* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
*****************************************************************************/
-#include <stdio.h>
-#include <string.h>
-
#include "common/common.h"
#include "macroblock.h"
-/* def_quant4_mf only for probe_skip; actual encoding uses matrices from set.c */
-/* FIXME this seems to make better decisions with cqm=jvt, but could screw up
- * with general custom matrices. */
-static const int def_quant4_mf[6][4][4] =
-{
- { { 13107, 8066, 13107, 8066 }, { 8066, 5243, 8066, 5243 },
- { 13107, 8066, 13107, 8066 }, { 8066, 5243, 8066, 5243 } },
- { { 11916, 7490, 11916, 7490 }, { 7490, 4660, 7490, 4660 },
- { 11916, 7490, 11916, 7490 }, { 7490, 4660, 7490, 4660 } },
- { { 10082, 6554, 10082, 6554 }, { 6554, 4194, 6554, 4194 },
- { 10082, 6554, 10082, 6554 }, { 6554, 4194, 6554, 4194 } },
- { { 9362, 5825, 9362, 5825 }, { 5825, 3647, 5825, 3647 },
- { 9362, 5825, 9362, 5825 }, { 5825, 3647, 5825, 3647 } },
- { { 8192, 5243, 8192, 5243 }, { 5243, 3355, 5243, 3355 },
- { 8192, 5243, 8192, 5243 }, { 5243, 3355, 5243, 3355 } },
- { { 7282, 4559, 7282, 4559 }, { 4559, 2893, 4559, 2893 },
- { 7282, 4559, 7282, 4559 }, { 4559, 2893, 4559, 2893 } }
-};
-
#define ZIG(i,y,x) level[i] = dct[x][y];
-static inline void zigzag_scan_2x2_dc( int level[4], int16_t dct[2][2] )
+static inline void zigzag_scan_2x2_dc( int16_t level[4], int16_t dct[2][2] )
{
ZIG(0,0,0)
ZIG(1,0,1)
}
#undef ZIG
-static void quant_8x8( x264_t *h, int16_t dct[8][8], int quant_mf[6][8][8], int i_qscale, int b_intra )
-{
- const int i_qbits = 16 + i_qscale / 6;
- const int i_mf = i_qscale % 6;
- const int f = h->mb.i_luma_deadzone[b_intra] << (i_qbits-6);
- h->quantf.quant_8x8_core( dct, quant_mf[i_mf], i_qbits, f );
-}
-static void quant_4x4( x264_t *h, int16_t dct[4][4], int quant_mf[6][4][4], int i_qscale, int b_intra )
-{
- const int i_qbits = 15 + i_qscale / 6;
- const int i_mf = i_qscale % 6;
- const int f = h->mb.i_luma_deadzone[b_intra] << (i_qbits-6);
- h->quantf.quant_4x4_core( dct, quant_mf[i_mf], i_qbits, f );
-}
-static void quant_4x4_chroma( x264_t *h, int16_t dct[4][4], int quant_mf[6][4][4], int i_qscale, int b_intra )
-{
- const int i_qbits = 15 + i_qscale / 6;
- const int i_mf = i_qscale % 6;
- const int f = ( 1 << (i_qbits + b_intra) ) / 6;
- h->quantf.quant_4x4_core( dct, quant_mf[i_mf], i_qbits, f );
-}
-static void quant_4x4_dc( x264_t *h, int16_t dct[4][4], int quant_mf[6][4][4], int i_qscale )
-{
- const int i_qbits = 16 + i_qscale / 6;
- const int i_mf = i_qscale % 6;
- const int f = h->mb.i_luma_deadzone[1] << (i_qbits-6);
- h->quantf.quant_4x4_dc_core( dct, quant_mf[i_mf][0][0], i_qbits, f );
-}
-static void quant_2x2_dc( x264_t *h, int16_t dct[2][2], int quant_mf[6][4][4], int i_qscale, int b_intra )
-{
- const int i_qbits = 16 + i_qscale / 6;
- const int i_mf = i_qscale % 6;
- const int f = ( 1 << (i_qbits + b_intra) ) / 6;
- h->quantf.quant_2x2_dc_core( dct, quant_mf[i_mf][0][0], i_qbits, f );
-}
-
/* (ref: JVT-B118)
* x264_mb_decimate_score: given dct coeffs it returns a score to see if we could empty this dct coeffs
* to 0 (low score means set it to null)
* for the complete mb: if score < 6 -> null
