har0ke/FFmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

vaapi_h264: Convert to use the new VAAPI hwaccel code

Browse Source 
(cherry picked from commit 2fe93244ab9465ef19c756a41524cbbf0580ee48)
This commit is contained in:
Mark Thompson 2016-08-06 18:19:32 +01:00
parent 79307ae563
commit c8b26d5954
1 changed files with 133 additions and 98 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

231
libavcodec/vaapi_h264.c
View File

@ -20,10 +20,9 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/ */
#include "vaapi_internal.h"
#include "h264dec.h" #include "h264dec.h"
#include "h264_ps.h" #include "h264_ps.h"
#include "mpegutils.h" #include "vaapi_decode.h"
/** /**
* @file * @file
@ -228,88 +227,102 @@ static int vaapi_h264_start_frame(AVCodecContext *avctx,
av_unused uint32_t size) av_unused uint32_t size)
{ {
const H264Context *h = avctx->priv_data; const H264Context *h = avctx->priv_data;
FFVAContext *vactx = ff_vaapi_get_context(avctx); VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
const PPS *pps = h->ps.pps; const PPS *pps = h->ps.pps;
const SPS *sps = h->ps.sps; const SPS *sps = h->ps.sps;
VAPictureParameterBufferH264 *pic_param; VAPictureParameterBufferH264 pic_param;
VAIQMatrixBufferH264 *iq_matrix; VAIQMatrixBufferH264 iq_matrix;
int err;
vactx->slice_param_size = sizeof(VASliceParameterBufferH264); pic->output_surface = ff_vaapi_get_surface_id(h->cur_pic_ptr->f);
/* Fill in VAPictureParameterBufferH264. */ pic_param = (VAPictureParameterBufferH264) {
pic_param = ff_vaapi_alloc_pic_param(vactx, sizeof(VAPictureParameterBufferH264)); .picture_width_in_mbs_minus1 = h->mb_width - 1,
if (!pic_param) .picture_height_in_mbs_minus1 = h->mb_height - 1,
return -1; .bit_depth_luma_minus8 = sps->bit_depth_luma - 8,
fill_vaapi_pic(&pic_param->CurrPic, h->cur_pic_ptr, h->picture_structure); .bit_depth_chroma_minus8 = sps->bit_depth_chroma - 8,
if (fill_vaapi_ReferenceFrames(pic_param, h) < 0) .num_ref_frames = sps->ref_frame_count,
return -1; .seq_fields.bits = {
pic_param->picture_width_in_mbs_minus1 = h->mb_width - 1; .chroma_format_idc = sps->chroma_format_idc,
pic_param->picture_height_in_mbs_minus1 = h->mb_height - 1; .residual_colour_transform_flag = sps->residual_color_transform_flag,
pic_param->bit_depth_luma_minus8 = sps->bit_depth_luma - 8; .gaps_in_frame_num_value_allowed_flag = sps->gaps_in_frame_num_allowed_flag,
pic_param->bit_depth_chroma_minus8 = sps->bit_depth_chroma - 8; .frame_mbs_only_flag = sps->frame_mbs_only_flag,
pic_param->num_ref_frames = sps->ref_frame_count; .mb_adaptive_frame_field_flag = sps->mb_aff,
pic_param->seq_fields.value = 0; /* reset all bits */ .direct_8x8_inference_flag = sps->direct_8x8_inference_flag,
pic_param->seq_fields.bits.chroma_format_idc = sps->chroma_format_idc; .MinLumaBiPredSize8x8 = sps->level_idc >= 31, /* A.3.3.2 */
pic_param->seq_fields.bits.residual_colour_transform_flag = sps->residual_color_transform_flag; /* XXX: only for 4:4:4 high profile? */ .log2_max_frame_num_minus4 = sps->log2_max_frame_num - 4,
pic_param->seq_fields.bits.gaps_in_frame_num_value_allowed_flag = sps->gaps_in_frame_num_allowed_flag; .pic_order_cnt_type = sps->poc_type,
pic_param->seq_fields.bits.frame_mbs_only_flag = sps->frame_mbs_only_flag; .log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4,
