >> 6;
220 dct[2] = ((b0 - b1) * dmf + 32) >> 6;
221 dct[3] = ((b2 - b3) * dmf + 32) >> 6;
222 dct[4] = ((b4 - b5) * dmf + 32) >> 6;
223 dct[5] = ((b6 - b7) * dmf + 32) >> 6;
224 dct[6] = ((b4 + b5) * dmf + 32) >> 6;
225 dct[7] = ((b6 + b7) * dmf + 32) >> 6;
226 }
227
optimize_chroma_idct_dequant_2x4(dctcoef out[8],dctcoef dct[8],int dmf)228 static ALWAYS_INLINE void optimize_chroma_idct_dequant_2x4( dctcoef out[8], dctcoef dct[8], int dmf )
229 {
230 IDCT_DEQUANT_2X4_START
231 out[0] = ((b0 + b1) * dmf + 2080) >> 6; /* 2080 = 32 + (32<<6) */
232 out[1] = ((b2 + b3) * dmf + 2080) >> 6;
233 out[2] = ((b0 - b1) * dmf + 2080) >> 6;
234 out[3] = ((b2 - b3) * dmf + 2080) >> 6;
235 out[4] = ((b4 - b5) * dmf + 2080) >> 6;
236 out[5] = ((b6 - b7) * dmf + 2080) >> 6;
237 out[6] = ((b4 + b5) * dmf + 2080) >> 6;
238 out[7] = ((b6 + b7) * dmf + 2080) >> 6;
239 }
240 #undef IDCT_DEQUANT_2X4_START
241
optimize_chroma_idct_dequant_2x2(dctcoef out[4],dctcoef dct[4],int dmf)242 static ALWAYS_INLINE void optimize_chroma_idct_dequant_2x2( dctcoef out[4], dctcoef dct[4], int dmf )
243 {
244 int d0 = dct[0] + dct[1];
245 int d1 = dct[2] + dct[3];
246 int d2 = dct[0] - dct[1];
247 int d3 = dct[2] - dct[3];
248 out[0] = ((d0 + d1) * dmf >> 5) + 32;
249 out[1] = ((d0 - d1) * dmf >> 5) + 32;
250 out[2] = ((d2 + d3) * dmf >> 5) + 32;
251 out[3] = ((d2 - d3) * dmf >> 5) + 32;
252 }
253
optimize_chroma_round(dctcoef * ref,dctcoef * dct,int dequant_mf,int chroma422)254 static ALWAYS_INLINE int optimize_chroma_round( dctcoef *ref, dctcoef *dct, int dequant_mf, int chroma422 )
255 {
256 dctcoef out[8];
257 int sum;
258 int i;
259
260 if( chroma422 )
261 optimize_chroma_idct_dequant_2x4( out, dct, dequant_mf );
262 else
263 optimize_chroma_idct_dequant_2x2( out, dct, dequant_mf );
264
265 sum = 0;
266 for( i = 0; i < (chroma422?8:4); i++ )
267 sum |= ref[i] ^ out[i];
268 return sum >> 6;
269 }
270
optimize_chroma_dc_internal(dctcoef * dct,int dequant_mf,int chroma422)271 static ALWAYS_INLINE int optimize_chroma_dc_internal( dctcoef *dct, int dequant_mf, int chroma422 )
272 {
273 /* dequant_mf = h->dequant4_mf[CQM_4IC + b_inter][i_qp%6][0] << i_qp/6, max 32*64 */
274 dctcoef dct_orig[8];
275 int coeff, nz;
276 int sum;
277 int i;
278
279 if( chroma422 )
280 optimize_chroma_idct_dequant_2x4( dct_orig, dct, dequant_mf );
281 else
282 optimize_chroma_idct_dequant_2x2( dct_orig, dct, dequant_mf );
283
284 /* If the DC coefficients already round to zero, terminate early. */
285 sum = 0;
286 for( i = 0; i < (chroma422?8:4); i++ )
287 sum |= dct_orig[i];
288 if( !(sum >> 6) )
289 return 0;
290
291 /* Start with the highest frequency coefficient... is this the best option? */
292 for( nz = 0, coeff = (chroma422?7:3); coeff >= 0; coeff-- )
293 {
294 int level = dct[coeff];
295 int sign = level>>31 | 1; /* dct[coeff] < 0 ? -1 : 1 */
296
297 while( level )
298 {
299 dct[coeff] = level - sign;
300 if( optimize_chroma_round( dct_orig, dct, dequant_mf, chroma422 ) )
