#include "mpegvideo.h"
#include "h264.h"
#include "h264data.h"
+#include "h264chroma.h"
#include "h264_mvpred.h"
#include "golomb.h"
#include "mathops.h"
#include "rectangle.h"
+#include "svq3.h"
#include "thread.h"
#include "vdpau_internal.h"
#include "libavutil/avassert.h"
};
static const enum AVPixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
+#if CONFIG_H264_DXVA2_HWACCEL
AV_PIX_FMT_DXVA2_VLD,
+#endif
+#if CONFIG_H264_VAAPI_HWACCEL
AV_PIX_FMT_VAAPI_VLD,
+#endif
+#if CONFIG_H264_VDA_HWACCEL
AV_PIX_FMT_VDA_VLD,
+#endif
+#if CONFIG_H264_VDPAU_HWACCEL
+ AV_PIX_FMT_VDPAU,
+#endif
AV_PIX_FMT_YUVJ420P,
AV_PIX_FMT_NONE
};
int height, int y_offset, int list)
{
int raw_my = h->mv_cache[list][scan8[n]][1];
- int filter_height = (raw_my & 3) ? 2 : 0;
+ int filter_height_up = (raw_my & 3) ? 2 : 0;
+ int filter_height_down = (raw_my & 3) ? 3 : 0;
int full_my = (raw_my >> 2) + y_offset;
- int top = full_my - filter_height;
- int bottom = full_my + filter_height + height;
+ int top = full_my - filter_height_up;
+ int bottom = full_my + filter_height_down + height;
return FFMAX(abs(top), bottom);
}
full_my < 0 - extra_height ||
full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
full_my + 16 /*FIXME*/ > pic_height + extra_height) {
- s->dsp.emulated_edge_mc(s->edge_emu_buffer,
- src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
- h->mb_linesize,
- 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
- full_my - 2, pic_width, pic_height);
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
+ h->mb_linesize,
+ 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
+ full_my - 2, pic_width, pic_height);
src_y = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
emu = 1;
}
if (chroma_idc == 3 /* yuv444 */) {
src_cb = pic->f.data[1] + offset;
if (emu) {
- s->dsp.emulated_edge_mc(s->edge_emu_buffer,
- src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
- h->mb_linesize,
- 16 + 5, 16 + 5 /*FIXME*/,
- full_mx - 2, full_my - 2,
- pic_width, pic_height);
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
+ h->mb_linesize,
+ 16 + 5, 16 + 5 /*FIXME*/,
+ full_mx - 2, full_my - 2,
+ pic_width, pic_height);
src_cb = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
}
qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
src_cr = pic->f.data[2] + offset;
if (emu) {
- s->dsp.emulated_edge_mc(s->edge_emu_buffer,
- src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
- h->mb_linesize,
- 16 + 5, 16 + 5 /*FIXME*/,
- full_mx - 2, full_my - 2,
- pic_width, pic_height);
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
+ src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
+ h->mb_linesize,
+ 16 + 5, 16 + 5 /*FIXME*/,
+ full_mx - 2, full_my - 2,
+ pic_width, pic_height);
src_cr = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
}
qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
(my >> ysh) * h->mb_uvlinesize;
if (emu) {
- s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
- 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
- pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
+ 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
+ pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
src_cb = s->edge_emu_buffer;
}
chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
if (emu) {
- s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
- 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
- pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
+ s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
+ 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
+ pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
src_cr = s->edge_emu_buffer;
}
chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
int off = (mx << pixel_shift) +
(my + (s->mb_x & 3) * 4) * h->mb_linesize +
(64 << pixel_shift);
- s->dsp.prefetch(src[0] + off, s->linesize, 4);
+ s->vdsp.prefetch(src[0] + off, s->linesize, 4);
if (chroma_idc == 3 /* yuv444 */) {
- s->dsp.prefetch(src[1] + off, s->linesize, 4);
- s->dsp.prefetch(src[2] + off, s->linesize, 4);
+ s->vdsp.prefetch(src[1] + off, s->linesize, 4);
+ s->vdsp.prefetch(src[2] + off, s->linesize, 4);
} else {
off = ((mx >> 1) << pixel_shift) +
((my >> 1) + (s->mb_x & 7)) * s->uvlinesize +
(64 << pixel_shift);
- s->dsp.prefetch(src[1] + off, src[2] - src[1], 2);
+ s->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
}
}
}
s->codec_id = s->avctx->codec->id;
ff_h264dsp_init(&h->h264dsp, 8, 1);
+ ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
+ ff_h264qpel_init(&h->h264qpel, 8);
ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
h->dequant_coeff_pps = -1;
/* needed so that IDCT permutation is known early */
ff_dsputil_init(&s->dsp, s->avctx);
+ ff_videodsp_init(&s->vdsp, 8);
memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
// reference lists
copy_fields(h, h1, ref_count, list_count);
- copy_fields(h, h1, ref_list, intra_gb);
+ copy_fields(h, h1, ref2frm, intra_gb);
copy_fields(h, h1, short_ref, cabac_init_idc);
copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
cur->f.repeat_pict = 1;
break;
case SEI_PIC_STRUCT_FRAME_DOUBLING:
- // Force progressive here, doubling interlaced frame is a bad idea.
cur->f.repeat_pict = 2;
break;
case SEI_PIC_STRUCT_FRAME_TRIPLING:
}
}
-static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth,
+static av_always_inline int dctcoef_get(int16_t *mb, int high_bit_depth,
int index)
{
if (high_bit_depth) {
return AV_RN16A(mb + index);
}
-static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth,
+static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth,
int index, int value)
{
if (high_bit_depth) {
uint8_t *dest_y, int p)
{
MpegEncContext *const s = &h->s;
- void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
- void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
+ void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
+ void (*idct_dc_add)(uint8_t *dst, int16_t *block, int stride);
int i;
int qscale = p == 0 ? s->qscale : h->chroma_qp[p - 1];
block_offset += 16 * p;
uint8_t *dest_y, int p)
{
MpegEncContext *const s = &h->s;
- void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
+ void (*idct_add)(uint8_t *dst, int16_t *block, int stride);
int i;
block_offset += 16 * p;
if (!IS_INTRA4x4(mb_type)) {
memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
- memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
h->sps.chroma_format_idc);
+ ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
+ ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma,
h->sps.chroma_format_idc);
s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
ff_dsputil_init(&s->dsp, s->avctx);
+ ff_videodsp_init(&s->vdsp, h->sps.bit_depth_luma);
} else {
av_log(s->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
h->sps.bit_depth_luma);
return ret;
}
} else {
- if ((ret = ff_MPV_common_init(s) < 0)) {
+ if ((ret = ff_MPV_common_init(s)) < 0) {
av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
return ret;
}
memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
c->h264dsp = h->h264dsp;
+ c->h264qpel = h->h264qpel;
+ c->h264chroma = h->h264chroma;
c->sps = h->sps;
c->pps = h->pps;
c->pixel_shift = h->pixel_shift;
int default_ref_list_done = 0;
int last_pic_structure, last_pic_droppable;
int needs_reinit = 0;
- enum AVPixelFormat pix_fmt;
- /* FIXME: 2tap qpel isn't implemented for high bit depth. */
- if ((s->avctx->flags2 & CODEC_FLAG2_FAST) &&
- !h->nal_ref_idc && !h->pixel_shift) {
- s->me.qpel_put = s->dsp.put_2tap_qpel_pixels_tab;
- s->me.qpel_avg = s->dsp.avg_2tap_qpel_pixels_tab;
- } else {
- s->me.qpel_put = s->dsp.put_h264_qpel_pixels_tab;
- s->me.qpel_avg = s->dsp.avg_h264_qpel_pixels_tab;
- }
+ s->me.qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
+ s->me.qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
first_mb_in_slice = get_ue_golomb(&s->gb);
if (h->pps.sps_id != h->current_sps_id ||
h->context_reinitialized ||
h0->sps_buffers[h->pps.sps_id]->new) {
+ SPS *new_sps = h0->sps_buffers[h->pps.sps_id];
+
h0->sps_buffers[h->pps.sps_id]->new = 0;
+ if (h->sps.chroma_format_idc != new_sps->chroma_format_idc ||
+ h->sps.bit_depth_luma != new_sps->bit_depth_luma)
+ needs_reinit = 1;
+
h->current_sps_id = h->pps.sps_id;
h->sps = *h0->sps_buffers[h->pps.sps_id];
s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
: AVCOL_RANGE_MPEG;
if (h->sps.colour_description_present_flag) {
+ if (s->avctx->colorspace != h->sps.colorspace)
+ needs_reinit = 1;
s->avctx->color_primaries = h->sps.color_primaries;
s->avctx->color_trc = h->sps.color_trc;
s->avctx->colorspace = h->sps.colorspace;
}
}
- ret = get_pixel_format(h);
- if (ret < 0)
- return ret;
- else
- pix_fmt = ret;
- if (s->avctx->pix_fmt == PIX_FMT_NONE)
- s->avctx->pix_fmt = pix_fmt;
-
if (s->context_initialized &&
(s->width != s->avctx->width ||
s->height != s->avctx->height ||
- pix_fmt != s->avctx->pix_fmt ||
needs_reinit ||
av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
return AVERROR_INVALIDDATA;
}
- av_log(h->s.avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
- "pix_fmt: %d\n", s->width, s->height, pix_fmt);
-
flush_change(h);
- s->avctx->pix_fmt = pix_fmt;
+ if ((ret = get_pixel_format(h)) < 0)
+ return ret;
+ s->avctx->pix_fmt = ret;
+
+ av_log(h->s.avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
+ "pix_fmt: %d\n", s->width, s->height, s->avctx->pix_fmt);
if ((ret = h264_slice_header_init(h, 1)) < 0) {
av_log(h->s.avctx, AV_LOG_ERROR,
"Cannot (re-)initialize context during parallel decoding.\n");
return -1;
}
+
+ if ((ret = get_pixel_format(h)) < 0)
+ return ret;
+ s->avctx->pix_fmt = ret;
+
if ((ret = h264_slice_header_init(h, 0)) < 0) {
av_log(h->s.avctx, AV_LOG_ERROR,
"h264_slice_header_init() failed\n");
av_log_ask_for_sample(s->avctx, NULL);
s->picture_structure = last_pic_structure;
s->droppable = last_pic_droppable;
- return AVERROR_INVALIDDATA;
+ return AVERROR_PATCHWELCOME;
}
/* Take ownership of this buffer. Note that if another thread owned
s->current_picture_ptr->frame_num = h->prev_frame_num;
ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 1);
- ff_generate_sliding_window_mmcos(h);
+ if ((ret = ff_generate_sliding_window_mmcos(h, 1)) < 0 &&
+ s->avctx->err_recognition & AV_EF_EXPLODE)
+ return ret;
if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
(s->avctx->err_recognition & AV_EF_EXPLODE))
return AVERROR_INVALIDDATA;
}
}
- if (h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
+ // If frame-mt is enabled, only update mmco tables for the first slice
+ // in a field. Subsequent slices can temporarily clobber h->mmco_index
+ // or h->mmco, which will cause ref list mix-ups and decoding errors
+ // further down the line. This may break decoding if the first slice is
+ // corrupt, thus we only do this if frame-mt is enabled.
+ if (h->nal_ref_idc &&
+ ff_h264_decode_ref_pic_marking(h0, &s->gb,
+ !(s->avctx->active_thread_type & FF_THREAD_FRAME) ||
+ h0->current_slice == 0) < 0 &&
(s->avctx->err_recognition & AV_EF_EXPLODE))
return AVERROR_INVALIDDATA;