Sub-8x8 Qpel-RD in P-frames

Improves quality when using p8x4/p4x8/p4x4 subpartitions
Benefit is proportional to how many sub-8x8 partitions are used; helps most at high bitrates and low resolutions.

diff --git a/encoder/analyse.c b/encoder/analyse.c
index 05ab9629983170be17d6ab126858c74a3574100b..91b3cb9ebb384208d1dd0ebc56f58b2f441af156 100644 (file)
--- a/encoder/analyse.c
+++ b/encoder/analyse.c
@@ -2377,7 +2377,7 @@ void x264_macroblock_analyse( x264_t *h )
                     x264_macroblock_cache_ref( h, 0, 0, 4, 2, 0, analysis.l0.me16x8[0].i_ref );
                     x264_macroblock_cache_ref( h, 0, 2, 4, 2, 0, analysis.l0.me16x8[1].i_ref );
                     x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[0], analysis.i_lambda2, 0, 0 );
-                    x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[1], analysis.i_lambda2, 2, 0 );
+                    x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[1], analysis.i_lambda2, 8, 0 );
                 }
                 else if( i_partition == D_8x16 )
                 {
@@ -2386,15 +2386,36 @@ void x264_macroblock_analyse( x264_t *h )
                     x264_macroblock_cache_ref( h, 0, 0, 2, 4, 0, analysis.l0.me8x16[0].i_ref );
                     x264_macroblock_cache_ref( h, 2, 0, 2, 4, 0, analysis.l0.me8x16[1].i_ref );
                     x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[0], analysis.i_lambda2, 0, 0 );
-                    x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[1], analysis.i_lambda2, 1, 0 );
+                    x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[1], analysis.i_lambda2, 4, 0 );
                 }
                 else if( i_partition == D_8x8 )
                 {
                     int i8x8;
                     x264_analyse_update_cache( h, &analysis );
                     for( i8x8 = 0; i8x8 < 4; i8x8++ )
-                         if( h->mb.i_sub_partition[i8x8] == D_L0_8x8 )
-                             x264_me_refine_qpel_rd( h, &analysis.l0.me8x8[i8x8], analysis.i_lambda2, i8x8, 0 );
+                    {
+                        if( h->mb.i_sub_partition[i8x8] == D_L0_8x8 )
+                        {
+                            x264_me_refine_qpel_rd( h, &analysis.l0.me8x8[i8x8], analysis.i_lambda2, i8x8*4, 0 );
+                        }
+                        else if( h->mb.i_sub_partition[i8x8] == D_L0_8x4 )
+                        {
+                           x264_me_refine_qpel_rd( h, &analysis.l0.me8x4[i8x8][0], analysis.i_lambda2, i8x8*4+0, 0 );
+                           x264_me_refine_qpel_rd( h, &analysis.l0.me8x4[i8x8][1], analysis.i_lambda2, i8x8*4+2, 0 );
+                        }
+                        else if( h->mb.i_sub_partition[i8x8] == D_L0_4x8 )
+                        {
+                           x264_me_refine_qpel_rd( h, &analysis.l0.me4x8[i8x8][0], analysis.i_lambda2, i8x8*4+0, 0 );
+                           x264_me_refine_qpel_rd( h, &analysis.l0.me4x8[i8x8][1], analysis.i_lambda2, i8x8*4+1, 0 );
+                        }
+                        else if( h->mb.i_sub_partition[i8x8] == D_L0_4x4 )
+                        {
+                           x264_me_refine_qpel_rd( h, &analysis.l0.me4x4[i8x8][0], analysis.i_lambda2, i8x8*4+0, 0 );
+                           x264_me_refine_qpel_rd( h, &analysis.l0.me4x4[i8x8][1], analysis.i_lambda2, i8x8*4+1, 0 );
+                           x264_me_refine_qpel_rd( h, &analysis.l0.me4x4[i8x8][2], analysis.i_lambda2, i8x8*4+2, 0 );
+                           x264_me_refine_qpel_rd( h, &analysis.l0.me4x4[i8x8][3], analysis.i_lambda2, i8x8*4+3, 0 );
+                        }
+                    }
                 }
             }
         }
@@ -2663,9 +2684,9 @@ void x264_macroblock_analyse( x264_t *h )
                     {
                         h->mb.i_sub_partition[i*2] = h->mb.i_sub_partition[i*2+1] = analysis.i_mb_partition16x8[i];
                         if( analysis.i_mb_partition16x8[i] == D_L0_8x8 )
-                            x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[i], analysis.i_lambda2, i*2, 0 );
+                            x264_me_refine_qpel_rd( h, &analysis.l0.me16x8[i], analysis.i_lambda2, i*8, 0 );
                         else if( analysis.i_mb_partition16x8[i] == D_L1_8x8 )
-                            x264_me_refine_qpel_rd( h, &analysis.l1.me16x8[i], analysis.i_lambda2, i*2, 1 );
+                            x264_me_refine_qpel_rd( h, &analysis.l1.me16x8[i], analysis.i_lambda2, i*8, 1 );
                         else if( analysis.i_mb_partition16x8[i] == D_BI_8x8 )
                             x264_me_refine_bidir_rd( h, &analysis.l0.me16x8[i], &analysis.l1.me16x8[i], i_biweight, i*2, analysis.i_lambda2 );
                     }
@@ -2676,9 +2697,9 @@ void x264_macroblock_analyse( x264_t *h )
                     {
                         h->mb.i_sub_partition[i] = h->mb.i_sub_partition[i+2] = analysis.i_mb_partition8x16[i];
                         if( analysis.i_mb_partition8x16[i] == D_L0_8x8 )
-                            x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[i], analysis.i_lambda2, i, 0 );
+                            x264_me_refine_qpel_rd( h, &analysis.l0.me8x16[i], analysis.i_lambda2, i*4, 0 );
                         else if( analysis.i_mb_partition8x16[i] == D_L1_8x8 )
-                            x264_me_refine_qpel_rd( h, &analysis.l1.me8x16[i], analysis.i_lambda2, i, 1 );
+                            x264_me_refine_qpel_rd( h, &analysis.l1.me8x16[i], analysis.i_lambda2, i*4, 1 );
                         else if( analysis.i_mb_partition8x16[i] == D_BI_8x8 )
                             x264_me_refine_bidir_rd( h, &analysis.l0.me8x16[i], &analysis.l1.me8x16[i], i_biweight, i, analysis.i_lambda2 );
                     }
@@ -2688,9 +2709,9 @@ void x264_macroblock_analyse( x264_t *h )
                     for( i = 0; i < 4; i++ )
                     {
                         if( h->mb.i_sub_partition[i] == D_L0_8x8 )
-                            x264_me_refine_qpel_rd( h, &analysis.l0.me8x8[i], analysis.i_lambda2, i, 0 );
+                            x264_me_refine_qpel_rd( h, &analysis.l0.me8x8[i], analysis.i_lambda2, i*4, 0 );
                         else if( h->mb.i_sub_partition[i] == D_L1_8x8 )
-                            x264_me_refine_qpel_rd( h, &analysis.l1.me8x8[i], analysis.i_lambda2, i, 1 );
+                            x264_me_refine_qpel_rd( h, &analysis.l1.me8x8[i], analysis.i_lambda2, i*4, 1 );
                         else if( h->mb.i_sub_partition[i] == D_BI_8x8 )
                             x264_me_refine_bidir_rd( h, &analysis.l0.me8x8[i], &analysis.l1.me8x8[i], i_biweight, i, analysis.i_lambda2 );
                     }

