Commit bbd0063b authored by Jingning Han's avatar Jingning Han
Browse files

Make dequant/idct block size independent

The unified dequantization, inverse transform, and adding functions
support rectangular block sizes. Also separate the operations on
luma and chroma components, in the consideration of the txfm_size
for uv components in rectangular block sizes.

Change-Id: I2a13246b2a9086b37d575d346070990d854cc110
parent e63099d1
......@@ -443,18 +443,61 @@ static void decode_4x4(VP9D_COMP *pbi, MACROBLOCKD *xd,
}
}
static void decode_sb_16x16(MACROBLOCKD *mb, int y_size) {
const int y_count = y_size * y_size;
const int uv_size = y_size / 2;
const int uv_count = uv_size * uv_size;
static INLINE void decode_sby_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) - 1, bw = 1 << bwl;
const int bhl = mb_height_log2(bsize) - 1, bh = 1 << bhl;
const int y_count = bw * bh;
int n;
for (n = 0; n < y_count; n++) {
const int x_idx = n % y_size;
const int y_idx = n / y_size;
const int x_idx = n & (bw - 1);
const int y_idx = n >> bwl;
const int y_offset = (y_idx * 32) * mb->dst.y_stride + (x_idx * 32);
vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 1024),
mb->block[0].dequant ,
mb->dst.y_buffer + y_offset,
mb->dst.y_buffer + y_offset,
mb->dst.y_stride, mb->dst.y_stride,
mb->plane[0].eobs[n * 64]);
}
}
static INLINE void decode_sbuv_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) - 1, bw = (1 << bwl) / 2;
const int bhl = mb_height_log2(bsize) - 1, bh = (1 << bhl) / 2;
const int uv_count = bw * bh;
int n;
for (n = 0; n < uv_count; n++) {
const int x_idx = n & (bw - 1);
const int y_idx = n >> (bwl - 1);
const int uv_offset = (y_idx * 32) * mb->dst.uv_stride + (x_idx * 32);
vp9_dequant_idct_add_32x32(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 1024),
mb->block[16].dequant,
mb->dst.u_buffer + uv_offset,
mb->dst.u_buffer + uv_offset,
mb->dst.uv_stride, mb->dst.uv_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->dst.v_buffer + uv_offset,
mb->dst.v_buffer + uv_offset,
mb->dst.uv_stride, mb->dst.uv_stride,
mb->plane[2].eobs[n * 64]);
}
}
static INLINE void decode_sby_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize), bw = 1 << bwl;
const int bhl = mb_height_log2(bsize), bh = 1 << bhl;
const int y_count = bw * bh;
int n;
for (n = 0; n < y_count; n++) {
const int x_idx = n & (bw - 1);
const int y_idx = n >> bwl;
const int y_offset = (y_idx * 16) * mb->dst.y_stride + (x_idx * 16);
const TX_TYPE tx_type = get_tx_type_16x16(mb,
(y_idx * (4 * y_size) + x_idx) * 4);
(y_idx * (4 * bw) + x_idx) * 4);
if (tx_type == DCT_DCT) {
vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[0].qcoeff, n, 256),
mb->block[0].dequant ,
......@@ -472,10 +515,19 @@ static void decode_sb_16x16(MACROBLOCKD *mb, int y_size) {
mb->plane[0].eobs[n * 16]);
}
}
}
static INLINE void decode_sbuv_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize), bw = (1 << bwl) / 2;
const int bhl = mb_height_log2(bsize), bh = (1 << bhl) / 2;
const int uv_count = bw * bh;
int n;
assert(bsize >= BLOCK_SIZE_SB32X32);
for (n = 0; n < uv_count; n++) {
const int x_idx = n % uv_size;
const int y_idx = n / uv_size;
const int x_idx = n & (bw - 1);
const int y_idx = n >> (bwl - 1);
const int uv_offset = (y_idx * 16) * mb->dst.uv_stride + (x_idx * 16);
vp9_dequant_idct_add_16x16(BLOCK_OFFSET(mb->plane[1].qcoeff, n, 256),
mb->block[16].dequant,
......@@ -492,19 +544,19 @@ static void decode_sb_16x16(MACROBLOCKD *mb, int y_size) {
}
}
static INLINE void decode_sb_8x8(MACROBLOCKD *xd, int y_size) {
const int y_count = y_size * y_size;
const int uv_size = y_size / 2;
const int uv_count = uv_size * uv_size;
static INLINE void decode_sby_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 1, bw = 1 << bwl;
const int bhl = mb_height_log2(bsize) + 1, bh = 1 << bhl;
const int y_count = bw * bh;
int n;
// luma
for (n = 0; n < y_count; n++) {
const int x_idx = n % y_size;
const int y_idx = n / y_size;
const int x_idx = n & (bw - 1);
const int y_idx = n >> bwl;
const int y_offset = (y_idx * 8) * xd->dst.y_stride + (x_idx * 8);
