X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;ds=sidebyside;f=encoder%2Fme.c;h=b5552fb40ade4183bdb198c6df4e2d19694af405;hb=867c1eca89477f4022b447207da3013acb538d57;hp=2c6a6257104c52a9ef0364b03956b2259ece90ab;hpb=213a99d070ebd4f9aeffe7cb3ed9bd7fe755ec7f;p=x264 diff --git a/encoder/me.c b/encoder/me.c index 2c6a6257..b5552fb4 100644 --- a/encoder/me.c +++ b/encoder/me.c @@ -1,7 +1,7 @@ /***************************************************************************** * me.c: motion estimation ***************************************************************************** - * Copyright (C) 2003-2010 x264 project + * Copyright (C) 2003-2016 x264 project * * Authors: Loren Merritt * Laurent Aimar @@ -46,6 +46,7 @@ static const uint8_t subpel_iterations[][4] = {0,0,2,2}, {0,0,4,10}, {0,0,4,10}, + {0,0,4,10}, {0,0,4,10}}; /* (x-1)%6 */ @@ -60,21 +61,22 @@ static void refine_subpel( x264_t *h, x264_me_t *m, int hpel_iters, int qpel_ite (p_cost_mvx[(mx)<<2] + p_cost_mvy[(my)<<2]) #define COST_MV( mx, my )\ +do\ {\ int cost = h->pixf.fpelcmp[i_pixel]( p_fenc, FENC_STRIDE,\ &p_fref_w[(my)*stride+(mx)], stride )\ + BITS_MVD(mx,my);\ COPY3_IF_LT( bcost, cost, bmx, mx, bmy, my );\ -} +} while(0) -#define COST_MV_HPEL( mx, my ) \ -{ \ - int stride2 = 16; \ - pixel *src = h->mc.get_ref( pix, &stride2, m->p_fref, stride, mx, my, bw, bh, &m->weight[0] ); \ - int cost = h->pixf.fpelcmp[i_pixel]( p_fenc, FENC_STRIDE, src, stride2 ) \ - + p_cost_mvx[ mx ] + p_cost_mvy[ my ]; \ - COPY3_IF_LT( bpred_cost, cost, bpred_mx, mx, bpred_my, my ); \ -} +#define COST_MV_HPEL( mx, my, cost )\ +do\ +{\ + intptr_t stride2 = 16;\ + pixel *src = h->mc.get_ref( pix, &stride2, m->p_fref, stride, mx, my, bw, bh, &m->weight[0] );\ + cost = h->pixf.fpelcmp[i_pixel]( p_fenc, FENC_STRIDE, src, stride2 )\ + + p_cost_mvx[ mx ] + p_cost_mvy[ my ];\ +} while(0) #define COST_MV_X3_DIR( m0x, m0y, m1x, m1y, m2x, m2y, costs )\ {\ @@ -173,6 +175,10 @@ static void refine_subpel( x264_t *h, x264_me_t *m, int hpel_iters, int qpel_ite }\ } +#define FPEL(mv) (((mv)+2)>>2) /* Convert subpel MV to fullpel with rounding... */ +#define SPEL(mv) ((mv)<<2) /* ... and the reverse. */ +#define SPELx2(mv) (SPEL(mv)&0xFFFCFFFC) /* for two packed MVs */ + void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, int *p_halfpel_thresh ) { const int bw = x264_pixel_size[m->i_pixel].w; @@ -180,99 +186,136 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, const int i_pixel = m->i_pixel; const int stride = m->i_stride[0]; int i_me_range = h->param.analyse.i_me_range; - int bmx, bmy, bcost; - int bpred_mx = 0, bpred_my = 0, bpred_cost = COST_MAX; + int bmx, bmy, bcost = COST_MAX; + int bpred_cost = COST_MAX; int omx, omy, pmx, pmy; pixel *p_fenc = m->p_fenc[0]; pixel *p_fref_w = m->p_fref_w; - ALIGNED_ARRAY_16( pixel, pix,[16*16] ); - - int costs[16]; - - int mv_x_min = h->mb.mv_min_fpel[0]; - int mv_y_min = h->mb.mv_min_fpel[1]; - int mv_x_max = h->mb.mv_max_fpel[0]; - int mv_y_max = h->mb.mv_max_fpel[1]; - int mv_x_min_qpel = mv_x_min << 2; - int mv_y_min_qpel = mv_y_min << 2; - int mv_x_max_qpel = mv_x_max << 2; - int mv_y_max_qpel = mv_y_max << 2; + ALIGNED_ARRAY_N( pixel, pix,[16*16] ); + ALIGNED_ARRAY_8( int16_t, mvc_temp,[16],[2] ); + + ALIGNED_ARRAY_16( int, costs,[16] ); + + int mv_x_min = h->mb.mv_limit_fpel[0][0]; + int mv_y_min = h->mb.mv_limit_fpel[0][1]; + int mv_x_max = h->mb.mv_limit_fpel[1][0]; + int mv_y_max = h->mb.mv_limit_fpel[1][1]; /* Special version of pack to allow shortcuts in CHECK_MVRANGE */ #define pack16to32_mask2(mx,my) ((mx<<16)|(my&0x7FFF)) uint32_t mv_min = pack16to32_mask2( -mv_x_min, -mv_y_min ); uint32_t mv_max = pack16to32_mask2( mv_x_max, mv_y_max )|0x8000; + uint32_t pmv, bpred_mv = 0; #define CHECK_MVRANGE(mx,my) (!(((pack16to32_mask2(mx,my) + mv_min) | (mv_max - pack16to32_mask2(mx,my))) & 0x80004000)) const uint16_t *p_cost_mvx = m->p_cost_mv - m->mvp[0]; const uint16_t *p_cost_mvy = m->p_cost_mv - m->mvp[1]; - uint32_t pmv; - bmx = x264_clip3( m->mvp[0], mv_x_min_qpel, mv_x_max_qpel ); - bmy = x264_clip3( m->mvp[1], mv_y_min_qpel, mv_y_max_qpel ); - pmx = ( bmx + 2 ) >> 2; - pmy = ( bmy + 2 ) >> 2; - bcost = COST_MAX; - - /* try extra predictors if provided */ + /* Try extra predictors if provided. If subme >= 3, check subpel predictors, + * otherwise round them to fullpel. */ if( h->mb.i_subpel_refine >= 3 ) { - pmv = pack16to32_mask(bmx,bmy); - if( i_mvc ) - COST_MV_HPEL( bmx, bmy ); - for( int i = 0; i < i_mvc; i++ ) + /* Calculate and check the MVP first */ + int bpred_mx = x264_clip3( m->mvp[0], SPEL(mv_x_min), SPEL(mv_x_max) ); + int bpred_my = x264_clip3( m->mvp[1], SPEL(mv_y_min), SPEL(mv_y_max) ); + pmv = pack16to32_mask( bpred_mx, bpred_my ); + pmx = FPEL( bpred_mx ); + pmy = FPEL( bpred_my ); + + COST_MV_HPEL( bpred_mx, bpred_my, bpred_cost ); + int pmv_cost = bpred_cost; + + if( i_mvc > 0 ) { - if( M32( mvc[i] ) && (pmv != M32( mvc[i] )) ) + /* Clip MV candidates and eliminate those equal to zero and pmv. */ + int valid_mvcs = x264_predictor_clip( mvc_temp+2, mvc, i_mvc, h->mb.mv_limit_fpel, pmv ); + if( valid_mvcs > 0 ) { - int mx = x264_clip3( mvc[i][0], mv_x_min_qpel, mv_x_max_qpel ); - int my = x264_clip3( mvc[i][1], mv_y_min_qpel, mv_y_max_qpel ); - COST_MV_HPEL( mx, my ); + int i = 1, cost; + /* We stuff pmv here to branchlessly pick between pmv and the various + * MV candidates. [0] gets skipped in order to maintain alignment for + * x264_predictor_clip. */ + M32( mvc_temp[1] ) = pmv; + bpred_cost <<= 4; + do + { + int mx = mvc_temp[i+1][0]; + int my = mvc_temp[i+1][1]; + COST_MV_HPEL( mx, my, cost ); + COPY1_IF_LT( bpred_cost, (cost << 4) + i ); + } while( ++i <= valid_mvcs ); + bpred_mx = mvc_temp[(bpred_cost&15)+1][0]; + bpred_my = mvc_temp[(bpred_cost&15)+1][1]; + bpred_cost >>= 4; } } - bmx = ( bpred_mx + 2 ) >> 2; - bmy = ( bpred_my + 2 ) >> 2; - COST_MV( bmx, bmy ); + + /* Round the best predictor back to fullpel and get the cost, since this is where + * we'll be starting the fullpel motion search. */ + bmx = FPEL( bpred_mx ); + bmy = FPEL( bpred_my ); + bpred_mv = pack16to32_mask(bpred_mx, bpred_my); + if( bpred_mv&0x00030003 ) /* Only test if the tested predictor is actually subpel... */ + COST_MV( bmx, bmy ); + else /* Otherwise just copy the cost (we already know it) */ + bcost = bpred_cost; + + /* Test the zero vector if it hasn't been tested yet. */ + if( pmv ) + { + if( bmx|bmy ) COST_MV( 0, 0 ); + } + /* If a subpel mv candidate was better than the zero vector, the previous + * fullpel check won't have gotten it even if the pmv was zero. So handle + * that possibility here. */ + else + { + COPY3_IF_LT( bcost, pmv_cost, bmx, 0, bmy, 0 ); + } } else { - /* check the MVP */ - bmx = pmx; - bmy = pmy; + /* Calculate and check the fullpel MVP first */ + bmx = pmx = x264_clip3( FPEL(m->mvp[0]), mv_x_min, mv_x_max ); + bmy = pmy = x264_clip3( FPEL(m->mvp[1]), mv_y_min, mv_y_max ); + pmv = pack16to32_mask( bmx, bmy ); + /* Because we are rounding the predicted motion vector to fullpel, there will be * an extra MV cost in 15 out of 16 cases. However, when the predicted MV is * chosen as the best predictor, it is often the case that the subpel search will - * result in a vector at or next to the predicted motion vector. Therefore, it is - * sensible to omit the cost of the MV from the rounded MVP to avoid unfairly - * biasing against use of the predicted motion vector. */ + * result in a vector at or next to the predicted motion vector. Therefore, we omit + * the cost of the MV from the rounded MVP to avoid unfairly biasing against use of + * the predicted motion vector. + * + * Disclaimer: this is a post-hoc rationalization for why this hack works. */ bcost = h->pixf.fpelcmp[i_pixel]( p_fenc, FENC_STRIDE, &p_fref_w[bmy*stride+bmx], stride ); - pmv = pack16to32_mask( bmx, bmy ); + if( i_mvc > 0 ) { - ALIGNED_ARRAY_8( int16_t, mvc_fpel,[16],[2] ); - x264_predictor_roundclip( mvc_fpel, mvc, i_mvc, mv_x_min, mv_x_max, mv_y_min, mv_y_max ); - bcost <<= 4; - for( int i = 1; i <= i_mvc; i++ ) + /* Like in subme>=3, except we also round the candidates to fullpel. */ + int valid_mvcs = x264_predictor_roundclip( mvc_temp+2, mvc, i_mvc, h->mb.mv_limit_fpel, pmv ); + if( valid_mvcs > 0 ) { - if( M32( mvc_fpel[i-1] ) && (pmv != M32( mvc[i-1] )) ) + int i = 1, cost; + M32( mvc_temp[1] ) = pmv; + bcost <<= 4; + do { - int mx = mvc_fpel[i-1][0]; - int my = mvc_fpel[i-1][1]; - int cost = h->pixf.fpelcmp[i_pixel]( p_fenc, FENC_STRIDE, &p_fref_w[my*stride+mx], stride ) + BITS_MVD( mx, my ); - cost = (cost << 4) + i; - COPY1_IF_LT( bcost, cost ); - } - } - if( bcost&15 ) - { - bmx = mvc_fpel[(bcost&15)-1][0]; - bmy = mvc_fpel[(bcost&15)-1][1]; + int mx = mvc_temp[i+1][0]; + int my = mvc_temp[i+1][1]; + cost = h->pixf.fpelcmp[i_pixel]( p_fenc, FENC_STRIDE, &p_fref_w[my*stride+mx], stride ) + BITS_MVD( mx, my ); + COPY1_IF_LT( bcost, (cost << 4) + i ); + } while( ++i <= valid_mvcs ); + bmx = mvc_temp[(bcost&15)+1][0]; + bmy = mvc_temp[(bcost&15)+1][1]; + bcost >>= 4; } - bcost >>= 4; } - } - if( pmv ) - COST_MV( 0, 0 ); + /* Same as above, except the condition is simpler. */ + if( pmv ) + COST_MV( 0, 0 ); + } switch( h->mb.i_me_method ) { @@ -299,7 +342,7 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, } case X264_ME_HEX: - { + { me_hex2: /* hexagon search, radius 2 */ #if 0 @@ -322,14 +365,14 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, /* hexagon */ COST_MV_X3_DIR( -2,0, -1, 2, 1, 2, costs ); - COST_MV_X3_DIR( 2,0, 1,-2, -1,-2, costs+3 ); + COST_MV_X3_DIR( 2,0, 1,-2, -1,-2, costs+4 ); /* +4 for 16-byte alignment */ bcost <<= 3; COPY1_IF_LT( bcost, (costs[0]<<3)+2 ); COPY1_IF_LT( bcost, (costs[1]<<3)+3 ); COPY1_IF_LT( bcost, (costs[2]<<3)+4 ); - COPY1_IF_LT( bcost, (costs[3]<<3)+5 ); - COPY1_IF_LT( bcost, (costs[4]<<3)+6 ); - COPY1_IF_LT( bcost, (costs[5]<<3)+7 ); + COPY1_IF_LT( bcost, (costs[4]<<3)+5 ); + COPY1_IF_LT( bcost, (costs[5]<<3)+6 ); + COPY1_IF_LT( bcost, (costs[6]<<3)+7 ); if( bcost&7 ) { @@ -359,19 +402,20 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, bcost >>= 3; #endif /* square refine */ - int dir = 0; + bcost <<= 4; COST_MV_X4_DIR( 0,-1, 0,1, -1,0, 1,0, costs ); - COPY2_IF_LT( bcost, costs[0], dir, 1 ); - COPY2_IF_LT( bcost, costs[1], dir, 2 ); - COPY2_IF_LT( bcost, costs[2], dir, 3 ); - COPY2_IF_LT( bcost, costs[3], dir, 4 ); + COPY1_IF_LT( bcost, (costs[0]<<4)+1 ); + COPY1_IF_LT( bcost, (costs[1]<<4)+2 ); + COPY1_IF_LT( bcost, (costs[2]<<4)+3 ); + COPY1_IF_LT( bcost, (costs[3]<<4)+4 ); COST_MV_X4_DIR( -1,-1, -1,1, 1,-1, 1,1, costs ); - COPY2_IF_LT( bcost, costs[0], dir, 5 ); - COPY2_IF_LT( bcost, costs[1], dir, 6 ); - COPY2_IF_LT( bcost, costs[2], dir, 7 ); - COPY2_IF_LT( bcost, costs[3], dir, 8 ); - bmx += square1[dir][0]; - bmy += square1[dir][1]; + COPY1_IF_LT( bcost, (costs[0]<<4)+5 ); + COPY1_IF_LT( bcost, (costs[1]<<4)+6 ); + COPY1_IF_LT( bcost, (costs[2]<<4)+7 ); + COPY1_IF_LT( bcost, (costs[3]<<4)+8 ); + bmx += square1[bcost&15][0]; + bmy += square1[bcost&15][1]; + bcost >>= 4; break; } @@ -583,21 +627,19 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, #if 0 /* plain old exhaustive search */ for( int my = min_y; my <= max_y; my++ ) - for( int mx = min_x; mx <= max_x; mx++ ) + for( int mx = min_x; mx < min_x + width; mx++ ) COST_MV( mx, my ); #else /* successive elimination by comparing DC before a full SAD, * because sum(abs(diff)) >= abs(diff(sum)). */ uint16_t *sums_base = m->integral; - /* due to a GCC bug on some platforms (win32?), zero[] may not actually be aligned. - * this is not a problem because it is not