libavformat/segment: add an option to write the header to a separate file
[ffmpeg.git] / libavcodec / h264_mb.c
index 31e372b..9036919 100644 (file)
@@ -2,20 +2,20 @@
  * H.26L/H.264/AVC/JVT/14496-10/... decoder
  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  *
- * 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
  */
 
 #include "thread.h"
 
 static inline int get_lowest_part_list_y(H264SliceContext *sl,
-                                         H264Picture *pic, int n,
-                                         int height, int y_offset, int list)
+                                         int n, int height, int y_offset, int list)
 {
     int raw_my             = sl->mv_cache[list][scan8[n]][1];
-    int filter_height_up   = (raw_my & 3) ? 2 : 0;
     int filter_height_down = (raw_my & 3) ? 3 : 0;
     int full_my            = (raw_my >> 2) + y_offset;
-    int top                = full_my - filter_height_up;
     int bottom             = full_my + filter_height_down + height;
 
-    return FFMAX(abs(top), bottom);
+    av_assert2(height >= 0);
+
+    return FFMAX(0, bottom);
 }
 
 static inline void get_lowest_part_y(const H264Context *h, H264SliceContext *sl,
-                                     int refs[2][48], int n,
+                                     int16_t refs[2][48], int n,
                                      int height, int y_offset, int list0,
                                      int list1, int *nrefs)
 {
@@ -61,14 +60,14 @@ static inline void get_lowest_part_y(const H264Context *h, H264SliceContext *sl,
 
     if (list0) {
         int ref_n = sl->ref_cache[0][scan8[n]];
-        H264Picture *ref = &sl->ref_list[0][ref_n];
+        H264Ref *ref = &sl->ref_list[0][ref_n];
 
         // Error resilience puts the current picture in the ref list.
         // Don't try to wait on these as it will cause a deadlock.
         // Fields can wait on each other, though.
-        if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
+        if (ref->parent->tf.progress->data != h->cur_pic.tf.progress->data ||
             (ref->reference & 3) != h->picture_structure) {
-            my = get_lowest_part_list_y(sl, ref, n, height, y_offset, 0);
+            my = get_lowest_part_list_y(sl, n, height, y_offset, 0);
             if (refs[0][ref_n] < 0)
                 nrefs[0] += 1;
             refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
@@ -77,11 +76,11 @@ static inline void get_lowest_part_y(const H264Context *h, H264SliceContext *sl,
 
     if (list1) {
         int ref_n    = sl->ref_cache[1][scan8[n]];
-        H264Picture *ref = &sl->ref_list[1][ref_n];
+        H264Ref *ref = &sl->ref_list[1][ref_n];
 
-        if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
+        if (ref->parent->tf.progress->data != h->cur_pic.tf.progress->data ||
             (ref->reference & 3) != h->picture_structure) {
-            my = get_lowest_part_list_y(sl, ref, n, height, y_offset, 1);
+            my = get_lowest_part_list_y(sl, n, height, y_offset, 1);
             if (refs[1][ref_n] < 0)
                 nrefs[1] += 1;
             refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
@@ -98,7 +97,7 @@ static void await_references(const H264Context *h, H264SliceContext *sl)
 {
     const int mb_xy   = sl->mb_xy;
     const int mb_type = h->cur_pic.mb_type[mb_xy];
-    int refs[2][48];
+    int16_t refs[2][48];
     int nrefs[2] = { 0 };
     int ref, list;
 
@@ -120,7 +119,7 @@ static void await_references(const H264Context *h, H264SliceContext *sl)
     } else {
         int i;
 
