[FFmpeg-devel] [PATCH] RV40 loop filter, try 2

Kostya kostya.shishkov
Thu Oct 30 15:51:19 CET 2008


$subj

I've tried to generalize, clean and document it.
-------------- next part --------------
Index: libavcodec/rv40.c
===================================================================
--- libavcodec/rv40.c	(revision 15732)
+++ libavcodec/rv40.c	(working copy)
@@ -247,7 +247,427 @@
     return 0;
 }
 
+#define CLIP_SYMM(a, b) av_clip(a, -(b), b)
 /**
+ * weaker deblocking very similar to the one described in 4.4.2 of JVT-A003r1
+ */
+static inline void rv40_weak_loop_filter(uint8_t *src, const int step,
+                                         const int filter_outer, const int filter_inner,
+                                         const int alpha,
+                                         const int lim_inner, const int lim_outer,
+                                         const int difflim, const int beta,
+                                         const int S0, const int S1,
+                                         const int S2, const int S3)
+{
+    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
+    int t, u, diff;
+
+    t = src[0*step] - src[-1*step];
+    if(!t){
+        return;
+    }
+    u = (alpha * FFABS(t)) >> 7;
+    if(u > 3 - (filter_inner && filter_outer)){
+        return;
+    }
+
+    t <<= 2;
+    if(filter_inner && filter_outer)
+        t += src[-2*step] - src[1*step];
+    diff = CLIP_SYMM((t + 4) >> 3, difflim);
+    src[-1*step] = cm[src[-1*step] + diff];
+    src[ 0*step] = cm[src[ 0*step] - diff];
+    if(FFABS(S2) <= beta && filter_outer){
+        t = (S0 + S2 - diff) >> 1;
+        src[-2*step] = cm[src[-2*step] - CLIP_SYMM(t, lim_outer)];
+    }
+    if(FFABS(S3) <= beta && filter_inner){
+        t = (S1 + S3 + diff) >> 1;
+        src[ 1*step] = cm[src[ 1*step] - CLIP_SYMM(t, lim_inner)];
+    }
+}
+
+/**
+ * This macro is used for calculating 25*x0+26*x1+26*x2+26*x3+25*x4
+ * or 25*x0+26*x1+51*x2+26*x3
+ *
+ * @param idx25 index of the value with coefficient = 25 (could be at the end of
+                coefficients or at the start)
+ * @param idx25_51 index of the value that takes coefficient 25 when this index
+                is before strt index or after (start+3) and 51 when it falls into
+                that range
+ */
+#define RV40_STRONG_FILTER(src, step, start, idx25_51, idx25) \
+     26*(src[start    *step] + src[(start+1)*step]  + src[(start+2)*step] \
+       + src[(start+3)*step] + src[idx25    *step]) - src[idx25_51 *step] \
+       - src[idx25    *step]
+
+/**
+ * Deblocking filter, the altered version from JVT-A003r1 H.26L draft.
+ */
+static inline void rv40_adaptive_loop_filter(uint8_t *src, const int step,
+                                             const int stride, const int dmode,
+                                             const int lim_inner, const int lim_outer,
+                                             const int alpha,
+                                             const int beta, const int beta2,
+                                             const int chroma, const int edge)
+{
+    int diffs[4][4];
+    int s0 = 0, s1 = 0, s2 = 0, s3 = 0;
+    uint8_t *ptr;
+    int flag_strong0 = 1, flag_strong1 = 1;
+    int strength0 = 3, strength1 = 3;
+    int i;
+    int lims;
+
+    for(i = 0, ptr = src; i < 4; i++, ptr += stride){
+        diffs[i][0] = ptr[-2*step] - ptr[-1*step];
+        diffs[i][1] = ptr[ 1*step] - ptr[ 0*step];
+        s0 += diffs[i][0];
+        s1 += diffs[i][1];
+    }
+    if(FFABS(s0) >= (beta<<2)){
+        strength0 = 1;
+    }
+    if(FFABS(s1) >= (beta<<2)){
+        strength1 = 1;
+    }
+    if(strength0 + strength1 <= 2){
+        return;
+    }
+
