]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/h264_cavlc.c
qsvdec: only access hwaccel_context is the pixel format is QSV
[ffmpeg] / libavcodec / h264_cavlc.c
index a46bc318930546a83f811c13ec30bb42252440ba..4fa2de01844aed07f11845486220208b6cd5a171 100644 (file)
@@ -21,7 +21,7 @@
 
 /**
  * @file
- * H.264 / AVC / MPEG4 part10 cavlc bitstream decoding.
+ * H.264 / AVC / MPEG-4 part10 cavlc bitstream decoding.
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
@@ -29,9 +29,9 @@
 
 #include "internal.h"
 #include "avcodec.h"
-#include "h264.h"
-#include "h264data.h" // FIXME FIXME FIXME
+#include "h264dec.h"
 #include "h264_mvpred.h"
+#include "h264data.h"
 #include "golomb.h"
 #include "mpegutils.h"
 
@@ -283,7 +283,7 @@ static int8_t cavlc_level_tab[7][1<<LEVEL_TAB_BITS][2];
  * Get the predicted number of non-zero coefficients.
  * @param n block index
  */
-static inline int pred_non_zero_count(H264Context *h, H264SliceContext *sl, int n)
+static inline int pred_non_zero_count(const H264Context *h, H264SliceContext *sl, int n)
 {
     const int index8= scan8[n];
     const int left = sl->non_zero_count_cache[index8 - 1];
@@ -292,7 +292,7 @@ static inline int pred_non_zero_count(H264Context *h, H264SliceContext *sl, int
 
     if(i<64) i= (i+1)>>1;
 
-    tprintf(h->avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
+    ff_tlog(h->avctx, "pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
 
     return i&31;
 }
@@ -412,9 +412,6 @@ av_cold void ff_h264_decode_init_vlc(void){
     }
 }
 
-/**
- *
- */
 static inline int get_level_prefix(GetBitContext *gb){
     unsigned int buf;
     int log;
@@ -424,10 +421,6 @@ static inline int get_level_prefix(GetBitContext *gb){
     buf=GET_CACHE(re, gb);
 
     log= 32 - av_log2(buf);
-#ifdef TRACE
-    print_bin(buf>>(32-log), log);
-    av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
-#endif
 
     LAST_SKIP_BITS(re, gb, log);
     CLOSE_READER(re, gb);
@@ -442,7 +435,7 @@ static inline int get_level_prefix(GetBitContext *gb){
  * @param max_coeff number of coefficients in the block
  * @return <0 if an error occurred
  */
-static int decode_residual(H264Context *h, H264SliceContext *sl,
+static int decode_residual(const H264Context *h, H264SliceContext *sl,
                            GetBitContext *gb, int16_t *block, int n,
                            const uint8_t *scantable, const uint32_t *qmul,
                            int max_coeff)
@@ -482,7 +475,7 @@ static int decode_residual(H264Context *h, H264SliceContext *sl,
     }
 
     trailing_ones= coeff_token&3;
-    tprintf(h->avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
+    ff_tlog(h->avctx, "trailing:%d, total:%d\n", trailing_ones, total_coeff);
     assert(total_coeff<=16);
 
     i = show_bits(gb, 3);
@@ -628,7 +621,12 @@ static int decode_residual(H264Context *h, H264SliceContext *sl,
     return 0;
 }
 
