h264: move [uv]linesize to the per-slice context

While it is a per-frame variable, it is only really used in the
low-level decoding code, so it is more efficient to store it in the
slice context.
pull/125/head
Anton Khirnov 10 years ago
parent b53569e068
commit c28ed1d743
  1. 2
      libavcodec/h264.h
  2. 8
      libavcodec/h264_mb.c
  3. 36
      libavcodec/h264_mb_template.c
  4. 43
      libavcodec/h264_slice.c
  5. 32
      libavcodec/svq3.c

@ -366,6 +366,7 @@ typedef struct H264SliceContext {
unsigned int topright_samples_available; unsigned int topright_samples_available;
unsigned int left_samples_available; unsigned int left_samples_available;
ptrdiff_t linesize, uvlinesize;
ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
ptrdiff_t mb_uvlinesize; ptrdiff_t mb_uvlinesize;
@ -478,7 +479,6 @@ typedef struct H264Context {
/* coded dimensions -- 16 * mb w/h */ /* coded dimensions -- 16 * mb w/h */
int width, height; int width, height;
ptrdiff_t linesize, uvlinesize;
int chroma_x_shift, chroma_y_shift; int chroma_x_shift, chroma_y_shift;
int droppable; int droppable;

@ -487,13 +487,13 @@ static av_always_inline void prefetch_motion(const H264Context *h, H264SliceCont
int off = (mx << pixel_shift) + int off = (mx << pixel_shift) +
(my + (sl->mb_x & 3) * 4) * sl->mb_linesize + (my + (sl->mb_x & 3) * 4) * sl->mb_linesize +
(64 << pixel_shift); (64 << pixel_shift);
h->vdsp.prefetch(src[0] + off, h->linesize, 4); h->vdsp.prefetch(src[0] + off, sl->linesize, 4);
if (chroma_idc == 3 /* yuv444 */) { if (chroma_idc == 3 /* yuv444 */) {
h->vdsp.prefetch(src[1] + off, h->linesize, 4); h->vdsp.prefetch(src[1] + off, sl->linesize, 4);
h->vdsp.prefetch(src[2] + off, h->linesize, 4); h->vdsp.prefetch(src[2] + off, sl->linesize, 4);
} else { } else {
off = ((mx >> 1) << pixel_shift) + off = ((mx >> 1) << pixel_shift) +
((my >> 1) + (sl->mb_x & 7)) * h->uvlinesize + ((my >> 1) + (sl->mb_x & 7)) * sl->uvlinesize +
(64 << pixel_shift); (64 << pixel_shift);
h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2); h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
} }

@ -57,23 +57,23 @@ static av_noinline void FUNC(hl_decode_mb)(const H264Context *h, H264SliceContex
const int block_h = 16 >> h->chroma_y_shift; const int block_h = 16 >> h->chroma_y_shift;
const int chroma422 = CHROMA422(h); const int chroma422 = CHROMA422(h);
dest_y = h->cur_pic.f.data[0] + ((mb_x << PIXEL_SHIFT) + mb_y * h->linesize) * 16; dest_y = h->cur_pic.f.data[0] + ((mb_x << PIXEL_SHIFT) + mb_y * sl->linesize) * 16;
dest_cb = h->cur_pic.f.data[1] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * h->uvlinesize * block_h; dest_cb = h->cur_pic.f.data[1] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * sl->uvlinesize * block_h;
dest_cr = h->cur_pic.f.data[2] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * h->uvlinesize * block_h; dest_cr = h->cur_pic.f.data[2] + (mb_x << PIXEL_SHIFT) * 8 + mb_y * sl->uvlinesize * block_h;
h->vdsp.prefetch(dest_y + (sl->mb_x & 3) * 4 * h->linesize + (64 << PIXEL_SHIFT), h->linesize, 4); h->vdsp.prefetch(dest_y + (sl->mb_x & 3) * 4 * sl->linesize + (64 << PIXEL_SHIFT), sl->linesize, 4);
h->vdsp.prefetch(dest_cb + (sl->mb_x & 7) * h->uvlinesize + (64 << PIXEL_SHIFT), dest_cr - dest_cb, 2); h->vdsp.prefetch(dest_cb + (sl->mb_x & 7) * sl->uvlinesize + (64 << PIXEL_SHIFT), dest_cr - dest_cb, 2);
h->list_counts[mb_xy] = sl->list_count; h->list_counts[mb_xy] = sl->list_count;
if (!SIMPLE && MB_FIELD(sl)) { if (!SIMPLE && MB_FIELD(sl)) {
linesize = sl->mb_linesize = h->linesize * 2; linesize = sl->mb_linesize = sl->linesize * 2;
uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2; uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
block_offset = &h->block_offset[48]; block_offset = &h->block_offset[48];
if (mb_y & 1) { // FIXME move out of this function? if (mb_y & 1) { // FIXME move out of this function?
dest_y -= h->linesize * 15; dest_y -= sl->linesize * 15;
dest_cb -= h->uvlinesize * (block_h - 1); dest_cb -= sl->uvlinesize * (block_h - 1);
dest_cr -= h->uvlinesize * (block_h - 1); dest_cr -= sl->uvlinesize * (block_h - 1);
} }
if (FRAME_MBAFF(h)) { if (FRAME_MBAFF(h)) {
int list; int list;
@ -94,8 +94,8 @@ static av_noinline void FUNC(hl_decode_mb)(const H264Context *h, H264SliceContex
} }
} }
} else { } else {
linesize = sl->mb_linesize = h->linesize; linesize = sl->mb_linesize = sl->linesize;
uvlinesize = sl->mb_uvlinesize = h->uvlinesize; uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
// dct_offset = s->linesize * 16; // dct_offset = s->linesize * 16;
} }
@ -287,19 +287,19 @@ static av_noinline void FUNC(hl_decode_mb_444)(const H264Context *h, H264SliceCo
for (p = 0; p < plane_count; p++) { for (p = 0; p < plane_count; p++) {
dest[p] = h->cur_pic.f.data[p] + dest[p] = h->cur_pic.f.data[p] +
((mb_x << PIXEL_SHIFT) + mb_y * h->linesize) * 16; ((mb_x << PIXEL_SHIFT) + mb_y * sl->linesize) * 16;
h->vdsp.prefetch(dest[p] + (sl->mb_x & 3) * 4 * h->linesize + (64 << PIXEL_SHIFT), h->vdsp.prefetch(dest[p] + (sl->mb_x & 3) * 4 * sl->linesize + (64 << PIXEL_SHIFT),
h->linesize, 4); sl->linesize, 4);
} }
h->list_counts[mb_xy] = sl->list_count; h->list_counts[mb_xy] = sl->list_count;
if (!SIMPLE && MB_FIELD(sl)) { if (!SIMPLE && MB_FIELD(sl)) {
linesize = sl->mb_linesize = sl->mb_uvlinesize = h->linesize * 2; linesize = sl->mb_linesize = sl->mb_uvlinesize = sl->linesize * 2;
block_offset = &h->block_offset[48]; block_offset = &h->block_offset[48];
if (mb_y & 1) // FIXME move out of this function? if (mb_y & 1) // FIXME move out of this function?
for (p = 0; p < 3; p++) for (p = 0; p < 3; p++)
dest[p] -= h->linesize * 15; dest[p] -= sl->linesize * 15;
if (FRAME_MBAFF(h)) { if (FRAME_MBAFF(h)) {
int list; int list;
for (list = 0; list < sl->list_count; list++) { for (list = 0; list < sl->list_count; list++) {
@ -319,7 +319,7 @@ static av_noinline void FUNC(hl_decode_mb_444)(const H264Context *h, H264SliceCo
} }
} }
} else { } else {
linesize = sl->mb_linesize = sl->mb_uvlinesize = h->linesize; linesize = sl->mb_linesize = sl->mb_uvlinesize = sl->linesize;
} }
if (!SIMPLE && IS_INTRA_PCM(mb_type)) { if (!SIMPLE && IS_INTRA_PCM(mb_type)) {

