mpegvideo: support encoding with chroma intra tables that differ from luma.

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

libavcodec/bfin/mpegvideo_bfin.c

@@ -61,8 +61,13 @@ static int dct_quantize_bfin (MpegEncContext *s,
         dc = block[0] = (block[0] + (q >> 1)) / q;
         start_i = 1;
         last_non_zero = 0;
-        bias = s->q_intra_matrix16[qscale][1];
-        qmat = s->q_intra_matrix16[qscale][0];
+        if(n<4){
+            bias = s->q_intra_matrix16[qscale][1];
+            qmat = s->q_intra_matrix16[qscale][0];
+        }else{
+            bias = s->q_chroma_intra_matrix16[qscale][1];
+            qmat = s->q_chroma_intra_matrix16[qscale][0];
+        }
 
     } else {
         start_i = 0;

libavcodec/dnxhdenc.c

@@ -77,7 +77,7 @@ static int dnxhd_10bit_dct_quantize(MpegEncContext *ctx, DCTELEM *block,
                                     int n, int qscale, int *overflow)
 {
     const uint8_t *scantable= ctx->intra_scantable.scantable;
-    const int *qmat = ctx->q_intra_matrix[qscale];
+    const int *qmat = n<4 ? ctx->q_intra_matrix[qscale] : ctx->q_chroma_intra_matrix[qscale];
     int last_non_zero = 0;
     int i;
 
@@ -208,6 +208,11 @@ static int dnxhd_init_qmat(DNXHDEncContext *ctx, int lbias, int cbias)
         }
     }
 
+    ctx->m.q_chroma_intra_matrix16 = ctx->qmatrix_c16;
+    ctx->m.q_chroma_intra_matrix   = ctx->qmatrix_c;
+    ctx->m.q_intra_matrix16        = ctx->qmatrix_l16;
+    ctx->m.q_intra_matrix          = ctx->qmatrix_l;
+
     return 0;
  fail:
     return -1;
@@ -498,12 +503,8 @@ static av_always_inline void dnxhd_get_blocks(DNXHDEncContext *ctx, int mb_x, in
 static av_always_inline int dnxhd_switch_matrix(DNXHDEncContext *ctx, int i)
 {
     if (i&2) {
-        ctx->m.q_intra_matrix16 = ctx->qmatrix_c16;
-        ctx->m.q_intra_matrix   = ctx->qmatrix_c;
         return 1 + (i&1);
     } else {
-        ctx->m.q_intra_matrix16 = ctx->qmatrix_l16;
-        ctx->m.q_intra_matrix   = ctx->qmatrix_l;
         return 0;
     }
 }
@@ -535,7 +536,7 @@ static int dnxhd_calc_bits_thread(AVCodecContext *avctx, void *arg, int jobnr, i
             int n = dnxhd_switch_matrix(ctx, i);
 
             memcpy(block, src_block, 64*sizeof(*block));
-            last_index = ctx->m.dct_quantize(&ctx->m, block, i, qscale, &overflow);
+            last_index = ctx->m.dct_quantize(&ctx->m, block, 4&(2*i), qscale, &overflow);
             ac_bits += dnxhd_calc_ac_bits(ctx, block, last_index);
 
             diff = block[0] - ctx->m.last_dc[n];
@@ -582,7 +583,7 @@ static int dnxhd_encode_thread(AVCodecContext *avctx, void *arg, int jobnr, int
             DCTELEM *block = ctx->blocks[i];
             int last_index, overflow;
             int n = dnxhd_switch_matrix(ctx, i);
-            last_index = ctx->m.dct_quantize(&ctx->m, block, i, qscale, &overflow);
+            last_index = ctx->m.dct_quantize(&ctx->m, block, 4&(2*i), qscale, &overflow);
             //START_TIMER;
             dnxhd_encode_block(ctx, block, last_index, n);
             //STOP_TIMER("encode_block");

libavcodec/mpegvideo.c

@@ -687,8 +687,10 @@ av_cold int MPV_common_init(MpegEncContext *s)
         FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
 
         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix  , 64*32   * sizeof(int), fail)
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->q_chroma_intra_matrix  , 64*32   * sizeof(int), fail)
         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix  , 64*32   * sizeof(int), fail)
         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t), fail)
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->q_chroma_intra_matrix16, 64*32*2 * sizeof(uint16_t), fail)
         FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t), fail)
         FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
         FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
@@ -846,6 +848,10 @@ void MPV_common_end(MpegEncContext *s)
     av_freep(&s->error_status_table);
     av_freep(&s->mb_index2xy);
     av_freep(&s->lambda_table);
+    if(s->q_chroma_intra_matrix   != s->q_intra_matrix  ) av_freep(&s->q_chroma_intra_matrix);
+    if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
+    s->q_chroma_intra_matrix=   NULL;
+    s->q_chroma_intra_matrix16= NULL;
     av_freep(&s->q_intra_matrix);
     av_freep(&s->q_inter_matrix);
     av_freep(&s->q_intra_matrix16);

libavcodec/mpegvideo.h

@@ -448,9 +448,11 @@ typedef struct MpegEncContext {
 
     /** precomputed matrix (combine qscale and DCT renorm) */
     int (*q_intra_matrix)[64];
+    int (*q_chroma_intra_matrix)[64];
     int (*q_inter_matrix)[64];
     /** identical to the above but for MMX & these are not permutated, second 64 entries are bias*/
     uint16_t (*q_intra_matrix16)[2][64];
+    uint16_t (*q_chroma_intra_matrix16)[2][64];
     uint16_t (*q_inter_matrix16)[2][64];
 
     /* noise reduction */

libavcodec/mpegvideo_enc.c

@@ -2825,6 +2825,13 @@ static int encode_picture(MpegEncContext *s, int picture_number)
         update_qscale(s);
     }
 
+    if(s->codec_id != CODEC_ID_AMV){
+        if(s->q_chroma_intra_matrix   != s->q_intra_matrix  ) av_freep(&s->q_chroma_intra_matrix);
+        if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
+        s->q_chroma_intra_matrix   = s->q_intra_matrix;
+        s->q_chroma_intra_matrix16 = s->q_intra_matrix16;
+    }
+
     s->mb_intra=0; //for the rate distortion & bit compare functions
     for(i=1; i<context_count; i++){
         ff_update_duplicate_context(s->thread_context[i], s);
@@ -3091,7 +3098,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
         block[0] = (block[0] + (q >> 1)) / q;
         start_i = 1;
         last_non_zero = 0;
-        qmat = s->q_intra_matrix[qscale];
+        qmat = n < 4 ? s->q_intra_matrix[qscale] : s->q_chroma_intra_matrix[qscale];
         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
             bias= 1<<(QMAT_SHIFT-1);
         length     = s->intra_ac_vlc_length;
@@ -3762,7 +3769,7 @@ int dct_quantize_c(MpegEncContext *s,
         block[0] = (block[0] + (q >> 1)) / q;
         start_i = 1;
         last_non_zero = 0;
-        qmat = s->q_intra_matrix[qscale];
+        qmat = n < 4 ? s->q_intra_matrix[qscale] : s->q_chroma_intra_matrix[qscale];
         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
     } else {
         start_i = 0;

libavcodec/ppc/mpegvideo_altivec.c

@@ -267,8 +267,13 @@ static int dct_quantize_altivec(MpegEncContext* s,
             baseVector = vec_cts(vec_splat(row0, 0), 0);
             vec_ste(baseVector, 0, &oldBaseValue);
 
-            qmat = (vector signed int*)s->q_intra_matrix[qscale];
-            biasAddr = &(s->intra_quant_bias);
+            if(n<4){
+                qmat = (vector signed int*)s->q_intra_matrix[qscale];
+                biasAddr = &(s->intra_quant_bias);
+            }else{
+                qmat = (vector signed int*)s->q_chroma_intra_matrix[qscale];
+                biasAddr = &(s->intra_quant_bias);
+            }
         } else {
             qmat = (vector signed int*)s->q_inter_matrix[qscale];
             biasAddr = &(s->inter_quant_bias);

libavcodec/x86/mpegvideo_mmx_template.c

@@ -110,10 +110,15 @@ static int RENAME(dct_quantize)(MpegEncContext *s,
 
     if (s->mb_intra) {
         int dummy;
-        if (n < 4)
+        if (n < 4){
             q = s->y_dc_scale;
-        else
+            bias = s->q_intra_matrix16[qscale][1];
+            qmat = s->q_intra_matrix16[qscale][0];
+        }else{
             q = s->c_dc_scale;
+            bias = s->q_chroma_intra_matrix16[qscale][1];
+            qmat = s->q_chroma_intra_matrix16[qscale][0];
+        }
         /* note: block[0] is assumed to be positive */
         if (!s->h263_aic) {
         __asm__ volatile (
@@ -128,8 +133,6 @@ static int RENAME(dct_quantize)(MpegEncContext *s,
         block[0]=0; //avoid fake overflow
 //        temp_block[0] = (block[0] + (q >> 1)) / q;
         last_non_zero_p1 = 1;
-        bias = s->q_intra_matrix16[qscale][1];
-        qmat = s->q_intra_matrix16[qscale][0];
     } else {
         last_non_zero_p1 = 0;
         bias = s->q_inter_matrix16[qscale][1];