avcodec/vc1: re-implement and expand VC-1 loop filtering

The existing implementation did loop filtering for progressive
frames only. This rewritten version implements loop filtering for
all applicable frame types for both progessive and
frame/field-interlace.

Signed-off-by: Jerome Borsboom <jerome.borsboom@carpalis.nl>

1 files changed, 1042 insertions, 0 deletions

diff --git a/libavcodec/vc1_loopfilter.c b/libavcodec/vc1_loopfilter.c
index 3122b1a258..7ef0fd1ea2 100644
--- a/libavcodec/vc1_loopfilter.c
+++ b/libavcodec/vc1_loopfilter.c
@@ -451,3 +451,1045 @@ void ff_vc1_apply_p_loop_filter(VC1Context *v)
         }
     }
 }
+
+#define LEFT_EDGE   (1 << 0)
+#define RIGHT_EDGE  (1 << 1)
+#define TOP_EDGE    (1 << 2)
+#define BOTTOM_EDGE (1 << 3)
+
+static av_always_inline void vc1_i_h_loop_filter(VC1Context *v, uint8_t *dest,
+                                                 uint32_t flags, int block_num)
+{
+    MpegEncContext *s  = &v->s;
+    int pq = v->pq;
+    uint8_t *dst;
+
+    if (block_num & 2)
+        return;
+
+    if (!(flags & LEFT_EDGE) || (block_num & 5) == 1) {
+        if (block_num > 3)
+            dst = dest;
+        else
+            dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
+
+        if (v->fcm == ILACE_FRAME)
+            if (block_num > 3) {
+                v->vc1dsp.vc1_h_loop_filter4(dst, 2 * s->uvlinesize, pq);
+                v->vc1dsp.vc1_h_loop_filter4(dst + s->uvlinesize, 2 * s->uvlinesize, pq);
+            } else {
+                v->vc1dsp.vc1_h_loop_filter8(dst, 2 * s->linesize, pq);
+                v->vc1dsp.vc1_h_loop_filter8(dst + s->linesize, 2 * s->linesize, pq);
+            }
+        else
+            if (block_num > 3)
+                v->vc1dsp.vc1_h_loop_filter8(dst, s->uvlinesize, pq);
+            else
+                v->vc1dsp.vc1_h_loop_filter16(dst, s->linesize, pq);
+    }
+}
+
+static av_always_inline void vc1_i_v_loop_filter(VC1Context *v, uint8_t *dest,
+                                                 uint32_t flags, uint8_t fieldtx,
+                                                 int block_num)
+{
+    MpegEncContext *s  = &v->s;
+    int pq = v->pq;
+    uint8_t *dst;
+
+    if ((block_num & 5) == 1)
+        return;
+
+    if (!(flags & TOP_EDGE) || block_num & 2) {
+        if (block_num > 3)
+            dst = dest;
+        else
+            dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
+
+        if (v->fcm == ILACE_FRAME) {
+            if (block_num > 3) {
+                v->vc1dsp.vc1_v_loop_filter8(dst, 2 * s->uvlinesize, pq);
+                v->vc1dsp.vc1_v_loop_filter8(dst + s->uvlinesize, 2 * s->uvlinesize, pq);
+            } else if (block_num < 2 || !fieldtx) {
+                v->vc1dsp.vc1_v_loop_filter16(dst, 2 * s->linesize, pq);
+                v->vc1dsp.vc1_v_loop_filter16(dst + s->linesize, 2 * s->linesize, pq);
+            }
+        } else
+            if (block_num > 3)
+                v->vc1dsp.vc1_v_loop_filter8(dst, s->uvlinesize, pq);
+            else
+                v->vc1dsp.vc1_v_loop_filter16(dst, s->linesize, pq);
+    }
+}
+
+void ff_vc1_i_loop_filter(VC1Context *v)
+{
+    MpegEncContext *s = &v->s;
+    int block_count = CONFIG_GRAY && (s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 4 : 6;
+    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
+    uint8_t *dest, fieldtx;
+    uint32_t flags = 0;
+    int i;
+
+    /* Within a MB, the vertical loop filter always runs before the horizontal.
+     * To accomplish that, we run the V loop filter on top and internal
+     * horizontal borders of the last overlap filtered MB. Then, we wait for
+     * the loop filter iteration on the next row to do V loop filter on the
+     * bottom edge of this MB, before moving over and running the H loop
+     * filter on the left and internal vertical borders. Therefore, the loop
+     * filter trails by one row and one column relative to the overlap filter
+     * and two rows and two colums relative to the decoding loop. */
+    if (!s->first_slice_line) {
+        dest = s->dest[0] - 16 * s->linesize - 16;
+        flags = s->mb_y == s->start_mb_y + 1 ? TOP_EDGE : 0;
+        if (s->mb_x) {
+            fieldtx = v->fieldtx_plane[mb_pos - s->mb_stride - 1];
+            for (i = 0; i < block_count; i++)
+                vc1_i_v_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 8 : dest, flags, fieldtx, i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            dest += 16;
+            fieldtx = v->fieldtx_plane[mb_pos - s->mb_stride];
+            for (i = 0; i < block_count; i++)
+                vc1_i_v_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize : dest, flags, fieldtx, i);
+        }
+    }
+    if (s->mb_y == s->end_mb_y - 1) {
+        dest = s->dest[0] - 16;
+        flags = s->first_slice_line ? TOP_EDGE | BOTTOM_EDGE : BOTTOM_EDGE;
+        if (s->mb_x) {
+            fieldtx = v->fieldtx_plane[mb_pos - 1];
+            for (i = 0; i < block_count; i++)
+                vc1_i_v_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 : dest, flags, fieldtx, i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            dest += 16;
+            fieldtx = v->fieldtx_plane[mb_pos];
+            for (i = 0; i < block_count; i++)
+                vc1_i_v_loop_filter(v, i > 3 ? s->dest[i - 3] : dest, flags, fieldtx, i);
+        }
+    }
+
+    if (s->mb_y >= s->start_mb_y + 2) {
+        dest = s->dest[0] - 32 * s->linesize - 16;
+        if (s->mb_x) {
+            flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+            for (i = 0; i < block_count; i++)
+                vc1_i_h_loop_filter(v, i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize - 8 : dest, flags, i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            dest += 16;
+            flags = s->mb_x == 0 ? LEFT_EDGE | RIGHT_EDGE : RIGHT_EDGE;
+            for (i = 0; i < block_count; i++)
+                vc1_i_h_loop_filter(v, i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize : dest, flags, i);
+        }
+    }
