Skip to content

GitLab

A
aom-rav1e
Commit f1d090e2 authored by Hui Su's avatar Hui Su Committed by Gerrit Code Review
Browse files
Options

Merge "Refactoring on transform types"

parents 697a8e6f d76e5b36
Hide whitespace changes
Inline Side-by-side
Showing with 452 additions and 243 deletions
vp10/common/idct.c
View file @ f1d090e2
......@@ -178,30 +178,73 @@ void vp10_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
vpx_idct32x32_1024_add(input, dest, stride);
}
// iht
void vp10_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
int stride, int eob) {
if (tx_type == DCT_DCT)
vp10_idct4x4_add(input, dest, stride, eob);
else
vp10_iht4x4_16_add(input, dest, stride, tx_type);
void vp10_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type,
void (*itxm_add_4x4)(const tran_low_t *input,
uint8_t *dest, int stride, int eob)) {
switch (tx_type) {
case DCT_DCT:
itxm_add_4x4(input, dest, stride, eob);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_iht4x4_16_add(input, dest, stride, tx_type);
break;
default:
assert(0);
break;
}
}
void vp10_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
int stride, int eob) {
if (tx_type == DCT_DCT) {
vp10_idct8x8_add(input, dest, stride, eob);
} else {
vp10_iht8x8_64_add(input, dest, stride, tx_type);
void vp10_inv_txfm_add_8x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
vp10_idct8x8_add(input, dest, stride, eob);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_iht8x8_64_add(input, dest, stride, tx_type);
break;
default:
assert(0);
break;
}
}
void vp10_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
int stride, int eob) {
if (tx_type == DCT_DCT) {
vp10_idct16x16_add(input, dest, stride, eob);
} else {
vp10_iht16x16_256_add(input, dest, stride, tx_type);
void vp10_inv_txfm_add_16x16(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
vp10_idct16x16_add(input, dest, stride, eob);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_iht16x16_256_add(input, dest, stride, tx_type);
break;
default:
assert(0);
break;
}
}
void vp10_inv_txfm_add_32x32(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
vp10_idct32x32_add(input, dest, stride, eob);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
assert(0);
break;
default:
assert(0);
break;
}
}
......@@ -373,30 +416,77 @@ void vp10_highbd_idct32x32_add(const tran_low_t *input, uint8_t *dest,
}
}
// iht
void vp10_highbd_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input,
uint8_t *dest, int stride, int eob, int bd) {
if (tx_type == DCT_DCT)
vp10_highbd_idct4x4_add(input, dest, stride, eob, bd);
else
vp10_highbd_iht4x4_16_add(input, dest, stride, tx_type, bd);
void vp10_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type,
void (*highbd_itxm_add_4x4)
(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd)) {
switch (tx_type) {
case DCT_DCT:
highbd_itxm_add_4x4(input, dest, stride, eob, bd);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_iht4x4_16_add(input, dest, stride, tx_type, bd);
break;
default:
assert(0);
break;
}
}
void vp10_highbd_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input,
uint8_t *dest, int stride, int eob, int bd) {
if (tx_type == DCT_DCT) {
vp10_highbd_idct8x8_add(input, dest, stride, eob, bd);
} else {
vp10_highbd_iht8x8_64_add(input, dest, stride, tx_type, bd);
void vp10_highbd_inv_txfm_add_8x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
vp10_highbd_idct8x8_add(input, dest, stride, eob, bd);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_iht8x8_64_add(input, dest, stride, tx_type, bd);
break;
default:
assert(0);
break;
}
}
void vp10_highbd_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input,
uint8_t *dest, int stride, int eob, int bd) {
if (tx_type == DCT_DCT) {
vp10_highbd_idct16x16_add(input, dest, stride, eob, bd);
} else {
vp10_highbd_iht16x16_256_add(input, dest, stride, tx_type, bd);
void vp10_highbd_inv_txfm_add_16x16(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
vp10_highbd_idct16x16_add(input, dest, stride, eob, bd);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_iht16x16_256_add(input, dest, stride, tx_type, bd);
break;
default:
assert(0);
break;
}
}
void vp10_highbd_inv_txfm_add_32x32(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
vp10_highbd_idct32x32_add(input, dest, stride, eob, bd);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
assert(0);
break;
default:
assert(0);
break;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
vp10/common/idct.h
View file @ f1d090e2
......@@ -42,19 +42,17 @@ void vp10_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob);
void vp10_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob);
void vp10_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob);
void vp10_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob);
void vp10_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob);
void vp10_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
int stride, int eob);
void vp10_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
int stride, int eob);
void vp10_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
int stride, int eob);
void vp10_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type,
void (*itxm_add_4x4)(const tran_low_t *input,
uint8_t *dest, int stride, int eob));
void vp10_inv_txfm_add_8x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type);
void vp10_inv_txfm_add_16x16(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type);
void vp10_inv_txfm_add_32x32(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type);
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
......@@ -67,12 +65,19 @@ void vp10_highbd_idct16x16_add(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd);
void vp10_highbd_idct32x32_add(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd);
void vp10_highbd_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input,
uint8_t *dest, int stride, int eob, int bd);
void vp10_highbd_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input,
uint8_t *dest, int stride, int eob, int bd);
void vp10_highbd_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input,
uint8_t *dest, int stride, int eob, int bd);
void vp10_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type,
void (*highbd_itxm_add_4x4)
(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd));
void vp10_highbd_inv_txfm_add_8x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type);
void vp10_highbd_inv_txfm_add_16x16(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type);
void vp10_highbd_inv_txfm_add_32x32(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type);
#endif // CONFIG_VP9_HIGHBITDEPTH
#ifdef __cplusplus
} // extern "C"
......
vp10/decoder/decodeframe.c
View file @ f1d090e2
......@@ -186,76 +186,59 @@ static void read_mv_probs(nmv_context *ctx, int allow_hp, vpx_reader *r) {
static void inverse_transform_block_inter(MACROBLOCKD* xd, int plane,
const TX_SIZE tx_size,
uint8_t *dst, int stride,
int eob) {
int eob, int block) {
struct macroblockd_plane *const pd = &xd->plane[plane];
TX_TYPE tx_type = get_tx_type(pd->plane_type, xd, block);
if (eob > 0) {
tran_low_t *const dqcoeff = pd->dqcoeff;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
if (xd->lossless) {
vp10_highbd_iwht4x4_add(dqcoeff, dst, stride, eob, xd->bd);
} else {
switch (tx_size) {
case TX_4X4:
vp10_highbd_idct4x4_add(dqcoeff, dst, stride, eob, xd->bd);
break;
case TX_8X8:
vp10_highbd_idct8x8_add(dqcoeff, dst, stride, eob, xd->bd);
break;
case TX_16X16:
vp10_highbd_idct16x16_add(dqcoeff, dst, stride, eob, xd->bd);
break;
case TX_32X32:
vp10_highbd_idct32x32_add(dqcoeff, dst, stride, eob, xd->bd);
break;
default:
assert(0 && "Invalid transform size");
}
}
} else {
if (xd->lossless) {
vp10_iwht4x4_add(dqcoeff, dst, stride, eob);
} else {
switch (tx_size) {
case TX_4X4:
vp10_idct4x4_add(dqcoeff, dst, stride, eob);
break;
case TX_8X8:
vp10_idct8x8_add(dqcoeff, dst, stride, eob);
break;
case TX_16X16:
vp10_idct16x16_add(dqcoeff, dst, stride, eob);
break;
case TX_32X32:
vp10_idct32x32_add(dqcoeff, dst, stride, eob);
break;
default:
assert(0 && "Invalid transform size");
return;
}
switch (tx_size) {
case TX_4X4:
vp10_highbd_inv_txfm_add_4x4(dqcoeff, dst, stride, eob, xd->bd,
tx_type, xd->lossless ?
vp10_highbd_iwht4x4_add :
vp10_highbd_idct4x4_add);
break;
case TX_8X8:
vp10_highbd_inv_txfm_add_8x8(dqcoeff, dst, stride, eob, xd->bd,
tx_type);
break;
case TX_16X16:
vp10_highbd_inv_txfm_add_16x16(dqcoeff, dst, stride, eob, xd->bd,
tx_type);
break;
case TX_32X32:
vp10_highbd_inv_txfm_add_32x32(dqcoeff, dst, stride, eob, xd->bd,
tx_type);
break;
default:
assert(0 && "Invalid transform size");
return;
}
}
#else
if (xd->lossless) {
vp10_iwht4x4_add(dqcoeff, dst, stride, eob);
} else {
#else // CONFIG_VP9_HIGHBITDEPTH
switch (tx_size) {
case TX_4X4:
vp10_idct4x4_add(dqcoeff, dst, stride, eob);
vp10_inv_txfm_add_4x4(dqcoeff, dst, stride, eob, tx_type,
xd->lossless ? vp10_iwht4x4_add :
vp10_idct4x4_add);
break;
case TX_8X8:
vp10_idct8x8_add(dqcoeff, dst, stride, eob);
vp10_inv_txfm_add_8x8(dqcoeff, dst, stride, eob, tx_type);
break;
case TX_16X16:
vp10_idct16x16_add(dqcoeff, dst, stride, eob);
vp10_inv_txfm_add_16x16(dqcoeff, dst, stride, eob, tx_type);
break;
case TX_32X32:
vp10_idct32x32_add(dqcoeff, dst, stride, eob);
vp10_inv_txfm_add_32x32(dqcoeff, dst, stride, eob, tx_type);
break;
default:
assert(0 && "Invalid transform size");
return;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
#if CONFIG_VP9_HIGHBITDEPTH
}
#endif // CONFIG_VP9_HIGHBITDEPTH
......@@ -282,70 +265,52 @@ static void inverse_transform_block_intra(MACROBLOCKD* xd, int plane,
tran_low_t *const dqcoeff = pd->dqcoeff;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
if (xd->lossless) {
vp10_highbd_iwht4x4_add(dqcoeff, dst, stride, eob, xd->bd);
} else {
switch (tx_size) {
case TX_4X4:
vp10_highbd_iht4x4_add(tx_type, dqcoeff, dst, stride, eob, xd->bd);
break;
case TX_8X8:
vp10_highbd_iht8x8_add(tx_type, dqcoeff, dst, stride, eob, xd->bd);
break;
case TX_16X16:
vp10_highbd_iht16x16_add(tx_type, dqcoeff, dst, stride, eob, xd->bd);
break;
case TX_32X32:
vp10_highbd_idct32x32_add(dqcoeff, dst, stride, eob, xd->bd);
break;
default:
assert(0 && "Invalid transform size");
}
}
} else {
if (xd->lossless) {
vp10_iwht4x4_add(dqcoeff, dst, stride, eob);
} else {
switch (tx_size) {
case TX_4X4:
vp10_iht4x4_add(tx_type, dqcoeff, dst, stride, eob);
break;
case TX_8X8:
vp10_iht8x8_add(tx_type, dqcoeff, dst, stride, eob);
break;
case TX_16X16:
vp10_iht16x16_add(tx_type, dqcoeff, dst, stride, eob);
break;
case TX_32X32:
vp10_idct32x32_add(dqcoeff, dst, stride, eob);
break;
default:
assert(0 && "Invalid transform size");
return;
}
switch (tx_size) {
case TX_4X4:
vp10_highbd_inv_txfm_add_4x4(dqcoeff, dst, stride, eob, xd->bd,
tx_type, xd->lossless ?
vp10_highbd_iwht4x4_add :
vp10_highbd_idct4x4_add);
break;
case TX_8X8:
vp10_highbd_inv_txfm_add_8x8(dqcoeff, dst, stride, eob, xd->bd,
tx_type);
break;
case TX_16X16:
vp10_highbd_inv_txfm_add_16x16(dqcoeff, dst, stride, eob, xd->bd,
tx_type);
break;
case TX_32X32:
vp10_highbd_inv_txfm_add_32x32(dqcoeff, dst, stride, eob, xd->bd,
tx_type);
break;
default:
assert(0 && "Invalid transform size");
return;
}
}
#else
if (xd->lossless) {
vp10_iwht4x4_add(dqcoeff, dst, stride, eob);
} else {
#else // CONFIG_VP9_HIGHBITDEPTH
switch (tx_size) {
case TX_4X4:
vp10_iht4x4_add(tx_type, dqcoeff, dst, stride, eob);
vp10_inv_txfm_add_4x4(dqcoeff, dst, stride, eob, tx_type,
xd->lossless ? vp10_iwht4x4_add :
vp10_idct4x4_add);
break;
case TX_8X8:
vp10_iht8x8_add(tx_type, dqcoeff, dst, stride, eob);
vp10_inv_txfm_add_8x8(dqcoeff, dst, stride, eob, tx_type);
break;
case TX_16X16:
vp10_iht16x16_add(tx_type, dqcoeff, dst, stride, eob);
vp10_inv_txfm_add_16x16(dqcoeff, dst, stride, eob, tx_type);
break;
case TX_32X32:
vp10_idct32x32_add(dqcoeff, dst, stride, eob);
vp10_inv_txfm_add_32x32(dqcoeff, dst, stride, eob, tx_type);
break;
default:
assert(0 && "Invalid transform size");
return;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
#if CONFIG_VP9_HIGHBITDEPTH
}
#endif // CONFIG_VP9_HIGHBITDEPTH
......@@ -406,7 +371,7 @@ static int reconstruct_inter_block(MACROBLOCKD *const xd, vpx_reader *r,
inverse_transform_block_inter(xd, plane, tx_size,
&pd->dst.buf[4 * row * pd->dst.stride + 4 * col],
pd->dst.stride, eob);
pd->dst.stride, eob, block_idx);
return eob;
}
......
vp10/encoder/encodemb.c
View file @ f1d090e2
......@@ -496,6 +496,146 @@ void vp10_xform_quant_dc(MACROBLOCK *x, int plane, int block,
}
}
void vp10_fwd_txfm_4x4(const int16_t *src_diff,
tran_low_t *coeff, int diff_stride, TX_TYPE tx_type,
void (*fwd_txm4x4)(const int16_t *input,
tran_low_t *output, int stride)) {
switch (tx_type) {
case DCT_DCT:
fwd_txm4x4(src_diff, coeff, diff_stride);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_fht4x4(src_diff, coeff, diff_stride, tx_type);
break;
default:
assert(0);
break;
}
}
static void fwd_txfm_8x8(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_fht8x8(src_diff, coeff, diff_stride, tx_type);
break;
default:
assert(0);
break;
}
}
static void fwd_txfm_16x16(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_fht16x16(src_diff, coeff, diff_stride, tx_type);
break;
default:
assert(0);
break;
}
}
static void fwd_txfm_32x32(int rd_transform, const int16_t *src_diff,
tran_low_t *coeff, int diff_stride,
TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
fdct32x32(rd_transform, src_diff, coeff, diff_stride);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
assert(0);
break;
default:
assert(0);
break;
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
void (*highbd_fwd_txm4x4)(const int16_t *input,
tran_low_t *output, int stride)) {
switch (tx_type) {
case DCT_DCT:
highbd_fwd_txm4x4(src_diff, coeff, diff_stride);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_fht4x4(src_diff, coeff, diff_stride, tx_type);
break;
default:
assert(0);
break;
}
}
static void highbd_fwd_txfm_8x8(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
vpx_highbd_fdct8x8(src_diff, coeff, diff_stride);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_fht8x8(src_diff, coeff, diff_stride, tx_type);
break;
default:
assert(0);
break;
}
}
static void highbd_fwd_txfm_16x16(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
vpx_highbd_fdct16x16(src_diff, coeff, diff_stride);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_fht16x16(src_diff, coeff, diff_stride, tx_type);
break;
default:
assert(0);
break;
}
}
static void highbd_fwd_txfm_32x32(int rd_transform, const int16_t *src_diff,
tran_low_t *coeff, int diff_stride,
TX_TYPE tx_type) {
switch (tx_type) {
case DCT_DCT:
highbd_fdct32x32(rd_transform, src_diff, coeff, diff_stride);
break;
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
assert(0);
break;
default:
assert(0);
break;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
void vp10_xform_quant(MACROBLOCK *x, int plane, int block,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
MACROBLOCKD *const xd = &x->e_mbd;
......@@ -518,28 +658,30 @@ void vp10_xform_quant(MACROBLOCK *x, int plane, int block,
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
switch (tx_size) {
case TX_32X32:
highbd_fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
highbd_fwd_txfm_32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride,
tx_type);
vpx_highbd_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin,
p->round, p->quant, p->quant_shift, qcoeff,
dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
break;
case TX_16X16:
vpx_highbd_fdct16x16(src_diff, coeff, diff_stride);
highbd_fwd_txfm_16x16(src_diff, coeff, diff_stride, tx_type);
vpx_highbd_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,