301 {
302 nz = 1;
303 dct[coeff] = level;
304 break;
305 }
306 level -= sign;
307 }
308 }
309
310 return nz;
311 }
312
optimize_chroma_2x2_dc(dctcoef dct[4],int dequant_mf)313 static int optimize_chroma_2x2_dc( dctcoef dct[4], int dequant_mf )
314 {
315 return optimize_chroma_dc_internal( dct, dequant_mf, 0 );
316 }
317
optimize_chroma_2x4_dc(dctcoef dct[8],int dequant_mf)318 static int optimize_chroma_2x4_dc( dctcoef dct[8], int dequant_mf )
319 {
320 return optimize_chroma_dc_internal( dct, dequant_mf, 1 );
321 }
322
x264_denoise_dct(dctcoef * dct,uint32_t * sum,udctcoef * offset,int size)323 static void x264_denoise_dct( dctcoef *dct, uint32_t *sum, udctcoef *offset, int size )
324 {
325 int i;
326
327 for( i = 0; i < size; i++ )
328 {
329 int level = dct[i];
330 int sign = level>>31;
331 level = (level+sign)^sign;
332 sum[i] += level;
333 level -= offset[i];
334 dct[i] = level<0 ? 0 : (level^sign)-sign;
335 }
336 }
337
338 /* (ref: JVT-B118)
339 * x264_mb_decimate_score: given dct coeffs it returns a score to see if we could empty this dct coeffs
340 * to 0 (low score means set it to null)
341 * Used in inter macroblock (luma and chroma)
342 * luma: for a 8x8 block: if score < 4 -> null
343 * for the complete mb: if score < 6 -> null
344 * chroma: for the complete mb: if score < 7 -> null
345 */
346
347 const uint8_t x264_decimate_table4[16] =
348 {
349 3,2,2,1,1,1,0,0,0,0,0,0,0,0,0,0
350 };
351 const uint8_t x264_decimate_table8[64] =
352 {
353 3,3,3,3,2,2,2,2,2,2,2,2,1,1,1,1,
354 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,
355 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
356 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
357 };
358
x264_decimate_score_internal(dctcoef * dct,int i_max)359 static int ALWAYS_INLINE x264_decimate_score_internal( dctcoef *dct, int i_max )
360 {
361 const uint8_t *ds_table = (i_max == 64) ? x264_decimate_table8 : x264_decimate_table4;
362 int i_score = 0;
363 int idx = i_max - 1;
364
365 while( idx >= 0 && dct[idx] == 0 )
366 idx--;
367 while( idx >= 0 )
368 {
369 int i_run;
370
371 if( (unsigned)(dct[idx--] + 1) > 2 )
372 return 9;
373
374 i_run = 0;
375 while( idx >= 0 && dct[idx] == 0 )
376 {
377 idx--;
378 i_run++;
379 }
380 i_score += ds_table[i_run];
381 }
382
383 return i_score;
384 }
385
x264_decimate_score15(dctcoef * dct)386 static int x264_decimate_score15( dctcoef *dct )
387 {
388 return x264_decimate_score_internal( dct+1, 15 );
389 }
x264_decimate_score16(dctcoef * dct)390 static int x264_decimate_score16( dctcoef *dct )
391 {
392 return x264_decimate_score_internal( dct, 16 );
393 }
x264_decimate_score64(dctcoef * dct)394 static int x264_decimate_score64( dctcoef *dct )
395 {
396 return x264_decimate_score_internal( dct, 64 );
397 }
398
399 #define last(num)\
400 static int x264_coeff_last##num( dctcoef *l )\
401 {\
402 int i_last = num-1;\
403 while( i_last >= 0 && l[i_last] == 0 )\