-static av_always_inline int decode_luma_residual(H264Context *h, H264SliceContext *sl, GetBitContext *gb, const uint8_t *scan, const uint8_t *scan8x8, int pixel_shift, int mb_type, int cbp, int p){
+static av_always_inline
+int decode_luma_residual(const H264Context *h, H264SliceContext *sl,
+                         GetBitContext *gb, const uint8_t *scan,
+                         const uint8_t *scan8x8, int pixel_shift,
+                         int mb_type, int cbp, int p)
+{
     int i4x4, i8x8;
     int qscale = p == 0 ? sl->qscale : sl->chroma_qp[p - 1];
     if(IS_INTRA16x16(mb_type)){
@@ -647,7 +645,7 @@ static av_always_inline int decode_luma_residual(H264Context *h, H264SliceContex
                 for(i4x4=0; i4x4<4; i4x4++){
                     const int index= i4x4 + 4*i8x8 + p*16;
                     if( decode_residual(h, sl, gb, sl->mb + (16*index << pixel_shift),
-                        index, scan + 1, h->dequant4_coeff[p][qscale], 15) < 0 ){
+                        index, scan + 1, h->ps.pps->dequant4_coeff[p][qscale], 15) < 0 ){
                         return -1;
                     }
                 }
@@ -669,7 +667,7 @@ static av_always_inline int decode_luma_residual(H264Context *h, H264SliceContex
                     for(i4x4=0; i4x4<4; i4x4++){
                         const int index= i4x4 + 4*i8x8 + p*16;
                         if( decode_residual(h, sl, gb, buf, index, scan8x8+16*i4x4,
-                                            h->dequant8_coeff[cqm][qscale], 16) < 0 )
+                                            h->ps.pps->dequant8_coeff[cqm][qscale], 16) < 0 )
                             return -1;
                     }
                     nnz = &sl->non_zero_count_cache[scan8[4 * i8x8 + p * 16]];
@@ -679,7 +677,7 @@ static av_always_inline int decode_luma_residual(H264Context *h, H264SliceContex
                     for(i4x4=0; i4x4<4; i4x4++){
                         const int index= i4x4 + 4*i8x8 + p*16;
                         if( decode_residual(h, sl, gb, sl->mb + (16*index << pixel_shift), index,
-                                            scan, h->dequant4_coeff[cqm][qscale], 16) < 0 ){
+                                            scan, h->ps.pps->dequant4_coeff[cqm][qscale], 16) < 0 ){
                             return -1;
                         }
                         new_cbp |= sl->non_zero_count_cache[scan8[index]] << i8x8;
@@ -694,18 +692,17 @@ static av_always_inline int decode_luma_residual(H264Context *h, H264SliceContex
     }
 }
 
-int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
+int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl)
 {
     int mb_xy;
     int partition_count;
     unsigned int mb_type, cbp;
-    int dct8x8_allowed= h->pps.transform_8x8_mode;
-    int decode_chroma = h->sps.chroma_format_idc == 1 || h->sps.chroma_format_idc == 2;
+    int dct8x8_allowed= h->ps.pps->transform_8x8_mode;
+    int decode_chroma = h->ps.sps->chroma_format_idc == 1 || h->ps.sps->chroma_format_idc == 2;
     const int pixel_shift = h->pixel_shift;
 
     mb_xy = sl->mb_xy = sl->mb_x + sl->mb_y*h->mb_stride;
 
-    tprintf(h->avctx, "pic:%d mb:%d/%d\n", h->frame_num, sl->mb_x, sl->mb_y);
     cbp = 0; /* avoid warning. FIXME: find a solution without slowing
                 down the code */
     if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
@@ -715,7 +712,7 @@ int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
         if (sl->mb_skip_run--) {
             if (FRAME_MBAFF(h) && (sl->mb_y & 1) == 0) {
                 if (sl->mb_skip_run == 0)
-                    h->mb_mbaff = sl->mb_field_decoding_flag = get_bits1(&sl->gb);
+                    sl->mb_mbaff = sl->mb_field_decoding_flag = get_bits1(&sl->gb);
             }
             decode_mb_skip(h, sl);
             return 0;
@@ -723,7 +720,7 @@ int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
     }
     if (FRAME_MBAFF(h)) {
         if ((sl->mb_y & 1) == 0)
-            h->mb_mbaff = sl->mb_field_decoding_flag = get_bits1(&sl->gb);
+            sl->mb_mbaff = sl->mb_field_decoding_flag = get_bits1(&sl->gb);
     }
 
