[FFmpeg-devel] [PATCH] RV40 Decoder - 1/3 - Loop Filter

Kostya kostya.shishkov
Fri Nov 28 19:17:17 CET 2008


On Fri, Nov 28, 2008 at 09:19:19AM +0100, Michael Niedermayer wrote:
> On Fri, Nov 28, 2008 at 09:03:04AM +0200, Kostya wrote:
> > On Thu, Nov 27, 2008 at 10:04:40PM +0100, Michael Niedermayer wrote:
> > > On Thu, Nov 27, 2008 at 09:23:21AM +0200, Kostya wrote:
> > > > $subj
> > > > 
> > > > Fixed several bugs in chroma loop filtering introduced during
> > > > refactoring, now it's bitexact again.
> > > 
> > > [...]
> > > 
> > > > +    for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++){
> > > > +        mb_pos = s->mb_y * s->mb_stride;
> > > > +        for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++, mb_pos++){
> > > > +            int mbtype = s->current_picture_ptr->mb_type[mb_pos];
> > > > +            if(IS_INTRA(mbtype) || IS_SEPARATE_DC(mbtype))
> > > > +                r->cbp_luma  [mb_pos] = 0xFFFF;
> > > > +            if(IS_INTRA(mbtype))
> > > > +                r->cbp_chroma[mb_pos] = 0xFF;
> > > > +        }
> > > > +    }
> > > > +    for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++){
> > > > +        mb_pos = s->mb_y * s->mb_stride;
> > > > +        for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++, mb_pos++){
> > > 
> > > loop filtering should be done after each block or row of blocks has been
> > > decoded to maximize the chance that things still are in the cache
> > 
> > I'd rather leave it as is since loop filter affects bottom macroblock edge
> > too so filtering the row above just decoded one is a bit strange.
> 
> please post benchmark scores for both, if theres no speed gain i dont mind
> leaving it
 
Hmm, there is ~1% gain, so row-based deblocking it is.
 
> [...]
> 
> -- 
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
-------------- next part --------------
Index: libavcodec/rv40.c
===================================================================
--- libavcodec/rv40.c	(revision 15920)
+++ libavcodec/rv40.c	(working copy)
@@ -389,7 +389,244 @@
                               alpha, beta, beta2, chroma, edge);
 }
 