-        assert(IS_8X8(mb_type));
+        av_assert2(IS_8X8(mb_type));
 
         for (i = 0; i < 4; i++) {
             const int sub_mb_type = sl->sub_mb_type[i];
@@ -152,7 +151,7 @@ static void await_references(const H264Context *h, H264SliceContext *sl)
                                   nrefs);
             } else {
                 int j;
-                assert(IS_SUB_4X4(sub_mb_type));
+                av_assert2(IS_SUB_4X4(sub_mb_type));
                 for (j = 0; j < 4; j++) {
                     int sub_y_offset = y_offset + 2 * (j & 2);
                     get_lowest_part_y(h, sl, refs, n + j, 4, sub_y_offset,
@@ -168,33 +167,33 @@ static void await_references(const H264Context *h, H264SliceContext *sl)
         for (ref = 0; ref < 48 && nrefs[list]; ref++) {
             int row = refs[list][ref];
             if (row >= 0) {
-                H264Picture *ref_pic  = &sl->ref_list[list][ref];
+                H264Ref *ref_pic  = &sl->ref_list[list][ref];
                 int ref_field         = ref_pic->reference - 1;
-                int ref_field_picture = ref_pic->field_picture;
+                int ref_field_picture = ref_pic->parent->field_picture;
                 int pic_height        = 16 * h->mb_height >> ref_field_picture;
 
                 row <<= MB_MBAFF(sl);
                 nrefs[list]--;
 
                 if (!FIELD_PICTURE(h) && ref_field_picture) { // frame referencing two fields
-                    ff_thread_await_progress(&ref_pic->tf,
+                    ff_thread_await_progress(&ref_pic->parent->tf,
                                              FFMIN((row >> 1) - !(row & 1),
                                                    pic_height - 1),
                                              1);
-                    ff_thread_await_progress(&ref_pic->tf,
+                    ff_thread_await_progress(&ref_pic->parent->tf,
                                              FFMIN((row >> 1), pic_height - 1),
                                              0);
                 } else if (FIELD_PICTURE(h) && !ref_field_picture) { // field referencing one field of a frame
-                    ff_thread_await_progress(&ref_pic->tf,
+                    ff_thread_await_progress(&ref_pic->parent->tf,
                                              FFMIN(row * 2 + ref_field,
                                                    pic_height - 1),
                                              0);
                 } else if (FIELD_PICTURE(h)) {
-                    ff_thread_await_progress(&ref_pic->tf,
+                    ff_thread_await_progress(&ref_pic->parent->tf,
                                              FFMIN(row, pic_height - 1),
                                              ref_field);
                 } else {
-                    ff_thread_await_progress(&ref_pic->tf,
+                    ff_thread_await_progress(&ref_pic->parent->tf,
                                              FFMIN(row, pic_height - 1),
                                              0);
                 }
@@ -203,7 +202,7 @@ static void await_references(const H264Context *h, H264SliceContext *sl)
 }
 
 static av_always_inline void mc_dir_part(const H264Context *h, H264SliceContext *sl,
-                                         H264Picture *pic,
+                                         H264Ref *pic,
                                          int n, int square, int height,
                                          int delta, int list,
                                          uint8_t *dest_y, uint8_t *dest_cb,
@@ -216,8 +215,8 @@ static av_always_inline void mc_dir_part(const H264Context *h, H264SliceContext
     const int mx      = sl->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
     int my            = sl->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
     const int luma_xy = (mx & 3) + ((my & 3) << 2);
-    ptrdiff_t offset  = ((mx >> 2) << pixel_shift) + (my >> 2) * sl->mb_linesize;
-    uint8_t *src_y    = pic->f.data[0] + offset;
+    ptrdiff_t offset  = (mx >> 2) * (1 << pixel_shift) + (my >> 2) * sl->mb_linesize;
+    uint8_t *src_y    = pic->data[0] + offset;
     uint8_t *src_cb, *src_cr;
     int extra_width  = 0;
     int extra_height = 0;
@@ -237,12 +236,12 @@ static av_always_inline void mc_dir_part(const H264Context *h, H264SliceContext
         full_my                <          0 - extra_height ||
         full_mx + 16 /*FIXME*/ > pic_width  + extra_width  ||
         full_my + 16 /*FIXME*/ > pic_height + extra_height) {
-        h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
+        h->vdsp.emulated_edge_mc(sl->edge_emu_buffer,
                                  src_y - (2 << pixel_shift) - 2 * sl->mb_linesize,
                                  sl->mb_linesize, sl->mb_linesize,
                                  16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
                                  full_my - 2, pic_width, pic_height);
-        src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * sl->mb_linesize;
+        src_y = sl->edge_emu_buffer + (2 << pixel_shift) + 2 * sl->mb_linesize;
         emu   = 1;
     }
 
