Commit bc6e0b64 authored by Ronald S. Bultje's avatar Ronald S. Bultje Committed by Anton Khirnov

vp9: split last/cur_frame from the reference buffers.

We need more information from last/cur_frame than from reference
buffers, so we can use a simplified structure for reference buffers,
and then store mvs and segmentation map information in last/cur.

This prepares the decoder for frame threading support.
Signed-off-by: 's avatarAnton Khirnov <anton@khirnov.net>
parent 04763c6f
......@@ -34,13 +34,77 @@
#define VP9_SYNCCODE 0x498342
#define MAX_PROB 255
static void vp9_frame_unref(AVCodecContext *avctx, VP9Frame *f)
{
ff_thread_release_buffer(avctx, &f->tf);
av_buffer_unref(&f->segmentation_map_buf);
av_buffer_unref(&f->mv_buf);
f->segmentation_map = NULL;
f->mv = NULL;
}
static int vp9_frame_alloc(AVCodecContext *avctx, VP9Frame *f)
{
VP9Context *s = avctx->priv_data;
int ret, sz;
ret = ff_thread_get_buffer(avctx, &f->tf, AV_GET_BUFFER_FLAG_REF);
if (ret < 0)
return ret;
sz = 64 * s->sb_cols * s->sb_rows;
f->segmentation_map_buf = av_buffer_allocz(sz * sizeof(*f->segmentation_map));
f->mv_buf = av_buffer_allocz(sz * sizeof(*f->mv));
if (!f->segmentation_map_buf || !f->mv_buf) {
vp9_frame_unref(avctx, f);
return AVERROR(ENOMEM);
}
f->segmentation_map = f->segmentation_map_buf->data;
f->mv = (VP9MVRefPair*)f->mv_buf->data;
if (s->segmentation.enabled && !s->segmentation.update_map &&
!s->keyframe && !s->intraonly)
memcpy(f->segmentation_map, s->frames[LAST_FRAME].segmentation_map, sz);
return 0;
}
static int vp9_frame_ref(VP9Frame *dst, VP9Frame *src)
{
int ret;
dst->segmentation_map_buf = av_buffer_ref(src->segmentation_map_buf);
dst->mv_buf = av_buffer_ref(src->mv_buf);
if (!dst->segmentation_map_buf || !dst->mv_buf) {
ret = AVERROR(ENOMEM);
goto fail;
}
ret = ff_thread_ref_frame(&dst->tf, &src->tf);
if (ret < 0)
goto fail;
dst->segmentation_map = src->segmentation_map;
dst->mv = src->mv;
return 0;
fail:
av_buffer_unref(&dst->segmentation_map_buf);
av_buffer_unref(&dst->mv_buf);
return ret;
}
static void vp9_decode_flush(AVCodecContext *avctx)
{
VP9Context *s = avctx->priv_data;
int i;
for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++)
vp9_frame_unref(avctx, &s->frames[i]);
for (i = 0; i < FF_ARRAY_ELEMS(s->refs); i++)
av_frame_unref(s->refs[i]);
ff_thread_release_buffer(avctx, &s->refs[i]);
}
static int update_size(AVCodecContext *avctx, int w, int h)
......@@ -66,8 +130,7 @@ static int update_size(AVCodecContext *avctx, int w, int h)
#define assign(var, type, n) var = (type)p; p += s->sb_cols * n * sizeof(*var)
av_free(s->above_partition_ctx);
p = av_malloc(s->sb_cols *
(240 + sizeof(*s->lflvl) + 16 * sizeof(*s->above_mv_ctx) +
64 * s->sb_rows * (1 + sizeof(*s->mv[0]) * 2)));
(240 + sizeof(*s->lflvl) + 16 * sizeof(*s->above_mv_ctx)));
if (!p)
return AVERROR(ENOMEM);
assign(s->above_partition_ctx, uint8_t *, 8);
......@@ -87,9 +150,6 @@ static int update_size(AVCodecContext *avctx, int w, int h)
assign(s->above_filter_ctx, uint8_t *, 8);
assign(s->lflvl, VP9Filter *, 1);
assign(s->above_mv_ctx, VP56mv(*)[2], 16);
assign(s->segmentation_map, uint8_t *, 64 * s->sb_rows);
assign(s->mv[0], VP9MVRefPair *, 64 * s->sb_rows);
assign(s->mv[1], VP9MVRefPair *, 64 * s->sb_rows);
#undef assign
