]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/dxva2_h264.c
h264: move the quantizers into the per-slice context
[ffmpeg] / libavcodec / dxva2_h264.c
index 22c76df601a2773659cbd7a056f186b0e55720d3..dbc5087619450e2ac07cf75fbcaaf2ea0b3d7fb5 100644 (file)
@@ -3,26 +3,27 @@
  *
  * copyright (c) 2009 Laurent Aimar
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 #include "dxva2_internal.h"
 #include "h264.h"
 #include "h264data.h"
+#include "mpegutils.h"
 
 struct dxva2_picture_context {
     DXVA_PicParams_H264   pp;
@@ -44,30 +45,29 @@ static void fill_picture_entry(DXVA_PicEntry_H264 *pic,
 static void fill_picture_parameters(struct dxva_context *ctx, const H264Context *h,
                                     DXVA_PicParams_H264 *pp)
 {
-    const MpegEncContext *s = &h->s;
-    const Picture *current_picture = s->current_picture_ptr;
-    int i;
+    const H264Picture *current_picture = h->cur_pic_ptr;
+    int i, j;
 
     memset(pp, 0, sizeof(*pp));
     /* Configure current picture */
     fill_picture_entry(&pp->CurrPic,
-                       ff_dxva2_get_surface_index(ctx, current_picture),
-                       s->picture_structure == PICT_BOTTOM_FIELD);
+                       ff_dxva2_get_surface_index(ctx, &current_picture->f),
+                       h->picture_structure == PICT_BOTTOM_FIELD);
     /* Configure the set of references */
     pp->UsedForReferenceFlags  = 0;
     pp->NonExistingFrameFlags  = 0;
-    for (i = 0; i < FF_ARRAY_ELEMS(pp->RefFrameList); i++) {
-        if (i < h->short_ref_count + h->long_ref_count) {
-            const Picture *r;
-            if (i < h->short_ref_count) {
-                r = h->short_ref[i];
-                assert(!r->long_ref);
-            } else {
-                r = h->long_ref[i - h->short_ref_count];
-                assert(r->long_ref);
-            }
+    for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->RefFrameList); i++) {
+        const H264Picture *r;
+        if (j < h->short_ref_count) {
+            r = h->short_ref[j++];
+        } else {
+            r = NULL;
+            while (!r && j < h->short_ref_count + 16)
+                r = h->long_ref[j++ - h->short_ref_count];
+        }
+        if (r) {
             fill_picture_entry(&pp->RefFrameList[i],
-                               ff_dxva2_get_surface_index(ctx, r),
+                               ff_dxva2_get_surface_index(ctx, &r->f),
                                r->long_ref != 0);
 
             if ((r->reference & PICT_TOP_FIELD) && r->field_poc[0] != INT_MAX)
@@ -88,14 +88,15 @@ static void fill_picture_parameters(struct dxva_context *ctx, const H264Context
         }
     }
 
-    pp->wFrameWidthInMbsMinus1        = s->mb_width  - 1;
-    pp->wFrameHeightInMbsMinus1       = s->mb_height - 1;
+    pp->wFrameWidthInMbsMinus1        = h->mb_width  - 1;
+    pp->wFrameHeightInMbsMinus1       = h->mb_height - 1;
     pp->num_ref_frames                = h->sps.ref_frame_count;
 
-    pp->wBitFields                    = ((s->picture_structure != PICT_FRAME) <<  0) |
-                                        (h->sps.mb_aff                        <<  1) |
+    pp->wBitFields                    = ((h->picture_structure != PICT_FRAME) <<  0) |
+                                        ((h->sps.mb_aff &&
+                                        (h->picture_structure == PICT_FRAME)) <<  1) |
                                         (h->sps.residual_color_transform_flag <<  2) |
-                                        /* sp_for_switch_flag (not implemented by FFmpeg) */
+                                        /* sp_for_switch_flag (not implemented by Libav) */
                                         (0                                    <<  3) |
                                         (h->sps.chroma_format_idc             <<  4) |
                                         ((h->nal_ref_idc != 0)                <<  6) |
@@ -108,19 +109,24 @@ static void fill_picture_parameters(struct dxva_context *ctx, const H264Context
                                         (h->pps.transform_8x8_mode            << 13) |
                                         ((h->sps.level_idc >= 31)             << 14) |
                                         /* IntraPicFlag (Modified if we detect a non
-                                         * intra slice in decode_slice) */
+                                         * intra slice in dxva2_h264_decode_slice) */
                                         (1                                    << 15);
 
