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

Merge "Moved dequantization into the token decoder" into experimental

Browse Source
This commit is contained in:
Scott LaVarnway
2013-04-25 08:53:42 -07:00
committed by Gerrit Code Review
parent bcf82cf503 e732bc298c
commit a426c7f343
12 changed files with 460 additions and 601 deletions
Download Patch File Download Diff File Expand all files Collapse all files

199
vp9/decoder/vp9_decodframe.c
View File

@@ -216,32 +216,29 @@ static void mb_init_dequantizer(VP9D_COMP *pbi, MACROBLOCKD *mb) {
assert(qindex == 0);
mb->inv_txm4x4_1 = vp9_short_iwalsh4x4_1;
mb->inv_txm4x4 = vp9_short_iwalsh4x4;
mb->itxm_add = vp9_dequant_idct_add_lossless_c;
mb->itxm_add_y_block = vp9_dequant_idct_add_y_block_lossless_c;
mb->itxm_add_uv_block = vp9_dequant_idct_add_uv_block_lossless_c;
mb->itxm_add = vp9_idct_add_lossless_c;
mb->itxm_add_y_block = vp9_idct_add_y_block_lossless_c;
mb->itxm_add_uv_block = vp9_idct_add_uv_block_lossless_c;
} else {
mb->inv_txm4x4_1 = vp9_short_idct4x4_1;
mb->inv_txm4x4 = vp9_short_idct4x4;
mb->itxm_add = vp9_dequant_idct_add;
mb->itxm_add_y_block = vp9_dequant_idct_add_y_block;
mb->itxm_add_uv_block = vp9_dequant_idct_add_uv_block;
mb->itxm_add = vp9_idct_add;
mb->itxm_add_y_block = vp9_idct_add_y_block;
mb->itxm_add_uv_block = vp9_idct_add_uv_block;
}
}
static void decode_16x16(MACROBLOCKD *xd) {
const TX_TYPE tx_type = get_tx_type_16x16(xd, 0);
vp9_dequant_iht_add_16x16_c(tx_type, xd->plane[0].qcoeff,
xd->block[0].dequant, xd->plane[0].dst.buf,
xd->plane[0].dst.stride, xd->plane[0].eobs[0]);
vp9_iht_add_16x16_c(tx_type, xd->plane[0].qcoeff, xd->plane[0].dst.buf,
xd->plane[0].dst.stride, xd->plane[0].eobs[0]);
vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
xd->plane[1].dst.buf, xd->plane[1].dst.stride,
xd->plane[1].eobs[0]);
vp9_idct_add_8x8(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
xd->plane[1].dst.stride, xd->plane[1].eobs[0]);
vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[20].dequant,
xd->plane[2].dst.buf, xd->plane[1].dst.stride,
xd->plane[2].eobs[0]);
vp9_idct_add_8x8(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
xd->plane[1].dst.stride, xd->plane[2].eobs[0]);
}
static void decode_8x8(MACROBLOCKD *xd) {
@@ -255,7 +252,6 @@ static void decode_8x8(MACROBLOCKD *xd) {
int ib = vp9_i8x8_block[i];
int idx = (ib & 0x02) ? (ib + 2) : ib;
int16_t *q = BLOCK_OFFSET(xd->plane[0].qcoeff, idx, 16);
int16_t *dq = xd->block[0].dequant;
uint8_t *dst = *(xd->block[ib].base_dst) + xd->block[ib].dst;
int stride = xd->plane[0].dst.stride;
if (mode == I8X8_PRED) {
@@ -264,13 +260,11 @@ static void decode_8x8(MACROBLOCKD *xd) {
vp9_intra8x8_predict(xd, b, i8x8mode, dst, stride);
}
tx_type = get_tx_type_8x8(xd, ib);
vp9_dequant_iht_add_8x8_c(tx_type, q, dq, dst, stride,
xd->plane[0].eobs[idx]);
vp9_iht_add_8x8_c(tx_type, q, dst, stride, xd->plane[0].eobs[idx]);
}
} else {
vp9_dequant_idct_add_y_block_8x8(xd->plane[0].qcoeff,
xd->block[0].dequant, xd->plane[0].dst.buf,
xd->plane[0].dst.stride, xd);
vp9_idct_add_y_block_8x8(xd->plane[0].qcoeff, xd->plane[0].dst.buf,
xd->plane[0].dst.stride, xd);
}
// chroma
@@ -285,29 +279,27 @@ static void decode_8x8(MACROBLOCKD *xd) {
vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
b->dst_stride);
xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
*(b->base_dst) + b->dst, b->dst_stride,
xd->plane[1].eobs[i]);
b = &xd->block[20 + i];
vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
b->dst_stride);
xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
*(b->base_dst) + b->dst, b->dst_stride,
xd->plane[2].eobs[i]);
}
} else if (mode == SPLITMV) {
xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
xd->plane[1].dst.buf, xd->plane[1].dst.stride, xd->plane[1].eobs);
xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
xd->plane[2].dst.buf, xd->plane[1].dst.stride, xd->plane[2].eobs);
xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
xd->plane[1].dst.stride, xd->plane[1].eobs);
xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
xd->plane[1].dst.stride, xd->plane[2].eobs);
} else {
vp9_dequant_idct_add_8x8(xd->plane[1].qcoeff, xd->block[16].dequant,
xd->plane[1].dst.buf, xd->plane[1].dst.stride,
xd->plane[1].eobs[0]);
vp9_idct_add_8x8(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
xd->plane[1].dst.stride, xd->plane[1].eobs[0]);
vp9_dequant_idct_add_8x8(xd->plane[2].qcoeff, xd->block[16].dequant,
xd->plane[2].dst.buf, xd->plane[1].dst.stride,
xd->plane[2].eobs[0]);
vp9_idct_add_8x8(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
xd->plane[1].dst.stride, xd->plane[2].eobs[0]);
}
}
@@ -315,13 +307,10 @@ static INLINE void dequant_add_y(MACROBLOCKD *xd, TX_TYPE tx_type, int idx) {
BLOCKD *const b = &xd->block[idx];
struct macroblockd_plane *const y = &xd->plane[0];
if (tx_type != DCT_DCT) {
vp9_dequant_iht_add_c(tx_type,
BLOCK_OFFSET(y->qcoeff, idx, 16),
b->dequant, *(b->base_dst) + b->dst,
b->dst_stride, y->eobs[idx]);
vp9_iht_add_c(tx_type, BLOCK_OFFSET(y->qcoeff, idx, 16),
*(b->base_dst) + b->dst, b->dst_stride, y->eobs[idx]);
} else {
xd->itxm_add(BLOCK_OFFSET(y->qcoeff, idx, 16),
b->dequant, *(b->base_dst) + b->dst,
xd->itxm_add(BLOCK_OFFSET(y->qcoeff, idx, 16), *(b->base_dst) + b->dst,
b->dst_stride, y->eobs[idx]);
}
}
@@ -348,13 +337,13 @@ static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd, vp9_reader *r) {
vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
b->dst_stride);
xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, i, 16),
b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
*(b->base_dst) + b->dst, b->dst_stride,
xd->plane[1].eobs[i]);
b = &xd->block[20 + i];
vp9_intra_uv4x4_predict(xd, b, i8x8mode, *(b->base_dst) + b->dst,
b->dst_stride);
xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, i, 16),
b->dequant, *(b->base_dst) + b->dst, b->dst_stride,
*(b->base_dst) + b->dst, b->dst_stride,
xd->plane[2].eobs[i]);
}
} else if (mode == I4X4_PRED) {
@@ -365,7 +354,7 @@ static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd, vp9_reader *r) {
xd->mode_info_context->bmi[i].as_mode.context = b->bmi.as_mode.context =
vp9_find_bpred_context(xd, b);
if (!xd->mode_info_context->mbmi.mb_skip_coeff)
vp9_decode_coefs_4x4(pbi, xd, r, PLANE_TYPE_Y_WITH_DC, i);
vp9_decode_coefs_4x4(pbi, xd, r, PLANE_TYPE_Y_WITH_DC, i, b->dequant);
#endif
vp9_intra4x4_predict(xd, b, b_mode, *(b->base_dst) + b->dst,
b->dst_stride);
@@ -374,32 +363,29 @@ static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd, vp9_reader *r) {
}
#if CONFIG_NEWBINTRAMODES
if (!xd->mode_info_context->mbmi.mb_skip_coeff)
vp9_decode_mb_tokens_4x4_uv(pbi, xd, r);
vp9_decode_mb_tokens_4x4_uv(pbi, xd, r, xd->block[16].dequant);
#endif
vp9_build_intra_predictors_sbuv_s(xd, BLOCK_SIZE_MB16X16);
xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
xd->plane[1].dst.buf, xd->plane[1].dst.stride, xd->plane[1].eobs);
xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
xd->plane[2].dst.buf, xd->plane[1].dst.stride, xd->plane[2].eobs);
xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
xd->plane[1].dst.stride, xd->plane[1].eobs);
xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
xd->plane[1].dst.stride, xd->plane[2].eobs);
} else if (mode == SPLITMV || get_tx_type_4x4(xd, 0) == DCT_DCT) {
xd->itxm_add_y_block(xd->plane[0].qcoeff,
xd->block[0].dequant,
xd->plane[0].dst.buf, xd->plane[0].dst.stride, xd);
xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
xd->plane[1].dst.buf, xd->plane[1].dst.stride, xd->plane[1].eobs);
xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
xd->plane[2].dst.buf, xd->plane[1].dst.stride, xd->plane[2].eobs);
xd->itxm_add_y_block(xd->plane[0].qcoeff, xd->plane[0].dst.buf,
xd->plane[0].dst.stride, xd);
xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
xd->plane[1].dst.stride, xd->plane[1].eobs);
xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
xd->plane[1].dst.stride, xd->plane[2].eobs);
} else {
for (i = 0; i < 16; i++) {
tx_type = get_tx_type_4x4(xd, i);
dequant_add_y(xd, tx_type, i);
}
xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->block[16].dequant,
xd->plane[1].dst.buf, xd->plane[1].dst.stride,
xd->plane[1].eobs);
xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->block[16].dequant,
xd->plane[2].dst.buf, xd->plane[1].dst.stride,
xd->plane[2].eobs);
xd->itxm_add_uv_block(xd->plane[1].qcoeff, xd->plane[1].dst.buf,
xd->plane[1].dst.stride, xd->plane[1].eobs);
xd->itxm_add_uv_block(xd->plane[2].qcoeff, xd->plane[2].dst.buf,
xd->plane[1].dst.stride, xd->plane[2].eobs);
}
}
@@ -413,11 +399,10 @@ static INLINE void decode_sby_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
const int x_idx = n & (bw - 1);
const int y_idx = n >> bwl;
const int y_offset = (y_idx * 32) * mb->plane[0].dst.stride + (x_idx * 32);
vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 1024),
mb->block[0].dequant ,
mb->plane[0].dst.buf + y_offset,
mb->plane[0].dst.stride,
mb->plane[0].eobs[n * 64]);
vp9_idct_add_32x32(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 1024),
mb->plane[0].dst.buf + y_offset,
mb->plane[0].dst.stride,
mb->plane[0].eobs[n * 64]);
}
}
@@ -431,16 +416,14 @@ static INLINE void decode_sbuv_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
const int y_idx = n >> (bwl - 1);
const int uv_offset = (y_idx * 32) * mb->plane[1].dst.stride +
(x_idx * 32);
vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 1024),
mb->block[16].dequant,
mb->plane[1].dst.buf + uv_offset,
mb->plane[1].dst.stride,
mb->plane[1].eobs[n * 64]);
vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 1024),
mb->block[20].dequant,
mb->plane[2].dst.buf + uv_offset,
mb->plane[1].dst.stride,
mb->plane[2].eobs[n * 64]);
vp9_idct_add_32x32(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 1024),
mb->plane[1].dst.buf + uv_offset,
mb->plane[1].dst.stride,
mb->plane[1].eobs[n * 64]);
vp9_idct_add_32x32(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 1024),
mb->plane[2].dst.buf + uv_offset,
mb->plane[1].dst.stride,
mb->plane[2].eobs[n * 64]);
}
}
@@ -456,12 +439,9 @@ static INLINE void decode_sby_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
const int y_offset = (y_idx * 16) * mb->plane[0].dst.stride + (x_idx * 16);
const TX_TYPE tx_type = get_tx_type_16x16(mb,
(y_idx * (4 * bw) + x_idx) * 4);
vp9_dequant_iht_add_16x16_c(tx_type,
BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
mb->block[0].dequant,
mb->plane[0].dst.buf + y_offset,
mb->plane[0].dst.stride,
mb->plane[0].eobs[n * 16]);
vp9_iht_add_16x16_c(tx_type, BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
mb->plane[0].dst.buf + y_offset,
mb->plane[0].dst.stride, mb->plane[0].eobs[n * 16]);
}
}
@@ -477,16 +457,12 @@ static INLINE void decode_sbuv_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
const int x_idx = n & (bw - 1);
const int y_idx = n >> (bwl - 1);
const int uv_offset = (y_idx * 16) * mb->plane[1].dst.stride + (x_idx * 16);
vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 256),
mb->block[16].dequant,
mb->plane[1].dst.buf + uv_offset,
mb->plane[1].dst.stride,
mb->plane[1].eobs[n * 16]);
vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 256),
mb->block[20].dequant,
mb->plane[2].dst.buf + uv_offset,
mb->plane[1].dst.stride,
mb->plane[2].eobs[n * 16]);