return 0;
......@@ -268,22 +328,22 @@ static int decode_frame_header(AVCodecContext *avctx,
s->signbias[1] = get_bits1(&s->gb);
s->refidx[2] = get_bits(&s->gb, 3);
s->signbias[2] = get_bits1(&s->gb);
if (!s->refs[s->refidx[0]]->buf[0] ||
!s->refs[s->refidx[1]]->buf[0] ||
!s->refs[s->refidx[2]]->buf[0]) {
if (!s->refs[s->refidx[0]].f->buf[0] ||
!s->refs[s->refidx[1]].f->buf[0] ||
!s->refs[s->refidx[2]].f->buf[0]) {
av_log(avctx, AV_LOG_ERROR,
"Not all references are available\n");
return AVERROR_INVALIDDATA;
}
if (get_bits1(&s->gb)) {
w = s->refs[s->refidx[0]]->width;
h = s->refs[s->refidx[0]]->height;
w = s->refs[s->refidx[0]].f->width;
h = s->refs[s->refidx[0]].f->height;
} else if (get_bits1(&s->gb)) {
w = s->refs[s->refidx[1]]->width;
h = s->refs[s->refidx[1]]->height;
w = s->refs[s->refidx[1]].f->width;
h = s->refs[s->refidx[1]].f->height;
} else if (get_bits1(&s->gb)) {
w = s->refs[s->refidx[2]]->width;
h = s->refs[s->refidx[2]]->height;
w = s->refs[s->refidx[2]].f->width;
h = s->refs[s->refidx[2]].f->height;
} else {
w = get_bits(&s->gb, 16) + 1;
h = get_bits(&s->gb, 16) + 1;
......@@ -679,6 +739,7 @@ static int decode_subblock(AVCodecContext *avctx, int row, int col,
ptrdiff_t yoff, ptrdiff_t uvoff, enum BlockLevel bl)
{
VP9Context *s = avctx->priv_data;
AVFrame *f = s->frames[CUR_FRAME].tf.f;
int c = ((s->above_partition_ctx[col] >> (3 - bl)) & 1) |
(((s->left_partition_ctx[row & 0x7] >> (3 - bl)) & 1) << 1);
int ret;
......@@ -702,8 +763,8 @@ static int decode_subblock(AVCodecContext *avctx, int row, int col,
ret = ff_vp9_decode_block(avctx, row, col, lflvl, yoff, uvoff,
bl, bp);
if (!ret) {
yoff += hbs * 8 * s->cur_frame->linesize[0];
uvoff += hbs * 4 * s->cur_frame->linesize[1];
yoff += hbs * 8 * f->linesize[0];
uvoff += hbs * 4 * f->linesize[1];
ret = ff_vp9_decode_block(avctx, row + hbs, col, lflvl,
yoff, uvoff, bl, bp);
}
......@@ -726,8 +787,8 @@ static int decode_subblock(AVCodecContext *avctx, int row, int col,
yoff + 8 * hbs, uvoff + 4 * hbs,
bl + 1);
if (!ret) {
yoff += hbs * 8 * s->cur_frame->linesize[0];
uvoff += hbs * 4 * s->cur_frame->linesize[1];
yoff += hbs * 8 * f->linesize[0];
uvoff += hbs * 4 * f->linesize[1];
ret = decode_subblock(avctx, row + hbs, col, lflvl,
yoff, uvoff, bl + 1);
if (!ret) {
......@@ -758,8 +819,8 @@ static int decode_subblock(AVCodecContext *avctx, int row, int col,
bp = PARTITION_SPLIT;
ret = decode_subblock(avctx, row, col, lflvl, yoff, uvoff, bl + 1);
if (!ret) {
yoff += hbs * 8 * s->cur_frame->linesize[0];
uvoff += hbs * 4 * s->cur_frame->linesize[1];
yoff += hbs * 8 * f->linesize[0];
uvoff += hbs * 4 * f->linesize[1];
ret = decode_subblock(avctx, row + hbs, col, lflvl,
yoff, uvoff, bl + 1);
}
......@@ -782,8 +843,10 @@ static void loopfilter_subblock(AVCodecContext *avctx, VP9Filter *lflvl,
ptrdiff_t yoff, ptrdiff_t uvoff)
{
VP9Context *s = avctx->priv_data;
uint8_t *dst = s->cur_frame->data[0] + yoff, *lvl = lflvl->level;
ptrdiff_t ls_y = s->cur_frame->linesize[0], ls_uv = s->cur_frame->linesize[1];
AVFrame *f = s->frames[CUR_FRAME].tf.f;
uint8_t *dst = f->data[0] + yoff;
ptrdiff_t ls_y = f->linesize[0], ls_uv = f->linesize[1];
uint8_t *lvl = lflvl->level;
int y, x, p;
/* FIXME: In how far can we interleave the v/h loopfilter calls? E.g.