* chroma: for the complete mb: if score < 7 -> null
*/
-static int x264_mb_decimate_score( int *dct, int i_max )
+static int x264_mb_decimate_score( int16_t *dct, int i_max )
{
static const int i_ds_table4[16] = {
3,2,2,1,1,1,0,0,0,0,0,0,0,0,0,0 };
{
int i_run;
- if( abs( dct[idx--] ) > 1 )
+ if( (unsigned)(dct[idx--] + 1) > 2 )
return 9;
i_run = 0;
int y = 4 * block_idx_y[idx];
uint8_t *p_src = &h->mb.pic.p_fenc[0][x+y*FENC_STRIDE];
uint8_t *p_dst = &h->mb.pic.p_fdec[0][x+y*FDEC_STRIDE];
- int16_t dct4x4[4][4];
+ DECLARE_ALIGNED_16( int16_t dct4x4[4][4] );
if( h->mb.b_lossless )
{
- h->zigzagf.sub_4x4( h->dct.block[idx].luma4x4, p_src, p_dst );
+ h->zigzagf.sub_4x4( h->dct.luma4x4[idx], p_src, p_dst );
return;
}
if( h->mb.b_trellis )
x264_quant_4x4_trellis( h, dct4x4, CQM_4IY, i_qscale, DCT_LUMA_4x4, 1 );
else
- quant_4x4( h, dct4x4, h->quant4_mf[CQM_4IY], i_qscale, 1 );
+ h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
- h->zigzagf.scan_4x4( h->dct.block[idx].luma4x4, dct4x4 );
+ h->zigzagf.scan_4x4( h->dct.luma4x4[idx], dct4x4 );
h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4IY], i_qscale );
/* output samples to fdec */
int y = 8 * (idx>>1);
uint8_t *p_src = &h->mb.pic.p_fenc[0][x+y*FENC_STRIDE];
uint8_t *p_dst = &h->mb.pic.p_fdec[0][x+y*FDEC_STRIDE];
- int16_t dct8x8[8][8];
+ DECLARE_ALIGNED_16( int16_t dct8x8[8][8] );
h->dctf.sub8x8_dct8( dct8x8, p_src, p_dst );
if( h->mb.b_trellis )
x264_quant_8x8_trellis( h, dct8x8, CQM_8IY, i_qscale, 1 );
else
- quant_8x8( h, dct8x8, h->quant8_mf[CQM_8IY], i_qscale, 1 );
+ h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8IY][i_qscale], h->quant8_bias[CQM_8IY][i_qscale] );
h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8 );
h->quantf.dequant_8x8( dct8x8, h->dequant8_mf[CQM_8IY], i_qscale );
uint8_t *p_src = h->mb.pic.p_fenc[0];
uint8_t *p_dst = h->mb.pic.p_fdec[0];
- int16_t dct4x4[16+1][4][4];
+ DECLARE_ALIGNED_16( int16_t dct4x4[16+1][4][4] );
int i;
{
int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
- h->zigzagf.sub_4x4ac( h->dct.block[i].residual_ac, p_src+oe, p_dst+od );
- dct4x4[0][block_idx_x[i]][block_idx_y[i]] = p_src[oe] - p_dst[od];
- p_dst[od] = p_src[oe];
+ h->zigzagf.sub_4x4( h->dct.luma4x4[i], p_src+oe, p_dst+od );
+ dct4x4[0][block_idx_x[i]][block_idx_y[i]] = h->dct.luma4x4[i][0];
+ h->dct.luma4x4[i][0] = 0;
}
h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct4x4[0] );
return;
{
/* copy dc coeff */
dct4x4[0][block_idx_y[i]][block_idx_x[i]] = dct4x4[1+i][0][0];
+ dct4x4[1+i][0][0] = 0;
/* quant/scan/dequant */
if( h->mb.b_trellis )
x264_quant_4x4_trellis( h, dct4x4[1+i], CQM_4IY, i_qscale, DCT_LUMA_AC, 1 );
else
- quant_4x4( h, dct4x4[1+i], h->quant4_mf[CQM_4IY], i_qscale, 1 );
+ h->quantf.quant_4x4( dct4x4[1+i], h->quant4_mf[CQM_4IY][i_qscale], h->quant4_bias[CQM_4IY][i_qscale] );
- h->zigzagf.scan_4x4ac( h->dct.block[i].residual_ac, dct4x4[1+i] );
+ h->zigzagf.scan_4x4( h->dct.luma4x4[i], dct4x4[1+i] );
h->quantf.dequant_4x4( dct4x4[1+i], h->dequant4_mf[CQM_4IY], i_qscale );
}
h->dctf.dct4x4dc( dct4x4[0] );
- quant_4x4_dc( h, dct4x4[0], h->quant4_mf[CQM_4IY], i_qscale );
+ h->quantf.quant_4x4_dc( dct4x4[0], h->quant4_mf[CQM_4IY][i_qscale][0]>>1, h->quant4_bias[CQM_4IY][i_qscale][0]<<1 );
