Merge commit '7acdd3a1275bcd9cad48f9632169f6bbaeb39d84'
[ffmpeg.git] / libavcodec / hevc_filter.c
index 91a1e1f..3aaf3e5 100644 (file)
@@ -5,20 +5,20 @@
  * Copyright (C) 2013 Seppo Tomperi
  * Copyright (C) 2013 Wassim Hamidouche
  *
- * This file is part of Libav.
+ * This file is part of FFmpeg.
  *
- * Libav is free software; you can redistribute it and/or
+ * FFmpeg is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
- * Libav is distributed in the hope that it will be useful,
+ * FFmpeg is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
+ * License along with FFmpeg; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
@@ -29,6 +29,8 @@
 #include "golomb.h"
 #include "hevc.h"
 
+#include "bit_depth_template.c"
+
 #define LUMA 0
 #define CB 1
 #define CR 2
@@ -58,29 +60,31 @@ static int chroma_tc(HEVCContext *s, int qp_y, int c_idx, int tc_offset)
     else
         offset = s->pps->cr_qp_offset;
 
-    qp_i = av_clip_c(qp_y + offset, 0, 57);
-    if (qp_i < 30)
-        qp = qp_i;
-    else if (qp_i > 43)
-        qp = qp_i - 6;
-    else
-        qp = qp_c[qp_i - 30];
+    qp_i = av_clip(qp_y + offset, 0, 57);
+    if (s->sps->chroma_format_idc == 1) {
+        if (qp_i < 30)
+            qp = qp_i;
+        else if (qp_i > 43)
+            qp = qp_i - 6;
+        else
+            qp = qp_c[qp_i - 30];
+    } else {
+        qp = av_clip(qp_i, 0, 51);
+    }
 
-    idxt = av_clip_c(qp + DEFAULT_INTRA_TC_OFFSET + tc_offset, 0, 53);
+    idxt = av_clip(qp + DEFAULT_INTRA_TC_OFFSET + tc_offset, 0, 53);
     return tctable[idxt];
 }
 
-static int get_qPy_pred(HEVCContext *s, int xC, int yC,
-                        int xBase, int yBase, int log2_cb_size)
+static int get_qPy_pred(HEVCContext *s, int xBase, int yBase, int log2_cb_size)
 {
-    HEVCLocalContext *lc     = &s->HEVClc;
+    HEVCLocalContext *lc     = s->HEVClc;
     int ctb_size_mask        = (1 << s->sps->log2_ctb_size) - 1;
     int MinCuQpDeltaSizeMask = (1 << (s->sps->log2_ctb_size -
                                       s->pps->diff_cu_qp_delta_depth)) - 1;
     int xQgBase              = xBase - (xBase & MinCuQpDeltaSizeMask);
     int yQgBase              = yBase - (yBase & MinCuQpDeltaSizeMask);
     int min_cb_width         = s->sps->min_cb_width;
-    int min_cb_height        = s->sps->min_cb_height;
     int x_cb                 = xQgBase >> s->sps->log2_min_cb_size;
     int y_cb                 = yQgBase >> s->sps->log2_min_cb_size;
     int availableA           = (xBase   & ctb_size_mask) &&
@@ -94,46 +98,7 @@ static int get_qPy_pred(HEVCContext *s, int xC, int yC,
         lc->first_qp_group = !lc->tu.is_cu_qp_delta_coded;
         qPy_pred = s->sh.slice_qp;
     } else {
-        qPy_pred = lc->qp_y;
-        if (log2_cb_size < s->sps->log2_ctb_size -
-                           s->pps->diff_cu_qp_delta_depth) {
-            static const int offsetX[8][8] = {
-                { -1, 1, 3, 1, 7, 1, 3, 1 },
-                {  0, 0, 0, 0, 0, 0, 0, 0 },
-                {  1, 3, 1, 3, 1, 3, 1, 3 },
-                {  2, 2, 2, 2, 2, 2, 2, 2 },
-                {  3, 5, 7, 5, 3, 5, 7, 5 },
-                {  4, 4, 4, 4, 4, 4, 4, 4 },
-                {  5, 7, 5, 7, 5, 7, 5, 7 },
-                {  6, 6, 6, 6, 6, 6, 6, 6 }
-            };
-            static const int offsetY[8][8] = {
-                { 7, 0, 1, 2, 3, 4, 5, 6 },
-                { 0, 1, 2, 3, 4, 5, 6, 7 },
-                { 1, 0, 3, 2, 5, 4, 7, 6 },
-                { 0, 1, 2, 3, 4, 5, 6, 7 },
-                { 3, 0, 1, 2, 7, 4, 5, 6 },
-                { 0, 1, 2, 3, 4, 5, 6, 7 },
-                { 1, 0, 3, 2, 5, 4, 7, 6 },
-                { 0, 1, 2, 3, 4, 5, 6, 7 }
-            };
-            int xC0b = (xC - (xC & ctb_size_mask)) >> s->sps->log2_min_cb_size;
-            int yC0b = (yC - (yC & ctb_size_mask)) >> s->sps->log2_min_cb_size;
-            int idxX = (xQgBase  & ctb_size_mask)  >> s->sps->log2_min_cb_size;
-            int idxY = (yQgBase  & ctb_size_mask)  >> s->sps->log2_min_cb_size;
-            int idx_mask = ctb_size_mask >> s->sps->log2_min_cb_size;
-            int x, y;
-
-            x = FFMIN(xC0b +  offsetX[idxX][idxY],             min_cb_width  - 1);
-            y = FFMIN(yC0b + (offsetY[idxX][idxY] & idx_mask), min_cb_height - 1);
-
-            if (xC0b == (lc->start_of_tiles_x >> s->sps->log2_min_cb_size) &&
-                offsetX[idxX][idxY] == -1) {
-                x = (lc->end_of_tiles_x >> s->sps->log2_min_cb_size) - 1;
-                y = yC0b - 1;
-            }
-            qPy_pred = s->qp_y_tab[y * min_cb_width + x];
-        }
+        qPy_pred = lc->qPy_pred;
     }
 