404 i_last--;\
405 return i_last;\
406 }
407
408 last(4)
409 last(8)
410 last(15)
411 last(16)
412 last(64)
413
414 #define level_run(num)\
415 static int x264_coeff_level_run##num( dctcoef *dct, x264_run_level_t *runlevel )\
416 {\
417 int i_last = runlevel->last = x264_coeff_last##num(dct);\
418 int i_total = 0;\
419 int mask = 0;\
420 do\
421 {\
422 runlevel->level[i_total++] = dct[i_last];\
423 mask |= 1 << (i_last);\
424 while( --i_last >= 0 && dct[i_last] == 0 );\
425 } while( i_last >= 0 );\
426 runlevel->mask = mask;\
427 return i_total;\
428 }
429
430 level_run(4)
431 level_run(8)
432 level_run(15)
433 level_run(16)
434
435 #if ARCH_X86_64
436 #define INIT_TRELLIS(cpu)\
437 pf->trellis_cabac_4x4 = x264_trellis_cabac_4x4_##cpu;\
438 pf->trellis_cabac_8x8 = x264_trellis_cabac_8x8_##cpu;\
439 pf->trellis_cabac_4x4_psy = x264_trellis_cabac_4x4_psy_##cpu;\
440 pf->trellis_cabac_8x8_psy = x264_trellis_cabac_8x8_psy_##cpu;\
441 pf->trellis_cabac_dc = x264_trellis_cabac_dc_##cpu;\
442 pf->trellis_cabac_chroma_422_dc = x264_trellis_cabac_chroma_422_dc_##cpu;
443 #else
444 # define INIT_TRELLIS(...)
445 #endif
446
x264_quant_init(x264_t * h,int cpu,x264_quant_function_t * pf)447 void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf )
448 {
449 pf->quant_8x8 = quant_8x8;
450 pf->quant_4x4 = quant_4x4;
451 pf->quant_4x4x4 = quant_4x4x4;
452 pf->quant_4x4_dc = quant_4x4_dc;
453 pf->quant_2x2_dc = quant_2x2_dc;
454
455 pf->dequant_4x4 = dequant_4x4;
456 pf->dequant_4x4_dc = dequant_4x4_dc;
457 pf->dequant_8x8 = dequant_8x8;
458
459 pf->idct_dequant_2x4_dc = idct_dequant_2x4_dc;
460 pf->idct_dequant_2x4_dconly = idct_dequant_2x4_dconly;
461
462 pf->optimize_chroma_2x2_dc = optimize_chroma_2x2_dc;
463 pf->optimize_chroma_2x4_dc = optimize_chroma_2x4_dc;
464
465 pf->denoise_dct = x264_denoise_dct;
466 pf->decimate_score15 = x264_decimate_score15;
467 pf->decimate_score16 = x264_decimate_score16;
468 pf->decimate_score64 = x264_decimate_score64;
469
470 pf->coeff_last4 = x264_coeff_last4;
471 pf->coeff_last8 = x264_coeff_last8;
472 pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15;
473 pf->coeff_last[ DCT_LUMA_4x4] = x264_coeff_last16;
474 pf->coeff_last[ DCT_LUMA_8x8] = x264_coeff_last64;
475 pf->coeff_level_run4 = x264_coeff_level_run4;
476 pf->coeff_level_run8 = x264_coeff_level_run8;
477 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15;
478 pf->coeff_level_run[ DCT_LUMA_4x4] = x264_coeff_level_run16;
479
480 #if HIGH_BIT_DEPTH
481 #if HAVE_MMX
482 INIT_TRELLIS( sse2 );
483 if( cpu&X264_CPU_MMX2 )
484 {
485 #if ARCH_X86
486 pf->denoise_dct = x264_denoise_dct_mmx;
487 pf->decimate_score15 = x264_decimate_score15_mmx2;
488 pf->decimate_score16 = x264_decimate_score16_mmx2;
489 pf->decimate_score64 = x264_decimate_score64_mmx2;
490 pf->coeff_last8 = x264_coeff_last8_mmx2;
491 pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_mmx2;
492 pf->coeff_last[ DCT_LUMA_4x4] = x264_coeff_last16_mmx2;