+    for(i = 0, ptr = src; i < 4; i++, ptr += stride){
+        diffs[i][2] = ptr[-2*step] - ptr[-3*step];
+        diffs[i][3] = ptr[ 1*step] - ptr[ 2*step];
+        s2 += diffs[i][2];
+        s3 += diffs[i][3];
+    }
+
+    if(!edge){
+        flag_strong0 = flag_strong1 = 0;
+    }else{
+        flag_strong0 = (strength0 == 3) && (FFABS(s2) < beta2);
+        flag_strong1 = (strength1 == 3) && (FFABS(s3) < beta2);
+    }
+
+    lims = (lim_inner + lim_outer + strength0 + strength1) >> 1;
+    if(flag_strong0 && flag_strong1){ /* strong filtering */
+        for(i = 0; i < 4; i++, src += stride){
+            int diff[2], sflag, p0, p1;
+            int t = src[0*step] - src[-1*step];
+
+            if(!t) continue;
+            sflag = (alpha * FFABS(t)) >> 7;
+            if(sflag > 1) continue;
+
+            p0 = (RV40_STRONG_FILTER(src, step, -3, 1, -3) + rv40_dither_l[dmode + i]) >> 7;
+            p1 = (RV40_STRONG_FILTER(src, step, -2, 2, -2) + rv40_dither_r[dmode + i]) >> 7;
+            diff[0] = src[-1*step];
+            diff[1] = src[ 0*step];
+            src[-1*step] = sflag ? av_clip(p0, src[-1*step] - lims, src[-1*step] + lims) : p0;
+            src[ 0*step] = sflag ? av_clip(p1, src[ 0*step] - lims, src[ 0*step] + lims) : p1;
+            diff[0] -= src[-1*step];
+            diff[1] -= src[ 0*step];
+            p0 = (RV40_STRONG_FILTER(src, step, -4, 0, -4) + rv40_dither_l[dmode + i] + diff[1]*25) >> 7;
+            p1 = (RV40_STRONG_FILTER(src, step, -1, 3, -1) + rv40_dither_r[dmode + i] + diff[0]*25) >> 7;
+            src[-2*step] = sflag ? av_clip(p0, src[-2*step] - lims, src[-2*step] + lims) : p0;
+            src[ 1*step] = sflag ? av_clip(p1, src[ 1*step] - lims, src[ 1*step] + lims) : p1;
+            if(!chroma){
+                src[-3*step] = (RV40_STRONG_FILTER(src, step, -4, -3, -1) + 64) >> 7;
+                src[ 2*step] = (RV40_STRONG_FILTER(src, step,  0,  2,  0) + 64) >> 7;
+            }
+        }
+    }else if(strength0 == 3 && strength1 == 3){
+        for(i = 0; i < 4; i++, src += stride)
+            rv40_weak_loop_filter(src, step, 1, 1, alpha, lim_inner, lim_outer, lims, beta,
+                                  diffs[i][0], diffs[i][1], diffs[i][2], diffs[i][3]);
+    }else{
+        for(i = 0; i < 4; i++, src += stride)
+            rv40_weak_loop_filter(src, step, strength0==3, strength1==3,
+                                  alpha, lim_inner>>1, lim_outer>>1, lims>>1, beta,
+                                  diffs[i][0], diffs[i][1], diffs[i][2], diffs[i][3]);
+    }
+}
+
+static void rv40_v_loop_filter(uint8_t *src, int stride, int dmode,
+                               int lim_inner, int lim_outer,
+                               int alpha, int beta, int beta2, int chroma, int edge){
+    rv40_adaptive_loop_filter(src, 1, stride, dmode, lim_inner, lim_outer,
+                              alpha, beta, beta2, chroma, edge);
+}
+static void rv40_h_loop_filter(uint8_t *src, int stride, int dmode,
+                               int lim_inner, int lim_outer,
+                               int alpha, int beta, int beta2, int chroma, int edge){
+    rv40_adaptive_loop_filter(src, stride, 1, dmode, lim_inner, lim_outer,
+                              alpha, beta, beta2, chroma, edge);
+}
+
+static int check_mv_difference(int16_t (*motion_val)[2], int step)
+{
+    int d;
+    d = motion_val[0][0] - motion_val[-step][0];
+    if(d < -3 || d > 3)
+        return 1;
+    d = motion_val[0][1] - motion_val[-step][1];
+    if(d < -3 || d > 3)
+        return 1;
+    return 0;
+}
+
+static int rv40_set_deblock_coef(RV34DecContext *r)
+{
+    MpegEncContext *s = &r->s;
+    int mvmask = 0, i, j;
+    int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