     pp->bit_depth_luma_minus8         = h->sps.bit_depth_luma - 8;
     pp->bit_depth_chroma_minus8       = h->sps.bit_depth_chroma - 8;
-    pp->Reserved16Bits                = 3; /* FIXME is there a way to detect the right mode ? */
+    if (ctx->workaround & FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG)
+        pp->Reserved16Bits            = 0;
+    else if (ctx->workaround & FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO)
+        pp->Reserved16Bits            = 0x34c;
+    else
+        pp->Reserved16Bits            = 3; /* FIXME is there a way to detect the right mode ? */
     pp->StatusReportFeedbackNumber    = 1 + ctx->report_id++;
     pp->CurrFieldOrderCnt[0] = 0;
-    if ((s->picture_structure & PICT_TOP_FIELD) &&
+    if ((h->picture_structure & PICT_TOP_FIELD) &&
         current_picture->field_poc[0] != INT_MAX)
         pp->CurrFieldOrderCnt[0] = current_picture->field_poc[0];
     pp->CurrFieldOrderCnt[1] = 0;
-    if ((s->picture_structure & PICT_BOTTOM_FIELD) &&
+    if ((h->picture_structure & PICT_BOTTOM_FIELD) &&
         current_picture->field_poc[1] != INT_MAX)
         pp->CurrFieldOrderCnt[1] = current_picture->field_poc[1];
     pp->pic_init_qs_minus26           = h->pps.init_qs - 26;
@@ -146,21 +152,33 @@ static void fill_picture_parameters(struct dxva_context *ctx, const H264Context
     pp->deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
     pp->redundant_pic_cnt_present_flag= h->pps.redundant_pic_cnt_present;
     pp->Reserved8BitsB                = 0;
-    pp->slice_group_change_rate_minus1= 0;  /* XXX not implemented by FFmpeg */
-    //pp->SliceGroupMap[810];               /* XXX not implemented by FFmpeg */
+    pp->slice_group_change_rate_minus1= 0;  /* XXX not implemented by Libav */
+    //pp->SliceGroupMap[810];               /* XXX not implemented by Libav */
 }
 
-static void fill_scaling_lists(const H264Context *h, DXVA_Qmatrix_H264 *qm)
+static void fill_scaling_lists(struct dxva_context *ctx, const H264Context *h, DXVA_Qmatrix_H264 *qm)
 {
     unsigned i, j;
     memset(qm, 0, sizeof(*qm));
-    for (i = 0; i < 6; i++)
-        for (j = 0; j < 16; j++)
-            qm->bScalingLists4x4[i][j] = h->pps.scaling_matrix4[i][zigzag_scan[j]];
+    if (ctx->workaround & FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG) {
+        for (i = 0; i < 6; i++)
+            for (j = 0; j < 16; j++)
+                qm->bScalingLists4x4[i][j] = h->pps.scaling_matrix4[i][j];
+
+        for (i = 0; i < 64; i++) {
+            qm->bScalingLists8x8[0][i] = h->pps.scaling_matrix8[0][i];
+            qm->bScalingLists8x8[1][i] = h->pps.scaling_matrix8[3][i];
+        }
+    } else {
+        for (i = 0; i < 6; i++)
+            for (j = 0; j < 16; j++)
+                qm->bScalingLists4x4[i][j] = h->pps.scaling_matrix4[i][zigzag_scan[j]];
 