     sl->prev_mb_skipped = 0;
@@ -731,16 +728,16 @@ int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl)
     mb_type= get_ue_golomb(&sl->gb);
     if (sl->slice_type_nos == AV_PICTURE_TYPE_B) {
         if(mb_type < 23){
-            partition_countb_mb_type_info[mb_type].partition_count;
-            mb_type=         b_mb_type_info[mb_type].type;
+            partition_count = ff_h264_b_mb_type_info[mb_type].partition_count;
+            mb_type         = ff_h264_b_mb_type_info[mb_type].type;
         }else{
             mb_type -= 23;
             goto decode_intra_mb;
         }
     } else if (sl->slice_type_nos == AV_PICTURE_TYPE_P) {
         if(mb_type < 5){
-            partition_countp_mb_type_info[mb_type].partition_count;
-            mb_type=         p_mb_type_info[mb_type].type;
+            partition_count = ff_h264_p_mb_type_info[mb_type].partition_count;
+            mb_type         = ff_h264_p_mb_type_info[mb_type].type;
         }else{
             mb_type -= 5;
             goto decode_intra_mb;
@@ -755,9 +752,9 @@ decode_intra_mb:
             return -1;
         }
         partition_count=0;
-        cbpi_mb_type_info[mb_type].cbp;
-        sl->intra16x16_pred_mode = i_mb_type_info[mb_type].pred_mode;
-        mb_typei_mb_type_info[mb_type].type;
+        cbp                      = ff_h264_i_mb_type_info[mb_type].cbp;
+        sl->intra16x16_pred_mode = ff_h264_i_mb_type_info[mb_type].pred_mode;
+        mb_type                  = ff_h264_i_mb_type_info[mb_type].type;
     }
 
     if (MB_FIELD(sl))
@@ -766,8 +763,8 @@ decode_intra_mb:
     h->slice_table[mb_xy] = sl->slice_num;
 
     if(IS_INTRA_PCM(mb_type)){
-        const int mb_size = ff_h264_mb_sizes[h->sps.chroma_format_idc] *
-                            h->sps.bit_depth_luma;
+        const int mb_size = ff_h264_mb_sizes[h->ps.sps->chroma_format_idc] *
+                            h->ps.sps->bit_depth_luma;
 
         // We assume these blocks are very rare so we do not optimize it.
         sl->intra_pcm_ptr = align_get_bits(&sl->gb);
@@ -816,15 +813,18 @@ decode_intra_mb:
                     sl->intra4x4_pred_mode_cache[scan8[i]] = mode;
             }
             write_back_intra_pred_mode(h, sl);
-            if (ff_h264_check_intra4x4_pred_mode(h, sl) < 0)
+            if (ff_h264_check_intra4x4_pred_mode(sl->intra4x4_pred_mode_cache, h->avctx,
+                                                 sl->top_samples_available, sl->left_samples_available) < 0)
                 return -1;
         }else{
-            sl->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h, sl, sl->intra16x16_pred_mode, 0);
+            sl->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h->avctx, sl->top_samples_available,
+                                                                     sl->left_samples_available, sl->intra16x16_pred_mode, 0);
             if (sl->intra16x16_pred_mode < 0)
                 return -1;
         }
         if(decode_chroma){
-            pred_mode= ff_h264_check_intra_pred_mode(h, sl, get_ue_golomb_31(&sl->gb), 1);
+            pred_mode= ff_h264_check_intra_pred_mode(h->avctx, sl->top_samples_available,
+                                                     sl->left_samples_available, get_ue_golomb_31(&sl->gb), 1);
             if(pred_mode < 0)
                 return -1;
             sl->chroma_pred_mode = pred_mode;
@@ -841,8 +841,8 @@ decode_intra_mb:
                     av_log(h->avctx, AV_LOG_ERROR, "B sub_mb_type %u out of range at %d %d\n", sl->sub_mb_type[i], sl->mb_x, sl->mb_y);
                     return -1;
                 }
-                sub_partition_count[i]= b_sub_mb_type_info[ sl->sub_mb_type[i] ].partition_count;
-                sl->sub_mb_type[i]=      b_sub_mb_type_info[ sl->sub_mb_type[i] ].type;
+                sub_partition_count[i] = ff_h264_b_sub_mb_type_info[sl->sub_mb_type[i]].partition_count;
+                sl->sub_mb_type[i]     = ff_h264_b_sub_mb_type_info[sl->sub_mb_type[i]].type;
             }
             if( IS_DIRECT(sl->sub_mb_type[0]|sl->sub_mb_type[1]|sl->sub_mb_type[2]|sl->sub_mb_type[3])) {
                 ff_h264_pred_direct_motion(h, sl, &mb_type);
@@ -859,13 +859,13 @@ decode_intra_mb:
                     av_log(h->avctx, AV_LOG_ERROR, "P sub_mb_type %u out of range at %d %d\n", sl->sub_mb_type[i], sl->mb_x, sl->mb_y);
                     return -1;
                 }
-                sub_partition_count[i]= p_sub_mb_type_info[ sl->sub_mb_type[i] ].partition_count;
-                sl->sub_mb_type[i]=      p_sub_mb_type_info[ sl->sub_mb_type[i] ].type;
+                sub_partition_count[i] = ff_h264_p_sub_mb_type_info[sl->sub_mb_type[i]].partition_count;
+                sl->sub_mb_type[i]     = ff_h264_p_sub_mb_type_info[sl->sub_mb_type[i]].type;
             }
         }
 