vp9_idct_add_16x16(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 256),
mb->plane[1].dst.buf + uv_offset,
mb->plane[1].dst.stride, mb->plane[1].eobs[n * 16]);
vp9_idct_add_16x16(BLOCK_OFFSET(mb->plane[2].qcoeff, n, 256),
mb->plane[2].dst.buf + uv_offset,
mb->plane[1].dst.stride, mb->plane[2].eobs[n * 16]);
}
}
@@ -504,12 +480,9 @@ static INLINE void decode_sby_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
const TX_TYPE tx_type = get_tx_type_8x8(xd,
(y_idx * (2 * bw) + x_idx) * 2);
vp9_dequant_iht_add_8x8_c(tx_type,
BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
xd->block[0].dequant,
xd->plane[0].dst.buf + y_offset,
xd->plane[0].dst.stride,
xd->plane[0].eobs[n * 4]);
vp9_iht_add_8x8_c(tx_type, BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
xd->plane[0].dst.buf + y_offset, xd->plane[0].dst.stride,
xd->plane[0].eobs[n * 4]);
}
}
@@ -524,16 +497,12 @@ static INLINE void decode_sbuv_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
const int x_idx = n & (bw - 1);
const int y_idx = n >> (bwl - 1);
const int uv_offset = (y_idx * 8) * xd->plane[1].dst.stride + (x_idx * 8);
vp9_dequant_idct_add_8x8(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 64),
xd->block[16].dequant,
xd->plane[1].dst.buf + uv_offset,
xd->plane[1].dst.stride,
xd->plane[1].eobs[n * 4]);
vp9_dequant_idct_add_8x8(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 64),
xd->block[20].dequant,
xd->plane[2].dst.buf + uv_offset,
xd->plane[1].dst.stride,
xd->plane[2].eobs[n * 4]);
vp9_idct_add_8x8(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 64),
xd->plane[1].dst.buf + uv_offset, xd->plane[1].dst.stride,
xd->plane[1].eobs[n * 4]);
vp9_idct_add_8x8(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 64),
xd->plane[2].dst.buf + uv_offset, xd->plane[1].dst.stride,
xd->plane[2].eobs[n * 4]);
}
}
@@ -550,15 +519,12 @@ static INLINE void decode_sby_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
const TX_TYPE tx_type = get_tx_type_4x4(xd, n);
if (tx_type == DCT_DCT) {
xd->itxm_add(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
xd->block[0].dequant,
xd->plane[0].dst.buf + y_offset, xd->plane[0].dst.stride,
xd->plane[0].eobs[n]);
} else {
vp9_dequant_iht_add_c(tx_type,
BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
xd->block[0].dequant,
xd->plane[0].dst.buf + y_offset,
xd->plane[0].dst.stride, xd->plane[0].eobs[n]);
vp9_iht_add_c(tx_type, BLOCK_OFFSET(xd->plane[0].qcoeff, n, 16),
xd->plane[0].dst.buf + y_offset, xd->plane[0].dst.stride,
xd->plane[0].eobs[n]);
}
}
}
@@ -574,11 +540,9 @@ static INLINE void decode_sbuv_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
const int y_idx = n >> (bwl - 1);
const int uv_offset = (y_idx * 4) * xd->plane[1].dst.stride + (x_idx * 4);
xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 16),
xd->block[16].dequant,
xd->plane[1].dst.buf + uv_offset, xd->plane[1].dst.stride,
xd->plane[1].eobs[n]);
xd->itxm_add(BLOCK_OFFSET(xd->plane[2].qcoeff, n, 16),
xd->block[20].dequant,
xd->plane[2].dst.buf + uv_offset, xd->plane[1].dst.stride,
xd->plane[2].eobs[n]);
}
@@ -632,7 +596,7 @@ static void decode_sb(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
mb_init_dequantizer(pbi, xd);
// dequantization and idct
eobtotal = vp9_decode_tokens(pbi, xd, r, bsize);
eobtotal = vp9_decode_tokens(pbi, xd, r, bsize, xd->block[0].dequant);
if (eobtotal == 0) { // skip loopfilter
for (n = 0; n < bw * bh; n++) {
const int x_idx = n & (bw - 1), y_idx = n >> bwl;
@@ -706,7 +670,8 @@ static void decode_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
#if CONFIG_NEWBINTRAMODES
if (mode != I4X4_PRED)
#endif
eobtotal = vp9_decode_tokens(pbi, xd, r, BLOCK_SIZE_MB16X16);
eobtotal = vp9_decode_tokens(pbi, xd, r, BLOCK_SIZE_MB16X16,
xd->block[0].dequant);
}
}