+    if (s->mb_y == s->end_mb_y - 1) {
+        if (s->mb_y >= s->start_mb_y + 1) {
+            dest = s->dest[0] - 16 * s->linesize - 16;
+            if (s->mb_x) {
+                flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+                for (i = 0; i < block_count; i++)
+                    vc1_i_h_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 8 : dest, flags, i);
+            }
+            if (s->mb_x == s->mb_width - 1) {
+                flags = s->mb_x == 0 ? LEFT_EDGE | RIGHT_EDGE : RIGHT_EDGE;
+                dest += 16;
+                for (i = 0; i < block_count; i++)
+                    vc1_i_h_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize : dest, flags, i);
+            }
+        }
+        dest = s->dest[0] - 16;
+        if (s->mb_x) {
+            flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+            for (i = 0; i < block_count; i++)
+                vc1_i_h_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 : dest, flags, i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            dest += 16;
+            flags = s->mb_x == 0 ? LEFT_EDGE | RIGHT_EDGE : RIGHT_EDGE;
+            for (i = 0; i < block_count; i++)
+                vc1_i_h_loop_filter(v, i > 3 ? s->dest[i - 3] : dest, flags, i);
+        }
+    }
+}
+
+static av_always_inline void vc1_p_h_loop_filter(VC1Context *v, uint8_t *dest, uint32_t *cbp,
+                                                 uint8_t *is_intra, int16_t (*mv)[2], uint8_t *mv_f,
+                                                 int *ttblk, uint32_t flags, int block_num)
+{
+    MpegEncContext *s  = &v->s;
+    int pq = v->pq;
+    uint32_t left_cbp = cbp[0] >> (block_num * 4), right_cbp;
+    uint8_t left_is_intra, right_is_intra;
+    int tt;
+    int idx, linesize  = block_num > 3 ? s->uvlinesize : s->linesize;
+    uint8_t *dst;
+
+    if (block_num > 3)
+        dst = dest;
+    else
+        dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
+
+    if (!(flags & RIGHT_EDGE) || !(block_num & 5)) {
+        left_is_intra = is_intra[0] & (1 << block_num);
+
+        if (block_num > 3) {
+            right_is_intra = is_intra[1] & (1 << block_num);
+            right_cbp = cbp[1] >> (block_num * 4);
+        } else if (block_num & 1) {
+            right_is_intra = is_intra[1] & (1 << block_num - 1);
+            right_cbp = cbp[1] >> ((block_num - 1) * 4);
+        } else {
+            right_is_intra = is_intra[0] & (1 << block_num + 1);
+            right_cbp = cbp[0] >> ((block_num + 1) * 4);
+        }
+
+        if (left_is_intra || right_is_intra ||
+            mv[0][0] != mv[1][0] || mv[0][1] != mv[1][1] ||
+            (v->fcm == ILACE_FIELD && mv_f[0] != mv_f[1]))
+            v->vc1dsp.vc1_h_loop_filter8(dst + 8, linesize, pq);
+        else {
+            idx = (left_cbp | (right_cbp >> 1)) & 5;
+            if (idx & 1)
+                v->vc1dsp.vc1_h_loop_filter4(dst + 4 * linesize + 8, linesize, pq);
+            if (idx & 4)
+                v->vc1dsp.vc1_h_loop_filter4(dst + 8, linesize, pq);
+        }
+    }
+
+    tt = ttblk[0] >> (block_num * 4) & 0xf;
+    if (tt == TT_4X4 || tt == TT_4X8) {
+        if (left_cbp & 3)
+            v->vc1dsp.vc1_h_loop_filter4(dst + 4 * linesize + 4, linesize, pq);
+        if (left_cbp & 12)
+            v->vc1dsp.vc1_h_loop_filter4(dst + 4, linesize, pq);
+    }
+}
+
+static av_always_inline void vc1_p_v_loop_filter(VC1Context *v, uint8_t *dest, uint32_t *cbp,
+                                                 uint8_t *is_intra, int16_t (*mv)[2], uint8_t *mv_f,
+                                                 int *ttblk, uint32_t flags, int block_num)
+{
+    MpegEncContext *s  = &v->s;
+    int pq = v->pq;
+    uint32_t top_cbp = cbp[0] >> (block_num * 4), bottom_cbp;
+    uint8_t top_is_intra, bottom_is_intra;
+    int tt;
+    int idx, linesize  = block_num > 3 ? s->uvlinesize : s->linesize;
+    uint8_t *dst;
+
+    if (block_num > 3)
+        dst = dest;
+    else
+        dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
+
+    if(!(flags & BOTTOM_EDGE) || block_num < 2) {
+        top_is_intra = is_intra[0] & (1 << block_num);
+
+        if (block_num > 3) {
+            bottom_is_intra = is_intra[s->mb_stride] & (1 << block_num);
+            bottom_cbp = cbp[s->mb_stride] >> (block_num * 4);
+        } else if (block_num < 2) {
+            bottom_is_intra = is_intra[0] & (1 << block_num + 2);
+            bottom_cbp = cbp[0] >> ((block_num + 2) * 4);
+        } else {
+            bottom_is_intra = is_intra[s->mb_stride] & (1 << block_num - 2);
+            bottom_cbp = cbp[s->mb_stride] >> ((block_num - 2) * 4);
+        }
+
+        if (top_is_intra || bottom_is_intra ||
+            mv[0][0] != mv[block_num > 3 ? s->mb_stride : s->b8_stride][0] ||
+            mv[0][1] != mv[block_num > 3 ? s->mb_stride : s->b8_stride][1] ||
+            (v->fcm == ILACE_FIELD && mv_f[0] != mv_f[block_num > 3 ? s->mb_stride : s->b8_stride]))
+            v->vc1dsp.vc1_v_loop_filter8(dst + 8 * linesize, linesize, pq);
+        else {
+            idx = (top_cbp | (bottom_cbp >> 2)) & 3;
+            if (idx & 1)
+                v->vc1dsp.vc1_v_loop_filter4(dst + 8 * linesize + 4, linesize, pq);
+            if (idx & 2)
+                v->vc1dsp.vc1_v_loop_filter4(dst + 8 * linesize, linesize, pq);
+        }
+    }
+
+    tt = ttblk[0] >> (block_num * 4) & 0xf;
+    if (tt == TT_4X4 || tt == TT_8X4) {
+        if (top_cbp & 5)
+            v->vc1dsp.vc1_v_loop_filter4(dst + 4 * linesize + 4, linesize, pq);
+        if (top_cbp & 10)
+            v->vc1dsp.vc1_v_loop_filter4(dst + 4 * linesize, linesize, pq);
+    }
+}
+
+void ff_vc1_p_loop_filter(VC1Context *v)
+{
+    MpegEncContext *s = &v->s;
+    int block_count = CONFIG_GRAY && (s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 4 : 6;
+    uint8_t *dest;
+    uint32_t *cbp;
+    uint8_t *is_intra;
+    int16_t (*uvmv)[2];
+    int *ttblk;
+    uint32_t flags;
+    int i;
+
+    /* Within a MB, the vertical loop filter always runs before the horizontal.