h->zigzagf.scan_4x4( h->dct.luma16x16_dc, dct4x4[0] );
/* output samples to fdec */
uint8_t *p_dst = h->mb.pic.p_fdec[1+ch];
int i_decimate_score = 0;
- int16_t dct2x2[2][2];
- int16_t dct4x4[4][4][4];
+ DECLARE_ALIGNED_16( int16_t dct2x2[2][2] );
+ DECLARE_ALIGNED_16( int16_t dct4x4[4][4][4] );
if( h->mb.b_lossless )
{
{
int oe = block_idx_x[i]*4 + block_idx_y[i]*4*FENC_STRIDE;
int od = block_idx_x[i]*4 + block_idx_y[i]*4*FDEC_STRIDE;
- h->zigzagf.sub_4x4ac( h->dct.block[16+i+ch*4].residual_ac, p_src+oe, p_dst+od );
- h->dct.chroma_dc[ch][i] = p_src[oe] - p_dst[od];
- p_dst[od] = p_src[oe];
+ h->zigzagf.sub_4x4( h->dct.luma4x4[16+i+ch*4], p_src+oe, p_dst+od );
+ h->dct.chroma_dc[ch][i] = h->dct.luma4x4[16+i+ch*4][0];
+ h->dct.luma4x4[16+i+ch*4][0] = 0;
}
continue;
}
{
/* copy dc coeff */
dct2x2[block_idx_y[i]][block_idx_x[i]] = dct4x4[i][0][0];
+ dct4x4[i][0][0] = 0;
/* no trellis; it doesn't seem to help chroma noticeably */
- quant_4x4_chroma( h, dct4x4[i], h->quant4_mf[CQM_4IC + b_inter], i_qscale, !b_inter );
- h->zigzagf.scan_4x4ac( h->dct.block[16+i+ch*4].residual_ac, dct4x4[i] );
+ h->quantf.quant_4x4( dct4x4[i], h->quant4_mf[CQM_4IC+b_inter][i_qscale], h->quant4_bias[CQM_4IC+b_inter][i_qscale] );
+ h->zigzagf.scan_4x4( h->dct.luma4x4[16+i+ch*4], dct4x4[i] );
if( b_decimate )
{
- i_decimate_score += x264_mb_decimate_score( h->dct.block[16+i+ch*4].residual_ac, 15 );
+ i_decimate_score += x264_mb_decimate_score( h->dct.luma4x4[16+i+ch*4]+1, 15 );
}
}
h->dctf.dct2x2dc( dct2x2 );
- quant_2x2_dc( h, dct2x2, h->quant4_mf[CQM_4IC + b_inter], i_qscale, !b_inter );
+ h->quantf.quant_2x2_dc( dct2x2, h->quant4_mf[CQM_4IC+b_inter][i_qscale][0]>>1, h->quant4_bias[CQM_4IC+b_inter][i_qscale][0]<<1 );
zigzag_scan_2x2_dc( h->dct.chroma_dc[ch], dct2x2 );
/* output samples to fdec */
if( b_decimate && i_decimate_score < 7 )
{
/* Near null chroma 8x8 block so make it null (bits saving) */
- memset( &h->dct.block[16+ch*4], 0, 4 * sizeof( *h->dct.block ) );
- if( !array_non_zero( (int*)dct2x2, sizeof(dct2x2)/sizeof(int) ) )
+ memset( &h->dct.luma4x4[16+ch*4], 0, 4 * sizeof( *h->dct.luma4x4 ) );
+ if( !array_non_zero( dct2x2 ) )
continue;
memset( dct4x4, 0, sizeof( dct4x4 ) );
}
h->mb.i_cbp_chroma = 0;
for( i = 0; i < 8; i++ )
{
- int nz = array_non_zero_count( h->dct.block[16+i].residual_ac, 15 );
+ int nz = array_non_zero( h->dct.luma4x4[16+i] );
h->mb.cache.non_zero_count[x264_scan8[16+i]] = nz;
h->mb.i_cbp_chroma |= nz;
}
if( h->mb.i_cbp_chroma )
h->mb.i_cbp_chroma = 2; /* dc+ac (we can't do only ac) */
- else if( array_non_zero( h->dct.chroma_dc[0], 8 ) )
+ else if( array_non_zero( h->dct.chroma_dc ) )
h->mb.i_cbp_chroma = 1; /* dc only */
}
static void x264_macroblock_encode_skip( x264_t *h )
{
- int i;
h->mb.i_cbp_luma = 0x00;
h->mb.i_cbp_chroma = 0x00;
-
- for( i = 0; i < 16+8; i++ )
- {
- h->mb.cache.non_zero_count[x264_scan8[i]] = 0;
- }
-
+ memset( h->mb.cache.non_zero_count, 0, X264_SCAN8_SIZE );
/* store cbp */
h->mb.cbp[h->mb.i_mb_xy] = 0;
}
const int mvy = x264_clip3( h->mb.cache.mv[0][x264_scan8[0]][1],
h->mb.mv_min[1], h->mb.mv_max[1] );
- /* Motion compensation XXX probably unneeded */