493 pf->coeff_last[ DCT_LUMA_8x8] = x264_coeff_last64_mmx2;
494 pf->coeff_level_run8 = x264_coeff_level_run8_mmx2;
495 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_mmx2;
496 pf->coeff_level_run[ DCT_LUMA_4x4] = x264_coeff_level_run16_mmx2;
497 #endif
498 pf->coeff_last4 = x264_coeff_last4_mmx2;
499 pf->coeff_level_run4 = x264_coeff_level_run4_mmx2;
500 if( cpu&X264_CPU_LZCNT )
501 pf->coeff_level_run4 = x264_coeff_level_run4_mmx2_lzcnt;
502 }
503 if( cpu&X264_CPU_SSE2 )
504 {
505 pf->quant_4x4 = x264_quant_4x4_sse2;
506 pf->quant_4x4x4 = x264_quant_4x4x4_sse2;
507 pf->quant_8x8 = x264_quant_8x8_sse2;
508 pf->quant_2x2_dc = x264_quant_2x2_dc_sse2;
509 pf->quant_4x4_dc = x264_quant_4x4_dc_sse2;
510 pf->dequant_4x4 = x264_dequant_4x4_sse2;
511 pf->dequant_8x8 = x264_dequant_8x8_sse2;
512 pf->dequant_4x4_dc = x264_dequant_4x4dc_sse2;
513 pf->denoise_dct = x264_denoise_dct_sse2;
514 pf->decimate_score15 = x264_decimate_score15_sse2;
515 pf->decimate_score16 = x264_decimate_score16_sse2;
516 pf->decimate_score64 = x264_decimate_score64_sse2;
517 pf->coeff_last8 = x264_coeff_last8_sse2;
518 pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_sse2;
519 pf->coeff_last[DCT_LUMA_4x4] = x264_coeff_last16_sse2;
520 pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_sse2;
521 pf->coeff_level_run8 = x264_coeff_level_run8_sse2;
522 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_sse2;
523 pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_sse2;
524 if( cpu&X264_CPU_LZCNT )
525 {
526 pf->coeff_last4 = x264_coeff_last4_mmx2_lzcnt;
527 pf->coeff_last8 = x264_coeff_last8_sse2_lzcnt;
528 pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_sse2_lzcnt;
529 pf->coeff_last[DCT_LUMA_4x4] = x264_coeff_last16_sse2_lzcnt;
530 pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_sse2_lzcnt;
531 pf->coeff_level_run8 = x264_coeff_level_run8_sse2_lzcnt;
532 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_sse2_lzcnt;
533 pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_sse2_lzcnt;
534 }
535 }
536 if( cpu&X264_CPU_SSSE3 )
537 {
538 pf->quant_4x4 = x264_quant_4x4_ssse3;
539 pf->quant_4x4x4 = x264_quant_4x4x4_ssse3;
540 pf->quant_8x8 = x264_quant_8x8_ssse3;
541 pf->quant_2x2_dc = x264_quant_2x2_dc_ssse3;
542 pf->quant_4x4_dc = x264_quant_4x4_dc_ssse3;
543 pf->denoise_dct = x264_denoise_dct_ssse3;
544 pf->decimate_score15 = x264_decimate_score15_ssse3;
545 pf->decimate_score16 = x264_decimate_score16_ssse3;
546 pf->decimate_score64 = x264_decimate_score64_ssse3;
547 INIT_TRELLIS( ssse3 );
548 }
549 if( cpu&X264_CPU_SSE4 )
550 {
551 pf->quant_2x2_dc = x264_quant_2x2_dc_sse4;
552 pf->quant_4x4_dc = x264_quant_4x4_dc_sse4;
553 pf->quant_4x4 = x264_quant_4x4_sse4;
554 pf->quant_4x4x4 = x264_quant_4x4x4_sse4;
555 pf->quant_8x8 = x264_quant_8x8_sse4;
556 }
557 if( cpu&X264_CPU_AVX )
558 {
559 pf->denoise_dct = x264_denoise_dct_avx;
560 }
561 if( cpu&X264_CPU_XOP )
