X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=common%2Fquant.c;h=cc4ea8649da84894b4973fb4901822d41843d3d8;hb=5eccc7d2263ffd63921b95b9e95152ffb6b3645c;hp=a7b72cfbbb56912757c15887bacbe29104ff7279;hpb=c91f43a4b09dab84953f417e6d6662ec0fa7acb1;p=x264 diff --git a/common/quant.c b/common/quant.c index a7b72cfb..cc4ea864 100644 --- a/common/quant.c +++ b/common/quant.c @@ -1,9 +1,10 @@ /***************************************************************************** - * quant.c: h264 encoder library + * quant.c: quantization and level-run ***************************************************************************** - * Copyright (C) 2005-2008 x264 project + * Copyright (C) 2005-2011 x264 project * * Authors: Loren Merritt + * Fiona Glaser * Christian Heine * * This program is free software; you can redistribute it and/or modify @@ -19,6 +20,9 @@ * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. + * + * This program is also available under a commercial proprietary license. + * For more information, contact us at licensing@x264.com. *****************************************************************************/ #include "common.h" @@ -42,7 +46,7 @@ nz |= (coef); \ } -static int quant_8x8( dctcoef dct[64], uint16_t mf[64], uint16_t bias[64] ) +static int quant_8x8( dctcoef dct[64], udctcoef mf[64], udctcoef bias[64] ) { int nz = 0; for( int i = 0; i < 64; i++ ) @@ -50,7 +54,7 @@ static int quant_8x8( dctcoef dct[64], uint16_t mf[64], uint16_t bias[64] ) return !!nz; } -static int quant_4x4( dctcoef dct[16], uint16_t mf[16], uint16_t bias[16] ) +static int quant_4x4( dctcoef dct[16], udctcoef mf[16], udctcoef bias[16] ) { int nz = 0; for( int i = 0; i < 16; i++ ) @@ -137,9 +141,69 @@ static void dequant_4x4_dc( dctcoef dct[16], int dequant_mf[6][16], int i_qp ) } } -static void x264_denoise_dct( dctcoef *dct, uint32_t *sum, uint16_t *offset, int size ) +static ALWAYS_INLINE void idct_dequant_2x2_dconly( dctcoef out[4], dctcoef dct[4], int dequant_mf ) +{ + int d0 = dct[0] + dct[1]; + int d1 = dct[2] + dct[3]; + int d2 = dct[0] - dct[1]; + int d3 = dct[2] - dct[3]; + out[0] = (d0 + d1) * dequant_mf >> 5; + out[1] = (d0 - d1) * dequant_mf >> 5; + out[2] = (d2 + d3) * dequant_mf >> 5; + out[3] = (d2 - d3) * dequant_mf >> 5; +} + +static ALWAYS_INLINE int idct_dequant_round_2x2_dc( dctcoef ref[4], dctcoef dct[4], int dequant_mf ) +{ + dctcoef out[4]; + idct_dequant_2x2_dconly( out, dct, dequant_mf ); + return ((ref[0] ^ (out[0]+32)) + | (ref[1] ^ (out[1]+32)) + | (ref[2] ^ (out[2]+32)) + | (ref[3] ^ (out[3]+32))) >> 6; +} + +static int optimize_chroma_dc( dctcoef dct[4], int dequant_mf ) { - for( int i = 1; i < size; i++ ) + /* dequant_mf = h->dequant4_mf[CQM_4IC + b_inter][i_qp%6][0] << i_qp/6, max 32*64 */ + dctcoef dct_orig[4]; + int coeff, nz; + + idct_dequant_2x2_dconly( dct_orig, dct, dequant_mf ); + dct_orig[0] += 32; + dct_orig[1] += 32; + dct_orig[2] += 32; + dct_orig[3] += 32; + + /* If the DC coefficients already round to zero, terminate early. */ + if( !