         for (list = 0; list < sl->list_count; list++) {
-            int ref_count = IS_REF0(mb_type) ? 1 : sl->ref_count[list] << MB_MBAFF(h);
+            int ref_count = IS_REF0(mb_type) ? 1 : sl->ref_count[list] << MB_MBAFF(sl);
             for(i=0; i<4; i++){
                 if(IS_DIRECT(sl->sub_mb_type[i])) continue;
                 if(IS_DIR(sl->sub_mb_type[i], 0, list)){
@@ -911,7 +911,7 @@ decode_intra_mb:
                         pred_motion(h, sl, index, block_width, list, sl->ref_cache[list][ scan8[index] ], &mx, &my);
                         mx += get_se_golomb(&sl->gb);
                         my += get_se_golomb(&sl->gb);
-                        tprintf(h->avctx, "final mv:%d %d\n", mx, my);
+                        ff_tlog(h->avctx, "final mv:%d %d\n", mx, my);
 
                         if(IS_SUB_8X8(sub_mb_type)){
                             mv_cache[ 1 ][0]=
@@ -937,7 +937,7 @@ decode_intra_mb:
         }
     }else if(IS_DIRECT(mb_type)){
         ff_h264_pred_direct_motion(h, sl, &mb_type);
-        dct8x8_allowed &= h->sps.direct_8x8_inference_flag;
+        dct8x8_allowed &= h->ps.sps->direct_8x8_inference_flag;
     }else{
         int list, mx, my, i;
          //FIXME we should set ref_idx_l? to 0 if we use that later ...
@@ -945,7 +945,7 @@ decode_intra_mb:
             for (list = 0; list < sl->list_count; list++) {
                     unsigned int val;
                     if(IS_DIR(mb_type, 0, list)){
-                        int rc = sl->ref_count[list] << MB_MBAFF(h);
+                        int rc = sl->ref_count[list] << MB_MBAFF(sl);
                         if (rc == 1) {
                             val= 0;
                         } else if (rc == 2) {
@@ -965,7 +965,7 @@ decode_intra_mb:
                     pred_motion(h, sl, 0, 4, list, sl->ref_cache[list][ scan8[0] ], &mx, &my);
                     mx += get_se_golomb(&sl->gb);
                     my += get_se_golomb(&sl->gb);
-                    tprintf(h->avctx, "final mv:%d %d\n", mx, my);
+                    ff_tlog(h->avctx, "final mv:%d %d\n", mx, my);
 
                     fill_rectangle(sl->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
                 }
@@ -976,7 +976,7 @@ decode_intra_mb:
                     for(i=0; i<2; i++){
                         unsigned int val;
                         if(IS_DIR(mb_type, i, list)){
-                            int rc = sl->ref_count[list] << MB_MBAFF(h);
+                            int rc = sl->ref_count[list] << MB_MBAFF(sl);
                             if (rc == 1) {
                                 val= 0;
                             } else if (rc == 2) {
@@ -1000,7 +1000,7 @@ decode_intra_mb:
                         pred_16x8_motion(h, sl, 8*i, list, sl->ref_cache[list][scan8[0] + 16*i], &mx, &my);
                         mx += get_se_golomb(&sl->gb);
                         my += get_se_golomb(&sl->gb);
-                        tprintf(h->avctx, "final mv:%d %d\n", mx, my);
+                        ff_tlog(h->avctx, "final mv:%d %d\n", mx, my);
 
                         val= pack16to32(mx,my);
                     }else
@@ -1014,7 +1014,7 @@ decode_intra_mb:
                     for(i=0; i<2; i++){
                         unsigned int val;
                         if(IS_DIR(mb_type, i, list)){ //FIXME optimize
-                            int rc = sl->ref_count[list] << MB_MBAFF(h);
+                            int rc = sl->ref_count[list] << MB_MBAFF(sl);
                             if (rc == 1) {
                                 val= 0;
                             } else if (rc == 2) {
@@ -1038,7 +1038,7 @@ decode_intra_mb:
                         pred_8x16_motion(h, sl, i*4, list, sl->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
                         mx += get_se_golomb(&sl->gb);
                         my += get_se_golomb(&sl->gb);
-                        tprintf(h->avctx, "final mv:%d %d\n", mx, my);
+                        ff_tlog(h->avctx, "final mv:%d %d\n", mx, my);
 