     // qPy_a
@@ -148,20 +113,22 @@ static int get_qPy_pred(HEVCContext *s, int xC, int yC,
     else
         qPy_b = s->qp_y_tab[x_cb + (y_cb - 1) * min_cb_width];
 
+    av_assert2(qPy_a >= -s->sps->qp_bd_offset && qPy_a < 52);
+    av_assert2(qPy_b >= -s->sps->qp_bd_offset && qPy_b < 52);
+
     return (qPy_a + qPy_b + 1) >> 1;
 }
 
-void ff_hevc_set_qPy(HEVCContext *s, int xC, int yC,
-                     int xBase, int yBase, int log2_cb_size)
+void ff_hevc_set_qPy(HEVCContext *s, int xBase, int yBase, int log2_cb_size)
 {
-    int qp_y = get_qPy_pred(s, xC, yC, xBase, yBase, log2_cb_size);
+    int qp_y = get_qPy_pred(s, xBase, yBase, log2_cb_size);
 
-    if (s->HEVClc.tu.cu_qp_delta != 0) {
+    if (s->HEVClc->tu.cu_qp_delta != 0) {
         int off = s->sps->qp_bd_offset;
-        s->HEVClc.qp_y = FFUMOD(qp_y + s->HEVClc.tu.cu_qp_delta + 52 + 2 * off,
-                                52 + off) - off;
+        s->HEVClc->qp_y = FFUMOD(qp_y + s->HEVClc->tu.cu_qp_delta + 52 + 2 * off,
+                                 52 + off) - off;
     } else
-        s->HEVClc.qp_y = qp_y;
+        s->HEVClc->qp_y = qp_y;
 }
 
 static int get_qPy(HEVCContext *s, int xC, int yC)
@@ -173,14 +140,46 @@ static int get_qPy(HEVCContext *s, int xC, int yC)
 }
 
 static void copy_CTB(uint8_t *dst, uint8_t *src,
-                     int width, int height, int stride)
+                     int width, int height, int stride_dst, int stride_src)
 {
     int i;
 
     for (i = 0; i < height; i++) {
         memcpy(dst, src, width);
-        dst += stride;
-        src += stride;
+        dst += stride_dst;
+        src += stride_src;
+    }
+}
+
+static void restore_tqb_pixels(HEVCContext *s, int x0, int y0, int width, int height, int c_idx)
+{
+    if ( s->pps->transquant_bypass_enable_flag ||
+            (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) {
+        int x, y;
+        ptrdiff_t stride_dst = s->sao_frame->linesize[c_idx];
+        ptrdiff_t stride_src = s->frame->linesize[c_idx];
+        int min_pu_size  = 1 << s->sps->log2_min_pu_size;
+        int hshift       = s->sps->hshift[c_idx];
+        int vshift       = s->sps->vshift[c_idx];
+        int x_min        = ((x0         ) >> s->sps->log2_min_pu_size);
+        int y_min        = ((y0         ) >> s->sps->log2_min_pu_size);
+        int x_max        = ((x0 + width ) >> s->sps->log2_min_pu_size);
+        int y_max        = ((y0 + height) >> s->sps->log2_min_pu_size);
+        int len          = min_pu_size >> hshift;
+        for (y = y_min; y < y_max; y++) {
+            for (x = x_min; x < x_max; x++) {
+                if (s->is_pcm[y * s->sps->min_pu_width + x]) {
+                    int n;
+                    uint8_t *src = &s->frame->data[c_idx][    ((y << s->sps->log2_min_pu_size) >> vshift) * stride_src + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
+                    uint8_t *dst = &s->sao_frame->data[c_idx][((y << s->sps->log2_min_pu_size) >> vshift) * stride_dst + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
+                    for (n = 0; n < (min_pu_size >> vshift); n++) {
+                        memcpy(src, dst, len);
+                        src += stride_src;
+                        dst += stride_dst;
+                    }
+                }
+            }
+        }
     }
 }
 
@@ -188,128 +187,127 @@ static void copy_CTB(uint8_t *dst, uint8_t *src,
 
 static void sao_filter_CTB(HEVCContext *s, int x, int y)
 {
-    //  TODO: This should be easily parallelizable
-    //  TODO: skip CBs when (cu_transquant_bypass_flag || (pcm_loop_filter_disable_flag && pcm_flag))
-    int c_idx = 0;
-    int class = 1, class_index;
+    int c_idx;
     int edges[4];  // 0 left 1 top 2 right 3 bottom
-    SAOParams *sao[4];
-    int classes[4];
-    int x_shift = 0, y_shift = 0;
-    int x_ctb = x >> s->sps->log2_ctb_size;
-    int y_ctb = y >> s->sps->log2_ctb_size;
-    int ctb_addr_rs = y_ctb * s->sps->ctb_width + x_ctb;
-    int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
-
+    int x_ctb                = x >> s->sps->log2_ctb_size;
+    int y_ctb                = y >> s->sps->log2_ctb_size;
+    int ctb_addr_rs          = y_ctb * s->sps->ctb_width + x_ctb;
+    int ctb_addr_ts          = s->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
+    SAOParams *sao           = &CTB(s->sao, x_ctb, y_ctb);
     // flags indicating unfilterable edges
-    uint8_t vert_edge[]  = { 0, 0, 0, 0 };
-    uint8_t horiz_edge[] = { 0, 0, 0, 0 };
-    uint8_t diag_edge[]  = { 0, 0, 0, 0 };
-    uint8_t lfase[3]; // current, above, left
-    uint8_t no_tile_filter = s->pps->tiles_enabled_flag &&
-                             !s->pps->loop_filter_across_tiles_enabled_flag;
-    uint8_t left_tile_edge = 0, up_tile_edge = 0;
-
-    sao[0]     = &CTB(s->sao, x_ctb, y_ctb);
+    uint8_t vert_edge[]      = { 0, 0 };
+    uint8_t horiz_edge[]     = { 0, 0 };
+    uint8_t diag_edge[]      = { 0, 0, 0, 0 };
+    uint8_t lfase            = CTB(s->filter_slice_edges, x_ctb, y_ctb);
+    uint8_t no_tile_filter   = s->pps->tiles_enabled_flag &&
+                               !s->pps->loop_filter_across_tiles_enabled_flag;
+    uint8_t restore          = no_tile_filter || !lfase;
+    uint8_t left_tile_edge   = 0;
+    uint8_t right_tile_edge  = 0;
+    uint8_t up_tile_edge     = 0;
+    uint8_t bottom_tile_edge = 0;
+
     edges[0]   = x_ctb == 0;
     edges[1]   = y_ctb == 0;
     edges[2]   = x_ctb == s->sps->ctb_width  - 1;
     edges[3]   = y_ctb == s->sps->ctb_height - 1;
-    lfase[0]   = CTB(s->filter_slice_edges, x_ctb, y_ctb);
-    classes[0] = 0;
-
-    if (!edges[0]) {
-        left_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
-        sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb);
-        vert_edge[0] = (!lfase[0] && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb)) || left_tile_edge;
-        vert_edge[2] = vert_edge[0];
-        lfase[2]     = CTB(s->filter_slice_edges, x_ctb - 1, y_ctb);
-        classes[class] = 2;
-        class++;
-        x_shift = 8;
-    }
-
-    if (!edges[1]) {
-        up_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]];
-        sao[class] = &CTB(s->sao, x_ctb, y_ctb - 1);
-        horiz_edge[0] = (!lfase[0] && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) || up_tile_edge;
-        horiz_edge[1] = horiz_edge[0];
-        lfase[1] = CTB(s->filter_slice_edges, x_ctb, y_ctb - 1);
-        classes[class] = 1;
-        class++;
-        y_shift = 4;
 