......@@ -860,7 +923,7 @@ static void loopfilter_subblock(AVCodecContext *avctx, VP9Filter *lflvl,
// block1
// filter edges between rows, Y plane (e.g. ------)
// block2
dst = s->cur_frame->data[0] + yoff;
dst = f->data[0] + yoff;
lvl = lflvl->level;
for (y = 0; y < 8; y++, dst += 8 * ls_y, lvl += 8) {
uint8_t *ptr = dst, *l = lvl, *vmask = lflvl->mask[0][1][y];
......@@ -924,7 +987,7 @@ static void loopfilter_subblock(AVCodecContext *avctx, VP9Filter *lflvl,
// same principle but for U/V planes
for (p = 0; p < 2; p++) {
lvl = lflvl->level;
dst = s->cur_frame->data[1 + p] + uvoff;
dst = f->data[1 + p] + uvoff;
for (y = 0; y < 8; y += 4, dst += 16 * ls_uv, lvl += 32) {
uint8_t *ptr = dst, *l = lvl, *hmask1 = lflvl->mask[1][0][y];
uint8_t *hmask2 = lflvl->mask[1][0][y + 2];
......@@ -971,7 +1034,7 @@ static void loopfilter_subblock(AVCodecContext *avctx, VP9Filter *lflvl,
}
}
lvl = lflvl->level;
dst = s->cur_frame->data[1 + p] + uvoff;
dst = f->data[1 + p] + uvoff;
for (y = 0; y < 8; y++, dst += 4 * ls_uv) {
uint8_t *ptr = dst, *l = lvl, *vmask = lflvl->mask[1][1][y];
unsigned vm = vmask[0] | vmask[1] | vmask[2];
......@@ -1030,6 +1093,7 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame,
int *got_frame, const uint8_t *data, int size)
{
VP9Context *s = avctx->priv_data;
AVFrame *f;
int ret, tile_row, tile_col, i, ref = -1, row, col;
ptrdiff_t yoff = 0, uvoff = 0;
......@@ -1037,13 +1101,13 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame,
if (ret < 0) {
return ret;
} else if (!ret) {
if (!s->refs[ref]->buf[0]) {
if (!s->refs[ref].f->buf[0]) {
av_log(avctx, AV_LOG_ERROR,
"Requested reference %d not available\n", ref);
return AVERROR_INVALIDDATA;
}
ret = av_frame_ref(frame, s->refs[ref]);
ret = av_frame_ref(frame, s->refs[ref].f);
if (ret < 0)
return ret;
*got_frame = 1;
......@@ -1052,15 +1116,21 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame,
data += ret;
size -= ret;
s->cur_frame = frame;
vp9_frame_unref(avctx, &s->frames[LAST_FRAME]);
if (!s->keyframe && s->frames[CUR_FRAME].tf.f->buf[0]) {
ret = vp9_frame_ref(&s->frames[LAST_FRAME], &s->frames[CUR_FRAME]);
if (ret < 0)
return ret;
}
av_frame_unref(s->cur_frame);
if ((ret = ff_get_buffer(avctx, s->cur_frame,
s->refreshrefmask ? AV_GET_BUFFER_FLAG_REF : 0)) < 0)
vp9_frame_unref(avctx, &s->frames[CUR_FRAME]);
ret = vp9_frame_alloc(avctx, &s->frames[CUR_FRAME]);
if (ret < 0)
return ret;
s->cur_frame->key_frame = s->keyframe;
s->cur_frame->pict_type = s->keyframe ? AV_PICTURE_TYPE_I
: AV_PICTURE_TYPE_P;
f = s->frames[CUR_FRAME].tf.f;
f->key_frame = s->keyframe;
f->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
if (s->fullrange)
avctx->color_range = AVCOL_RANGE_JPEG;