diff --git a/encoder/cabac.c b/encoder/cabac.c
index 111b82ef559187f47a8cd2b0801d79fa62ae0a91..17863780400d91d154d2cd9c3afb3c47f55cc71f 100644 (file)
--- a/encoder/cabac.c
+++ b/encoder/cabac.c
@@ -1077,6 +1077,19 @@ static void x264_partition_size_cabac( x264_t *h, x264_cabac_t *cb, int i8, int
     }
 }
 
+static void x264_subpartition_size_cabac( x264_t *h, x264_cabac_t *cb, int i4, int i_pixel )
+{
+    int b_8x4 = i_pixel == PIXEL_8x4;
+    block_residual_write_cabac( h, cb, DCT_LUMA_4x4, i4, h->dct.luma4x4[i4], 16 );
+    if( i_pixel == PIXEL_4x4 )
+        x264_cabac_mb_mvd( h, cb, 0, i4, 1, 1 );
+    else
+    {
+        x264_cabac_mb_mvd( h, cb, 0, i4, 1+b_8x4, 2-b_8x4 );
+        block_residual_write_cabac( h, cb, DCT_LUMA_4x4, i4+2-b_8x4, h->dct.luma4x4[i4+2-b_8x4], 16 );
+    }
+}
+
 static void x264_partition_i8x8_size_cabac( x264_t *h, x264_cabac_t *cb, int i8, int i_mode )
 {
     const int i_pred = x264_mb_predict_intra4x4_mode( h, 4*i8 );

diff --git a/encoder/cavlc.c b/encoder/cavlc.c
index df084e2243a0b40a0b2c69f5659de46e302b319c..7f3588d4036bc41808fa172c03e7877c9359d72c 100644 (file)
--- a/encoder/cavlc.c
+++ b/encoder/cavlc.c
@@ -685,6 +685,24 @@ static int x264_partition_size_cavlc( x264_t *h, int i8, int i_pixel )
     return s.i_bits_encoded;
 }
 
+static int x264_subpartition_size_cavlc( x264_t *h, int i4, int i_pixel )
+{
+    bs_t s;
+    int b_8x4 = i_pixel == PIXEL_8x4;
+    s.i_bits_encoded = 0;
+    cavlc_mb_mvd( h, &s, 0, i4, 1+b_8x4 );
+    h->mb.cache.non_zero_count[x264_scan8[i4]] = array_non_zero_count( h->dct.luma4x4[i4] );
+    block_residual_write_cavlc( h, &s, i4, h->dct.luma4x4[i4], 16 );
+    if( i_pixel != PIXEL_4x4 )
+    {
+        i4 += 2-b_8x4;
+        h->mb.cache.non_zero_count[x264_scan8[i4]] = array_non_zero_count( h->dct.luma4x4[i4] );
+        block_residual_write_cavlc( h, &s, i4, h->dct.luma4x4[i4], 16 );
+    }
+
+    return s.i_bits_encoded;
+}
+
 static int cavlc_intra4x4_pred_size( x264_t *h, int i4, int i_mode )
 {
     if( x264_mb_predict_intra4x4_mode( h, i4 ) == x264_mb_pred_mode4x4_fix( i_mode ) )

diff --git a/encoder/macroblock.c b/encoder/macroblock.c
index 3bb4e0dd008c9502c8c0f44b423d00fa5c3ae604..a24ae54c52f4bd999a14e498312b4dc0e1bb3ea2 100644 (file)
--- a/encoder/macroblock.c
+++ b/encoder/macroblock.c
@@ -921,3 +921,33 @@ void x264_macroblock_encode_p8x8( x264_t *h, int i8 )
     h->mb.i_cbp_luma |= nnz8x8 << i8;
     h->mb.i_cbp_chroma = 0x02;
 }
+
+/*****************************************************************************
+ * RD only, luma only
+ *****************************************************************************/
+void x264_macroblock_encode_p4x4( x264_t *h, int i4 )
+{
+    int i_qp = h->mb.i_qp;
+    uint8_t *p_fenc = &h->mb.pic.p_fenc[0][block_idx_xy_fenc[i4]];
+    uint8_t *p_fdec = &h->mb.pic.p_fdec[0][block_idx_xy_fdec[i4]];
+    const int i_ref = h->mb.cache.ref[0][x264_scan8[i4]];
+    const int mvx   = x264_clip3( h->mb.cache.mv[0][x264_scan8[i4]][0], h->mb.mv_min[0], h->mb.mv_max[0] );
+    const int mvy   = x264_clip3( h->mb.cache.mv[0][x264_scan8[i4]][1], h->mb.mv_min[1], h->mb.mv_max[1] );
+
+    h->mc.mc_luma( p_fdec, FDEC_STRIDE, h->mb.pic.p_fref[0][i_ref], h->mb.pic.i_stride[0], mvx + 4*4*block_idx_x[i4], mvy + 4*4*block_idx_y[i4], 4, 4 );
+
+    if( h->mb.b_lossless )
+        h->zigzagf.sub_4x4( h->dct.luma4x4[i4], p_fenc, p_fdec );
+    else
+    {
+        DECLARE_ALIGNED_16( int16_t dct4x4[4][4] );
+        h->dctf.sub4x4_dct( dct4x4, p_fenc, p_fdec );
+        x264_quant_4x4( h, dct4x4, i_qp, DCT_LUMA_4x4, 0, i4 );
+        h->zigzagf.scan_4x4( h->dct.luma4x4[i4], dct4x4 );
+        if( array_non_zero( dct4x4 ) )
+        {
+            h->quantf.dequant_4x4( dct4x4, h->dequant4_mf[CQM_4PY], i_qp );
+            h->dctf.add4x4_idct( p_fdec, dct4x4 );
+        }
+    }
+}