@@ -254,29 +253,29 @@ static av_always_inline void mc_dir_part(const H264Context *h, H264SliceContext
         return;
 
     if (chroma_idc == 3 /* yuv444 */) {
-        src_cb = pic->f.data[1] + offset;
+        src_cb = pic->data[1] + offset;
         if (emu) {
-            h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
+            h->vdsp.emulated_edge_mc(sl->edge_emu_buffer,
                                      src_cb - (2 << pixel_shift) - 2 * sl->mb_linesize,
                                      sl->mb_linesize, sl->mb_linesize,
                                      16 + 5, 16 + 5 /*FIXME*/,
                                      full_mx - 2, full_my - 2,
                                      pic_width, pic_height);
-            src_cb = h->edge_emu_buffer + (2 << pixel_shift) + 2 * sl->mb_linesize;
+            src_cb = sl->edge_emu_buffer + (2 << pixel_shift) + 2 * sl->mb_linesize;
         }
         qpix_op[luma_xy](dest_cb, src_cb, sl->mb_linesize); // FIXME try variable height perhaps?
         if (!square)
             qpix_op[luma_xy](dest_cb + delta, src_cb + delta, sl->mb_linesize);
 
-        src_cr = pic->f.data[2] + offset;
+        src_cr = pic->data[2] + offset;
         if (emu) {
-            h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
+            h->vdsp.emulated_edge_mc(sl->edge_emu_buffer,
                                      src_cr - (2 << pixel_shift) - 2 * sl->mb_linesize,
                                      sl->mb_linesize, sl->mb_linesize,
                                      16 + 5, 16 + 5 /*FIXME*/,
                                      full_mx - 2, full_my - 2,
                                      pic_width, pic_height);
-            src_cr = h->edge_emu_buffer + (2 << pixel_shift) + 2 * sl->mb_linesize;
+            src_cr = sl->edge_emu_buffer + (2 << pixel_shift) + 2 * sl->mb_linesize;
         }
         qpix_op[luma_xy](dest_cr, src_cr, sl->mb_linesize); // FIXME try variable height perhaps?
         if (!square)
@@ -291,31 +290,31 @@ static av_always_inline void mc_dir_part(const H264Context *h, H264SliceContext
         emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
     }
 
-    src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
+    src_cb = pic->data[1] + ((mx >> 3) * (1 << pixel_shift)) +
              (my >> ysh) * sl->mb_uvlinesize;
-    src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
+    src_cr = pic->data[2] + ((mx >> 3) * (1 << pixel_shift)) +
              (my >> ysh) * sl->mb_uvlinesize;
 
     if (emu) {
-        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb,
+        h->vdsp.emulated_edge_mc(sl->edge_emu_buffer, src_cb,
                                  sl->mb_uvlinesize, sl->mb_uvlinesize,
                                  9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
                                  pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
-        src_cb = h->edge_emu_buffer;
+        src_cb = sl->edge_emu_buffer;
     }
     chroma_op(dest_cb, src_cb, sl->mb_uvlinesize,
               height >> (chroma_idc == 1 /* yuv420 */),
-              mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
+              mx & 7, ((unsigned)my << (chroma_idc == 2 /* yuv422 */)) & 7);
 
     if (emu) {
-        h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr,
+        h->vdsp.emulated_edge_mc(sl->edge_emu_buffer, src_cr,
                                  sl->mb_uvlinesize, sl->mb_uvlinesize,
                                  9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
                                  pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
-        src_cr = h->edge_emu_buffer;
+        src_cr = sl->edge_emu_buffer;
     }
     chroma_op(dest_cr, src_cr, sl->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
-              mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
+              mx & 7, ((unsigned)my << (chroma_idc == 2 /* yuv422 */)) & 7);
 }
 