pic_param->seq_fields.bits.mb_adaptive_frame_field_flag = sps->mb_aff; .delta_pic_order_always_zero_flag = sps->delta_pic_order_always_zero_flag,
pic_param->seq_fields.bits.direct_8x8_inference_flag = sps->direct_8x8_inference_flag; },
pic_param->seq_fields.bits.MinLumaBiPredSize8x8 = sps->level_idc >= 31; /* A.3.3.2 */ .num_slice_groups_minus1 = pps->slice_group_count - 1,
pic_param->seq_fields.bits.log2_max_frame_num_minus4 = sps->log2_max_frame_num - 4; .slice_group_map_type = pps->mb_slice_group_map_type,
pic_param->seq_fields.bits.pic_order_cnt_type = sps->poc_type; .slice_group_change_rate_minus1 = 0, /* FMO is not implemented */
pic_param->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4; .pic_init_qp_minus26 = pps->init_qp - 26,
pic_param->seq_fields.bits.delta_pic_order_always_zero_flag = sps->delta_pic_order_always_zero_flag; .pic_init_qs_minus26 = pps->init_qs - 26,
pic_param->num_slice_groups_minus1 = pps->slice_group_count - 1; .chroma_qp_index_offset = pps->chroma_qp_index_offset[0],
pic_param->slice_group_map_type = pps->mb_slice_group_map_type; .second_chroma_qp_index_offset = pps->chroma_qp_index_offset[1],
pic_param->slice_group_change_rate_minus1 = 0; /* XXX: unimplemented in FFmpeg */ .pic_fields.bits = {
pic_param->pic_init_qp_minus26 = pps->init_qp - 26; .entropy_coding_mode_flag = pps->cabac,
pic_param->pic_init_qs_minus26 = pps->init_qs - 26; .weighted_pred_flag = pps->weighted_pred,
pic_param->chroma_qp_index_offset = pps->chroma_qp_index_offset[0]; .weighted_bipred_idc = pps->weighted_bipred_idc,
pic_param->second_chroma_qp_index_offset = pps->chroma_qp_index_offset[1]; .transform_8x8_mode_flag = pps->transform_8x8_mode,
pic_param->pic_fields.value = 0; /* reset all bits */ .field_pic_flag = h->picture_structure != PICT_FRAME,
pic_param->pic_fields.bits.entropy_coding_mode_flag = pps->cabac; .constrained_intra_pred_flag = pps->constrained_intra_pred,
pic_param->pic_fields.bits.weighted_pred_flag = pps->weighted_pred; .pic_order_present_flag = pps->pic_order_present,
pic_param->pic_fields.bits.weighted_bipred_idc = pps->weighted_bipred_idc; .deblocking_filter_control_present_flag = pps->deblocking_filter_parameters_present,
pic_param->pic_fields.bits.transform_8x8_mode_flag = pps->transform_8x8_mode; .redundant_pic_cnt_present_flag = pps->redundant_pic_cnt_present,
pic_param->pic_fields.bits.field_pic_flag = h->picture_structure != PICT_FRAME; .reference_pic_flag = h->nal_ref_idc != 0,
pic_param->pic_fields.bits.constrained_intra_pred_flag = pps->constrained_intra_pred; },
pic_param->pic_fields.bits.pic_order_present_flag = pps->pic_order_present; .frame_num = h->poc.frame_num,
pic_param->pic_fields.bits.deblocking_filter_control_present_flag = pps->deblocking_filter_parameters_present; };
pic_param->pic_fields.bits.redundant_pic_cnt_present_flag = pps->redundant_pic_cnt_present;
pic_param->pic_fields.bits.reference_pic_flag = h->nal_ref_idc != 0; fill_vaapi_pic(&pic_param.CurrPic, h->cur_pic_ptr, h->picture_structure);
pic_param->frame_num = h->poc.frame_num; err = fill_vaapi_ReferenceFrames(&pic_param, h);
if (err < 0)
goto fail;
err = ff_vaapi_decode_make_param_buffer(avctx, pic,
VAPictureParameterBufferType,
&pic_param, sizeof(pic_param));
if (err < 0)
goto fail;
memcpy(iq_matrix.ScalingList4x4,
pps->scaling_matrix4, sizeof(iq_matrix.ScalingList4x4));
memcpy(iq_matrix.ScalingList8x8[0],
pps->scaling_matrix8[0], sizeof(iq_matrix.ScalingList8x8[0]));