- h->mc.mc_luma( h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
- h->mb.pic.p_fdec[0], FDEC_STRIDE,
- mvx, mvy, 16, 16 );
+ /* don't do pskip motion compensation if it was already done in macroblock_analyse */
+ if( !h->mb.b_skip_mc )
+ {
+ h->mc.mc_luma( h->mb.pic.p_fdec[0], FDEC_STRIDE,
+ h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
+ mvx, mvy, 16, 16 );
- /* Chroma MC */
- h->mc.mc_chroma( h->mb.pic.p_fref[0][0][4], h->mb.pic.i_stride[1],
- h->mb.pic.p_fdec[1], FDEC_STRIDE,
- mvx, mvy, 8, 8 );
+ h->mc.mc_chroma( h->mb.pic.p_fdec[1], FDEC_STRIDE,
+ h->mb.pic.p_fref[0][0][4], h->mb.pic.i_stride[1],
+ mvx, mvy, 8, 8 );
- h->mc.mc_chroma( h->mb.pic.p_fref[0][0][5], h->mb.pic.i_stride[2],
- h->mb.pic.p_fdec[2], FDEC_STRIDE,
- mvx, mvy, 8, 8 );
+ h->mc.mc_chroma( h->mb.pic.p_fdec[2], FDEC_STRIDE,
+ h->mb.pic.p_fref[0][0][5], h->mb.pic.i_stride[2],
+ mvx, mvy, 8, 8 );
+ }
x264_macroblock_encode_skip( h );
}
int i_qp = h->mb.i_qp;
int b_decimate = h->sh.i_type == SLICE_TYPE_B || h->param.analyse.b_dct_decimate;
int b_force_no_skip = 0;
- int i;
+ int i,j,idx;
+ uint8_t nnz8x8[4] = {1,1,1,1};
if( h->sh.b_mbaff
&& h->mb.i_mb_xy == h->sh.i_first_mb + h->mb.i_mb_stride
}
if( h->mb.i_type == B_SKIP )
{
- /* XXX motion compensation is probably unneeded */
- x264_mb_mc( h );
+ /* don't do bskip motion compensation if it was already done in macroblock_analyse */
+ if( !h->mb.b_skip_mc )
+ x264_mb_mc( h );
x264_macroblock_encode_skip( h );
return;
}
}
else if( h->mb.i_type == I_8x8 )
{
- DECLARE_ALIGNED( uint8_t, edge[33], 8 );
+ DECLARE_ALIGNED_16( uint8_t edge[33] );
h->mb.b_transform_8x8 = 1;
- for( i = 0; i < 4; i++ )
+ /* If we already encoded 3 of the 4 i8x8 blocks, we don't have to do them again. */
+ if( h->mb.i_skip_intra )
+ {
+ h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fdec[0], FDEC_STRIDE, h->mb.pic.i8x8_fdec_buf, 16, 16 );
+ /* In RD mode, restore the now-overwritten DCT data. */
+ if( h->mb.i_skip_intra == 2 )
+ h->mc.memcpy_aligned( h->dct.luma8x8, h->mb.pic.i8x8_dct_buf, sizeof(h->mb.pic.i8x8_dct_buf) );
+ }
+ for( i = h->mb.i_skip_intra ? 3 : 0 ; i < 4; i++ )
{
uint8_t *p_dst = &h->mb.pic.p_fdec[0][8 * (i&1) + 8 * (i>>1) * FDEC_STRIDE];
int i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[4*i]];
h->predict_8x8[i_mode]( p_dst, edge );
x264_mb_encode_i8x8( h, i, i_qp );
}
+ for( i = 0; i < 4; i++ )
+ nnz8x8[i] = array_non_zero( h->dct.luma8x8[i] );
}
else if( h->mb.i_type == I_4x4 )
{
h->mb.b_transform_8x8 = 0;
- for( i = 0; i < 16; i++ )
+ /* If we already encoded 15 of the 16 i4x4 blocks, we don't have to do them again. */
+ if( h->mb.i_skip_intra )
+ {
+ h->mc.copy[PIXEL_16x16]( h->mb.pic.p_fdec[0], FDEC_STRIDE, h->mb.pic.i4x4_fdec_buf, 16, 16 );
+ /* In RD mode, restore the now-overwritten DCT data. */
+ if( h->mb.i_skip_intra == 2 )
+ h->mc.memcpy_aligned( h->dct.luma4x4, h->mb.pic.i4x4_dct_buf, sizeof(h->mb.pic.i4x4_dct_buf) );
+ }
+ for( i = h->mb.i_skip_intra ? 15 : 0 ; i < 16; i++ )
{
uint8_t *p_dst = &h->mb.pic.p_fdec[0][4 * block_idx_x[i] + 4 * block_idx_y[i] * FDEC_STRIDE];
int i_mode = h->mb.cache.intra4x4_pred_mode[x264_scan8[i]];
}
else /* Inter MB */
{
- int i8x8, i4x4, idx;
+ int i8x8, i4x4;
int i_decimate_mb = 0;