 static av_always_inline void mc_part_std(const H264Context *h, H264SliceContext *sl,
@@ -349,7 +348,7 @@ static av_always_inline void mc_part_std(const H264Context *h, H264SliceContext
     y_offset += 8 * (sl->mb_y >> MB_FIELD(sl));
 
     if (list0) {
-        H264Picture *ref = &sl->ref_list[0][sl->ref_cache[0][scan8[n]]];
+        H264Ref *ref = &sl->ref_list[0][sl->ref_cache[0][scan8[n]]];
         mc_dir_part(h, sl, ref, n, square, height, delta, 0,
                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_op, chroma_op, pixel_shift, chroma_idc);
@@ -359,7 +358,7 @@ static av_always_inline void mc_part_std(const H264Context *h, H264SliceContext
     }
 
     if (list1) {
-        H264Picture *ref = &sl->ref_list[1][sl->ref_cache[1][scan8[n]]];
+        H264Ref *ref = &sl->ref_list[1][sl->ref_cache[1][scan8[n]]];
         mc_dir_part(h, sl, ref, n, square, height, delta, 1,
                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_op, chroma_op, pixel_shift, chroma_idc);
@@ -405,9 +404,9 @@ static av_always_inline void mc_part_weighted(const H264Context *h, H264SliceCon
     if (list0 && list1) {
         /* don't optimize for luma-only case, since B-frames usually
          * use implicit weights => chroma too. */
-        uint8_t *tmp_cb = h->bipred_scratchpad;
-        uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
-        uint8_t *tmp_y  = h->bipred_scratchpad + 16 * sl->mb_uvlinesize;
+        uint8_t *tmp_cb = sl->bipred_scratchpad;
+        uint8_t *tmp_cr = sl->bipred_scratchpad + (16 << pixel_shift);
+        uint8_t *tmp_y  = sl->bipred_scratchpad + 16 * sl->mb_uvlinesize;
         int refn0       = sl->ref_cache[0][scan8[n]];
         int refn1       = sl->ref_cache[1][scan8[n]];
 
@@ -425,10 +424,12 @@ static av_always_inline void mc_part_weighted(const H264Context *h, H264SliceCon
             int weight1 = 64 - weight0;
             luma_weight_avg(dest_y, tmp_y, sl->mb_linesize,
                             height, 5, weight0, weight1, 0);
-            chroma_weight_avg(dest_cb, tmp_cb, sl->mb_uvlinesize,
-                              chroma_height, 5, weight0, weight1, 0);
-            chroma_weight_avg(dest_cr, tmp_cr, sl->mb_uvlinesize,
-                              chroma_height, 5, weight0, weight1, 0);
+            if (!CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
+                chroma_weight_avg(dest_cb, tmp_cb, sl->mb_uvlinesize,
+                                  chroma_height, 5, weight0, weight1, 0);
+                chroma_weight_avg(dest_cr, tmp_cr, sl->mb_uvlinesize,
+                                  chroma_height, 5, weight0, weight1, 0);
+            }
         } else {
             luma_weight_avg(dest_y, tmp_y, sl->mb_linesize, height,
                             sl->luma_log2_weight_denom,
@@ -436,23 +437,25 @@ static av_always_inline void mc_part_weighted(const H264Context *h, H264SliceCon
                             sl->luma_weight[refn1][1][0],
                             sl->luma_weight[refn0][0][1] +
                             sl->luma_weight[refn1][1][1]);
-            chroma_weight_avg(dest_cb, tmp_cb, sl->mb_uvlinesize, chroma_height,
-                              sl->chroma_log2_weight_denom,
-                              sl->chroma_weight[refn0][0][0][0],
-                              sl->chroma_weight[refn1][1][0][0],
-                              sl->chroma_weight[refn0][0][0][1] +
-                              sl->chroma_weight[refn1][1][0][1]);
-            chroma_weight_avg(dest_cr, tmp_cr, sl->mb_uvlinesize, chroma_height,
-                              sl->chroma_log2_weight_denom,
-                              sl->chroma_weight[refn0][0][1][0],
-                              sl->chroma_weight[refn1][1][1][0],
-                              sl->chroma_weight[refn0][0][1][1] +
-                              sl->chroma_weight[refn1][1][1][1]);
+            if (!CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
+                chroma_weight_avg(dest_cb, tmp_cb, sl->mb_uvlinesize, chroma_height,
+                                  sl->chroma_log2_weight_denom,
+                                  sl->chroma_weight[refn0][0][0][0],
+                                  sl->chroma_weight[refn1][1][0][0],
+                                  sl->chroma_weight[refn0][0][0][1] +
+                                  sl->chroma_weight[refn1][1][0][1]);
+                chroma_weight_avg(dest_cr, tmp_cr, sl->mb_uvlinesize, chroma_height,
+                                  sl->chroma_log2_weight_denom,
+                                  sl->chroma_weight[refn0][0][1][0],
+                                  sl->chroma_weight[refn1][1][1][0],
+                                  sl->chroma_weight[refn0][0][1][1] +
+                                  sl->chroma_weight[refn1][1][1][1]);
+            }
         }
     } else {
         int list     = list1 ? 1 : 0;
         int refn     = sl->ref_cache[list][scan8[n]];
-        H264Picture *ref = &sl->ref_list[list][refn];
+        H264Ref *ref = &sl->ref_list[list][refn];
         mc_dir_part(h, sl, ref, n, square, height, delta, list,
                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put, chroma_put, pixel_shift, chroma_idc);
@@ -461,15 +464,17 @@ static av_always_inline void mc_part_weighted(const H264Context *h, H264SliceCon
                        sl->luma_log2_weight_denom,
                        sl->luma_weight[refn][list][0],
                        sl->luma_weight[refn][list][1]);
-        if (sl->use_weight_chroma) {
-            chroma_weight_op(dest_cb, sl->mb_uvlinesize, chroma_height,
-                             sl->chroma_log2_weight_denom,
-                             sl->chroma_weight[refn][list][0][0],
-                             sl->chroma_weight[refn][list][0][1]);
-            chroma_weight_op(dest_cr, sl->mb_uvlinesize, chroma_height,
-                             sl->chroma_log2_weight_denom,
-                             sl->chroma_weight[refn][list][1][0],
-                             sl->chroma_weight[refn][list][1][1]);
+        if (!CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
+            if (sl->use_weight_chroma) {
+                chroma_weight_op(dest_cb, sl->mb_uvlinesize, chroma_height,
+                                 sl->chroma_log2_weight_denom,
+                                 sl->chroma_weight[refn][list][0][0],
+                                 sl->chroma_weight[refn][list][0][1]);
+                chroma_weight_op(dest_cr, sl->mb_uvlinesize, chroma_height,
+                                 sl->chroma_log2_weight_denom,
+                                 sl->chroma_weight[refn][list][1][0],
+                                 sl->chroma_weight[refn][list][1][1]);
+            }
         }
     }
 }
@@ -484,18 +489,16 @@ static av_always_inline void prefetch_motion(const H264Context *h, H264SliceCont
     if (refn >= 0) {
         const int mx  = (sl->mv_cache[list][scan8[0]][0] >> 2) + 16 * sl->mb_x + 8;
         const int my  = (sl->mv_cache[list][scan8[0]][1] >> 2) + 16 * sl->mb_y;
-        uint8_t **src = sl->ref_list[list][refn].f.data;
-        int off       = (mx << pixel_shift) +
+        uint8_t **src = sl->ref_list[list][refn].data;
+        int off       =  mx * (1<< pixel_shift) +
                         (my + (sl->mb_x & 3) * 4) * sl->mb_linesize +
                         (64 << pixel_shift);
-        h->vdsp.prefetch(src[0] + off, h->linesize, 4);
+        h->vdsp.prefetch(src[0] + off, sl->linesize, 4);
         if (chroma_idc == 3 /* yuv444 */) {
-            h->vdsp.prefetch(src[1] + off, h->linesize, 4);
-            h->vdsp.prefetch(src[2] + off, h->linesize, 4);
+            h->vdsp.prefetch(src[1] + off, sl->linesize, 4);
+            h->vdsp.prefetch(src[2] + off, sl->linesize, 4);
         } else {
-            off = ((mx >> 1) << pixel_shift) +
-                  ((my >> 1) + (sl->mb_x & 7)) * h->uvlinesize +
-                  (64 << pixel_shift);
+            off= ((mx>>1)+64) * (1<<pixel_shift) + ((my>>1) + (sl->mb_x&7))*sl->uvlinesize;
             h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
         }
     }
@@ -535,8 +538,8 @@ static av_always_inline void xchg_mb_border(const H264Context *h, H264SliceConte
     src_cb -= uvlinesize + 1 + pixel_shift;
     src_cr -= uvlinesize + 1 + pixel_shift;
 
-    top_border_m1 = h->top_borders[top_idx][sl->mb_x - 1];
-    top_border    = h->top_borders[top_idx][sl->mb_x];
+    top_border_m1 = sl->top_borders[top_idx][sl->mb_x - 1];
+    top_border    = sl->top_borders[top_idx][sl->mb_x];
 
 #define XCHG(a, b, xchg)                        \
     if (pixel_shift) {                          \
@@ -559,13 +562,11 @@ static av_always_inline void xchg_mb_border(const H264Context *h, H264SliceConte
         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
         if (sl->mb_x + 1 < h->mb_width) {
-            XCHG(h->top_borders[top_idx][sl->mb_x + 1],
+            XCHG(sl->top_borders[top_idx][sl->mb_x + 1],
                  src_y + (17 << pixel_shift), 1);
         }
-    }
-    if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
-        if (chroma444) {
-            if (deblock_top) {
+        if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
+            if (chroma444) {
                 if (deblock_topleft) {
                     XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
                     XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
@@ -575,12 +576,10 @@ static av_always_inline void xchg_mb_border(const H264Context *h, H264SliceConte
                 XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
                 XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
                 if (sl->mb_x + 1 < h->mb_width) {
-                    XCHG(h->top_borders[top_idx][sl->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
-                    XCHG(h->top_borders[top_idx][sl->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
+                    XCHG(sl->top_borders[top_idx][sl->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
+                    XCHG(sl->top_borders[top_idx][sl->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
                 }
-            }
-        } else {
-            if (deblock_top) {
+            } else {
                 if (deblock_topleft) {
                     XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
                     XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
@@ -638,7 +637,12 @@ static av_always_inline void hl_decode_mb_predict_luma(const H264Context *h,
                 uint8_t *const ptr = dest_y + block_offset[i];
                 const int dir      = sl->intra4x4_pred_mode_cache[scan8[i]];
                 if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
-                    h->hpc.pred8x8l_add[dir](ptr, sl->mb + (i * 16 + p * 256 << pixel_shift), linesize);
+                    if (h->x264_build != -1) {
+                        h->hpc.pred8x8l_add[dir](ptr, sl->mb + (i * 16 + p * 256 << pixel_shift), linesize);
+                    } else
+                        h->hpc.pred8x8l_filter_add[dir](ptr, sl->mb + (i * 16 + p * 256 << pixel_shift),
+                                                        (sl-> topleft_samples_available << i) & 0x8000,
+                                                        (sl->topright_samples_available << i) & 0x4000, linesize);
                 } else {
                     const int nnz = sl->non_zero_count_cache[scan8[i + p * 16]];
                     h->hpc.pred8x8l[dir](ptr, (sl->topleft_samples_available << i) & 0x8000,
@@ -671,7 +675,7 @@ static av_always_inline void hl_decode_mb_predict_luma(const H264Context *h,
                     uint64_t tr_high;
                     if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
                         const int topright_avail = (sl->topright_samples_available << i) & 0x8000;
-                        assert(sl->mb_y || linesize <= block_offset[i]);
+                        av_assert2(sl->mb_y || linesize <= block_offset[i]);
                         if (!topright_avail) {
                             if (pixel_shift) {
                                 tr_high  = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;