memcpy(iq_matrix.ScalingList8x8[1],
pps->scaling_matrix8[3], sizeof(iq_matrix.ScalingList8x8[0]));
err = ff_vaapi_decode_make_param_buffer(avctx, pic,
VAIQMatrixBufferType,
&iq_matrix, sizeof(iq_matrix));
if (err < 0)
goto fail;
/* Fill in VAIQMatrixBufferH264. */
iq_matrix = ff_vaapi_alloc_iq_matrix(vactx, sizeof(VAIQMatrixBufferH264));
if (!iq_matrix)
return -1;
memcpy(iq_matrix->ScalingList4x4, pps->scaling_matrix4, sizeof(iq_matrix->ScalingList4x4));
memcpy(iq_matrix->ScalingList8x8[0], pps->scaling_matrix8[0], sizeof(iq_matrix->ScalingList8x8[0]));
memcpy(iq_matrix->ScalingList8x8[1], pps->scaling_matrix8[3], sizeof(iq_matrix->ScalingList8x8[0]));
return 0; return 0;
fail:
ff_vaapi_decode_cancel(avctx, pic);
return err;
} }
/** End a hardware decoding based frame. */ /** End a hardware decoding based frame. */
static int vaapi_h264_end_frame(AVCodecContext *avctx) static int vaapi_h264_end_frame(AVCodecContext *avctx)
{ {
FFVAContext *vactx = ff_vaapi_get_context(avctx);
const H264Context *h = avctx->priv_data; const H264Context *h = avctx->priv_data;
VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
H264SliceContext *sl = &h->slice_ctx[0]; H264SliceContext *sl = &h->slice_ctx[0];
int ret; int ret;
ret = ff_vaapi_commit_slices(vactx); ret = ff_vaapi_decode_issue(avctx, pic);
if (ret < 0)
goto finish;
ret = ff_vaapi_render_picture(vactx, ff_vaapi_get_surface_id(h->cur_pic_ptr->f));
if (ret < 0) if (ret < 0)
goto finish; goto finish;
ff_h264_draw_horiz_band(h, sl, 0, h->avctx->height); ff_h264_draw_horiz_band(h, sl, 0, h->avctx->height);
finish: finish:
ff_vaapi_common_end_frame(avctx);
return ret; return ret;
} }
@ -318,50 +331,72 @@ static int vaapi_h264_decode_slice(AVCodecContext *avctx,
const uint8_t *buffer, const uint8_t *buffer,
uint32_t size) uint32_t size)
{ {
FFVAContext *vactx = ff_vaapi_get_context(avctx);
const H264Context *h = avctx->priv_data; const H264Context *h = avctx->priv_data;
VAAPIDecodePicture *pic = h->cur_pic_ptr->hwaccel_picture_private;
const H264SliceContext *sl = &h->slice_ctx[0]; const H264SliceContext *sl = &h->slice_ctx[0];
VASliceParameterBufferH264 *slice_param; VASliceParameterBufferH264 slice_param;
int err;
/* Fill in VASliceParameterBufferH264. */ slice_param = (VASliceParameterBufferH264) {
slice_param = (VASliceParameterBufferH264 *)ff_vaapi_alloc_slice(vactx, buffer, size); .slice_data_size = size,
if (!slice_param) .slice_data_offset = 0,
return -1; .slice_data_flag = VA_SLICE_DATA_FLAG_ALL,
slice_param->slice_data_bit_offset = get_bits_count(&sl->gb); .slice_data_bit_offset = get_bits_count(&sl->gb),
slice_param->first_mb_in_slice = (sl->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + sl->mb_x; .first_mb_in_slice = (sl->mb_y >> FIELD_OR_MBAFF_PICTURE(h)) * h->mb_width + sl->mb_x,
slice_param->slice_type = ff_h264_get_slice_type(sl); .slice_type = ff_h264_get_slice_type(sl),
slice_param->direct_spatial_mv_pred_flag = sl->slice_type == AV_PICTURE_TYPE_B ? sl->direct_spatial_mv_pred : 0; .direct_spatial_mv_pred_flag = sl->slice_type == AV_PICTURE_TYPE_B ? sl->direct_spatial_mv_pred : 0,
slice_param->num_ref_idx_l0_active_minus1 = sl->list_count > 0 ? sl->ref_count[0] - 1 : 0; .num_ref_idx_l0_active_minus1 = sl->list_count > 0 ? sl->ref_count[0] - 1 : 0,