- /* Motion compensation */
- x264_mb_mc( h );
+ /* Don't repeat motion compensation if it was already done in non-RD transform analysis */
+ if( !h->mb.b_skip_mc )
+ x264_mb_mc( h );
if( h->mb.b_lossless )
{
{
int x = 4*block_idx_x[i4x4];
int y = 4*block_idx_y[i4x4];
- h->zigzagf.sub_4x4( h->dct.block[i4x4].luma4x4,
+ h->zigzagf.sub_4x4( h->dct.luma4x4[i4x4],
h->mb.pic.p_fenc[0]+x+y*FENC_STRIDE,
h->mb.pic.p_fdec[0]+x+y*FDEC_STRIDE );
}
}
else if( h->mb.b_transform_8x8 )
{
- int16_t dct8x8[4][8][8];
- int nnz8x8[4] = {1,1,1,1};
+ DECLARE_ALIGNED_16( int16_t dct8x8[4][8][8] );
b_decimate &= !h->mb.b_trellis; // 8x8 trellis is inherently optimal decimation
h->dctf.sub16x16_dct8( dct8x8, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
if( h->mb.b_trellis )
x264_quant_8x8_trellis( h, dct8x8[idx], CQM_8PY, i_qp, 0 );
else
- quant_8x8( h, dct8x8[idx], h->quant8_mf[CQM_8PY], i_qp, 0 );
+ h->quantf.quant_8x8( dct8x8[idx], h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
h->zigzagf.scan_8x8( h->dct.luma8x8[idx], dct8x8[idx] );
int i_decimate_8x8 = x264_mb_decimate_score( h->dct.luma8x8[idx], 64 );
i_decimate_mb += i_decimate_8x8;
if( i_decimate_8x8 < 4 )
- {
- memset( h->dct.luma8x8[idx], 0, sizeof( h->dct.luma8x8[idx] ) );
- memset( dct8x8[idx], 0, sizeof( dct8x8[idx] ) );
nnz8x8[idx] = 0;
- }
}
else
- nnz8x8[idx] = array_non_zero( (int*)dct8x8[idx], sizeof(*dct8x8)/sizeof(int) );
+ nnz8x8[idx] = array_non_zero( dct8x8[idx] );
}
if( i_decimate_mb < 6 && b_decimate )
- memset( h->dct.luma8x8, 0, sizeof( h->dct.luma8x8 ) );
+ *(uint32_t*)nnz8x8 = 0;
else
{
for( idx = 0; idx < 4; idx++ )
}
else
{
- int16_t dct4x4[16][4][4];
- int nnz8x8[4] = {1,1,1,1};
+ DECLARE_ALIGNED_16( int16_t dct4x4[16][4][4] );
h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0], h->mb.pic.p_fdec[0] );
for( i8x8 = 0; i8x8 < 4; i8x8++ )
if( h->mb.b_trellis )
x264_quant_4x4_trellis( h, dct4x4[idx], CQM_4PY, i_qp, DCT_LUMA_4x4, 0 );
else
- quant_4x4( h, dct4x4[idx], h->quant4_mf[CQM_4PY], i_qp, 0 );
+ h->quantf.quant_4x4( dct4x4[idx], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
+
+ h->zigzagf.scan_4x4( h->dct.luma4x4[idx], dct4x4[idx] );
- h->zigzagf.scan_4x4( h->dct.block[idx].luma4x4, dct4x4[idx] );
-
- if( b_decimate )
- i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[idx].luma4x4, 16 );
+ if( b_decimate && i_decimate_8x8 <= 6 )
+ i_decimate_8x8 += x264_mb_decimate_score( h->dct.luma4x4[idx], 16 );
}
/* decimate this 8x8 block */
i_decimate_mb += i_decimate_8x8;
if( i_decimate_8x8 < 4 && b_decimate )
- {
- memset( &dct4x4[i8x8*4], 0, 4 * sizeof( *dct4x4 ) );
- memset( &h->dct.block[i8x8*4], 0, 4 * sizeof( *h->dct.block ) );
nnz8x8[i8x8] = 0;
- }
}
if( i_decimate_mb < 6 && b_decimate )
- memset( h->dct.block, 0, 16 * sizeof( *h->dct.block ) );
+ *(uint32_t*)nnz8x8 = 0;
else
{
for( i8x8 = 0; i8x8 < 4; i8x8++ )
{
for( i = 0; i < 16; i++ )
{
- const int nz = array_non_zero_count( h->dct.block[i].residual_ac, 15 );
+ int nz = array_non_zero( h->dct.luma4x4[i] );
h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
- if( nz > 0 )
- h->mb.i_cbp_luma = 0x0f;
- }
- }
- else if( h->mb.b_transform_8x8 )
- {
- /* coded_block_flag is enough for CABAC.
