1 /*****************************************************************************
2 * quant.c: h264 encoder library
3 *****************************************************************************
4 * Copyright (C) 2005-2008 x264 project
6 * Authors: Loren Merritt <lorenm@u.washington.edu>
7 * Christian Heine <sennindemokrit@gmx.net>
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 *****************************************************************************/
27 #include "x86/quant.h"
30 # include "ppc/quant.h"
33 #define QUANT_ONE( coef, mf, f ) \
36 (coef) = (f + (coef)) * (mf) >> 16; \
38 (coef) = - ((f - (coef)) * (mf) >> 16); \
41 static void quant_8x8( int16_t dct[8][8], uint16_t mf[64], uint16_t bias[64] )
44 for( i = 0; i < 64; i++ )
45 QUANT_ONE( dct[0][i], mf[i], bias[i] );
48 static void quant_4x4( int16_t dct[4][4], uint16_t mf[16], uint16_t bias[16] )
51 for( i = 0; i < 16; i++ )
52 QUANT_ONE( dct[0][i], mf[i], bias[i] );
55 static void quant_4x4_dc( int16_t dct[4][4], int mf, int bias )
58 for( i = 0; i < 16; i++ )
59 QUANT_ONE( dct[0][i], mf, bias );
62 static void quant_2x2_dc( int16_t dct[2][2], int mf, int bias )
64 QUANT_ONE( dct[0][0], mf, bias );
65 QUANT_ONE( dct[0][1], mf, bias );
66 QUANT_ONE( dct[0][2], mf, bias );
67 QUANT_ONE( dct[0][3], mf, bias );
70 #define DEQUANT_SHL( x ) \
71 dct[y][x] = ( dct[y][x] * dequant_mf[i_mf][y][x] ) << i_qbits
73 #define DEQUANT_SHR( x ) \
74 dct[y][x] = ( dct[y][x] * dequant_mf[i_mf][y][x] + f ) >> (-i_qbits)
76 static void dequant_4x4( int16_t dct[4][4], int dequant_mf[6][4][4], int i_qp )
78 const int i_mf = i_qp%6;
79 const int i_qbits = i_qp/6 - 4;
84 for( y = 0; y < 4; y++ )
94 const int f = 1 << (-i_qbits-1);
95 for( y = 0; y < 4; y++ )
105 static void dequant_8x8( int16_t dct[8][8], int dequant_mf[6][8][8], int i_qp )
107 const int i_mf = i_qp%6;
108 const int i_qbits = i_qp/6 - 6;
113 for( y = 0; y < 8; y++ )
127 const int f = 1 << (-i_qbits-1);
128 for( y = 0; y < 8; y++ )
142 void x264_mb_dequant_2x2_dc( int16_t dct[2][2], int dequant_mf[6][4][4], int i_qp )
144 const int i_qbits = i_qp/6 - 5;
148 const int i_dmf = dequant_mf[i_qp%6][0][0] << i_qbits;
156 const int i_dmf = dequant_mf[i_qp%6][0][0];
157 // chroma DC is truncated, not rounded
158 dct[0][0] = ( dct[0][0] * i_dmf ) >> (-i_qbits);
159 dct[0][1] = ( dct[0][1] * i_dmf ) >> (-i_qbits);
160 dct[1][0] = ( dct[1][0] * i_dmf ) >> (-i_qbits);
161 dct[1][1] = ( dct[1][1] * i_dmf ) >> (-i_qbits);
165 void x264_mb_dequant_4x4_dc( int16_t dct[4][4], int dequant_mf[6][4][4], int i_qp )
167 const int i_qbits = i_qp/6 - 6;
172 const int i_dmf = dequant_mf[i_qp%6][0][0] << i_qbits;
174 for( y = 0; y < 4; y++ )
184 const int i_dmf = dequant_mf[i_qp%6][0][0];
185 const int f = 1 << (-i_qbits-1);
187 for( y = 0; y < 4; y++ )
189 dct[y][0] = ( dct[y][0] * i_dmf + f ) >> (-i_qbits);
190 dct[y][1] = ( dct[y][1] * i_dmf + f ) >> (-i_qbits);
191 dct[y][2] = ( dct[y][2] * i_dmf + f ) >> (-i_qbits);
192 dct[y][3] = ( dct[y][3] * i_dmf + f ) >> (-i_qbits);
197 static void x264_denoise_dct( int16_t *dct, uint32_t *sum, uint16_t *offset, int size )
200 for( i=1; i<size; i++ )
203 int sign = level>>15;
204 level = (level+sign)^sign;
207 dct[i] = level<0 ? 0 : (level^sign)-sign;
212 * x264_mb_decimate_score: given dct coeffs it returns a score to see if we could empty this dct coeffs
213 * to 0 (low score means set it to null)
214 * Used in inter macroblock (luma and chroma)
215 * luma: for a 8x8 block: if score < 4 -> null
216 * for the complete mb: if score < 6 -> null
217 * chroma: for the complete mb: if score < 7 -> null
220 const uint8_t x264_decimate_table4[16] = {