562 {
563 pf->dequant_4x4_dc = x264_dequant_4x4dc_xop;
564 if( h->param.i_cqm_preset != X264_CQM_FLAT )
565 {
566 pf->dequant_4x4 = x264_dequant_4x4_xop;
567 pf->dequant_8x8 = x264_dequant_8x8_xop;
568 }
569 }
570 if( cpu&X264_CPU_AVX2 )
571 {
572 pf->quant_4x4 = x264_quant_4x4_avx2;
573 pf->quant_4x4_dc = x264_quant_4x4_dc_avx2;
574 pf->quant_8x8 = x264_quant_8x8_avx2;
575 pf->quant_4x4x4 = x264_quant_4x4x4_avx2;
576 pf->dequant_4x4 = x264_dequant_4x4_avx2;
577 pf->dequant_8x8 = x264_dequant_8x8_avx2;
578 pf->dequant_4x4_dc = x264_dequant_4x4dc_avx2;
579 pf->denoise_dct = x264_denoise_dct_avx2;
580 }
581 #endif // HAVE_MMX
582 #else // !HIGH_BIT_DEPTH
583 #if HAVE_MMX
584 INIT_TRELLIS( sse2 );
585 if( cpu&X264_CPU_MMX )
586 {
587 #if ARCH_X86
588 pf->quant_4x4 = x264_quant_4x4_mmx;
589 pf->quant_8x8 = x264_quant_8x8_mmx;
590 pf->dequant_4x4 = x264_dequant_4x4_mmx;
591 pf->dequant_4x4_dc = x264_dequant_4x4dc_mmx2;
592 pf->dequant_8x8 = x264_dequant_8x8_mmx;
593 if( h->param.i_cqm_preset == X264_CQM_FLAT )
594 {
595 pf->dequant_4x4 = x264_dequant_4x4_flat16_mmx;
596 pf->dequant_8x8 = x264_dequant_8x8_flat16_mmx;
597 }
598 pf->denoise_dct = x264_denoise_dct_mmx;
599 #endif
600 }
601
602 if( cpu&X264_CPU_MMX2 )
603 {
604 pf->quant_2x2_dc = x264_quant_2x2_dc_mmx2;
605 #if ARCH_X86
606 pf->quant_4x4_dc = x264_quant_4x4_dc_mmx2;
607 pf->decimate_score15 = x264_decimate_score15_mmx2;
608 pf->decimate_score16 = x264_decimate_score16_mmx2;
609 pf->decimate_score64 = x264_decimate_score64_mmx2;
610 pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_mmx2;
611 pf->coeff_last[ DCT_LUMA_4x4] = x264_coeff_last16_mmx2;
612 pf->coeff_last[ DCT_LUMA_8x8] = x264_coeff_last64_mmx2;
613 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_mmx2;
614 pf->coeff_level_run[ DCT_LUMA_4x4] = x264_coeff_level_run16_mmx2;
615 #endif
616 pf->coeff_last4 = x264_coeff_last4_mmx2;
617 pf->coeff_last8 = x264_coeff_last8_mmx2;
618 pf->coeff_level_run4 = x264_coeff_level_run4_mmx2;
619 pf->coeff_level_run8 = x264_coeff_level_run8_mmx2;
620 if( cpu&X264_CPU_LZCNT )
621 {
622 pf->coeff_last4 = x264_coeff_last4_mmx2_lzcnt;
623 pf->coeff_last8 = x264_coeff_last8_mmx2_lzcnt;
624 pf->coeff_level_run4 = x264_coeff_level_run4_mmx2_lzcnt;
625 pf->coeff_level_run8 = x264_coeff_level_run8_mmx2_lzcnt;
626 }
627 }
628
629 if( cpu&X264_CPU_SSE2 )
630 {
631 pf->quant_4x4_dc = x264_quant_4x4_dc_sse2;
632 pf->quant_4x4 = x264_quant_4x4_sse2;
633 pf->quant_4x4x4 = x264_quant_4x4x4_sse2;
634 pf->quant_8x8 = x264_quant_8x8_sse2;
635 pf->dequant_4x4 = x264_dequant_4x4_sse2;
636 pf->dequant_4x4_dc = x264_dequant_4x4dc_sse2;
637 pf->dequant_8x8 = x264_dequant_8x8_sse2;
638 if( h->param.i_cqm_preset == X264_CQM_FLAT )
639 {
640 pf->dequant_4x4 = x264_dequant_4x4_flat16_sse2;
641 pf->dequant_8x8 = x264_dequant_8x8_flat16_sse2;
642 }
643 pf->optimize_chroma_2x2_dc = x264_optimize_chroma_2x2_dc_sse2;
644 pf->denoise_dct = x264_denoise_dct_sse2;