......@@ -1110,8 +1180,8 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame,
for (row = s->tiling.tile_row_start;
row < s->tiling.tile_row_end;
row += 8, yoff += s->cur_frame->linesize[0] * 64,
uvoff += s->cur_frame->linesize[1] * 32) {
row += 8, yoff += f->linesize[0] * 64,
uvoff += f->linesize[1] * 32) {
VP9Filter *lflvl = s->lflvl;
ptrdiff_t yoff2 = yoff, uvoff2 = uvoff;
......@@ -1149,16 +1219,16 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame,
// prediction of next row of sb64s
if (row + 8 < s->rows) {
memcpy(s->intra_pred_data[0],
s->cur_frame->data[0] + yoff +
63 * s->cur_frame->linesize[0],
f->data[0] + yoff +
63 * f->linesize[0],
8 * s->cols);
memcpy(s->intra_pred_data[1],
s->cur_frame->data[1] + uvoff +
31 * s->cur_frame->linesize[1],
f->data[1] + uvoff +
31 * f->linesize[1],
4 * s->cols);
memcpy(s->intra_pred_data[2],
s->cur_frame->data[2] + uvoff +
31 * s->cur_frame->linesize[2],
f->data[2] + uvoff +
31 * f->linesize[2],
4 * s->cols);
}
......@@ -1194,21 +1264,23 @@ static int vp9_decode_frame(AVCodecContext *avctx, AVFrame *frame,
ff_vp9_adapt_probs(s);
}
}
FFSWAP(VP9MVRefPair *, s->mv[0], s->mv[1]);
// ref frame setup
for (i = 0; i < 8; i++)
if (s->refreshrefmask & (1 << i)) {
av_frame_unref(s->refs[i]);
ret = av_frame_ref(s->refs[i], s->cur_frame);
ff_thread_release_buffer(avctx, &s->refs[i]);
ret = ff_thread_ref_frame(&s->refs[i], &s->frames[CUR_FRAME].tf);
if (ret < 0)
return ret;
}
if (s->invisible)
av_frame_unref(s->cur_frame);
else
if (!s->invisible) {
av_frame_unref(frame);
ret = av_frame_ref(frame, s->frames[CUR_FRAME].tf.f);
if (ret < 0)
return ret;
*got_frame = 1;
}
return 0;
}
......@@ -1267,8 +1339,15 @@ static av_cold int vp9_decode_free(AVCodecContext *avctx)
VP9Context *s = avctx->priv_data;
int i;
for (i = 0; i < FF_ARRAY_ELEMS(s->refs); i++)
av_frame_free(&s->refs[i]);
for (i = 0; i < FF_ARRAY_ELEMS(s->frames); i++) {
vp9_frame_unref(avctx, &s->frames[i]);
av_frame_free(&s->frames[i].tf.f);
}
for (i = 0; i < FF_ARRAY_ELEMS(s->refs); i++) {
ff_thread_release_buffer(avctx, &s->refs[i]);
av_frame_free(&s->refs[i].f);
}
av_freep(&s->c_b);
av_freep(&s->above_partition_ctx);
......@@ -1286,17 +1365,23 @@ static av_cold int vp9_decode_init(AVCodecContext *avctx)
ff_vp9dsp_init(&s->dsp);
ff_videodsp_init(&s->vdsp, 8);
s->frames[0].tf.f = av_frame_alloc();
s->frames[1].tf.f = av_frame_alloc();
if (!s->frames[0].tf.f || !s->frames[1].tf.f)
goto fail;
for (i = 0; i < FF_ARRAY_ELEMS(s->refs); i++) {
s->refs[i] = av_frame_alloc();
if (!s->refs[i]) {
vp9_decode_free(avctx);
return AVERROR(ENOMEM);
}
s->refs[i].f = av_frame_alloc();
if (!s->refs[i].f)
goto fail;
}
s->filter.sharpness = -1;
return 0;
fail:
vp9_decode_free(avctx);
return AVERROR(ENOMEM);
}
AVCodec ff_vp9_decoder = {
......
