Skip to content

Commit

Permalink
h264: decouple h264_ps from the h264 decoder
Browse files Browse the repository at this point in the history 
Make the SPS/PPS parsing independent of the H264Context, to allow
decoupling the parser from the decoder. The change is modelled after the
one done earlier for HEVC.

Move the dequant buffers to the PPS to avoid complex checks whether they
changed and an expensive copy for frame threads.
  • Loading branch information
@elenril
elenril committed Apr 24, 2016
1 parent 44d16df commit 3176217
Show file tree
Hide file tree
Showing 17 changed files with 677 additions and 630 deletions.
80 changes: 41 additions & 39 deletions libavcodec/dxva2_h264.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -50,6 +50,8 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *
DXVA_PicParams_H264 *pp)
{
const H264Picture *current_picture = h->cur_pic_ptr;
const SPS *sps = h->ps.sps;
const PPS *pps = h->ps.pps;
int i, j;

memset(pp, 0, sizeof(*pp));
Expand Down Expand Up @@ -94,30 +96,30 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *

pp->wFrameWidthInMbsMinus1 = h->mb_width - 1;
pp->wFrameHeightInMbsMinus1 = h->mb_height - 1;
pp->num_ref_frames = h->sps.ref_frame_count;
pp->num_ref_frames = sps->ref_frame_count;