645 pf->decimate_score15 = x264_decimate_score15_sse2;
646 pf->decimate_score16 = x264_decimate_score16_sse2;
647 pf->decimate_score64 = x264_decimate_score64_sse2;
648 pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_sse2;
649 pf->coeff_last[DCT_LUMA_4x4] = x264_coeff_last16_sse2;
650 pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_sse2;
651 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_sse2;
652 pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_sse2;
653 if( cpu&X264_CPU_LZCNT )
654 {
655 pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_sse2_lzcnt;
656 pf->coeff_last[DCT_LUMA_4x4] = x264_coeff_last16_sse2_lzcnt;
657 pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_sse2_lzcnt;
658 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_sse2_lzcnt;
659 pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_sse2_lzcnt;
660 }
661 }
662
663 if( cpu&X264_CPU_SSSE3 )
664 {
665 pf->quant_2x2_dc = x264_quant_2x2_dc_ssse3;
666 pf->quant_4x4_dc = x264_quant_4x4_dc_ssse3;
667 pf->quant_4x4 = x264_quant_4x4_ssse3;
668 pf->quant_4x4x4 = x264_quant_4x4x4_ssse3;
669 pf->quant_8x8 = x264_quant_8x8_ssse3;
670 pf->optimize_chroma_2x2_dc = x264_optimize_chroma_2x2_dc_ssse3;
671 pf->denoise_dct = x264_denoise_dct_ssse3;
672 pf->decimate_score15 = x264_decimate_score15_ssse3;
673 pf->decimate_score16 = x264_decimate_score16_ssse3;
674 pf->decimate_score64 = x264_decimate_score64_ssse3;
675 INIT_TRELLIS( ssse3 );
676 pf->coeff_level_run4 = x264_coeff_level_run4_ssse3;
677 pf->coeff_level_run8 = x264_coeff_level_run8_ssse3;
678 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_ssse3;
679 pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_ssse3;
680 if( cpu&X264_CPU_LZCNT )
681 {
682 pf->coeff_level_run4 = x264_coeff_level_run4_ssse3;
683 pf->coeff_level_run8 = x264_coeff_level_run8_ssse3;
684 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_ssse3_lzcnt;
685 pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_ssse3_lzcnt;
686 }
687 }
688
689 if( cpu&X264_CPU_SSE4 )
690 {
691 pf->quant_4x4_dc = x264_quant_4x4_dc_sse4;
692 pf->quant_4x4 = x264_quant_4x4_sse4;
693 pf->quant_8x8 = x264_quant_8x8_sse4;
694 pf->optimize_chroma_2x2_dc = x264_optimize_chroma_2x2_dc_sse4;
695 }
696
697 if( cpu&X264_CPU_AVX )
698 {
699 pf->dequant_4x4_dc = x264_dequant_4x4dc_avx;
700 if( h->param.i_cqm_preset != X264_CQM_FLAT )
701 {
702 pf->dequant_4x4 = x264_dequant_4x4_avx;
703 pf->dequant_8x8 = x264_dequant_8x8_avx;
704 }
705 pf->optimize_chroma_2x2_dc = x264_optimize_chroma_2x2_dc_avx;
706 pf->denoise_dct = x264_denoise_dct_avx;
707 }
708
709 if( cpu&X264_CPU_XOP )
710 {
711 if( h->param.i_cqm_preset != X264_CQM_FLAT )
712 {
713 pf->dequant_4x4 = x264_dequant_4x4_xop;
714 pf->dequant_8x8 = x264_dequant_8x8_xop;
715 }
716 }
717
718 if( cpu&X264_CPU_AVX2 )
719 {
720 pf->quant_4x4 = x264_quant_4x4_avx2;
721 pf->quant_4x4_dc = x264_quant_4x4_dc_avx2;
722 pf->quant_8x8 = x264_quant_8x8_avx2;