-    for (i = 0; i < 2; i++)
-        for (j = 0; j < 64; j++)
-            qm->bScalingLists8x8[i][j] = h->pps.scaling_matrix8[i][ff_zigzag_direct[j]];
+        for (i = 0; i < 64; i++) {
+            qm->bScalingLists8x8[0][i] = h->pps.scaling_matrix8[0][ff_zigzag_direct[i]];
+            qm->bScalingLists8x8[1][i] = h->pps.scaling_matrix8[3][ff_zigzag_direct[i]];
+        }
+    }
 }
 
 static int is_slice_short(struct dxva_context *ctx)
@@ -179,12 +197,22 @@ static void fill_slice_short(DXVA_Slice_H264_Short *slice,
     slice->wBadSliceChopping     = 0;
 }
 
+static int get_refpic_index(const DXVA_PicParams_H264 *pp, int surface_index)
+{
+    int i;
+    for (i = 0; i < FF_ARRAY_ELEMS(pp->RefFrameList); i++) {
+        if ((pp->RefFrameList[i].bPicEntry & 0x7f) == surface_index)
+          return i;
+    }
+    return 0x7f;
+}
+
 static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice,
-                            unsigned position, unsigned size)
+                            const DXVA_PicParams_H264 *pp, unsigned position, unsigned size)
 {
     const H264Context *h = avctx->priv_data;
+    H264SliceContext *sl = &h->slice_ctx[0];
     struct dxva_context *ctx = avctx->hwaccel_context;
-    const MpegEncContext *s = &h->s;
     unsigned list;
 
     memset(slice, 0, sizeof(*slice));
@@ -192,42 +220,46 @@ static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice,
     slice->SliceBytesInBuffer    = size;
     slice->wBadSliceChopping     = 0;
 
-    slice->first_mb_in_slice     = (s->mb_y >> FIELD_OR_MBAFF_PICTURE) * s->mb_width + s->mb_x;
+    slice->first_mb_in_slice     = (h->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + h->mb_x;
     slice->NumMbsForSlice        = 0; /* XXX it is set once we have all slices */
-    slice->BitOffsetToSliceData  = get_bits_count(&s->gb) + 8;
+    slice->BitOffsetToSliceData  = get_bits_count(&h->gb);
     slice->slice_type            = ff_h264_get_slice_type(h);
     if (h->slice_type_fixed)
         slice->slice_type += 5;
-    slice->luma_log2_weight_denom       = h->luma_log2_weight_denom;
-    slice->chroma_log2_weight_denom     = h->chroma_log2_weight_denom;
+    slice->luma_log2_weight_denom       = sl->luma_log2_weight_denom;
+    slice->chroma_log2_weight_denom     = sl->chroma_log2_weight_denom;
     if (h->list_count > 0)
         slice->num_ref_idx_l0_active_minus1 = h->ref_count[0] - 1;
     if (h->list_count > 1)
         slice->num_ref_idx_l1_active_minus1 = h->ref_count[1] - 1;
-    slice->slice_alpha_c0_offset_div2   = h->slice_alpha_c0_offset / 2 - 26;
-    slice->slice_beta_offset_div2       = h->slice_beta_offset     / 2 - 26;
+    slice->slice_alpha_c0_offset_div2   = h->slice_alpha_c0_offset / 2;
+    slice->slice_beta_offset_div2       = h->slice_beta_offset     / 2;
     slice->Reserved8Bits                = 0;
 