221 3,2,2,1,1,1,0,0,0,0,0,0,0,0,0,0 };
222 const uint8_t x264_decimate_table8[64] = {
223 3,3,3,3,2,2,2,2,2,2,2,2,1,1,1,1,
224 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,
225 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
226 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0 };
228 static int ALWAYS_INLINE x264_decimate_score_internal( int16_t *dct, int i_max )
230 const uint8_t *ds_table = (i_max == 64) ? x264_decimate_table8 : x264_decimate_table4;
234 /* Yes, dct[idx-1] is guaranteed to be 32-bit aligned. idx>=0 instead of 1 works correctly for the same reason */
235 while( idx >= 0 && *(uint32_t*)&dct[idx-1] == 0 )
237 if( idx >= 0 && dct[idx] == 0 )
243 if( (unsigned)(dct[idx--] + 1) > 2 )
247 while( idx >= 0 && dct[idx] == 0 )
252 i_score += ds_table[i_run];
258 static int x264_decimate_score15( int16_t *dct )
260 return x264_decimate_score_internal( dct+1, 15 );
262 static int x264_decimate_score16( int16_t *dct )
264 return x264_decimate_score_internal( dct, 16 );
266 static int x264_decimate_score64( int16_t *dct )
268 return x264_decimate_score_internal( dct, 64 );
271 void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf )
273 pf->quant_8x8 = quant_8x8;
274 pf->quant_4x4 = quant_4x4;
275 pf->quant_4x4_dc = quant_4x4_dc;
276 pf->quant_2x2_dc = quant_2x2_dc;
278 pf->dequant_4x4 = dequant_4x4;
279 pf->dequant_8x8 = dequant_8x8;
281 pf->denoise_dct = x264_denoise_dct;
282 pf->decimate_score15 = x264_decimate_score15;
283 pf->decimate_score16 = x264_decimate_score16;
284 pf->decimate_score64 = x264_decimate_score64;
287 if( cpu&X264_CPU_MMX )
290 pf->quant_4x4 = x264_quant_4x4_mmx;
291 pf->quant_8x8 = x264_quant_8x8_mmx;
292 pf->dequant_4x4 = x264_dequant_4x4_mmx;
293 pf->dequant_8x8 = x264_dequant_8x8_mmx;
294 if( h->param.i_cqm_preset == X264_CQM_FLAT )
296 pf->dequant_4x4 = x264_dequant_4x4_flat16_mmx;
297 pf->dequant_8x8 = x264_dequant_8x8_flat16_mmx;
299 pf->denoise_dct = x264_denoise_dct_mmx;
303 if( cpu&X264_CPU_MMXEXT )
305 pf->quant_2x2_dc = x264_quant_2x2_dc_mmxext;
307 pf->quant_4x4_dc = x264_quant_4x4_dc_mmxext;
308 pf->decimate_score15 = x264_decimate_score15_mmxext;
309 pf->decimate_score16 = x264_decimate_score16_mmxext;
310 pf->decimate_score64 = x264_decimate_score64_mmxext;
314 if( cpu&X264_CPU_SSE2 )
316 pf->quant_4x4_dc = x264_quant_4x4_dc_sse2;
317 pf->quant_4x4 = x264_quant_4x4_sse2;
318 pf->quant_8x8 = x264_quant_8x8_sse2;
319 pf->dequant_4x4 = x264_dequant_4x4_sse2;
320 pf->dequant_8x8 = x264_dequant_8x8_sse2;
321 if( h->param.i_cqm_preset == X264_CQM_FLAT )
323 pf->dequant_4x4 = x264_dequant_4x4_flat16_sse2;
324 pf->dequant_8x8 = x264_dequant_8x8_flat16_sse2;
326 pf->denoise_dct = x264_denoise_dct_sse2;
327 pf->decimate_score15 = x264_decimate_score15_sse2;
328 pf->decimate_score16 = x264_decimate_score16_sse2;
329 pf->decimate_score64 = x264_decimate_score64_sse2;
332 if( cpu&X264_CPU_SSSE3 )
334 pf->quant_2x2_dc = x264_quant_2x2_dc_ssse3;
335 pf->quant_4x4_dc = x264_quant_4x4_dc_ssse3;
336 pf->quant_4x4 = x264_quant_4x4_ssse3;
337 pf->quant_8x8 = x264_quant_8x8_ssse3;
338 pf->denoise_dct = x264_denoise_dct_ssse3;
339 pf->decimate_score15 = x264_decimate_score15_ssse3;
340 pf->decimate_score16 = x264_decimate_score16_ssse3;
341 pf->decimate_score64 = x264_decimate_score64_ssse3;
346 if( cpu&X264_CPU_ALTIVEC ) {
347 pf->quant_2x2_dc = x264_quant_2x2_dc_altivec;
348 pf->quant_4x4_dc = x264_quant_4x4_dc_altivec;
349 pf->quant_4x4 = x264_quant_4x4_altivec;
350 pf->quant_8x8 = x264_quant_8x8_altivec;
352 pf->dequant_4x4 = x264_dequant_4x4_altivec;
353 pf->dequant_8x8 = x264_dequant_8x8_altivec;