const TX_TYPE tx_type = get_tx_type_8x8(xd,
(y_idx * (2 * y_size) + x_idx) * 2);
(y_idx * (2 * bw) + x_idx) * 2);
if (tx_type == DCT_DCT) {
vp9_dequant_idct_add_8x8_c(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 64),
xd->block[0].dequant,
......@@ -522,11 +574,18 @@ static INLINE void decode_sb_8x8(MACROBLOCKD *xd, int y_size) {
xd->plane[0].eobs[n * 4]);
}
}
}
static INLINE void decode_sbuv_8x8(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 1, bw = 1 << (bwl - 1);
const int bhl = mb_height_log2(bsize) + 1, bh = 1 << (bhl - 1);
const int uv_count = bw * bh;
int n;
// chroma
for (n = 0; n < uv_count; n++) {
const int x_idx = n % uv_size;
const int y_idx = n / uv_size;
const int x_idx = n & (bw - 1);
const int y_idx = n >> (bwl - 1);
const int uv_offset = (y_idx * 8) * xd->dst.uv_stride + (x_idx * 8);
vp9_dequant_idct_add_8x8_c(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 64),
xd->block[16].dequant,
......@@ -543,18 +602,17 @@ static INLINE void decode_sb_8x8(MACROBLOCKD *xd, int y_size) {
}
}
static void decode_sb_4x4(MACROBLOCKD *xd, int y_size) {
const int y_count = y_size * y_size;
const int uv_size = y_size / 2;
const int uv_count = uv_size * uv_size;
static INLINE void decode_sby_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 2, bw = 1 << bwl;
const int bhl = mb_height_log2(bsize) + 2, bh = 1 << bhl;
const int y_count = bw * bh;
int n;
for (n = 0; n < y_count; n++) {
const int x_idx = n % y_size;
const int y_idx = n / y_size;
const int x_idx = n & (bw - 1);
const int y_idx = n >> bwl;
const int y_offset = (y_idx * 4) * xd->dst.y_stride + (x_idx * 4);
const TX_TYPE tx_type = get_tx_type_4x4(xd, y_idx * y_size + x_idx);
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,
......@@ -573,10 +631,17 @@ static void decode_sb_4x4(MACROBLOCKD *xd, int y_size) {
xd->plane[0].eobs[n]);
}
}
}
static INLINE void decode_sbuv_4x4(MACROBLOCKD *xd, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize) + 2, bw = 1 << (bwl - 1);
const int bhl = mb_height_log2(bsize) + 2, bh = 1 << (bhl - 1);
const int uv_count = bw * bh;
int n;
for (n = 0; n < uv_count; n++) {
const int x_idx = n % uv_size;
const int y_idx = n / uv_size;
const int x_idx = n & (bw - 1);
const int y_idx = n >> (bwl - 1);
const int uv_offset = (y_idx * 4) * xd->dst.uv_stride + (x_idx * 4);
xd->itxm_add(BLOCK_OFFSET(xd->plane[1].qcoeff, n, 16),
xd->block[16].dequant,
......@@ -591,14 +656,34 @@ static void decode_sb_4x4(MACROBLOCKD *xd, int y_size) {
}
}
static void decode_sb64(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
BOOL_DECODER* const bc) {
// TODO(jingning): combine luma and chroma dequantization and inverse
// transform into a single function looping over planes.
static void decode_sb_32x32(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
decode_sby_32x32(mb, bsize);
if (bsize == BLOCK_SIZE_SB64X64)
decode_sbuv_32x32(mb, bsize);
else
decode_sbuv_16x16(mb, bsize);
}
static void decode_sb_16x16(MACROBLOCKD *mb, BLOCK_SIZE_TYPE bsize) {
decode_sby_16x16(mb, bsize);
if (bsize >= BLOCK_SIZE_SB32X32)
decode_sbuv_16x16(mb, bsize);
else
decode_sbuv_8x8(mb, bsize);
}
static void decode_sb(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
BOOL_DECODER* const bc, BLOCK_SIZE_TYPE bsize) {
const int bwl = mb_width_log2(bsize), bhl = mb_height_log2(bsize);
const int bw = 1 << bwl, bh = 1 << bhl;
int n, eobtotal;
VP9_COMMON *const pc = &pbi->common;
MODE_INFO *mi = xd->mode_info_context;
const int mis = pc->mode_info_stride;
assert(mi->mbmi.sb_type == BLOCK_SIZE_SB64X64);
assert(mi->mbmi.sb_type == bsize);
if (pbi->common.frame_type != KEY_FRAME)
vp9_setup_interp_filters(xd, mi->mbmi.interp_filter, pc);
......@@ -608,7 +693,7 @@ static void decode_sb64(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
mb_init_dequantizer(pbi, xd);
if (mi->mbmi.mb_skip_coeff) {
vp9_reset_sb_tokens_context(xd, BLOCK_SIZE_SB64X64);
vp9_reset_sb_tokens_context(xd, bsize);
// Special case: Force the loopfilter to skip when eobtotal and
// mb_skip_coeff are zero.