diff --git a/encoder/macroblock.h b/encoder/macroblock.h
index b25509a895f349a329159072084ba7c0e44be5d4..4cc599aa37cba017f66b4326923ab7af92bfe2f4 100644 (file)
--- a/encoder/macroblock.h
+++ b/encoder/macroblock.h
@@ -48,6 +48,7 @@ void x264_macroblock_write_cabac ( x264_t *h, x264_cabac_t *cb );
 void x264_macroblock_write_cavlc ( x264_t *h, bs_t *s );
 
 void x264_macroblock_encode_p8x8( x264_t *h, int i8 );
+void x264_macroblock_encode_p4x4( x264_t *h, int i4 );
 void x264_mb_encode_i4x4( x264_t *h, int idx, int i_qp );
 void x264_mb_encode_i8x8( x264_t *h, int idx, int i_qp );
 void x264_mb_encode_8x8_chroma( x264_t *h, int b_inter, int i_qp );

diff --git a/encoder/me.c b/encoder/me.c
index 06a0415eb1dcb0e16d7c34a8d78652520e553074..8d1d4750f03e95984aa1b47fc78b462f4b2b73ae 100644 (file)
--- a/encoder/me.c
+++ b/encoder/me.c
@@ -944,7 +944,7 @@ void x264_me_refine_bidir_rd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_wei
 #undef COST_MV_SATD
 #define COST_MV_SATD( mx, my, dst, avoid_mvp ) \
 { \
-    if( !avoid_pmv || !(mx == pmx && my == pmy) ) \
+    if( !avoid_mvp || !(mx == pmx && my == pmy) ) \
     { \
         int stride = 16; \
         uint8_t *src = h->mc.get_ref( pix, &stride, m->p_fref, m->i_stride[0], mx, my, bw*4, bh*4 ); \
@@ -962,16 +962,16 @@ void x264_me_refine_bidir_rd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_wei
     { \
         uint64_t cost; \
         *(uint32_t*)cache_mv = *(uint32_t*)cache_mv2 = pack16to32_mask(mx,my); \
-        cost = x264_rd_cost_part( h, i_lambda2, i8, m->i_pixel ); \
+        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 ); \
     } \
 }
 
-void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8, int i_list )
+void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i4, int i_list )
 {
     // don't have to fill the whole mv cache rectangle
-    static const int pixel_mv_offs[] = { 0, 4, 4*8, 0 };
-    int16_t *cache_mv = h->mb.cache.mv[i_list][x264_scan8[i8*4]];
+    static const int pixel_mv_offs[] = { 0, 4, 4*8, 0, 2, 2*8, 0 };
+    int16_t *cache_mv = h->mb.cache.mv[i_list][x264_scan8[i4]];
     int16_t *cache_mv2 = cache_mv + pixel_mv_offs[m->i_pixel];
     const int16_t *p_cost_mvx, *p_cost_mvy;
     const int bw = x264_pixel_size[m->i_pixel].w>>2;
@@ -990,8 +990,8 @@ void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8, int
     int dir = -2;
     int satds[8];
 
-    if( m->i_pixel != PIXEL_16x16 && i8 != 0 )
-        x264_mb_predict_mv( h, i_list, i8*4, bw, m->mvp );
+    if( m->i_pixel != PIXEL_16x16 && i4 != 0 )
+        x264_mb_predict_mv( h, i_list, i4, bw, m->mvp );
     pmx = m->mvp[0];
     pmy = m->mvp[1];
     p_cost_mvx = m->p_cost_mv - pmx;
@@ -1051,7 +1051,6 @@ void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8, int
     m->cost = bcost;
     m->mv[0] = bmx;
     m->mv[1] = bmy;
-    x264_macroblock_cache_mv ( h, 2*(i8&1), i8&2, bw, bh, i_list, pack16to32_mask(bmx, bmy) );
-    x264_macroblock_cache_mvd( h, 2*(i8&1), i8&2, bw, bh, i_list, pack16to32_mask(bmx - m->mvp[0], bmy - m->mvp[1]) );
+    x264_macroblock_cache_mv ( h, block_idx_x[i4], block_idx_y[i4], bw, bh, i_list, pack16to32_mask(bmx, bmy) );
+    x264_macroblock_cache_mvd( h, block_idx_x[i4], block_idx_y[i4], bw, bh, i_list, pack16to32_mask(bmx - m->mvp[0], bmy - m->mvp[1]) );
 }
-