+enum RV40BlockPos{
+    POS_CUR,
+    POS_TOP,
+    POS_LEFT,
+    POS_BOTTOM,
+};
+
+#define MASK_CUR          0x0001
+#define MASK_RIGHT        0x0008
+#define MASK_BOTTOM       0x0010
+#define MASK_TOP          0x1000
+#define MASK_Y_TOP_ROW    0x000F
+#define MASK_Y_LAST_ROW   0xF000
+#define MASK_Y_LEFT_COL   0x1111
+#define MASK_Y_RIGHT_COL  0x8888
+#define MASK_C_TOP_ROW    0x0003
+#define MASK_C_LAST_ROW   0x000C
+#define MASK_C_LEFT_COL   0x0005
+#define MASK_C_RIGHT_COL  0x000A
+
+static const int neighbour_offs_x[4] = { 0,  0, -1, 0 };
+static const int neighbour_offs_y[4] = { 0, -1,  0, 1 };
+
 /**
+ * RV40 loop filtering function
+ */
+static void rv40_loop_filter(RV34DecContext *r, int row)
+{
+    MpegEncContext *s = &r->s;
+    int mb_pos, mb_x;
+    int i, j, k;
+    uint8_t *Y, *C;
+    int alpha, beta, betaY, betaC;
+    int q;
+    int mbtype[4];   ///< current macroblock and its neighbours types
+    /**
+     * flags indicating that macroblock can be filtered with strong filter
+     * it is set only for intra coded MB and MB with DCs coded separately
+     */
+    int mb_strong[4];
+    int clip[4];     ///< MB filter clipping value calculated from filtering strength
+    /**
+     * coded block patterns for luma part of current macroblock and its neighbours
+     * Format:
+     * LSB corresponds to the top left block,
+     * each nibble represents one row of subblocks.
+     */
+    int cbp[4];
+    /**
+     * coded block patterns for chroma part of current macroblock and its neighbours
+     * Format is the same as for luma with two subblocks in a row.
+     */
+    int uvcbp[4][2];
+    /**
+     * This mask represents the pattern of luma subblocks that should be filtered
+     * in addition to the coded ones because because they lie at the edge of
+     * 8x8 block with different enough motion vectors
+     */
+    int mvmasks[4];
+
+    mb_pos = row * s->mb_stride;
+    for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
+        int mbtype = s->current_picture_ptr->mb_type[mb_pos];
+        if(IS_INTRA(mbtype) || IS_SEPARATE_DC(mbtype))
+            r->cbp_luma  [mb_pos] = 0xFFFF;
+        if(IS_INTRA(mbtype))
+            r->cbp_chroma[mb_pos] = 0xFF;
+    }
+    mb_pos = row * s->mb_stride;
+    for(mb_x = 0; mb_x < s->mb_width; mb_x++, mb_pos++){
+        int y_h_deblock, y_v_deblock;
+        int c_v_deblock[2], c_h_deblock[2];
+        int clip_left;
+        int avail[4];
+        int y_to_deblock, c_to_deblock[2];
+
+        q = s->current_picture_ptr->qscale_table[mb_pos];
+        alpha = rv40_alpha_tab[q];
+        beta  = rv40_beta_tab [q];
+        betaY = betaC = beta * 3;
+        if(s->width * s->height <= 176*144)
+            betaY += beta;
+
+        avail[0] = 1;
+        avail[1] = row;
+        avail[2] = mb_x;
+        avail[3] = row < s->mb_height - 1;
+        for(i = 0; i < 4; i++){
+            if(avail[i]){
+                int pos = mb_pos + neighbour_offs_x[i] + neighbour_offs_y[i]*s->mb_stride;
+                mvmasks[i] = r->deblock_coefs[pos];
+                mbtype [i] = s->current_picture_ptr->mb_type[pos];
+                cbp    [i] = r->cbp_luma[pos];
+                uvcbp[i][0] = r->cbp_chroma[pos] & 0xF;
+                uvcbp[i][1] = r->cbp_chroma[pos] >> 4;
+            }else{
+                mvmasks[i] = 0;
+                mbtype [i] = mbtype[0];
+                cbp    [i] = 0;
+                uvcbp[i][0] = uvcbp[i][1] = 0;
+            }
+            mb_strong[i] = IS_INTRA(mbtype[i]) || IS_SEPARATE_DC(mbtype[i]);
+            clip[i] = rv40_filter_clip_tbl[mb_strong[i] + 1][q];
+        }
+        y_to_deblock =  cbp[POS_CUR]
+                     | (cbp[POS_BOTTOM]     << 16)
+                     |  mvmasks[POS_CUR]
+                     | (mvmasks[POS_BOTTOM] << 16);
+        /* This pattern contains bits signalling that horizontal edges of
+         * the current block can be filtered.
+         * That happens when either of adjacent subblocks is coded or lies on
+         * the edge of 8x8 blocks with motion vectors differing by more than
+         * 3/4 pel in any component (any edge orientation for some reason).
+         */
+        y_h_deblock =   y_to_deblock
+                    | ((cbp[POS_CUR]                           <<  4) & ~MASK_Y_TOP_ROW)
+                    | ((cbp[POS_TOP]        & MASK_Y_LAST_ROW) >> 12);
+        /* This pattern contains bits signalling that vertical edges of
+         * the current block can be filtered.
+         * That happens when either of adjacent subblocks is coded or lies on
+         * the edge of 8x8 blocks with motion vectors differing by more than
+         * 3/4 pel in any component (any edge orientation for some reason).
+         */
+        y_v_deblock =   y_to_deblock
+                    | ((cbp[POS_CUR]                      << 1) & ~MASK_Y_LEFT_COL)
+                    | ((cbp[POS_LEFT] & MASK_Y_RIGHT_COL) >> 3);
+        if(!mb_x)
+            y_v_deblock &= ~MASK_Y_LEFT_COL;
+        if(!row)
+            y_h_deblock &= ~MASK_Y_TOP_ROW;
+        if(row == s->mb_height - 1 || (mb_strong[POS_CUR] || mb_strong[POS_BOTTOM]))
+            y_h_deblock &= ~(MASK_Y_TOP_ROW << 16);
+        /* Calculating chroma patterns is similar and easier since there is
+         * no motion vector pattern for them.
+         */
+        for(i = 0; i < 2; i++){
+            c_to_deblock[i] = (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i];
+            c_v_deblock[i] =   c_to_deblock[i]
+                           | ((uvcbp[POS_CUR] [i]                       << 1) & ~MASK_C_LEFT_COL)
+                           | ((uvcbp[POS_LEFT][i]   & MASK_C_RIGHT_COL) >> 1);
+            c_h_deblock[i] =   c_to_deblock[i]
+                           | ((uvcbp[POS_TOP][i]    & MASK_C_LAST_ROW)  >> 2)
+                           |  (uvcbp[POS_CUR][i]                        << 2);
+            if(!mb_x)
