vc1dec: redesign the intensity compensation

The existing implementation had little to do with VC1.
This could be implemented by adjusting the reference frames
ithemselfs but that would make frame multi-threading difficult.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

3 files changed, 112 insertions, 37 deletions

diff --git a/libavcodec/vc1.c b/libavcodec/vc1.c
index 07f5fb5a60..208fef33a5 100644
--- a/libavcodec/vc1.c
+++ b/libavcodec/vc1.c
@@ -576,8 +576,27 @@ int ff_vc1_decode_entry_point(AVCodecContext *avctx, VC1Context *v, GetBitContex
     return 0;
 }
 
+static void rotate_luts(VC1Context *v)
+{
+#define ROTATE(DEF, L, N, C, A) do {\
+        if (v->s.pict_type == AV_PICTURE_TYPE_BI || v->s.pict_type == AV_PICTURE_TYPE_B) {\
+            C = A;\
+        } else {\
+            DEF;\
+            memcpy(&tmp, &L  , sizeof(tmp));\
+            memcpy(&L  , &N  , sizeof(tmp));\
+            memcpy(&N  , &tmp, sizeof(tmp));\
+            C = N;\
+        }\
+    }while(0)
+
+        ROTATE(int tmp            , v->last_use_ic, v->next_use_ic, v->curr_use_ic, v->aux_use_ic);
+        ROTATE(uint8_t tmp[2][256], v->last_luty , v->next_luty , v->curr_luty , v->aux_luty);
+        ROTATE(uint8_t tmp[2][256], v->last_lutuv, v->next_lutuv, v->curr_lutuv, v->aux_lutuv);
+}
+
 /* fill lookup tables for intensity compensation */
-#define INIT_LUT(lumscale, lumshift, luty, lutuv)   do {\
+#define INIT_LUT(lumscale, lumshift, luty, lutuv, chain)   do {\
     int scale, shift, i;                            \
     if (!lumscale) {                                \
         scale = -64;                                \
@@ -592,11 +611,14 @@ int ff_vc1_decode_entry_point(AVCodecContext *avctx, VC1Context *v, GetBitContex
             shift = lumshift << 6;                  \
     }                                               \
     for (i = 0; i < 256; i++) {                     \
-        luty[i]  = av_clip_uint8((scale * i + shift + 32) >> 6);           \
-        lutuv[i] = av_clip_uint8((scale * (i - 128) + 128*64 + 32) >> 6);  \
+        int iy = chain ? luty[i] : i;               \
+        int iu = chain ? lutuv[i] : i;              \
+        luty[i]  = av_clip_uint8((scale * iy + shift + 32) >> 6);           \
+        lutuv[i] = av_clip_uint8((scale * (iu - 128) + 128*64 + 32) >> 6);  \
     } \
     }while(0)
 
+
 int ff_vc1_parse_frame_header(VC1Context *v, GetBitContext* gb)
 {
     int pqindex, lowquant, status;
@@ -685,8 +707,12 @@ int ff_vc1_parse_frame_header(VC1Context *v, GetBitContext* gb)
             (v->s.pict_type == AV_PICTURE_TYPE_P) ? 'P' : ((v->s.pict_type == AV_PICTURE_TYPE_I) ? 'I' : 'B'),
             pqindex, v->pq, v->halfpq, v->rangeredfrm);
 
-    if (v->s.pict_type == AV_PICTURE_TYPE_I || v->s.pict_type == AV_PICTURE_TYPE_P)
-        v->use_ic = 0;
+    if(v->first_pic_header_flag) {
+        rotate_luts(v);
+        INIT_LUT(32, 0 , v->curr_luty[0] , v->curr_lutuv[0] , 0);
+        INIT_LUT(32, 0 , v->curr_luty[1] , v->curr_lutuv[1] , 0);
+        v->curr_use_ic = 0;
+    }
 
     switch (v->s.pict_type) {
     case AV_PICTURE_TYPE_P:
@@ -700,9 +726,10 @@ int ff_vc1_parse_frame_header(VC1Context *v, GetBitContext* gb)
             v->mv_mode2 = ff_vc1_mv_pmode_table2[lowquant][get_unary(gb, 1, 3)];
             v->lumscale = get_bits(gb, 6);
             v->lumshift = get_bits(gb, 6);
-            v->use_ic   = 1;
+            v->last_use_ic   = 1;
             /* fill lookup tables for intensity compensation */
-            INIT_LUT(v->lumscale, v->lumshift, v->luty, v->lutuv);
+            INIT_LUT(v->lumscale, v->lumshift , v->last_luty[0] , v->last_lutuv[0] , 1);
+            INIT_LUT(v->lumscale, v->lumshift , v->last_luty[1] , v->last_lutuv[1] , 1);
         }
         v->qs_last = v->s.quarter_sample;
         if (v->mv_mode == MV_PMODE_1MV_HPEL || v->mv_mode == MV_PMODE_1MV_HPEL_BILIN)
@@ -957,12 +984,16 @@ int ff_vc1_parse_frame_header_adv(VC1Context *v, GetBitContext* gb)
     if (v->postprocflag)
         v->postproc = get_bits(gb, 2);
 