     for (list = 0; list < 2; list++) {
         unsigned i;
         for (i = 0; i < FF_ARRAY_ELEMS(slice->RefPicList[list]); i++) {
             if (list < h->list_count && i < h->ref_count[list]) {
-                const Picture *r = &h->ref_list[list][i];
+                const H264Picture *r = &h->ref_list[list][i];
                 unsigned plane;
-                fill_picture_entry(&slice->RefPicList[list][i],
-                                   ff_dxva2_get_surface_index(ctx, r),
+                unsigned index;
+                if (ctx->workaround & FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO)
+                    index = ff_dxva2_get_surface_index(ctx, &r->f);
+                else
+                    index = get_refpic_index(pp, ff_dxva2_get_surface_index(ctx, &r->f));
+                fill_picture_entry(&slice->RefPicList[list][i], index,
                                    r->reference == PICT_BOTTOM_FIELD);
                 for (plane = 0; plane < 3; plane++) {
                     int w, o;
-                    if (plane == 0 && h->luma_weight_flag[list]) {
-                        w = h->luma_weight[i][list][0];
-                        o = h->luma_weight[i][list][1];
-                    } else if (plane >= 1 && h->chroma_weight_flag[list]) {
-                        w = h->chroma_weight[i][list][plane-1][0];
-                        o = h->chroma_weight[i][list][plane-1][1];
+                    if (plane == 0 && sl->luma_weight_flag[list]) {
+                        w = sl->luma_weight[i][list][0];
+                        o = sl->luma_weight[i][list][1];
+                    } else if (plane >= 1 && sl->chroma_weight_flag[list]) {
+                        w = sl->chroma_weight[i][list][plane-1][0];
+                        o = sl->chroma_weight[i][list][plane-1][1];
                     } else {
-                        w = 1 << (plane == 0 ? h->luma_log2_weight_denom :
-                                               h->chroma_log2_weight_denom);
+                        w = 1 << (plane == 0 ? sl->luma_log2_weight_denom :
+                                               sl->chroma_log2_weight_denom);
                         o = 0;
                     }
                     slice->Weights[list][i][plane][0] = w;
@@ -243,10 +275,10 @@ static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice,
             }
         }
     }
-    slice->slice_qs_delta    = 0; /* XXX not implemented by FFmpeg */
-    slice->slice_qp_delta    = s->qscale - h->pps.init_qp;
+    slice->slice_qs_delta    = 0; /* XXX not implemented by Libav */
+    slice->slice_qp_delta    = sl->qscale - h->pps.init_qp;
     slice->redundant_pic_cnt = h->redundant_pic_count;
-    if (h->slice_type == FF_B_TYPE)
+    if (h->slice_type == AV_PICTURE_TYPE_B)
         slice->direct_spatial_mv_pred_flag = h->direct_spatial_mv_pred;
     slice->cabac_init_idc = h->pps.cabac ? h->cabac_init_idc : 0;
     if (h->deblocking_filter < 2)
@@ -261,12 +293,12 @@ static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
                                              DXVA2_DecodeBufferDesc *sc)
 {
     const H264Context *h = avctx->priv_data;
-    const MpegEncContext *s = &h->s;
-    const unsigned mb_count = s->mb_width * s->mb_height;
+    const unsigned mb_count = h->mb_width * h->mb_height;
     struct dxva_context *ctx = avctx->hwaccel_context;
-    const Picture *current_picture = h->s.current_picture_ptr;
+    const H264Picture *current_picture = h->cur_pic_ptr;
     struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
     DXVA_Slice_H264_Short *slice = NULL;
+    void     *dxva_data_ptr;
     uint8_t  *dxva_data, *current, *end;
     unsigned dxva_size;
     void     *slice_data;
@@ -276,9 +308,11 @@ static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
 
     /* Create an annex B bitstream buffer with only slice NAL and finalize slice */
     if (FAILED(IDirectXVideoDecoder_GetBuffer(ctx->decoder,
-                                               DXVA2_BitStreamDateBufferType,
-                                               &dxva_data, &dxva_size)))
+                                              DXVA2_BitStreamDateBufferType,
+                                              &dxva_data_ptr, &dxva_size)))
         return -1;
+
+    dxva_data = dxva_data_ptr;
     current = dxva_data;
     end = dxva_data + dxva_size;
 
@@ -354,13 +388,13 @@ static int commit_bitstream_and_slice_buffer(AVCodecContext *avctx,
 }
 
 
-static int start_frame(AVCodecContext *avctx,
-                       av_unused const uint8_t *buffer,
-                       av_unused uint32_t size)
+static int dxva2_h264_start_frame(AVCodecContext *avctx,
+                                  av_unused const uint8_t *buffer,
+                                  av_unused uint32_t size)
 {
     const H264Context *h = avctx->priv_data;
     struct dxva_context *ctx = avctx->hwaccel_context;
-    struct dxva2_picture_context *ctx_pic = h->s.current_picture_ptr->hwaccel_picture_private;
+    struct dxva2_picture_context *ctx_pic = h->cur_pic_ptr->hwaccel_picture_private;
 
     if (!ctx->decoder || !ctx->cfg || ctx->surface_count <= 0)
         return -1;
@@ -370,7 +404,7 @@ static int start_frame(AVCodecContext *avctx,
     fill_picture_parameters(ctx, h, &ctx_pic->pp);
 