- * the full non_zero_count is done only in CAVLC. */
- for( i = 0; i < 4; i++ )
- {
- const int nz = array_non_zero( h->dct.luma8x8[i], 64 );
- int j;
- for( j = 0; j < 4; j++ )
- h->mb.cache.non_zero_count[x264_scan8[4*i+j]] = nz;
- if( nz > 0 )
- h->mb.i_cbp_luma |= 1 << i;
+ h->mb.i_cbp_luma |= nz;
}
+ h->mb.i_cbp_luma *= 0xf;
}
else
{
- for( i = 0; i < 16; i++ )
+ for( i = 0; i < 4; i++)
{
- const int nz = array_non_zero_count( h->dct.block[i].luma4x4, 16 );
- h->mb.cache.non_zero_count[x264_scan8[i]] = nz;
- if( nz > 0 )
- h->mb.i_cbp_luma |= 1 << (i/4);
+ if(!nnz8x8[i])
+ {
+ *(uint16_t*)&h->mb.cache.non_zero_count[x264_scan8[0+i*4]] = 0;
+ *(uint16_t*)&h->mb.cache.non_zero_count[x264_scan8[2+i*4]] = 0;
+ }
+ else if( h->mb.b_transform_8x8 )
+ {
+ *(uint16_t*)&h->mb.cache.non_zero_count[x264_scan8[0+4*i]] = nnz8x8[i] * 0x0101;
+ *(uint16_t*)&h->mb.cache.non_zero_count[x264_scan8[2+4*i]] = nnz8x8[i] * 0x0101;
+ h->mb.i_cbp_luma |= nnz8x8[i] << i;
+ }
+ else
+ {
+ int nz, cbp = 0;
+ for( j = 0; j < 4; j++ )
+ {
+ nz = array_non_zero( h->dct.luma4x4[j+4*i] );
+ h->mb.cache.non_zero_count[x264_scan8[j+4*i]] = nz;
+ cbp |= nz;
+ }
+ h->mb.i_cbp_luma |= cbp << i;
+ }
}
}
if( h->param.b_cabac )
{
- i_cbp_dc = ( h->mb.i_type == I_16x16 && array_non_zero( h->dct.luma16x16_dc, 16 ) )
- | array_non_zero( h->dct.chroma_dc[0], 4 ) << 1
- | array_non_zero( h->dct.chroma_dc[1], 4 ) << 2;
+ i_cbp_dc = ( h->mb.i_type == I_16x16 && array_non_zero( h->dct.luma16x16_dc ) )
+ | array_non_zero( h->dct.chroma_dc[0] ) << 1
+ | array_non_zero( h->dct.chroma_dc[1] ) << 2;
}
/* store cbp */
if( !b_force_no_skip )
{
if( h->mb.i_type == P_L0 && h->mb.i_partition == D_16x16 &&
- h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma == 0x00 &&
- h->mb.cache.mv[0][x264_scan8[0]][0] == h->mb.cache.pskip_mv[0] &&
- h->mb.cache.mv[0][x264_scan8[0]][1] == h->mb.cache.pskip_mv[1] &&
- h->mb.cache.ref[0][x264_scan8[0]] == 0 )
+ !(h->mb.i_cbp_luma | h->mb.i_cbp_chroma) &&
+ *(uint32_t*)h->mb.cache.mv[0][x264_scan8[0]] == *(uint32_t*)h->mb.cache.pskip_mv
+ && h->mb.cache.ref[0][x264_scan8[0]] == 0 )
{
h->mb.i_type = P_SKIP;
}
/* Check for B_SKIP */
- if( h->mb.i_type == B_DIRECT &&
- h->mb.i_cbp_luma == 0x00 && h->mb.i_cbp_chroma== 0x00 )
+ if( h->mb.i_type == B_DIRECT && !(h->mb.i_cbp_luma | h->mb.i_cbp_chroma) )
{
h->mb.i_type = B_SKIP;
}
* Check if the current MB could be encoded as a [PB]_SKIP (it supposes you use
* the previous QP
*****************************************************************************/
-int x264_macroblock_probe_skip( x264_t *h, int b_bidir )
+int x264_macroblock_probe_skip( x264_t *h, const int b_bidir )
{
- DECLARE_ALIGNED( int16_t, dct4x4[16][4][4], 16 );
- DECLARE_ALIGNED( int16_t, dct2x2[2][2], 16 );
- DECLARE_ALIGNED( int, dctscan[16], 16 );