used for any SSE instructions. */ - ALIGNED_16( static pixel zero[8*FENC_STRIDE] ); + ALIGNED_16( static pixel zero[8*FENC_STRIDE] ) = {0}; ALIGNED_ARRAY_16( int, enc_dc,[4] ); int sad_size = i_pixel <= PIXEL_8x8 ? PIXEL_8x8 : PIXEL_4x4; int delta = x264_pixel_size[sad_size].w; int16_t *xs = h->scratch_buffer; int xn; - uint16_t *cost_fpel_mvx = h->cost_mv_fpel[x264_lambda_tab[h->mb.i_qp]][-m->mvp[0]&3] + (-m->mvp[0]>>2); + uint16_t *cost_fpel_mvx = h->cost_mv_fpel[h->mb.i_qp][-m->mvp[0]&3] + (-m->mvp[0]>>2); h->pixf.sad_x4[sad_size]( zero, p_fenc, p_fenc+delta, p_fenc+delta*FENC_STRIDE, p_fenc+delta+delta*FENC_STRIDE, @@ -612,7 +654,7 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, if( h->mb.i_me_method == X264_ME_TESA ) { // ADS threshold, then SAD threshold, then keep the best few SADs, then SATD - mvsad_t *mvsads = (mvsad_t *)(xs + ((width+15)&~15) + 4); + mvsad_t *mvsads = (mvsad_t *)(xs + ((width+31)&~31) + 4); int nmvsad = 0, limit; int sad_thresh = i_me_range <= 16 ? 10 : i_me_range <= 24 ? 11 : 12; int bsad = h->pixf.sad[i_pixel]( p_fenc, FENC_STRIDE, p_fref_w+bmy*stride+bmx, stride ) @@ -629,7 +671,7 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, for( i = 0; i < xn-2; i += 3 ) { pixel *ref = p_fref_w+min_x+my*stride; - int sads[3]; + ALIGNED_ARRAY_16( int, sads,[4] ); /* padded to [4] for asm */ h->pixf.sad_x3[i_pixel]( p_fenc, ref+xs[i], ref+xs[i+1], ref+xs[i+2], stride, sads ); for( int j = 0; j < 3; j++ ) { @@ -665,10 +707,11 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, sad_thresh = bsad*sad_thresh>>3; while( nmvsad > limit*2 && sad_thresh > bsad ) { - int i; + int i = 0; // halve the range if the domain is too large... eh, close enough sad_thresh = (sad_thresh + bsad) >> 1; - for( i = 0; i < nmvsad && mvsads[i].sad <= sad_thresh; i++ ); + while( i < nmvsad && mvsads[i].sad <= sad_thresh ) + i++; for( int j = i; j < nmvsad; j++ ) { uint32_t sad; @@ -730,24 +773,22 @@ void x264_me_search_ref( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], int i_mvc, } /* -> qpel mv */ - if( bpred_cost < bcost ) + uint32_t bmv = pack16to32_mask(bmx,bmy); + uint32_t bmv_spel = SPELx2(bmv); + if( h->mb.i_subpel_refine < 3 ) { - m->mv[0] = bpred_mx; - m->mv[1] = bpred_my; - m->cost = bpred_cost; + m->cost_mv = p_cost_mvx[bmx<<2] + p_cost_mvy[bmy<<2]; + m->cost = bcost; + /* compute the real cost */ + if( bmv == pmv ) m->cost += m->cost_mv; + M32( m->mv ) = bmv_spel; } else { - m->mv[0] = bmx << 2; - m->mv[1] = bmy << 2; - m->cost = bcost; + M32(m->mv) = bpred_cost < bcost ? bpred_mv : bmv_spel; + m->cost = X264_MIN( bpred_cost, bcost ); } - /* compute the real cost */ - m->cost_mv = p_cost_mvx[ m->mv[0] ] + p_cost_mvy[ m->mv[1] ]; - if( bmx == pmx && bmy == pmy && h->mb.i_subpel_refine < 3 ) - m->cost += m->cost_mv; - /* subpel refine */ if( h->mb.i_subpel_refine >= 2 ) { @@ -776,7 +817,7 @@ void x264_me_refine_qpel_refdupe( x264_t *h, x264_me_t *m, int *p_halfpel_thresh #define COST_MV_SAD( mx, my ) \ { \ - int stride = 16; \ + intptr_t stride = 16; \ pixel *src = h->mc.get_ref( pix, &stride, m->p_fref, m->i_stride[0], mx, my, bw, bh, &m->weight[0] ); \ int cost = h->pixf.fpelcmp[i_pixel]( m->p_fenc[0], FENC_STRIDE, src, stride ) \ + p_cost_mvx[ mx ] + p_cost_mvy[ my ]; \ @@ -786,23 +827,37 @@ void x264_me_refine_qpel_refdupe( x264_t *h, x264_me_t *m, int *p_halfpel_thresh #define COST_MV_SATD( mx, my, dir ) \ if( b_refine_qpel || (dir^1) != odir ) \ { \ - int stride = 16; \ - pixel *src = h->mc.get_ref( pix, &stride, m->p_fref, m->i_stride[0], mx, my, bw, bh, &m->weight[0] ); \ + intptr_t stride = 16; \ + pixel *src = h->mc.get_ref( pix, &stride, &m->p_fref[0], m->i_stride[0], mx, my, bw, bh, &m->weight[0] ); \ int cost = h->pixf.mbcmp_unaligned[i_pixel]( m->p_fenc[0], FENC_STRIDE, src, stride ) \ + p_cost_mvx[ mx ] + p_cost_mvy[ my ]; \ if( b_chroma_me && cost < bcost ) \ { \ - h->mc.mc_chroma( pix, pix+8, 16, m->p_fref[4], m->i_stride[1], mx, my + mvy_offset, bw>>1, bh>>1 ); \ - if( m->weight[1].weightfn ) \ - m->weight[1].weightfn[x264_pixel_size[i_pixel].w>>3]( pix, 16, pix, 16, \ - &m->weight[1], x264_pixel_size[i_pixel].h>>1 ); \ - cost += h->pixf.mbcmp[i_pixel+3]( m->p_fenc[1], FENC_STRIDE, pix, 16 ); \ - if( cost < bcost ) \ + if( CHROMA444 ) \ { \ - if( m->weight[2].weightfn ) \ - m->weight[2].weightfn[x264_pixel_size[i_pixel].w>>3]( pix+8, 16, pix+8, 16, \ - &m->weight[2], x264_pixel_size[i_pixel].h>>1 ); \ - cost += h->pixf.mbcmp[i_pixel+3]( m->p_fenc[2], FENC_STRIDE, pix+8, 16 ); \ + stride = 16; \ + src = h->mc.get_ref( pix, &stride, &m->p_fref[4], m->i_stride[1], mx, my, bw, bh, &m->weight[1] ); \ + cost += h->pixf.mbcmp_unaligned[i_pixel]( m->p_fenc[1], FENC_STRIDE, src, stride ); \ + if( cost < bcost ) \ + { \ + stride = 16; \ + src = h->mc.get_ref( pix, &stride, &m->p_fref[8], m->i_stride[2], mx, my, bw, bh, &m->weight[2] ); \ + cost += h->pixf.mbcmp_unaligned[i_pixel]( m->p_fenc[2], FENC_STRIDE, src, stride ); \ + } \ + } \ + else \ + { \ + h->mc.mc_chroma( pix, pix+8, 16, m->p_fref[4], m->i_stride[1], \ + mx, 2*(my+mvy_offset)>>chroma_v_shift, bw>>1, bh>>chroma_v_shift ); \ + if( m->weight[1].weightfn ) \ + m->weight[1].weightfn[bw>>3]( pix, 16, pix, 16, &m->weight[1], bh>>chroma_v_shift ); \ + cost += h->pixf.mbcmp[chromapix]( m->p_fenc[1], FENC_STRIDE, pix, 16 ); \ + if( cost < bcost ) \ + { \ + if( m->weight[2].weightfn ) \ + m->weight[2].weightfn[bw>>3]( pix+8, 16, pix+8, 16, &m->weight[2], bh>>chroma_v_shift ); \ + cost += h->pixf.mbcmp[chromapix]( m->p_fenc[2], FENC_STRIDE, pix+8, 16 ); \ + } \ } \ } \ COPY4_IF_LT( bcost, cost, bmx, mx, bmy, my, bdir, dir ); \ @@ -815,43 +870,57 @@ static void refine_subpel( x264_t *h, x264_me_t *m, int hpel_iters, int qpel_ite const uint16_t *p_cost_mvx = m->p_cost_mv - m->mvp[0]; const uint16_t *p_cost_mvy = m->p_cost_mv - m->mvp[1]; const int i_pixel = m->i_pixel; - const int b_chroma_me = h->mb.b_chroma_me && i_pixel <= PIXEL_8x8; - const int mvy_offset = h->mb.b_interlaced & m->i_ref ? (h->mb.i_mb_y & 1)*4 - 2 : 0; + const int b_chroma_me = h->mb.b_chroma_me && (i_pixel <= PIXEL_8x8 || CHROMA444); + int chromapix = h->luma2chroma_pixel[i_pixel]; + int chroma_v_shift = CHROMA_V_SHIFT; + int mvy_offset = chroma_v_shift & MB_INTERLACED & m->i_ref ? (h->mb.i_mb_y & 1)*4 - 2 : 0; - ALIGNED_ARRAY_16( pixel, pix,[64*18] ); // really 17x17x2, but round up for alignment + ALIGNED_ARRAY_N( pixel, pix,[64*18] ); // really 17x17x2, but round up for alignment + ALIGNED_ARRAY_16( int, costs,[4] ); int bmx = m->mv[0]; int bmy = m->mv[1]; int bcost = m->cost; int odir = -1, bdir; - /* try the subpel component of the predicted mv */ - if( hpel_iters && h->mb.i_subpel_refine < 3 ) - { - int mx = x264_clip3( m->mvp[0], h->mb.mv_min_spel[0]+2, h->mb.mv_max_spel[0]-2 ); - int my = x264_clip3( m->mvp[1], h->mb.mv_min_spel[1]+2, h->mb.mv_max_spel[1]-2 ); - if( (mx-bmx)|(my-bmy) ) - COST_MV_SAD( mx, my ); - } - /* halfpel diamond search */ - for( int i = hpel_iters; i > 0; i-- ) + if( hpel_iters ) { - int omx = bmx, omy = bmy; - int costs[4]; - int stride = 64; // candidates are either all hpel or all qpel, so one stride is enough - pixel *src0, *src1, *src2, *src3; - src0 = h->mc.get_ref( pix, &stride, m->p_fref, m->i_stride[0], omx, omy-2, bw, bh+1, &m->weight[0] ); - src2 = h->mc.get_ref( pix+32, &stride, m->p_fref, m->i_stride[0], omx-2, omy, bw+4, bh, &m->weight[0] ); - src1 = src0 + stride; - src3 = src2 + 1; - h->pixf.fpelcmp_x4[i_pixel]( m->p_fenc[0], src0, src1, src2, src3, stride, costs ); - COPY2_IF_LT( bcost, costs[0] + p_cost_mvx[omx ] + p_cost_mvy[omy-2], bmy, omy-2 ); - COPY2_IF_LT( bcost, costs[1] + p_cost_mvx[omx ] + p_cost_mvy[omy+2], bmy, omy+2 ); - COPY3_IF_LT( bcost, costs[2] + p_cost_mvx[omx-2] + p_cost_mvy[omy ], bmx, omx-2, bmy, omy ); - COPY3_IF_LT( bcost, costs[3] + p_cost_mvx[omx+2] + p_cost_mvy[omy ], bmx, omx+2, bmy, omy ); - if( (bmx == omx) & (bmy == omy) ) - break; + /* try the subpel component of the predicted mv */ + if( h->mb.i_subpel_refine < 3 ) + { + int mx = x264_clip3( m->mvp[0], h->mb.mv_min_spel[0]+2, h->mb.mv_max_spel[0]-2 ); + int my = x264_clip3( m->mvp[1], h->mb.mv_min_spel[1]+2, h->mb.mv_max_spel[1]-2 ); + if( (mx-bmx)|(my-bmy) ) + COST_MV_SAD( mx, my ); + } + + bcost <<= 6; + for( int i = hpel_iters; i > 0; i-- ) + { + int omx = bmx, omy = bmy; + intptr_t stride = 64; // candidates are either all hpel or all qpel, so one stride is enough + pixel *src0, *src1, *src2, *src3; + src0 = h->mc.get_ref( pix, &stride, m->p_fref, m->i_stride[0], omx, omy-2, bw, bh+1, &m->weight[0] ); + src2 = h->mc.get_ref( pix+32, &stride, m->p_fref, m->i_stride[0], omx-2, omy, bw+4, bh, &m->weight[0] ); + src1 = src0 + stride; + src3 = src2 + 1; + h->pixf.fpelcmp_x4[i_pixel]( m->p_fenc[0], src0, src1, src2, src3, stride, costs ); + costs[0] += p_cost_mvx[omx ] + p_cost_mvy[omy-2]; + costs[1] += p_cost_mvx[omx ] + p_cost_mvy[omy+2]; + costs[2] += p_cost_mvx[omx-2] + p_cost_mvy[omy ]; + costs[3] += p_cost_mvx[omx+2] + p_cost_mvy[omy ]; + COPY1_IF_LT( bcost, (costs[0]<<6)+2 ); + COPY1_IF_LT( bcost, (costs[1]<<6)+6 ); + COPY1_IF_LT( bcost, (costs[2]<<6)+16 ); + COPY1_IF_LT( bcost, (costs[3]<<6)+48 ); + if( !(bcost&63) ) + break; + bmx -= (bcost<<26)>>29; + bmy -= (bcost<<29)>>29; + bcost &= ~63; + } + bcost >>= 6; } if( !b_refine_qpel && (h->pixf.mbcmp_unaligned[0] != h->pixf.fpelcmp[0] || b_chroma_me) ) @@ -896,7 +965,6 @@ static void refine_subpel( x264_t *h, x264_me_t *m, int hpel_iters, int qpel_ite /* Special simplified case for subme=1 */ else if( bmy > h->mb.mv_min_spel[1] && bmy < h->mb.mv_max_spel[1] && bmx > h->mb.mv_min_spel[0] && bmx < h->mb.mv_max_spel[0] ) { - int costs[4]; int omx = bmx, omy = bmy; /* We have to use mc_luma because all strides must be the same to use fpelcmp_x4 */ h->mc.mc_luma( pix , 64, m->p_fref, m->i_stride[0], omx, omy-1, bw, bh, &m->weight[0] ); @@ -904,10 +972,18 @@ static void refine_subpel( x264_t *h, x264_me_t *m, int hpel_iters, int qpel_ite h->mc.mc_luma( pix+32, 64, m->p_fref, m->i_stride[0], omx-1, omy, bw, bh, &m->weight[0] ); h->mc.mc_luma( pix+48, 64, m->p_fref, m->i_stride[0], omx+1, omy, bw, bh, &m->weight[0] ); h->pixf.fpelcmp_x4[i_pixel]( m->p_fenc[0], pix, pix+16, pix+32, pix+48, 64, costs ); - COPY2_IF_LT( bcost, costs[0] + p_cost_mvx[omx ] + p_cost_mvy[omy-1], bmy, omy-1 ); - COPY2_IF_LT( bcost, costs[1] + p_cost_mvx[omx ] + p_cost_mvy[omy+1], bmy, omy+1 ); - COPY3_IF_LT( bcost, costs[2] + p_cost_mvx[omx-1] + p_cost_mvy[omy ], bmx, omx-1, bmy, omy ); - COPY3_IF_LT( bcost, costs[3] + p_cost_mvx[omx+1] + p_cost_mvy[omy ], bmx, omx+1, bmy, omy ); + costs[0] += p_cost_mvx[omx ] + p_cost_mvy[omy-1]; + costs[1] += p_cost_mvx[omx ] + p_cost_mvy[omy+1]; + costs[2] += p_cost_mvx[omx-1] + p_cost_mvy[omy ]; + costs[3] += p_cost_mvx[omx+1] + p_cost_mvy[omy ]; + bcost <<= 4; + COPY1_IF_LT( bcost, (costs[0]<<4)+1 ); + COPY1_IF_LT( bcost, (costs[1]<<4)+3 ); + COPY1_IF_LT( bcost, (costs[2]<<4)+4 ); + COPY1_IF_LT( bcost, (costs[3]<<4)+12 ); + bmx -= (bcost<<28)>>30; + bmy -= (bcost<<30)>>30; + bcost >>= 4; } m->cost = bcost; @@ -922,13 +998,27 @@ static void refine_subpel( x264_t *h, x264_me_t *m, int hpel_iters, int qpel_ite int i = 4 + 3*dx + dy;\ int mvx = bm##list##x+dx;\ int mvy = bm##list##y+dy;\ - stride[list][i] = bw;\ - src[list][i] = h->mc.get_ref( pixy_buf[list][i], &stride[list][i], m->p_fref, m->i_stride[0], mvx, mvy, bw, bh, weight_none );\ + stride[0][list][i] = bw;\ + src[0][list][i] = h->mc.get_ref( pixy_buf[list][i], &stride[0][list][i], &m->p_fref[0],\ + m->i_stride[0], mvx, mvy, bw, bh, x264_weight_none );\ if( rd )\ - h->mc.mc_chroma( pixu_buf[list][i], pixv_buf[list][i], 8, m->p_fref[4], m->i_stride[1], mvx, mvy + mv##list##y_offset, bw>>1, bh>>1 );\ + {\ + if( CHROMA444 )\ + {\ + stride[1][list][i] = bw;\ + src[1][list][i] = h->mc.get_ref( pixu_buf[list][i], &stride[1][list][i], &m->p_fref[4],\ + m->i_stride[1], mvx, mvy, bw, bh, x264_weight_none );\ + stride[2][list][i] = bw;\ + src[2][list][i] = h->mc.get_ref( pixv_buf[list][i], &stride[2][list][i], &m->p_fref[8],\ + m->i_stride[2], mvx, mvy, bw, bh, x264_weight_none );\ + }\ + else\ + h->mc.mc_chroma( pixu_buf[list][i], pixv_buf[list][i], 8, m->p_fref[4], m->i_stride[1],\ + mvx, 2*(mvy+mv##list##y_offset)>>chroma_v_shift, bw>>1, bh>>chroma_v_shift );\ + }\ } -#define SATD_THRESH 17/16 +#define SATD_THRESH(cost) (cost+(cost>>4)) /* Don't unroll the BIME_CACHE loop. I couldn't find any way to force this * other than making its iteration count not a compile-time constant. */ @@ -944,18 +1034,22 @@ static void ALWAYS_INLINE x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_m const int i_pixel = m0->i_pixel; const int bw = x264_pixel_size[i_pixel].w; const int bh = x264_pixel_size[i_pixel].h; - ALIGNED_ARRAY_16( pixel, pixy_buf,[2],[9][16*16] ); - ALIGNED_ARRAY_8( pixel, pixu_buf,[2],[9][8*8] ); - ALIGNED_ARRAY_8( pixel, pixv_buf,[2],[9][8*8] ); - pixel *src[2][9]; + ALIGNED_ARRAY_N( pixel, pixy_buf,[2],[9][16*16] ); + ALIGNED_ARRAY_N( pixel, pixu_buf,[2],[9][16*16] ); + ALIGNED_ARRAY_N( pixel, pixv_buf,[2],[9][16*16] ); + pixel *src[3][2][9]; + int chromapix = h->luma2chroma_pixel[i_pixel]; + int chroma_v_shift = CHROMA_V_SHIFT; + int chroma_x = (8 >> CHROMA_H_SHIFT) * x; + int chroma_y = (8 >> chroma_v_shift) * y; pixel *pix = &h->mb.pic.p_fdec[0][8*x + 8*y*FDEC_STRIDE]; - pixel *pixu = &h->mb.pic.p_fdec[1][4*x + 4*y*FDEC_STRIDE]; - pixel *pixv = &h->mb.pic.p_fdec[2][4*x + 4*y*FDEC_STRIDE]; + pixel *pixu = &h->mb.pic.p_fdec[1][chroma_x + chroma_y*FDEC_STRIDE]; + pixel *pixv = &h->mb.pic.p_fdec[2][chroma_x + chroma_y*FDEC_STRIDE]; int ref0 = h->mb.cache.ref[0][s8]; int ref1 = h->mb.cache.ref[1][s8]; - const int mv0y_offset = h->mb.b_interlaced & ref0 ? (h->mb.i_mb_y & 1)*4 - 2 : 0; - const int mv1y_offset = h->mb.b_interlaced & ref1 ? (h->mb.i_mb_y & 1)*4 - 2 : 0; - int stride[2][9]; + const int mv0y_offset = chroma_v_shift & MB_INTERLACED & ref0 ? (h->mb.i_mb_y & 1)*4 - 2 : 0; + const int mv1y_offset = chroma_v_shift & MB_INTERLACED & ref1 ? (h->mb.i_mb_y & 1)*4 - 2 : 0; + intptr_t stride[3][2][9]; int bm0x = m0->mv[0]; int bm0y = m0->mv[1]; int bm1x = m1->mv[0]; @@ -965,9 +1059,9 @@ static void ALWAYS_INLINE x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_m uint64_t bcostrd = COST_MAX64; uint16_t amvd; /* each byte of visited represents 8 possible m1y positions, so a 4D array isn't needed */ - ALIGNED_ARRAY_16( uint8_t, visited,[8],[8][8] ); + ALIGNED_ARRAY_N( uint8_t, visited,[8],[8][8] ); /* all permutations of an offset in up to 2 of the dimensions */ - static const int8_t dia4d[33][4] = + ALIGNED_4( static const int8_t dia4d[33][4] ) = { {0,0,0,0}, {0,0,0,1}, {0,0,0,-1}, {0,0,1,0}, {0,0,-1,0}, @@ -1025,18 +1119,26 @@ static void ALWAYS_INLINE x264_me_refine_bidir( x264_t *h, x264_me_t *m0, x264_m int i0 = 4 + 3*dia4d[j][0] + dia4d[j][1]; int i1 = 4 + 3*dia4d[j][2] + dia4d[j][3]; visited[(m0x)&7][(m0y)&7][(m1x)&7] |= (1<<((m1y)&7)); - h->mc.avg[i_pixel]( pix, FDEC_STRIDE, src[0][i0], stride[0][i0], src[1][i1], stride[1][i1], i_weight ); + h->mc.avg[i_pixel]( pix, FDEC_STRIDE, src[0][0][i0], stride[0][0][i0], src[0][1][i1], stride[0][1][i1], i_weight ); int cost = h->pixf.mbcmp[i_pixel]( m0->p_fenc[0], FENC_STRIDE, pix, FDEC_STRIDE ) + p_cost_m0x[m0x] + p_cost_m0y[m0y] + p_cost_m1x[m1x] + p_cost_m1y[m1y]; if( rd ) { - if( cost < bcost * SATD_THRESH ) + if( cost < SATD_THRESH(bcost) ) { bcost = X264_MIN( cost, bcost ); M32( cache0_mv ) = pack16to32_mask(m0x,m0y); M32( cache1_mv ) = pack16to32_mask(m1x,m1y); - h->mc.avg[i_pixel+3]( pixu, FDEC_STRIDE, pixu_buf[0][i0], 8, pixu_buf[1][i1], 8, i_weight ); - h->mc.avg[i_pixel+3]( pixv, FDEC_STRIDE, pixv_buf[0][i0], 8, pixv_buf[1][i1], 8, i_weight ); + if( CHROMA444 ) + { + h->mc.avg[i_pixel]( pixu, FDEC_STRIDE, src[1][0][i0], stride[1][0][i0], src[1][1][i1], stride[1][1][i1], i_weight ); + h->mc.avg[i_pixel]( pixv, FDEC_STRIDE, src[2][0][i0], stride[2][0][i0], src[2][1][i1], stride[2][1][i1], i_weight ); + } + else + { + h->mc.avg[chromapix]( pixu, FDEC_STRIDE, pixu_buf[0][i0], 8, pixu_buf[1][i1], 8, i_weight ); + h->mc.avg[chromapix]( pixv, FDEC_STRIDE, pixv_buf[0][i0], 8, pixv_buf[1][i1], 8, i_weight ); + } uint64_t costrd = x264_rd_cost_part( h, i_lambda2, i8*4, m0->i_pixel ); COPY2_IF_LT( bcostrd, costrd, bestj, j ); } @@ -1105,12 +1207,24 @@ void x264_me_refine_bidir_rd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_wei #define COST_MV_RD( mx, my, satd, do_dir, mdir ) \ { \ - if( satd <= bsatd * SATD_THRESH ) \ + if( satd <= SATD_THRESH(bsatd) ) \ { \ uint64_t cost; \ M32( cache_mv ) = pack16to32_mask(mx,my); \ - if( m->i_pixel <= PIXEL_8x8 ) \ - h->mc.mc_chroma( pixu, pixv, FDEC_STRIDE, m->p_fref[4], m->i_stride[1], mx, my + mvy_offset, bw>>1, bh>>1 ); \ + if( CHROMA444 ) \ + { \ + h->mc.mc_luma( pixu, FDEC_STRIDE, &m->p_fref[4], m->i_stride[1], mx, my, bw, bh, &m->weight[1] ); \ + h->mc.mc_luma( pixv, FDEC_STRIDE, &m->p_fref[8], m->i_stride[2], mx, my, bw, bh, &m->weight[2] ); \ + } \ + else if( m->i_pixel <= PIXEL_8x8 ) \ + { \ + h->mc.mc_chroma( pixu, pixv, FDEC_STRIDE, m->p_fref[4], m->i_stride[1], \ + mx, 2*(my+mvy_offset)>>chroma_v_shift, bw>>1, bh>>chroma_v_shift ); \ + if( m->weight[1].weightfn ) \ + m->weight[1].weightfn[bw>>3]( pixu, FDEC_STRIDE, pixu, FDEC_STRIDE, &m->weight[1], bh>>chroma_v_shift ); \ + if( m->weight[2].weightfn ) \ + m->weight[2].weightfn[bw>>3]( pixv, FDEC_STRIDE, pixv, FDEC_STRIDE, &m->weight[2], bh>>chroma_v_shift ); \ + } \ cost = x264_rd_cost_part( h, i_lambda2, i4, m->i_pixel ); \ COPY4_IF_LT( bcost, cost, bmx, mx, bmy, my, dir, do_dir?mdir:dir ); \ } \ @@ -1123,21 +1237,30 @@ void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i4, int const int bw = x264_pixel_size[m->i_pixel].w; const int bh = x264_pixel_size[m->i_pixel].h; const int i_pixel = m->i_pixel; - const int mvy_offset = h->mb.b_interlaced & m->i_ref ? (h->mb.i_mb_y & 1)*4 - 2 : 0; + int chroma_v_shift = CHROMA_V_SHIFT; + int mvy_offset = chroma_v_shift & MB_INTERLACED & m->i_ref ? (h->mb.i_mb_y & 1)*4 - 2 : 0; uint64_t bcost = COST_MAX64; int bmx = m->mv[0]; int bmy = m->mv[1]; int omx, omy, pmx, pmy; - unsigned bsatd; - int satd; + int satd, bsatd; int dir = -2; int i8 = i4>>2; uint16_t amvd; pixel *pix = &h->mb.pic.p_fdec[0][block_idx_xy_fdec[i4]]; - pixel *pixu = &h->mb.pic.p_fdec[1][(i8>>1)*4*FDEC_STRIDE+(i8&1)*4]; - pixel *pixv = &h->mb.pic.p_fdec[2][(i8>>1)*4*FDEC_STRIDE+(i8&1)*4]; + pixel *pixu, *pixv; + if( CHROMA444 ) + { + pixu = &h->mb.pic.p_fdec[1][block_idx_xy_fdec[i4]]; + pixv = &h->mb.pic.p_fdec[2][block_idx_xy_fdec[i4]]; + } + else + { + pixu = &h->mb.pic.p_fdec[1][(i8>>1)*(8*FDEC_STRIDE>>chroma_v_shift)+(i8&1)*4]; + pixv = &h->mb.pic.p_fdec[2][(i8>>1)*(8*FDEC_STRIDE>>chroma_v_shift)+(i8&1)*4]; + } h->mb.b_skip_mc = 1; @@ -1221,7 +1344,7 @@ void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i4, int m->mv[0] = bmx; m->mv[1] = bmy; x264_macroblock_cache_mv ( h, block_idx_x[i4], block_idx_y[i4], bw>>2, bh>>2, i_list, pack16to32_mask(bmx, bmy) ); - amvd = pack8to16( X264_MIN(abs(bmx - m->mvp[0]),33), X264_MIN(abs(bmy - m->mvp[1]),33) ); + amvd = pack8to16( X264_MIN(abs(bmx - m->mvp[0]),66), X264_MIN(abs(bmy - m->mvp[1]),66) ); x264_macroblock_cache_mvd( h, block_idx_x[i4], block_idx_y[i4], bw>>2, bh>>2, i_list, amvd ); h->mb.b_skip_mc = 0; }