723 pf->quant_4x4x4 = x264_quant_4x4x4_avx2;
724 pf->dequant_4x4 = x264_dequant_4x4_avx2;
725 pf->dequant_8x8 = x264_dequant_8x8_avx2;
726 pf->dequant_4x4_dc = x264_dequant_4x4dc_avx2;
727 if( h->param.i_cqm_preset == X264_CQM_FLAT )
728 {
729 pf->dequant_4x4 = x264_dequant_4x4_flat16_avx2;
730 pf->dequant_8x8 = x264_dequant_8x8_flat16_avx2;
731 }
732 pf->decimate_score64 = x264_decimate_score64_avx2;
733 pf->denoise_dct = x264_denoise_dct_avx2;
734 if( cpu&X264_CPU_LZCNT )
735 {
736 pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_avx2_lzcnt;
737 pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_avx2_lzcnt;
738 pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_avx2_lzcnt;
739 }
740 }
741 #endif // HAVE_MMX
742
743 #if HAVE_ALTIVEC
744 if( cpu&X264_CPU_ALTIVEC ) {
745 pf->quant_2x2_dc = x264_quant_2x2_dc_altivec;
746 pf->quant_4x4_dc = x264_quant_4x4_dc_altivec;
747 pf->quant_4x4 = x264_quant_4x4_altivec;
748 pf->quant_8x8 = x264_quant_8x8_altivec;
749
750 pf->dequant_4x4 = x264_dequant_4x4_altivec;
751 pf->dequant_8x8 = x264_dequant_8x8_altivec;
752 }
753 #endif
754
755 #if HAVE_ARMV6
756 if( cpu&X264_CPU_ARMV6 )
757 {
758 pf->coeff_last4 = x264_coeff_last4_arm;
759 pf->coeff_last8 = x264_coeff_last8_arm;
760 }
761 #endif
762 #if HAVE_ARMV6 || ARCH_AARCH64
763 if( cpu&X264_CPU_NEON )
764 {
765 pf->quant_2x2_dc = x264_quant_2x2_dc_neon;
766 pf->quant_4x4 = x264_quant_4x4_neon;
767 pf->quant_4x4_dc = x264_quant_4x4_dc_neon;
768 pf->quant_4x4x4 = x264_quant_4x4x4_neon;
769 pf->quant_8x8 = x264_quant_8x8_neon;
770 pf->dequant_4x4 = x264_dequant_4x4_neon;
771 pf->dequant_4x4_dc = x264_dequant_4x4_dc_neon;
772 pf->dequant_8x8 = x264_dequant_8x8_neon;
773 pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_neon;
774 pf->coeff_last[DCT_LUMA_4x4] = x264_coeff_last16_neon;
775 pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_neon;
776 }
777 #endif
778 #if ARCH_AARCH64
779 if( cpu&X264_CPU_ARMV8 )
780 {
781 pf->coeff_last4 = x264_coeff_last4_aarch64;
782 pf->coeff_last8 = x264_coeff_last8_aarch64;
783 }
784 #endif
785 #endif // HIGH_BIT_DEPTH
786 pf->coeff_last[DCT_LUMA_DC] = pf->coeff_last[DCT_CHROMAU_DC] = pf->coeff_last[DCT_CHROMAV_DC] =
787 pf->coeff_last[DCT_CHROMAU_4x4] = pf->coeff_last[DCT_CHROMAV_4x4] = pf->coeff_last[DCT_LUMA_4x4];
788 pf->coeff_last[DCT_CHROMA_AC] = pf->coeff_last[DCT_CHROMAU_AC] =
789 pf->coeff_last[DCT_CHROMAV_AC] = pf->coeff_last[DCT_LUMA_AC];
790 pf->coeff_last[DCT_CHROMAU_8x8] = pf->coeff_last[DCT_CHROMAV_8x8] = pf->coeff_last[DCT_LUMA_8x8];
791
792 pf->coeff_level_run[DCT_LUMA_DC] = pf->coeff_level_run[DCT_CHROMAU_DC] = pf->coeff_level_run[DCT_CHROMAV_DC] =
793 pf->coeff_level_run[DCT_CHROMAU_4x4] = pf->coeff_level_run[DCT_CHROMAV_4x4] = pf->coeff_level_run[DCT_LUMA_4x4];
794 pf->coeff_level_run[DCT_CHROMA_AC] = pf->coeff_level_run[DCT_CHROMAU_AC] =
795 pf->coeff_level_run[DCT_CHROMAV_AC] = pf->coeff_level_run[DCT_LUMA_AC];
796 }
797