diff --git a/encoder/me.h b/encoder/me.h
index 7b94a19799f0fa1def81540d05cf51b4c51a3150..eba4957a9a9865b24a47eeac2ce029ef455d4698 100644 (file)
--- a/encoder/me.h
+++ b/encoder/me.h
@@ -53,7 +53,7 @@ static inline void x264_me_search( x264_t *h, x264_me_t *m, int16_t (*mvc)[2], i
     { x264_me_search_ref( h, m, mvc, i_mvc, NULL ); }
 
 void x264_me_refine_qpel( x264_t *h, x264_me_t *m );
-void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i8, int i_list );
+void x264_me_refine_qpel_rd( x264_t *h, x264_me_t *m, int i_lambda2, int i4, int i_list );
 void x264_me_refine_bidir_rd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight, int i8, int i_lambda2 );
 void x264_me_refine_bidir_satd( x264_t *h, x264_me_t *m0, x264_me_t *m1, int i_weight );
 uint64_t x264_rd_cost_part( x264_t *h, int i_lambda2, int i8, int i_pixel );

diff --git a/encoder/rdo.c b/encoder/rdo.c
index 296a3f386c8726dc08768b327543b9658b3cbde5..accc8d30763a6b4625765f7e139fffa4201d5acc 100644 (file)
--- a/encoder/rdo.c
+++ b/encoder/rdo.c
@@ -159,12 +159,40 @@ static int x264_rd_cost_mb( x264_t *h, int i_lambda2 )
     return i_ssd + i_bits;
 }
 
-/* subpartition RD functions use 8 bits more precision to avoid large rounding errors at low QPs */
+/* partition RD functions use 8 bits more precision to avoid large rounding errors at low QPs */
 
-uint64_t x264_rd_cost_part( x264_t *h, int i_lambda2, int i8, int i_pixel )
+static uint64_t x264_rd_cost_subpart( x264_t *h, int i_lambda2, int i4, int i_pixel )
 {
     uint64_t i_ssd, i_bits;
 
+    x264_macroblock_encode_p4x4( h, i4 );
+    if( i_pixel == PIXEL_8x4 )
+        x264_macroblock_encode_p4x4( h, i4+1 );
+    if( i_pixel == PIXEL_4x8 )
+        x264_macroblock_encode_p4x4( h, i4+2 );
+
+    i_ssd = ssd_plane( h, i_pixel, 0, block_idx_x[i4]*4, block_idx_y[i4]*4 );
+
+    if( h->param.b_cabac )
+    {
+        x264_cabac_t cabac_tmp;
+        COPY_CABAC;
+        x264_subpartition_size_cabac( h, &cabac_tmp, i4, i_pixel );
+        i_bits = ( (uint64_t)cabac_tmp.f8_bits_encoded * i_lambda2 + 128 ) >> 8;
+    }
+    else
+    {
+        i_bits = x264_subpartition_size_cavlc( h, i4, i_pixel );
+    }
+
+    return (i_ssd<<8) + i_bits;
+}
+
+uint64_t x264_rd_cost_part( x264_t *h, int i_lambda2, int i4, int i_pixel )
+{
+    uint64_t i_ssd, i_bits;
+    int i8 = i4 >> 2;
+
     if( i_pixel == PIXEL_16x16 )
     {
         int type_bak = h->mb.i_type;
@@ -173,6 +201,9 @@ uint64_t x264_rd_cost_part( x264_t *h, int i_lambda2, int i8, int i_pixel )
         return i_cost;
     }
 
+    if( i_pixel > PIXEL_8x8 )
+        return x264_rd_cost_subpart( h, i_lambda2, i4, i_pixel );
+
     x264_macroblock_encode_p8x8( h, i8 );
     if( i_pixel == PIXEL_16x8 )
         x264_macroblock_encode_p8x8( h, i8+1 );