@ -228,9 +228,6 @@ static int alloc_picture(H264Context *h, H264Picture *pic)
if (ret < 0) if (ret < 0)
goto fail; goto fail;
h->linesize = pic->f.linesize[0];
h->uvlinesize = pic->f.linesize[1];
if (h->avctx->hwaccel) { if (h->avctx->hwaccel) {
const AVHWAccel *hwaccel = h->avctx->hwaccel; const AVHWAccel *hwaccel = h->avctx->hwaccel;
av_assert0(!pic->hwaccel_picture_private); av_assert0(!pic->hwaccel_picture_private);
@ -458,11 +455,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
} }
context_reinitialized = 1; context_reinitialized = 1;
/* update linesize on resize. The decoder doesn't
* necessarily call h264_frame_start in the new thread */
h->linesize = h1->linesize;
h->uvlinesize = h1->uvlinesize;
/* copy block_offset since frame_start may not be called */ /* copy block_offset since frame_start may not be called */
memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset)); memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
} }
@ -644,17 +636,15 @@ static int h264_frame_start(H264Context *h)
if (CONFIG_ERROR_RESILIENCE) if (CONFIG_ERROR_RESILIENCE)
ff_er_frame_start(&h->slice_ctx[0].er); ff_er_frame_start(&h->slice_ctx[0].er);
assert(h->linesize && h->uvlinesize);
for (i = 0; i < 16; i++) { for (i = 0; i < 16; i++) {
h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3); h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f.linesize[0] * ((scan8[i] - scan8[0]) >> 3);
h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3); h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f.linesize[0] * ((scan8[i] - scan8[0]) >> 3);
} }
for (i = 0; i < 16; i++) { for (i = 0; i < 16; i++) {
h->block_offset[16 + i] = h->block_offset[16 + i] =
h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f.linesize[1] * ((scan8[i] - scan8[0]) >> 3);
h->block_offset[48 + 16 + i] = h->block_offset[48 + 16 + i] =
h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f.linesize[1] * ((scan8[i] - scan8[0]) >> 3);
} }
/* Some macroblocks can be accessed before they're available in case /* Some macroblocks can be accessed before they're available in case
@ -1976,26 +1966,26 @@ static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x,
sl->mb_x = mb_x; sl->mb_x = mb_x;
sl->mb_y = mb_y; sl->mb_y = mb_y;
dest_y = h->cur_pic.f.data[0] + dest_y = h->cur_pic.f.data[0] +
((mb_x << pixel_shift) + mb_y * h->linesize) * 16; ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
dest_cb = h->cur_pic.f.data[1] + dest_cb = h->cur_pic.f.data[1] +
(mb_x << pixel_shift) * (8 << CHROMA444(h)) + (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
mb_y * h->uvlinesize * block_h; mb_y * sl->uvlinesize * block_h;
dest_cr = h->cur_pic.f.data[2] + dest_cr = h->cur_pic.f.data[2] +
(mb_x << pixel_shift) * (8 << CHROMA444(h)) + (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
mb_y * h->uvlinesize * block_h; mb_y * sl->uvlinesize * block_h;
// FIXME simplify above // FIXME simplify above
if (MB_FIELD(sl)) { if (MB_FIELD(sl)) {
linesize = sl->mb_linesize = h->linesize * 2; linesize = sl->mb_linesize = sl->linesize * 2;
uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2; uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
if (mb_y & 1) { // FIXME move out of this function? if (mb_y & 1) { // FIXME move out of this function?
dest_y -= h->linesize * 15; dest_y -= sl->linesize * 15;
dest_cb -= h->uvlinesize * (block_h - 1); dest_cb -= sl->uvlinesize * (block_h - 1);
dest_cr -= h->uvlinesize * (block_h - 1); dest_cr -= sl->uvlinesize * (block_h - 1);
} }
} else { } else {
linesize = sl->mb_linesize = h->linesize; linesize = sl->mb_linesize = sl->linesize;
uvlinesize = sl->mb_uvlinesize = h->uvlinesize; uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
} }
backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize, backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
uvlinesize, 0); uvlinesize, 0);
@ -2083,7 +2073,10 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
int lf_x_start = sl->mb_x; int lf_x_start = sl->mb_x;
int ret; int ret;
ret = alloc_scratch_buffers(sl, h->linesize); sl->linesize = h->cur_pic_ptr->f.linesize[0];
sl->uvlinesize = h->cur_pic_ptr->f.linesize[1];
ret = alloc_scratch_buffers(sl, sl->linesize);
if (ret < 0) if (ret < 0)
return ret; return ret;

@ -313,23 +313,23 @@ static inline void svq3_mc_dir_part(SVQ3Context *s,
} }
/* form component predictions */ /* form component predictions */
dest = h->cur_pic.f.data[0] + x + y * h->linesize; dest = h->cur_pic.f.data[0] + x + y * sl->linesize;
src = pic->f.data[0] + mx + my * h->linesize; src = pic->f.data[0] + mx + my * sl->linesize;
if (emu) { if (emu) {
h->vdsp.emulated_edge_mc(sl->edge_emu_buffer, src, h->vdsp.emulated_edge_mc(sl->edge_emu_buffer, src,
h->linesize, h->linesize, sl->linesize, sl->linesize,
width + 1, height + 1, width + 1, height + 1,
mx, my, s->h_edge_pos, s->v_edge_pos); mx, my, s->h_edge_pos, s->v_edge_pos);
src = sl->edge_emu_buffer; src = sl->edge_emu_buffer;
} }
if (thirdpel) if (thirdpel)
(avg ? s->tdsp.avg_tpel_pixels_tab (avg ? s->tdsp.avg_tpel_pixels_tab
: s->tdsp.put_tpel_pixels_tab)[dxy](dest, src, h->linesize, : s->tdsp.put_tpel_pixels_tab)[dxy](dest, src, sl->linesize,
width, height); width, height);
else else
(avg ? s->hdsp.avg_pixels_tab (avg ? s->hdsp.avg_pixels_tab
: s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, h->linesize, : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, sl->linesize,
height); height);
if (!(h->flags & CODEC_FLAG_GRAY)) { if (!(h->flags & CODEC_FLAG_GRAY)) {
@ -340,12 +340,12 @@ static inline void svq3_mc_dir_part(SVQ3Context *s,
blocksize++; blocksize++;
for (i = 1; i < 3; i++) { for (i = 1; i < 3; i++) {
dest = h->cur_pic.f.data[i] + (x >> 1) + (y >> 1) * h->uvlinesize; dest = h->cur_pic.f.data[i] + (x >> 1) + (y >> 1) * sl->uvlinesize;
src = pic->f.data[i] + mx + my * h->uvlinesize; src = pic->f.data[i] + mx + my * sl->uvlinesize;
if (emu) { if (emu) {
h->vdsp.emulated_edge_mc(sl->edge_emu_buffer, src, h->vdsp.emulated_edge_mc(sl->edge_emu_buffer, src,
h->uvlinesize, h->uvlinesize, sl->uvlinesize, sl->uvlinesize,
width + 1, height + 1, width + 1, height + 1,
mx, my, (s->h_edge_pos >> 1), mx, my, (s->h_edge_pos >> 1),
s->v_edge_pos >> 1); s->v_edge_pos >> 1);
@ -354,12 +354,12 @@ static inline void svq3_mc_dir_part(SVQ3Context *s,
if (thirdpel) if (thirdpel)
(avg ? s->tdsp.avg_tpel_pixels_tab (avg ? s->tdsp.avg_tpel_pixels_tab
: s->tdsp.put_tpel_pixels_tab)[dxy](dest, src, : s->tdsp.put_tpel_pixels_tab)[dxy](dest, src,
h->uvlinesize, sl->uvlinesize,
width, height); width, height);
else else
(avg ? s->hdsp.avg_pixels_tab (avg ? s->hdsp.avg_pixels_tab
: s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src,
h->uvlinesize, sl->uvlinesize,
height); height);
} }
} }
@ -1101,8 +1101,8 @@ static int get_buffer(AVCodecContext *avctx, H264Picture *pic)
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
} }
h->linesize = pic->f.linesize[0]; sl->linesize = pic->f.linesize[0];
h->uvlinesize = pic->f.linesize[1]; sl->uvlinesize = pic->f.linesize[1];
return 0; return 0;
fail: fail:
@ -1162,14 +1162,14 @@ static int svq3_decode_frame(AVCodecContext *avctx, void *data,
return ret; return ret;
for (i = 0; i < 16; i++) { for (i = 0; i < 16; i++) {
h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3); h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * sl->linesize * ((scan8[i] - scan8[0]) >> 3);
h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3); h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * sl->linesize * ((scan8[i] - scan8[0]) >> 3);
} }
for (i = 0; i < 16; i++) { for (i = 0; i < 16; i++) {
h->block_offset[16 + i] = h->block_offset[16 + i] =
h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * sl->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
h->block_offset[48 + 16 + i] = h->block_offset[48 + 16 + i] =
h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * sl->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
} }
if (h->pict_type != AV_PICTURE_TYPE_I) { if (h->pict_type != AV_PICTURE_TYPE_I) {

Loading…
Cancel
Save