+ DECLARE_ALIGNED_16( int16_t dct4x4[4][4][4] );
+ DECLARE_ALIGNED_16( int16_t dct2x2[2][2] );
+ DECLARE_ALIGNED_16( int16_t dctscan[16] );
int i_qp = h->mb.i_qp;
int mvp[2];
mvp[1] = x264_clip3( h->mb.cache.pskip_mv[1], h->mb.mv_min[1], h->mb.mv_max[1] );
/* Motion compensation */
- h->mc.mc_luma( h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
- h->mb.pic.p_fdec[0], FDEC_STRIDE,
+ h->mc.mc_luma( h->mb.pic.p_fdec[0], FDEC_STRIDE,
+ h->mb.pic.p_fref[0][0], h->mb.pic.i_stride[0],
mvp[0], mvp[1], 16, 16 );
}
- /* get luma diff */
- h->dctf.sub16x16_dct( dct4x4, h->mb.pic.p_fenc[0],
- h->mb.pic.p_fdec[0] );
-
for( i8x8 = 0, i_decimate_mb = 0; i8x8 < 4; i8x8++ )
{
+ int fenc_offset = (i8x8&1) * 8 + (i8x8>>1) * FENC_STRIDE * 8;
+ int fdec_offset = (i8x8&1) * 8 + (i8x8>>1) * FDEC_STRIDE * 8;
+ /* get luma diff */
+ h->dctf.sub8x8_dct( dct4x4, h->mb.pic.p_fenc[0] + fenc_offset,
+ h->mb.pic.p_fdec[0] + fdec_offset );
/* encode one 4x4 block */
for( i4x4 = 0; i4x4 < 4; i4x4++ )
{
- const int idx = i8x8 * 4 + i4x4;
-
- quant_4x4( h, dct4x4[idx], (int(*)[4][4])def_quant4_mf, i_qp, 0 );
- h->zigzagf.scan_4x4( dctscan, dct4x4[idx] );
-
+ h->quantf.quant_4x4( dct4x4[i4x4], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
+ if( !array_non_zero(dct4x4[i4x4]) )
+ continue;
+ h->zigzagf.scan_4x4( dctscan, dct4x4[i4x4] );
i_decimate_mb += x264_mb_decimate_score( dctscan, 16 );
-
if( i_decimate_mb >= 6 )
- {
- /* not as P_SKIP */
return 0;
- }
}
}
if( !b_bidir )
{
- h->mc.mc_chroma( h->mb.pic.p_fref[0][0][4+ch], h->mb.pic.i_stride[1+ch],
- h->mb.pic.p_fdec[1+ch], FDEC_STRIDE,
+ h->mc.mc_chroma( h->mb.pic.p_fdec[1+ch], FDEC_STRIDE,
+ h->mb.pic.p_fref[0][0][4+ch], h->mb.pic.i_stride[1+ch],
mvp[0], mvp[1], 8, 8 );
}
dct2x2[1][0] = dct4x4[2][0][0];
dct2x2[1][1] = dct4x4[3][0][0];
h->dctf.dct2x2dc( dct2x2 );
- quant_2x2_dc( h, dct2x2, (int(*)[4][4])def_quant4_mf, i_qp, 0 );
- if( dct2x2[0][0] || dct2x2[0][1] || dct2x2[1][0] || dct2x2[1][1] )
- {
- /* can't be */
+ h->quantf.quant_2x2_dc( dct2x2, h->quant4_mf[CQM_4PC][i_qp][0]>>1, h->quant4_bias[CQM_4PC][i_qp][0]<<1 );
+ if( array_non_zero(dct2x2) )
return 0;
- }
/* calculate dct coeffs */
for( i4x4 = 0, i_decimate_mb = 0; i4x4 < 4; i4x4++ )
{
- quant_4x4_chroma( h, dct4x4[i4x4], (int(*)[4][4])def_quant4_mf, i_qp, 0 );
- h->zigzagf.scan_4x4ac( dctscan, dct4x4[i4x4] );
-
- i_decimate_mb += x264_mb_decimate_score( dctscan, 15 );
+ h->quantf.quant_4x4( dct4x4[i4x4], h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
+ if( !array_non_zero(dct4x4[i4x4]) )
+ continue;
+ h->zigzagf.scan_4x4( dctscan, dct4x4[i4x4] );
+ i_decimate_mb += x264_mb_decimate_score( dctscan+1, 15 );
if( i_decimate_mb >= 7 )
- {
return 0;
- }
}
}
+ h->mb.b_skip_mc = 1;
return 1;
}
for( cat = 0; cat < 2; cat++ )
{
int size = cat ? 64 : 16;
- const int *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
+ const uint16_t *weight = cat ? x264_dct8_weight2_tab : x264_dct4_weight2_tab;
if( h->nr_count[cat] > (cat ? (1<<16) : (1<<18)) )
{
if( h->mb.b_transform_8x8 )
{
- int16_t dct8x8[8][8];
+ DECLARE_ALIGNED_16( int16_t dct8x8[8][8] );
h->dctf.sub8x8_dct8( dct8x8, p_fenc, p_fdec );
- quant_8x8( h, dct8x8, h->quant8_mf[CQM_8PY], i_qp, 0 );
+ h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8PY][i_qp], h->quant8_bias[CQM_8PY][i_qp] );
h->zigzagf.scan_8x8( h->dct.luma8x8[i8], dct8x8 );
if( b_decimate )
nnz8x8 = 4 <= x264_mb_decimate_score( h->dct.luma8x8[i8], 64 );
else
- nnz8x8 = array_non_zero( (int*)dct8x8, sizeof(dct8x8)/sizeof(int) );
+ nnz8x8 = array_non_zero( dct8x8 );
if( nnz8x8 )
{
else
{
int i4;
- int16_t dct4x4[4][4][4];
+ DECLARE_ALIGNED_16( int16_t dct4x4[4][4][4] );
h->dctf.sub8x8_dct( dct4x4, p_fenc, p_fdec );
- quant_4x4( h, dct4x4[0], h->quant4_mf[CQM_4PY], i_qp, 0 );
- quant_4x4( h, dct4x4[1], h->quant4_mf[CQM_4PY], i_qp, 0 );
- quant_4x4( h, dct4x4[2], h->quant4_mf[CQM_4PY], i_qp, 0 );
- quant_4x4( h, dct4x4[3], h->quant4_mf[CQM_4PY], i_qp, 0 );
for( i4 = 0; i4 < 4; i4++ )
- h->zigzagf.scan_4x4( h->dct.block[i8*4+i4].luma4x4, dct4x4[i4] );
+ h->quantf.quant_4x4( dct4x4[i4], h->quant4_mf[CQM_4PY][i_qp], h->quant4_bias[CQM_4PY][i_qp] );
+ for( i4 = 0; i4 < 4; i4++ )
+ h->zigzagf.scan_4x4( h->dct.luma4x4[i8*4+i4], dct4x4[i4] );
if( b_decimate )
{
int i_decimate_8x8 = 0;
for( i4 = 0; i4 < 4 && i_decimate_8x8 < 4; i4++ )
- i_decimate_8x8 += x264_mb_decimate_score( h->dct.block[i8*4+i4].luma4x4, 16 );
+ i_decimate_8x8 += x264_mb_decimate_score( h->dct.luma4x4[i8*4+i4], 16 );
nnz8x8 = 4 <= i_decimate_8x8;
}
else
- nnz8x8 = array_non_zero( (int*)dct4x4, sizeof(dct4x4)/sizeof(int) );
+ nnz8x8 = array_non_zero( dct4x4 );
if( nnz8x8 )
{
for( ch = 0; ch < 2; ch++ )
{
- int16_t dct4x4[4][4];
+ DECLARE_ALIGNED_16( int16_t dct4x4[4][4] );
p_fenc = h->mb.pic.p_fenc[1+ch] + (i8&1)*4 + (i8>>1)*4*FENC_STRIDE;
p_fdec = h->mb.pic.p_fdec[1+ch] + (i8&1)*4 + (i8>>1)*4*FDEC_STRIDE;
h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
- quant_4x4_chroma( h, dct4x4, h->quant4_mf[CQM_4PC], i_qp, 0 );
- h->zigzagf.scan_4x4ac( h->dct.block[16+i8+ch*4].residual_ac, dct4x4 );
- if( array_non_zero( (int*)dct4x4, sizeof(dct4x4)/sizeof(int) ) )
+ h->quantf.quant_4x4( dct4x4, h->quant4_mf[CQM_4PC][i_qp], h->quant4_bias[CQM_4PC][i_qp] );
+ h->zigzagf.scan_4x4( h->dct.luma4x4[16+i8+ch*4], dct4x4 );
+ if( array_non_zero( dct4x4 ) )
{
h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PC], i_qp );
h->dctf.add4x4_idct( p_fdec, dct4x4 );
}
}
- if( nnz8x8 )
- h->mb.i_cbp_luma |= (1 << i8);
- else
- h->mb.i_cbp_luma &= ~(1 << i8);
+ h->mb.i_cbp_luma &= ~(1 << i8);
+ h->mb.i_cbp_luma |= nnz8x8 << i8;
h->mb.i_cbp_chroma = 0x02;
}
projects / x264 / blobdiff