pp->wBitFields = ((h->picture_structure != PICT_FRAME) << 0) |
((h->sps.mb_aff &&
((sps->mb_aff &&
(h->picture_structure == PICT_FRAME)) << 1) |
(h->sps.residual_color_transform_flag << 2) |
(sps->residual_color_transform_flag << 2) |
/* sp_for_switch_flag (not implemented by Libav) */
(0 << 3) |
(h->sps.chroma_format_idc << 4) |
(sps->chroma_format_idc << 4) |
((h->nal_ref_idc != 0) << 6) |
(h->pps.constrained_intra_pred << 7) |
(h->pps.weighted_pred << 8) |
(h->pps.weighted_bipred_idc << 9) |
(pps->constrained_intra_pred << 7) |
(pps->weighted_pred << 8) |
(pps->weighted_bipred_idc << 9) |
/* MbsConsecutiveFlag */
(1 << 11) |
(h->sps.frame_mbs_only_flag << 12) |
(h->pps.transform_8x8_mode << 13) |
((h->sps.level_idc >= 31) << 14) |
(sps->frame_mbs_only_flag << 12) |
(pps->transform_8x8_mode << 13) |
((sps->level_idc >= 31) << 14) |
/* IntraPicFlag (Modified if we detect a non
* 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->bit_depth_luma_minus8 = sps->bit_depth_luma - 8;
pp->bit_depth_chroma_minus8 = sps->bit_depth_chroma - 8;
if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG)
pp->Reserved16Bits = 0;
else if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO)
Expand All @@ -133,28 +135,28 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *
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;
pp->chroma_qp_index_offset = h->pps.chroma_qp_index_offset[0];
pp->second_chroma_qp_index_offset = h->pps.chroma_qp_index_offset[1];
pp->pic_init_qs_minus26 = pps->init_qs - 26;
pp->chroma_qp_index_offset = pps->chroma_qp_index_offset[0];
pp->second_chroma_qp_index_offset = pps->chroma_qp_index_offset[1];
pp->ContinuationFlag = 1;
pp->pic_init_qp_minus26 = h->pps.init_qp - 26;
pp->num_ref_idx_l0_active_minus1 = h->pps.ref_count[0] - 1;
pp->num_ref_idx_l1_active_minus1 = h->pps.ref_count[1] - 1;
pp->pic_init_qp_minus26 = pps->init_qp - 26;
pp->num_ref_idx_l0_active_minus1 = pps->ref_count[0] - 1;
pp->num_ref_idx_l1_active_minus1 = pps->ref_count[1] - 1;
pp->Reserved8BitsA = 0;
pp->frame_num = h->frame_num;
pp->log2_max_frame_num_minus4 = h->sps.log2_max_frame_num - 4;
pp->pic_order_cnt_type = h->sps.poc_type;
if (h->sps.poc_type == 0)
pp->log2_max_pic_order_cnt_lsb_minus4 = h->sps.log2_max_poc_lsb - 4;
else if (h->sps.poc_type == 1)
pp->delta_pic_order_always_zero_flag = h->sps.delta_pic_order_always_zero_flag;
pp->direct_8x8_inference_flag = h->sps.direct_8x8_inference_flag;
pp->entropy_coding_mode_flag = h->pps.cabac;
pp->pic_order_present_flag = h->pps.pic_order_present;
pp->num_slice_groups_minus1 = h->pps.slice_group_count - 1;
pp->slice_group_map_type = h->pps.mb_slice_group_map_type;
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->log2_max_frame_num_minus4 = sps->log2_max_frame_num - 4;
pp->pic_order_cnt_type = sps->poc_type;
if (sps->poc_type == 0)
pp->log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4;
else if (sps->poc_type == 1)
pp->delta_pic_order_always_zero_flag = sps->delta_pic_order_always_zero_flag;
pp->direct_8x8_inference_flag = sps->direct_8x8_inference_flag;
pp->entropy_coding_mode_flag = pps->cabac;
pp->pic_order_present_flag = pps->pic_order_present;
pp->num_slice_groups_minus1 = pps->slice_group_count - 1;
pp->slice_group_map_type = pps->mb_slice_group_map_type;
pp->deblocking_filter_control_present_flag = pps->deblocking_filter_parameters_present;
pp->redundant_pic_cnt_present_flag= pps->redundant_pic_cnt_present;
pp->Reserved8BitsB = 0;
pp->slice_group_change_rate_minus1= 0; /* XXX not implemented by Libav */
//pp->SliceGroupMap[810]; /* XXX not implemented by Libav */
Expand All @@ -167,20 +169,20 @@ static void fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx,
if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & 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];
qm->bScalingLists4x4[i][j] = 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];
qm->bScalingLists8x8[0][i] = pps->scaling_matrix8[0][i];
qm->bScalingLists8x8[1][i] = 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][ff_zigzag_scan[j]];
qm->bScalingLists4x4[i][j] = pps->scaling_matrix4[i][ff_zigzag_scan[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]];
qm->bScalingLists8x8[0][i] = pps->scaling_matrix8[0][ff_zigzag_direct[i]];
qm->bScalingLists8x8[1][i] = pps->scaling_matrix8[3][ff_zigzag_direct[i]];
}
}
}
Expand Down Expand Up @@ -280,11 +282,11 @@ static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice,
}
}
slice->slice_qs_delta = 0; /* XXX not implemented by Libav */
slice->slice_qp_delta = sl->qscale - h->pps.init_qp;
slice->slice_qp_delta = sl->qscale - h->ps.pps->init_qp;
slice->redundant_pic_cnt = sl->redundant_pic_count;
if (sl->slice_type == AV_PICTURE_TYPE_B)
slice->direct_spatial_mv_pred_flag = sl->direct_spatial_mv_pred;
slice->cabac_init_idc = h->pps.cabac ? sl->cabac_init_idc : 0;
slice->cabac_init_idc = h->ps.pps->cabac ? sl->cabac_init_idc : 0;
if (sl->deblocking_filter < 2)
slice->disable_deblocking_filter_idc = 1 - sl->deblocking_filter;
else
Expand Down
69 changes: 33 additions & 36 deletions libavcodec/h264.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -206,9 +206,6 @@ int ff_h264_alloc_tables(H264Context *h)
h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
}

if (!h->dequant4_coeff[0])
ff_h264_init_dequant_tables(h);

return 0;

fail:
Expand Down Expand Up @@ -397,7 +394,6 @@ static int h264_init_context(AVCodecContext *avctx, H264Context *h)
int i;

h->avctx = avctx;
h->dequant_coeff_pps = -1;

h->picture_structure = PICT_FRAME;
h->slice_context_count = 1;
Expand Down Expand Up @@ -474,9 +470,9 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx)
}
}

if (h->sps.bitstream_restriction_flag &&
h->avctx->has_b_frames < h->sps.num_reorder_frames) {
h->avctx->has_b_frames = h->sps.num_reorder_frames;
if (h->ps.sps && h->ps.sps->bitstream_restriction_flag &&
h->avctx->has_b_frames < h->ps.sps->num_reorder_frames) {
h->avctx->has_b_frames = h->ps.sps->num_reorder_frames;
h->low_delay = 0;
}