......@@ -27,9 +27,11 @@
#include <stddef.h>
#include <stdint.h>
#include "libavutil/buffer.h"
#include "libavutil/internal.h"
#include "avcodec.h"
#include "thread.h"
#include "vp56.h"
enum TxfmMode {
......@@ -225,6 +227,16 @@ typedef struct VP9Filter {
[8 /* rows */][4 /* 0=16, 1=8, 2=4, 3=inner4 */];
} VP9Filter;
typedef struct VP9Frame {
ThreadFrame tf;
uint8_t *segmentation_map;
VP9MVRefPair *mv;
AVBufferRef *segmentation_map_buf;
AVBufferRef *mv_buf;
} VP9Frame;
enum BlockLevel {
BL_64X64,
BL_32X32,
......@@ -293,8 +305,12 @@ typedef struct VP9Context {
uint8_t refidx[3];
uint8_t signbias[3];
uint8_t varcompref[2];
AVFrame *refs[8];
AVFrame *cur_frame;
ThreadFrame refs[8];
#define CUR_FRAME 0
#define LAST_FRAME 1
VP9Frame frames[2];
struct {
uint8_t level;
......@@ -392,8 +408,6 @@ typedef struct VP9Context {
// whole-frame cache
uint8_t *intra_pred_data[3];
uint8_t *segmentation_map;
VP9MVRefPair *mv[2];
VP9Filter *lflvl;
DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[71 * 80];
......
......@@ -70,13 +70,14 @@ static void decode_mode(VP9Context *s, VP9Block *const b)
vp56_rac_get_prob_branchy(&s->c,
s->prob.segpred[s->above_segpred_ctx[col] +
s->left_segpred_ctx[row7]]))) {
uint8_t *refsegmap = s->frames[LAST_FRAME].segmentation_map;
int pred = MAX_SEGMENT - 1;
int x;
for (y = 0; y < h4; y++)
for (x = 0; x < w4; x++)
pred = FFMIN(pred,
s->segmentation_map[(y + row) * 8 * s->sb_cols + x + col]);
refsegmap[(y + row) * 8 * s->sb_cols + x + col]);
b->seg_id = pred;
memset(&s->above_segpred_ctx[col], 1, w4);
......@@ -89,8 +90,10 @@ static void decode_mode(VP9Context *s, VP9Block *const b)
memset(&s->left_segpred_ctx[row7], 0, h4);
}
if ((s->segmentation.enabled && s->segmentation.update_map) || s->keyframe) {
uint8_t *segmap = s->frames[CUR_FRAME].segmentation_map;
for (y = 0; y < h4; y++)
memset(&s->segmentation_map[(y + row) * 8 * s->sb_cols + col],
memset(&segmap[(y + row) * 8 * s->sb_cols + col],
b->seg_id, w4);
}
......@@ -684,24 +687,25 @@ static void decode_mode(VP9Context *s, VP9Block *const b)
// FIXME kinda ugly
for (y = 0; y < h4; y++) {
int x, o = (row + y) * s->sb_cols * 8 + col;
VP9MVRefPair *mv = &s->frames[CUR_FRAME].mv[o];
if (b->intra) {
for (x = 0; x < w4; x++) {
s->mv[0][o + x].ref[0] =
s->mv[0][o + x].ref[1] = -1;
mv[x].ref[0] =
mv[x].ref[1] = -1;
}
} else if (b->comp) {
for (x = 0; x < w4; x++) {
s->mv[0][o + x].ref[0] = b->ref[0];
s->mv[0][o + x].ref[1] = b->ref[1];
AV_COPY32(&s->mv[0][o + x].mv[0], &b->mv[3][0]);
AV_COPY32(&s->mv[0][o + x].mv[1], &b->mv[3][1]);
mv[x].ref[0] = b->ref[0];
mv[x].ref[1] = b->ref[1];
AV_COPY32(&mv[x].mv[0], &b->mv[3][0]);
AV_COPY32(&mv[x].mv[1], &b->mv[3][1]);
}
} else {
for (x = 0; x < w4; x++) {