+    if (restore) {
         if (!edges[0]) {
-            classes[class] = 3;
-            sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb - 1);
-            class++;
-
-            // Tile check here is done current CTB row/col, not above/left like you'd expect,
-            //but that is because the tile boundary always extends through the whole pic
-            vert_edge[1] = (!lfase[1] && CTB(s->tab_slice_address, x_ctb, y_ctb - 1) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || left_tile_edge;
-            vert_edge[3] = vert_edge[1];
-            horiz_edge[2] = (!lfase[2] && CTB(s->tab_slice_address, x_ctb - 1, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || up_tile_edge;
-            horiz_edge[3] = horiz_edge[2];
-            diag_edge[0] = (!lfase[0] && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || left_tile_edge || up_tile_edge;
-            diag_edge[3] = diag_edge[0];
-
-            // Does left CTB comes after above CTB?
-            if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) >
-                CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
-                diag_edge[2] = !lfase[2] || left_tile_edge || up_tile_edge;
-                diag_edge[1] = diag_edge[2];
-            } else if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) <
-                       CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
-                diag_edge[1] = !lfase[1] || left_tile_edge || up_tile_edge;
-                diag_edge[2] = diag_edge[1];
-            } else {
-                // Same slice, only consider tiles
-                diag_edge[2] = left_tile_edge || up_tile_edge;
-                diag_edge[1] = diag_edge[2];
-            }
+            left_tile_edge  = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
+            vert_edge[0]    = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb)) || left_tile_edge;
+        }
+        if (!edges[2]) {
+            right_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs+1]];
+            vert_edge[1]    = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb)) || right_tile_edge;
+        }
+        if (!edges[1]) {
+            up_tile_edge     = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]];
+            horiz_edge[0]    = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) || up_tile_edge;
+        }
+        if (!edges[3]) {
+            bottom_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs + s->sps->ctb_width]];
+            horiz_edge[1]    = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb + 1)) || bottom_tile_edge;
+        }
+        if (!edges[0] && !edges[1]) {
+            diag_edge[0] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || left_tile_edge || up_tile_edge;
+        }
+        if (!edges[1] && !edges[2]) {
+            diag_edge[1] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb - 1)) || right_tile_edge || up_tile_edge;
+        }
+        if (!edges[2] && !edges[3]) {
+            diag_edge[2] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb + 1)) || right_tile_edge || bottom_tile_edge;
+        }
+        if (!edges[0] && !edges[3]) {
+            diag_edge[3] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb + 1)) || left_tile_edge || bottom_tile_edge;
         }
     }
 
     for (c_idx = 0; c_idx < 3; c_idx++) {
-        int chroma = c_idx ? 1 : 0;
-        int x0 = x >> chroma;
-        int y0 = y >> chroma;
-        int stride = s->frame->linesize[c_idx];
-        int ctb_size = (1 << (s->sps->log2_ctb_size)) >> s->sps->hshift[c_idx];
-        int width = FFMIN(ctb_size,
-                          (s->sps->width >> s->sps->hshift[c_idx]) - x0);
-        int height = FFMIN(ctb_size,
-                           (s->sps->height >> s->sps->vshift[c_idx]) - y0);
-
-        uint8_t *src = &s->frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)];
-        uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)];
-        int offset = (y_shift >> chroma) * stride + ((x_shift >> chroma) << s->sps->pixel_shift);
-
-        copy_CTB(dst - offset, src - offset,
-                 (edges[2] ? width  + (x_shift >> chroma) : width)  << s->sps->pixel_shift,
-                 (edges[3] ? height + (y_shift >> chroma) : height), stride);
-
-        for (class_index = 0; class_index < class; class_index++) {
-
-            switch (sao[class_index]->type_idx[c_idx]) {
-            case SAO_BAND:
-                s->hevcdsp.sao_band_filter[classes[class_index]](dst, src,
-                                                                 stride,
-                                                                 sao[class_index],
-                                                                 edges, width,
-                                                                 height, c_idx);
-                break;
-            case SAO_EDGE:
-                s->hevcdsp.sao_edge_filter[classes[class_index]](dst, src,
-                                                                 stride,
-                                                                 sao[class_index],
-                                                                 edges, width,
-                                                                 height, c_idx,
-                                                                 vert_edge[classes[class_index]],
-                                                                 horiz_edge[classes[class_index]],
-                                                                 diag_edge[classes[class_index]]);
-                break;
+        int x0       = x >> s->sps->hshift[c_idx];
+        int y0       = y >> s->sps->vshift[c_idx];
+        int stride_src = s->frame->linesize[c_idx];
+        int stride_dst = s->sao_frame->linesize[c_idx];
+        int ctb_size_h = (1 << (s->sps->log2_ctb_size)) >> s->sps->hshift[c_idx];
+        int ctb_size_v = (1 << (s->sps->log2_ctb_size)) >> s->sps->vshift[c_idx];
+        int width    = FFMIN(ctb_size_h, (s->sps->width  >> s->sps->hshift[c_idx]) - x0);
+        int height   = FFMIN(ctb_size_v, (s->sps->height >> s->sps->vshift[c_idx]) - y0);
+        uint8_t *src = &s->frame->data[c_idx][y0 * stride_src + (x0 << s->sps->pixel_shift)];
+        uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride_dst + (x0 << s->sps->pixel_shift)];
+
+        switch (sao->type_idx[c_idx]) {
+        case SAO_BAND:
+            copy_CTB(dst, src, width << s->sps->pixel_shift, height, stride_dst, stride_src);
+            s->hevcdsp.sao_band_filter(src, dst,
+                                       stride_src, stride_dst,
+                                       sao,
+                                       edges, width,
+                                       height, c_idx);
+            restore_tqb_pixels(s, x, y, width, height, c_idx);
+            sao->type_idx[c_idx] = SAO_APPLIED;
+            break;
+        case SAO_EDGE:
+        {
+            uint8_t left_pixels = !edges[0] && (CTB(s->sao, x_ctb-1, y_ctb).type_idx[c_idx] != SAO_APPLIED);
+            if (!edges[1]) {
+                uint8_t top_left  = !edges[0] && (CTB(s->sao, x_ctb-1, y_ctb-1).type_idx[c_idx] != SAO_APPLIED);
+                uint8_t top_right = !edges[2] && (CTB(s->sao, x_ctb+1, y_ctb-1).type_idx[c_idx] != SAO_APPLIED);
+                if (CTB(s->sao, x_ctb  , y_ctb-1).type_idx[c_idx] == 0)
+                    memcpy( dst - stride_dst - (top_left << s->sps->pixel_shift),
+                            src - stride_src - (top_left << s->sps->pixel_shift),
+                            (top_left + width + top_right) << s->sps->pixel_shift);
+                else {
+                    if (top_left)
+                        memcpy( dst - stride_dst - (1 << s->sps->pixel_shift),
+                                src - stride_src - (1 << s->sps->pixel_shift),
+                                1 << s->sps->pixel_shift);
+                    if(top_right)
+                        memcpy( dst - stride_dst + (width << s->sps->pixel_shift),
+                                src - stride_src + (width << s->sps->pixel_shift),
+                                1 << s->sps->pixel_shift);
+                }
+            }
+            if (!edges[3]) {                                                                // bottom and bottom right
+                uint8_t bottom_left = !edges[0] && (CTB(s->sao, x_ctb-1, y_ctb+1).type_idx[c_idx] != SAO_APPLIED);
+                memcpy( dst + height * stride_dst - (bottom_left << s->sps->pixel_shift),
+                        src + height * stride_src - (bottom_left << s->sps->pixel_shift),
+                        (width + 1 + bottom_left) << s->sps->pixel_shift);
             }
+            copy_CTB(dst - (left_pixels << s->sps->pixel_shift),
+                     src - (left_pixels << s->sps->pixel_shift),
+                     (width + 1 + left_pixels) << s->sps->pixel_shift, height, stride_dst, stride_src);
+            s->hevcdsp.sao_edge_filter[restore](src, dst,
+                                                stride_src, stride_dst,
+                                                sao,
+                                                edges, width,
+                                                height, c_idx,
+                                                vert_edge,
+                                                horiz_edge,
+                                                diag_edge);
+            restore_tqb_pixels(s, x, y, width, height, c_idx);
+            sao->type_idx[c_idx] = SAO_APPLIED;
+            break;
+        }
         }
     }
 }
@@ -338,18 +336,21 @@ static int get_pcm(HEVCContext *s, int x, int y)
 static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
 {
     uint8_t *src;
-    int x, y, x_end, y_end, chroma;
-    int c_tc[2], tc[2], beta;
+    int x, y;
+    int chroma, beta;
+    int32_t c_tc[2], tc[2];
     uint8_t no_p[2] = { 0 };
     uint8_t no_q[2] = { 0 };
 
     int log2_ctb_size = s->sps->log2_ctb_size;
+    int x_end, x_end2, y_end;
     int ctb_size        = 1 << log2_ctb_size;
     int ctb             = (x0 >> log2_ctb_size) +
                           (y0 >> log2_ctb_size) * s->sps->ctb_width;
     int cur_tc_offset   = s->deblock[ctb].tc_offset;
     int cur_beta_offset = s->deblock[ctb].beta_offset;
-    int tc_offset, left_tc_offset, beta_offset, left_beta_offset;
+    int left_tc_offset, left_beta_offset;
+    int tc_offset, beta_offset;
     int pcmf = (s->sps->pcm_enabled_flag &&
                 s->sps->pcm.loop_filter_disable_flag) ||
                s->pps->transquant_bypass_enable_flag;
@@ -357,6 +358,9 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
     if (x0) {
         left_tc_offset   = s->deblock[ctb - 1].tc_offset;
         left_beta_offset = s->deblock[ctb - 1].beta_offset;
+    } else {
+        left_tc_offset   = 0;
+        left_beta_offset = 0;
     }
 
     x_end = x0 + ctb_size;
@@ -369,11 +373,14 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
     tc_offset   = cur_tc_offset;
     beta_offset = cur_beta_offset;
 