......@@ -616,19 +701,32 @@ static void decode_sb64(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
return;
}
// do prediction
if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
vp9_build_intra_predictors_sb64y_s(xd);
vp9_build_intra_predictors_sb64uv_s(xd);
// TODO(jingning): need to combine intra/inter predictor functions and
// make them block size independent.
// generate prediction
if (bsize == BLOCK_SIZE_SB64X64) {
assert(bsize == BLOCK_SIZE_SB64X64);
if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
vp9_build_intra_predictors_sb64y_s(xd);
vp9_build_intra_predictors_sb64uv_s(xd);
} else {
vp9_build_inter64x64_predictors_sb(xd, mb_row, mb_col);
}
} else {
vp9_build_inter64x64_predictors_sb(xd, mb_row, mb_col);
assert(bsize == BLOCK_SIZE_SB32X32);
if (xd->mode_info_context->mbmi.ref_frame == INTRA_FRAME) {
vp9_build_intra_predictors_sby_s(xd);
vp9_build_intra_predictors_sbuv_s(xd);
} else {
vp9_build_inter32x32_predictors_sb(xd, mb_row, mb_col);
}
}
// dequantization and idct
eobtotal = vp9_decode_tokens(pbi, xd, bc, BLOCK_SIZE_SB64X64);
eobtotal = vp9_decode_tokens(pbi, xd, bc, bsize);
if (eobtotal == 0) { // skip loopfilter
for (n = 0; n < 16; n++) {
const int x_idx = n & 3, y_idx = n >> 2;
for (n = 0; n < bw * bh; n++) {
const int x_idx = n & (bw - 1), y_idx = n >> bwl;
if (mb_col + x_idx < pc->mb_cols && mb_row + y_idx < pc->mb_rows)
mi[y_idx * mis + x_idx].mbmi.mb_skip_coeff = mi->mbmi.mb_skip_coeff;
......@@ -636,108 +734,18 @@ static void decode_sb64(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
} else {
switch (xd->mode_info_context->mbmi.txfm_size) {
case TX_32X32:
for (n = 0; n < 4; n++) {
const int x_idx = n & 1, y_idx = n >> 1;
const int y_offset = x_idx * 32 + y_idx * xd->dst.y_stride * 32;
vp9_dequant_idct_add_32x32(BLOCK_OFFSET(xd->plane[0].qcoeff, n, 1024),
xd->block[0].dequant,
xd->dst.y_buffer + y_offset,
xd->dst.y_buffer + y_offset,
xd->dst.y_stride, xd->dst.y_stride, xd->plane[0].eobs[n * 64]);
}
vp9_dequant_idct_add_32x32(xd->plane[1].qcoeff,
xd->block[16].dequant, xd->dst.u_buffer, xd->dst.u_buffer,
xd->dst.uv_stride, xd->dst.uv_stride, xd->plane[1].eobs[0]);
vp9_dequant_idct_add_32x32(xd->plane[2].qcoeff,
xd->block[20].dequant, xd->dst.v_buffer, xd->dst.v_buffer,
xd->dst.uv_stride, xd->dst.uv_stride, xd->plane[2].eobs[0]);
break;
case TX_16X16:
decode_sb_16x16(xd, 4);
break;
case TX_8X8:
decode_sb_8x8(xd, 8);
break;
case TX_4X4:
decode_sb_4x4(xd, 16);
break;
default: assert(0);
}
}
#if CONFIG_CODE_NONZEROCOUNT
propagate_nzcs(&pbi->common, xd);
#endif
}
static void decode_sb32(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
BOOL_DECODER* const bc) {
int eobtotal;
VP9_COMMON *const pc = &pbi->common;
MODE_INFO *mi = xd->mode_info_context;
const int mis = pc->mode_info_stride;
assert(mi->mbmi.sb_type == BLOCK_SIZE_SB32X32);
if (pbi->common.frame_type != KEY_FRAME)
vp9_setup_interp_filters(xd, mi->mbmi.interp_filter, pc);
// re-initialize macroblock dequantizer before detokenization
if (xd->segmentation_enabled)
mb_init_dequantizer(pbi, xd);
if (mi->mbmi.mb_skip_coeff) {
vp9_reset_sb_tokens_context(xd, BLOCK_SIZE_SB32X32);
// Special case: Force the loopfilter to skip when eobtotal and
// mb_skip_coeff are zero.