+     * To accomplish that, we run the V loop filter on all applicable
+     * horizontal borders of the MB above the last overlap filtered MB. Then,
+     * we wait for the next loop filter iteration to do H loop filter on all
+     * applicable vertical borders of this MB. Therefore, the loop filter
+     * trails by one row and one column relative to the overlap filter and two
+     * rows and two colums relative to the decoding loop. */
+    if (s->mb_y >= s->start_mb_y + 2) {
+        if (s->mb_x) {
+            dest = s->dest[0] - 32 * s->linesize - 16;
+            cbp = &v->cbp[s->mb_x - 2 * s->mb_stride - 1];
+            is_intra = &v->is_intra[s->mb_x - 2 * s->mb_stride - 1];
+            uvmv = &v->luma_mv[s->mb_x - 2 * s->mb_stride - 1];
+            ttblk = &v->ttblk[s->mb_x - 2 * s->mb_stride - 1];
+            flags = s->mb_y == s->start_mb_y + 2 ? TOP_EDGE : 0;
+            for (i = 0; i < block_count; i++)
+                vc1_p_v_loop_filter(v,
+                                    i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize - 8 : dest,
+                                    cbp,
+                                    is_intra,
+                                    i > 3 ? uvmv :
+                                            &s->current_picture.motion_val[0][s->block_index[i] - 4 * s->b8_stride - 2 + v->blocks_off],
+                                    i > 3 ? &v->mv_f[0][s->block_index[i] - 2 * s->mb_stride - 1 + v->mb_off] :
+                                            &v->mv_f[0][s->block_index[i] - 4 * s->b8_stride - 2 + v->blocks_off],
+                                    ttblk,
+                                    flags,
+                                    i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            dest = s->dest[0] - 32 * s->linesize;
+            cbp = &v->cbp[s->mb_x - 2 * s->mb_stride];
+            is_intra = &v->is_intra[s->mb_x - 2 * s->mb_stride];
+            uvmv = &v->luma_mv[s->mb_x - 2 * s->mb_stride];
+            ttblk = &v->ttblk[s->mb_x - 2 * s->mb_stride];
+            flags = s->mb_y == s->start_mb_y + 2 ? TOP_EDGE : 0;
+            for (i = 0; i < block_count; i++)
+                vc1_p_v_loop_filter(v,
+                                    i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize : dest,
+                                    cbp,
+                                    is_intra,
+                                    i > 3 ? uvmv :
+                                            &s->current_picture.motion_val[0][s->block_index[i] - 4 * s->b8_stride + v->blocks_off],
+                                    i > 3 ? &v->mv_f[0][s->block_index[i] - 2 * s->mb_stride + v->mb_off] :
+                                            &v->mv_f[0][s->block_index[i] - 4 * s->b8_stride + v->blocks_off],
+                                    ttblk,
+                                    flags,
+                                    i);
+        }
+    }
+    if (s->mb_y == s->end_mb_y - 1) {
+        if (s->mb_x) {
+            if (s->mb_y >= s->start_mb_y + 1) {
+                dest = s->dest[0] - 16 * s->linesize - 16;
+                cbp = &v->cbp[s->mb_x - s->mb_stride - 1];
+                is_intra = &v->is_intra[s->mb_x - s->mb_stride - 1];
+                uvmv = &v->luma_mv[s->mb_x - s->mb_stride - 1];
+                ttblk = &v->ttblk[s->mb_x - s->mb_stride - 1];
+                flags = s->mb_y == s->start_mb_y + 1 ? TOP_EDGE : 0;
+                for (i = 0; i < block_count; i++)
+                    vc1_p_v_loop_filter(v,
+                                        i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 8 : dest,
+                                        cbp,
+                                        is_intra,
+                                        i > 3 ? uvmv :
+                                                &s->current_picture.motion_val[0][s->block_index[i] - 2 * s->b8_stride - 2 + v->blocks_off],
+                                        i > 3 ? &v->mv_f[0][s->block_index[i] - s->mb_stride - 1 + v->mb_off] :
+                                                &v->mv_f[0][s->block_index[i] - 2 * s->b8_stride - 2 + v->blocks_off],
+                                        ttblk,
+                                        flags,
+                                        i);
+            }
+            dest = s->dest[0] - 16;
+            cbp = &v->cbp[s->mb_x - 1];
+            is_intra = &v->is_intra[s->mb_x - 1];
+            uvmv = &v->luma_mv[s->mb_x - 1];
+            ttblk = &v->ttblk[s->mb_x - 1];
+            flags = s->mb_y == s->start_mb_y ? TOP_EDGE | BOTTOM_EDGE : BOTTOM_EDGE;
+            for (i = 0; i < block_count; i++)
+                vc1_p_v_loop_filter(v,
+                                    i > 3 ? s->dest[i - 3] - 8 : dest,
+                                    cbp,
+                                    is_intra,
+                                    i > 3 ? uvmv :
+                                            &s->current_picture.motion_val[0][s->block_index[i] - 2 + v->blocks_off],
+                                    i > 3 ? &v->mv_f[0][s->block_index[i] - 1 + v->mb_off] :
+                                            &v->mv_f[0][s->block_index[i] - 2 + v->blocks_off],
+                                    ttblk,
+                                    flags,
+                                    i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            if (s->mb_y >= s->start_mb_y + 1) {
+                dest = s->dest[0] - 16 * s->linesize;
+                cbp = &v->cbp[s->mb_x - s->mb_stride];
+                is_intra = &v->is_intra[s->mb_x - s->mb_stride];
+                uvmv = &v->luma_mv[s->mb_x - s->mb_stride];
+                ttblk = &v->ttblk[s->mb_x - s->mb_stride];
+                flags = s->mb_y == s->start_mb_y + 1 ? TOP_EDGE : 0;
+                for (i = 0; i < block_count; i++)
+                    vc1_p_v_loop_filter(v,
+                                        i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize : dest,
+                                        cbp,
+                                        is_intra,
+                                        i > 3 ? uvmv :
+                                                &s->current_picture.motion_val[0][s->block_index[i] - 2 * s->b8_stride + v->blocks_off],
+                                        i > 3 ? &v->mv_f[0][s->block_index[i] - s->mb_stride + v->mb_off] :
+                                                &v->mv_f[0][s->block_index[i] - 2 * s->b8_stride + v->blocks_off],
+                                        ttblk,
+                                        flags,
+                                        i);
+            }
+            dest = s->dest[0];
+            cbp = &v->cbp[s->mb_x];
+            is_intra = &v->is_intra[s->mb_x];
+            uvmv = &v->luma_mv[s->mb_x];
+            ttblk = &v->ttblk[s->mb_x];
+            flags = s->mb_y == s->start_mb_y ? TOP_EDGE | BOTTOM_EDGE : BOTTOM_EDGE;
+            for (i = 0; i < block_count; i++)
+                vc1_p_v_loop_filter(v,
+                                    i > 3 ? s->dest[i - 3] : dest,
+                                    cbp,
+                                    is_intra,
+                                    i > 3 ? uvmv :
+                                            &s->current_picture.motion_val[0][s->block_index[i] + v->blocks_off],
+                                    i > 3 ? &v->mv_f[0][s->block_index[i] + v->mb_off] :
+                                            &v->mv_f[0][s->block_index[i] + v->blocks_off],
+                                    ttblk,
+                                    flags,
+                                    i);
+        }
+    }
+
+    if (s->mb_y >= s->start_mb_y + 2) {
+        if (s->mb_x >= 2) {
+            dest = s->dest[0] - 32 * s->linesize - 32;
+            cbp = &v->cbp[s->mb_x - 2 * s->mb_stride - 2];
+            is_intra = &v->is_intra[s->mb_x - 2 * s->mb_stride - 2];
+            uvmv = &v->luma_mv[s->mb_x - 2 * s->mb_stride - 2];
+            ttblk = &v->ttblk[s->mb_x - 2 * s->mb_stride - 2];
+            flags = s->mb_x == 2 ? LEFT_EDGE : 0;
+            for (i = 0; i < block_count; i++)
+                vc1_p_h_loop_filter(v,
+                                    i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize - 16 : dest,
+                                    cbp,
+                                    is_intra,
+                                    i > 3 ? uvmv :
+                                            &s->current_picture.motion_val[0][s->block_index[i] - 4 * s->b8_stride - 4 + v->blocks_off],
+                                    i > 3 ? &v->mv_f[0][s->block_index[i] - 2 * s->mb_stride - 2 + v->mb_off] :
+                                            &v->mv_f[0][s->block_index[i] - 4 * s->b8_stride - 4 + v->blocks_off],
+                                    ttblk,
+                                    flags,
+                                    i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            if (s->mb_x >= 1) {
+                dest = s->dest[0] - 32 * s->linesize - 16;
+                cbp = &v->cbp[s->mb_x - 2 * s->mb_stride - 1];
+                is_intra = &v->is_intra[s->mb_x - 2 * s->mb_stride - 1];
+                uvmv = &v->luma_mv[s->mb_x - 2 * s->mb_stride - 1];
+                ttblk = &v->ttblk[s->mb_x - 2 * s->mb_stride - 1];
+                flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+                for (i = 0; i < block_count; i++)
+                        vc1_p_h_loop_filter(v,
+                                            i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize - 8 : dest,
+                                            cbp,
+                                            is_intra,
+                                            i > 3 ? uvmv :
+                                                    &s->current_picture.motion_val[0][s->block_index[i] - 4 * s->b8_stride - 2 + v->blocks_off],
+                                            i > 3 ? &v->mv_f[0][s->block_index[i] - 2 * s->mb_stride - 1 + v->mb_off] :
+                                                    &v->mv_f[0][s->block_index[i] - 4 * s->b8_stride - 2 + v->blocks_off],
+                                            ttblk,
+                                            flags,
+                                            i);
+            }
+            dest = s->dest[0] - 32 * s->linesize;
+            cbp = &v->cbp[s->mb_x - 2 * s->mb_stride];
+            is_intra = &v->is_intra[s->mb_x - 2 * s->mb_stride];
+            uvmv = &v->luma_mv[s->mb_x - 2 * s->mb_stride];
+            ttblk = &v->ttblk[s->mb_x - 2 * s->mb_stride];
+            flags = s->mb_x ? RIGHT_EDGE : LEFT_EDGE | RIGHT_EDGE;
+            for (i = 0; i < block_count; i++)
+                vc1_p_h_loop_filter(v,
+                                    i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize : dest,
+                                    cbp,
+                                    is_intra,
+                                    i > 3 ? uvmv :
+                                            &s->current_picture.motion_val[0][s->block_index[i] - 4 * s->b8_stride + v->blocks_off],
+                                    i > 3 ? &v->mv_f[0][s->block_index[i] - 2 * s->mb_stride + v->mb_off] :
+                                            &v->mv_f[0][s->block_index[i] - 4 * s->b8_stride + v->blocks_off],
+                                    ttblk,
+                                    flags,
+                                    i);
+        }
+    }
+    if (s->mb_y == s->end_mb_y - 1) {
+        if (s->mb_y >= s->start_mb_y + 1) {
+            if (s->mb_x >= 2) {
+                dest = s->dest[0] - 16 * s->linesize - 32;
+                cbp = &v->cbp[s->mb_x - s->mb_stride - 2];
+                is_intra = &v->is_intra[s->mb_x - s->mb_stride - 2];
+                uvmv = &v->luma_mv[s->mb_x - s->mb_stride - 2];
+                ttblk = &v->ttblk[s->mb_x - s->mb_stride - 2];
+                flags = s->mb_x == 2 ? LEFT_EDGE : 0;
+                for (i = 0; i < block_count; i++)
+                    vc1_p_h_loop_filter(v,