+    int16_t (*motion_val)[2] = s->current_picture_ptr->motion_val[0][midx];
+    if(s->pict_type == FF_I_TYPE)
+        return 0;
+    for(i = 0; i < 2; i++){
+        if(!s->first_slice_line && check_mv_difference(motion_val + i, s->b8_stride)){
+            mvmask |= 0x03 << (i*2);
+        }
+        if(check_mv_difference(motion_val + s->b8_stride + i, s->b8_stride)){
+            mvmask |= 0x03 << (i*2 + 8);
+        }
+    }
+    for(j = 0; j < 2; j++){
+        if(s->mb_x && check_mv_difference(motion_val, 1)){
+            mvmask |= 0x11 << (j*8);
+        }
+        if(check_mv_difference(motion_val + 1, 1)){
+            mvmask |= 0x11 << (2 + j*8);
+        }
+        motion_val += s->b8_stride;
+    }
+    return mvmask;
+}
+
+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 void rv40_loop_filter(RV34DecContext *r)
+{
+    MpegEncContext *s = &r->s;
+    int mb_pos;
+    int i, j, k;
+    uint8_t *Y, *C;
+    int alpha, beta, betaY, betaC;
+    int q;
+    int mbtype[4], clip[4], mvmasks[4], cbp[4], uvcbp[4][2], strength[4];
+
+    if(s->pict_type == FF_B_TYPE)
+        return;
+
+    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 btype = s->current_picture_ptr->mb_type[mb_pos];
+            if(IS_INTRA(btype) || IS_SEPARATE_DC(btype)){
+                r->cbp_luma  [mb_pos] = 0xFFFF;
+            }
+            if(IS_INTRA(btype)){
+                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++){
+            int y_h_deblock, y_v_deblock;
+            int c_v_deblock[2], c_h_deblock[2];
+
+            ff_init_block_index(s);
+            ff_update_block_index(s);
+            Y = s->dest[0];
+            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;
+            }
+
+            mvmasks[POS_CUR] = r->deblock_coefs[mb_pos];
+            mbtype [POS_CUR] = s->current_picture_ptr->mb_type[mb_pos];
+            cbp    [POS_CUR] = r->cbp_luma[mb_pos];
+            uvcbp[POS_CUR][0] = r->cbp_chroma[mb_pos] & 0xF;
+            uvcbp[POS_CUR][1] = r->cbp_chroma[mb_pos] >> 4;
+            for(i = 1; i < 4; i++){
+                mvmasks[i] = 0;
+                mbtype [i] = mbtype[0];
+                cbp    [i] = 0;
+                uvcbp[1][0] = uvcbp[1][1] = 0;
+            }
+            if(s->mb_y){
+                mvmasks[POS_TOP] = r->deblock_coefs[mb_pos - s->mb_stride] & MASK_Y_LAST_ROW;
+                mbtype [POS_TOP] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
+                cbp    [POS_TOP] = r->cbp_luma[mb_pos - s->mb_stride] & MASK_Y_LAST_ROW;
+                uvcbp[POS_TOP][0] =  r->cbp_chroma[mb_pos - s->mb_stride]       & MASK_C_LAST_ROW;
+                uvcbp[POS_TOP][1] = (r->cbp_chroma[mb_pos - s->mb_stride] >> 4) & MASK_C_LAST_ROW;
+            }
+            if(s->mb_x){
+                mvmasks[POS_LEFT] = r->deblock_coefs[mb_pos - 1] & MASK_Y_RIGHT_COL;
+                mbtype [POS_LEFT] = s->current_picture_ptr->mb_type[mb_pos - 1];
+                cbp    [POS_LEFT] = r->cbp_luma[mb_pos - 1] & MASK_Y_RIGHT_COL;
+                uvcbp[POS_LEFT][0] =  r->cbp_chroma[mb_pos - 1]       & MASK_C_RIGHT_COL;
+                uvcbp[POS_LEFT][1] = (r->cbp_chroma[mb_pos - 1] >> 4) & MASK_C_RIGHT_COL;
+            }
+            if(s->mb_y < s->mb_height - 1){
+                mvmasks[POS_BOTTOM] = r->deblock_coefs[mb_pos + s->mb_stride] & MASK_Y_TOP_ROW;
+                mbtype [POS_BOTTOM] = s->current_picture_ptr->mb_type[mb_pos + s->mb_stride];
+                cbp    [POS_BOTTOM] = r->cbp_luma[mb_pos + s->mb_stride] & MASK_Y_TOP_ROW;
+                uvcbp[POS_BOTTOM][0] =  r->cbp_chroma[mb_pos + s->mb_stride]       & MASK_C_TOP_ROW;
+                uvcbp[POS_BOTTOM][1] = (r->cbp_chroma[mb_pos + s->mb_stride] >> 4) & MASK_C_TOP_ROW;
