Merge changes Iea45fd22,If174d8dd,I9f539491 into nextgenv2

* changes:
  Add facade to inverse txfm
  Create hybrid_fwd_txfm.c
  merge txfm_#x#_1 into txfm_#x#
This commit is contained in:
Angie Chiang 2015-12-03 22:29:03 +00:00 коммит произвёл Gerrit Code Review
Родитель 3e2273fcee a245d9f88c
Коммит 2b3f1d36b3
8 изменённых файлов: 633 добавлений и 667 удалений

Просмотреть файл

@ -1282,3 +1282,66 @@ void vp10_highbd_inv_txfm_add_32x32(const tran_low_t *input, uint8_t *dest,
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
void inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
INV_TXFM_PARAM *inv_txfm_param) {
const TX_TYPE tx_type = inv_txfm_param->tx_type;
const TX_SIZE tx_size = inv_txfm_param->tx_size;
const int eob = inv_txfm_param->eob;
const int lossless = inv_txfm_param->lossless;
switch (tx_size) {
case TX_32X32:
vp10_inv_txfm_add_32x32(input, dest, stride, eob, tx_type);
break;
case TX_16X16:
vp10_inv_txfm_add_16x16(input, dest, stride, eob, tx_type);
break;
case TX_8X8:
vp10_inv_txfm_add_8x8(input, dest, stride, eob, tx_type);
break;
case TX_4X4:
// this is like vp10_short_idct4x4 but has a special case around eob<=1
// which is significant (not just an optimization) for the lossless
// case.
vp10_inv_txfm_add_4x4(input, dest, stride, eob, tx_type,
lossless);
break;
default:
assert(0 && "Invalid transform size");
break;
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void highbd_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
INV_TXFM_PARAM *inv_txfm_param) {
const TX_TYPE tx_type = inv_txfm_param->tx_type;
const TX_SIZE tx_size = inv_txfm_param->tx_size;
const int eob = inv_txfm_param->eob;
const int bd = inv_txfm_param->bd;
const int lossless = inv_txfm_param->lossless;
switch (tx_size) {
case TX_32X32:
vp10_highbd_inv_txfm_add_32x32(input, dest, stride, eob, bd, tx_type);
break;
case TX_16X16:
vp10_highbd_inv_txfm_add_16x16(input, dest, stride, eob, bd, tx_type);
break;
case TX_8X8:
vp10_highbd_inv_txfm_add_8x8(input, dest, stride, eob, bd, tx_type);
break;
case TX_4X4:
// this is like vp10_short_idct4x4 but has a special case around eob<=1
// which is significant (not just an optimization) for the lossless
// case.
vp10_highbd_inv_txfm_add_4x4(input, dest, stride, eob, bd, tx_type,
lossless);
break;
default:
assert(0 && "Invalid transform size");
break;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH

Просмотреть файл

@ -24,6 +24,16 @@
extern "C" {
#endif
typedef struct INV_TXFM_PARAM {
TX_TYPE tx_type;
TX_SIZE tx_size;
int eob;
int lossless;
#if CONFIG_VP9_HIGHBITDEPTH
int bd;
#endif
} INV_TXFM_PARAM;
typedef void (*transform_1d)(const tran_low_t*, tran_low_t*);
typedef struct {
@ -51,7 +61,8 @@ 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);
void inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
INV_TXFM_PARAM *inv_txfm_param);
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd);
@ -74,6 +85,8 @@ void vp10_highbd_inv_txfm_add_16x16(const tran_low_t *input, uint8_t *dest,
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);
void highbd_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
INV_TXFM_PARAM *inv_txfm_param);
#endif // CONFIG_VP9_HIGHBITDEPTH
#ifdef __cplusplus
} // extern "C"

Разница между файлами не показана из-за своего большого размера Загрузить разницу

Просмотреть файл

@ -43,14 +43,6 @@ void vp10_encode_block_intra(int plane, int block, int blk_row, int blk_col,
void vp10_encode_intra_block_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane);
void vp10_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type, int lossless);
#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, int lossless);
#endif // CONFIG_VP9_HIGHBITDEPTH
#ifdef __cplusplus
} // extern "C"
#endif

Просмотреть файл

@ -0,0 +1,406 @@
/*
* Copyright (c) 2015 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "./vp10_rtcd.h"
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
#include "vp10/common/idct.h"
#include "vp10/encoder/hybrid_fwd_txfm.h"
static INLINE void fdct32x32(int rd_transform, const int16_t *src,
tran_low_t *dst, int src_stride) {
if (rd_transform)
vpx_fdct32x32_rd(src, dst, src_stride);
else
vpx_fdct32x32(src, dst, src_stride);
}
#if CONFIG_VP9_HIGHBITDEPTH
static INLINE void highbd_fdct32x32(int rd_transform, const int16_t *src,
tran_low_t *dst, int src_stride) {
if (rd_transform)
vpx_highbd_fdct32x32_rd(src, dst, src_stride);
else
vpx_highbd_fdct32x32(src, dst, src_stride);
}
#endif // CONFIG_VP9_HIGHBITDEPTH
#if CONFIG_EXT_TX
// Forward identity transform.
static void fwd_idtx_c(const int16_t *src_diff, tran_low_t *coeff, int stride,
int bs) {
int r, c;
const int shift = bs < 32 ? 3 : 2;
for (r = 0; r < bs; ++r) {
for (c = 0; c < bs; ++c) coeff[c] = src_diff[c] << shift;
src_diff += stride;
coeff += bs;
}
}
#endif // CONFIG_EXT_TX
void vp10_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type, int lossless) {
if (lossless) {
assert(tx_type == DCT_DCT);
vp10_fwht4x4(src_diff, coeff, diff_stride);
return;
}
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_fht4x4(src_diff, coeff, diff_stride, tx_type);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
case DCT_FLIPADST:
case FLIPADST_FLIPADST:
case ADST_FLIPADST:
case FLIPADST_ADST:
vp10_fht4x4(src_diff, coeff, diff_stride, tx_type);
break;
case DST_DST:
case DCT_DST:
case DST_DCT:
case DST_ADST:
case ADST_DST:
case DST_FLIPADST:
case FLIPADST_DST:
// Use C version since DST exists only in C
vp10_fht4x4_c(src_diff, coeff, diff_stride, tx_type);
break;
case IDTX:
fwd_idtx_c(src_diff, coeff, diff_stride, 4);
break;
#endif // CONFIG_EXT_TX
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,
FWD_TXFM_OPT fwd_txfm_opt) {
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
if (fwd_txfm_opt == FWD_TXFM_OPT_NORMAL)
vp10_fht8x8(src_diff, coeff, diff_stride, tx_type);
else // FWD_TXFM_OPT_DC
vpx_fdct8x8_1(src_diff, coeff, diff_stride);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
case DCT_FLIPADST:
case FLIPADST_FLIPADST:
case ADST_FLIPADST:
case FLIPADST_ADST:
vp10_fht8x8(src_diff, coeff, diff_stride, tx_type);
break;
case DST_DST:
case DCT_DST:
case DST_DCT:
case DST_ADST:
case ADST_DST:
case DST_FLIPADST:
case FLIPADST_DST:
// Use C version since DST exists only in C
vp10_fht8x8_c(src_diff, coeff, diff_stride, tx_type);
break;
case IDTX:
fwd_idtx_c(src_diff, coeff, diff_stride, 8);
break;
#endif // CONFIG_EXT_TX
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,
FWD_TXFM_OPT fwd_txfm_opt) {
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
if (fwd_txfm_opt == FWD_TXFM_OPT_NORMAL)
vp10_fht16x16(src_diff, coeff, diff_stride, tx_type);
else // FWD_TXFM_OPT_DC
vpx_fdct16x16_1(src_diff, coeff, diff_stride);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
case DCT_FLIPADST:
case FLIPADST_FLIPADST:
case ADST_FLIPADST:
case FLIPADST_ADST:
vp10_fht16x16(src_diff, coeff, diff_stride, tx_type);
break;
case DST_DST:
case DCT_DST:
case DST_DCT:
case DST_ADST:
case ADST_DST:
case DST_FLIPADST:
case FLIPADST_DST:
// Use C version since DST exists only in C
vp10_fht16x16_c(src_diff, coeff, diff_stride, tx_type);
break;
case IDTX:
fwd_idtx_c(src_diff, coeff, diff_stride, 16);
break;
#endif // CONFIG_EXT_TX
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,
FWD_TXFM_OPT fwd_txfm_opt) {
switch (tx_type) {
case DCT_DCT:
if (fwd_txfm_opt == FWD_TXFM_OPT_NORMAL)
fdct32x32(rd_transform, src_diff, coeff, diff_stride);
else // FWD_TXFM_OPT_DC
vpx_fdct32x32_1(src_diff, coeff, diff_stride);
break;
#if CONFIG_EXT_TX
case IDTX:
fwd_idtx_c(src_diff, coeff, diff_stride, 32);
break;
#endif // CONFIG_EXT_TX
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, int lossless) {
if (lossless) {
assert(tx_type == DCT_DCT);
vp10_highbd_fwht4x4(src_diff, coeff, diff_stride);
return;
}
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_fht4x4(src_diff, coeff, diff_stride, tx_type);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
case DCT_FLIPADST:
case FLIPADST_FLIPADST:
case ADST_FLIPADST:
case FLIPADST_ADST:
vp10_highbd_fht4x4(src_diff, coeff, diff_stride, tx_type);
break;
case DST_DST:
case DCT_DST:
case DST_DCT:
case DST_ADST:
case ADST_DST:
case DST_FLIPADST:
case FLIPADST_DST:
// Use C version since DST exists only in C
vp10_highbd_fht4x4_c(src_diff, coeff, diff_stride, tx_type);
break;
case IDTX:
fwd_idtx_c(src_diff, coeff, diff_stride, 4);
break;
#endif // CONFIG_EXT_TX
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,
FWD_TXFM_OPT fwd_txfm_opt) {
(void)fwd_txfm_opt;
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_fht8x8(src_diff, coeff, diff_stride, tx_type);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
case DCT_FLIPADST:
case FLIPADST_FLIPADST:
case ADST_FLIPADST:
case FLIPADST_ADST:
vp10_highbd_fht8x8(src_diff, coeff, diff_stride, tx_type);
break;
case DST_DST:
case DCT_DST:
case DST_DCT:
case DST_ADST:
case ADST_DST:
case DST_FLIPADST:
case FLIPADST_DST:
// Use C version since DST exists only in C
vp10_highbd_fht8x8_c(src_diff, coeff, diff_stride, tx_type);
break;
case IDTX:
fwd_idtx_c(src_diff, coeff, diff_stride, 8);
break;
#endif // CONFIG_EXT_TX
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,
FWD_TXFM_OPT fwd_txfm_opt) {
(void)fwd_txfm_opt;
switch (tx_type) {
case DCT_DCT:
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
vp10_highbd_fht16x16(src_diff, coeff, diff_stride, tx_type);
break;
#if CONFIG_EXT_TX
case FLIPADST_DCT:
case DCT_FLIPADST:
case FLIPADST_FLIPADST:
case ADST_FLIPADST:
case FLIPADST_ADST:
vp10_highbd_fht16x16(src_diff, coeff, diff_stride, tx_type);
break;
case DST_DST:
case DCT_DST:
case DST_DCT:
case DST_ADST:
case ADST_DST:
case DST_FLIPADST:
case FLIPADST_DST:
// Use C version since DST exists only in C
vp10_highbd_fht16x16_c(src_diff, coeff, diff_stride, tx_type);
break;
case IDTX:
fwd_idtx_c(src_diff, coeff, diff_stride, 16);
break;
#endif // CONFIG_EXT_TX
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, FWD_TXFM_OPT fwd_txfm_opt) {
switch (tx_type) {
case DCT_DCT:
if (fwd_txfm_opt == FWD_TXFM_OPT_NORMAL)
highbd_fdct32x32(rd_transform, src_diff, coeff, diff_stride);
else // FWD_TXFM_OPT_DC
vpx_highbd_fdct32x32_1(src_diff, coeff, diff_stride);
break;
#if CONFIG_EXT_TX
case IDTX:
fwd_idtx_c(src_diff, coeff, diff_stride, 32);
break;
#endif // CONFIG_EXT_TX
case ADST_DCT:
case DCT_ADST:
case ADST_ADST:
assert(0);
break;
default:
assert(0);
break;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
void fwd_txfm(const int16_t *src_diff, tran_low_t *coeff, int diff_stride,
FWD_TXFM_PARAM *fwd_txfm_param) {
const int fwd_txfm_opt = fwd_txfm_param->fwd_txfm_opt;
const TX_TYPE tx_type = fwd_txfm_param->tx_type;
const TX_SIZE tx_size = fwd_txfm_param->tx_size;
const int rd_transform = fwd_txfm_param->rd_transform;
const int lossless = fwd_txfm_param->lossless;
switch (tx_size) {
case TX_32X32:
fwd_txfm_32x32(rd_transform, src_diff, coeff, diff_stride, tx_type,
fwd_txfm_opt);
break;
case TX_16X16:
fwd_txfm_16x16(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt);
break;
case TX_8X8:
fwd_txfm_8x8(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt);
break;
case TX_4X4:
vp10_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type, lossless);
break;
default:
assert(0);
break;
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void highbd_fwd_txfm(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, FWD_TXFM_PARAM *fwd_txfm_param) {
const int fwd_txfm_opt = fwd_txfm_param->fwd_txfm_opt;
const TX_TYPE tx_type = fwd_txfm_param->tx_type;
const TX_SIZE tx_size = fwd_txfm_param->tx_size;
const int rd_transform = fwd_txfm_param->rd_transform;
const int lossless = fwd_txfm_param->lossless;
switch (tx_size) {
case TX_32X32:
highbd_fwd_txfm_32x32(rd_transform, src_diff, coeff, diff_stride, tx_type,
fwd_txfm_opt);
break;
case TX_16X16:
highbd_fwd_txfm_16x16(src_diff, coeff, diff_stride, tx_type,
fwd_txfm_opt);
break;
case TX_8X8:
highbd_fwd_txfm_8x8(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt);
break;
case TX_4X4:
vp10_highbd_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type, lossless);
break;
default:
assert(0);
break;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH

Просмотреть файл

@ -0,0 +1,62 @@
/*
* Copyright (c) 2015 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP10_ENCODER_HYBRID_FWD_TXFM_H_
#define VP10_ENCODER_HYBRID_FWD_TXFM_H_
#include "./vpx_config.h"
typedef enum FWD_TXFM_OPT { FWD_TXFM_OPT_NORMAL, FWD_TXFM_OPT_DC } FWD_TXFM_OPT;
typedef struct FWD_TXFM_PARAM {
TX_TYPE tx_type;
TX_SIZE tx_size;
FWD_TXFM_OPT fwd_txfm_opt;
int rd_transform;
int lossless;
} FWD_TXFM_PARAM;
#ifdef __cplusplus
extern "C" {
#endif
void fwd_txfm(const int16_t *src_diff, tran_low_t *coeff, int diff_stride,
FWD_TXFM_PARAM *fwd_txfm_param);
void vp10_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type, int lossless);
#if CONFIG_VP9_HIGHBITDEPTH
void highbd_fwd_txfm(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, FWD_TXFM_PARAM *fwd_txfm_param);
void vp10_highbd_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type, int lossless);
#endif // CONFIG_VP9_HIGHBITDEPTH
static INLINE int get_tx1d_size(TX_SIZE tx_size) {
switch (tx_size) {
case TX_32X32:
return 32;
case TX_16X16:
return 16;
case TX_8X8:
return 8;
case TX_4X4:
return 4;
default:
assert(0);
return -1;
}
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP10_ENCODER_HYBRID_FWD_TXFM_H_

Просмотреть файл

@ -35,6 +35,7 @@
#include "vp10/encoder/encodemb.h"
#include "vp10/encoder/encodemv.h"
#include "vp10/encoder/encoder.h"
#include "vp10/encoder/hybrid_fwd_txfm.h"
#include "vp10/encoder/mcomp.h"
#include "vp10/encoder/palette.h"
#include "vp10/encoder/quantize.h"

Просмотреть файл

@ -24,6 +24,8 @@ VP10_CX_SRCS-yes += encoder/context_tree.h
VP10_CX_SRCS-yes += encoder/cost.h
VP10_CX_SRCS-yes += encoder/cost.c
VP10_CX_SRCS-yes += encoder/dct.c
VP10_CX_SRCS-yes += encoder/hybrid_fwd_txfm.c
VP10_CX_SRCS-yes += encoder/hybrid_fwd_txfm.h
VP10_CX_SRCS-$(CONFIG_VP9_TEMPORAL_DENOISING) += encoder/denoiser.c
VP10_CX_SRCS-$(CONFIG_VP9_TEMPORAL_DENOISING) += encoder/denoiser.h
VP10_CX_SRCS-yes += encoder/encodeframe.c