-    if (v->s.pict_type == AV_PICTURE_TYPE_I || v->s.pict_type == AV_PICTURE_TYPE_P)
-        v->use_ic = 0;
-
     if (v->parse_only)
         return 0;
 
+    if(v->first_pic_header_flag) {
+        rotate_luts(v);
+        INIT_LUT(32, 0 , v->curr_luty[0] , v->curr_lutuv[0] , 0);
+        INIT_LUT(32, 0 , v->curr_luty[1] , v->curr_lutuv[1] , 0);
+        v->curr_use_ic = 0;
+    }
+
     switch (v->s.pict_type) {
     case AV_PICTURE_TYPE_I:
     case AV_PICTURE_TYPE_BI:
@@ -1013,7 +1044,9 @@ int ff_vc1_parse_frame_header_adv(VC1Context *v, GetBitContext* gb)
                 if (v->intcomp) {
                     v->lumscale = get_bits(gb, 6);
                     v->lumshift = get_bits(gb, 6);
-                    INIT_LUT(v->lumscale, v->lumshift, v->luty, v->lutuv);
+                    INIT_LUT(v->lumscale, v->lumshift, v->last_luty[0], v->last_lutuv[0], 1);
+                    INIT_LUT(v->lumscale, v->lumshift, v->last_luty[1], v->last_lutuv[1], 1);
+                    v->last_use_ic = 1;
                 }
                 status = bitplane_decoding(v->s.mbskip_table, &v->skip_is_raw, v);
                 av_log(v->s.avctx, AV_LOG_DEBUG, "SKIPMB plane encoding: "
@@ -1056,17 +1089,38 @@ int ff_vc1_parse_frame_header_adv(VC1Context *v, GetBitContext* gb)
                 int mvmode2;
                 mvmode2 = get_unary(gb, 1, 3);
                 v->mv_mode2 = ff_vc1_mv_pmode_table2[lowquant][mvmode2];
-                if (v->field_mode)
-                    v->intcompfield = decode210(gb);
-                v->lumscale = get_bits(gb, 6);
-                v->lumshift = get_bits(gb, 6);
-                INIT_LUT(v->lumscale, v->lumshift, v->luty, v->lutuv);
-                if ((v->field_mode) && !v->intcompfield) {
+                if (v->field_mode) {
+                    v->intcompfield = decode210(gb)^3;
+                } else
+                    v->intcompfield = 3;
+
+                v->lumscale2 = v->lumscale = 32;
+                v->lumshift2 = v->lumshift =  0;
+                if (v->intcompfield & 1) {
+                    v->lumscale = get_bits(gb, 6);
+                    v->lumshift = get_bits(gb, 6);
+                }
+                if ((v->intcompfield & 2) && v->field_mode) {
                     v->lumscale2 = get_bits(gb, 6);
                     v->lumshift2 = get_bits(gb, 6);
-                    INIT_LUT(v->lumscale2, v->lumshift2, v->luty2, v->lutuv2);
+                } else if(!v->field_mode) {
+                    v->lumscale2 = v->lumscale;
+                    v->lumshift2 = v->lumshift;
+                }
+                if (v->field_mode && v->second_field) {
+                    if (v->cur_field_type) {
+                        INIT_LUT(v->lumscale , v->lumshift , v->curr_luty[v->cur_field_type^1], v->curr_lutuv[v->cur_field_type^1], 0);
+                        INIT_LUT(v->lumscale2, v->lumshift2, v->last_luty[v->cur_field_type  ], v->last_lutuv[v->cur_field_type  ], 1);
+                    } else {
+                        INIT_LUT(v->lumscale2, v->lumshift2, v->curr_luty[v->cur_field_type^1], v->curr_lutuv[v->cur_field_type^1], 0);
+                        INIT_LUT(v->lumscale , v->lumshift , v->last_luty[v->cur_field_type  ], v->last_lutuv[v->cur_field_type  ], 1);
+                    }
+                    v->curr_use_ic = 1;
+                } else {
+                    INIT_LUT(v->lumscale , v->lumshift , v->last_luty[0], v->last_lutuv[0], 1);
+                    INIT_LUT(v->lumscale2, v->lumshift2, v->last_luty[1], v->last_lutuv[1], 1);
                 }
-                v->use_ic = 1;
+                v->last_use_ic = 1;
             }
             v->qs_last = v->s.quarter_sample;
             if (v->mv_mode == MV_PMODE_1MV_HPEL || v->mv_mode == MV_PMODE_1MV_HPEL_BILIN)
diff --git a/libavcodec/vc1.h b/libavcodec/vc1.h
index 596d4d35e5..9e2ec687ba 100644
--- a/libavcodec/vc1.h
+++ b/libavcodec/vc1.h
@@ -296,8 +296,11 @@ typedef struct VC1Context{
     int dmb_is_raw;                 ///< direct mb plane is raw
     int fmb_is_raw;                 ///< forward mb plane is raw
     int skip_is_raw;                ///< skip mb plane is not coded
-    uint8_t luty[256], lutuv[256];  ///< lookup tables used for intensity compensation
-    int use_ic;                     ///< use intensity compensation in B-frames
+    uint8_t last_luty[2][256], last_lutuv[2][256];  ///< lookup tables used for intensity compensation
+    uint8_t  aux_luty[2][256],  aux_lutuv[2][256];  ///< lookup tables used for intensity compensation
+    uint8_t next_luty[2][256], next_lutuv[2][256];  ///< lookup tables used for intensity compensation
+    uint8_t (*curr_luty)[256]  ,(*curr_lutuv)[256];
+    int last_use_ic, curr_use_ic, next_use_ic, aux_use_ic;
     int rnd;                        ///< rounding control
 