+            }
+            for(i = 0; i < 4; i++){
+                strength[i] = (IS_INTRA(mbtype[i]) || IS_SEPARATE_DC(mbtype[i])) ? 2 : 1;
+                clip[i] = rv40_filter_clip_tbl[strength[i]][q];
+            }
+            y_h_deblock = cbp[POS_CUR]
+                        // this mask corresponds to checking whether top edge can be filtered
+                        | ((cbp[POS_CUR] << 4) & ~MASK_Y_TOP_ROW) | (cbp[POS_TOP] >> 12)
+                        // this mask corresponds to checking whether bottom edge can be filtered
+                        | ((cbp[POS_BOTTOM] << 20) & ~MASK_Y_TOP_ROW) | (cbp[POS_BOTTOM] << 16)
+                        // this mask corresponds to checking whether top edge can be filtered
+                        // because of motion vector differences between blocks
+                        | mvmasks[POS_CUR] | (mvmasks[POS_CUR] << 16);
+            y_v_deblock = cbp[POS_CUR]
+                        // this mask corresponds to checking whether left edge can be filtered
+                        | ((cbp[POS_CUR] << 1) & ~MASK_Y_LEFT_COL) | (cbp[POS_LEFT] >> 3)
+                        // this mask corresponds to checking whether left edge can be filtered
+                        // because of motion vector differences between blocks
+                        | mvmasks[0];
+            if(!s->mb_x){
+                y_v_deblock &= ~MASK_Y_LEFT_COL;
+            }
+            if(!s->mb_y){
+                y_h_deblock &= ~MASK_Y_TOP_ROW;
+            }
+            if(s->mb_y == s->mb_height - 1 || (strength[POS_CUR] == 2 || strength[POS_BOTTOM] == 2)){
+                y_h_deblock &= ~(MASK_C_TOP_ROW << 16);
+            }
+            cbp[POS_CUR] = cbp[POS_CUR] | (cbp[POS_BOTTOM] << 16)
+                         | mvmasks[POS_CUR] | (mvmasks[POS_BOTTOM] << 16);
+            for(i = 0; i < 2; i++){
+                c_v_deblock[i] = ((uvcbp[POS_CUR][i] << 1) & ~MASK_C_RIGHT_COL) | (uvcbp[POS_LEFT][i] >> 1)
+                               | (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i];
+                c_h_deblock[i] = (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i] | (uvcbp[POS_TOP][i] >> 2)
+                               | (uvcbp[POS_BOTTOM][i] << 6) | (uvcbp[POS_CUR][i] << 2);
+                uvcbp[POS_CUR][i] = (uvcbp[POS_BOTTOM][i] << 4) | uvcbp[POS_CUR][i];
+                if(!s->mb_x){
+                    c_v_deblock[i] &= ~MASK_C_LEFT_COL;
+                }
+                if(!s->mb_y){
+                    c_h_deblock[i] &= ~MASK_C_TOP_ROW;
+                }
+                if(s->mb_y == s->mb_height - 1 || strength[POS_CUR] == 2 || strength[POS_BOTTOM] == 2){
+                    c_h_deblock[i] &= ~(MASK_C_TOP_ROW << 4);
+                }
+            }
+
+            for(j = 0; j < 16; j += 4){
+                Y = s->dest[0] + j * s->linesize;
+                for(i = 0; i < 4; i++, Y += 4){
+                    int ij = i + j;
+                    int clip_cur = cbp[POS_CUR] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
+                    int dither = j ? ij : i*4;
+                    if(y_h_deblock & (MASK_BOTTOM << ij)){
+                        rv40_h_loop_filter(Y+4*s->linesize, s->linesize, dither,
+                                           cbp[POS_CUR] & (MASK_BOTTOM << ij) ? clip[POS_CUR] : 0,
+                                           clip_cur,
+                                           alpha, beta, betaY, 0, 0);
+                    }
+                    if(y_v_deblock & (MASK_CUR << ij) && (i || !(strength[POS_CUR] == 2 || strength[POS_LEFT] == 2))){
+                        int clip_left;
+                        if(!i){
+                            clip_left = cbp[POS_LEFT] & (MASK_RIGHT << j) ? clip[POS_LEFT] : 0;
+                        }else{
+                            clip_left = cbp[POS_CUR] & (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);
+                    }
+                    if(!j && y_h_deblock & (MASK_CUR << i) && (strength[POS_CUR] == 2 || strength[POS_TOP] == 2)){
+                        rv40_h_loop_filter(Y, s->linesize, dither,
+                                           clip_cur,
+                                           cbp[POS_TOP] & (MASK_TOP << i) ? clip[POS_TOP] : 0,
+                                           alpha, beta, betaY, 0, 1);
+                    }
+                    if(y_v_deblock & (MASK_CUR << ij) && !i && (strength[POS_CUR] == 2 || strength[POS_LEFT] == 2)){
+                        int clip_left = cbp[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->dest[k] + j * 4 * s->uvlinesize;
+                    for(i = 0; i < 2; i++, C += 4){
+                        int ij = i + j*2;
+                        int cur_clip = uvcbp[POS_CUR][k] & (MASK_CUR << ij) ? clip[POS_CUR] : 0;
+                        if(c_h_deblock[k] & (MASK_CUR << (ij+2))){
+                            int bot_clip = uvcbp[POS_CUR][k] & (MASK_CUR << (ij+2)) ? clip[POS_CUR] : 0;
+                            rv40_h_loop_filter(C+4*s->uvlinesize, s->uvlinesize, i*8,
+                                               bot_clip,
+                                               cur_clip,
+                                               alpha, beta, betaC, 1, 0);
+                        }
+                        if((c_v_deblock[k] & (MASK_CUR << ij)) && (i || !(strength[POS_CUR] == 2 || strength[POS_LEFT] == 2))){
+                            int left_clip;
+                            if(!i){
+                                left_clip = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
+                            }else{
+                                left_clip = uvcbp[POS_CUR][k]  & (MASK_CUR << (ij-1))  ? clip[POS_CUR]  : 0;
+                            }
+                            rv40_v_loop_filter(C, s->uvlinesize, j*8,
+                                               cur_clip,
+                                               left_clip,
+                                               alpha, beta, betaY, 1, 0);
+                        }
+                        if(!j && c_h_deblock[k] & (MASK_CUR << ij) && (strength[POS_CUR] == 2 || strength[POS_TOP] == 2)){
+                            int top_clip = uvcbp[POS_TOP][k] & (MASK_CUR << (ij+2)) ? clip[POS_TOP] : 0;
+                            rv40_h_loop_filter(C, s->uvlinesize, i*8,
+                                               cur_clip,
+                                               top_clip,
+                                               alpha, beta, betaC, 1, 1);
+                        }
+                        if(c_v_deblock[k] & (MASK_CUR << ij) && !i && (strength[POS_CUR] == 2 || strength[POS_LEFT] == 2)){
+                            int left_clip = uvcbp[POS_LEFT][k] & (MASK_CUR << (2*j+1)) ? clip[POS_LEFT] : 0;
+                            rv40_v_loop_filter(C, s->uvlinesize, j*8,
+                                               cur_clip,
+                                               left_clip,
+                                               alpha, beta, betaY, 1, 1);
+                        }
+                    }
+                }
+            }
+        }
+    }
+}
+
+/**
  * Initialize decoder.
  */
 static av_cold int rv40_decode_init(AVCodecContext *avctx)
@@ -261,6 +681,8 @@
     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->set_deblock_coef   = rv40_set_deblock_coef;
     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