     /* Fill up DXVA_Qmatrix_H264 */
-    fill_scaling_lists(h, &ctx_pic->qm);
+    fill_scaling_lists(ctx, h, &ctx_pic->qm);
 
     ctx_pic->slice_count    = 0;
     ctx_pic->bitstream_size = 0;
@@ -378,12 +412,13 @@ static int start_frame(AVCodecContext *avctx,
     return 0;
 }
 
-static int decode_slice(AVCodecContext *avctx,
-                        const uint8_t *buffer, uint32_t size)
+static int dxva2_h264_decode_slice(AVCodecContext *avctx,
+                                   const uint8_t *buffer,
+                                   uint32_t size)
 {
     const H264Context *h = avctx->priv_data;
     struct dxva_context *ctx = avctx->hwaccel_context;
-    const Picture *current_picture = h->s.current_picture_ptr;
+    const H264Picture *current_picture = h->cur_pic_ptr;
     struct dxva2_picture_context *ctx_pic = current_picture->hwaccel_picture_private;
     unsigned position;
 
@@ -400,38 +435,39 @@ static int decode_slice(AVCodecContext *avctx,
                          position, size);
     else
         fill_slice_long(avctx, &ctx_pic->slice_long[ctx_pic->slice_count],
-                        position, size);
+                        &ctx_pic->pp, position, size);
     ctx_pic->slice_count++;
 
-    if (h->slice_type != FF_I_TYPE && h->slice_type != FF_SI_TYPE)
+    if (h->slice_type != AV_PICTURE_TYPE_I && h->slice_type != AV_PICTURE_TYPE_SI)
         ctx_pic->pp.wBitFields &= ~(1 << 15); /* Set IntraPicFlag to 0 */
     return 0;
 }
 
-static int end_frame(AVCodecContext *avctx)
+static int dxva2_h264_end_frame(AVCodecContext *avctx)
 {
     H264Context *h = avctx->priv_data;
-    MpegEncContext *s = &h->s;
     struct dxva2_picture_context *ctx_pic =
-        h->s.current_picture_ptr->hwaccel_picture_private;
+        h->cur_pic_ptr->hwaccel_picture_private;
+    int ret;
 
     if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0)
         return -1;
-    return ff_dxva2_common_end_frame(avctx, s,
-                                     &ctx_pic->pp, sizeof(ctx_pic->pp),
-                                     &ctx_pic->qm, sizeof(ctx_pic->qm),
-                                     commit_bitstream_and_slice_buffer);
+    ret = ff_dxva2_common_end_frame(avctx, &h->cur_pic_ptr->f,
+                                    &ctx_pic->pp, sizeof(ctx_pic->pp),
+                                    &ctx_pic->qm, sizeof(ctx_pic->qm),
+                                    commit_bitstream_and_slice_buffer);
+    if (!ret)
+        ff_h264_draw_horiz_band(h, 0, h->avctx->height);
+    return ret;
 }
 
-AVHWAccel h264_dxva2_hwaccel = {
+AVHWAccel ff_h264_dxva2_hwaccel = {
     .name           = "h264_dxva2",
     .type           = AVMEDIA_TYPE_VIDEO,
-    .id             = CODEC_ID_H264,
-    .pix_fmt        = PIX_FMT_DXVA2_VLD,
-    .capabilities   = 0,
-    .start_frame    = start_frame,
-    .decode_slice   = decode_slice,
-    .end_frame      = end_frame,
-    .priv_data_size = sizeof(struct dxva2_picture_context),
+    .id             = AV_CODEC_ID_H264,
+    .pix_fmt        = AV_PIX_FMT_DXVA2_VLD,
+    .start_frame    = dxva2_h264_start_frame,
+    .decode_slice   = dxva2_h264_decode_slice,
+    .end_frame      = dxva2_h264_end_frame,
+    .frame_priv_data_size = sizeof(struct dxva2_picture_context),
 };
-