+                                        i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 16 : dest,
+                                        cbp,
+                                        is_intra,
+                                        i > 3 ? uvmv :
+                                                &s->current_picture.motion_val[0][s->block_index[i] - 2 * s->b8_stride - 4 + v->blocks_off],
+                                        i > 3 ? &v->mv_f[0][s->block_index[i] - s->mb_stride - 2 + v->mb_off] :
+                                                &v->mv_f[0][s->block_index[i] - 2 * s->b8_stride - 4 + v->blocks_off],
+                                        ttblk,
+                                        flags,
+                                        i);
+            }
+            if (s->mb_x == s->mb_width - 1) {
+                if (s->mb_x >= 1) {
+                    dest = s->dest[0] - 16 * s->linesize - 16;
+                    cbp = &v->cbp[s->mb_x - s->mb_stride - 1];
+                    is_intra = &v->is_intra[s->mb_x - s->mb_stride - 1];
+                    uvmv = &v->luma_mv[s->mb_x - s->mb_stride - 1];
+                    ttblk = &v->ttblk[s->mb_x - s->mb_stride - 1];
+                    flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+                    for (i = 0; i < block_count; i++)
+                            vc1_p_h_loop_filter(v,
+                                                i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 8 : dest,
+                                                cbp,
+                                                is_intra,
+                                                i > 3 ? uvmv :
+                                                        &s->current_picture.motion_val[0][s->block_index[i] - 2 * s->b8_stride - 2 + v->blocks_off],
+                                                i > 3 ? &v->mv_f[0][s->block_index[i] - s->mb_stride - 1 + v->mb_off] :
+                                                        &v->mv_f[0][s->block_index[i] - 2 * s->b8_stride - 2 + v->blocks_off],
+                                                ttblk,
+                                                flags,
+                                                i);
+                }
+                dest = s->dest[0] - 16 * s->linesize;
+                cbp = &v->cbp[s->mb_x - s->mb_stride];
+                is_intra = &v->is_intra[s->mb_x - s->mb_stride];
+                uvmv = &v->luma_mv[s->mb_x - s->mb_stride];
+                ttblk = &v->ttblk[s->mb_x - s->mb_stride];
+                flags = s->mb_x ? RIGHT_EDGE : LEFT_EDGE | RIGHT_EDGE;
+                for (i = 0; i < block_count; i++)
+                    vc1_p_h_loop_filter(v,
+                                        i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize : dest,
+                                        cbp,
+                                        is_intra,
+                                        i > 3 ? uvmv :
+                                                &s->current_picture.motion_val[0][s->block_index[i] - 2 * s->b8_stride + v->blocks_off],
+                                        i > 3 ? &v->mv_f[0][s->block_index[i] - s->mb_stride + v->mb_off] :
+                                                &v->mv_f[0][s->block_index[i] - 2 * s->b8_stride + v->blocks_off],
+                                        ttblk,
+                                        flags,
+                                        i);
+            }
+        }
+        if (s->mb_x >= 2) {
+            dest = s->dest[0] - 32;
+            cbp = &v->cbp[s->mb_x - 2];
+            is_intra = &v->is_intra[s->mb_x - 2];
+            uvmv = &v->luma_mv[s->mb_x - 2];
+            ttblk = &v->ttblk[s->mb_x - 2];
+            flags = s->mb_x == 2 ? LEFT_EDGE : 0;
+            for (i = 0; i < block_count; i++)
+                vc1_p_h_loop_filter(v,
+                                    i > 3 ? s->dest[i - 3] - 16 : dest,
+                                    cbp,
+                                    is_intra,
+                                    i > 3 ? uvmv :
+                                            &s->current_picture.motion_val[0][s->block_index[i] - 4 + v->blocks_off],
+                                    i > 3 ? &v->mv_f[0][s->block_index[i] - 2 + v->mb_off] :
+                                            &v->mv_f[0][s->block_index[i] - 4 + v->blocks_off],
+                                    ttblk,
+                                    flags,
+                                    i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            if (s->mb_x >= 1) {
+                dest = s->dest[0] - 16;
+                cbp = &v->cbp[s->mb_x - 1];
+                is_intra = &v->is_intra[s->mb_x - 1];
+                uvmv = &v->luma_mv[s->mb_x - 1];
+                ttblk = &v->ttblk[s->mb_x - 1];
+                flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+                for (i = 0; i < block_count; i++)
+                    vc1_p_h_loop_filter(v,
+                                        i > 3 ? s->dest[i - 3] - 8 : dest,
+                                        cbp,
+                                        is_intra,
+                                        i > 3 ? uvmv :
+                                                &s->current_picture.motion_val[0][s->block_index[i] - 2 + v->blocks_off],
+                                        i > 3 ? &v->mv_f[0][s->block_index[i] - 1 + v->mb_off] :
+                                                &v->mv_f[0][s->block_index[i] - 2 + v->blocks_off],
+                                        ttblk,
+                                        flags,
+                                        i);
+            }
+            dest = s->dest[0];
+            cbp = &v->cbp[s->mb_x];
+            is_intra = &v->is_intra[s->mb_x];
+            uvmv = &v->luma_mv[s->mb_x];
+            ttblk = &v->ttblk[s->mb_x];
+            flags = s->mb_x ? RIGHT_EDGE : LEFT_EDGE | RIGHT_EDGE;
+            for (i = 0; i < block_count; i++)
+                vc1_p_h_loop_filter(v,
+                                    i > 3 ? s->dest[i - 3] : dest,