((dct_orig[0]|dct_orig[1]|dct_orig[2]|dct_orig[3])>>6) ) + return 0; + + /* Start with the highest frequency coefficient... is this the best option? */ + for( nz = 0, coeff = 3; coeff >= 0; coeff-- ) + { + int level = dct[coeff]; + int sign = level>>31 | 1; /* dct2x2[coeff] < 0 ? -1 : 1 */ + + while( level ) + { + dct[coeff] = level - sign; + if( idct_dequant_round_2x2_dc( dct_orig, dct, dequant_mf ) ) + { + nz = 1; + dct[coeff] = level; + break; + } + level -= sign; + } + } + + return nz; +} + +static void x264_denoise_dct( dctcoef *dct, uint32_t *sum, udctcoef *offset, int size ) +{ + for( int i = 0; i < size; i++ ) { int level = dct[i]; int sign = level>>31; @@ -268,6 +332,8 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf ) pf->dequant_4x4_dc = dequant_4x4_dc; pf->dequant_8x8 = dequant_8x8; + pf->optimize_chroma_dc = optimize_chroma_dc; + pf->denoise_dct = x264_denoise_dct; pf->decimate_score15 = x264_decimate_score15; pf->decimate_score16 = x264_decimate_score16; @@ -281,7 +347,89 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf ) pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15; pf->coeff_level_run[ DCT_LUMA_4x4] = x264_coeff_level_run16; -#if !X264_HIGH_BIT_DEPTH +#if HIGH_BIT_DEPTH +#if HAVE_MMX + if( cpu&X264_CPU_MMXEXT ) + { +#if ARCH_X86 + pf->denoise_dct = x264_denoise_dct_mmx; + pf->decimate_score15 = x264_decimate_score15_mmxext; + pf->decimate_score16 = x264_decimate_score16_mmxext; + if( cpu&X264_CPU_SLOW_CTZ ) + { + pf->decimate_score15 = x264_decimate_score15_mmxext_slowctz; + pf->decimate_score16 = x264_decimate_score16_mmxext_slowctz; + } + pf->decimate_score64 = x264_decimate_score64_mmxext; + pf->coeff_last[DCT_CHROMA_DC] = x264_coeff_last4_mmxext; + pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_mmxext; + pf->coeff_last[ DCT_LUMA_4x4] = x264_coeff_last16_mmxext; + pf->coeff_last[ DCT_LUMA_8x8] = x264_coeff_last64_mmxext; + pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_mmxext; + pf->coeff_level_run[ DCT_LUMA_4x4] = x264_coeff_level_run16_mmxext; +#endif + pf->coeff_level_run[DCT_CHROMA_DC] = x264_coeff_level_run4_mmxext; + if( cpu&X264_CPU_LZCNT ) + pf->coeff_level_run[DCT_CHROMA_DC] = x264_coeff_level_run4_mmxext_lzcnt; + } + if( cpu&X264_CPU_SSE2 ) + { + pf->quant_4x4 = x264_quant_4x4_sse2; + pf->quant_8x8 = x264_quant_8x8_sse2; + pf->quant_2x2_dc = x264_quant_2x2_dc_sse2; + pf->quant_4x4_dc = x264_quant_4x4_dc_sse2; + pf->dequant_4x4 = x264_dequant_4x4_sse2; + pf->dequant_8x8 = x264_dequant_8x8_sse2; + pf->dequant_4x4_dc = x264_dequant_4x4dc_sse2; + pf->denoise_dct = x264_denoise_dct_sse2; + pf->decimate_score15 = x264_decimate_score15_sse2; + pf->decimate_score16 = x264_decimate_score16_sse2; + pf->decimate_score64 = x264_decimate_score64_sse2; + if( cpu&X264_CPU_SLOW_CTZ ) + { + pf->decimate_score15 = x264_decimate_score15_sse2_slowctz; + pf->decimate_score16 = x264_decimate_score16_sse2_slowctz; + } + pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_sse2; + pf->coeff_last[DCT_LUMA_4x4] = x264_coeff_last16_sse2; + pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_sse2; + pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_sse2; + pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_sse2; + if( cpu&X264_CPU_LZCNT ) + { + pf->coeff_last[DCT_CHROMA_DC] = x264_coeff_last4_mmxext_lzcnt; + pf->coeff_last[ DCT_LUMA_AC] = x264_coeff_last15_sse2_lzcnt; + pf->coeff_last[DCT_LUMA_4x4] = x264_coeff_last16_sse2_lzcnt; + pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_sse2_lzcnt; + pf->coeff_level_run[ DCT_LUMA_AC] = x264_coeff_level_run15_sse2_lzcnt; + pf->coeff_level_run[DCT_LUMA_4x4] = x264_coeff_level_run16_sse2_lzcnt; + } + } + if( cpu&X264_CPU_SSSE3 ) + { + pf->quant_4x4 = x264_quant_4x4_ssse3; + pf->quant_8x8 = x264_quant_8x8_ssse3; + pf->quant_2x2_dc = x264_quant_2x2_dc_ssse3; + pf->quant_4x4_dc = x264_quant_4x4_dc_ssse3; + pf->denoise_dct = x264_denoise_dct_ssse3; + pf->decimate_score15 = x264_decimate_score15_ssse3; + pf->decimate_score16 = x264_decimate_score16_ssse3; + if( cpu&X264_CPU_SLOW_CTZ ) + { + pf->decimate_score15 = x264_decimate_score15_ssse3_slowctz; + pf->decimate_score16 = x264_decimate_score16_ssse3_slowctz; + } + pf->decimate_score64 = x264_decimate_score64_ssse3; + } + if( cpu&X264_CPU_SSE4 ) + { + pf->quant_2x2_dc = x264_quant_2x2_dc_sse4; + pf->quant_4x4_dc = x264_quant_4x4_dc_sse4; + pf->quant_4x4 = x264_quant_4x4_sse4; + pf->quant_8x8 = x264_quant_8x8_sse4; + } +#endif // HAVE_MMX +#else // !HIGH_BIT_DEPTH #if HAVE_MMX if( cpu&X264_CPU_MMX ) { @@ -341,6 +489,7 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf ) pf->dequant_4x4 = x264_dequant_4x4_flat16_sse2; pf->dequant_8x8 = x264_dequant_8x8_flat16_sse2; } + pf->optimize_chroma_dc = x264_optimize_chroma_dc_sse2; pf->denoise_dct = x264_denoise_dct_sse2; pf->decimate_score15 = x264_decimate_score15_sse2; pf->decimate_score16 = x264_decimate_score16_sse2; @@ -371,6 +520,7 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf ) pf->quant_4x4_dc = x264_quant_4x4_dc_ssse3; pf->quant_4x4 = x264_quant_4x4_ssse3; pf->quant_8x8 = x264_quant_8x8_ssse3; + pf->optimize_chroma_dc = x264_optimize_chroma_dc_ssse3; pf->denoise_dct = x264_denoise_dct_ssse3; pf->decimate_score15 = x264_decimate_score15_ssse3; pf->decimate_score16 = x264_decimate_score16_ssse3; @@ -387,6 +537,16 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf ) pf->quant_4x4_dc = x264_quant_4x4_dc_sse4; pf->quant_4x4 = x264_quant_4x4_sse4; pf->quant_8x8 = x264_quant_8x8_sse4; + pf->optimize_chroma_dc = x264_optimize_chroma_dc_sse4; + } + + if( cpu&X264_CPU_AVX ) + { + pf->dequant_4x4 = x264_dequant_4x4_avx; + pf->dequant_8x8 = x264_dequant_8x8_avx; + pf->dequant_4x4_dc = x264_dequant_4x4dc_avx; + pf->optimize_chroma_dc = x264_optimize_chroma_dc_avx; + pf->denoise_dct = x264_denoise_dct_avx; } #endif // HAVE_MMX @@ -420,7 +580,7 @@ void x264_quant_init( x264_t *h, int cpu, x264_quant_function_t *pf ) pf->coeff_last[DCT_LUMA_8x8] = x264_coeff_last64_neon; } #endif -#endif // !X264_HIGH_BIT_DEPTH +#endif // HIGH_BIT_DEPTH pf->coeff_last[ DCT_LUMA_DC] = pf->coeff_last[DCT_LUMA_4x4]; pf->coeff_last[DCT_CHROMA_AC] = pf->coeff_last[ DCT_LUMA_AC]; pf->coeff_level_run[ DCT_LUMA_DC] = pf->coeff_level_run[DCT_LUMA_4x4];