Expand Down Expand Up @@ -520,6 +516,7 @@ static int decode_init_thread_copy(AVCodecContext *avctx)
*/
static void decode_postinit(H264Context *h, int setup_finished)
{
const SPS *sps = h->ps.sps;
H264Picture *out = h->cur_pic_ptr;
H264Picture *cur = h->cur_pic_ptr;
int i, pics, out_of_order, out_idx;
Expand Down Expand Up @@ -547,7 +544,7 @@ static void decode_postinit(H264Context *h, int setup_finished)
/* Prioritize picture timing SEI information over used
* decoding process if it exists. */

if (h->sps.pic_struct_present_flag) {
if (sps->pic_struct_present_flag) {
switch (h->sei_pic_struct) {
case SEI_PIC_STRUCT_FRAME:
break;
Expand Down Expand Up @@ -591,7 +588,7 @@ static void decode_postinit(H264Context *h, int setup_finished)
/* Derive top_field_first from field pocs. */
cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
} else {
if (cur->f->interlaced_frame || h->sps.pic_struct_present_flag) {
if (cur->f->interlaced_frame || sps->pic_struct_present_flag) {
/* Use picture timing SEI information. Even if it is a
* information of a past frame, better than nothing. */
if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
Expand Down Expand Up @@ -684,9 +681,9 @@ static void decode_postinit(H264Context *h, int setup_finished)
// FIXME do something with unavailable reference frames

/* Sort B-frames into display order */
if (h->sps.bitstream_restriction_flag ||
if (sps->bitstream_restriction_flag ||
h->avctx->strict_std_compliance >= FF_COMPLIANCE_NORMAL) {
h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, h->sps.num_reorder_frames);
h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, sps->num_reorder_frames);
}
h->low_delay = !h->avctx->has_b_frames;

Expand Down Expand Up @@ -746,8 +743,8 @@ static void decode_postinit(H264Context *h, int setup_finished)
out_of_order = !out->f->key_frame && !h->mmco_reset &&
(out->poc < h->next_outputed_poc);

if (h->sps.bitstream_restriction_flag &&
h->avctx->has_b_frames >= h->sps.num_reorder_frames) {
if (sps->bitstream_restriction_flag &&
h->avctx->has_b_frames >= sps->num_reorder_frames) {
} else if (out_of_order && pics - 1 == h->avctx->has_b_frames &&
h->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
Expand Down Expand Up @@ -862,15 +859,16 @@ static void flush_dpb(AVCodecContext *avctx)

int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
{
const int max_frame_num = 1 << h->sps.log2_max_frame_num;
const SPS *sps = h->ps.sps;
const int max_frame_num = 1 << sps->log2_max_frame_num;
int field_poc[2];

h->frame_num_offset = h->prev_frame_num_offset;
if (h->frame_num < h->prev_frame_num)
h->frame_num_offset += max_frame_num;

if (h->sps.poc_type == 0) {
const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
if (sps->poc_type == 0) {
const int max_poc_lsb = 1 << sps->log2_max_poc_lsb;

if (h->poc_lsb < h->prev_poc_lsb &&
h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
Expand All @@ -884,11 +882,11 @@ int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
field_poc[1] = h->poc_msb + h->poc_lsb;
if (h->picture_structure == PICT_FRAME)
field_poc[1] += h->delta_poc_bottom;
} else if (h->sps.poc_type == 1) {
} else if (sps->poc_type == 1) {
int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
int i;

if (h->sps.poc_cycle_length != 0)
if (sps->poc_cycle_length != 0)
abs_frame_num = h->frame_num_offset + h->frame_num;
else
abs_frame_num = 0;
Expand All @@ -897,25 +895,25 @@ int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
abs_frame_num--;

expected_delta_per_poc_cycle = 0;
for (i = 0; i < h->sps.poc_cycle_length; i++)
for (i = 0; i < sps->poc_cycle_length; i++)
// FIXME integrate during sps parse
expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
expected_delta_per_poc_cycle += sps->offset_for_ref_frame[i];

if (abs_frame_num > 0) {
int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
int poc_cycle_cnt = (abs_frame_num - 1) / sps->poc_cycle_length;
int frame_num_in_poc_cycle = (abs_frame_num - 1) % sps->poc_cycle_length;

expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
for (i = 0; i <= frame_num_in_poc_cycle; i++)
expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
expectedpoc = expectedpoc + sps->offset_for_ref_frame[i];
} else
expectedpoc = 0;

if (h->nal_ref_idc == 0)
expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
expectedpoc = expectedpoc + sps->offset_for_non_ref_pic;

field_poc[0] = expectedpoc + h->delta_poc[0];
field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
field_poc[1] = field_poc[0] + sps->offset_for_top_to_bottom_field;

if (h->picture_structure == PICT_FRAME)
field_poc[1] += h->delta_poc[1];
Expand Down Expand Up @@ -945,7 +943,7 @@ int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
*
* @return profile as defined by FF_PROFILE_H264_*
*/
int ff_h264_get_profile(SPS *sps)
int ff_h264_get_profile(const SPS *sps)
{
int profile = sps->profile_idc;

Expand Down Expand Up @@ -1067,7 +1065,7 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,

if (h->sei_recovery_frame_cnt >= 0 && h->recovery_frame < 0) {
h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) &
((1 << h->sps.log2_max_frame_num) - 1);
((1 << h->ps.sps->log2_max_frame_num) - 1);
}

h->cur_pic_ptr->f->key_frame |=
Expand Down Expand Up @@ -1121,14 +1119,13 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
goto end;
break;
case NAL_SPS:
h->gb = nal->gb;
ret = ff_h264_decode_seq_parameter_set(h);
ret = ff_h264_decode_seq_parameter_set(&nal->gb, avctx, &h->ps);
if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
goto end;
break;
case NAL_PPS:
h->gb = nal->gb;
ret = ff_h264_decode_picture_parameter_set(h, nal->size_bits);
ret = ff_h264_decode_picture_parameter_set(&nal->gb, avctx, &h->ps,
nal->size_bits);
if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
goto end;
break;
Expand Down Expand Up @@ -1201,14 +1198,14 @@ static int output_frame(H264Context *h, AVFrame *dst, AVFrame *src)
if (ret < 0)
return ret;

if (!h->sps.crop)
if (!h->ps.sps || !h->ps.sps->crop)
return 0;

for (i = 0; i < 3; i++) {
int hshift = (i > 0) ? h->chroma_x_shift : 0;
int vshift = (i > 0) ? h->chroma_y_shift : 0;
int off = ((h->sps.crop_left >> hshift) << h->pixel_shift) +
(h->sps.crop_top >> vshift) * dst->linesize[i];
int off = ((h->ps.sps->crop_left >> hshift) << h->pixel_shift) +
(h->ps.sps->crop_top >> vshift) * dst->linesize[i];
dst->data[i] += off;
}
return 0;
Expand Down Expand Up @@ -1321,10 +1318,10 @@ av_cold void ff_h264_free_context(H264Context *h)
h->nb_slice_ctx = 0;

for (i = 0; i < MAX_SPS_COUNT; i++)
av_freep(h->sps_buffers + i);
av_buffer_unref(&h->ps.sps_list[i]);

for (i = 0; i < MAX_PPS_COUNT; i++)
av_freep(h->pps_buffers + i);
av_buffer_unref(&h->ps.pps_list[i]);

ff_h2645_packet_uninit(&h->pkt);
}
Expand Down
Loading

0 comments on commit 3176217

Please sign in to comment.