generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Reduce dqcoeff array size in decoder

Browse Source 
The decoding process handles detokenization and reconstruction per
transform block sequentially. There is no need to offset the dqcoeff
buffer according to the transform block index. This allows to
reduce the memory spill and improve cache performance.

Change-Id: Ibb8bfe532a7a08fcabaf6d42cbec1e986901d32d
This commit is contained in:
Jingning Han 2015-07-07 11:36:05 -07:00
parent 0ede9f52b7
commit cccad1c5de
3 changed files with 6 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
vp9/common/vp9_blockd.h
View File

@ -188,8 +188,11 @@ typedef struct macroblockd {
#endif #endif
/* dqcoeff are shared by all the planes. So planes must be decoded serially */ /* dqcoeff are shared by all the planes. So planes must be decoded serially */
#if CONFIG_VP9_ENCODER
DECLARE_ALIGNED(16, tran_low_t, dqcoeff[64 * 64]); DECLARE_ALIGNED(16, tran_low_t, dqcoeff[64 * 64]);
#else
DECLARE_ALIGNED(16, tran_low_t, dqcoeff[32 * 32]);
#endif
int lossless; int lossless;
int corrupted; int corrupted;

2
vp9/decoder/vp9_decodeframe.c
View File

@ -188,7 +188,7 @@ static void inverse_transform_block(MACROBLOCKD* xd, int plane, int block,
struct macroblockd_plane *const pd = &xd->plane[plane]; struct macroblockd_plane *const pd = &xd->plane[plane];
if (eob > 0) { if (eob > 0) {
TX_TYPE tx_type = DCT_DCT; TX_TYPE tx_type = DCT_DCT;
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block); tran_low_t *const dqcoeff = pd->dqcoeff;
#if CONFIG_VP9_HIGHBITDEPTH #if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
if (xd->lossless) { if (xd->lossless) {

2
vp9/decoder/vp9_detokenize.c
View File

@ -217,7 +217,7 @@ int vp9_decode_block_tokens(MACROBLOCKD *xd,
pd->left_context + y); pd->left_context + y);
const scan_order *so = get_scan(xd, tx_size, pd->plane_type, block); const scan_order *so = get_scan(xd, tx_size, pd->plane_type, block);
const int eob = decode_coefs(xd, pd->plane_type, const int eob = decode_coefs(xd, pd->plane_type,
BLOCK_OFFSET(pd->dqcoeff, block), tx_size, pd->dqcoeff, tx_size,
dequant, ctx, so->scan, so->neighbors, r); dequant, ctx, so->scan, so->neighbors, r);
vp9_set_contexts(xd, pd, plane_bsize, tx_size, eob > 0, x, y); vp9_set_contexts(xd, pd, plane_bsize, tx_size, eob > 0, x, y);
return eob; return eob;