skip_recon_mb(pbi, xd, mb_row, mb_col);
return;
}
// do prediction
if (mi->mbmi.ref_frame == INTRA_FRAME) {
vp9_build_intra_predictors_sby_s(xd);
vp9_build_intra_predictors_sbuv_s(xd);
} else {
vp9_build_inter32x32_predictors_sb(xd, mb_row, mb_col);
}
// dequantization and idct
eobtotal = vp9_decode_tokens(pbi, xd, bc, BLOCK_SIZE_SB32X32);
if (eobtotal == 0) { // skip loopfilter
mi->mbmi.mb_skip_coeff = 1;
if (mb_col + 1 < pc->mb_cols)
mi[1].mbmi.mb_skip_coeff = 1;
if (mb_row + 1 < pc->mb_rows) {
mi[mis].mbmi.mb_skip_coeff = 1;
if (mb_col + 1 < pc->mb_cols)
mi[mis + 1].mbmi.mb_skip_coeff = 1;
}
} else {
switch (xd->mode_info_context->mbmi.txfm_size) {
case TX_32X32:
vp9_dequant_idct_add_32x32(xd->plane[0].qcoeff, xd->block[0].dequant,
xd->dst.y_buffer, xd->dst.y_buffer,
xd->dst.y_stride, xd->dst.y_stride,
xd->plane[0].eobs[0]);
vp9_dequant_idct_add_16x16(xd->plane[1].qcoeff, xd->block[16].dequant,
xd->dst.u_buffer, xd->dst.u_buffer,
xd->dst.uv_stride, xd->dst.uv_stride,
xd->plane[1].eobs[0]);
vp9_dequant_idct_add_16x16(xd->plane[2].qcoeff, xd->block[16].dequant,
xd->dst.v_buffer, xd->dst.v_buffer,
xd->dst.uv_stride, xd->dst.uv_stride,
xd->plane[2].eobs[0]);
decode_sb_32x32(xd, bsize);
break;
case TX_16X16:
decode_sb_16x16(xd, 2);
decode_sb_16x16(xd, bsize);
break;
case TX_8X8:
decode_sb_8x8(xd, 4);
decode_sby_8x8(xd, bsize);
decode_sbuv_8x8(xd, bsize);
break;
case TX_4X4:
decode_sb_4x4(xd, 8);
decode_sby_4x4(xd, bsize);
decode_sbuv_4x4(xd, bsize);
break;
default: assert(0);
}
......@@ -747,6 +755,8 @@ static void decode_sb32(VP9D_COMP *pbi, MACROBLOCKD *xd, int mb_row, int mb_col,
#endif
}
// TODO(jingning): Need to merge SB and MB decoding. The MB decoding currently
// couples special handles on I8x8, B_PRED, and splitmv modes.
static void decode_mb(VP9D_COMP *pbi, MACROBLOCKD *xd,
int mb_row, int mb_col,
BOOL_DECODER* const bc) {
......@@ -943,7 +953,7 @@ static void decode_sb_row(VP9D_COMP *pbi, int mb_row, vp9_reader* r) {
set_offsets(pbi, 64, mb_row, mb_col);
vp9_decode_mb_mode_mv(pbi, xd, mb_row, mb_col, r);
set_refs(pbi, 64, mb_row, mb_col);
decode_sb64(pbi, xd, mb_row, mb_col, r);
decode_sb(pbi, xd, mb_row, mb_col, r, BLOCK_SIZE_SB64X64);
xd->corrupted |= bool_error(r);
} else {
// not SB64
......@@ -962,7 +972,7 @@ static void decode_sb_row(VP9D_COMP *pbi, int mb_row, vp9_reader* r) {
set_offsets(pbi, 32, y_idx_sb, x_idx_sb);
vp9_decode_mb_mode_mv(pbi, xd, y_idx_sb, x_idx_sb, r);
set_refs(pbi, 32, y_idx_sb, x_idx_sb);
decode_sb32(pbi, xd, y_idx_sb, x_idx_sb, r);
decode_sb(pbi, xd, y_idx_sb, x_idx_sb, r, BLOCK_SIZE_SB32X32);
xd->corrupted |= bool_error(r);
} else {
// not SB32
......
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