-    // vertical filtering luma
+    x_end2 = x_end;
+    if (x_end2 != s->sps->width)
+        x_end2 -= 8;
     for (y = y0; y < y_end; y += 8) {
+        // vertical filtering luma
         for (x = x0 ? x0 : 8; x < x_end; x += 8) {
-            const int bs0 = s->vertical_bs[(x >> 3) + (y       >> 2) * s->bs_width];
-            const int bs1 = s->vertical_bs[(x >> 3) + ((y + 4) >> 2) * s->bs_width];
+            const int bs0 = s->vertical_bs[(x +  y      * s->bs_width) >> 2];
+            const int bs1 = s->vertical_bs[(x + (y + 4) * s->bs_width) >> 2];
             if (bs0 || bs1) {
                 const int qp = (get_qPy(s, x - 1, y)     + get_qPy(s, x, y)     + 1) >> 1;
 
@@ -396,45 +403,14 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
                                                        beta, tc, no_p, no_q);
             }
         }
-    }
-
-    // vertical filtering chroma
-    for (chroma = 1; chroma <= 2; chroma++) {
-        for (y = y0; y < y_end; y += 16) {
-            for (x = x0 ? x0 : 16; x < x_end; x += 16) {
-                const int bs0 = s->vertical_bs[(x >> 3) + (y       >> 2) * s->bs_width];
-                const int bs1 = s->vertical_bs[(x >> 3) + ((y + 8) >> 2) * s->bs_width];
-                if ((bs0 == 2) || (bs1 == 2)) {
-                    const int qp0 = (get_qPy(s, x - 1, y)     + get_qPy(s, x, y)     + 1) >> 1;
-                    const int qp1 = (get_qPy(s, x - 1, y + 8) + get_qPy(s, x, y + 8) + 1) >> 1;
 
-                    c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
-                    c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
-                    src     = &s->frame->data[chroma][y / 2 * s->frame->linesize[chroma] + ((x / 2) << s->sps->pixel_shift)];
-                    if (pcmf) {
-                        no_p[0] = get_pcm(s, x - 1, y);
-                        no_p[1] = get_pcm(s, x - 1, y + 8);
-                        no_q[0] = get_pcm(s, x, y);
-                        no_q[1] = get_pcm(s, x, y + 8);
-                        s->hevcdsp.hevc_v_loop_filter_chroma_c(src,
-                                                               s->frame->linesize[chroma],
-                                                               c_tc, no_p, no_q);
-                    } else
-                        s->hevcdsp.hevc_v_loop_filter_chroma(src,
-                                                             s->frame->linesize[chroma],
-                                                             c_tc, no_p, no_q);
-                }
-            }
-        }
-    }
+        if(!y)
+             continue;
 
-    // horizontal filtering luma
-    if (x_end != s->sps->width)
-        x_end -= 8;
-    for (y = y0 ? y0 : 8; y < y_end; y += 8) {
-        for (x = x0 ? x0 - 8 : 0; x < x_end; x += 8) {
-            const int bs0 = s->horizontal_bs[(x +     y * s->bs_width) >> 2];
-            const int bs1 = s->horizontal_bs[(x + 4 + y * s->bs_width) >> 2];
+        // horizontal filtering luma
+        for (x = x0 ? x0 - 8 : 0; x < x_end2; x += 8) {
+            const int bs0 = s->horizontal_bs[( x      + y * s->bs_width) >> 2];
+            const int bs1 = s->horizontal_bs[((x + 4) + y * s->bs_width) >> 2];
             if (bs0 || bs1) {
                 const int qp = (get_qPy(s, x, y - 1)     + get_qPy(s, x, y)     + 1) >> 1;
 
@@ -461,37 +437,61 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
         }
     }
 
-    // horizontal filtering chroma
     for (chroma = 1; chroma <= 2; chroma++) {
-        for (y = y0 ? y0 : 16; y < y_end; y += 16) {
-            for (x = x0 - 8; x < x_end; x += 16) {
-                int bs0, bs1;
-                // to make sure no memory access over boundary when x = -8
-                // TODO: simplify with row based deblocking
-                if (x < 0) {
-                    bs0 = 0;
-                    bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
-                } else if (x >= x_end - 8) {
-                    bs0 = s->horizontal_bs[(x +     y * s->bs_width) >> 2];
-                    bs1 = 0;
-                } else {
-                    bs0 = s->horizontal_bs[(x + y     * s->bs_width) >> 2];
-                    bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
+        int h = 1 << s->sps->hshift[chroma];
+        int v = 1 << s->sps->vshift[chroma];
+
+        // vertical filtering chroma
+        for (y = y0; y < y_end; y += (8 * v)) {
+            for (x = x0 ? x0 : 8 * h; x < x_end; x += (8 * h)) {
+                const int bs0 = s->vertical_bs[(x +  y            * s->bs_width) >> 2];
+                const int bs1 = s->vertical_bs[(x + (y + (4 * v)) * s->bs_width) >> 2];
+
+                if ((bs0 == 2) || (bs1 == 2)) {
+                    const int qp0 = (get_qPy(s, x - 1, y)           + get_qPy(s, x, y)           + 1) >> 1;
+                    const int qp1 = (get_qPy(s, x - 1, y + (4 * v)) + get_qPy(s, x, y + (4 * v)) + 1) >> 1;
+
+                    c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
+                    c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
+                    src       = &s->frame->data[chroma][(y >> s->sps->vshift[chroma]) * s->frame->linesize[chroma] + ((x >> s->sps->hshift[chroma]) << s->sps->pixel_shift)];
+                    if (pcmf) {
+                        no_p[0] = get_pcm(s, x - 1, y);
+                        no_p[1] = get_pcm(s, x - 1, y + (4 * v));
+                        no_q[0] = get_pcm(s, x, y);
+                        no_q[1] = get_pcm(s, x, y + (4 * v));
+                        s->hevcdsp.hevc_v_loop_filter_chroma_c(src,
+                                                               s->frame->linesize[chroma],
+                                                               c_tc, no_p, no_q);
+                    } else
+                        s->hevcdsp.hevc_v_loop_filter_chroma(src,
+                                                             s->frame->linesize[chroma],
+                                                             c_tc, no_p, no_q);
                 }
+            }
 