+                c_v_deblock[i] &= ~MASK_C_LEFT_COL;
+            if(!row)
+                c_h_deblock[i] &= ~MASK_C_TOP_ROW;
+            if(row == s->mb_height - 1 || mb_strong[POS_CUR] || mb_strong[POS_BOTTOM])
+                c_h_deblock[i] &= ~(MASK_C_TOP_ROW << 4);
+        }
+
+        for(j = 0; j < 16; j += 4){
+            Y = s->current_picture_ptr->data[0] + mb_x*16 + (row*16 + j) * s->linesize;
+            for(i = 0; i < 4; i++, Y += 4){
+                int ij = i + j;
+                int clip_cur = y_to_deblock & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
+                int dither = j ? ij : i*4;
+
+                // if bottom block is coded then we can filter its top edge
+                // (or bottom edge of this block, which is the same)
+                if(y_h_deblock & (MASK_BOTTOM << ij)){
+                    rv40_h_loop_filter(Y+4*s->linesize, s->linesize, dither,
+                                       y_to_deblock & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
+                                       clip_cur,
+                                       alpha, beta, betaY, 0, 0);
+                }
+                // filter left block edge in ordinary mode (with low filtering strength)
+                if(y_v_deblock & (MASK_CUR << ij) && (i || !(mb_strong[POS_CUR] || mb_strong[POS_LEFT]))){
+                    if(!i)
+                        clip_left = (cbp[POS_LEFT] | mvmasks[POS_LEFT]) & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
+                    else
+                        clip_left = y_to_deblock & (MASK_CUR << (ij-1)) ? clip[POS_CUR] : 0;
+                    rv40_v_loop_filter(Y, s->linesize, dither,
+                                       clip_cur,
+                                       clip_left,
+                                       alpha, beta, betaY, 0, 0);
+                }
+                // filter top edge of the current macroblock when filtering strength is high
+                if(!j && y_h_deblock & (MASK_CUR << i) && (mb_strong[POS_CUR] || mb_strong[POS_TOP])){
+                    rv40_h_loop_filter(Y, s->linesize, dither,
+                                       clip_cur,
+                                       (cbp[POS_TOP] | mvmasks[POS_TOP]) & (MASK_TOP << i) ? clip[POS_TOP] : 0,
+                                       alpha, beta, betaY, 0, 1);
+                }
+                // filter left block edge in edge mode (with high filtering strength)
+                if(y_v_deblock & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] || mb_strong[POS_LEFT])){
+                    clip_left = (cbp[POS_LEFT] | mvmasks[POS_LEFT]) & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
+                    rv40_v_loop_filter(Y, s->linesize, dither,
+                                       clip_cur,
+                                       clip_left,
+                                       alpha, beta, betaY, 0, 1);
+                }
+            }
+        }
+        for(k = 0; k < 2; k++){
+            for(j = 0; j < 2; j++){
+                C = s->current_picture_ptr->data[k+1] + mb_x*8 + (row*8 + j*4) * s->uvlinesize;
+                for(i = 0; i < 2; i++, C += 4){
+                    int ij = i + j*2;
+                    int clip_cur = c_to_deblock[k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
+                    if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
+                        int clip_bot = c_to_deblock[k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
+                        rv40_h_loop_filter(C+4*s->uvlinesize, s->uvlinesize, i*8,
+                                           clip_bot,
+                                           clip_cur,
+                                           alpha, beta, betaC, 1, 0);
+                    }
+                    if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(mb_strong[POS_CUR] || mb_strong[POS_LEFT]))){
+                        if(!i)
+                            clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
+                        else
+                            clip_left = c_to_deblock[k]    & (MASK_CUR << (ij-1))  ? clip[POS_CUR]  : 0;
+                        rv40_v_loop_filter(C, s->uvlinesize, j*8,
+                                           clip_cur,
+                                           clip_left,
+                                           alpha, beta, betaC, 1, 0);
+                    }
+                    if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (mb_strong[POS_CUR] || mb_strong[POS_TOP])){
+                        int clip_top = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
+                        rv40_h_loop_filter(C, s->uvlinesize, i*8,
+                                           clip_cur,
+                                           clip_top,
+                                           alpha, beta, betaC, 1, 1);
+                    }
+                    if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (mb_strong[POS_CUR] || mb_strong[POS_LEFT])){
+                        clip_left = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
+                        rv40_v_loop_filter(C, s->uvlinesize, j*8,
+                                           clip_cur,
+                                           clip_left,
+                                           alpha, beta, betaC, 1, 1);
+                    }
+                }
+            }
+        }
+    }
+}
+
+/**
  * Initialize decoder.
  */
 static av_cold int rv40_decode_init(AVCodecContext *avctx)
@@ -403,6 +640,7 @@
     r->parse_slice_header = rv40_parse_slice_header;
     r->decode_intra_types = rv40_decode_intra_types;
     r->decode_mb_info     = rv40_decode_mb_info;
+    r->loop_filter        = rv40_loop_filter;
     r->luma_dc_quant_i = rv40_luma_dc_quant[0];
     r->luma_dc_quant_p = rv40_luma_dc_quant[1];
     return 0;



More information about the ffmpeg-devel mailing list