     /** Frame decoding info for S/M profiles only */
@@ -340,7 +343,6 @@ typedef struct VC1Context{
     int intcomp;
     uint8_t lumscale2;  ///< for interlaced field P picture
     uint8_t lumshift2;
-    uint8_t luty2[256], lutuv2[256]; // lookup tables used for intensity compensation
     VLC* mbmode_vlc;
     VLC* imv_vlc;
     VLC* twomvbp_vlc;
diff --git a/libavcodec/vc1dec.c b/libavcodec/vc1dec.c
index d79d484c7e..050bd99405 100644
--- a/libavcodec/vc1dec.c
+++ b/libavcodec/vc1dec.c
@@ -351,6 +351,7 @@ static void vc1_mc_1mv(VC1Context *v, int dir)
     int off, off_uv;
     int v_edge_pos = s->v_edge_pos >> v->field_mode;
     int i;
+    const uint8_t *luty, *lutuv;
 
     if ((!v->field_mode ||
          (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
@@ -389,15 +390,21 @@ static void vc1_mc_1mv(VC1Context *v, int dir)
             srcY = s->current_picture.f.data[0];
             srcU = s->current_picture.f.data[1];
             srcV = s->current_picture.f.data[2];
+            luty = v->curr_luty [v->ref_field_type[dir]];
+            lutuv= v->curr_lutuv[v->ref_field_type[dir]];
         } else {
             srcY = s->last_picture.f.data[0];
             srcU = s->last_picture.f.data[1];
             srcV = s->last_picture.f.data[2];
+            luty = v->last_luty [v->ref_field_type[dir]];
+            lutuv= v->last_lutuv[v->ref_field_type[dir]];
         }
     } else {
         srcY = s->next_picture.f.data[0];
         srcU = s->next_picture.f.data[1];
         srcV = s->next_picture.f.data[2];
+        luty = v->next_luty [v->ref_field_type[dir]];
+        lutuv= v->next_lutuv[v->ref_field_type[dir]];
     }
 
     if(!srcY)
@@ -484,15 +491,15 @@ static void vc1_mc_1mv(VC1Context *v, int dir)
             src = srcY;
             for (j = 0; j < 17 + s->mspel * 2; j++) {
                 for (i = 0; i < 17 + s->mspel * 2; i++)
-                    src[i] = v->luty[src[i]];
+                    src[i] = luty[src[i]];
                 src += s->linesize;
             }
             src  = srcU;
             src2 = srcV;
             for (j = 0; j < 9; j++) {
                 for (i = 0; i < 9; i++) {
-                    src[i]  = v->lutuv[src[i]];
-                    src2[i] = v->lutuv[src2[i]];
+                    src[i]  = lutuv[src[i]];
+                    src2[i] = lutuv[src2[i]];
                 }
                 src  += s->uvlinesize;
                 src2 += s->uvlinesize;
@@ -552,6 +559,7 @@ static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
     int off;
     int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
     int v_edge_pos = s->v_edge_pos >> v->field_mode;
+    const uint8_t *luty;
 
     if ((!v->field_mode ||
          (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
@@ -564,10 +572,15 @@ static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
     if (!dir) {
         if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
             srcY = s->current_picture.f.data[0];
-        } else
+            luty = v->curr_luty[v->ref_field_type[dir]];
+        } else {
             srcY = s->last_picture.f.data[0];
-    } else
+            luty = v->last_luty[v->ref_field_type[dir]];
+        }
+    } else {
         srcY = s->next_picture.f.data[0];
+        luty = v->next_luty[v->ref_field_type[dir]];
+    }
 
     if(!srcY)
         return;
@@ -699,7 +712,7 @@ static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg)
             src = srcY;
             for (j = 0; j < 9 + s->mspel * 2; j++) {
                 for (i = 0; i < 9 + s->mspel * 2; i++)
-                    src[i] = v->luty[src[i]];
+                    src[i] = luty[src[i]];
                 src += s->linesize << fieldmv;
             }
         }
@@ -787,6 +800,7 @@ static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
     int valid_count;
     int chroma_ref_type = v->cur_field_type, off = 0;
     int v_edge_pos = s->v_edge_pos >> v->field_mode;
+    const uint8_t *lutuv;
 
     if (!v->field_mode && !v->s.last_picture.f.data[0])
         return;
@@ -852,13 +866,16 @@ static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
         if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) {
             srcU = s->current_picture.f.data[1];
             srcV = s->current_picture.f.data[2];
+            lutuv= v->curr_lutuv[chroma_ref_type];
         } else {
             srcU = s->last_picture.f.data[1];
             srcV = s->last_picture.f.data[2];
+            lutuv= v->last_lutuv[chroma_ref_type];
         }
     } else {
         srcU = s->next_picture.f.data[1];
         srcV = s->next_picture.f.data[2];
+        lutuv= v->next_lutuv[chroma_ref_type];
     }
 
     if(!srcU)
@@ -913,8 +930,8 @@ static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
             src2 = srcV;
             for (j = 0; j < 9; j++) {
                 for (i = 0; i < 9; i++) {
-                    src[i]  = v->lutuv[src[i]];
-                    src2[i] = v->lutuv[src2[i]];
+                    src[i]  = lutuv[src[i]];
+                    src2[i] = lutuv[src2[i]];
                 }
                 src  += s->uvlinesize;
                 src2 += s->uvlinesize;
@@ -998,13 +1015,14 @@ static void vc1_mc_4mv_chroma4(VC1Context *v)
             if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
                 int i, j;
                 uint8_t *src, *src2;
+                const uint8_t *lutuv = v->last_lutuv[v->ref_field_type[0]];
 
                 src  = srcU;
                 src2 = srcV;
                 for (j = 0; j < 5; j++) {
                     for (i = 0; i < 5; i++) {
-                        src[i]  = v->lutuv[src[i]];
-                        src2[i] = v->lutuv[src2[i]];
+                        src[i]  = lutuv[src[i]];
+                        src2[i] = lutuv[src2[i]];
                     }
                     src  += s->uvlinesize << 1;
                     src2 += s->uvlinesize << 1;
@@ -1998,29 +2016,30 @@ static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2],
                             int direct, int mode)
 {
-    if (v->use_ic) {
+    int use_ic = v->next_use_ic || v->curr_use_ic || v->last_use_ic;
+    if (use_ic) {
         v->mv_mode2 = v->mv_mode;
         v->mv_mode  = MV_PMODE_INTENSITY_COMP;
     }
     if (direct) {
         vc1_mc_1mv(v, 0);
         vc1_interp_mc(v);
-        if (v->use_ic)
+        if (use_ic)
             v->mv_mode = v->mv_mode2;
         return;
     }
     if (mode == BMV_TYPE_INTERPOLATED) {
         vc1_mc_1mv(v, 0);
         vc1_interp_mc(v);
-        if (v->use_ic)
+        if (use_ic)
             v->mv_mode = v->mv_mode2;
         return;
     }
 
-    if (v->use_ic && (mode == BMV_TYPE_BACKWARD))
+    if (use_ic && (mode == BMV_TYPE_BACKWARD))
         v->mv_mode = v->mv_mode2;
     vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
-    if (v->use_ic)
+    if (use_ic)
         v->mv_mode = v->mv_mode2;
 }