+            if(!y)
+                 continue;
+
+            // horizontal filtering chroma
+            tc_offset = x0 ? left_tc_offset : cur_tc_offset;
+            x_end2 = x_end;
+            if (x_end != s->sps->width)
+                x_end2 = x_end - 8 * h;
+            for (x = x0 ? x0 - 8 * h : 0; x < x_end2; x += (8 * h)) {
+                const int bs0 = s->horizontal_bs[( x          + y * s->bs_width) >> 2];
+                const int bs1 = s->horizontal_bs[((x + 4 * h) + y * s->bs_width) >> 2];
                 if ((bs0 == 2) || (bs1 == 2)) {
-                    const int qp0 = bs0 == 2 ? (get_qPy(s, x,     y - 1) + get_qPy(s, x,     y) + 1) >> 1 : 0;
-                    const int qp1 = bs1 == 2 ? (get_qPy(s, x + 8, y - 1) + get_qPy(s, x + 8, y) + 1) >> 1 : 0;
+                    const int qp0 = bs0 == 2 ? (get_qPy(s, x,           y - 1) + get_qPy(s, x,           y) + 1) >> 1 : 0;
+                    const int qp1 = bs1 == 2 ? (get_qPy(s, x + (4 * h), y - 1) + get_qPy(s, x + (4 * h), y) + 1) >> 1 : 0;
 
-                    tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
                     c_tc[0]   = bs0 == 2 ? chroma_tc(s, qp0, chroma, tc_offset)     : 0;
                     c_tc[1]   = bs1 == 2 ? chroma_tc(s, qp1, chroma, cur_tc_offset) : 0;
-                    src       = &s->frame->data[chroma][y / 2 * s->frame->linesize[chroma] + ((x / 2) << s->sps->pixel_shift)];
+                    src       = &s->frame->data[chroma][(y >> s->sps->vshift[1]) * s->frame->linesize[chroma] + ((x >> s->sps->hshift[1]) << s->sps->pixel_shift)];
                     if (pcmf) {
-                        no_p[0] = get_pcm(s, x, y - 1);
-                        no_p[1] = get_pcm(s, x + 8, y - 1);
-                        no_q[0] = get_pcm(s, x, y);
-                        no_q[1] = get_pcm(s, x + 8, y);
+                        no_p[0] = get_pcm(s, x,           y - 1);
+                        no_p[1] = get_pcm(s, x + (4 * h), y - 1);
+                        no_q[0] = get_pcm(s, x,           y);
+                        no_q[1] = get_pcm(s, x + (4 * h), y);
                         s->hevcdsp.hevc_h_loop_filter_chroma_c(src,
                                                                s->frame->linesize[chroma],
                                                                c_tc, no_p, no_q);
@@ -505,79 +505,65 @@ static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
     }
 }
 
-static int boundary_strength(HEVCContext *s, MvField *curr,
-                             uint8_t curr_cbf_luma, MvField *neigh,
-                             uint8_t neigh_cbf_luma,
-                             RefPicList *neigh_refPicList,
-                             int tu_border)
+static int boundary_strength(HEVCContext *s, MvField *curr, MvField *neigh,
+                             RefPicList *neigh_refPicList)
 {
-    int mvs = curr->pred_flag[0] + curr->pred_flag[1];
-
-    if (tu_border) {
-        if (curr->is_intra || neigh->is_intra)
-            return 2;
-        if (curr_cbf_luma || neigh_cbf_luma)
-            return 1;
-    }
-
-    if (mvs == neigh->pred_flag[0] + neigh->pred_flag[1]) {
-        if (mvs == 2) {
-            // same L0 and L1
-            if (s->ref->refPicList[0].list[curr->ref_idx[0]] == neigh_refPicList[0].list[neigh->ref_idx[0]]  &&
-                s->ref->refPicList[0].list[curr->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]] &&
-                neigh_refPicList[0].list[neigh->ref_idx[0]] == neigh_refPicList[1].list[neigh->ref_idx[1]]) {
-                if ((abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
-                     abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4) &&
-                    (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
-                     abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4))
-                    return 1;
-                else
-                    return 0;
-            } else if (neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
-                       neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
-                if (abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
-                    abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4)
-                    return 1;
-                else
-                    return 0;
-            } else if (neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
-                       neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
-                if (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
-                    abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4)
-                    return 1;
-                else
-                    return 0;
-            } else {
+    if (curr->pred_flag == PF_BI &&  neigh->pred_flag == PF_BI) {
+        // same L0 and L1
+        if (s->ref->refPicList[0].list[curr->ref_idx[0]] == neigh_refPicList[0].list[neigh->ref_idx[0]]  &&
+            s->ref->refPicList[0].list[curr->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]] &&
+            neigh_refPicList[0].list[neigh->ref_idx[0]] == neigh_refPicList[1].list[neigh->ref_idx[1]]) {
+            if ((FFABS(neigh->mv[0].x - curr->mv[0].x) >= 4 || FFABS(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
+                 FFABS(neigh->mv[1].x - curr->mv[1].x) >= 4 || FFABS(neigh->mv[1].y - curr->mv[1].y) >= 4) &&
+                (FFABS(neigh->mv[1].x - curr->mv[0].x) >= 4 || FFABS(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
+                 FFABS(neigh->mv[0].x - curr->mv[1].x) >= 4 || FFABS(neigh->mv[0].y - curr->mv[1].y) >= 4))
                 return 1;
-            }
-        } else { // 1 MV
-            Mv A, B;
-            int ref_A, ref_B;
-
-            if (curr->pred_flag[0]) {
-                A     = curr->mv[0];
-                ref_A = s->ref->refPicList[0].list[curr->ref_idx[0]];
-            } else {
-                A     = curr->mv[1];
-                ref_A = s->ref->refPicList[1].list[curr->ref_idx[1]];
-            }
+            else
+                return 0;
+        } else if (neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
+                   neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
+            if (FFABS(neigh->mv[0].x - curr->mv[0].x) >= 4 || FFABS(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
+                FFABS(neigh->mv[1].x - curr->mv[1].x) >= 4 || FFABS(neigh->mv[1].y - curr->mv[1].y) >= 4)
+                return 1;
+            else
+                return 0;
+        } else if (neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
+                   neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
+            if (FFABS(neigh->mv[1].x - curr->mv[0].x) >= 4 || FFABS(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
+                FFABS(neigh->mv[0].x - curr->mv[1].x) >= 4 || FFABS(neigh->mv[0].y - curr->mv[1].y) >= 4)
+                return 1;
+            else
+                return 0;
+        } else {
+            return 1;
+        }
+    } else if ((curr->pred_flag != PF_BI) && (neigh->pred_flag != PF_BI)){ // 1 MV
+        Mv A, B;
+        int ref_A, ref_B;
+
+        if (curr->pred_flag & 1) {
+            A     = curr->mv[0];
+            ref_A = s->ref->refPicList[0].list[curr->ref_idx[0]];
+        } else {
+            A     = curr->mv[1];
+            ref_A = s->ref->refPicList[1].list[curr->ref_idx[1]];
+        }
 