+                                    cbp,
+                                    is_intra,
+                                    i > 3 ? uvmv :
+                                            &s->current_picture.motion_val[0][s->block_index[i] + v->blocks_off],
+                                    i > 3 ? &v->mv_f[0][s->block_index[i] + v->mb_off] :
+                                            &v->mv_f[0][s->block_index[i] + v->blocks_off],
+                                    ttblk,
+                                    flags,
+                                    i);
+        }
+    }
+}
+
+static av_always_inline void vc1_p_h_intfr_loop_filter(VC1Context *v, uint8_t *dest, int *ttblk,
+                                                       uint32_t flags, uint8_t fieldtx, int block_num)
+{
+    MpegEncContext *s  = &v->s;
+    int pq = v->pq;
+    int tt;
+    int linesize  = block_num > 3 ? s->uvlinesize : s->linesize;
+    uint8_t *dst;
+
+    if (block_num > 3)
+        dst = dest;
+    else
+        dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
+
+    tt = ttblk[0] >> (block_num * 4) & 0xf;
+    if (block_num < 4) {
+        if (fieldtx) {
+            if (block_num < 2) {
+                if (tt == TT_4X4 || tt == TT_4X8)
+                    v->vc1dsp.vc1_h_loop_filter8(dst + 4, 2 * linesize, pq);
+                if (!(flags & RIGHT_EDGE) || block_num == 0)
+                    v->vc1dsp.vc1_h_loop_filter8(dst + 8, 2 * linesize, pq);
+            } else {
+                if (tt == TT_4X4 || tt == TT_4X8)
+                    v->vc1dsp.vc1_h_loop_filter8(dst - 7 * linesize + 4, 2 * linesize, pq);
+                if (!(flags & RIGHT_EDGE) || block_num == 2)
+                    v->vc1dsp.vc1_h_loop_filter8(dst - 7 * linesize + 8, 2 * linesize, pq);
+            }
+        } else {
+            if(tt == TT_4X4 || tt == TT_4X8) {
+                v->vc1dsp.vc1_h_loop_filter4(dst + 4, 2 * linesize, pq);
+                v->vc1dsp.vc1_h_loop_filter4(dst + linesize + 4, 2 * linesize, pq);
+            }
+            if (!(flags & RIGHT_EDGE) || !(block_num & 5)) {
+                v->vc1dsp.vc1_h_loop_filter4(dst + 8, 2 * linesize, pq);
+                v->vc1dsp.vc1_h_loop_filter4(dst + linesize + 8, 2 * linesize, pq);
+            }
+        }
+    } else {
+        if (tt == TT_4X4 || tt == TT_4X8) {
+            v->vc1dsp.vc1_h_loop_filter4(dst + 4, 2 * linesize, pq);
+            v->vc1dsp.vc1_h_loop_filter4(dst + linesize + 4, 2 * linesize, pq);
+        }
+        if (!(flags & RIGHT_EDGE)) {
+            v->vc1dsp.vc1_h_loop_filter4(dst + 8, 2 * linesize, pq);
+            v->vc1dsp.vc1_h_loop_filter4(dst + linesize + 8, 2 * linesize, pq);
+        }
+    }
+}
+
+static av_always_inline void vc1_p_v_intfr_loop_filter(VC1Context *v, uint8_t *dest, int *ttblk,
+                                                       uint32_t flags, uint8_t fieldtx, int block_num)
+{
+    MpegEncContext *s  = &v->s;
+    int pq = v->pq;
+    int tt;
+    int linesize  = block_num > 3 ? s->uvlinesize : s->linesize;
+    uint8_t *dst;
+
+    if (block_num > 3)
+        dst = dest;
+    else
+        dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
+
+    tt = ttblk[0] >> (block_num * 4) & 0xf;
+    if (block_num < 4) {
+        if (fieldtx) {
+            if (block_num < 2) {
+                if (tt == TT_4X4 || tt == TT_8X4)
+                    v->vc1dsp.vc1_v_loop_filter8(dst + 8 * linesize, 2 * linesize, pq);
+                if (!(flags & BOTTOM_EDGE))
+                    v->vc1dsp.vc1_v_loop_filter8(dst + 16 * linesize, 2 * linesize, pq);
+            } else {
+                if (tt == TT_4X4 || tt == TT_8X4)
+                    v->vc1dsp.vc1_v_loop_filter8(dst + linesize, 2 * linesize, pq);
+                if (!(flags & BOTTOM_EDGE))
+                    v->vc1dsp.vc1_v_loop_filter8(dst + 9 * linesize, 2 * linesize, pq);
+            }
+        } else {
+            if (block_num < 2) {
+                if (!(flags & TOP_EDGE) && (tt == TT_4X4 || tt == TT_8X4)) {
+                    v->vc1dsp.vc1_v_loop_filter8(dst + 4 * linesize, 2 * linesize, pq);
+                    v->vc1dsp.vc1_v_loop_filter8(dst + 5 * linesize, 2 * linesize, pq);
+                }
+                v->vc1dsp.vc1_v_loop_filter8(dst + 8 * linesize, 2 * linesize, pq);
+                v->vc1dsp.vc1_v_loop_filter8(dst + 9 * linesize, 2 * linesize, pq);
+            } else if (!(flags & BOTTOM_EDGE)) {
+                if (tt == TT_4X4 || tt == TT_8X4) {
+                    v->vc1dsp.vc1_v_loop_filter8(dst + 4 * linesize, 2 * linesize, pq);
+                    v->vc1dsp.vc1_v_loop_filter8(dst + 5 * linesize, 2 * linesize, pq);
+                }
+                v->vc1dsp.vc1_v_loop_filter8(dst + 8 * linesize, 2 * linesize, pq);
+                v->vc1dsp.vc1_v_loop_filter8(dst + 9 * linesize, 2 * linesize, pq);
+            }
+        }
+    } else {
+        if (!(flags & BOTTOM_EDGE)) {
+            if (!(flags & TOP_EDGE) && (tt == TT_4X4 || tt == TT_8X4)) {
+                v->vc1dsp.vc1_v_loop_filter8(dst + 4 * linesize, 2 * linesize, pq);
+                v->vc1dsp.vc1_v_loop_filter8(dst + 5 * linesize, 2 * linesize, pq);
+            }
+                v->vc1dsp.vc1_v_loop_filter8(dst + 8 * linesize, 2 * linesize, pq);
+                v->vc1dsp.vc1_v_loop_filter8(dst + 9 * linesize, 2 * linesize, pq);
+        }
+    }
+}
+
+void ff_vc1_p_intfr_loop_filter(VC1Context *v)
+{
+    MpegEncContext *s = &v->s;
+    int block_count = CONFIG_GRAY && (s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 4 : 6;
+    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
+    uint8_t *dest;
+    int *ttblk;
+    uint32_t flags;
+    uint8_t fieldtx;
+    int i;
+
+    /* Within a MB, the vertical loop filter always runs before the horizontal.
+     * To accomplish that, we run the V loop filter on all applicable
+     * horizontal borders of the MB above the last overlap filtered MB. Then,
+     * we wait for the loop filter iteration on the next row and next column to
+     * do H loop filter on all applicable vertical borders of this MB.