slice_param->num_ref_idx_l1_active_minus1 = sl->list_count > 1 ? sl->ref_count[1] - 1 : 0; .num_ref_idx_l1_active_minus1 = sl->list_count > 1 ? sl->ref_count[1] - 1 : 0,
slice_param->cabac_init_idc = sl->cabac_init_idc; .cabac_init_idc = sl->cabac_init_idc,
slice_param->slice_qp_delta = sl->qscale - h->ps.pps->init_qp; .slice_qp_delta = sl->qscale - h->ps.pps->init_qp,
slice_param->disable_deblocking_filter_idc = sl->deblocking_filter < 2 ? !sl->deblocking_filter : sl->deblocking_filter; .disable_deblocking_filter_idc = sl->deblocking_filter < 2 ? !sl->deblocking_filter : sl->deblocking_filter,
slice_param->slice_alpha_c0_offset_div2 = sl->slice_alpha_c0_offset / 2; .slice_alpha_c0_offset_div2 = sl->slice_alpha_c0_offset / 2,
slice_param->slice_beta_offset_div2 = sl->slice_beta_offset / 2; .slice_beta_offset_div2 = sl->slice_beta_offset / 2,
slice_param->luma_log2_weight_denom = sl->pwt.luma_log2_weight_denom; .luma_log2_weight_denom = sl->pwt.luma_log2_weight_denom,
slice_param->chroma_log2_weight_denom = sl->pwt.chroma_log2_weight_denom; .chroma_log2_weight_denom = sl->pwt.chroma_log2_weight_denom,
};
fill_vaapi_RefPicList(slice_param->RefPicList0, sl->ref_list[0], sl->list_count > 0 ? sl->ref_count[0] : 0); fill_vaapi_RefPicList(slice_param.RefPicList0, sl->ref_list[0],
fill_vaapi_RefPicList(slice_param->RefPicList1, sl->ref_list[1], sl->list_count > 1 ? sl->ref_count[1] : 0); sl->list_count > 0 ? sl->ref_count[0] : 0);
fill_vaapi_RefPicList(slice_param.RefPicList1, sl->ref_list[1],
sl->list_count > 1 ? sl->ref_count[1] : 0);
fill_vaapi_plain_pred_weight_table(h, 0, fill_vaapi_plain_pred_weight_table(h, 0,
&slice_param->luma_weight_l0_flag, slice_param->luma_weight_l0, slice_param->luma_offset_l0, &slice_param.luma_weight_l0_flag,
&slice_param->chroma_weight_l0_flag, slice_param->chroma_weight_l0, slice_param->chroma_offset_l0); slice_param.luma_weight_l0,
slice_param.luma_offset_l0,
&slice_param.chroma_weight_l0_flag,
slice_param.chroma_weight_l0,
slice_param.chroma_offset_l0);
fill_vaapi_plain_pred_weight_table(h, 1, fill_vaapi_plain_pred_weight_table(h, 1,
&slice_param->luma_weight_l1_flag, slice_param->luma_weight_l1, slice_param->luma_offset_l1, &slice_param.luma_weight_l1_flag,
&slice_param->chroma_weight_l1_flag, slice_param->chroma_weight_l1, slice_param->chroma_offset_l1); slice_param.luma_weight_l1,
slice_param.luma_offset_l1,
&slice_param.chroma_weight_l1_flag,
slice_param.chroma_weight_l1,
slice_param.chroma_offset_l1);
err = ff_vaapi_decode_make_slice_buffer(avctx, pic,
&slice_param, sizeof(slice_param),
buffer, size);
if (err) {
ff_vaapi_decode_cancel(avctx, pic);
return err;
}
return 0; return 0;
} }
AVHWAccel ff_h264_vaapi_hwaccel = { AVHWAccel ff_h264_vaapi_hwaccel = {
.name = "h264_vaapi", .name = "h264_vaapi",
.type = AVMEDIA_TYPE_VIDEO, .type = AVMEDIA_TYPE_VIDEO,
.id = AV_CODEC_ID_H264, .id = AV_CODEC_ID_H264,
.pix_fmt = AV_PIX_FMT_VAAPI, .pix_fmt = AV_PIX_FMT_VAAPI,
.start_frame = vaapi_h264_start_frame, .start_frame = &vaapi_h264_start_frame,
.end_frame = vaapi_h264_end_frame, .end_frame = &vaapi_h264_end_frame,
.decode_slice = vaapi_h264_decode_slice, .decode_slice = &vaapi_h264_decode_slice,
.init = ff_vaapi_context_init, .frame_priv_data_size = sizeof(VAAPIDecodePicture),
.uninit = ff_vaapi_context_fini, .init = &ff_vaapi_decode_init,
.priv_data_size = sizeof(FFVAContext), .uninit = &ff_vaapi_decode_uninit,
.priv_data_size = sizeof(VAAPIDecodeContext),
}; };