-            if (neigh->pred_flag[0]) {
-                B     = neigh->mv[0];
-                ref_B = neigh_refPicList[0].list[neigh->ref_idx[0]];
-            } else {
-                B     = neigh->mv[1];
-                ref_B = neigh_refPicList[1].list[neigh->ref_idx[1]];
-            }
+        if (neigh->pred_flag & 1) {
+            B     = neigh->mv[0];
+            ref_B = neigh_refPicList[0].list[neigh->ref_idx[0]];
+        } else {
+            B     = neigh->mv[1];
+            ref_B = neigh_refPicList[1].list[neigh->ref_idx[1]];
+        }
 
-            if (ref_A == ref_B) {
-                if (abs(A.x - B.x) >= 4 || abs(A.y - B.y) >= 4)
-                    return 1;
-                else
-                    return 0;
-            } else
+        if (ref_A == ref_B) {
+            if (FFABS(A.x - B.x) >= 4 || FFABS(A.y - B.y) >= 4)
                 return 1;
-        }
+            else
+                return 0;
+        } else
+            return 1;
     }
 
     return 1;
@@ -586,14 +572,14 @@ static int boundary_strength(HEVCContext *s, MvField *curr,
 void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
                                            int log2_trafo_size)
 {
-    HEVCLocalContext *lc = &s->HEVClc;
+    HEVCLocalContext *lc = s->HEVClc;
     MvField *tab_mvf     = s->ref->tab_mvf;
     int log2_min_pu_size = s->sps->log2_min_pu_size;
     int log2_min_tu_size = s->sps->log2_min_tb_size;
     int min_pu_width     = s->sps->min_pu_width;
     int min_tu_width     = s->sps->min_tb_width;
     int is_intra = tab_mvf[(y0 >> log2_min_pu_size) * min_pu_width +
-                           (x0 >> log2_min_pu_size)].is_intra;
+                           (x0 >> log2_min_pu_size)].pred_flag == PF_INTRA;
     int boundary_upper, boundary_left;
     int i, j, bs;
 
@@ -611,37 +597,11 @@ void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
         RefPicList *rpl_top = (lc->boundary_flags & BOUNDARY_UPPER_SLICE) ?
                               ff_hevc_get_ref_list(s, s->ref, x0, y0 - 1) :
                               s->ref->refPicList;
-
         int yp_pu = (y0 - 1) >> log2_min_pu_size;
         int yq_pu =  y0      >> log2_min_pu_size;
         int yp_tu = (y0 - 1) >> log2_min_tu_size;
         int yq_tu =  y0      >> log2_min_tu_size;
 
-        for (i = 0; i < (1 << log2_trafo_size); i += 4) {
-            int x_pu = (x0 + i) >> log2_min_pu_size;
-            int x_tu = (x0 + i) >> log2_min_tu_size;
-            MvField *top  = &tab_mvf[yp_pu * min_pu_width + x_pu];
-            MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
-            uint8_t top_cbf_luma  = s->cbf_luma[yp_tu * min_tu_width + x_tu];
-            uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
-
-            bs = boundary_strength(s, curr, curr_cbf_luma,
-                                   top, top_cbf_luma, rpl_top, 1);
-            if (bs)
-                s->horizontal_bs[((x0 + i) + y0 * s->bs_width) >> 2] = bs;
-        }
-    }
-
-    // bs for TU internal horizontal PU boundaries
-    if (log2_trafo_size > s->sps->log2_min_pu_size && !is_intra) {
-        RefPicList *rpl = s->ref->refPicList;
-
-        for (j = 8; j < (1 << log2_trafo_size); j += 8) {
-            int yp_pu = (y0 + j - 1) >> log2_min_pu_size;
-            int yq_pu = (y0 + j)     >> log2_min_pu_size;
-            int yp_tu = (y0 + j - 1) >> log2_min_tu_size;
-            int yq_tu = (y0 + j)     >> log2_min_tu_size;
-
             for (i = 0; i < (1 << log2_trafo_size); i += 4) {
                 int x_pu = (x0 + i) >> log2_min_pu_size;
                 int x_tu = (x0 + i) >> log2_min_tu_size;
@@ -650,12 +610,14 @@ void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
                 uint8_t top_cbf_luma  = s->cbf_luma[yp_tu * min_tu_width + x_tu];
                 uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
 
-                bs = boundary_strength(s, curr, curr_cbf_luma,
-                                       top, top_cbf_luma, rpl, 0);
-                if (bs)
-                    s->horizontal_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
+                if (curr->pred_flag == PF_INTRA || top->pred_flag == PF_INTRA)
+                    bs = 2;
+                else if (curr_cbf_luma || top_cbf_luma)
+                    bs = 1;
+                else
+                    bs = boundary_strength(s, curr, top, rpl_top);
+                s->horizontal_bs[((x0 + i) + y0 * s->bs_width) >> 2] = bs;
             }
-        }
     }
 