+     * Therefore, the loop filter trails by two rows and one column relative to
+     * the overlap filter and two rows and two colums relative to the decoding
+     * loop. */
+    if (s->mb_x) {
+        if (s->mb_y >= s->start_mb_y + 1) {
+            dest = s->dest[0] - 16 * s->linesize - 16;
+            ttblk = &v->ttblk[s->mb_x - s->mb_stride - 1];
+            flags = s->mb_y == s->start_mb_y + 1 ? TOP_EDGE : 0;
+            fieldtx = v->fieldtx_plane[mb_pos - s->mb_stride - 1];
+            for (i = 0; i < block_count; i++)
+                vc1_p_v_intfr_loop_filter(v,
+                                          i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 8 : dest,
+                                          ttblk,
+                                          flags,
+                                          fieldtx,
+                                          i);
+        }
+    }
+    if (s->mb_x == s->mb_width - 1) {
+        if (s->mb_y >= s->start_mb_y + 1) {
+            dest = s->dest[0] - 16 * s->linesize;
+            ttblk = &v->ttblk[s->mb_x - s->mb_stride];
+            flags = s->mb_y == s->start_mb_y + 1 ? TOP_EDGE : 0;
+            fieldtx = v->fieldtx_plane[mb_pos - s->mb_stride];
+            for (i = 0; i < block_count; i++)
+                vc1_p_v_intfr_loop_filter(v,
+                                          i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize : dest,
+                                          ttblk,
+                                          flags,
+                                          fieldtx,
+                                          i);
+        }
+    }
+    if (s->mb_y == s->end_mb_y - 1) {
+        if (s->mb_x) {
+            dest = s->dest[0] - 16;
+            ttblk = &v->ttblk[s->mb_x - 1];
+            flags = s->mb_y == s->start_mb_y ? TOP_EDGE | BOTTOM_EDGE : BOTTOM_EDGE;
+            fieldtx = v->fieldtx_plane[mb_pos - 1];
+            for (i = 0; i < block_count; i++)
+                vc1_p_v_intfr_loop_filter(v,
+                                          i > 3 ? s->dest[i - 3] - 8 : dest,
+                                          ttblk,
+                                          flags,
+                                          fieldtx,
+                                          i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            dest = s->dest[0];
+            ttblk = &v->ttblk[s->mb_x];
+            flags = s->mb_y == s->start_mb_y ? TOP_EDGE | BOTTOM_EDGE : BOTTOM_EDGE;
+            fieldtx = v->fieldtx_plane[mb_pos];
+            for (i = 0; i < block_count; i++)
+                vc1_p_v_intfr_loop_filter(v,
+                                          i > 3 ? s->dest[i - 3] : dest,
+                                          ttblk,
+                                          flags,
+                                          fieldtx,
+                                          i);
+        }
+    }
+
+    if (s->mb_y >= s->start_mb_y + 2) {
+        if (s->mb_x >= 2) {
+            dest = s->dest[0] - 32 * s->linesize - 32;
+            ttblk = &v->ttblk[s->mb_x - 2 * s->mb_stride - 2];
+            flags = s->mb_x == 2 ? LEFT_EDGE : 0;
+            fieldtx = v->fieldtx_plane[mb_pos - 2 * s->mb_stride - 2];
+            for (i = 0; i < block_count; i++)
+                vc1_p_h_intfr_loop_filter(v,
+                                          i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize - 16 : dest,
+                                          ttblk,
+                                          flags,
+                                          fieldtx,
+                                          i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            if (s->mb_x >= 1) {
+                dest = s->dest[0] - 32 * s->linesize - 16;
+                ttblk = &v->ttblk[s->mb_x - 2 * s->mb_stride - 1];
+                flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+                fieldtx = v->fieldtx_plane[mb_pos - 2 * s->mb_stride - 1];
+                for (i = 0; i < block_count; i++)
+                    vc1_p_h_intfr_loop_filter(v,
+                                              i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize - 8 : dest,
+                                              ttblk,
+                                              flags,
+                                              fieldtx,
+                                              i);
+            }
+            dest = s->dest[0] - 32 * s->linesize;
+            ttblk = &v->ttblk[s->mb_x - 2 * s->mb_stride];
+            flags = s->mb_x ? RIGHT_EDGE : LEFT_EDGE | RIGHT_EDGE;
+            fieldtx = v->fieldtx_plane[mb_pos - 2 * s->mb_stride];
+            for (i = 0; i < block_count; i++)
+                vc1_p_h_intfr_loop_filter(v,
+                                          i > 3 ? s->dest[i - 3] - 16 * s->uvlinesize : dest,
+                                          ttblk,
+                                          flags,
+                                          fieldtx,
+                                          i);
+        }
+    }
+    if (s->mb_y == s->end_mb_y - 1) {
+        if (s->mb_y >= s->start_mb_y + 1) {
+            if (s->mb_x >= 2) {
+                dest = s->dest[0] - 16 * s->linesize - 32;
+                ttblk = &v->ttblk[s->mb_x - s->mb_stride - 2];
+                flags = s->mb_x == 2 ? LEFT_EDGE : 0;
+                fieldtx = v->fieldtx_plane[mb_pos - s->mb_stride - 2];
+                for (i = 0; i < block_count; i++)
+                    vc1_p_h_intfr_loop_filter(v,
+                                              i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 16 : dest,
+                                              ttblk,
+                                              flags,
+                                              fieldtx,
+                                              i);
+            }
+            if (s->mb_x == s->mb_width - 1) {
+                if (s->mb_x >= 1) {
+                    dest = s->dest[0] - 16 * s->linesize - 16;
+                    ttblk = &v->ttblk[s->mb_x - s->mb_stride - 1];
+                    flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+                    fieldtx = v->fieldtx_plane[mb_pos - s->mb_stride - 1];
+                    for (i = 0; i < block_count; i++)
+                        vc1_p_h_intfr_loop_filter(v,
+                                                  i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 8 : dest,
+                                                  ttblk,
+                                                  flags,
+                                                  fieldtx,
+                                                  i);
+                }
+                dest = s->dest[0] - 16 * s->linesize;
+                ttblk = &v->ttblk[s->mb_x - s->mb_stride];
+                flags = s->mb_x ? RIGHT_EDGE : LEFT_EDGE | RIGHT_EDGE;
+                fieldtx = v->fieldtx_plane[mb_pos - s->mb_stride];
+                for (i = 0; i < block_count; i++)
+                    vc1_p_h_intfr_loop_filter(v,
+                                              i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize : dest,
+                                              ttblk,
+                                              flags,
+                                              fieldtx,
+                                              i);
+            }
+        }
+        if (s->mb_x >= 2) {
+            dest = s->dest[0] - 32;
+            ttblk = &v->ttblk[s->mb_x - 2];
+            flags = s->mb_x == 2 ? LEFT_EDGE : 0;
+            fieldtx = v->fieldtx_plane[mb_pos - 2];
+            for (i = 0; i < block_count; i++)
+                vc1_p_h_intfr_loop_filter(v,
+                                          i > 3 ? s->dest[i - 3] - 16 : dest,
+                                          ttblk,
+                                          flags,
+                                          fieldtx,
+                                          i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            if (s->mb_x >= 1) {
+                dest = s->dest[0] - 16;
+                ttblk = &v->ttblk[s->mb_x - 1];
+                flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+                fieldtx = v->fieldtx_plane[mb_pos - 1];
+                for (i = 0; i < block_count; i++)
+                    vc1_p_h_intfr_loop_filter(v,
+                                              i > 3 ? s->dest[i - 3] - 8 : dest,
+                                              ttblk,
+                                              flags,
+                                              fieldtx,
+                                              i);
+            }
+            dest = s->dest[0];
+            ttblk = &v->ttblk[s->mb_x];
+            flags = s->mb_x ? RIGHT_EDGE : LEFT_EDGE | RIGHT_EDGE;
+            fieldtx = v->fieldtx_plane[mb_pos];
+            for (i = 0; i < block_count; i++)
+                vc1_p_h_intfr_loop_filter(v,
+                                          i > 3 ? s->dest[i - 3] : dest,
+                                          ttblk,
+                                          flags,
+                                          fieldtx,
+                                          i);
+        }
+    }
+}
+
+static av_always_inline void vc1_b_h_intfi_loop_filter(VC1Context *v, uint8_t *dest, uint32_t *cbp,
+                                                       int *ttblk, uint32_t flags, int block_num)
+{
+    MpegEncContext *s  = &v->s;
+    int pq = v->pq;
+    uint8_t *dst;
+    uint32_t block_cbp = cbp[0] >> (block_num * 4);
+    int tt;
+    int idx, linesize  = block_num > 3 ? s->uvlinesize : s->linesize;
+
+    if (block_num > 3)
+        dst = dest;
+    else
+        dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
+
+    if (!(flags & RIGHT_EDGE) || !(block_num & 5)) {
+        if (block_num > 3)
+            v->vc1dsp.vc1_h_loop_filter8(dst + 8, linesize, pq);
+        else
+            v->vc1dsp.vc1_h_loop_filter8(dst + 8, linesize, pq);
+    }
+
+    tt = ttblk[0] >> (block_num * 4) & 0xf;
+    if (tt == TT_4X4 || tt == TT_4X8) {
+        idx = (block_cbp | (block_cbp >> 1)) & 5;
+        if (idx & 1)
+            v->vc1dsp.vc1_h_loop_filter4(dst + 4 * linesize + 4, linesize, pq);
+        if (idx & 4)
+            v->vc1dsp.vc1_h_loop_filter4(dst + 4, linesize, pq);
+    }
+}
+
+static av_always_inline void vc1_b_v_intfi_loop_filter(VC1Context *v, uint8_t *dest, uint32_t *cbp,
+                                                       int *ttblk, uint32_t flags, int block_num)
+{
+    MpegEncContext *s  = &v->s;
+    int pq = v->pq;
+    uint8_t *dst;
+    uint32_t block_cbp = cbp[0] >> (block_num * 4);
+    int tt;
+    int idx, linesize  = block_num > 3 ? s->uvlinesize : s->linesize;
+
+    if (block_num > 3)
+        dst = dest;
+    else
+        dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
+
+    if(!(flags & BOTTOM_EDGE) || block_num < 2)
+        v->vc1dsp.vc1_v_loop_filter8(dst + 8 * linesize, linesize, pq);
+
+    tt = ttblk[0] >> (block_num * 4) & 0xf;
+    if (tt == TT_4X4 || tt == TT_8X4) {
+        idx = (block_cbp | (block_cbp >> 2)) & 3;
+        if (idx & 1)
+            v->vc1dsp.vc1_v_loop_filter4(dst + 4 * linesize + 4, linesize, pq);
+        if (idx & 2)
+            v->vc1dsp.vc1_v_loop_filter4(dst + 4 * linesize, linesize, pq);
+    }
+}
+
+void ff_vc1_b_intfi_loop_filter(VC1Context *v)
+{
+    MpegEncContext *s = &v->s;
+    int block_count = CONFIG_GRAY && (s->avctx->flags & AV_CODEC_FLAG_GRAY) ? 4 : 6;
+    uint8_t *dest;
+    uint32_t *cbp;
+    int *ttblk;
+    uint32_t flags = 0;
+    int i;
+
+    /* Within a MB, the vertical loop filter always runs before the horizontal.
+     * To accomplish that, we run the V loop filter on all applicable
+     * horizontal borders of the MB above the currently decoded MB. Then,
+     * we wait for the next loop filter iteration to do H loop filter on all
+     * applicable vertical borders of this MB. Therefore, the loop filter
+     * trails by one row and one column relative to the decoding loop. */
+    if (!s->first_slice_line) {
+        dest = s->dest[0] - 16 * s->linesize;
+        cbp = &v->cbp[s->mb_x - s->mb_stride];
+        ttblk = &v->ttblk[s->mb_x - s->mb_stride];
+        flags = s->mb_y == s->start_mb_y + 1 ? TOP_EDGE : 0;
+        for (i = 0; i < block_count; i++)
+            vc1_b_v_intfi_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize : dest, cbp, ttblk, flags, i);
+    }
+    if (s->mb_y == s->end_mb_y - 1) {
+        dest = s->dest[0];
+        cbp = &v->cbp[s->mb_x];
+        ttblk = &v->ttblk[s->mb_x];
+        flags = s->first_slice_line ? TOP_EDGE | BOTTOM_EDGE : BOTTOM_EDGE;
+        for (i = 0; i < block_count; i++)
+            vc1_b_v_intfi_loop_filter(v, i > 3 ? s->dest[i - 3] : dest, cbp, ttblk, flags, i);
+    }
+
+    if (!s->first_slice_line) {
+        dest = s->dest[0] - 16 * s->linesize - 16;
+        cbp = &v->cbp[s->mb_x - s->mb_stride - 1];
+        ttblk = &v->ttblk[s->mb_x - s->mb_stride - 1];
+        if (s->mb_x) {
+            flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+            for (i = 0; i < block_count; i++)
+                vc1_b_h_intfi_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize - 8 : dest, cbp, ttblk, flags, i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            dest += 16;
+            cbp++;
+            ttblk++;
+            flags = s->mb_x == 0 ? LEFT_EDGE | RIGHT_EDGE : RIGHT_EDGE;
+            for (i = 0; i < block_count; i++)
+                vc1_b_h_intfi_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 * s->uvlinesize : dest, cbp, ttblk, flags, i);
+        }
+    }
+    if (s->mb_y == s->end_mb_y - 1) {
+        dest = s->dest[0] - 16;
+        cbp = &v->cbp[s->mb_x - 1];
+        ttblk = &v->ttblk[s->mb_x - 1];
+        if (s->mb_x) {
+            flags = s->mb_x == 1 ? LEFT_EDGE : 0;
+            for (i = 0; i < block_count; i++)
+                vc1_b_h_intfi_loop_filter(v, i > 3 ? s->dest[i - 3] - 8 : dest, cbp, ttblk, flags, i);
+        }
+        if (s->mb_x == s->mb_width - 1) {
+            dest += 16;
+            cbp++;
+            ttblk++;
+            flags = s->mb_x == 0 ? LEFT_EDGE | RIGHT_EDGE : RIGHT_EDGE;
+            for (i = 0; i < block_count; i++)
+                vc1_b_h_intfi_loop_filter(v, i > 3 ? s->dest[i - 3] : dest, cbp, ttblk, flags, i);
+        }
+    }
+}