s->mv[0][o + x].ref[0] = b->ref[0];
s->mv[0][o + x].ref[1] = -1;
AV_COPY32(&s->mv[0][o + x].mv[0], &b->mv[3][0]);
mv[x].ref[0] = b->ref[0];
mv[x].ref[1] = -1;
AV_COPY32(&mv[x].mv[0], &b->mv[3][0]);
}
}
}
......@@ -1071,6 +1075,7 @@ static void intra_recon(AVCodecContext *avctx, ptrdiff_t y_off, ptrdiff_t uv_off
{
VP9Context *s = avctx->priv_data;
VP9Block *const b = &s->b;
AVFrame *f = s->frames[CUR_FRAME].tf.f;
int row = b->row, col = b->col;
int w4 = bwh_tab[1][b->bs][0] << 1, step1d = 1 << b->tx, n;
int h4 = bwh_tab[1][b->bs][1] << 1, x, y, step = 1 << (b->tx * 2);
......@@ -1078,7 +1083,7 @@ static void intra_recon(AVCodecContext *avctx, ptrdiff_t y_off, ptrdiff_t uv_off
int end_y = FFMIN(2 * (s->rows - row), h4);
int tx = 4 * s->lossless + b->tx, uvtx = b->uvtx + 4 * s->lossless;
int uvstep1d = 1 << b->uvtx, p;
uint8_t *dst = b->dst[0], *dst_r = s->cur_frame->data[0] + y_off;
uint8_t *dst = b->dst[0], *dst_r = f->data[0] + y_off;
for (n = 0, y = 0; y < end_y; y += step1d) {
uint8_t *ptr = dst, *ptr_r = dst_r;
......@@ -1092,7 +1097,7 @@ static void intra_recon(AVCodecContext *avctx, ptrdiff_t y_off, ptrdiff_t uv_off
int eob = b->tx > TX_8X8 ? AV_RN16A(&s->eob[n]) : s->eob[n];
mode = check_intra_mode(s, mode, &a, ptr_r,
s->cur_frame->linesize[0],
f->linesize[0],
ptr, b->y_stride, l,
col, x, w4, row, y, b->tx, 0);
s->dsp.intra_pred[b->tx][mode](ptr, b->y_stride, l, a);
......@@ -1100,7 +1105,7 @@ static void intra_recon(AVCodecContext *avctx, ptrdiff_t y_off, ptrdiff_t uv_off
s->dsp.itxfm_add[tx][txtp](ptr, b->y_stride,
s->block + 16 * n, eob);
}
dst_r += 4 * s->cur_frame->linesize[0] * step1d;
dst_r += 4 * f->linesize[0] * step1d;
dst += 4 * b->y_stride * step1d;
}
......@@ -1112,7 +1117,7 @@ static void intra_recon(AVCodecContext *avctx, ptrdiff_t y_off, ptrdiff_t uv_off
step = 1 << (b->uvtx * 2);
for (p = 0; p < 2; p++) {
dst = b->dst[1 + p];
dst_r = s->cur_frame->data[1 + p] + uv_off;
dst_r = f->data[1 + p] + uv_off;
for (n = 0, y = 0; y < end_y; y += uvstep1d) {
uint8_t *ptr = dst, *ptr_r = dst_r;
for (x = 0; x < end_x;
......@@ -1125,7 +1130,7 @@ static void intra_recon(AVCodecContext *avctx, ptrdiff_t y_off, ptrdiff_t uv_off
: s->uveob[p][n];
mode = check_intra_mode(s, mode, &a, ptr_r,
s->cur_frame->linesize[1],
f->linesize[1],
ptr, b->uv_stride, l,
col, x, w4, row, y, b->uvtx, p + 1);
s->dsp.intra_pred[b->uvtx][mode](ptr, b->uv_stride, l, a);
......@@ -1134,7 +1139,7 @@ static void intra_recon(AVCodecContext *avctx, ptrdiff_t y_off, ptrdiff_t uv_off
s->uvblock[p] + 16 * n,
eob);
}
dst_r += 4 * uvstep1d * s->cur_frame->linesize[1];
dst_r += 4 * uvstep1d * f->linesize[1];
dst += 4 * uvstep1d * b->uv_stride;
}
}
......@@ -1224,8 +1229,12 @@ static int inter_recon(AVCodecContext *avctx)