     // bs for vertical TU boundaries
@@ -673,50 +635,59 @@ void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
         RefPicList *rpl_left = (lc->boundary_flags & BOUNDARY_LEFT_SLICE) ?
                                ff_hevc_get_ref_list(s, s->ref, x0 - 1, y0) :
                                s->ref->refPicList;
-
         int xp_pu = (x0 - 1) >> log2_min_pu_size;
         int xq_pu =  x0      >> log2_min_pu_size;
         int xp_tu = (x0 - 1) >> log2_min_tu_size;
         int xq_tu =  x0      >> log2_min_tu_size;
 
-        for (i = 0; i < (1 << log2_trafo_size); i += 4) {
-            int y_pu      = (y0 + i) >> log2_min_pu_size;
-            int y_tu      = (y0 + i) >> log2_min_tu_size;
-            MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
-            MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
-
-            uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
-            uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
+            for (i = 0; i < (1 << log2_trafo_size); i += 4) {
+                int y_pu      = (y0 + i) >> log2_min_pu_size;
+                int y_tu      = (y0 + i) >> log2_min_tu_size;
+                MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
+                MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
+                uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
+                uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
 
-            bs = boundary_strength(s, curr, curr_cbf_luma,
-                                   left, left_cbf_luma, rpl_left, 1);
-            if (bs)
-                s->vertical_bs[(x0 >> 3) + ((y0 + i) >> 2) * s->bs_width] = bs;
-        }
+                if (curr->pred_flag == PF_INTRA || left->pred_flag == PF_INTRA)
+                    bs = 2;
+                else if (curr_cbf_luma || left_cbf_luma)
+                    bs = 1;
+                else
+                    bs = boundary_strength(s, curr, left, rpl_left);
+                s->vertical_bs[(x0 + (y0 + i) * s->bs_width) >> 2] = bs;
+            }
     }
 
-    // bs for TU internal vertical PU boundaries
     if (log2_trafo_size > log2_min_pu_size && !is_intra) {
         RefPicList *rpl = s->ref->refPicList;
 
+        // bs for TU internal horizontal PU boundaries
+        for (j = 8; j < (1 << log2_trafo_size); j += 8) {
+            int yp_pu = (y0 + j - 1) >> log2_min_pu_size;
+            int yq_pu = (y0 + j)     >> log2_min_pu_size;
+
+            for (i = 0; i < (1 << log2_trafo_size); i += 4) {
+                int x_pu = (x0 + i) >> log2_min_pu_size;
+                MvField *top  = &tab_mvf[yp_pu * min_pu_width + x_pu];
+                MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
+
+                bs = boundary_strength(s, curr, top, rpl);
+                s->horizontal_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
+            }
+        }
+
+        // bs for TU internal vertical PU boundaries
         for (j = 0; j < (1 << log2_trafo_size); j += 4) {
             int y_pu = (y0 + j) >> log2_min_pu_size;
-            int y_tu = (y0 + j) >> log2_min_tu_size;
 
             for (i = 8; i < (1 << log2_trafo_size); i += 8) {
                 int xp_pu = (x0 + i - 1) >> log2_min_pu_size;
                 int xq_pu = (x0 + i)     >> log2_min_pu_size;
-                int xp_tu = (x0 + i - 1) >> log2_min_tu_size;
-                int xq_tu = (x0 + i)     >> log2_min_tu_size;
                 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
                 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
-                uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
-                uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
 
-                bs = boundary_strength(s, curr, curr_cbf_luma,
-                                       left, left_cbf_luma, rpl, 0);
-                if (bs)
-                    s->vertical_bs[((x0 + i) >> 3) + ((y0 + j) >> 2) * s->bs_width] = bs;
+                bs = boundary_strength(s, curr, left, rpl);
+                s->vertical_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
             }
         }
     }
@@ -726,21 +697,38 @@ void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
 #undef CB
 #undef CR
 
-void ff_hevc_hls_filter(HEVCContext *s, int x, int y)
+void ff_hevc_hls_filter(HEVCContext *s, int x, int y, int ctb_size)
 {
+    int x_end = x >= s->sps->width  - ctb_size;
     deblocking_filter_CTB(s, x, y);
-    if (s->sps->sao_enabled)
-        sao_filter_CTB(s, x, y);
+    if (s->sps->sao_enabled) {
+        int y_end = y >= s->sps->height - ctb_size;
+        if (y && x)
+            sao_filter_CTB(s, x - ctb_size, y - ctb_size);
+        if (x && y_end)
+            sao_filter_CTB(s, x - ctb_size, y);
+        if (y && x_end) {
+            sao_filter_CTB(s, x, y - ctb_size);
+            if (s->threads_type & FF_THREAD_FRAME )
+                ff_thread_report_progress(&s->ref->tf, y, 0);
+        }
+        if (x_end && y_end) {
+            sao_filter_CTB(s, x , y);
+            if (s->threads_type & FF_THREAD_FRAME )
+                ff_thread_report_progress(&s->ref->tf, y + ctb_size, 0);
+        }
+    } else if (s->threads_type & FF_THREAD_FRAME && x_end)
+        ff_thread_report_progress(&s->ref->tf, y + ctb_size - 4, 0);
 }
 
 void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size)
 {
+    int x_end = x_ctb >= s->sps->width  - ctb_size;
+    int y_end = y_ctb >= s->sps->height - ctb_size;
     if (y_ctb && x_ctb)
-        ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size);
-    if (y_ctb && x_ctb >= s->sps->width - ctb_size) {
-        ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size);
-        ff_thread_report_progress(&s->ref->tf, y_ctb - ctb_size, 0);
-    }
-    if (x_ctb && y_ctb >= s->sps->height - ctb_size)
-        ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb);
+        ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size, ctb_size);
+    if (y_ctb && x_end)
+        ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size, ctb_size);
+    if (x_ctb && y_end)
+        ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb, ctb_size);
 }