                         val= pack16to32(mx,my);
                     }else
@@ -1060,8 +1060,10 @@ decode_intra_mb:
                 av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, sl->mb_x, sl->mb_y);
                 return -1;
             }
-            if(IS_INTRA4x4(mb_type)) cbp= golomb_to_intra4x4_cbp[cbp];
-            else                     cbp= golomb_to_inter_cbp   [cbp];
+            if (IS_INTRA4x4(mb_type))
+                cbp = ff_h264_golomb_to_intra4x4_cbp[cbp];
+            else
+                cbp = ff_h264_golomb_to_inter_cbp[cbp];
         }else{
             if(cbp > 15){
                 av_log(h->avctx, AV_LOG_ERROR, "cbp too large (%u) at %d %d\n", cbp, sl->mb_x, sl->mb_y);
@@ -1085,7 +1087,7 @@ decode_intra_mb:
         int ret;
         GetBitContext *gb = &sl->gb;
         const uint8_t *scan, *scan8x8;
-        const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
+        const int max_qp = 51 + 6 * (h->ps.sps->bit_depth_luma - 8);
 
         if(IS_INTERLACED(mb_type)){
             scan8x8 = sl->qscale ? h->field_scan8x8_cavlc : h->field_scan8x8_cavlc_q0;
@@ -1108,8 +1110,8 @@ decode_intra_mb:
             }
         }
 
-        sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
-        sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
+        sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
+        sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
 
         if ((ret = decode_luma_residual(h, sl, gb, scan, scan8x8, pixel_shift, mb_type, cbp, 0)) < 0 ) {
             return -1;
@@ -1126,7 +1128,7 @@ decode_intra_mb:
             if(cbp&0x30){
                 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
                     if (decode_residual(h, sl, gb, sl->mb + ((256 + 16*16*chroma_idx) << pixel_shift),
-                                        CHROMA_DC_BLOCK_INDEX+chroma_idx, chroma422_dc_scan,
+                                        CHROMA_DC_BLOCK_INDEX + chroma_idx, ff_h264_chroma422_dc_scan,
                                         NULL, 8) < 0) {
                         return -1;
                     }
@@ -1134,7 +1136,7 @@ decode_intra_mb:
 
             if(cbp&0x20){
                 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
-                    const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][sl->chroma_qp[chroma_idx]];
+                    const uint32_t *qmul = h->ps.pps->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][sl->chroma_qp[chroma_idx]];
                     int16_t *mb = sl->mb + (16*(16 + 16*chroma_idx) << pixel_shift);
                     for (i8x8 = 0; i8x8 < 2; i8x8++) {
                         for (i4x4 = 0; i4x4 < 4; i4x4++) {
@@ -1152,14 +1154,15 @@ decode_intra_mb:
         } else /* yuv420 */ {
             if(cbp&0x30){
                 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
-                    if( decode_residual(h, sl, gb, sl->mb + ((256 + 16*16*chroma_idx) << pixel_shift), CHROMA_DC_BLOCK_INDEX+chroma_idx, chroma_dc_scan, NULL, 4) < 0){
+                    if (decode_residual(h, sl, gb, sl->mb + ((256 + 16 * 16 * chroma_idx) << pixel_shift),
+                                        CHROMA_DC_BLOCK_INDEX + chroma_idx, ff_h264_chroma_dc_scan, NULL, 4) < 0) {
                         return -1;
                     }
             }
 
             if(cbp&0x20){
                 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
-                    const uint32_t *qmul = h->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][sl->chroma_qp[chroma_idx]];
+                    const uint32_t *qmul = h->ps.pps->dequant4_coeff[chroma_idx+1+(IS_INTRA( mb_type ) ? 0:3)][sl->chroma_qp[chroma_idx]];
                     for(i4x4=0; i4x4<4; i4x4++){
                         const int index= 16 + 16*chroma_idx + i4x4;
                         if( decode_residual(h, sl, gb, sl->mb + (16*index << pixel_shift), index, scan + 1, qmul, 15) < 0){