VP9Context *s = avctx->priv_data;
VP9Block *const b = &s->b;
int row = b->row, col = b->col;
AVFrame *ref1 = s->refs[s->refidx[b->ref[0]]];
AVFrame *ref2 = b->comp ? s->refs[s->refidx[b->ref[1]]] : NULL;
ThreadFrame *tref1 = &s->refs[s->refidx[b->ref[0]]];
ThreadFrame *tref2 = b->comp ? &s->refs[s->refidx[b->ref[1]]] : NULL;
AVFrame *ref1 = tref1->f;
AVFrame *ref2 = tref2 ? tref2->f : NULL;
int w = avctx->width, h = avctx->height;
ptrdiff_t ls_y = b->y_stride, ls_uv = b->uv_stride;
......@@ -1547,6 +1556,7 @@ int ff_vp9_decode_block(AVCodecContext *avctx, int row, int col,
{
VP9Context *s = avctx->priv_data;
VP9Block *const b = &s->b;
AVFrame *f = s->frames[CUR_FRAME].tf.f;
enum BlockSize bs = bl * 3 + bp;
int ret, y, w4 = bwh_tab[1][bs][0], h4 = bwh_tab[1][bs][1], lvl;
int emu[2];
......@@ -1582,25 +1592,25 @@ int ff_vp9_decode_block(AVCodecContext *avctx, int row, int col,
/* Emulated overhangs if the stride of the target buffer can't hold.
* This allows to support emu-edge and so on even if we have large
* block overhangs. */
emu[0] = (col + w4) * 8 > s->cur_frame->linesize[0] ||
emu[0] = (col + w4) * 8 > f->linesize[0] ||
(row + h4) > s->rows;
emu[1] = (col + w4) * 4 > s->cur_frame->linesize[1] ||
emu[1] = (col + w4) * 4 > f->linesize[1] ||
(row + h4) > s->rows;
if (emu[0]) {
b->dst[0] = s->tmp_y;
b->y_stride = 64;
} else {
b->dst[0] = s->cur_frame->data[0] + yoff;
b->y_stride = s->cur_frame->linesize[0];
b->dst[0] = f->data[0] + yoff;
b->y_stride = f->linesize[0];
}
if (emu[1]) {
b->dst[1] = s->tmp_uv[0];
b->dst[2] = s->tmp_uv[1];
b->uv_stride = 32;
} else {
b->dst[1] = s->cur_frame->data[1] + uvoff;
b->dst[2] = s->cur_frame->data[2] + uvoff;
b->uv_stride = s->cur_frame->linesize[1];
b->dst[1] = f->data[1] + uvoff;
b->dst[2] = f->data[2] + uvoff;
b->uv_stride = f->linesize[1];
}
if (b->intra) {
intra_recon(avctx, yoff, uvoff);
......@@ -1618,9 +1628,9 @@ int ff_vp9_decode_block(AVCodecContext *avctx, int row, int col,
av_assert2(n <= 4);
if (w & bw) {
s->dsp.mc[n][0][0][0][0](s->cur_frame->data[0] + yoff + o,
s->dsp.mc[n][0][0][0][0](f->data[0] + yoff + o,
s->tmp_y + o,
s->cur_frame->linesize[0],
f->linesize[0],
64, h, 0, 0);
o += bw;
}
......@@ -1636,13 +1646,13 @@ int ff_vp9_decode_block(AVCodecContext *avctx, int row, int col,
av_assert2(n <= 4);
if (w & bw) {
s->dsp.mc[n][0][0][0][0](s->cur_frame->data[1] + uvoff + o,
s->dsp.mc[n][0][0][0][0](f->data[1] + uvoff + o,
s->tmp_uv[0] + o,
s->cur_frame->linesize[1],
f->linesize[1],
32, h, 0, 0);
s->dsp.mc[n][0][0][0][0](s->cur_frame->data[2] + uvoff + o,
s->dsp.mc[n][0][0][0][0](f->data[2] + uvoff + o,
s->tmp_uv[1] + o,
s->cur_frame->linesize[2],
f->linesize[2],
32, h, 0, 0);
o += bw;
}
......
......@@ -125,7 +125,7 @@ static void find_ref_mvs(VP9Context *s,
} while (0)
if (row > 0) {
VP9MVRefPair *mv = &s->mv[0][(row - 1) * s->sb_cols * 8 + col];
VP9MVRefPair *mv = &s->frames[CUR_FRAME].mv[(row - 1) * s->sb_cols * 8 + col];
if (mv->ref[0] == ref)
RETURN_MV(s->above_mv_ctx[2 * col + (sb & 1)][0]);
......@@ -133,7 +133,7 @@ static void find_ref_mvs(VP9Context *s,
RETURN_MV(s->above_mv_ctx[2 * col + (sb & 1)][1]);
}
if (col > s->tiling.tile_col_start) {
VP9MVRefPair *mv = &s->mv[0][row * s->sb_cols * 8 + col - 1];
VP9MVRefPair *mv = &s->frames[CUR_FRAME].mv[row * s->sb_cols * 8 + col - 1];
if (mv->ref[0] == ref)
RETURN_MV(s->left_mv_ctx[2 * row7 + (sb >> 1)][0]);
......@@ -151,7 +151,7 @@ static void find_ref_mvs(VP9Context *s,
if (c >= s->tiling.tile_col_start && c < s->cols &&
r >= 0 && r < s->rows) {
VP9MVRefPair *mv = &s->mv[0][r * s->sb_cols * 8 + c];
VP9MVRefPair *mv = &s->frames[CUR_FRAME].mv[r * s->sb_cols * 8 + c];
if (mv->ref[0] == ref)
RETURN_MV(mv->mv[0]);
......@@ -162,7 +162,7 @@ static void find_ref_mvs(VP9Context *s,
// MV at this position in previous frame, using same reference frame
if (s->use_last_frame_mvs) {
VP9MVRefPair *mv = &s->mv[1][row * s->sb_cols * 8 + col];
VP9MVRefPair *mv = &s->frames[LAST_FRAME].mv[row * s->sb_cols * 8 + col];
if (mv->ref[0] == ref)
RETURN_MV(mv->mv[0]);
......@@ -186,7 +186,7 @@ static void find_ref_mvs(VP9Context *s,
if (c >= s->tiling.tile_col_start && c < s->cols &&
r >= 0 && r < s->rows) {
VP9MVRefPair *mv = &s->mv[0][r * s->sb_cols * 8 + c];
VP9MVRefPair *mv = &s->frames[CUR_FRAME].mv[r * s->sb_cols * 8 + c];
if (mv->ref[0] != ref && mv->ref[0] >= 0)
RETURN_SCALE_MV(mv->mv[0],
......@@ -203,7 +203,7 @@ static void find_ref_mvs(VP9Context *s,
// MV at this position in previous frame, using different reference frame
if (s->use_last_frame_mvs) {
VP9MVRefPair *mv = &s->mv[1][row * s->sb_cols * 8 + col];
VP9MVRefPair *mv = &s->frames[LAST_FRAME].mv[row * s->sb_cols * 8 + col];
if (mv->ref[0] != ref && mv->ref[0] >= 0)
RETURN_SCALE_MV(mv->mv[0],
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment