Commit d0bf649f authored by Fiona Glaser's avatar Fiona Glaser
Browse files

CABAC trellis optimizations: use SIMD quant

Significant speed increase, minor change in output due to rounding.
parent 6767f967
......@@ -498,6 +498,8 @@ struct x264_t
udctcoef (*quant8_mf[4])[64]; /* [4][52][64] */
udctcoef (*quant4_bias[4])[16]; /* [4][52][16] */
udctcoef (*quant8_bias[4])[64]; /* [4][52][64] */
udctcoef (*quant4_bias0[4])[16]; /* [4][52][16] */
udctcoef (*quant8_bias0[4])[64]; /* [4][52][64] */
udctcoef (*nr_offset_emergency)[4][64];
/* mv/ref cost arrays. */
......
......@@ -112,9 +112,15 @@ int x264_cqm_init( x264_t *h )
!memcmp( h->pps->scaling_list[i], h->pps->scaling_list[j], size*sizeof(uint8_t) ) )
break;
if( j < i )
{
h->quant4_bias[i] = h->quant4_bias[j];
h->quant4_bias0[i] = h->quant4_bias0[j];
}
else
{
CHECKED_MALLOC( h->quant4_bias[i], (QP_MAX+1)*size*sizeof(udctcoef) );
CHECKED_MALLOC( h->quant4_bias0[i], (QP_MAX+1)*size*sizeof(udctcoef) );
}
}
for( int q = 0; q < 6; q++ )
......@@ -163,6 +169,7 @@ int x264_cqm_init( x264_t *h )
}
// round to nearest, unless that would cause the deadzone to be negative
h->quant4_bias[i_list][q][i] = X264_MIN( DIV(deadzone[i_list]<<10, j), (1<<15)/j );
h->quant4_bias0[i_list][q][i] = (1<<15)/j;
if( j > 0xffff && q > max_qp_err && (i_list == CQM_4IY || i_list == CQM_4PY) )
max_qp_err = q;
if( j > 0xffff && q > max_chroma_qp_err && (i_list == CQM_4IC || i_list == CQM_4PC) )
......@@ -182,6 +189,7 @@ int x264_cqm_init( x264_t *h )
continue;
}
h->quant8_bias[i_list][q][i] = X264_MIN( DIV(deadzone[i_list]<<10, j), (1<<15)/j );
h->quant8_bias0[i_list][q][i] = (1<<15)/j;
if( j > 0xffff && q > max_qp_err && (i_list == CQM_8IY || i_list == CQM_8PY) )
max_qp_err = q;
if( j > 0xffff && q > max_chroma_qp_err && (i_list == CQM_8IC || i_list == CQM_8PC) )
......@@ -272,7 +280,10 @@ fail:
if( h->quant##n##_bias[i] == h->quant##n##_bias[j] )\
break;\
if( j == i )\
{\
x264_free( h->quant##n##_bias[i] );\
x264_free( h->quant##n##_bias0[i] );\
}\
}
void x264_cqm_delete( x264_t *h )
......
......@@ -431,15 +431,17 @@ typedef struct
// comparable to the input. so unquant is the direct inverse of quant,
// and uses the dct scaling factors, not the idct ones.
#define SIGN(x,y) ((x^(y >> 31))-(y >> 31))
static ALWAYS_INLINE
int quant_trellis_cabac( x264_t *h, dctcoef *dct,
const udctcoef *quant_mf, const int *unquant_mf,
udctcoef *quant_mf, udctcoef *quant_bias, const int *unquant_mf,
const uint16_t *coef_weight, const uint8_t *zigzag,
int ctx_block_cat, int i_lambda2, int b_ac,
int b_chroma, int dc, int i_coefs, int idx )
{
udctcoef abs_coefs[64];
int8_t signs[64];
ALIGNED_ARRAY_16( dctcoef, coefs, [64] );
ALIGNED_ARRAY_16( dctcoef, quant_coefs, [64] );
trellis_node_t nodes[2][8];
trellis_node_t *nodes_cur = nodes[0];
trellis_node_t *nodes_prev = nodes[1];
......@@ -448,9 +450,6 @@ int quant_trellis_cabac( x264_t *h, dctcoef *dct,
uint8_t *cabac_state_sig = &h->cabac.state[ significant_coeff_flag_offset[b_interlaced][ctx_block_cat] ];
uint8_t *cabac_state_last = &h->cabac.state[ last_coeff_flag_offset[b_interlaced][ctx_block_cat] ];
const uint8_t *levelgt1_ctx = b_chroma && dc ? coeff_abs_levelgt1_ctx_chroma_dc : coeff_abs_levelgt1_ctx;
const int f = 1 << 15; // no deadzone
int i_last_nnz;
int i;
// (# of coefs) * (# of ctx) * (# of levels tried) = 1024
// we don't need to keep all of those: (# of coefs) * (# of ctx) would be enough,
......@@ -462,30 +461,36 @@ int quant_trellis_cabac( x264_t *h, dctcoef *dct,
} level_tree[64*8*2];
int i_levels_used = 1;
/* init coefs */
for( i = i_coefs-1; i >= b_ac; i-- )
if( (unsigned)(dct[zigzag[i]] * (dc?quant_mf[0]>>1:quant_mf[zigzag[i]]) + f-1) >= 2*f )
break;
if( i < b_ac )
if( i_coefs == 64 )
{
/* We only need to zero an empty 4x4 block. 8x8 can be
implicitly emptied via zero nnz, as can dc. */
if( i_coefs == 16 && !dc )
memset( dct, 0, 16 * sizeof(dctcoef) );
return 0;
h->mc.memcpy_aligned( coefs, dct, sizeof(dctcoef)*64 );
if( !h->quantf.quant_8x8( dct, quant_mf, quant_bias ) )
return 0;
h->zigzagf.scan_8x8( quant_coefs, dct );
}
i_last_nnz = i;
idx &= i_coefs == 64 ? 3 : 15;
for( ; i >= b_ac; i-- )
else if( i_coefs == 16 )
{
memcpy( coefs, dct, sizeof(dctcoef)*16 );
if( !h->quantf.quant_4x4( dct, quant_mf, quant_bias ) )
return 0;
h->zigzagf.scan_4x4( quant_coefs, dct );
}
else
{
int coef = dct[zigzag[i]];
abs_coefs[i] = abs(coef);
signs[i] = coef>>31 | 1;
memcpy( coefs, dct, sizeof(dctcoef)*i_coefs );
int nz = h->quantf.quant_2x2_dc( &dct[0], quant_mf[0] >> 1, quant_bias[0] << 1 );
if( i_coefs == 8 )
nz |= h->quantf.quant_2x2_dc( &dct[4], quant_mf[0] >> 1, quant_bias[0] << 1 );
if( !nz )
return 0;
for( int i = 0; i < i_coefs; i++ )
quant_coefs[i] = dct[zigzag[i]];
}
int i_last_nnz = h->quantf.coeff_last[ctx_block_cat]( quant_coefs+b_ac )+b_ac;
idx &= i_coefs == 64 ? 3 : 15;
/* init trellis */
for( int j = 1; j < 8; j++ )
nodes_cur[j].score = TRELLIS_SCORE_MAX;
......@@ -504,15 +509,11 @@ int quant_trellis_cabac( x264_t *h, dctcoef *dct,
memcpy( nodes_cur[0].cabac_state, &h->cabac.state[ coeff_abs_level_m1_offset[ctx_block_cat] ], 10 );
int i;
for( i = i_last_nnz; i >= b_ac; i-- )
{
int i_coef = abs_coefs[i];
int q = ( f + i_coef * (dc?quant_mf[0]>>1:quant_mf[zigzag[i]]) ) >> 16;
int cost_sig[2], cost_last[2];
trellis_node_t n;
// skip 0s: this doesn't affect the output, but saves some unnecessary computation.
if( q == 0 )
if( !quant_coefs[i] )
{
// no need to calculate ssd of 0s: it's the same in all nodes.
// no need to modify level_tree for ctx=0: it starts with an infinite loop of 0s.
......@@ -537,6 +538,12 @@ int quant_trellis_cabac( x264_t *h, dctcoef *dct,
continue;
}
int sign_coef = coefs[zigzag[i]];
int i_coef = abs( sign_coef );
int q = abs( quant_coefs[i] );
int cost_sig[2], cost_last[2];
trellis_node_t n;
XCHG( trellis_node_t*, nodes_cur, nodes_prev );
for( int j = 0; j < 8; j++ )
......@@ -572,8 +579,8 @@ int quant_trellis_cabac( x264_t *h, dctcoef *dct,
if( h->mb.i_psy_trellis && i && !dc && !b_chroma )
{
int orig_coef = (i_coefs == 64) ? h->mb.pic.fenc_dct8[idx][zigzag[i]] : h->mb.pic.fenc_dct4[idx][zigzag[i]];
int predicted_coef = orig_coef - i_coef * signs[i];
int psy_value = h->mb.i_psy_trellis * abs(predicted_coef + unquant_abs_level * signs[i]);
int predicted_coef = orig_coef - sign_coef;
int psy_value = h->mb.i_psy_trellis * abs(predicted_coef + SIGN(unquant_abs_level, sign_coef));
int psy_weight = (i_coefs == 64) ? x264_dct8_weight_tab[zigzag[i]] : x264_dct4_weight_tab[zigzag[i]];
ssd = (int64_t)d*d * coef_weight[i] - psy_weight * psy_value;
}
......@@ -620,7 +627,7 @@ int quant_trellis_cabac( x264_t *h, dctcoef *dct,
/* Optimize rounding for DC coefficients in DC-only luma 4x4/8x8 blocks. */
else
{
d = i_coef * signs[0] - ((unquant_abs_level * signs[0] + 8)&~15);
d = sign_coef - ((SIGN(unquant_abs_level, sign_coef) + 8)&~15);
n.score += (int64_t)d*d * coef_weight[i];
}
......@@ -642,19 +649,19 @@ int quant_trellis_cabac( x264_t *h, dctcoef *dct,
if( bnode == &nodes_cur[0] )
{
/* We only need to zero an empty 4x4 block. 8x8 can be
implicitly emptied via zero nnz, as can dc. */
if( i_coefs == 16 && !dc )
memset( dct, 0, 16 * sizeof(dctcoef) );
return 0;
}
int level = bnode->level_idx;
for( i = b_ac; level; i++ )
for( i = b_ac; i <= i_last_nnz; i++ )
{
dct[zigzag[i]] = level_tree[level].abs_level * signs[i];
dct[zigzag[i]] = SIGN(level_tree[level].abs_level, coefs[zigzag[i]]);
level = level_tree[level].next;
}
for( ; i < i_coefs; i++ )
dct[zigzag[i]] = 0;
return 1;
}
......@@ -839,10 +846,8 @@ int quant_trellis_cavlc( x264_t *h, dctcoef *dct,
if( coef_mask )
{
for( i = b_ac, j = start; i <= i_last_nnz; i++, j += step )
for( i = b_ac, j = start; i < i_coefs; i++, j += step )
dct[zigzag[j]] = coefs[i];
for( ; j <= end; j += step )
dct[zigzag[j]] = 0;
return 1;
}
......@@ -862,7 +867,8 @@ int x264_quant_luma_dc_trellis( x264_t *h, dctcoef *dct, int i_quant_cat, int i_
{
if( h->param.b_cabac )
return quant_trellis_cabac( h, dct,
h->quant4_mf[i_quant_cat][i_qp], h->unquant4_mf[i_quant_cat][i_qp], NULL, x264_zigzag_scan4[MB_INTERLACED],
h->quant4_mf[i_quant_cat][i_qp], h->quant4_bias0[i_quant_cat][i_qp],
h->unquant4_mf[i_quant_cat][i_qp], NULL, x264_zigzag_scan4[MB_INTERLACED],
ctx_block_cat, h->mb.i_trellis_lambda2[0][b_intra], 0, 0, 1, 16, idx );
return quant_trellis_cavlc( h, dct,
......@@ -892,7 +898,8 @@ int x264_quant_chroma_dc_trellis( x264_t *h, dctcoef *dct, int i_qp, int b_intra
if( h->param.b_cabac )
return quant_trellis_cabac( h, dct,
h->quant4_mf[quant_cat][i_qp], h->unquant4_mf[quant_cat][i_qp], NULL, zigzag,
h->quant4_mf[quant_cat][i_qp], h->quant4_bias0[quant_cat][i_qp],
h->unquant4_mf[quant_cat][i_qp], NULL, zigzag,
DCT_CHROMA_DC, h->mb.i_trellis_lambda2[1][b_intra], 0, 1, 1, num_coefs, idx );
return quant_trellis_cavlc( h, dct,
......@@ -907,8 +914,8 @@ int x264_quant_4x4_trellis( x264_t *h, dctcoef *dct, int i_quant_cat,
int b_ac = ctx_ac[ctx_block_cat];
if( h->param.b_cabac )
return quant_trellis_cabac( h, dct,
h->quant4_mf[i_quant_cat][i_qp], h->unquant4_mf[i_quant_cat][i_qp],
x264_dct4_weight2_zigzag[MB_INTERLACED],
h->quant4_mf[i_quant_cat][i_qp], h->quant4_bias0[i_quant_cat][i_qp],
h->unquant4_mf[i_quant_cat][i_qp], x264_dct4_weight2_zigzag[MB_INTERLACED],
x264_zigzag_scan4[MB_INTERLACED],
ctx_block_cat, h->mb.i_trellis_lambda2[b_chroma][b_intra], b_ac, b_chroma, 0, 16, idx );
......@@ -925,8 +932,8 @@ int x264_quant_8x8_trellis( x264_t *h, dctcoef *dct, int i_quant_cat,
if( h->param.b_cabac )
{
return quant_trellis_cabac( h, dct,
h->quant8_mf[i_quant_cat][i_qp], h->unquant8_mf[i_quant_cat][i_qp],
x264_dct8_weight2_zigzag[MB_INTERLACED],
h->quant8_mf[i_quant_cat][i_qp], h->quant8_bias0[i_quant_cat][i_qp],
h->unquant8_mf[i_quant_cat][i_qp], x264_dct8_weight2_zigzag[MB_INTERLACED],
x264_zigzag_scan8[MB_INTERLACED],
ctx_block_cat, h->mb.i_trellis_lambda2[b_chroma][b_intra], 0, b_chroma, 0, 64, idx );
}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment