Revert "VC1: merge idct8x8, coeff adjustments and put_pixels."

This reverts commit f8bed30d8b176fa030f6737765338bb4a2bcabc9. The reason
for this is that the overlap filter, which runs after IDCT, should run
on unclamped values, and thus IDCT and put_pixels() cannot be merged if
we want to attempt to be bitexact.

1 files changed, 25 insertions, 33 deletions

diff --git a/libavcodec/vc1dec.c b/libavcodec/vc1dec.c
index 1d891ad418..70c7d268bc 100644
--- a/libavcodec/vc1dec.c
+++ b/libavcodec/vc1dec.c
@@ -2016,7 +2016,8 @@ static int vc1_decode_p_block(VC1Context *v, DCTELEM block[64], int n, int mquan
             if(i==1)
                 v->vc1dsp.vc1_inv_trans_8x8_dc(dst, linesize, block);
             else{
-                v->vc1dsp.vc1_inv_trans_8x8_add(dst, linesize, block);
+                v->vc1dsp.vc1_inv_trans_8x8(block);
+                s->dsp.add_pixels_clamped(block, dst, linesize);
             }
         }
         break;
@@ -2258,7 +2259,7 @@ static int vc1_decode_p_mb(VC1Context *v)
 {
     MpegEncContext *s = &v->s;
     GetBitContext *gb = &s->gb;
-    int i;
+    int i, j;
     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
     int cbp; /* cbp decoding stuff */
     int mqdiff, mquant; /* MB quantization */
@@ -2288,8 +2289,6 @@ static int vc1_decode_p_mb(VC1Context *v)
     {
         if (!skipped)
         {
-            vc1_idct_func idct8x8_fn;
-
             GET_MVDATA(dmv_x, dmv_y);
 
             if (s->mb_intra) {
@@ -2324,7 +2323,6 @@ static int vc1_decode_p_mb(VC1Context *v)
                                 VC1_TTMB_VLC_BITS, 2);
             if(!s->mb_intra) vc1_mc_1mv(v, 0);
             dst_idx = 0;
-            idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
             for (i=0; i<6; i++)
             {
                 s->dc_val[0][s->block_index[i]] = 0;
@@ -2342,9 +2340,9 @@ static int vc1_decode_p_mb(VC1Context *v)
 
                     vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
                     if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
-                    idct8x8_fn(s->dest[dst_idx] + off,
-                               i & 4 ? s->uvlinesize : s->linesize,
-                               s->block[i]);
+                    v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
+                    if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
+                    s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
                     if(v->pq >= 9 && v->overlap) {
                         if(v->c_avail)
                             v->vc1dsp.vc1_h_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
@@ -2380,7 +2378,6 @@ static int vc1_decode_p_mb(VC1Context *v)
         {
             int intra_count = 0, coded_inter = 0;
             int is_intra[6], is_coded[6];
-            vc1_idct_func idct8x8_fn;
             /* Get CBPCY */
             cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
             for (i=0; i<6; i++)
@@ -2431,7 +2428,6 @@ static int vc1_decode_p_mb(VC1Context *v)
             }
             if (!v->ttmbf && coded_inter)
                 ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
-            idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
             for (i=0; i<6; i++)
             {
                 dst_idx += i >> 2;
@@ -2447,9 +2443,9 @@ static int vc1_decode_p_mb(VC1Context *v)
 
                     vc1_decode_intra_block(v, s->block[i], i, is_coded[i], mquant, (i&4)?v->codingset2:v->codingset);
                     if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
-                    idct8x8_fn(s->dest[dst_idx] + off,
-                               (i&4)?s->uvlinesize:s->linesize,
-                               s->block[i]);
+                    v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
+                    if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
+                    s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize);
                     if(v->pq >= 9 && v->overlap) {
                         if(v->c_avail)
                             v->vc1dsp.vc1_h_overlap(s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
@@ -2497,7 +2493,7 @@ static void vc1_decode_b_mb(VC1Context *v)
 {
     MpegEncContext *s = &v->s;
     GetBitContext *gb = &s->gb;
-    int i;
+    int i, j;
     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
     int cbp = 0; /* cbp decoding stuff */
     int mqdiff, mquant; /* MB quantization */
@@ -2510,7 +2506,6 @@ static void vc1_decode_b_mb(VC1Context *v)
     int skipped, direct;
     int dmv_x[2], dmv_y[2];
     int bmvtype = BMV_TYPE_BACKWARD;
-    vc1_idct_func idct8x8_fn;
 
     mquant = v->pq; /* Loosy initialization */
     s->mb_intra = 0;
@@ -2608,7 +2603,6 @@ static void vc1_decode_b_mb(VC1Context *v)
         }
     }
     dst_idx = 0;
-    idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
     for (i=0; i<6; i++)
     {
         s->dc_val[0][s->block_index[i]] = 0;
@@ -2626,9 +2620,9 @@ static void vc1_decode_b_mb(VC1Context *v)
 
             vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
             if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
-            idct8x8_fn(s->dest[dst_idx] + off,
-                       i & 4 ? s->uvlinesize : s->linesize,
-                       s->block[i]);
+            v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
+            if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
+            s->dsp.put_signed_pixels_clamped(s->block[i], s->dest[dst_idx] + off, i & 4 ? s->uvlinesize : s->linesize);
         } else if(val) {
             vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block, s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize, (i&4) && (s->flags & CODEC_FLAG_GRAY), NULL);
             if(!v->ttmbf && ttmb < 8) ttmb = -1;
@@ -2641,12 +2635,11 @@ static void vc1_decode_b_mb(VC1Context *v)
  */
 static void vc1_decode_i_blocks(VC1Context *v)
 {
-    int k;
+    int k, j;
     MpegEncContext *s = &v->s;
     int cbp, val;
     uint8_t *coded_val;
     int mb_pos;
-    vc1_idct_func idct8x8_fn;
 
     /* select codingmode used for VLC tables selection */
     switch(v->y_ac_table_index){
@@ -2681,10 +2674,6 @@ static void vc1_decode_i_blocks(VC1Context *v)
     s->mb_x = s->mb_y = 0;
     s->mb_intra = 1;
     s->first_slice_line = 1;
-    if(v->pq >= 9 && v->overlap) {
-        idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[!!v->rangeredfrm];
-    } else
-        idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put[!!v->rangeredfrm];
     for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
         s->mb_x = 0;
         ff_init_block_index(s);
@@ -2721,9 +2710,14 @@ static void vc1_decode_i_blocks(VC1Context *v)
                 vc1_decode_i_block(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2);
 
                 if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue;
-                idct8x8_fn(dst[k],
-                           k & 4 ? s->uvlinesize : s->linesize,
-                           s->block[k]);
+                v->vc1dsp.vc1_inv_trans_8x8(s->block[k]);
+                if(v->pq >= 9 && v->overlap) {
+                    if (v->rangeredfrm) for(j = 0; j < 64; j++) s->block[k][j] <<= 1;
+                    s->dsp.put_signed_pixels_clamped(s->block[k], dst[k], k & 4 ? s->uvlinesize : s->linesize);
+                } else {
+                    if (v->rangeredfrm) for(j = 0; j < 64; j++) s->block[k][j] = (s->block[k][j] - 64) << 1;
+                    s->dsp.put_pixels_clamped(s->block[k], dst[k], k & 4 ? s->uvlinesize : s->linesize);
+                }
             }
 
             if(v->pq >= 9 && v->overlap) {
@@ -2781,7 +2775,6 @@ static void vc1_decode_i_blocks_adv(VC1Context *v, int mby_start, int mby_end)
     int mqdiff;
     int overlap;
     GetBitContext *gb = &s->gb;
-    vc1_idct_func idct8x8_fn;
 
     /* select codingmode used for VLC tables selection */
     switch(v->y_ac_table_index){
@@ -2819,7 +2812,6 @@ static void vc1_decode_i_blocks_adv(VC1Context *v, int mby_start, int mby_end)
         memset(&s->coded_block[s->block_index[0]-s->b8_stride], 0,
                s->b8_stride * sizeof(*s->coded_block));
     }
-    idct8x8_fn = v->vc1dsp.vc1_inv_trans_8x8_put_signed[0];
     for(; s->mb_y < mby_end; s->mb_y++) {
         s->mb_x = 0;
         ff_init_block_index(s);
@@ -2876,9 +2868,9 @@ static void vc1_decode_i_blocks_adv(VC1Context *v, int mby_start, int mby_end)
                 vc1_decode_i_block_adv(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2, mquant);
 
                 if (k > 3 && (s->flags & CODEC_FLAG_GRAY)) continue;
-                idct8x8_fn(dst[k],
-                           k & 4 ? s->uvlinesize : s->linesize,
-                           s->block[k]);
+                v->vc1dsp.vc1_inv_trans_8x8(s->block[k]);
+                s->dsp.put_signed_pixels_clamped(s->block[k], dst[k],
+                                                 k & 4 ? s->uvlinesize : s->linesize);
             }
 
             if(overlap) {