avcodec/vc1: fix overlap filter for frame interlaced pictures
authorJerome Borsboom <jerome.borsboom@carpalis.nl>
Wed, 20 Jun 2018 11:11:07 +0000 (13:11 +0200)
committerCarl Eugen Hoyos <ceffmpeg@gmail.com>
Thu, 28 Jun 2018 23:18:44 +0000 (01:18 +0200)
The overlap filter is not correct for vertical edges in frame interlaced
I and P pictures. When filtering macroblocks with different FIELDTX values,
we have to match the lines at both sides of the vertical border. In addition,
we have to use the correct rounding values, depending on the line we are
filtering.

Signed-off-by: Jerome Borsboom <jerome.borsboom@carpalis.nl>
libavcodec/mips/vc1dsp_mmi.c
libavcodec/vc1_loopfilter.c
libavcodec/vc1dsp.c
libavcodec/vc1dsp.h

index 01e7f9f..ec2fdca 100644 (file)
@@ -1019,12 +1019,13 @@ void ff_vc1_h_overlap_mmi(uint8_t *src, int stride)
     }
 }
 
-void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right)
+void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags)
 {
     int i;
     int a, b, c, d;
     int d1, d2;
-    int rnd1 = 4, rnd2 = 3;
+    int rnd1 = flags & 2 ? 3 : 4;
+    int rnd2 = 7 - rnd1;
     for (i = 0; i < 8; i++) {
         a  = left[6];
         b  = left[7];
@@ -1038,10 +1039,12 @@ void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right)
         right[0] = ((c << 3) + d2 + rnd1) >> 3;
         right[1] = ((d << 3) + d1 + rnd2) >> 3;
 
-        right += 8;
-        left  += 8;
-        rnd2   = 7 - rnd2;
-        rnd1   = 7 - rnd1;
+        right += right_stride;
+        left  += left_stride;
+        if (flags & 1) {
+            rnd2   = 7 - rnd2;
+            rnd1   = 7 - rnd1;
+        }
     }
 }
 
index 5d5630d..d43fa5b 100644 (file)
 #include "vc1dsp.h"
 
 static av_always_inline void vc1_h_overlap_filter(VC1Context *v, int16_t (*left_block)[64],
-                                                  int16_t (*right_block)[64], int block_num)
+                                                  int16_t (*right_block)[64], int left_fieldtx,
+                                                  int right_fieldtx, int block_num)
 {
-    if (block_num > 3)
-        v->vc1dsp.vc1_h_s_overlap(left_block[block_num], right_block[block_num]);
-    else if (block_num & 1)
-        v->vc1dsp.vc1_h_s_overlap(right_block[block_num - 1], right_block[block_num]);
-    else
-        v->vc1dsp.vc1_h_s_overlap(left_block[block_num + 1], right_block[block_num]);
+    switch (block_num) {
+    case 0:
+        v->vc1dsp.vc1_h_s_overlap(left_block[2],
+                                  right_block[0],
+                                  left_fieldtx ^ right_fieldtx ? 16 - 8 * left_fieldtx : 8,
+                                  left_fieldtx ^ right_fieldtx ? 16 - 8 * right_fieldtx : 8,
+                                  left_fieldtx || right_fieldtx ? 0 : 1);
+        break;
+
+    case 1:
+        v->vc1dsp.vc1_h_s_overlap(right_block[0],
+                                  right_block[2],
+                                  8,
+                                  8,
+                                  right_fieldtx ? 0 : 1);
+        break;
+
+    case 2:
+        v->vc1dsp.vc1_h_s_overlap(!left_fieldtx && right_fieldtx ? left_block[2] + 8 : left_block[3],
+                                  left_fieldtx && !right_fieldtx ? right_block[0] + 8 : right_block[1],
+                                  left_fieldtx ^ right_fieldtx ? 16 - 8 * left_fieldtx : 8,
+                                  left_fieldtx ^ right_fieldtx ? 16 - 8 * right_fieldtx : 8,
+                                  left_fieldtx || right_fieldtx ? 2 : 1);
+        break;
+
+    case 3:
+        v->vc1dsp.vc1_h_s_overlap(right_block[1],
+                                  right_block[3],
+                                  8,
+                                  8,
+                                  right_fieldtx ? 2 : 1);
+        break;
+
+    case 4:
+    case 5:
+        v->vc1dsp.vc1_h_s_overlap(left_block[block_num], right_block[block_num], 8, 8, 1);
+        break;
+    }
 }
 
 static av_always_inline void vc1_v_overlap_filter(VC1Context *v, int16_t (*top_block)[64],
                                                   int16_t (*bottom_block)[64], int block_num)
 {
-    if (block_num > 3)
+    switch (block_num) {
+    case 0:
+        v->vc1dsp.vc1_v_s_overlap(top_block[1], bottom_block[0]);
+        break;
+
+    case 1:
+        v->vc1dsp.vc1_v_s_overlap(top_block[3], bottom_block[2]);
+        break;
+
+    case 2:
+        v->vc1dsp.vc1_v_s_overlap(bottom_block[0], bottom_block[1]);
+        break;
+
+    case 3:
+        v->vc1dsp.vc1_v_s_overlap(bottom_block[2], bottom_block[3]);
+        break;
+
+    case 4:
+    case 5:
         v->vc1dsp.vc1_v_s_overlap(top_block[block_num], bottom_block[block_num]);
-    else if (block_num & 2)
-        v->vc1dsp.vc1_v_s_overlap(bottom_block[block_num - 2], bottom_block[block_num]);
-    else
-        v->vc1dsp.vc1_v_s_overlap(top_block[block_num + 2], bottom_block[block_num]);
+        break;
+    }
 }
 
 void ff_vc1_i_overlap_filter(VC1Context *v)
@@ -82,7 +131,11 @@ void ff_vc1_i_overlap_filter(VC1Context *v)
                            (v->condover == CONDOVER_ALL ||
                             (v->over_flags_plane[mb_pos] &&
                              ((i & 5) == 1 || v->over_flags_plane[mb_pos - 1])))))
-            vc1_h_overlap_filter(v, s->mb_x ? left_blk : cur_blk, cur_blk, i);
+            vc1_h_overlap_filter(v,
+                                 s->mb_x ? left_blk : cur_blk, cur_blk,
+                                 v->fcm == ILACE_FRAME && s->mb_x && v->fieldtx_plane[mb_pos - 1],
+                                 v->fcm == ILACE_FRAME && v->fieldtx_plane[mb_pos],
+                                 i);
     }
 
     if (v->fcm != ILACE_FRAME)
@@ -110,6 +163,7 @@ void ff_vc1_p_overlap_filter(VC1Context *v)
     MpegEncContext *s = &v->s;
     int16_t (*topleft_blk)[64], (*top_blk)[64], (*left_blk)[64], (*cur_blk)[64];
     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;
     int i;
 
     topleft_blk = v->block[v->topleft_blk_idx];
@@ -122,7 +176,11 @@ void ff_vc1_p_overlap_filter(VC1Context *v)
             continue;
 
         if (v->mb_type[0][s->block_index[i]] && v->mb_type[0][s->block_index[i] - 1])
-            vc1_h_overlap_filter(v, s->mb_x ? left_blk : cur_blk, cur_blk, i);
+            vc1_h_overlap_filter(v,
+                                 s->mb_x ? left_blk : cur_blk, cur_blk,
+                                 v->fcm == ILACE_FRAME && s->mb_x && v->fieldtx_plane[mb_pos - 1],
+                                 v->fcm == ILACE_FRAME && v->fieldtx_plane[mb_pos],
+                                 i);
     }
 
     if (v->fcm != ILACE_FRAME)
index 9239a4a..778b811 100644 (file)
@@ -107,12 +107,13 @@ static void vc1_v_s_overlap_c(int16_t *top, int16_t *bottom)
     }
 }
 
-static void vc1_h_s_overlap_c(int16_t *left, int16_t *right)
+static void vc1_h_s_overlap_c(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags)
 {
     int i;
     int a, b, c, d;
     int d1, d2;
-    int rnd1 = 4, rnd2 = 3;
+    int rnd1 = flags & 2 ? 3 : 4;
+    int rnd2 = 7 - rnd1;
     for (i = 0; i < 8; i++) {
         a  = left[6];
         b  = left[7];
@@ -126,10 +127,12 @@ static void vc1_h_s_overlap_c(int16_t *left, int16_t *right)
         right[0] = ((c << 3) + d2 + rnd1) >> 3;
         right[1] = ((d << 3) + d1 + rnd2) >> 3;
 
-        right += 8;
-        left  += 8;
-        rnd2   = 7 - rnd2;
-        rnd1   = 7 - rnd1;
+        right += right_stride;
+        left  += left_stride;
+        if (flags & 1) {
+            rnd2   = 7 - rnd2;
+            rnd1   = 7 - rnd1;
+        }
     }
 }
 
index 16b3528..75db62b 100644 (file)
@@ -45,7 +45,7 @@ typedef struct VC1DSPContext {
     void (*vc1_v_overlap)(uint8_t *src, int stride);
     void (*vc1_h_overlap)(uint8_t *src, int stride);
     void (*vc1_v_s_overlap)(int16_t *top,  int16_t *bottom);
-    void (*vc1_h_s_overlap)(int16_t *left, int16_t *right);
+    void (*vc1_h_s_overlap)(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags);
     void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq);
     void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq);
     void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq);