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

Fork VP9 and VP10 codebase

Browse Source 
This commit folks the VP9 and VP10 codebase and makes libvpx
support VP8, VP9, and VP10.

Change-Id: I81782e0b809acb3c9844bee8c8ec8f4d5e8fa356
This commit is contained in:
Jingning Han 2015-08-05 19:00:31 -07:00
parent b04dad328c
commit 3ee6db6c81
200 changed files with 67352 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
configure vendored
View File

@ -37,6 +37,7 @@ Advanced options:
${toggle_vp9_highbitdepth} use VP9 high bit depth (10/12) profiles
${toggle_vp8} VP8 codec support
${toggle_vp9} VP9 codec support
${toggle_vp10} VP10 codec support
${toggle_internal_stats} output of encoder internal stats for debug, if supported (encoders)
${toggle_postproc} postprocessing
${toggle_vp9_postproc} vp9 specific postprocessing
@ -191,6 +192,7 @@ fi
# disable codecs when their source directory does not exist
[ -d "${source_path}/vp8" ] || disable_feature vp8
[ -d "${source_path}/vp9" ] || disable_feature vp9
[ -d "${source_path}/vp10" ] || disable_feature vp10
# install everything except the sources, by default. sources will have
# to be enabled when doing dist builds, since that's no longer a common
@ -212,10 +214,13 @@ CODECS="
vp8_decoder
vp9_encoder
vp9_decoder
vp10_encoder
vp10_decoder
"
CODEC_FAMILIES="
vp8
vp9
vp10
"
ARCH_LIST="

34
libs.mk
View File

@ -109,6 +109,40 @@ endif
VP9_PREFIX=vp9/
$(BUILD_PFX)$(VP9_PREFIX)%.c.o: CFLAGS += -Wextra
# VP10 make file
ifneq ($(CONFIG_VP10_ENCODER)$(CONFIG_VP10_DECODER),)
VP10_PREFIX=vp10/
include $(SRC_PATH_BARE)/$(VP10_PREFIX)vp10_common.mk
endif
ifeq ($(CONFIG_VP10_ENCODER),yes)
VP10_PREFIX=vp10/
include $(SRC_PATH_BARE)/$(VP10_PREFIX)vp10cx.mk
CODEC_SRCS-yes += $(addprefix $(VP10_PREFIX),$(call enabled,VP10_CX_SRCS))
CODEC_EXPORTS-yes += $(addprefix $(VP10_PREFIX),$(VP10_CX_EXPORTS))
CODEC_SRCS-yes += $(VP10_PREFIX)vp10cx.mk vpx/vp8.h vpx/vp8cx.h
INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8cx.h
INSTALL-LIBS-$(CONFIG_SPATIAL_SVC) += include/vpx/svc_context.h
INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP10_PREFIX)/%
CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8cx.h
CODEC_DOC_SECTIONS += vp9 vp9_encoder
endif
ifeq ($(CONFIG_VP10_DECODER),yes)
VP10_PREFIX=vp10/
include $(SRC_PATH_BARE)/$(VP10_PREFIX)vp10dx.mk
CODEC_SRCS-yes += $(addprefix $(VP10_PREFIX),$(call enabled,VP10_DX_SRCS))
CODEC_EXPORTS-yes += $(addprefix $(VP10_PREFIX),$(VP10_DX_EXPORTS))
CODEC_SRCS-yes += $(VP10_PREFIX)vp10dx.mk vpx/vp8.h vpx/vp8dx.h
INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8dx.h
INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP10_PREFIX)/%
CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8dx.h
CODEC_DOC_SECTIONS += vp9 vp9_decoder
endif
VP10_PREFIX=vp10/
$(BUILD_PFX)$(VP10_PREFIX)%.c.o: CFLAGS += -Wextra
ifeq ($(CONFIG_ENCODERS),yes)
CODEC_DOC_SECTIONS += encoder
endif

97
test/codec_factory.h
View File

@ -13,10 +13,10 @@
#include "./vpx_config.h"
#include "vpx/vpx_decoder.h"
#include "vpx/vpx_encoder.h"
#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
#include "vpx/vp8cx.h"
#endif
#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER
#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER || CONFIG_VP10_DECODER
#include "vpx/vp8dx.h"
#endif
@ -233,6 +233,8 @@ class VP9CodecFactory : public CodecFactory {
int usage) const {
#if CONFIG_VP9_ENCODER
return vpx_codec_enc_config_default(&vpx_codec_vp9_cx_algo, cfg, usage);
#elif CONFIG_VP10_ENCODER
return vpx_codec_enc_config_default(&vpx_codec_vp10_cx_algo, cfg, usage);
#else
return VPX_CODEC_INCAPABLE;
#endif
@ -251,7 +253,96 @@ const libvpx_test::VP9CodecFactory kVP9;
#define VP9_INSTANTIATE_TEST_CASE(test, ...)
#endif // CONFIG_VP9
/*
* VP10 Codec Definitions
*/
#if CONFIG_VP10
class VP10Decoder : public Decoder {
public:
VP10Decoder(vpx_codec_dec_cfg_t cfg, unsigned long deadline)
: Decoder(cfg, deadline) {}
VP10Decoder(vpx_codec_dec_cfg_t cfg, const vpx_codec_flags_t flag,
unsigned long deadline) // NOLINT
: Decoder(cfg, flag, deadline) {}
protected:
virtual vpx_codec_iface_t* CodecInterface() const {
#if CONFIG_VP10_DECODER
return &vpx_codec_vp10_dx_algo;
#else
return NULL;
#endif
}
};
class VP10Encoder : public Encoder {
public:
VP10Encoder(vpx_codec_enc_cfg_t cfg, unsigned long deadline,
const unsigned long init_flags, TwopassStatsStore *stats)
: Encoder(cfg, deadline, init_flags, stats) {}
protected:
virtual vpx_codec_iface_t* CodecInterface() const {
#if CONFIG_VP10_ENCODER
return &vpx_codec_vp10_cx_algo;
#else
return NULL;
#endif
}
};
class VP10CodecFactory : public CodecFactory {
public:
VP10CodecFactory() : CodecFactory() {}
virtual Decoder* CreateDecoder(vpx_codec_dec_cfg_t cfg,
unsigned long deadline) const {
return CreateDecoder(cfg, 0, deadline);
}
virtual Decoder* CreateDecoder(vpx_codec_dec_cfg_t cfg,
const vpx_codec_flags_t flags,
unsigned long deadline) const { // NOLINT
#if CONFIG_VP10_DECODER
return new VP10Decoder(cfg, flags, deadline);
#else
return NULL;
#endif
}
virtual Encoder* CreateEncoder(vpx_codec_enc_cfg_t cfg,
unsigned long deadline,
const unsigned long init_flags,
TwopassStatsStore *stats) const {
#if CONFIG_VP10_ENCODER
return new VP10Encoder(cfg, deadline, init_flags, stats);
#else
return NULL;
#endif
}
virtual vpx_codec_err_t DefaultEncoderConfig(vpx_codec_enc_cfg_t *cfg,
int usage) const {
#if CONFIG_VP10_ENCODER
return vpx_codec_enc_config_default(&vpx_codec_vp10_cx_algo, cfg, usage);
#else
return VPX_CODEC_INCAPABLE;
#endif
}
};
const libvpx_test::VP10CodecFactory kVP10;
#define VP10_INSTANTIATE_TEST_CASE(test, ...)\
INSTANTIATE_TEST_CASE_P(VP10, test, \
::testing::Combine( \
::testing::Values(static_cast<const libvpx_test::CodecFactory*>( \
&libvpx_test::kVP10)), \
__VA_ARGS__))
#else
#define VP10_INSTANTIATE_TEST_CASE(test, ...)
#endif // CONFIG_VP10
} // namespace libvpx_test
#endif // TEST_CODEC_FACTORY_H_

3
test/decode_api_test.cc
View File

@ -26,6 +26,9 @@ TEST(DecodeAPI, InvalidParams) {
#endif
#if CONFIG_VP9_DECODER
&vpx_codec_vp9_dx_algo,
#endif
#if CONFIG_VP10_DECODER
&vpx_codec_vp10_dx_algo,
#endif
};
uint8_t buf[1] = {0};

4
test/encode_test_driver.h
View File

@ -16,7 +16,7 @@
#include "third_party/googletest/src/include/gtest/gtest.h"
#include "./vpx_config.h"
#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
#include "vpx/vp8cx.h"
#endif
#include "vpx/vpx_encoder.h"
@ -138,7 +138,7 @@ class Encoder {
const vpx_codec_err_t res = vpx_codec_control_(&encoder_, ctrl_id, arg);
ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
}
#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
void Control(int ctrl_id, vpx_active_map_t *arg) {
const vpx_codec_err_t res = vpx_codec_control_(&encoder_, ctrl_id, arg);
ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();

12
tools_common.c
View File

@ -16,11 +16,11 @@
#include "./tools_common.h"
#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
#include "vpx/vp8cx.h"
#endif
#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER
#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER || CONFIG_VP10_DECODER
#include "vpx/vp8dx.h"
#endif
@ -138,6 +138,10 @@ static const VpxInterface vpx_encoders[] = {
#if CONFIG_VP9_ENCODER
{"vp9", VP9_FOURCC, &vpx_codec_vp9_cx},
#endif
#if CONFIG_VP10_ENCODER
{"vp10", VP10_FOURCC, &vpx_codec_vp10_cx},
#endif
};
int get_vpx_encoder_count(void) {
@ -168,6 +172,10 @@ static const VpxInterface vpx_decoders[] = {
#if CONFIG_VP9_DECODER
{"vp9", VP9_FOURCC, &vpx_codec_vp9_dx},
#endif
#if CONFIG_VP10_DECODER
{"vp10", VP10_FOURCC, &vpx_codec_vp10_dx},
#endif
};
int get_vpx_decoder_count(void) {

1
tools_common.h
View File

@ -62,6 +62,7 @@
#define VP8_FOURCC 0x30385056
#define VP9_FOURCC 0x30395056
#define VP10_FOURCC 0x303a5056
enum VideoFileType {
FILE_TYPE_RAW,

248
vp10/common/arm/neon/vp9_iht4x4_add_neon.c Normal file
View File

@ -0,0 +1,248 @@
/*
* Copyright (c) 2014 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 <arm_neon.h>
#include <assert.h>
#include "./vp10_rtcd.h"
#include "./vpx_config.h"
#include "vp10/common/vp9_common.h"
static int16_t sinpi_1_9 = 0x14a3;
static int16_t sinpi_2_9 = 0x26c9;
static int16_t sinpi_3_9 = 0x3441;
static int16_t sinpi_4_9 = 0x3b6c;
static int16_t cospi_8_64 = 0x3b21;
static int16_t cospi_16_64 = 0x2d41;
static int16_t cospi_24_64 = 0x187e;
static INLINE void TRANSPOSE4X4(
int16x8_t *q8s16,
int16x8_t *q9s16) {
int32x4_t q8s32, q9s32;
int16x4x2_t d0x2s16, d1x2s16;
int32x4x2_t q0x2s32;
d0x2s16 = vtrn_s16(vget_low_s16(*q8s16), vget_high_s16(*q8s16));
d1x2s16 = vtrn_s16(vget_low_s16(*q9s16), vget_high_s16(*q9s16));
q8s32 = vreinterpretq_s32_s16(vcombine_s16(d0x2s16.val[0], d0x2s16.val[1]));
q9s32 = vreinterpretq_s32_s16(vcombine_s16(d1x2s16.val[0], d1x2s16.val[1]));
q0x2s32 = vtrnq_s32(q8s32, q9s32);
*q8s16 = vreinterpretq_s16_s32(q0x2s32.val[0]);
*q9s16 = vreinterpretq_s16_s32(q0x2s32.val[1]);
return;
}
static INLINE void GENERATE_COSINE_CONSTANTS(
int16x4_t *d0s16,
int16x4_t *d1s16,
int16x4_t *d2s16) {
*d0s16 = vdup_n_s16(cospi_8_64);
*d1s16 = vdup_n_s16(cospi_16_64);
*d2s16 = vdup_n_s16(cospi_24_64);
return;
}
static INLINE void GENERATE_SINE_CONSTANTS(
int16x4_t *d3s16,
int16x4_t *d4s16,
int16x4_t *d5s16,
int16x8_t *q3s16) {
*d3s16 = vdup_n_s16(sinpi_1_9);
*d4s16 = vdup_n_s16(sinpi_2_9);
*q3s16 = vdupq_n_s16(sinpi_3_9);
*d5s16 = vdup_n_s16(sinpi_4_9);
return;
}
static INLINE void IDCT4x4_1D(
int16x4_t *d0s16,
int16x4_t *d1s16,
int16x4_t *d2s16,
int16x8_t *q8s16,
int16x8_t *q9s16) {
int16x4_t d16s16, d17s16, d18s16, d19s16, d23s16, d24s16;
int16x4_t d26s16, d27s16, d28s16, d29s16;
int32x4_t q10s32, q13s32, q14s32, q15s32;
int16x8_t q13s16, q14s16;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d23s16 = vadd_s16(d16s16, d18s16);
d24s16 = vsub_s16(d16s16, d18s16);
q15s32 = vmull_s16(d17s16, *d2s16);
q10s32 = vmull_s16(d17s16, *d0s16);
q13s32 = vmull_s16(d23s16, *d1s16);
q14s32 = vmull_s16(d24s16, *d1s16);
q15s32 = vmlsl_s16(q15s32, d19s16, *d0s16);
q10s32 = vmlal_s16(q10s32, d19s16, *d2s16);
d26s16 = vqrshrn_n_s32(q13s32, 14);
d27s16 = vqrshrn_n_s32(q14s32, 14);
d29s16 = vqrshrn_n_s32(q15s32, 14);
d28s16 = vqrshrn_n_s32(q10s32, 14);
q13s16 = vcombine_s16(d26s16, d27s16);
q14s16 = vcombine_s16(d28s16, d29s16);
*q8s16 = vaddq_s16(q13s16, q14s16);
*q9s16 = vsubq_s16(q13s16, q14s16);
*q9s16 = vcombine_s16(vget_high_s16(*q9s16),
vget_low_s16(*q9s16)); // vswp
return;
}
static INLINE void IADST4x4_1D(
int16x4_t *d3s16,
int16x4_t *d4s16,
int16x4_t *d5s16,
int16x8_t *q3s16,
int16x8_t *q8s16,
int16x8_t *q9s16) {
int16x4_t d6s16, d16s16, d17s16, d18s16, d19s16;
int32x4_t q8s32, q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32;
d6s16 = vget_low_s16(*q3s16);
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
q10s32 = vmull_s16(*d3s16, d16s16);
q11s32 = vmull_s16(*d4s16, d16s16);
q12s32 = vmull_s16(d6s16, d17s16);
q13s32 = vmull_s16(*d5s16, d18s16);
q14s32 = vmull_s16(*d3s16, d18s16);
q15s32 = vmovl_s16(d16s16);
q15s32 = vaddw_s16(q15s32, d19s16);
q8s32 = vmull_s16(*d4s16, d19s16);
q15s32 = vsubw_s16(q15s32, d18s16);
q9s32 = vmull_s16(*d5s16, d19s16);
q10s32 = vaddq_s32(q10s32, q13s32);
q10s32 = vaddq_s32(q10s32, q8s32);
q11s32 = vsubq_s32(q11s32, q14s32);
q8s32 = vdupq_n_s32(sinpi_3_9);
q11s32 = vsubq_s32(q11s32, q9s32);
q15s32 = vmulq_s32(q15s32, q8s32);
q13s32 = vaddq_s32(q10s32, q12s32);
q10s32 = vaddq_s32(q10s32, q11s32);
q14s32 = vaddq_s32(q11s32, q12s32);
q10s32 = vsubq_s32(q10s32, q12s32);
d16s16 = vqrshrn_n_s32(q13s32, 14);
d17s16 = vqrshrn_n_s32(q14s32, 14);
d18s16 = vqrshrn_n_s32(q15s32, 14);
d19s16 = vqrshrn_n_s32(q10s32, 14);
*q8s16 = vcombine_s16(d16s16, d17s16);
*q9s16 = vcombine_s16(d18s16, d19s16);
return;
}
void vp10_iht4x4_16_add_neon(const tran_low_t *input, uint8_t *dest,
int dest_stride, int tx_type) {
uint8x8_t d26u8, d27u8;
int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16;
uint32x2_t d26u32, d27u32;
int16x8_t q3s16, q8s16, q9s16;
uint16x8_t q8u16, q9u16;
d26u32 = d27u32 = vdup_n_u32(0);
q8s16 = vld1q_s16(input);
q9s16 = vld1q_s16(input + 8);
TRANSPOSE4X4(&q8s16, &q9s16);
switch (tx_type) {
case 0: // idct_idct is not supported. Fall back to C
vp10_iht4x4_16_add_c(input, dest, dest_stride, tx_type);
return;
break;
case 1: // iadst_idct
// generate constants
GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16);
GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
// first transform rows
IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16);
// transpose the matrix
TRANSPOSE4X4(&q8s16, &q9s16);
// then transform columns
IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
break;
case 2: // idct_iadst
// generate constantsyy
GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16);
GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
// first transform rows
IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
// transpose the matrix
TRANSPOSE4X4(&q8s16, &q9s16);
// then transform columns
IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16);
break;
case 3: // iadst_iadst
// generate constants
GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
// first transform rows
IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
// transpose the matrix
TRANSPOSE4X4(&q8s16, &q9s16);
// then transform columns
IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
break;
default: // iadst_idct
assert(0);
break;
}
q8s16 = vrshrq_n_s16(q8s16, 4);
q9s16 = vrshrq_n_s16(q9s16, 4);
d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 0);
dest += dest_stride;
d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 1);
dest += dest_stride;
d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 0);
dest += dest_stride;
d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 1);
q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), vreinterpret_u8_u32(d26u32));
q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), vreinterpret_u8_u32(d27u32));
d26u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16));
d27u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16));
vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 1);
dest -= dest_stride;
vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 0);
dest -= dest_stride;
vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 1);
dest -= dest_stride;
vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 0);
return;
}

624
vp10/common/arm/neon/vp9_iht8x8_add_neon.c Normal file
View File

@ -0,0 +1,624 @@
/*
* Copyright (c) 2014 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 <arm_neon.h>
#include <assert.h>
#include "./vp10_rtcd.h"
#include "./vpx_config.h"
#include "vp10/common/vp9_common.h"
static int16_t cospi_2_64 = 16305;
static int16_t cospi_4_64 = 16069;
static int16_t cospi_6_64 = 15679;
static int16_t cospi_8_64 = 15137;
static int16_t cospi_10_64 = 14449;
static int16_t cospi_12_64 = 13623;
static int16_t cospi_14_64 = 12665;
static int16_t cospi_16_64 = 11585;
static int16_t cospi_18_64 = 10394;
static int16_t cospi_20_64 = 9102;
static int16_t cospi_22_64 = 7723;
static int16_t cospi_24_64 = 6270;
static int16_t cospi_26_64 = 4756;
static int16_t cospi_28_64 = 3196;
static int16_t cospi_30_64 = 1606;
static INLINE void TRANSPOSE8X8(
int16x8_t *q8s16,
int16x8_t *q9s16,
int16x8_t *q10s16,
int16x8_t *q11s16,
int16x8_t *q12s16,
int16x8_t *q13s16,
int16x8_t *q14s16,
int16x8_t *q15s16) {
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int32x4x2_t q0x2s32, q1x2s32, q2x2s32, q3x2s32;
int16x8x2_t q0x2s16, q1x2s16, q2x2s16, q3x2s16;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
*q8s16 = vcombine_s16(d16s16, d24s16); // vswp d17, d24
*q9s16 = vcombine_s16(d18s16, d26s16); // vswp d19, d26
*q10s16 = vcombine_s16(d20s16, d28s16); // vswp d21, d28
*q11s16 = vcombine_s16(d22s16, d30s16); // vswp d23, d30
*q12s16 = vcombine_s16(d17s16, d25s16);
*q13s16 = vcombine_s16(d19s16, d27s16);
*q14s16 = vcombine_s16(d21s16, d29s16);
*q15s16 = vcombine_s16(d23s16, d31s16);
q0x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q8s16),
vreinterpretq_s32_s16(*q10s16));
q1x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q9s16),
vreinterpretq_s32_s16(*q11s16));
q2x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q12s16),
vreinterpretq_s32_s16(*q14s16));
q3x2s32 = vtrnq_s32(vreinterpretq_s32_s16(*q13s16),
vreinterpretq_s32_s16(*q15s16));
q0x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[0]), // q8
vreinterpretq_s16_s32(q1x2s32.val[0])); // q9
q1x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q0x2s32.val[1]), // q10
vreinterpretq_s16_s32(q1x2s32.val[1])); // q11
q2x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[0]), // q12
vreinterpretq_s16_s32(q3x2s32.val[0])); // q13
q3x2s16 = vtrnq_s16(vreinterpretq_s16_s32(q2x2s32.val[1]), // q14
vreinterpretq_s16_s32(q3x2s32.val[1])); // q15
*q8s16 = q0x2s16.val[0];
*q9s16 = q0x2s16.val[1];
*q10s16 = q1x2s16.val[0];
*q11s16 = q1x2s16.val[1];
*q12s16 = q2x2s16.val[0];
*q13s16 = q2x2s16.val[1];
*q14s16 = q3x2s16.val[0];
*q15s16 = q3x2s16.val[1];
return;
}
static INLINE void IDCT8x8_1D(
int16x8_t *q8s16,
int16x8_t *q9s16,
int16x8_t *q10s16,
int16x8_t *q11s16,
int16x8_t *q12s16,
int16x8_t *q13s16,
int16x8_t *q14s16,
int16x8_t *q15s16) {
int16x4_t d0s16, d1s16, d2s16, d3s16;
int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int16x8_t q0s16, q1s16, q2s16, q3s16, q4s16, q5s16, q6s16, q7s16;
int32x4_t q2s32, q3s32, q5s32, q6s32, q8s32, q9s32;
int32x4_t q10s32, q11s32, q12s32, q13s32, q15s32;
d0s16 = vdup_n_s16(cospi_28_64);
d1s16 = vdup_n_s16(cospi_4_64);
d2s16 = vdup_n_s16(cospi_12_64);
d3s16 = vdup_n_s16(cospi_20_64);
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
q2s32 = vmull_s16(d18s16, d0s16);
q3s32 = vmull_s16(d19s16, d0s16);
q5s32 = vmull_s16(d26s16, d2s16);
q6s32 = vmull_s16(d27s16, d2s16);
q2s32 = vmlsl_s16(q2s32, d30s16, d1s16);
q3s32 = vmlsl_s16(q3s32, d31s16, d1s16);
q5s32 = vmlsl_s16(q5s32, d22s16, d3s16);
q6s32 = vmlsl_s16(q6s32, d23s16, d3s16);
d8s16 = vqrshrn_n_s32(q2s32, 14);
d9s16 = vqrshrn_n_s32(q3s32, 14);
d10s16 = vqrshrn_n_s32(q5s32, 14);
d11s16 = vqrshrn_n_s32(q6s32, 14);
q4s16 = vcombine_s16(d8s16, d9s16);
q5s16 = vcombine_s16(d10s16, d11s16);
q2s32 = vmull_s16(d18s16, d1s16);
q3s32 = vmull_s16(d19s16, d1s16);
q9s32 = vmull_s16(d26s16, d3s16);
q13s32 = vmull_s16(d27s16, d3s16);
q2s32 = vmlal_s16(q2s32, d30s16, d0s16);
q3s32 = vmlal_s16(q3s32, d31s16, d0s16);
q9s32 = vmlal_s16(q9s32, d22s16, d2s16);
q13s32 = vmlal_s16(q13s32, d23s16, d2s16);
d14s16 = vqrshrn_n_s32(q2s32, 14);
d15s16 = vqrshrn_n_s32(q3s32, 14);
d12s16 = vqrshrn_n_s32(q9s32, 14);
d13s16 = vqrshrn_n_s32(q13s32, 14);
q6s16 = vcombine_s16(d12s16, d13s16);
q7s16 = vcombine_s16(d14s16, d15s16);
d0s16 = vdup_n_s16(cospi_16_64);
q2s32 = vmull_s16(d16s16, d0s16);
q3s32 = vmull_s16(d17s16, d0s16);
q13s32 = vmull_s16(d16s16, d0s16);
q15s32 = vmull_s16(d17s16, d0s16);
q2s32 = vmlal_s16(q2s32, d24s16, d0s16);
q3s32 = vmlal_s16(q3s32, d25s16, d0s16);
q13s32 = vmlsl_s16(q13s32, d24s16, d0s16);
q15s32 = vmlsl_s16(q15s32, d25s16, d0s16);
d0s16 = vdup_n_s16(cospi_24_64);
d1s16 = vdup_n_s16(cospi_8_64);
d18s16 = vqrshrn_n_s32(q2s32, 14);
d19s16 = vqrshrn_n_s32(q3s32, 14);
d22s16 = vqrshrn_n_s32(q13s32, 14);
d23s16 = vqrshrn_n_s32(q15s32, 14);
*q9s16 = vcombine_s16(d18s16, d19s16);
*q11s16 = vcombine_s16(d22s16, d23s16);
q2s32 = vmull_s16(d20s16, d0s16);
q3s32 = vmull_s16(d21s16, d0s16);
q8s32 = vmull_s16(d20s16, d1s16);
q12s32 = vmull_s16(d21s16, d1s16);
q2s32 = vmlsl_s16(q2s32, d28s16, d1s16);
q3s32 = vmlsl_s16(q3s32, d29s16, d1s16);
q8s32 = vmlal_s16(q8s32, d28s16, d0s16);
q12s32 = vmlal_s16(q12s32, d29s16, d0s16);
d26s16 = vqrshrn_n_s32(q2s32, 14);
d27s16 = vqrshrn_n_s32(q3s32, 14);
d30s16 = vqrshrn_n_s32(q8s32, 14);
d31s16 = vqrshrn_n_s32(q12s32, 14);
*q13s16 = vcombine_s16(d26s16, d27s16);
*q15s16 = vcombine_s16(d30s16, d31s16);
q0s16 = vaddq_s16(*q9s16, *q15s16);
q1s16 = vaddq_s16(*q11s16, *q13s16);
q2s16 = vsubq_s16(*q11s16, *q13s16);
q3s16 = vsubq_s16(*q9s16, *q15s16);
*q13s16 = vsubq_s16(q4s16, q5s16);
q4s16 = vaddq_s16(q4s16, q5s16);
*q14s16 = vsubq_s16(q7s16, q6s16);
q7s16 = vaddq_s16(q7s16, q6s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d16s16 = vdup_n_s16(cospi_16_64);
q9s32 = vmull_s16(d28s16, d16s16);
q10s32 = vmull_s16(d29s16, d16s16);
q11s32 = vmull_s16(d28s16, d16s16);
q12s32 = vmull_s16(d29s16, d16s16);
q9s32 = vmlsl_s16(q9s32, d26s16, d16s16);
q10s32 = vmlsl_s16(q10s32, d27s16, d16s16);
q11s32 = vmlal_s16(q11s32, d26s16, d16s16);
q12s32 = vmlal_s16(q12s32, d27s16, d16s16);
d10s16 = vqrshrn_n_s32(q9s32, 14);
d11s16 = vqrshrn_n_s32(q10s32, 14);
d12s16 = vqrshrn_n_s32(q11s32, 14);
d13s16 = vqrshrn_n_s32(q12s32, 14);
q5s16 = vcombine_s16(d10s16, d11s16);
q6s16 = vcombine_s16(d12s16, d13s16);
*q8s16 = vaddq_s16(q0s16, q7s16);
*q9s16 = vaddq_s16(q1s16, q6s16);
*q10s16 = vaddq_s16(q2s16, q5s16);
*q11s16 = vaddq_s16(q3s16, q4s16);
*q12s16 = vsubq_s16(q3s16, q4s16);
*q13s16 = vsubq_s16(q2s16, q5s16);
*q14s16 = vsubq_s16(q1s16, q6s16);
*q15s16 = vsubq_s16(q0s16, q7s16);
return;
}
static INLINE void IADST8X8_1D(
int16x8_t *q8s16,
int16x8_t *q9s16,
int16x8_t *q10s16,
int16x8_t *q11s16,
int16x8_t *q12s16,
int16x8_t *q13s16,
int16x8_t *q14s16,
int16x8_t *q15s16) {
int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16;
int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16, d23s16;
int16x4_t d24s16, d25s16, d26s16, d27s16, d28s16, d29s16, d30s16, d31s16;
int16x8_t q2s16, q4s16, q5s16, q6s16;
int32x4_t q0s32, q1s32, q2s32, q3s32, q4s32, q5s32, q6s32, q7s32, q8s32;
int32x4_t q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32;
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
d20s16 = vget_low_s16(*q10s16);
d21s16 = vget_high_s16(*q10s16);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
d26s16 = vget_low_s16(*q13s16);
d27s16 = vget_high_s16(*q13s16);
d28s16 = vget_low_s16(*q14s16);
d29s16 = vget_high_s16(*q14s16);
d30s16 = vget_low_s16(*q15s16);
d31s16 = vget_high_s16(*q15s16);
d14s16 = vdup_n_s16(cospi_2_64);
d15s16 = vdup_n_s16(cospi_30_64);
q1s32 = vmull_s16(d30s16, d14s16);
q2s32 = vmull_s16(d31s16, d14s16);
q3s32 = vmull_s16(d30s16, d15s16);
q4s32 = vmull_s16(d31s16, d15s16);
d30s16 = vdup_n_s16(cospi_18_64);
d31s16 = vdup_n_s16(cospi_14_64);
q1s32 = vmlal_s16(q1s32, d16s16, d15s16);
q2s32 = vmlal_s16(q2s32, d17s16, d15s16);
q3s32 = vmlsl_s16(q3s32, d16s16, d14s16);
q4s32 = vmlsl_s16(q4s32, d17s16, d14s16);
q5s32 = vmull_s16(d22s16, d30s16);
q6s32 = vmull_s16(d23s16, d30s16);
q7s32 = vmull_s16(d22s16, d31s16);
q8s32 = vmull_s16(d23s16, d31s16);
q5s32 = vmlal_s16(q5s32, d24s16, d31s16);
q6s32 = vmlal_s16(q6s32, d25s16, d31s16);
q7s32 = vmlsl_s16(q7s32, d24s16, d30s16);
q8s32 = vmlsl_s16(q8s32, d25s16, d30s16);
q11s32 = vaddq_s32(q1s32, q5s32);
q12s32 = vaddq_s32(q2s32, q6s32);
q1s32 = vsubq_s32(q1s32, q5s32);
q2s32 = vsubq_s32(q2s32, q6s32);
d22s16 = vqrshrn_n_s32(q11s32, 14);
d23s16 = vqrshrn_n_s32(q12s32, 14);
*q11s16 = vcombine_s16(d22s16, d23s16);
q12s32 = vaddq_s32(q3s32, q7s32);
q15s32 = vaddq_s32(q4s32, q8s32);
q3s32 = vsubq_s32(q3s32, q7s32);
q4s32 = vsubq_s32(q4s32, q8s32);
d2s16 = vqrshrn_n_s32(q1s32, 14);
d3s16 = vqrshrn_n_s32(q2s32, 14);
d24s16 = vqrshrn_n_s32(q12s32, 14);
d25s16 = vqrshrn_n_s32(q15s32, 14);
d6s16 = vqrshrn_n_s32(q3s32, 14);
d7s16 = vqrshrn_n_s32(q4s32, 14);
*q12s16 = vcombine_s16(d24s16, d25s16);
d0s16 = vdup_n_s16(cospi_10_64);
d1s16 = vdup_n_s16(cospi_22_64);
q4s32 = vmull_s16(d26s16, d0s16);
q5s32 = vmull_s16(d27s16, d0s16);
q2s32 = vmull_s16(d26s16, d1s16);
q6s32 = vmull_s16(d27s16, d1s16);
d30s16 = vdup_n_s16(cospi_26_64);
d31s16 = vdup_n_s16(cospi_6_64);
q4s32 = vmlal_s16(q4s32, d20s16, d1s16);
q5s32 = vmlal_s16(q5s32, d21s16, d1s16);
q2s32 = vmlsl_s16(q2s32, d20s16, d0s16);
q6s32 = vmlsl_s16(q6s32, d21s16, d0s16);
q0s32 = vmull_s16(d18s16, d30s16);
q13s32 = vmull_s16(d19s16, d30s16);
q0s32 = vmlal_s16(q0s32, d28s16, d31s16);
q13s32 = vmlal_s16(q13s32, d29s16, d31s16);
q10s32 = vmull_s16(d18s16, d31s16);
q9s32 = vmull_s16(d19s16, d31s16);
q10s32 = vmlsl_s16(q10s32, d28s16, d30s16);
q9s32 = vmlsl_s16(q9s32, d29s16, d30s16);
q14s32 = vaddq_s32(q2s32, q10s32);
q15s32 = vaddq_s32(q6s32, q9s32);
q2s32 = vsubq_s32(q2s32, q10s32);
q6s32 = vsubq_s32(q6s32, q9s32);
d28s16 = vqrshrn_n_s32(q14s32, 14);
d29s16 = vqrshrn_n_s32(q15s32, 14);
d4s16 = vqrshrn_n_s32(q2s32, 14);
d5s16 = vqrshrn_n_s32(q6s32, 14);
*q14s16 = vcombine_s16(d28s16, d29s16);
q9s32 = vaddq_s32(q4s32, q0s32);
q10s32 = vaddq_s32(q5s32, q13s32);
q4s32 = vsubq_s32(q4s32, q0s32);
q5s32 = vsubq_s32(q5s32, q13s32);
d30s16 = vdup_n_s16(cospi_8_64);
d31s16 = vdup_n_s16(cospi_24_64);
d18s16 = vqrshrn_n_s32(q9s32, 14);
d19s16 = vqrshrn_n_s32(q10s32, 14);
d8s16 = vqrshrn_n_s32(q4s32, 14);
d9s16 = vqrshrn_n_s32(q5s32, 14);
*q9s16 = vcombine_s16(d18s16, d19s16);
q5s32 = vmull_s16(d2s16, d30s16);
q6s32 = vmull_s16(d3s16, d30s16);
q7s32 = vmull_s16(d2s16, d31s16);
q0s32 = vmull_s16(d3s16, d31s16);
q5s32 = vmlal_s16(q5s32, d6s16, d31s16);
q6s32 = vmlal_s16(q6s32, d7s16, d31s16);
q7s32 = vmlsl_s16(q7s32, d6s16, d30s16);
q0s32 = vmlsl_s16(q0s32, d7s16, d30s16);
q1s32 = vmull_s16(d4s16, d30s16);
q3s32 = vmull_s16(d5s16, d30s16);
q10s32 = vmull_s16(d4s16, d31s16);
q2s32 = vmull_s16(d5s16, d31s16);
q1s32 = vmlsl_s16(q1s32, d8s16, d31s16);
q3s32 = vmlsl_s16(q3s32, d9s16, d31s16);
q10s32 = vmlal_s16(q10s32, d8s16, d30s16);
q2s32 = vmlal_s16(q2s32, d9s16, d30s16);
*q8s16 = vaddq_s16(*q11s16, *q9s16);
*q11s16 = vsubq_s16(*q11s16, *q9s16);
q4s16 = vaddq_s16(*q12s16, *q14s16);
*q12s16 = vsubq_s16(*q12s16, *q14s16);
q14s32 = vaddq_s32(q5s32, q1s32);
q15s32 = vaddq_s32(q6s32, q3s32);
q5s32 = vsubq_s32(q5s32, q1s32);
q6s32 = vsubq_s32(q6s32, q3s32);
d18s16 = vqrshrn_n_s32(q14s32, 14);
d19s16 = vqrshrn_n_s32(q15s32, 14);
d10s16 = vqrshrn_n_s32(q5s32, 14);
d11s16 = vqrshrn_n_s32(q6s32, 14);
*q9s16 = vcombine_s16(d18s16, d19s16);
q1s32 = vaddq_s32(q7s32, q10s32);
q3s32 = vaddq_s32(q0s32, q2s32);
q7s32 = vsubq_s32(q7s32, q10s32);
q0s32 = vsubq_s32(q0s32, q2s32);
d28s16 = vqrshrn_n_s32(q1s32, 14);
d29s16 = vqrshrn_n_s32(q3s32, 14);
d14s16 = vqrshrn_n_s32(q7s32, 14);
d15s16 = vqrshrn_n_s32(q0s32, 14);
*q14s16 = vcombine_s16(d28s16, d29s16);
d30s16 = vdup_n_s16(cospi_16_64);
d22s16 = vget_low_s16(*q11s16);
d23s16 = vget_high_s16(*q11s16);
q2s32 = vmull_s16(d22s16, d30s16);
q3s32 = vmull_s16(d23s16, d30s16);
q13s32 = vmull_s16(d22s16, d30s16);
q1s32 = vmull_s16(d23s16, d30s16);
d24s16 = vget_low_s16(*q12s16);
d25s16 = vget_high_s16(*q12s16);
q2s32 = vmlal_s16(q2s32, d24s16, d30s16);
q3s32 = vmlal_s16(q3s32, d25s16, d30s16);
q13s32 = vmlsl_s16(q13s32, d24s16, d30s16);
q1s32 = vmlsl_s16(q1s32, d25s16, d30s16);
d4s16 = vqrshrn_n_s32(q2s32, 14);
d5s16 = vqrshrn_n_s32(q3s32, 14);
d24s16 = vqrshrn_n_s32(q13s32, 14);
d25s16 = vqrshrn_n_s32(q1s32, 14);
q2s16 = vcombine_s16(d4s16, d5s16);
*q12s16 = vcombine_s16(d24s16, d25s16);
q13s32 = vmull_s16(d10s16, d30s16);
q1s32 = vmull_s16(d11s16, d30s16);
q11s32 = vmull_s16(d10s16, d30s16);
q0s32 = vmull_s16(d11s16, d30s16);
q13s32 = vmlal_s16(q13s32, d14s16, d30s16);
q1s32 = vmlal_s16(q1s32, d15s16, d30s16);
q11s32 = vmlsl_s16(q11s32, d14s16, d30s16);
q0s32 = vmlsl_s16(q0s32, d15s16, d30s16);
d20s16 = vqrshrn_n_s32(q13s32, 14);
d21s16 = vqrshrn_n_s32(q1s32, 14);
d12s16 = vqrshrn_n_s32(q11s32, 14);
d13s16 = vqrshrn_n_s32(q0s32, 14);
*q10s16 = vcombine_s16(d20s16, d21s16);
q6s16 = vcombine_s16(d12s16, d13s16);
q5s16 = vdupq_n_s16(0);
*q9s16 = vsubq_s16(q5s16, *q9s16);
*q11s16 = vsubq_s16(q5s16, q2s16);
*q13s16 = vsubq_s16(q5s16, q6s16);
*q15s16 = vsubq_s16(q5s16, q4s16);
return;
}
void vp10_iht8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
int dest_stride, int tx_type) {
int i;
uint8_t *d1, *d2;
uint8x8_t d0u8, d1u8, d2u8, d3u8;
uint64x1_t d0u64, d1u64, d2u64, d3u64;
int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16, q13s16, q14s16, q15s16;
uint16x8_t q8u16, q9u16, q10u16, q11u16;
q8s16 = vld1q_s16(input);
q9s16 = vld1q_s16(input + 8);
q10s16 = vld1q_s16(input + 8 * 2);
q11s16 = vld1q_s16(input + 8 * 3);
q12s16 = vld1q_s16(input + 8 * 4);
q13s16 = vld1q_s16(input + 8 * 5);
q14s16 = vld1q_s16(input + 8 * 6);
q15s16 = vld1q_s16(input + 8 * 7);
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
switch (tx_type) {
case 0: // idct_idct is not supported. Fall back to C
vp10_iht8x8_64_add_c(input, dest, dest_stride, tx_type);
return;
break;
case 1: // iadst_idct
// generate IDCT constants
// GENERATE_IDCT_CONSTANTS
// first transform rows
IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
// generate IADST constants
// GENERATE_IADST_CONSTANTS
// then transform columns
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
break;
case 2: // idct_iadst
// generate IADST constants
// GENERATE_IADST_CONSTANTS
// first transform rows
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
// generate IDCT constants
// GENERATE_IDCT_CONSTANTS
// then transform columns
IDCT8x8_1D(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
break;
case 3: // iadst_iadst
// generate IADST constants
// GENERATE_IADST_CONSTANTS
// first transform rows
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
// transpose the matrix
TRANSPOSE8X8(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
// then transform columns
IADST8X8_1D(&q8s16, &q9s16, &q10s16, &q11s16,
&q12s16, &q13s16, &q14s16, &q15s16);
break;
default: // iadst_idct
assert(0);
break;
}
q8s16 = vrshrq_n_s16(q8s16, 5);
q9s16 = vrshrq_n_s16(q9s16, 5);
q10s16 = vrshrq_n_s16(q10s16, 5);
q11s16 = vrshrq_n_s16(q11s16, 5);
q12s16 = vrshrq_n_s16(q12s16, 5);
q13s16 = vrshrq_n_s16(q13s16, 5);
q14s16 = vrshrq_n_s16(q14s16, 5);
q15s16 = vrshrq_n_s16(q15s16, 5);
for (d1 = d2 = dest, i = 0; i < 2; i++) {
if (i != 0) {
q8s16 = q12s16;
q9s16 = q13s16;
q10s16 = q14s16;
q11s16 = q15s16;
}
d0u64 = vld1_u64((uint64_t *)d1);
d1 += dest_stride;
d1u64 = vld1_u64((uint64_t *)d1);
d1 += dest_stride;
d2u64 = vld1_u64((uint64_t *)d1);
d1 += dest_stride;
d3u64 = vld1_u64((uint64_t *)d1);
d1 += dest_stride;
q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16),
vreinterpret_u8_u64(d0u64));
q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16),
vreinterpret_u8_u64(d1u64));
q10u16 = vaddw_u8(vreinterpretq_u16_s16(q10s16),
vreinterpret_u8_u64(d2u64));
q11u16 = vaddw_u8(vreinterpretq_u16_s16(q11s16),
vreinterpret_u8_u64(d3u64));
d0u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16));
d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16));
d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16));
d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16));
vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d0u8));
d2 += dest_stride;
vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d1u8));
d2 += dest_stride;
vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8));
d2 += dest_stride;
vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d3u8));
d2 += dest_stride;
}
return;
}

108
vp10/common/mips/dspr2/vp9_itrans16_dspr2.c Normal file
View File

@ -0,0 +1,108 @@
/*
* Copyright (c) 2013 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 <assert.h>
#include <stdio.h>
#include "./vpx_config.h"
#include "./vp10_rtcd.h"
#include "vp10/common/vp9_common.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_idct.h"
#include "vpx_dsp/mips/inv_txfm_dspr2.h"
#include "vpx_dsp/txfm_common.h"
#include "vpx_ports/mem.h"
#if HAVE_DSPR2
void vp10_iht16x16_256_add_dspr2(const int16_t *input, uint8_t *dest,
int pitch, int tx_type) {
int i, j;
DECLARE_ALIGNED(32, int16_t, out[16 * 16]);
int16_t *outptr = out;
int16_t temp_out[16];
uint32_t pos = 45;
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
switch (tx_type) {
case DCT_DCT: // DCT in both horizontal and vertical
idct16_rows_dspr2(input, outptr, 16);
idct16_cols_add_blk_dspr2(out, dest, pitch);
break;
case ADST_DCT: // ADST in vertical, DCT in horizontal
idct16_rows_dspr2(input, outptr, 16);
outptr = out;
for (i = 0; i < 16; ++i) {
iadst16_dspr2(outptr, temp_out);
for (j = 0; j < 16; ++j)
dest[j * pitch + i] =
clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+ dest[j * pitch + i]);
outptr += 16;
}
break;
case DCT_ADST: // DCT in vertical, ADST in horizontal
{
int16_t temp_in[16 * 16];
for (i = 0; i < 16; ++i) {
/* prefetch row */
prefetch_load((const uint8_t *)(input + 16));
iadst16_dspr2(input, outptr);
input += 16;
outptr += 16;
}
for (i = 0; i < 16; ++i)
for (j = 0; j < 16; ++j)
temp_in[j * 16 + i] = out[i * 16 + j];
idct16_cols_add_blk_dspr2(temp_in, dest, pitch);
}
break;
case ADST_ADST: // ADST in both directions
{
int16_t temp_in[16];
for (i = 0; i < 16; ++i) {
/* prefetch row */
prefetch_load((const uint8_t *)(input + 16));
iadst16_dspr2(input, outptr);
input += 16;
outptr += 16;
}
for (i = 0; i < 16; ++i) {
for (j = 0; j < 16; ++j)
temp_in[j] = out[j * 16 + i];
iadst16_dspr2(temp_in, temp_out);
for (j = 0; j < 16; ++j)
dest[j * pitch + i] =
clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+ dest[j * pitch + i]);
}
}
break;
default:
printf("vp10_short_iht16x16_add_dspr2 : Invalid tx_type\n");
break;
}
}
#endif // #if HAVE_DSPR2

97
vp10/common/mips/dspr2/vp9_itrans4_dspr2.c Normal file
View File

@ -0,0 +1,97 @@
/*
* Copyright (c) 2013 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 <assert.h>
#include <stdio.h>
#include "./vpx_config.h"
#include "./vp10_rtcd.h"
#include "vp10/common/vp9_common.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_idct.h"
#include "vpx_dsp/mips/inv_txfm_dspr2.h"
#include "vpx_dsp/txfm_common.h"
#include "vpx_ports/mem.h"
#if HAVE_DSPR2
void vp10_iht4x4_16_add_dspr2(const int16_t *input, uint8_t *dest,
int dest_stride, int tx_type) {
int i, j;
DECLARE_ALIGNED(32, int16_t, out[4 * 4]);
int16_t *outptr = out;
int16_t temp_in[4 * 4], temp_out[4];
uint32_t pos = 45;
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
switch (tx_type) {
case DCT_DCT: // DCT in both horizontal and vertical
vpx_idct4_rows_dspr2(input, outptr);
vpx_idct4_columns_add_blk_dspr2(&out[0], dest, dest_stride);
break;
case ADST_DCT: // ADST in vertical, DCT in horizontal
vpx_idct4_rows_dspr2(input, outptr);
outptr = out;
for (i = 0; i < 4; ++i) {
iadst4_dspr2(outptr, temp_out);
for (j = 0; j < 4; ++j)
dest[j * dest_stride + i] =
clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+ dest[j * dest_stride + i]);
outptr += 4;
}
break;
case DCT_ADST: // DCT in vertical, ADST in horizontal
for (i = 0; i < 4; ++i) {
iadst4_dspr2(input, outptr);
input += 4;
outptr += 4;
}
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j) {
temp_in[i * 4 + j] = out[j * 4 + i];
}
}
vpx_idct4_columns_add_blk_dspr2(&temp_in[0], dest, dest_stride);
break;
case ADST_ADST: // ADST in both directions
for (i = 0; i < 4; ++i) {
iadst4_dspr2(input, outptr);
input += 4;
outptr += 4;
}
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j)
temp_in[j] = out[j * 4 + i];
iadst4_dspr2(temp_in, temp_out);
for (j = 0; j < 4; ++j)
dest[j * dest_stride + i] =
clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+ dest[j * dest_stride + i]);
}
break;
default:
printf("vp10_short_iht4x4_add_dspr2 : Invalid tx_type\n");
break;
}
}
#endif // #if HAVE_DSPR2

93
vp10/common/mips/dspr2/vp9_itrans8_dspr2.c Normal file
View File

@ -0,0 +1,93 @@
/*
* Copyright (c) 2013 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 <assert.h>
#include <stdio.h>
#include "./vpx_config.h"
#include "./vp10_rtcd.h"
#include "vp10/common/vp9_common.h"
#include "vp10/common/vp9_blockd.h"
#include "vpx_dsp/mips/inv_txfm_dspr2.h"
#include "vpx_dsp/txfm_common.h"
#include "vpx_ports/mem.h"
#if HAVE_DSPR2
void vp10_iht8x8_64_add_dspr2(const int16_t *input, uint8_t *dest,
int dest_stride, int tx_type) {
int i, j;
DECLARE_ALIGNED(32, int16_t, out[8 * 8]);
int16_t *outptr = out;
int16_t temp_in[8 * 8], temp_out[8];
uint32_t pos = 45;
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
switch (tx_type) {
case DCT_DCT: // DCT in both horizontal and vertical
idct8_rows_dspr2(input, outptr, 8);
idct8_columns_add_blk_dspr2(&out[0], dest, dest_stride);
break;
case ADST_DCT: // ADST in vertical, DCT in horizontal
idct8_rows_dspr2(input, outptr, 8);
for (i = 0; i < 8; ++i) {
iadst8_dspr2(&out[i * 8], temp_out);
for (j = 0; j < 8; ++j)
dest[j * dest_stride + i] =
clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+ dest[j * dest_stride + i]);
}
break;
case DCT_ADST: // DCT in vertical, ADST in horizontal
for (i = 0; i < 8; ++i) {
iadst8_dspr2(input, outptr);
input += 8;
outptr += 8;
}
for (i = 0; i < 8; ++i) {
for (j = 0; j < 8; ++j) {
temp_in[i * 8 + j] = out[j * 8 + i];
}
}
idct8_columns_add_blk_dspr2(&temp_in[0], dest, dest_stride);
break;
case ADST_ADST: // ADST in both directions
for (i = 0; i < 8; ++i) {
iadst8_dspr2(input, outptr);
input += 8;
outptr += 8;
}
for (i = 0; i < 8; ++i) {
for (j = 0; j < 8; ++j)
temp_in[j] = out[j * 8 + i];
iadst8_dspr2(temp_in, temp_out);
for (j = 0; j < 8; ++j)
dest[j * dest_stride + i] =
clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+ dest[j * dest_stride + i]);
}
break;
default:
printf("vp10_short_iht8x8_add_dspr2 : Invalid tx_type\n");
break;
}
}
#endif // #if HAVE_DSPR2

81
vp10/common/mips/msa/vp9_idct16x16_msa.c Normal file
View File

@ -0,0 +1,81 @@
/*
* 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 <assert.h>
#include "vp10/common/vp9_enums.h"
#include "vpx_dsp/mips/inv_txfm_msa.h"
void vp10_iht16x16_256_add_msa(const int16_t *input, uint8_t *dst,
int32_t dst_stride, int32_t tx_type) {
int32_t i;
DECLARE_ALIGNED(32, int16_t, out[16 * 16]);
int16_t *out_ptr = &out[0];
switch (tx_type) {
case DCT_DCT:
/* transform rows */
for (i = 0; i < 2; ++i) {
/* process 16 * 8 block */
vpx_idct16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
}
/* transform columns */
for (i = 0; i < 2; ++i) {
/* process 8 * 16 block */
vpx_idct16_1d_columns_addblk_msa((out_ptr + (i << 3)), (dst + (i << 3)),
dst_stride);
}
break;
case ADST_DCT:
/* transform rows */
for (i = 0; i < 2; ++i) {
/* process 16 * 8 block */
vpx_idct16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
}
/* transform columns */
for (i = 0; i < 2; ++i) {
vpx_iadst16_1d_columns_addblk_msa((out_ptr + (i << 3)),
(dst + (i << 3)), dst_stride);
}
break;
case DCT_ADST:
/* transform rows */
for (i = 0; i < 2; ++i) {
/* process 16 * 8 block */
vpx_iadst16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
}
/* transform columns */
for (i = 0; i < 2; ++i) {
/* process 8 * 16 block */
vpx_idct16_1d_columns_addblk_msa((out_ptr + (i << 3)), (dst + (i << 3)),
dst_stride);
}
break;
case ADST_ADST:
/* transform rows */
for (i = 0; i < 2; ++i) {
/* process 16 * 8 block */
vpx_iadst16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7)));
}
/* transform columns */
for (i = 0; i < 2; ++i) {
vpx_iadst16_1d_columns_addblk_msa((out_ptr + (i << 3)),
(dst + (i << 3)), dst_stride);
}
break;
default:
assert(0);
break;
}
}

62
vp10/common/mips/msa/vp9_idct4x4_msa.c Normal file
View File

@ -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.
*/
#include <assert.h>
#include "vp10/common/vp9_enums.h"
#include "vpx_dsp/mips/inv_txfm_msa.h"
void vp10_iht4x4_16_add_msa(const int16_t *input, uint8_t *dst,
int32_t dst_stride, int32_t tx_type) {
v8i16 in0, in1, in2, in3;
/* load vector elements of 4x4 block */
LD4x4_SH(input, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
switch (tx_type) {
case DCT_DCT:
/* DCT in horizontal */
VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
/* DCT in vertical */
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_DCT:
/* DCT in horizontal */
VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
/* ADST in vertical */
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case DCT_ADST:
/* ADST in horizontal */
VP9_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
/* DCT in vertical */
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_ADST:
/* ADST in horizontal */
VP9_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
/* ADST in vertical */
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
default:
assert(0);
break;
}
/* final rounding (add 2^3, divide by 2^4) and shift */
SRARI_H4_SH(in0, in1, in2, in3, 4);
/* add block and store 4x4 */
ADDBLK_ST4x4_UB(in0, in1, in2, in3, dst, dst_stride);
}

80
vp10/common/mips/msa/vp9_idct8x8_msa.c Normal file
View File

@ -0,0 +1,80 @@
/*
* 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 <assert.h>
#include "vp10/common/vp9_enums.h"
#include "vpx_dsp/mips/inv_txfm_msa.h"
void vp10_iht8x8_64_add_msa(const int16_t *input, uint8_t *dst,
int32_t dst_stride, int32_t tx_type) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
/* load vector elements of 8x8 block */
LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
switch (tx_type) {
case DCT_DCT:
/* DCT in horizontal */
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* DCT in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_DCT:
/* DCT in horizontal */
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* ADST in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case DCT_ADST:
/* ADST in horizontal */
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* DCT in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_ADST:
/* ADST in horizontal */
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* ADST in vertical */
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
default:
assert(0);
break;
}
/* final rounding (add 2^4, divide by 2^5) and shift */
SRARI_H4_SH(in0, in1, in2, in3, 5);
SRARI_H4_SH(in4, in5, in6, in7, 5);
/* add block and store 8x8 */
VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3);
dst += (4 * dst_stride);
VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7);
}

137
vp10/common/mips/msa/vp9_mfqe_msa.c Normal file
View File

@ -0,0 +1,137 @@
/*
* 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 "vp10/common/vp9_onyxc_int.h"
#include "vpx_dsp/mips/macros_msa.h"
static void filter_by_weight8x8_msa(const uint8_t *src_ptr, int32_t src_stride,
uint8_t *dst_ptr, int32_t dst_stride,
int32_t src_weight) {
int32_t dst_weight = (1 << MFQE_PRECISION) - src_weight;
int32_t row;
uint64_t src0_d, src1_d, dst0_d, dst1_d;
v16i8 src0 = { 0 };
v16i8 src1 = { 0 };
v16i8 dst0 = { 0 };
v16i8 dst1 = { 0 };
v8i16 src_wt, dst_wt, res_h_r, res_h_l, src_r, src_l, dst_r, dst_l;
src_wt = __msa_fill_h(src_weight);
dst_wt = __msa_fill_h(dst_weight);
for (row = 2; row--;) {
LD2(src_ptr, src_stride, src0_d, src1_d);
src_ptr += (2 * src_stride);
LD2(dst_ptr, dst_stride, dst0_d, dst1_d);
INSERT_D2_SB(src0_d, src1_d, src0);
INSERT_D2_SB(dst0_d, dst1_d, dst0);
LD2(src_ptr, src_stride, src0_d, src1_d);
src_ptr += (2 * src_stride);
LD2((dst_ptr + 2 * dst_stride), dst_stride, dst0_d, dst1_d);
INSERT_D2_SB(src0_d, src1_d, src1);
INSERT_D2_SB(dst0_d, dst1_d, dst1);
UNPCK_UB_SH(src0, src_r, src_l);
UNPCK_UB_SH(dst0, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
dst0 = (v16i8)__msa_pckev_b((v16i8)res_h_l, (v16i8)res_h_r);
ST8x2_UB(dst0, dst_ptr, dst_stride);
dst_ptr += (2 * dst_stride);
UNPCK_UB_SH(src1, src_r, src_l);
UNPCK_UB_SH(dst1, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
dst1 = (v16i8)__msa_pckev_b((v16i8)res_h_l, (v16i8)res_h_r);
ST8x2_UB(dst1, dst_ptr, dst_stride);
dst_ptr += (2 * dst_stride);
}
}
static void filter_by_weight16x16_msa(const uint8_t *src_ptr,
int32_t src_stride,
uint8_t *dst_ptr,
int32_t dst_stride,
int32_t src_weight) {
int32_t dst_weight = (1 << MFQE_PRECISION) - src_weight;
int32_t row;
v16i8 src0, src1, src2, src3, dst0, dst1, dst2, dst3;
v8i16 src_wt, dst_wt, res_h_r, res_h_l, src_r, src_l, dst_r, dst_l;
src_wt = __msa_fill_h(src_weight);
dst_wt = __msa_fill_h(dst_weight);
for (row = 4; row--;) {
LD_SB4(src_ptr, src_stride, src0, src1, src2, src3);
src_ptr += (4 * src_stride);
LD_SB4(dst_ptr, dst_stride, dst0, dst1, dst2, dst3);
UNPCK_UB_SH(src0, src_r, src_l);
UNPCK_UB_SH(dst0, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr);
dst_ptr += dst_stride;
UNPCK_UB_SH(src1, src_r, src_l);
UNPCK_UB_SH(dst1, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr);
dst_ptr += dst_stride;
UNPCK_UB_SH(src2, src_r, src_l);
UNPCK_UB_SH(dst2, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr);
dst_ptr += dst_stride;
UNPCK_UB_SH(src3, src_r, src_l);
UNPCK_UB_SH(dst3, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr);
dst_ptr += dst_stride;
}
}
void vp10_filter_by_weight8x8_msa(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
int src_weight) {
filter_by_weight8x8_msa(src, src_stride, dst, dst_stride, src_weight);
}
void vp10_filter_by_weight16x16_msa(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
int src_weight) {
filter_by_weight16x16_msa(src, src_stride, dst, dst_stride, src_weight);
}

19
vp10/common/vp10_rtcd.c Normal file
View File

@ -0,0 +1,19 @@
/*
* Copyright (c) 2011 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 "./vpx_config.h"
#define RTCD_C
#include "./vp10_rtcd.h"
#include "vpx_ports/vpx_once.h"
void vp10_rtcd() {
// TODO(JBB): Remove this once, by insuring that both the encoder and
// decoder setup functions are protected by once();
once(setup_rtcd_internal);
}

362
vp10/common/vp10_rtcd_defs.pl Normal file
View File

@ -0,0 +1,362 @@
sub vp10_common_forward_decls() {
print <<EOF
/*
* VP9
*/
#include "vpx/vpx_integer.h"
#include "vp10/common/vp9_common.h"
#include "vp10/common/vp9_enums.h"
struct macroblockd;
/* Encoder forward decls */
struct macroblock;
struct vp9_variance_vtable;
struct search_site_config;
struct mv;
union int_mv;
struct yv12_buffer_config;
EOF
}
forward_decls qw/vp10_common_forward_decls/;
# x86inc.asm had specific constraints. break it out so it's easy to disable.
# zero all the variables to avoid tricky else conditions.
$mmx_x86inc = $sse_x86inc = $sse2_x86inc = $ssse3_x86inc = $avx_x86inc =
$avx2_x86inc = '';
$mmx_x86_64_x86inc = $sse_x86_64_x86inc = $sse2_x86_64_x86inc =
$ssse3_x86_64_x86inc = $avx_x86_64_x86inc = $avx2_x86_64_x86inc = '';
if (vpx_config("CONFIG_USE_X86INC") eq "yes") {
$mmx_x86inc = 'mmx';
$sse_x86inc = 'sse';
$sse2_x86inc = 'sse2';
$ssse3_x86inc = 'ssse3';
$avx_x86inc = 'avx';
$avx2_x86inc = 'avx2';
if ($opts{arch} eq "x86_64") {
$mmx_x86_64_x86inc = 'mmx';
$sse_x86_64_x86inc = 'sse';
$sse2_x86_64_x86inc = 'sse2';
$ssse3_x86_64_x86inc = 'ssse3';
$avx_x86_64_x86inc = 'avx';
$avx2_x86_64_x86inc = 'avx2';
}
}
# functions that are 64 bit only.
$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
if ($opts{arch} eq "x86_64") {
$mmx_x86_64 = 'mmx';
$sse2_x86_64 = 'sse2';
$ssse3_x86_64 = 'ssse3';
$avx_x86_64 = 'avx';
$avx2_x86_64 = 'avx2';
}
#
# post proc
#
if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {
add_proto qw/void vp10_mbpost_proc_down/, "uint8_t *dst, int pitch, int rows, int cols, int flimit";
specialize qw/vp10_mbpost_proc_down sse2/;
$vp10_mbpost_proc_down_sse2=vp10_mbpost_proc_down_xmm;
add_proto qw/void vp10_mbpost_proc_across_ip/, "uint8_t *src, int pitch, int rows, int cols, int flimit";
specialize qw/vp10_mbpost_proc_across_ip sse2/;
$vp10_mbpost_proc_across_ip_sse2=vp10_mbpost_proc_across_ip_xmm;
add_proto qw/void vp10_post_proc_down_and_across/, "const uint8_t *src_ptr, uint8_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
specialize qw/vp10_post_proc_down_and_across sse2/;
$vp10_post_proc_down_and_across_sse2=vp10_post_proc_down_and_across_xmm;
add_proto qw/void vp10_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
specialize qw/vp10_plane_add_noise sse2/;
$vp10_plane_add_noise_sse2=vp10_plane_add_noise_wmt;
add_proto qw/void vp10_filter_by_weight16x16/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
specialize qw/vp10_filter_by_weight16x16 sse2 msa/;
add_proto qw/void vp10_filter_by_weight8x8/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
specialize qw/vp10_filter_by_weight8x8 sse2 msa/;
}
#
# dct
#
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
# Note as optimized versions of these functions are added we need to add a check to ensure
# that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x4_16_add/;
add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x8_64_add/;
add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
specialize qw/vp10_iht16x16_256_add/;
} else {
# Force C versions if CONFIG_EMULATE_HARDWARE is 1
if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x4_16_add/;
add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x8_64_add/;
add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
specialize qw/vp10_iht16x16_256_add/;
} else {
add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x4_16_add sse2 neon dspr2 msa/;
add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x8_64_add sse2 neon dspr2 msa/;
add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
specialize qw/vp10_iht16x16_256_add sse2 dspr2 msa/;
}
}
# High bitdepth functions
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
#
# Sub Pixel Filters
#
add_proto qw/void vp10_highbd_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/vp10_highbd_convolve_copy/;
add_proto qw/void vp10_highbd_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/vp10_highbd_convolve_avg/;
add_proto qw/void vp10_highbd_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/vp10_highbd_convolve8/, "$sse2_x86_64";
add_proto qw/void vp10_highbd_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/vp10_highbd_convolve8_horiz/, "$sse2_x86_64";
add_proto qw/void vp10_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/vp10_highbd_convolve8_vert/, "$sse2_x86_64";
add_proto qw/void vp10_highbd_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/vp10_highbd_convolve8_avg/, "$sse2_x86_64";
add_proto qw/void vp10_highbd_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/vp10_highbd_convolve8_avg_horiz/, "$sse2_x86_64";
add_proto qw/void vp10_highbd_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
specialize qw/vp10_highbd_convolve8_avg_vert/, "$sse2_x86_64";
#
# post proc
#
if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {
add_proto qw/void vp10_highbd_mbpost_proc_down/, "uint16_t *dst, int pitch, int rows, int cols, int flimit";
specialize qw/vp10_highbd_mbpost_proc_down/;
add_proto qw/void vp10_highbd_mbpost_proc_across_ip/, "uint16_t *src, int pitch, int rows, int cols, int flimit";
specialize qw/vp10_highbd_mbpost_proc_across_ip/;
add_proto qw/void vp10_highbd_post_proc_down_and_across/, "const uint16_t *src_ptr, uint16_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
specialize qw/vp10_highbd_post_proc_down_and_across/;
add_proto qw/void vp10_highbd_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
specialize qw/vp10_highbd_plane_add_noise/;
}
#
# dct
#
# Note as optimized versions of these functions are added we need to add a check to ensure
# that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
add_proto qw/void vp10_highbd_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht4x4_16_add/;
add_proto qw/void vp10_highbd_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht8x8_64_add/;
add_proto qw/void vp10_highbd_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type, int bd";
specialize qw/vp10_highbd_iht16x16_256_add/;
}
#
# Encoder functions below this point.
#
if (vpx_config("CONFIG_VP10_ENCODER") eq "yes") {
add_proto qw/unsigned int vp10_avg_8x8/, "const uint8_t *, int p";
specialize qw/vp10_avg_8x8 sse2 neon msa/;
add_proto qw/unsigned int vp10_avg_4x4/, "const uint8_t *, int p";
specialize qw/vp10_avg_4x4 sse2 msa/;
add_proto qw/void vp10_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
specialize qw/vp10_minmax_8x8 sse2/;
add_proto qw/void vp10_hadamard_8x8/, "int16_t const *src_diff, int src_stride, int16_t *coeff";
specialize qw/vp10_hadamard_8x8 sse2/, "$ssse3_x86_64_x86inc";
add_proto qw/void vp10_hadamard_16x16/, "int16_t const *src_diff, int src_stride, int16_t *coeff";
specialize qw/vp10_hadamard_16x16 sse2/;
add_proto qw/int16_t vp10_satd/, "const int16_t *coeff, int length";
specialize qw/vp10_satd sse2/;
add_proto qw/void vp10_int_pro_row/, "int16_t *hbuf, uint8_t const *ref, const int ref_stride, const int height";
specialize qw/vp10_int_pro_row sse2 neon/;
add_proto qw/int16_t vp10_int_pro_col/, "uint8_t const *ref, const int width";
specialize qw/vp10_int_pro_col sse2 neon/;
add_proto qw/int vp10_vector_var/, "int16_t const *ref, int16_t const *src, const int bwl";
specialize qw/vp10_vector_var neon sse2/;
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/unsigned int vp10_highbd_avg_8x8/, "const uint8_t *, int p";
specialize qw/vp10_highbd_avg_8x8/;
add_proto qw/unsigned int vp10_highbd_avg_4x4/, "const uint8_t *, int p";
specialize qw/vp10_highbd_avg_4x4/;
add_proto qw/void vp10_highbd_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
specialize qw/vp10_highbd_minmax_8x8/;
}
# ENCODEMB INVOKE
#
# Denoiser
#
if (vpx_config("CONFIG_VP9_TEMPORAL_DENOISING") eq "yes") {
add_proto qw/int vp10_denoiser_filter/, "const uint8_t *sig, int sig_stride, const uint8_t *mc_avg, int mc_avg_stride, uint8_t *avg, int avg_stride, int increase_denoising, BLOCK_SIZE bs, int motion_magnitude";
specialize qw/vp10_denoiser_filter sse2/;
}
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
# the transform coefficients are held in 32-bit
# values, so the assembler code for vp10_block_error can no longer be used.
add_proto qw/int64_t vp10_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
specialize qw/vp10_block_error/;
add_proto qw/void vp10_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp10_quantize_fp/;
add_proto qw/void vp10_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp10_quantize_fp_32x32/;
add_proto qw/void vp10_fdct8x8_quant/, "const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp10_fdct8x8_quant/;
} else {
add_proto qw/int64_t vp10_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
specialize qw/vp10_block_error avx2 msa/, "$sse2_x86inc";
add_proto qw/int64_t vp10_block_error_fp/, "const int16_t *coeff, const int16_t *dqcoeff, int block_size";
specialize qw/vp10_block_error_fp neon/, "$sse2_x86inc";
add_proto qw/void vp10_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp10_quantize_fp neon sse2/, "$ssse3_x86_64_x86inc";
add_proto qw/void vp10_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp10_quantize_fp_32x32/, "$ssse3_x86_64_x86inc";
add_proto qw/void vp10_fdct8x8_quant/, "const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp10_fdct8x8_quant sse2 ssse3 neon/;
}
#
# Structured Similarity (SSIM)
#
if (vpx_config("CONFIG_INTERNAL_STATS") eq "yes") {
add_proto qw/void vp10_ssim_parms_8x8/, "uint8_t *s, int sp, uint8_t *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr";
specialize qw/vp10_ssim_parms_8x8/, "$sse2_x86_64";
add_proto qw/void vp10_ssim_parms_16x16/, "uint8_t *s, int sp, uint8_t *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr";
specialize qw/vp10_ssim_parms_16x16/, "$sse2_x86_64";
}
# fdct functions
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht4x4 sse2/;
add_proto qw/void vp10_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht8x8 sse2/;
add_proto qw/void vp10_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht16x16 sse2/;
add_proto qw/void vp10_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp10_fwht4x4/, "$mmx_x86inc";
} else {
add_proto qw/void vp10_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht4x4 sse2 msa/;
add_proto qw/void vp10_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht8x8 sse2 msa/;
add_proto qw/void vp10_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht16x16 sse2 msa/;
add_proto qw/void vp10_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp10_fwht4x4 msa/, "$mmx_x86inc";
}
#
# Motion search
#
add_proto qw/int vp10_full_search_sad/, "const struct macroblock *x, const struct mv *ref_mv, int sad_per_bit, int distance, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv, struct mv *best_mv";
specialize qw/vp10_full_search_sad sse3 sse4_1/;
$vp10_full_search_sad_sse3=vp10_full_search_sadx3;
$vp10_full_search_sad_sse4_1=vp10_full_search_sadx8;
add_proto qw/int vp10_diamond_search_sad/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv";
specialize qw/vp10_diamond_search_sad/;
add_proto qw/int vp10_full_range_search/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv";
specialize qw/vp10_full_range_search/;
add_proto qw/void vp10_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
specialize qw/vp10_temporal_filter_apply sse2 msa/;
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
# ENCODEMB INVOKE
add_proto qw/int64_t vp10_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
specialize qw/vp10_highbd_block_error sse2/;
add_proto qw/void vp10_highbd_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp10_highbd_quantize_fp/;
add_proto qw/void vp10_highbd_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vp10_highbd_quantize_fp_32x32/;
#
# Structured Similarity (SSIM)
#
if (vpx_config("CONFIG_INTERNAL_STATS") eq "yes") {
add_proto qw/void vp10_highbd_ssim_parms_8x8/, "uint16_t *s, int sp, uint16_t *r, int rp, uint32_t *sum_s, uint32_t *sum_r, uint32_t *sum_sq_s, uint32_t *sum_sq_r, uint32_t *sum_sxr";
specialize qw/vp10_highbd_ssim_parms_8x8/;
}
# fdct functions
add_proto qw/void vp10_highbd_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht4x4/;
add_proto qw/void vp10_highbd_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht8x8/;
add_proto qw/void vp10_highbd_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht16x16/;
add_proto qw/void vp10_highbd_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp10_highbd_fwht4x4/;
add_proto qw/void vp10_highbd_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
specialize qw/vp10_highbd_temporal_filter_apply/;
}
# End vp10_high encoder functions
}
# end encoder functions
1;

165
vp10/common/vp9_alloccommon.c Normal file
View File

@ -0,0 +1,165 @@
/*
* Copyright (c) 2010 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 "./vpx_config.h"
#include "vpx_mem/vpx_mem.h"
#include "vp10/common/vp9_alloccommon.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_entropymode.h"
#include "vp10/common/vp9_entropymv.h"
#include "vp10/common/vp9_onyxc_int.h"
#include "vp10/common/vp9_systemdependent.h"
void vp10_set_mb_mi(VP9_COMMON *cm, int width, int height) {
const int aligned_width = ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2);
const int aligned_height = ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2);
cm->mi_cols = aligned_width >> MI_SIZE_LOG2;
cm->mi_rows = aligned_height >> MI_SIZE_LOG2;
cm->mi_stride = calc_mi_size(cm->mi_cols);
cm->mb_cols = (cm->mi_cols + 1) >> 1;
cm->mb_rows = (cm->mi_rows + 1) >> 1;
cm->MBs = cm->mb_rows * cm->mb_cols;
}
static int alloc_seg_map(VP9_COMMON *cm, int seg_map_size) {
int i;
for (i = 0; i < NUM_PING_PONG_BUFFERS; ++i) {
cm->seg_map_array[i] = (uint8_t *)vpx_calloc(seg_map_size, 1);
if (cm->seg_map_array[i] == NULL)
return 1;
}
cm->seg_map_alloc_size = seg_map_size;
// Init the index.
cm->seg_map_idx = 0;
cm->prev_seg_map_idx = 1;
cm->current_frame_seg_map = cm->seg_map_array[cm->seg_map_idx];
if (!cm->frame_parallel_decode)
cm->last_frame_seg_map = cm->seg_map_array[cm->prev_seg_map_idx];
return 0;
}
static void free_seg_map(VP9_COMMON *cm) {
int i;
for (i = 0; i < NUM_PING_PONG_BUFFERS; ++i) {
vpx_free(cm->seg_map_array[i]);
cm->seg_map_array[i] = NULL;
}
cm->current_frame_seg_map = NULL;
if (!cm->frame_parallel_decode) {
cm->last_frame_seg_map = NULL;
}
}
void vp10_free_ref_frame_buffers(BufferPool *pool) {
int i;
for (i = 0; i < FRAME_BUFFERS; ++i) {
if (pool->frame_bufs[i].ref_count > 0 &&
pool->frame_bufs[i].raw_frame_buffer.data != NULL) {
pool->release_fb_cb(pool->cb_priv, &pool->frame_bufs[i].raw_frame_buffer);
pool->frame_bufs[i].ref_count = 0;
}
vpx_free(pool->frame_bufs[i].mvs);
pool->frame_bufs[i].mvs = NULL;
vp9_free_frame_buffer(&pool->frame_bufs[i].buf);
}
}
void vp10_free_postproc_buffers(VP9_COMMON *cm) {
#if CONFIG_VP9_POSTPROC
vp9_free_frame_buffer(&cm->post_proc_buffer);
vp9_free_frame_buffer(&cm->post_proc_buffer_int);
#else
(void)cm;
#endif
}
void vp10_free_context_buffers(VP9_COMMON *cm) {
cm->free_mi(cm);
free_seg_map(cm);
vpx_free(cm->above_context);
cm->above_context = NULL;
vpx_free(cm->above_seg_context);
cm->above_seg_context = NULL;
}
int vp10_alloc_context_buffers(VP9_COMMON *cm, int width, int height) {
int new_mi_size;
vp10_set_mb_mi(cm, width, height);
new_mi_size = cm->mi_stride * calc_mi_size(cm->mi_rows);
if (cm->mi_alloc_size < new_mi_size) {
cm->free_mi(cm);
if (cm->alloc_mi(cm, new_mi_size))
goto fail;
}
if (cm->seg_map_alloc_size < cm->mi_rows * cm->mi_cols) {
// Create the segmentation map structure and set to 0.
free_seg_map(cm);
if (alloc_seg_map(cm, cm->mi_rows * cm->mi_cols))
goto fail;
}
if (cm->above_context_alloc_cols < cm->mi_cols) {
vpx_free(cm->above_context);
cm->above_context = (ENTROPY_CONTEXT *)vpx_calloc(
2 * mi_cols_aligned_to_sb(cm->mi_cols) * MAX_MB_PLANE,
sizeof(*cm->above_context));
if (!cm->above_context) goto fail;
vpx_free(cm->above_seg_context);
cm->above_seg_context = (PARTITION_CONTEXT *)vpx_calloc(
mi_cols_aligned_to_sb(cm->mi_cols), sizeof(*cm->above_seg_context));
if (!cm->above_seg_context) goto fail;
cm->above_context_alloc_cols = cm->mi_cols;
}
return 0;
fail:
vp10_free_context_buffers(cm);
return 1;
}
void vp10_remove_common(VP9_COMMON *cm) {
vp10_free_context_buffers(cm);
vpx_free(cm->fc);
cm->fc = NULL;
vpx_free(cm->frame_contexts);
cm->frame_contexts = NULL;
}
void vp10_init_context_buffers(VP9_COMMON *cm) {
cm->setup_mi(cm);
if (cm->last_frame_seg_map && !cm->frame_parallel_decode)
memset(cm->last_frame_seg_map, 0, cm->mi_rows * cm->mi_cols);
}
void vp10_swap_current_and_last_seg_map(VP9_COMMON *cm) {
// Swap indices.
const int tmp = cm->seg_map_idx;
cm->seg_map_idx = cm->prev_seg_map_idx;
cm->prev_seg_map_idx = tmp;
cm->current_frame_seg_map = cm->seg_map_array[cm->seg_map_idx];
cm->last_frame_seg_map = cm->seg_map_array[cm->prev_seg_map_idx];
}

44
vp10/common/vp9_alloccommon.h Normal file
View File

@ -0,0 +1,44 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_ALLOCCOMMON_H_
#define VP9_COMMON_VP9_ALLOCCOMMON_H_
#define INVALID_IDX -1 // Invalid buffer index.
#ifdef __cplusplus
extern "C" {
#endif
struct VP9Common;
struct BufferPool;
void vp10_remove_common(struct VP9Common *cm);
int vp10_alloc_context_buffers(struct VP9Common *cm, int width, int height);
void vp10_init_context_buffers(struct VP9Common *cm);
void vp10_free_context_buffers(struct VP9Common *cm);
void vp10_free_ref_frame_buffers(struct BufferPool *pool);
void vp10_free_postproc_buffers(struct VP9Common *cm);
int vp10_alloc_state_buffers(struct VP9Common *cm, int width, int height);
void vp10_free_state_buffers(struct VP9Common *cm);
void vp10_set_mb_mi(struct VP9Common *cm, int width, int height);
void vp10_swap_current_and_last_seg_map(struct VP9Common *cm);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_ALLOCCOMMON_H_

136
vp10/common/vp9_blockd.c Normal file
View File

@ -0,0 +1,136 @@
/*
* Copyright (c) 2014 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/common/vp9_blockd.h"
PREDICTION_MODE vp10_left_block_mode(const MODE_INFO *cur_mi,
const MODE_INFO *left_mi, int b) {
if (b == 0 || b == 2) {
if (!left_mi || is_inter_block(&left_mi->mbmi))
return DC_PRED;
return get_y_mode(left_mi, b + 1);
} else {
assert(b == 1 || b == 3);
return cur_mi->bmi[b - 1].as_mode;
}
}
PREDICTION_MODE vp10_above_block_mode(const MODE_INFO *cur_mi,
const MODE_INFO *above_mi, int b) {
if (b == 0 || b == 1) {
if (!above_mi || is_inter_block(&above_mi->mbmi))
return DC_PRED;
return get_y_mode(above_mi, b + 2);
} else {
assert(b == 2 || b == 3);
return cur_mi->bmi[b - 2].as_mode;
}
}
void vp10_foreach_transformed_block_in_plane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
foreach_transformed_block_visitor visit, void *arg) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
const MB_MODE_INFO* mbmi = &xd->mi[0]->mbmi;
// block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
// 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
// transform size varies per plane, look it up in a common way.
const TX_SIZE tx_size = plane ? get_uv_tx_size(mbmi, pd)
: mbmi->tx_size;
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
const int step = 1 << (tx_size << 1);
int i = 0, r, c;
// If mb_to_right_edge is < 0 we are in a situation in which
// the current block size extends into the UMV and we won't
// visit the sub blocks that are wholly within the UMV.
const int max_blocks_wide = num_4x4_w + (xd->mb_to_right_edge >= 0 ? 0 :
xd->mb_to_right_edge >> (5 + pd->subsampling_x));
const int max_blocks_high = num_4x4_h + (xd->mb_to_bottom_edge >= 0 ? 0 :
xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
const int extra_step = ((num_4x4_w - max_blocks_wide) >> tx_size) * step;
// Keep track of the row and column of the blocks we use so that we know
// if we are in the unrestricted motion border.
for (r = 0; r < max_blocks_high; r += (1 << tx_size)) {
// Skip visiting the sub blocks that are wholly within the UMV.
for (c = 0; c < max_blocks_wide; c += (1 << tx_size)) {
visit(plane, i, plane_bsize, tx_size, arg);
i += step;
}
i += extra_step;
}
}
void vp10_foreach_transformed_block(const MACROBLOCKD* const xd,
BLOCK_SIZE bsize,
foreach_transformed_block_visitor visit,
void *arg) {
int plane;
for (plane = 0; plane < MAX_MB_PLANE; ++plane)
vp10_foreach_transformed_block_in_plane(xd, bsize, plane, visit, arg);
}
void vp10_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size, int has_eob,
int aoff, int loff) {
ENTROPY_CONTEXT *const a = pd->above_context + aoff;
ENTROPY_CONTEXT *const l = pd->left_context + loff;
const int tx_size_in_blocks = 1 << tx_size;
// above
if (has_eob && xd->mb_to_right_edge < 0) {
int i;
const int blocks_wide = num_4x4_blocks_wide_lookup[plane_bsize] +
(xd->mb_to_right_edge >> (5 + pd->subsampling_x));
int above_contexts = tx_size_in_blocks;
if (above_contexts + aoff > blocks_wide)
above_contexts = blocks_wide - aoff;
for (i = 0; i < above_contexts; ++i)
a[i] = has_eob;
for (i = above_contexts; i < tx_size_in_blocks; ++i)
a[i] = 0;
} else {
memset(a, has_eob, sizeof(ENTROPY_CONTEXT) * tx_size_in_blocks);
}
// left
if (has_eob && xd->mb_to_bottom_edge < 0) {
int i;
const int blocks_high = num_4x4_blocks_high_lookup[plane_bsize] +
(xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
int left_contexts = tx_size_in_blocks;
if (left_contexts + loff > blocks_high)
left_contexts = blocks_high - loff;
for (i = 0; i < left_contexts; ++i)
l[i] = has_eob;
for (i = left_contexts; i < tx_size_in_blocks; ++i)
l[i] = 0;
} else {
memset(l, has_eob, sizeof(ENTROPY_CONTEXT) * tx_size_in_blocks);
}
}
void vp10_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y) {
int i;
for (i = 0; i < MAX_MB_PLANE; i++) {
xd->plane[i].plane_type = i ? PLANE_TYPE_UV : PLANE_TYPE_Y;
xd->plane[i].subsampling_x = i ? ss_x : 0;
xd->plane[i].subsampling_y = i ? ss_y : 0;
}
}

318
vp10/common/vp9_blockd.h Normal file
View File

@ -0,0 +1,318 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_BLOCKD_H_
#define VP9_COMMON_VP9_BLOCKD_H_
#include "./vpx_config.h"
#include "vpx_ports/mem.h"
#include "vpx_scale/yv12config.h"
#include "vp10/common/vp9_common_data.h"
#include "vp10/common/vp9_entropy.h"
#include "vp10/common/vp9_entropymode.h"
#include "vp10/common/vp9_mv.h"
#include "vp10/common/vp9_scale.h"
#include "vp10/common/vp9_seg_common.h"
#include "vp10/common/vp9_tile_common.h"
#ifdef __cplusplus
extern "C" {
#endif
#define MAX_MB_PLANE 3
typedef enum {
KEY_FRAME = 0,
INTER_FRAME = 1,
FRAME_TYPES,
} FRAME_TYPE;
static INLINE int is_inter_mode(PREDICTION_MODE mode) {
return mode >= NEARESTMV && mode <= NEWMV;
}
/* For keyframes, intra block modes are predicted by the (already decoded)
modes for the Y blocks to the left and above us; for interframes, there
is a single probability table. */
typedef struct {
PREDICTION_MODE as_mode;
int_mv as_mv[2]; // first, second inter predictor motion vectors
} b_mode_info;
// Note that the rate-distortion optimization loop, bit-stream writer, and
// decoder implementation modules critically rely on the defined entry values
// specified herein. They should be refactored concurrently.
#define NONE -1
#define INTRA_FRAME 0
#define LAST_FRAME 1
#define GOLDEN_FRAME 2
#define ALTREF_FRAME 3
#define MAX_REF_FRAMES 4
typedef int8_t MV_REFERENCE_FRAME;
// This structure now relates to 8x8 block regions.
typedef struct {
// Common for both INTER and INTRA blocks
BLOCK_SIZE sb_type;
PREDICTION_MODE mode;
TX_SIZE tx_size;
int8_t skip;
int8_t segment_id;
int8_t seg_id_predicted; // valid only when temporal_update is enabled
// Only for INTRA blocks
PREDICTION_MODE uv_mode;
// Only for INTER blocks
INTERP_FILTER interp_filter;
MV_REFERENCE_FRAME ref_frame[2];
// TODO(slavarnway): Delete and use bmi[3].as_mv[] instead.
int_mv mv[2];
} MB_MODE_INFO;
typedef struct MODE_INFO {
MB_MODE_INFO mbmi;
b_mode_info bmi[4];
} MODE_INFO;
static INLINE PREDICTION_MODE get_y_mode(const MODE_INFO *mi, int block) {
return mi->mbmi.sb_type < BLOCK_8X8 ? mi->bmi[block].as_mode
: mi->mbmi.mode;
}
static INLINE int is_inter_block(const MB_MODE_INFO *mbmi) {
return mbmi->ref_frame[0] > INTRA_FRAME;
}
static INLINE int has_second_ref(const MB_MODE_INFO *mbmi) {
return mbmi->ref_frame[1] > INTRA_FRAME;
}
PREDICTION_MODE vp10_left_block_mode(const MODE_INFO *cur_mi,
const MODE_INFO *left_mi, int b);
PREDICTION_MODE vp10_above_block_mode(const MODE_INFO *cur_mi,
const MODE_INFO *above_mi, int b);
enum mv_precision {
MV_PRECISION_Q3,
MV_PRECISION_Q4
};
struct buf_2d {
uint8_t *buf;
int stride;
};
struct macroblockd_plane {
tran_low_t *dqcoeff;
PLANE_TYPE plane_type;
int subsampling_x;
int subsampling_y;
struct buf_2d dst;
struct buf_2d pre[2];
ENTROPY_CONTEXT *above_context;
ENTROPY_CONTEXT *left_context;
int16_t seg_dequant[MAX_SEGMENTS][2];
// number of 4x4s in current block
uint16_t n4_w, n4_h;
// log2 of n4_w, n4_h
uint8_t n4_wl, n4_hl;
// encoder
const int16_t *dequant;
};
#define BLOCK_OFFSET(x, i) ((x) + (i) * 16)
typedef struct RefBuffer {
// TODO(dkovalev): idx is not really required and should be removed, now it
// is used in vp10_onyxd_if.c
int idx;
YV12_BUFFER_CONFIG *buf;
struct scale_factors sf;
} RefBuffer;
typedef struct macroblockd {
struct macroblockd_plane plane[MAX_MB_PLANE];
uint8_t bmode_blocks_wl;
uint8_t bmode_blocks_hl;
FRAME_COUNTS *counts;
TileInfo tile;
int mi_stride;
MODE_INFO **mi;
MODE_INFO *left_mi;
MODE_INFO *above_mi;
MB_MODE_INFO *left_mbmi;
MB_MODE_INFO *above_mbmi;
int up_available;
int left_available;
const vpx_prob (*partition_probs)[PARTITION_TYPES - 1];
/* Distance of MB away from frame edges */
int mb_to_left_edge;
int mb_to_right_edge;
int mb_to_top_edge;
int mb_to_bottom_edge;
FRAME_CONTEXT *fc;
int frame_parallel_decoding_mode;
/* pointers to reference frames */
RefBuffer *block_refs[2];
/* pointer to current frame */
const YV12_BUFFER_CONFIG *cur_buf;
ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16];
PARTITION_CONTEXT *above_seg_context;
PARTITION_CONTEXT left_seg_context[8];
#if CONFIG_VP9_HIGHBITDEPTH
/* Bit depth: 8, 10, 12 */
int bd;
#endif
int lossless;
int corrupted;
struct vpx_internal_error_info *error_info;
} MACROBLOCKD;
static INLINE BLOCK_SIZE get_subsize(BLOCK_SIZE bsize,
PARTITION_TYPE partition) {
return subsize_lookup[partition][bsize];
}
static const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES] = {
DCT_DCT, // DC
ADST_DCT, // V
DCT_ADST, // H
DCT_DCT, // D45
ADST_ADST, // D135
ADST_DCT, // D117
DCT_ADST, // D153
DCT_ADST, // D207
ADST_DCT, // D63
ADST_ADST, // TM
};
static INLINE TX_TYPE get_tx_type(PLANE_TYPE plane_type,
const MACROBLOCKD *xd) {
const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
if (plane_type != PLANE_TYPE_Y || xd->lossless || is_inter_block(mbmi))
return DCT_DCT;
return intra_mode_to_tx_type_lookup[mbmi->mode];
}
static INLINE TX_TYPE get_tx_type_4x4(PLANE_TYPE plane_type,
const MACROBLOCKD *xd, int ib) {
const MODE_INFO *const mi = xd->mi[0];
if (plane_type != PLANE_TYPE_Y || xd->lossless || is_inter_block(&mi->mbmi))
return DCT_DCT;
return intra_mode_to_tx_type_lookup[get_y_mode(mi, ib)];
}
void vp10_setup_block_planes(MACROBLOCKD *xd, int ss_x, int ss_y);
static INLINE TX_SIZE get_uv_tx_size_impl(TX_SIZE y_tx_size, BLOCK_SIZE bsize,
int xss, int yss) {
if (bsize < BLOCK_8X8) {
return TX_4X4;
} else {
const BLOCK_SIZE plane_bsize = ss_size_lookup[bsize][xss][yss];
return MIN(y_tx_size, max_txsize_lookup[plane_bsize]);
}
}
static INLINE TX_SIZE get_uv_tx_size(const MB_MODE_INFO *mbmi,
const struct macroblockd_plane *pd) {
return get_uv_tx_size_impl(mbmi->tx_size, mbmi->sb_type, pd->subsampling_x,
pd->subsampling_y);
}
static INLINE BLOCK_SIZE get_plane_block_size(BLOCK_SIZE bsize,
const struct macroblockd_plane *pd) {
return ss_size_lookup[bsize][pd->subsampling_x][pd->subsampling_y];
}
static INLINE void reset_skip_context(MACROBLOCKD *xd, BLOCK_SIZE bsize) {
int i;
for (i = 0; i < MAX_MB_PLANE; i++) {
struct macroblockd_plane *const pd = &xd->plane[i];
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize, pd);
memset(pd->above_context, 0,
sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_wide_lookup[plane_bsize]);
memset(pd->left_context, 0,
sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_high_lookup[plane_bsize]);
}
}
static INLINE const vpx_prob *get_y_mode_probs(const MODE_INFO *mi,
const MODE_INFO *above_mi,
const MODE_INFO *left_mi,
int block) {
const PREDICTION_MODE above = vp10_above_block_mode(mi, above_mi, block);
const PREDICTION_MODE left = vp10_left_block_mode(mi, left_mi, block);
return vp10_kf_y_mode_prob[above][left];
}
typedef void (*foreach_transformed_block_visitor)(int plane, int block,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size,
void *arg);
void vp10_foreach_transformed_block_in_plane(
const MACROBLOCKD *const xd, BLOCK_SIZE bsize, int plane,
foreach_transformed_block_visitor visit, void *arg);
void vp10_foreach_transformed_block(
const MACROBLOCKD* const xd, BLOCK_SIZE bsize,
foreach_transformed_block_visitor visit, void *arg);
static INLINE void txfrm_block_to_raster_xy(BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, int block,
int *x, int *y) {
const int bwl = b_width_log2_lookup[plane_bsize];
const int tx_cols_log2 = bwl - tx_size;
const int tx_cols = 1 << tx_cols_log2;
const int raster_mb = block >> (tx_size << 1);
*x = (raster_mb & (tx_cols - 1)) << tx_size;
*y = (raster_mb >> tx_cols_log2) << tx_size;
}
void vp10_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size, int has_eob,
int aoff, int loff);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_BLOCKD_H_

75
vp10/common/vp9_common.h Normal file
View File

@ -0,0 +1,75 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_COMMON_H_
#define VP9_COMMON_VP9_COMMON_H_
/* Interface header for common constant data structures and lookup tables */
#include <assert.h>
#include "./vpx_config.h"
#include "vpx_dsp/vpx_dsp_common.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx/vpx_integer.h"
#include "vp10/common/vp9_systemdependent.h"
#ifdef __cplusplus
extern "C" {
#endif
// Only need this for fixed-size arrays, for structs just assign.
#define vp10_copy(dest, src) { \
assert(sizeof(dest) == sizeof(src)); \
memcpy(dest, src, sizeof(src)); \
}
// Use this for variably-sized arrays.
#define vp10_copy_array(dest, src, n) { \
assert(sizeof(*dest) == sizeof(*src)); \
memcpy(dest, src, n * sizeof(*src)); \
}
#define vp10_zero(dest) memset(&(dest), 0, sizeof(dest))
#define vp10_zero_array(dest, n) memset(dest, 0, n * sizeof(*dest))
static INLINE int get_unsigned_bits(unsigned int num_values) {
return num_values > 0 ? get_msb(num_values) + 1 : 0;
}
#if CONFIG_DEBUG
#define CHECK_MEM_ERROR(cm, lval, expr) do { \
lval = (expr); \
if (!lval) \
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, \
"Failed to allocate "#lval" at %s:%d", \
__FILE__, __LINE__); \
} while (0)
#else
#define CHECK_MEM_ERROR(cm, lval, expr) do { \
lval = (expr); \
if (!lval) \
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR, \
"Failed to allocate "#lval); \
} while (0)
#endif
#define VP9_SYNC_CODE_0 0x49
#define VP9_SYNC_CODE_1 0x83
#define VP9_SYNC_CODE_2 0x42
#define VP9_FRAME_MARKER 0x2
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_COMMON_H_

13
vp10/common/vp9_common_data.c Normal file
View File

@ -0,0 +1,13 @@
/*
* Copyright (c) 2010 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/common/vp9_common_data.h"

174
vp10/common/vp9_common_data.h Normal file
View File

@ -0,0 +1,174 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_COMMON_DATA_H_
#define VP9_COMMON_VP9_COMMON_DATA_H_
#include "vp10/common/vp9_enums.h"
#include "vpx/vpx_integer.h"
#ifdef __cplusplus
extern "C" {
#endif
// Log 2 conversion lookup tables for block width and height
static const uint8_t b_width_log2_lookup[BLOCK_SIZES] =
{0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4};
static const uint8_t b_height_log2_lookup[BLOCK_SIZES] =
{0, 1, 0, 1, 2, 1, 2, 3, 2, 3, 4, 3, 4};
static const uint8_t num_4x4_blocks_wide_lookup[BLOCK_SIZES] =
{1, 1, 2, 2, 2, 4, 4, 4, 8, 8, 8, 16, 16};
static const uint8_t num_4x4_blocks_high_lookup[BLOCK_SIZES] =
{1, 2, 1, 2, 4, 2, 4, 8, 4, 8, 16, 8, 16};
// Log 2 conversion lookup tables for modeinfo width and height
static const uint8_t mi_width_log2_lookup[BLOCK_SIZES] =
{0, 0, 0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3};
static const uint8_t num_8x8_blocks_wide_lookup[BLOCK_SIZES] =
{1, 1, 1, 1, 1, 2, 2, 2, 4, 4, 4, 8, 8};
static const uint8_t num_8x8_blocks_high_lookup[BLOCK_SIZES] =
{1, 1, 1, 1, 2, 1, 2, 4, 2, 4, 8, 4, 8};
// MIN(3, MIN(b_width_log2(bsize), b_height_log2(bsize)))
static const uint8_t size_group_lookup[BLOCK_SIZES] =
{0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3};
static const uint8_t num_pels_log2_lookup[BLOCK_SIZES] =
{4, 5, 5, 6, 7, 7, 8, 9, 9, 10, 11, 11, 12};
static const PARTITION_TYPE partition_lookup[][BLOCK_SIZES] = {
{ // 4X4
// 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64
PARTITION_NONE, PARTITION_INVALID, PARTITION_INVALID,
PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID,
PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID,
PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID,
PARTITION_INVALID
}, { // 8X8
// 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64
PARTITION_SPLIT, PARTITION_VERT, PARTITION_HORZ, PARTITION_NONE,
PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID,
PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID,
PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID
}, { // 16X16
// 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64
PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT,
PARTITION_VERT, PARTITION_HORZ, PARTITION_NONE, PARTITION_INVALID,
PARTITION_INVALID, PARTITION_INVALID, PARTITION_INVALID,
PARTITION_INVALID, PARTITION_INVALID
}, { // 32X32
// 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64
PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT,
PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_VERT,
PARTITION_HORZ, PARTITION_NONE, PARTITION_INVALID,
PARTITION_INVALID, PARTITION_INVALID
}, { // 64X64
// 4X4, 4X8,8X4,8X8,8X16,16X8,16X16,16X32,32X16,32X32,32X64,64X32,64X64
PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT,
PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_SPLIT,
PARTITION_SPLIT, PARTITION_SPLIT, PARTITION_VERT, PARTITION_HORZ,
PARTITION_NONE
}
};
static const BLOCK_SIZE subsize_lookup[PARTITION_TYPES][BLOCK_SIZES] = {
{ // PARTITION_NONE
BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
BLOCK_8X8, BLOCK_8X16, BLOCK_16X8,
BLOCK_16X16, BLOCK_16X32, BLOCK_32X16,
BLOCK_32X32, BLOCK_32X64, BLOCK_64X32,
BLOCK_64X64,
}, { // PARTITION_HORZ
BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_8X4, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_16X8, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_32X16, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_64X32,
}, { // PARTITION_VERT
BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_4X8, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_8X16, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_16X32, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_32X64,
}, { // PARTITION_SPLIT
BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_4X4, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_8X8, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_16X16, BLOCK_INVALID, BLOCK_INVALID,
BLOCK_32X32,
}
};
static const TX_SIZE max_txsize_lookup[BLOCK_SIZES] = {
TX_4X4, TX_4X4, TX_4X4,
TX_8X8, TX_8X8, TX_8X8,
TX_16X16, TX_16X16, TX_16X16,
TX_32X32, TX_32X32, TX_32X32, TX_32X32
};
static const BLOCK_SIZE txsize_to_bsize[TX_SIZES] = {
BLOCK_4X4, // TX_4X4
BLOCK_8X8, // TX_8X8
BLOCK_16X16, // TX_16X16
BLOCK_32X32, // TX_32X32
};
static const TX_SIZE tx_mode_to_biggest_tx_size[TX_MODES] = {
TX_4X4, // ONLY_4X4
TX_8X8, // ALLOW_8X8
TX_16X16, // ALLOW_16X16
TX_32X32, // ALLOW_32X32
TX_32X32, // TX_MODE_SELECT
};
static const BLOCK_SIZE ss_size_lookup[BLOCK_SIZES][2][2] = {
// ss_x == 0 ss_x == 0 ss_x == 1 ss_x == 1
// ss_y == 0 ss_y == 1 ss_y == 0 ss_y == 1
{{BLOCK_4X4, BLOCK_INVALID}, {BLOCK_INVALID, BLOCK_INVALID}},
{{BLOCK_4X8, BLOCK_4X4}, {BLOCK_INVALID, BLOCK_INVALID}},
{{BLOCK_8X4, BLOCK_INVALID}, {BLOCK_4X4, BLOCK_INVALID}},
{{BLOCK_8X8, BLOCK_8X4}, {BLOCK_4X8, BLOCK_4X4}},
{{BLOCK_8X16, BLOCK_8X8}, {BLOCK_INVALID, BLOCK_4X8}},
{{BLOCK_16X8, BLOCK_INVALID}, {BLOCK_8X8, BLOCK_8X4}},
{{BLOCK_16X16, BLOCK_16X8}, {BLOCK_8X16, BLOCK_8X8}},
{{BLOCK_16X32, BLOCK_16X16}, {BLOCK_INVALID, BLOCK_8X16}},
{{BLOCK_32X16, BLOCK_INVALID}, {BLOCK_16X16, BLOCK_16X8}},
{{BLOCK_32X32, BLOCK_32X16}, {BLOCK_16X32, BLOCK_16X16}},
{{BLOCK_32X64, BLOCK_32X32}, {BLOCK_INVALID, BLOCK_16X32}},
{{BLOCK_64X32, BLOCK_INVALID}, {BLOCK_32X32, BLOCK_32X16}},
{{BLOCK_64X64, BLOCK_64X32}, {BLOCK_32X64, BLOCK_32X32}},
};
// Generates 4 bit field in which each bit set to 1 represents
// a blocksize partition 1111 means we split 64x64, 32x32, 16x16
// and 8x8. 1000 means we just split the 64x64 to 32x32
static const struct {
PARTITION_CONTEXT above;
PARTITION_CONTEXT left;
} partition_context_lookup[BLOCK_SIZES]= {
{15, 15}, // 4X4 - {0b1111, 0b1111}
{15, 14}, // 4X8 - {0b1111, 0b1110}
{14, 15}, // 8X4 - {0b1110, 0b1111}
{14, 14}, // 8X8 - {0b1110, 0b1110}
{14, 12}, // 8X16 - {0b1110, 0b1100}
{12, 14}, // 16X8 - {0b1100, 0b1110}
{12, 12}, // 16X16 - {0b1100, 0b1100}
{12, 8 }, // 16X32 - {0b1100, 0b1000}
{8, 12}, // 32X16 - {0b1000, 0b1100}
{8, 8 }, // 32X32 - {0b1000, 0b1000}
{8, 0 }, // 32X64 - {0b1000, 0b0000}
{0, 8 }, // 64X32 - {0b0000, 0b1000}
{0, 0 }, // 64X64 - {0b0000, 0b0000}
};
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_COMMON_DATA_H_

91
vp10/common/vp9_debugmodes.c Normal file
View File

@ -0,0 +1,91 @@
/*
* Copyright (c) 2010 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 <stdio.h>
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_onyxc_int.h"
static void log_frame_info(VP9_COMMON *cm, const char *str, FILE *f) {
fprintf(f, "%s", str);
fprintf(f, "(Frame %d, Show:%d, Q:%d): \n", cm->current_video_frame,
cm->show_frame, cm->base_qindex);
}
/* This function dereferences a pointer to the mbmi structure
* and uses the passed in member offset to print out the value of an integer
* for each mbmi member value in the mi structure.
*/
static void print_mi_data(VP9_COMMON *cm, FILE *file, const char *descriptor,
size_t member_offset) {
int mi_row, mi_col;
MODE_INFO **mi = cm->mi_grid_visible;
int rows = cm->mi_rows;
int cols = cm->mi_cols;
char prefix = descriptor[0];
log_frame_info(cm, descriptor, file);
for (mi_row = 0; mi_row < rows; mi_row++) {
fprintf(file, "%c ", prefix);
for (mi_col = 0; mi_col < cols; mi_col++) {
fprintf(file, "%2d ",
*((int*) ((char *) (&mi[0]->mbmi) +
member_offset)));
mi++;
}
fprintf(file, "\n");
mi += 8;
}
fprintf(file, "\n");
}
void vp10_print_modes_and_motion_vectors(VP9_COMMON *cm, const char *file) {
int mi_row;
int mi_col;
FILE *mvs = fopen(file, "a");
MODE_INFO **mi = cm->mi_grid_visible;
int rows = cm->mi_rows;
int cols = cm->mi_cols;
print_mi_data(cm, mvs, "Partitions:", offsetof(MB_MODE_INFO, sb_type));
print_mi_data(cm, mvs, "Modes:", offsetof(MB_MODE_INFO, mode));
print_mi_data(cm, mvs, "Ref frame:", offsetof(MB_MODE_INFO, ref_frame[0]));
print_mi_data(cm, mvs, "Transform:", offsetof(MB_MODE_INFO, tx_size));
print_mi_data(cm, mvs, "UV Modes:", offsetof(MB_MODE_INFO, uv_mode));
// output skip infomation.
log_frame_info(cm, "Skips:", mvs);
for (mi_row = 0; mi_row < rows; mi_row++) {
fprintf(mvs, "S ");
for (mi_col = 0; mi_col < cols; mi_col++) {
fprintf(mvs, "%2d ", mi[0]->mbmi.skip);
mi++;
}
fprintf(mvs, "\n");
mi += 8;
}
fprintf(mvs, "\n");
// output motion vectors.
log_frame_info(cm, "Vectors ", mvs);
mi = cm->mi_grid_visible;
for (mi_row = 0; mi_row < rows; mi_row++) {
fprintf(mvs, "V ");
for (mi_col = 0; mi_col < cols; mi_col++) {
fprintf(mvs, "%4d:%4d ", mi[0]->mbmi.mv[0].as_mv.row,
mi[0]->mbmi.mv[0].as_mv.col);
mi++;
}
fprintf(mvs, "\n");
mi += 8;
}
fprintf(mvs, "\n");
fclose(mvs);
}

817
vp10/common/vp9_entropy.c Normal file
View File

@ -0,0 +1,817 @@
/*
* Copyright (c) 2010 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/common/vp9_entropy.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_onyxc_int.h"
#include "vp10/common/vp9_entropymode.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx/vpx_integer.h"
// Unconstrained Node Tree
const vpx_tree_index vp10_coef_con_tree[TREE_SIZE(ENTROPY_TOKENS)] = {
2, 6, // 0 = LOW_VAL
-TWO_TOKEN, 4, // 1 = TWO
-THREE_TOKEN, -FOUR_TOKEN, // 2 = THREE
8, 10, // 3 = HIGH_LOW
-CATEGORY1_TOKEN, -CATEGORY2_TOKEN, // 4 = CAT_ONE
12, 14, // 5 = CAT_THREEFOUR
-CATEGORY3_TOKEN, -CATEGORY4_TOKEN, // 6 = CAT_THREE
-CATEGORY5_TOKEN, -CATEGORY6_TOKEN // 7 = CAT_FIVE
};
const vpx_prob vp10_cat1_prob[] = { 159 };
const vpx_prob vp10_cat2_prob[] = { 165, 145 };
const vpx_prob vp10_cat3_prob[] = { 173, 148, 140 };
const vpx_prob vp10_cat4_prob[] = { 176, 155, 140, 135 };
const vpx_prob vp10_cat5_prob[] = { 180, 157, 141, 134, 130 };
const vpx_prob vp10_cat6_prob[] = {
254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129
};
#if CONFIG_VP9_HIGHBITDEPTH
const vpx_prob vp10_cat1_prob_high10[] = { 159 };
const vpx_prob vp10_cat2_prob_high10[] = { 165, 145 };
const vpx_prob vp10_cat3_prob_high10[] = { 173, 148, 140 };
const vpx_prob vp10_cat4_prob_high10[] = { 176, 155, 140, 135 };
const vpx_prob vp10_cat5_prob_high10[] = { 180, 157, 141, 134, 130 };
const vpx_prob vp10_cat6_prob_high10[] = {
255, 255, 254, 254, 254, 252, 249, 243,
230, 196, 177, 153, 140, 133, 130, 129
};
const vpx_prob vp10_cat1_prob_high12[] = { 159 };
const vpx_prob vp10_cat2_prob_high12[] = { 165, 145 };
const vpx_prob vp10_cat3_prob_high12[] = { 173, 148, 140 };
const vpx_prob vp10_cat4_prob_high12[] = { 176, 155, 140, 135 };
const vpx_prob vp10_cat5_prob_high12[] = { 180, 157, 141, 134, 130 };
const vpx_prob vp10_cat6_prob_high12[] = {
255, 255, 255, 255, 254, 254, 254, 252, 249,
243, 230, 196, 177, 153, 140, 133, 130, 129
};
#endif
const uint8_t vp10_coefband_trans_8x8plus[1024] = {
0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4,
4, 4, 4, 4, 4, 5,
// beyond MAXBAND_INDEX+1 all values are filled as 5
5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
};
const uint8_t vp10_coefband_trans_4x4[16] = {
0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 5, 5, 5,
};
const uint8_t vp10_pt_energy_class[ENTROPY_TOKENS] = {
0, 1, 2, 3, 3, 4, 4, 5, 5, 5, 5, 5
};
// Model obtained from a 2-sided zero-centerd distribuition derived
// from a Pareto distribution. The cdf of the distribution is:
// cdf(x) = 0.5 + 0.5 * sgn(x) * [1 - {alpha/(alpha + |x|)} ^ beta]
//
// For a given beta and a given probablity of the 1-node, the alpha
// is first solved, and then the {alpha, beta} pair is used to generate
// the probabilities for the rest of the nodes.
// beta = 8
// Every odd line in this table can be generated from the even lines
// by averaging :
// vp10_pareto8_full[l][node] = (vp10_pareto8_full[l-1][node] +
// vp10_pareto8_full[l+1][node] ) >> 1;
const vpx_prob vp10_pareto8_full[COEFF_PROB_MODELS][MODEL_NODES] = {
{ 3, 86, 128, 6, 86, 23, 88, 29},
{ 6, 86, 128, 11, 87, 42, 91, 52},
{ 9, 86, 129, 17, 88, 61, 94, 76},
{ 12, 86, 129, 22, 88, 77, 97, 93},
{ 15, 87, 129, 28, 89, 93, 100, 110},
{ 17, 87, 129, 33, 90, 105, 103, 123},
{ 20, 88, 130, 38, 91, 118, 106, 136},
{ 23, 88, 130, 43, 91, 128, 108, 146},
{ 26, 89, 131, 48, 92, 139, 111, 156},
{ 28, 89, 131, 53, 93, 147, 114, 163},
{ 31, 90, 131, 58, 94, 156, 117, 171},
{ 34, 90, 131, 62, 94, 163, 119, 177},
{ 37, 90, 132, 66, 95, 171, 122, 184},
{ 39, 90, 132, 70, 96, 177, 124, 189},
{ 42, 91, 132, 75, 97, 183, 127, 194},
{ 44, 91, 132, 79, 97, 188, 129, 198},
{ 47, 92, 133, 83, 98, 193, 132, 202},
{ 49, 92, 133, 86, 99, 197, 134, 205},
{ 52, 93, 133, 90, 100, 201, 137, 208},
{ 54, 93, 133, 94, 100, 204, 139, 211},
{ 57, 94, 134, 98, 101, 208, 142, 214},
{ 59, 94, 134, 101, 102, 211, 144, 216},
{ 62, 94, 135, 105, 103, 214, 146, 218},
{ 64, 94, 135, 108, 103, 216, 148, 220},
{ 66, 95, 135, 111, 104, 219, 151, 222},
{ 68, 95, 135, 114, 105, 221, 153, 223},
{ 71, 96, 136, 117, 106, 224, 155, 225},
{ 73, 96, 136, 120, 106, 225, 157, 226},
{ 76, 97, 136, 123, 107, 227, 159, 228},
{ 78, 97, 136, 126, 108, 229, 160, 229},
{ 80, 98, 137, 129, 109, 231, 162, 231},
{ 82, 98, 137, 131, 109, 232, 164, 232},
{ 84, 98, 138, 134, 110, 234, 166, 233},
{ 86, 98, 138, 137, 111, 235, 168, 234},
{ 89, 99, 138, 140, 112, 236, 170, 235},
{ 91, 99, 138, 142, 112, 237, 171, 235},
{ 93, 100, 139, 145, 113, 238, 173, 236},
{ 95, 100, 139, 147, 114, 239, 174, 237},
{ 97, 101, 140, 149, 115, 240, 176, 238},
{ 99, 101, 140, 151, 115, 241, 177, 238},
{101, 102, 140, 154, 116, 242, 179, 239},
{103, 102, 140, 156, 117, 242, 180, 239},
{105, 103, 141, 158, 118, 243, 182, 240},
{107, 103, 141, 160, 118, 243, 183, 240},
{109, 104, 141, 162, 119, 244, 185, 241},
{111, 104, 141, 164, 119, 244, 186, 241},
{113, 104, 142, 166, 120, 245, 187, 242},
{114, 104, 142, 168, 121, 245, 188, 242},
{116, 105, 143, 170, 122, 246, 190, 243},
{118, 105, 143, 171, 122, 246, 191, 243},
{120, 106, 143, 173, 123, 247, 192, 244},
{121, 106, 143, 175, 124, 247, 193, 244},
{123, 107, 144, 177, 125, 248, 195, 244},
{125, 107, 144, 178, 125, 248, 196, 244},
{127, 108, 145, 180, 126, 249, 197, 245},
{128, 108, 145, 181, 127, 249, 198, 245},
{130, 109, 145, 183, 128, 249, 199, 245},
{132, 109, 145, 184, 128, 249, 200, 245},
{134, 110, 146, 186, 129, 250, 201, 246},
{135, 110, 146, 187, 130, 250, 202, 246},
{137, 111, 147, 189, 131, 251, 203, 246},
{138, 111, 147, 190, 131, 251, 204, 246},
{140, 112, 147, 192, 132, 251, 205, 247},
{141, 112, 147, 193, 132, 251, 206, 247},
{143, 113, 148, 194, 133, 251, 207, 247},
{144, 113, 148, 195, 134, 251, 207, 247},
{146, 114, 149, 197, 135, 252, 208, 248},
{147, 114, 149, 198, 135, 252, 209, 248},
{149, 115, 149, 199, 136, 252, 210, 248},
{150, 115, 149, 200, 137, 252, 210, 248},
{152, 115, 150, 201, 138, 252, 211, 248},
{153, 115, 150, 202, 138, 252, 212, 248},
{155, 116, 151, 204, 139, 253, 213, 249},
{156, 116, 151, 205, 139, 253, 213, 249},
{158, 117, 151, 206, 140, 253, 214, 249},
{159, 117, 151, 207, 141, 253, 215, 249},
{161, 118, 152, 208, 142, 253, 216, 249},
{162, 118, 152, 209, 142, 253, 216, 249},
{163, 119, 153, 210, 143, 253, 217, 249},
{164, 119, 153, 211, 143, 253, 217, 249},
{166, 120, 153, 212, 144, 254, 218, 250},
{167, 120, 153, 212, 145, 254, 219, 250},
{168, 121, 154, 213, 146, 254, 220, 250},
{169, 121, 154, 214, 146, 254, 220, 250},
{171, 122, 155, 215, 147, 254, 221, 250},
{172, 122, 155, 216, 147, 254, 221, 250},
{173, 123, 155, 217, 148, 254, 222, 250},
{174, 123, 155, 217, 149, 254, 222, 250},
{176, 124, 156, 218, 150, 254, 223, 250},
{177, 124, 156, 219, 150, 254, 223, 250},
{178, 125, 157, 220, 151, 254, 224, 251},
{179, 125, 157, 220, 151, 254, 224, 251},
{180, 126, 157, 221, 152, 254, 225, 251},
{181, 126, 157, 221, 152, 254, 225, 251},
{183, 127, 158, 222, 153, 254, 226, 251},
{184, 127, 158, 223, 154, 254, 226, 251},
{185, 128, 159, 224, 155, 255, 227, 251},
{186, 128, 159, 224, 155, 255, 227, 251},
{187, 129, 160, 225, 156, 255, 228, 251},
{188, 130, 160, 225, 156, 255, 228, 251},
{189, 131, 160, 226, 157, 255, 228, 251},
{190, 131, 160, 226, 158, 255, 228, 251},
{191, 132, 161, 227, 159, 255, 229, 251},
{192, 132, 161, 227, 159, 255, 229, 251},
{193, 133, 162, 228, 160, 255, 230, 252},
{194, 133, 162, 229, 160, 255, 230, 252},
{195, 134, 163, 230, 161, 255, 231, 252},
{196, 134, 163, 230, 161, 255, 231, 252},
{197, 135, 163, 231, 162, 255, 231, 252},
{198, 135, 163, 231, 162, 255, 231, 252},
{199, 136, 164, 232, 163, 255, 232, 252},
{200, 136, 164, 232, 164, 255, 232, 252},
{201, 137, 165, 233, 165, 255, 233, 252},
{201, 137, 165, 233, 165, 255, 233, 252},
{202, 138, 166, 233, 166, 255, 233, 252},
{203, 138, 166, 233, 166, 255, 233, 252},
{204, 139, 166, 234, 167, 255, 234, 252},
{205, 139, 166, 234, 167, 255, 234, 252},
{206, 140, 167, 235, 168, 255, 235, 252},
{206, 140, 167, 235, 168, 255, 235, 252},
{207, 141, 168, 236, 169, 255, 235, 252},
{208, 141, 168, 236, 170, 255, 235, 252},
{209, 142, 169, 237, 171, 255, 236, 252},
{209, 143, 169, 237, 171, 255, 236, 252},
{210, 144, 169, 237, 172, 255, 236, 252},
{211, 144, 169, 237, 172, 255, 236, 252},
{212, 145, 170, 238, 173, 255, 237, 252},
{213, 145, 170, 238, 173, 255, 237, 252},
{214, 146, 171, 239, 174, 255, 237, 253},
{214, 146, 171, 239, 174, 255, 237, 253},
{215, 147, 172, 240, 175, 255, 238, 253},
{215, 147, 172, 240, 175, 255, 238, 253},
{216, 148, 173, 240, 176, 255, 238, 253},
{217, 148, 173, 240, 176, 255, 238, 253},
{218, 149, 173, 241, 177, 255, 239, 253},
{218, 149, 173, 241, 178, 255, 239, 253},
{219, 150, 174, 241, 179, 255, 239, 253},
{219, 151, 174, 241, 179, 255, 239, 253},
{220, 152, 175, 242, 180, 255, 240, 253},
{221, 152, 175, 242, 180, 255, 240, 253},
{222, 153, 176, 242, 181, 255, 240, 253},
{222, 153, 176, 242, 181, 255, 240, 253},
{223, 154, 177, 243, 182, 255, 240, 253},
{223, 154, 177, 243, 182, 255, 240, 253},
{224, 155, 178, 244, 183, 255, 241, 253},
{224, 155, 178, 244, 183, 255, 241, 253},
{225, 156, 178, 244, 184, 255, 241, 253},
{225, 157, 178, 244, 184, 255, 241, 253},
{226, 158, 179, 244, 185, 255, 242, 253},
{227, 158, 179, 244, 185, 255, 242, 253},
{228, 159, 180, 245, 186, 255, 242, 253},
{228, 159, 180, 245, 186, 255, 242, 253},
{229, 160, 181, 245, 187, 255, 242, 253},
{229, 160, 181, 245, 187, 255, 242, 253},
{230, 161, 182, 246, 188, 255, 243, 253},
{230, 162, 182, 246, 188, 255, 243, 253},
{231, 163, 183, 246, 189, 255, 243, 253},
{231, 163, 183, 246, 189, 255, 243, 253},
{232, 164, 184, 247, 190, 255, 243, 253},
{232, 164, 184, 247, 190, 255, 243, 253},
{233, 165, 185, 247, 191, 255, 244, 253},
{233, 165, 185, 247, 191, 255, 244, 253},
{234, 166, 185, 247, 192, 255, 244, 253},
{234, 167, 185, 247, 192, 255, 244, 253},
{235, 168, 186, 248, 193, 255, 244, 253},
{235, 168, 186, 248, 193, 255, 244, 253},
{236, 169, 187, 248, 194, 255, 244, 253},
{236, 169, 187, 248, 194, 255, 244, 253},
{236, 170, 188, 248, 195, 255, 245, 253},
{236, 170, 188, 248, 195, 255, 245, 253},
{237, 171, 189, 249, 196, 255, 245, 254},
{237, 172, 189, 249, 196, 255, 245, 254},
{238, 173, 190, 249, 197, 255, 245, 254},
{238, 173, 190, 249, 197, 255, 245, 254},
{239, 174, 191, 249, 198, 255, 245, 254},
{239, 174, 191, 249, 198, 255, 245, 254},
{240, 175, 192, 249, 199, 255, 246, 254},
{240, 176, 192, 249, 199, 255, 246, 254},
{240, 177, 193, 250, 200, 255, 246, 254},
{240, 177, 193, 250, 200, 255, 246, 254},
{241, 178, 194, 250, 201, 255, 246, 254},
{241, 178, 194, 250, 201, 255, 246, 254},
{242, 179, 195, 250, 202, 255, 246, 254},
{242, 180, 195, 250, 202, 255, 246, 254},
{242, 181, 196, 250, 203, 255, 247, 254},
{242, 181, 196, 250, 203, 255, 247, 254},
{243, 182, 197, 251, 204, 255, 247, 254},
{243, 183, 197, 251, 204, 255, 247, 254},
{244, 184, 198, 251, 205, 255, 247, 254},
{244, 184, 198, 251, 205, 255, 247, 254},
{244, 185, 199, 251, 206, 255, 247, 254},
{244, 185, 199, 251, 206, 255, 247, 254},
{245, 186, 200, 251, 207, 255, 247, 254},
{245, 187, 200, 251, 207, 255, 247, 254},
{246, 188, 201, 252, 207, 255, 248, 254},
{246, 188, 201, 252, 207, 255, 248, 254},
{246, 189, 202, 252, 208, 255, 248, 254},
{246, 190, 202, 252, 208, 255, 248, 254},
{247, 191, 203, 252, 209, 255, 248, 254},
{247, 191, 203, 252, 209, 255, 248, 254},
{247, 192, 204, 252, 210, 255, 248, 254},
{247, 193, 204, 252, 210, 255, 248, 254},
{248, 194, 205, 252, 211, 255, 248, 254},
{248, 194, 205, 252, 211, 255, 248, 254},
{248, 195, 206, 252, 212, 255, 249, 254},
{248, 196, 206, 252, 212, 255, 249, 254},
{249, 197, 207, 253, 213, 255, 249, 254},
{249, 197, 207, 253, 213, 255, 249, 254},
{249, 198, 208, 253, 214, 255, 249, 254},
{249, 199, 209, 253, 214, 255, 249, 254},
{250, 200, 210, 253, 215, 255, 249, 254},
{250, 200, 210, 253, 215, 255, 249, 254},
{250, 201, 211, 253, 215, 255, 249, 254},
{250, 202, 211, 253, 215, 255, 249, 254},
{250, 203, 212, 253, 216, 255, 249, 254},
{250, 203, 212, 253, 216, 255, 249, 254},
{251, 204, 213, 253, 217, 255, 250, 254},
{251, 205, 213, 253, 217, 255, 250, 254},
{251, 206, 214, 254, 218, 255, 250, 254},
{251, 206, 215, 254, 218, 255, 250, 254},
{252, 207, 216, 254, 219, 255, 250, 254},
{252, 208, 216, 254, 219, 255, 250, 254},
{252, 209, 217, 254, 220, 255, 250, 254},
{252, 210, 217, 254, 220, 255, 250, 254},
{252, 211, 218, 254, 221, 255, 250, 254},
{252, 212, 218, 254, 221, 255, 250, 254},
{253, 213, 219, 254, 222, 255, 250, 254},
{253, 213, 220, 254, 222, 255, 250, 254},
{253, 214, 221, 254, 223, 255, 250, 254},
{253, 215, 221, 254, 223, 255, 250, 254},
{253, 216, 222, 254, 224, 255, 251, 254},
{253, 217, 223, 254, 224, 255, 251, 254},
{253, 218, 224, 254, 225, 255, 251, 254},
{253, 219, 224, 254, 225, 255, 251, 254},
{254, 220, 225, 254, 225, 255, 251, 254},
{254, 221, 226, 254, 225, 255, 251, 254},
{254, 222, 227, 255, 226, 255, 251, 254},
{254, 223, 227, 255, 226, 255, 251, 254},
{254, 224, 228, 255, 227, 255, 251, 254},
{254, 225, 229, 255, 227, 255, 251, 254},
{254, 226, 230, 255, 228, 255, 251, 254},
{254, 227, 230, 255, 229, 255, 251, 254},
{255, 228, 231, 255, 230, 255, 251, 254},
{255, 229, 232, 255, 230, 255, 251, 254},
{255, 230, 233, 255, 231, 255, 252, 254},
{255, 231, 234, 255, 231, 255, 252, 254},
{255, 232, 235, 255, 232, 255, 252, 254},
{255, 233, 236, 255, 232, 255, 252, 254},
{255, 235, 237, 255, 233, 255, 252, 254},
{255, 236, 238, 255, 234, 255, 252, 254},
{255, 238, 240, 255, 235, 255, 252, 255},
{255, 239, 241, 255, 235, 255, 252, 254},
{255, 241, 243, 255, 236, 255, 252, 254},
{255, 243, 245, 255, 237, 255, 252, 254},
{255, 246, 247, 255, 239, 255, 253, 255},
{255, 246, 247, 255, 239, 255, 253, 255},
};
static const vp10_coeff_probs_model default_coef_probs_4x4[PLANE_TYPES] = {
{ // Y plane
{ // Intra
{ // Band 0
{ 195, 29, 183 }, { 84, 49, 136 }, { 8, 42, 71 }
}, { // Band 1
{ 31, 107, 169 }, { 35, 99, 159 }, { 17, 82, 140 },
{ 8, 66, 114 }, { 2, 44, 76 }, { 1, 19, 32 }
}, { // Band 2
{ 40, 132, 201 }, { 29, 114, 187 }, { 13, 91, 157 },
{ 7, 75, 127 }, { 3, 58, 95 }, { 1, 28, 47 }
}, { // Band 3
{ 69, 142, 221 }, { 42, 122, 201 }, { 15, 91, 159 },
{ 6, 67, 121 }, { 1, 42, 77 }, { 1, 17, 31 }
}, { // Band 4
{ 102, 148, 228 }, { 67, 117, 204 }, { 17, 82, 154 },
{ 6, 59, 114 }, { 2, 39, 75 }, { 1, 15, 29 }
}, { // Band 5
{ 156, 57, 233 }, { 119, 57, 212 }, { 58, 48, 163 },
{ 29, 40, 124 }, { 12, 30, 81 }, { 3, 12, 31 }
}
}, { // Inter
{ // Band 0
{ 191, 107, 226 }, { 124, 117, 204 }, { 25, 99, 155 }
}, { // Band 1
{ 29, 148, 210 }, { 37, 126, 194 }, { 8, 93, 157 },
{ 2, 68, 118 }, { 1, 39, 69 }, { 1, 17, 33 }
}, { // Band 2
{ 41, 151, 213 }, { 27, 123, 193 }, { 3, 82, 144 },
{ 1, 58, 105 }, { 1, 32, 60 }, { 1, 13, 26 }
}, { // Band 3
{ 59, 159, 220 }, { 23, 126, 198 }, { 4, 88, 151 },
{ 1, 66, 114 }, { 1, 38, 71 }, { 1, 18, 34 }
}, { // Band 4
{ 114, 136, 232 }, { 51, 114, 207 }, { 11, 83, 155 },
{ 3, 56, 105 }, { 1, 33, 65 }, { 1, 17, 34 }
}, { // Band 5
{ 149, 65, 234 }, { 121, 57, 215 }, { 61, 49, 166 },
{ 28, 36, 114 }, { 12, 25, 76 }, { 3, 16, 42 }
}
}
}, { // UV plane
{ // Intra
{ // Band 0
{ 214, 49, 220 }, { 132, 63, 188 }, { 42, 65, 137 }
}, { // Band 1
{ 85, 137, 221 }, { 104, 131, 216 }, { 49, 111, 192 },
{ 21, 87, 155 }, { 2, 49, 87 }, { 1, 16, 28 }
}, { // Band 2
{ 89, 163, 230 }, { 90, 137, 220 }, { 29, 100, 183 },
{ 10, 70, 135 }, { 2, 42, 81 }, { 1, 17, 33 }
}, { // Band 3
{ 108, 167, 237 }, { 55, 133, 222 }, { 15, 97, 179 },
{ 4, 72, 135 }, { 1, 45, 85 }, { 1, 19, 38 }
}, { // Band 4
{ 124, 146, 240 }, { 66, 124, 224 }, { 17, 88, 175 },
{ 4, 58, 122 }, { 1, 36, 75 }, { 1, 18, 37 }
}, { // Band 5
{ 141, 79, 241 }, { 126, 70, 227 }, { 66, 58, 182 },
{ 30, 44, 136 }, { 12, 34, 96 }, { 2, 20, 47 }
}
}, { // Inter
{ // Band 0
{ 229, 99, 249 }, { 143, 111, 235 }, { 46, 109, 192 }
}, { // Band 1
{ 82, 158, 236 }, { 94, 146, 224 }, { 25, 117, 191 },
{ 9, 87, 149 }, { 3, 56, 99 }, { 1, 33, 57 }
}, { // Band 2
{ 83, 167, 237 }, { 68, 145, 222 }, { 10, 103, 177 },
{ 2, 72, 131 }, { 1, 41, 79 }, { 1, 20, 39 }
}, { // Band 3
{ 99, 167, 239 }, { 47, 141, 224 }, { 10, 104, 178 },
{ 2, 73, 133 }, { 1, 44, 85 }, { 1, 22, 47 }
}, { // Band 4
{ 127, 145, 243 }, { 71, 129, 228 }, { 17, 93, 177 },
{ 3, 61, 124 }, { 1, 41, 84 }, { 1, 21, 52 }
}, { // Band 5
{ 157, 78, 244 }, { 140, 72, 231 }, { 69, 58, 184 },
{ 31, 44, 137 }, { 14, 38, 105 }, { 8, 23, 61 }
}
}
}
};
static const vp10_coeff_probs_model default_coef_probs_8x8[PLANE_TYPES] = {
{ // Y plane
{ // Intra
{ // Band 0
{ 125, 34, 187 }, { 52, 41, 133 }, { 6, 31, 56 }
}, { // Band 1
{ 37, 109, 153 }, { 51, 102, 147 }, { 23, 87, 128 },
{ 8, 67, 101 }, { 1, 41, 63 }, { 1, 19, 29 }
}, { // Band 2
{ 31, 154, 185 }, { 17, 127, 175 }, { 6, 96, 145 },
{ 2, 73, 114 }, { 1, 51, 82 }, { 1, 28, 45 }
}, { // Band 3
{ 23, 163, 200 }, { 10, 131, 185 }, { 2, 93, 148 },
{ 1, 67, 111 }, { 1, 41, 69 }, { 1, 14, 24 }
}, { // Band 4
{ 29, 176, 217 }, { 12, 145, 201 }, { 3, 101, 156 },
{ 1, 69, 111 }, { 1, 39, 63 }, { 1, 14, 23 }
}, { // Band 5
{ 57, 192, 233 }, { 25, 154, 215 }, { 6, 109, 167 },
{ 3, 78, 118 }, { 1, 48, 69 }, { 1, 21, 29 }
}
}, { // Inter
{ // Band 0
{ 202, 105, 245 }, { 108, 106, 216 }, { 18, 90, 144 }
}, { // Band 1
{ 33, 172, 219 }, { 64, 149, 206 }, { 14, 117, 177 },
{ 5, 90, 141 }, { 2, 61, 95 }, { 1, 37, 57 }
}, { // Band 2
{ 33, 179, 220 }, { 11, 140, 198 }, { 1, 89, 148 },
{ 1, 60, 104 }, { 1, 33, 57 }, { 1, 12, 21 }
}, { // Band 3
{ 30, 181, 221 }, { 8, 141, 198 }, { 1, 87, 145 },
{ 1, 58, 100 }, { 1, 31, 55 }, { 1, 12, 20 }
}, { // Band 4
{ 32, 186, 224 }, { 7, 142, 198 }, { 1, 86, 143 },
{ 1, 58, 100 }, { 1, 31, 55 }, { 1, 12, 22 }
}, { // Band 5
{ 57, 192, 227 }, { 20, 143, 204 }, { 3, 96, 154 },
{ 1, 68, 112 }, { 1, 42, 69 }, { 1, 19, 32 }
}
}
}, { // UV plane
{ // Intra
{ // Band 0
{ 212, 35, 215 }, { 113, 47, 169 }, { 29, 48, 105 }
}, { // Band 1
{ 74, 129, 203 }, { 106, 120, 203 }, { 49, 107, 178 },
{ 19, 84, 144 }, { 4, 50, 84 }, { 1, 15, 25 }
}, { // Band 2
{ 71, 172, 217 }, { 44, 141, 209 }, { 15, 102, 173 },
{ 6, 76, 133 }, { 2, 51, 89 }, { 1, 24, 42 }
}, { // Band 3
{ 64, 185, 231 }, { 31, 148, 216 }, { 8, 103, 175 },
{ 3, 74, 131 }, { 1, 46, 81 }, { 1, 18, 30 }
}, { // Band 4
{ 65, 196, 235 }, { 25, 157, 221 }, { 5, 105, 174 },
{ 1, 67, 120 }, { 1, 38, 69 }, { 1, 15, 30 }
}, { // Band 5
{ 65, 204, 238 }, { 30, 156, 224 }, { 7, 107, 177 },
{ 2, 70, 124 }, { 1, 42, 73 }, { 1, 18, 34 }
}
}, { // Inter
{ // Band 0
{ 225, 86, 251 }, { 144, 104, 235 }, { 42, 99, 181 }
}, { // Band 1
{ 85, 175, 239 }, { 112, 165, 229 }, { 29, 136, 200 },
{ 12, 103, 162 }, { 6, 77, 123 }, { 2, 53, 84 }
}, { // Band 2
{ 75, 183, 239 }, { 30, 155, 221 }, { 3, 106, 171 },
{ 1, 74, 128 }, { 1, 44, 76 }, { 1, 17, 28 }
}, { // Band 3
{ 73, 185, 240 }, { 27, 159, 222 }, { 2, 107, 172 },
{ 1, 75, 127 }, { 1, 42, 73 }, { 1, 17, 29 }
}, { // Band 4
{ 62, 190, 238 }, { 21, 159, 222 }, { 2, 107, 172 },
{ 1, 72, 122 }, { 1, 40, 71 }, { 1, 18, 32 }
}, { // Band 5
{ 61, 199, 240 }, { 27, 161, 226 }, { 4, 113, 180 },
{ 1, 76, 129 }, { 1, 46, 80 }, { 1, 23, 41 }
}
}
}
};
static const vp10_coeff_probs_model default_coef_probs_16x16[PLANE_TYPES] = {
{ // Y plane
{ // Intra
{ // Band 0
{ 7, 27, 153 }, { 5, 30, 95 }, { 1, 16, 30 }
}, { // Band 1
{ 50, 75, 127 }, { 57, 75, 124 }, { 27, 67, 108 },
{ 10, 54, 86 }, { 1, 33, 52 }, { 1, 12, 18 }
}, { // Band 2
{ 43, 125, 151 }, { 26, 108, 148 }, { 7, 83, 122 },
{ 2, 59, 89 }, { 1, 38, 60 }, { 1, 17, 27 }
}, { // Band 3
{ 23, 144, 163 }, { 13, 112, 154 }, { 2, 75, 117 },
{ 1, 50, 81 }, { 1, 31, 51 }, { 1, 14, 23 }
}, { // Band 4
{ 18, 162, 185 }, { 6, 123, 171 }, { 1, 78, 125 },
{ 1, 51, 86 }, { 1, 31, 54 }, { 1, 14, 23 }
}, { // Band 5
{ 15, 199, 227 }, { 3, 150, 204 }, { 1, 91, 146 },
{ 1, 55, 95 }, { 1, 30, 53 }, { 1, 11, 20 }
}
}, { // Inter
{ // Band 0
{ 19, 55, 240 }, { 19, 59, 196 }, { 3, 52, 105 }
}, { // Band 1
{ 41, 166, 207 }, { 104, 153, 199 }, { 31, 123, 181 },
{ 14, 101, 152 }, { 5, 72, 106 }, { 1, 36, 52 }
}, { // Band 2
{ 35, 176, 211 }, { 12, 131, 190 }, { 2, 88, 144 },
{ 1, 60, 101 }, { 1, 36, 60 }, { 1, 16, 28 }
}, { // Band 3
{ 28, 183, 213 }, { 8, 134, 191 }, { 1, 86, 142 },
{ 1, 56, 96 }, { 1, 30, 53 }, { 1, 12, 20 }
}, { // Band 4
{ 20, 190, 215 }, { 4, 135, 192 }, { 1, 84, 139 },
{ 1, 53, 91 }, { 1, 28, 49 }, { 1, 11, 20 }
}, { // Band 5
{ 13, 196, 216 }, { 2, 137, 192 }, { 1, 86, 143 },
{ 1, 57, 99 }, { 1, 32, 56 }, { 1, 13, 24 }
}
}
}, { // UV plane
{ // Intra
{ // Band 0
{ 211, 29, 217 }, { 96, 47, 156 }, { 22, 43, 87 }
}, { // Band 1
{ 78, 120, 193 }, { 111, 116, 186 }, { 46, 102, 164 },
{ 15, 80, 128 }, { 2, 49, 76 }, { 1, 18, 28 }
}, { // Band 2
{ 71, 161, 203 }, { 42, 132, 192 }, { 10, 98, 150 },
{ 3, 69, 109 }, { 1, 44, 70 }, { 1, 18, 29 }
}, { // Band 3
{ 57, 186, 211 }, { 30, 140, 196 }, { 4, 93, 146 },
{ 1, 62, 102 }, { 1, 38, 65 }, { 1, 16, 27 }
}, { // Band 4
{ 47, 199, 217 }, { 14, 145, 196 }, { 1, 88, 142 },
{ 1, 57, 98 }, { 1, 36, 62 }, { 1, 15, 26 }
}, { // Band 5
{ 26, 219, 229 }, { 5, 155, 207 }, { 1, 94, 151 },
{ 1, 60, 104 }, { 1, 36, 62 }, { 1, 16, 28 }
}
}, { // Inter
{ // Band 0
{ 233, 29, 248 }, { 146, 47, 220 }, { 43, 52, 140 }
}, { // Band 1
{ 100, 163, 232 }, { 179, 161, 222 }, { 63, 142, 204 },
{ 37, 113, 174 }, { 26, 89, 137 }, { 18, 68, 97 }
}, { // Band 2
{ 85, 181, 230 }, { 32, 146, 209 }, { 7, 100, 164 },
{ 3, 71, 121 }, { 1, 45, 77 }, { 1, 18, 30 }
}, { // Band 3
{ 65, 187, 230 }, { 20, 148, 207 }, { 2, 97, 159 },
{ 1, 68, 116 }, { 1, 40, 70 }, { 1, 14, 29 }
}, { // Band 4
{ 40, 194, 227 }, { 8, 147, 204 }, { 1, 94, 155 },
{ 1, 65, 112 }, { 1, 39, 66 }, { 1, 14, 26 }
}, { // Band 5
{ 16, 208, 228 }, { 3, 151, 207 }, { 1, 98, 160 },
{ 1, 67, 117 }, { 1, 41, 74 }, { 1, 17, 31 }
}
}
}
};
static const vp10_coeff_probs_model default_coef_probs_32x32[PLANE_TYPES] = {
{ // Y plane
{ // Intra
{ // Band 0
{ 17, 38, 140 }, { 7, 34, 80 }, { 1, 17, 29 }
}, { // Band 1
{ 37, 75, 128 }, { 41, 76, 128 }, { 26, 66, 116 },
{ 12, 52, 94 }, { 2, 32, 55 }, { 1, 10, 16 }
}, { // Band 2
{ 50, 127, 154 }, { 37, 109, 152 }, { 16, 82, 121 },
{ 5, 59, 85 }, { 1, 35, 54 }, { 1, 13, 20 }
}, { // Band 3
{ 40, 142, 167 }, { 17, 110, 157 }, { 2, 71, 112 },
{ 1, 44, 72 }, { 1, 27, 45 }, { 1, 11, 17 }
}, { // Band 4
{ 30, 175, 188 }, { 9, 124, 169 }, { 1, 74, 116 },
{ 1, 48, 78 }, { 1, 30, 49 }, { 1, 11, 18 }
}, { // Band 5
{ 10, 222, 223 }, { 2, 150, 194 }, { 1, 83, 128 },
{ 1, 48, 79 }, { 1, 27, 45 }, { 1, 11, 17 }
}
}, { // Inter
{ // Band 0
{ 36, 41, 235 }, { 29, 36, 193 }, { 10, 27, 111 }
}, { // Band 1
{ 85, 165, 222 }, { 177, 162, 215 }, { 110, 135, 195 },
{ 57, 113, 168 }, { 23, 83, 120 }, { 10, 49, 61 }
}, { // Band 2
{ 85, 190, 223 }, { 36, 139, 200 }, { 5, 90, 146 },
{ 1, 60, 103 }, { 1, 38, 65 }, { 1, 18, 30 }
}, { // Band 3
{ 72, 202, 223 }, { 23, 141, 199 }, { 2, 86, 140 },
{ 1, 56, 97 }, { 1, 36, 61 }, { 1, 16, 27 }
}, { // Band 4
{ 55, 218, 225 }, { 13, 145, 200 }, { 1, 86, 141 },
{ 1, 57, 99 }, { 1, 35, 61 }, { 1, 13, 22 }
}, { // Band 5
{ 15, 235, 212 }, { 1, 132, 184 }, { 1, 84, 139 },
{ 1, 57, 97 }, { 1, 34, 56 }, { 1, 14, 23 }
}
}
}, { // UV plane
{ // Intra
{ // Band 0
{ 181, 21, 201 }, { 61, 37, 123 }, { 10, 38, 71 }
}, { // Band 1
{ 47, 106, 172 }, { 95, 104, 173 }, { 42, 93, 159 },
{ 18, 77, 131 }, { 4, 50, 81 }, { 1, 17, 23 }
}, { // Band 2
{ 62, 147, 199 }, { 44, 130, 189 }, { 28, 102, 154 },
{ 18, 75, 115 }, { 2, 44, 65 }, { 1, 12, 19 }
}, { // Band 3
{ 55, 153, 210 }, { 24, 130, 194 }, { 3, 93, 146 },
{ 1, 61, 97 }, { 1, 31, 50 }, { 1, 10, 16 }
}, { // Band 4
{ 49, 186, 223 }, { 17, 148, 204 }, { 1, 96, 142 },
{ 1, 53, 83 }, { 1, 26, 44 }, { 1, 11, 17 }
}, { // Band 5
{ 13, 217, 212 }, { 2, 136, 180 }, { 1, 78, 124 },
{ 1, 50, 83 }, { 1, 29, 49 }, { 1, 14, 23 }
}
}, { // Inter
{ // Band 0
{ 197, 13, 247 }, { 82, 17, 222 }, { 25, 17, 162 }
}, { // Band 1
{ 126, 186, 247 }, { 234, 191, 243 }, { 176, 177, 234 },
{ 104, 158, 220 }, { 66, 128, 186 }, { 55, 90, 137 }
}, { // Band 2
{ 111, 197, 242 }, { 46, 158, 219 }, { 9, 104, 171 },
{ 2, 65, 125 }, { 1, 44, 80 }, { 1, 17, 91 }
}, { // Band 3
{ 104, 208, 245 }, { 39, 168, 224 }, { 3, 109, 162 },
{ 1, 79, 124 }, { 1, 50, 102 }, { 1, 43, 102 }
}, { // Band 4
{ 84, 220, 246 }, { 31, 177, 231 }, { 2, 115, 180 },
{ 1, 79, 134 }, { 1, 55, 77 }, { 1, 60, 79 }
}, { // Band 5
{ 43, 243, 240 }, { 8, 180, 217 }, { 1, 115, 166 },
{ 1, 84, 121 }, { 1, 51, 67 }, { 1, 16, 6 }
}
}
}
};
static void extend_to_full_distribution(vpx_prob *probs, vpx_prob p) {
memcpy(probs, vp10_pareto8_full[p = 0 ? 0 : p - 1],
MODEL_NODES * sizeof(vpx_prob));
}
void vp10_model_to_full_probs(const vpx_prob *model, vpx_prob *full) {
if (full != model)
memcpy(full, model, sizeof(vpx_prob) * UNCONSTRAINED_NODES);
extend_to_full_distribution(&full[UNCONSTRAINED_NODES], model[PIVOT_NODE]);
}
void vp10_default_coef_probs(VP9_COMMON *cm) {
vp10_copy(cm->fc->coef_probs[TX_4X4], default_coef_probs_4x4);
vp10_copy(cm->fc->coef_probs[TX_8X8], default_coef_probs_8x8);
vp10_copy(cm->fc->coef_probs[TX_16X16], default_coef_probs_16x16);
vp10_copy(cm->fc->coef_probs[TX_32X32], default_coef_probs_32x32);
}
#define COEF_COUNT_SAT 24
#define COEF_MAX_UPDATE_FACTOR 112
#define COEF_COUNT_SAT_KEY 24
#define COEF_MAX_UPDATE_FACTOR_KEY 112
#define COEF_COUNT_SAT_AFTER_KEY 24
#define COEF_MAX_UPDATE_FACTOR_AFTER_KEY 128
static void adapt_coef_probs(VP9_COMMON *cm, TX_SIZE tx_size,
unsigned int count_sat,
unsigned int update_factor) {
const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
vp10_coeff_probs_model *const probs = cm->fc->coef_probs[tx_size];
const vp10_coeff_probs_model *const pre_probs = pre_fc->coef_probs[tx_size];
vp10_coeff_count_model *counts = cm->counts.coef[tx_size];
unsigned int (*eob_counts)[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] =
cm->counts.eob_branch[tx_size];
int i, j, k, l, m;
for (i = 0; i < PLANE_TYPES; ++i)
for (j = 0; j < REF_TYPES; ++j)
for (k = 0; k < COEF_BANDS; ++k)
for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l) {
const int n0 = counts[i][j][k][l][ZERO_TOKEN];
const int n1 = counts[i][j][k][l][ONE_TOKEN];
const int n2 = counts[i][j][k][l][TWO_TOKEN];
const int neob = counts[i][j][k][l][EOB_MODEL_TOKEN];
const unsigned int branch_ct[UNCONSTRAINED_NODES][2] = {
{ neob, eob_counts[i][j][k][l] - neob },
{ n0, n1 + n2 },
{ n1, n2 }
};
for (m = 0; m < UNCONSTRAINED_NODES; ++m)
probs[i][j][k][l][m] = merge_probs(pre_probs[i][j][k][l][m],
branch_ct[m],
count_sat, update_factor);
}
}
void vp10_adapt_coef_probs(VP9_COMMON *cm) {
TX_SIZE t;
unsigned int count_sat, update_factor;
if (frame_is_intra_only(cm)) {
update_factor = COEF_MAX_UPDATE_FACTOR_KEY;
count_sat = COEF_COUNT_SAT_KEY;
} else if (cm->last_frame_type == KEY_FRAME) {
update_factor = COEF_MAX_UPDATE_FACTOR_AFTER_KEY; /* adapt quickly */
count_sat = COEF_COUNT_SAT_AFTER_KEY;
} else {
update_factor = COEF_MAX_UPDATE_FACTOR;
count_sat = COEF_COUNT_SAT;
}
for (t = TX_4X4; t <= TX_32X32; t++)
adapt_coef_probs(cm, t, count_sat, update_factor);
}

215
vp10/common/vp9_entropy.h Normal file
View File

@ -0,0 +1,215 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_ENTROPY_H_
#define VP9_COMMON_VP9_ENTROPY_H_
#include "vpx/vpx_integer.h"
#include "vpx_dsp/prob.h"
#include "vp10/common/vp9_common.h"
#include "vp10/common/vp9_enums.h"
#ifdef __cplusplus
extern "C" {
#endif
#define DIFF_UPDATE_PROB 252
// Coefficient token alphabet
#define ZERO_TOKEN 0 // 0 Extra Bits 0+0
#define ONE_TOKEN 1 // 1 Extra Bits 0+1
#define TWO_TOKEN 2 // 2 Extra Bits 0+1
#define THREE_TOKEN 3 // 3 Extra Bits 0+1
#define FOUR_TOKEN 4 // 4 Extra Bits 0+1
#define CATEGORY1_TOKEN 5 // 5-6 Extra Bits 1+1
#define CATEGORY2_TOKEN 6 // 7-10 Extra Bits 2+1
#define CATEGORY3_TOKEN 7 // 11-18 Extra Bits 3+1
#define CATEGORY4_TOKEN 8 // 19-34 Extra Bits 4+1
#define CATEGORY5_TOKEN 9 // 35-66 Extra Bits 5+1
#define CATEGORY6_TOKEN 10 // 67+ Extra Bits 14+1
#define EOB_TOKEN 11 // EOB Extra Bits 0+0
#define ENTROPY_TOKENS 12
#define ENTROPY_NODES 11
DECLARE_ALIGNED(16, extern const uint8_t, vp10_pt_energy_class[ENTROPY_TOKENS]);
#define CAT1_MIN_VAL 5
#define CAT2_MIN_VAL 7
#define CAT3_MIN_VAL 11
#define CAT4_MIN_VAL 19
#define CAT5_MIN_VAL 35
#define CAT6_MIN_VAL 67
// Extra bit probabilities.
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat1_prob[1]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat2_prob[2]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat3_prob[3]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat4_prob[4]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat5_prob[5]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat6_prob[14]);
#if CONFIG_VP9_HIGHBITDEPTH
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat1_prob_high10[1]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat2_prob_high10[2]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat3_prob_high10[3]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat4_prob_high10[4]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat5_prob_high10[5]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat6_prob_high10[16]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat1_prob_high12[1]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat2_prob_high12[2]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat3_prob_high12[3]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat4_prob_high12[4]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat5_prob_high12[5]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_cat6_prob_high12[18]);
#endif // CONFIG_VP9_HIGHBITDEPTH
#define EOB_MODEL_TOKEN 3
typedef struct {
const vpx_tree_index *tree;
const vpx_prob *prob;
int len;
int base_val;
const int16_t *cost;
} vp10_extra_bit;
// indexed by token value
extern const vp10_extra_bit vp10_extra_bits[ENTROPY_TOKENS];
#if CONFIG_VP9_HIGHBITDEPTH
extern const vp10_extra_bit vp10_extra_bits_high10[ENTROPY_TOKENS];
extern const vp10_extra_bit vp10_extra_bits_high12[ENTROPY_TOKENS];
#endif // CONFIG_VP9_HIGHBITDEPTH
#define DCT_MAX_VALUE 16384
#if CONFIG_VP9_HIGHBITDEPTH
#define DCT_MAX_VALUE_HIGH10 65536
#define DCT_MAX_VALUE_HIGH12 262144
#endif // CONFIG_VP9_HIGHBITDEPTH
/* Coefficients are predicted via a 3-dimensional probability table. */
#define REF_TYPES 2 // intra=0, inter=1
/* Middle dimension reflects the coefficient position within the transform. */
#define COEF_BANDS 6
/* Inside dimension is measure of nearby complexity, that reflects the energy
of nearby coefficients are nonzero. For the first coefficient (DC, unless
block type is 0), we look at the (already encoded) blocks above and to the
left of the current block. The context index is then the number (0,1,or 2)
of these blocks having nonzero coefficients.
After decoding a coefficient, the measure is determined by the size of the
most recently decoded coefficient.
Note that the intuitive meaning of this measure changes as coefficients
are decoded, e.g., prior to the first token, a zero means that my neighbors
are empty while, after the first token, because of the use of end-of-block,
a zero means we just decoded a zero and hence guarantees that a non-zero
coefficient will appear later in this block. However, this shift
in meaning is perfectly OK because our context depends also on the
coefficient band (and since zigzag positions 0, 1, and 2 are in
distinct bands). */
#define COEFF_CONTEXTS 6
#define BAND_COEFF_CONTEXTS(band) ((band) == 0 ? 3 : COEFF_CONTEXTS)
// #define ENTROPY_STATS
typedef unsigned int vp10_coeff_count[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS]
[ENTROPY_TOKENS];
typedef unsigned int vp10_coeff_stats[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS]
[ENTROPY_NODES][2];
#define SUBEXP_PARAM 4 /* Subexponential code parameter */
#define MODULUS_PARAM 13 /* Modulus parameter */
struct VP9Common;
void vp10_default_coef_probs(struct VP9Common *cm);
void vp10_adapt_coef_probs(struct VP9Common *cm);
// This is the index in the scan order beyond which all coefficients for
// 8x8 transform and above are in the top band.
// This macro is currently unused but may be used by certain implementations
#define MAXBAND_INDEX 21
DECLARE_ALIGNED(16, extern const uint8_t, vp10_coefband_trans_8x8plus[1024]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_coefband_trans_4x4[16]);
static INLINE const uint8_t *get_band_translate(TX_SIZE tx_size) {
return tx_size == TX_4X4 ? vp10_coefband_trans_4x4
: vp10_coefband_trans_8x8plus;
}
// 128 lists of probabilities are stored for the following ONE node probs:
// 1, 3, 5, 7, ..., 253, 255
// In between probabilities are interpolated linearly
#define COEFF_PROB_MODELS 256
#define UNCONSTRAINED_NODES 3
#define PIVOT_NODE 2 // which node is pivot
#define MODEL_NODES (ENTROPY_NODES - UNCONSTRAINED_NODES)
extern const vpx_tree_index vp10_coef_con_tree[TREE_SIZE(ENTROPY_TOKENS)];
extern const vpx_prob vp10_pareto8_full[COEFF_PROB_MODELS][MODEL_NODES];
typedef vpx_prob vp10_coeff_probs_model[REF_TYPES][COEF_BANDS]
[COEFF_CONTEXTS][UNCONSTRAINED_NODES];
typedef unsigned int vp10_coeff_count_model[REF_TYPES][COEF_BANDS]
[COEFF_CONTEXTS]
[UNCONSTRAINED_NODES + 1];
void vp10_model_to_full_probs(const vpx_prob *model, vpx_prob *full);
typedef char ENTROPY_CONTEXT;
static INLINE int combine_entropy_contexts(ENTROPY_CONTEXT a,
ENTROPY_CONTEXT b) {
return (a != 0) + (b != 0);
}
static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
const ENTROPY_CONTEXT *l) {
ENTROPY_CONTEXT above_ec = 0, left_ec = 0;
switch (tx_size) {
case TX_4X4:
above_ec = a[0] != 0;
left_ec = l[0] != 0;
break;
case TX_8X8:
above_ec = !!*(const uint16_t *)a;
left_ec = !!*(const uint16_t *)l;
break;
case TX_16X16:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
case TX_32X32:
above_ec = !!*(const uint64_t *)a;
left_ec = !!*(const uint64_t *)l;
break;
default:
assert(0 && "Invalid transform size.");
break;
}
return combine_entropy_contexts(above_ec, left_ec);
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_ENTROPY_H_

469
vp10/common/vp9_entropymode.c Normal file
View File

@ -0,0 +1,469 @@
/*
* Copyright (c) 2010 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 "vpx_mem/vpx_mem.h"
#include "vp10/common/vp9_onyxc_int.h"
#include "vp10/common/vp9_seg_common.h"
const vpx_prob vp10_kf_y_mode_prob[INTRA_MODES][INTRA_MODES][INTRA_MODES - 1] = {
{ // above = dc
{ 137, 30, 42, 148, 151, 207, 70, 52, 91 }, // left = dc
{ 92, 45, 102, 136, 116, 180, 74, 90, 100 }, // left = v
{ 73, 32, 19, 187, 222, 215, 46, 34, 100 }, // left = h
{ 91, 30, 32, 116, 121, 186, 93, 86, 94 }, // left = d45
{ 72, 35, 36, 149, 68, 206, 68, 63, 105 }, // left = d135
{ 73, 31, 28, 138, 57, 124, 55, 122, 151 }, // left = d117
{ 67, 23, 21, 140, 126, 197, 40, 37, 171 }, // left = d153
{ 86, 27, 28, 128, 154, 212, 45, 43, 53 }, // left = d207
{ 74, 32, 27, 107, 86, 160, 63, 134, 102 }, // left = d63
{ 59, 67, 44, 140, 161, 202, 78, 67, 119 } // left = tm
}, { // above = v
{ 63, 36, 126, 146, 123, 158, 60, 90, 96 }, // left = dc
{ 43, 46, 168, 134, 107, 128, 69, 142, 92 }, // left = v
{ 44, 29, 68, 159, 201, 177, 50, 57, 77 }, // left = h
{ 58, 38, 76, 114, 97, 172, 78, 133, 92 }, // left = d45
{ 46, 41, 76, 140, 63, 184, 69, 112, 57 }, // left = d135
{ 38, 32, 85, 140, 46, 112, 54, 151, 133 }, // left = d117
{ 39, 27, 61, 131, 110, 175, 44, 75, 136 }, // left = d153
{ 52, 30, 74, 113, 130, 175, 51, 64, 58 }, // left = d207
{ 47, 35, 80, 100, 74, 143, 64, 163, 74 }, // left = d63
{ 36, 61, 116, 114, 128, 162, 80, 125, 82 } // left = tm
}, { // above = h
{ 82, 26, 26, 171, 208, 204, 44, 32, 105 }, // left = dc
{ 55, 44, 68, 166, 179, 192, 57, 57, 108 }, // left = v
{ 42, 26, 11, 199, 241, 228, 23, 15, 85 }, // left = h
{ 68, 42, 19, 131, 160, 199, 55, 52, 83 }, // left = d45
{ 58, 50, 25, 139, 115, 232, 39, 52, 118 }, // left = d135
{ 50, 35, 33, 153, 104, 162, 64, 59, 131 }, // left = d117
{ 44, 24, 16, 150, 177, 202, 33, 19, 156 }, // left = d153
{ 55, 27, 12, 153, 203, 218, 26, 27, 49 }, // left = d207
{ 53, 49, 21, 110, 116, 168, 59, 80, 76 }, // left = d63
{ 38, 72, 19, 168, 203, 212, 50, 50, 107 } // left = tm
}, { // above = d45
{ 103, 26, 36, 129, 132, 201, 83, 80, 93 }, // left = dc
{ 59, 38, 83, 112, 103, 162, 98, 136, 90 }, // left = v
{ 62, 30, 23, 158, 200, 207, 59, 57, 50 }, // left = h
{ 67, 30, 29, 84, 86, 191, 102, 91, 59 }, // left = d45
{ 60, 32, 33, 112, 71, 220, 64, 89, 104 }, // left = d135
{ 53, 26, 34, 130, 56, 149, 84, 120, 103 }, // left = d117
{ 53, 21, 23, 133, 109, 210, 56, 77, 172 }, // left = d153
{ 77, 19, 29, 112, 142, 228, 55, 66, 36 }, // left = d207
{ 61, 29, 29, 93, 97, 165, 83, 175, 162 }, // left = d63
{ 47, 47, 43, 114, 137, 181, 100, 99, 95 } // left = tm
}, { // above = d135
{ 69, 23, 29, 128, 83, 199, 46, 44, 101 }, // left = dc
{ 53, 40, 55, 139, 69, 183, 61, 80, 110 }, // left = v
{ 40, 29, 19, 161, 180, 207, 43, 24, 91 }, // left = h
{ 60, 34, 19, 105, 61, 198, 53, 64, 89 }, // left = d45
{ 52, 31, 22, 158, 40, 209, 58, 62, 89 }, // left = d135
{ 44, 31, 29, 147, 46, 158, 56, 102, 198 }, // left = d117
{ 35, 19, 12, 135, 87, 209, 41, 45, 167 }, // left = d153
{ 55, 25, 21, 118, 95, 215, 38, 39, 66 }, // left = d207
{ 51, 38, 25, 113, 58, 164, 70, 93, 97 }, // left = d63
{ 47, 54, 34, 146, 108, 203, 72, 103, 151 } // left = tm
}, { // above = d117
{ 64, 19, 37, 156, 66, 138, 49, 95, 133 }, // left = dc
{ 46, 27, 80, 150, 55, 124, 55, 121, 135 }, // left = v
{ 36, 23, 27, 165, 149, 166, 54, 64, 118 }, // left = h
{ 53, 21, 36, 131, 63, 163, 60, 109, 81 }, // left = d45
{ 40, 26, 35, 154, 40, 185, 51, 97, 123 }, // left = d135
{ 35, 19, 34, 179, 19, 97, 48, 129, 124 }, // left = d117
{ 36, 20, 26, 136, 62, 164, 33, 77, 154 }, // left = d153
{ 45, 18, 32, 130, 90, 157, 40, 79, 91 }, // left = d207
{ 45, 26, 28, 129, 45, 129, 49, 147, 123 }, // left = d63
{ 38, 44, 51, 136, 74, 162, 57, 97, 121 } // left = tm
}, { // above = d153
{ 75, 17, 22, 136, 138, 185, 32, 34, 166 }, // left = dc
{ 56, 39, 58, 133, 117, 173, 48, 53, 187 }, // left = v
{ 35, 21, 12, 161, 212, 207, 20, 23, 145 }, // left = h
{ 56, 29, 19, 117, 109, 181, 55, 68, 112 }, // left = d45
{ 47, 29, 17, 153, 64, 220, 59, 51, 114 }, // left = d135
{ 46, 16, 24, 136, 76, 147, 41, 64, 172 }, // left = d117
{ 34, 17, 11, 108, 152, 187, 13, 15, 209 }, // left = d153
{ 51, 24, 14, 115, 133, 209, 32, 26, 104 }, // left = d207
{ 55, 30, 18, 122, 79, 179, 44, 88, 116 }, // left = d63
{ 37, 49, 25, 129, 168, 164, 41, 54, 148 } // left = tm
}, { // above = d207
{ 82, 22, 32, 127, 143, 213, 39, 41, 70 }, // left = dc
{ 62, 44, 61, 123, 105, 189, 48, 57, 64 }, // left = v
{ 47, 25, 17, 175, 222, 220, 24, 30, 86 }, // left = h
{ 68, 36, 17, 106, 102, 206, 59, 74, 74 }, // left = d45
{ 57, 39, 23, 151, 68, 216, 55, 63, 58 }, // left = d135
{ 49, 30, 35, 141, 70, 168, 82, 40, 115 }, // left = d117
{ 51, 25, 15, 136, 129, 202, 38, 35, 139 }, // left = d153
{ 68, 26, 16, 111, 141, 215, 29, 28, 28 }, // left = d207
{ 59, 39, 19, 114, 75, 180, 77, 104, 42 }, // left = d63
{ 40, 61, 26, 126, 152, 206, 61, 59, 93 } // left = tm
}, { // above = d63
{ 78, 23, 39, 111, 117, 170, 74, 124, 94 }, // left = dc
{ 48, 34, 86, 101, 92, 146, 78, 179, 134 }, // left = v
{ 47, 22, 24, 138, 187, 178, 68, 69, 59 }, // left = h
{ 56, 25, 33, 105, 112, 187, 95, 177, 129 }, // left = d45
{ 48, 31, 27, 114, 63, 183, 82, 116, 56 }, // left = d135
{ 43, 28, 37, 121, 63, 123, 61, 192, 169 }, // left = d117
{ 42, 17, 24, 109, 97, 177, 56, 76, 122 }, // left = d153
{ 58, 18, 28, 105, 139, 182, 70, 92, 63 }, // left = d207
{ 46, 23, 32, 74, 86, 150, 67, 183, 88 }, // left = d63
{ 36, 38, 48, 92, 122, 165, 88, 137, 91 } // left = tm
}, { // above = tm
{ 65, 70, 60, 155, 159, 199, 61, 60, 81 }, // left = dc
{ 44, 78, 115, 132, 119, 173, 71, 112, 93 }, // left = v
{ 39, 38, 21, 184, 227, 206, 42, 32, 64 }, // left = h
{ 58, 47, 36, 124, 137, 193, 80, 82, 78 }, // left = d45
{ 49, 50, 35, 144, 95, 205, 63, 78, 59 }, // left = d135
{ 41, 53, 52, 148, 71, 142, 65, 128, 51 }, // left = d117
{ 40, 36, 28, 143, 143, 202, 40, 55, 137 }, // left = d153
{ 52, 34, 29, 129, 183, 227, 42, 35, 43 }, // left = d207
{ 42, 44, 44, 104, 105, 164, 64, 130, 80 }, // left = d63
{ 43, 81, 53, 140, 169, 204, 68, 84, 72 } // left = tm
}
};
const vpx_prob vp10_kf_uv_mode_prob[INTRA_MODES][INTRA_MODES - 1] = {
{ 144, 11, 54, 157, 195, 130, 46, 58, 108 }, // y = dc
{ 118, 15, 123, 148, 131, 101, 44, 93, 131 }, // y = v
{ 113, 12, 23, 188, 226, 142, 26, 32, 125 }, // y = h
{ 120, 11, 50, 123, 163, 135, 64, 77, 103 }, // y = d45
{ 113, 9, 36, 155, 111, 157, 32, 44, 161 }, // y = d135
{ 116, 9, 55, 176, 76, 96, 37, 61, 149 }, // y = d117
{ 115, 9, 28, 141, 161, 167, 21, 25, 193 }, // y = d153
{ 120, 12, 32, 145, 195, 142, 32, 38, 86 }, // y = d207
{ 116, 12, 64, 120, 140, 125, 49, 115, 121 }, // y = d63
{ 102, 19, 66, 162, 182, 122, 35, 59, 128 } // y = tm
};
static const vpx_prob default_if_y_probs[BLOCK_SIZE_GROUPS][INTRA_MODES - 1] = {
{ 65, 32, 18, 144, 162, 194, 41, 51, 98 }, // block_size < 8x8
{ 132, 68, 18, 165, 217, 196, 45, 40, 78 }, // block_size < 16x16
{ 173, 80, 19, 176, 240, 193, 64, 35, 46 }, // block_size < 32x32
{ 221, 135, 38, 194, 248, 121, 96, 85, 29 } // block_size >= 32x32
};
static const vpx_prob default_if_uv_probs[INTRA_MODES][INTRA_MODES - 1] = {
{ 120, 7, 76, 176, 208, 126, 28, 54, 103 }, // y = dc
{ 48, 12, 154, 155, 139, 90, 34, 117, 119 }, // y = v
{ 67, 6, 25, 204, 243, 158, 13, 21, 96 }, // y = h
{ 97, 5, 44, 131, 176, 139, 48, 68, 97 }, // y = d45
{ 83, 5, 42, 156, 111, 152, 26, 49, 152 }, // y = d135
{ 80, 5, 58, 178, 74, 83, 33, 62, 145 }, // y = d117
{ 86, 5, 32, 154, 192, 168, 14, 22, 163 }, // y = d153
{ 85, 5, 32, 156, 216, 148, 19, 29, 73 }, // y = d207
{ 77, 7, 64, 116, 132, 122, 37, 126, 120 }, // y = d63
{ 101, 21, 107, 181, 192, 103, 19, 67, 125 } // y = tm
};
const vpx_prob vp10_kf_partition_probs[PARTITION_CONTEXTS]
[PARTITION_TYPES - 1] = {
// 8x8 -> 4x4
{ 158, 97, 94 }, // a/l both not split
{ 93, 24, 99 }, // a split, l not split
{ 85, 119, 44 }, // l split, a not split
{ 62, 59, 67 }, // a/l both split
// 16x16 -> 8x8
{ 149, 53, 53 }, // a/l both not split
{ 94, 20, 48 }, // a split, l not split
{ 83, 53, 24 }, // l split, a not split
{ 52, 18, 18 }, // a/l both split
// 32x32 -> 16x16
{ 150, 40, 39 }, // a/l both not split
{ 78, 12, 26 }, // a split, l not split
{ 67, 33, 11 }, // l split, a not split
{ 24, 7, 5 }, // a/l both split
// 64x64 -> 32x32
{ 174, 35, 49 }, // a/l both not split
{ 68, 11, 27 }, // a split, l not split
{ 57, 15, 9 }, // l split, a not split
{ 12, 3, 3 }, // a/l both split
};
static const vpx_prob default_partition_probs[PARTITION_CONTEXTS]
[PARTITION_TYPES - 1] = {
// 8x8 -> 4x4
{ 199, 122, 141 }, // a/l both not split
{ 147, 63, 159 }, // a split, l not split
{ 148, 133, 118 }, // l split, a not split
{ 121, 104, 114 }, // a/l both split
// 16x16 -> 8x8
{ 174, 73, 87 }, // a/l both not split
{ 92, 41, 83 }, // a split, l not split
{ 82, 99, 50 }, // l split, a not split
{ 53, 39, 39 }, // a/l both split
// 32x32 -> 16x16
{ 177, 58, 59 }, // a/l both not split
{ 68, 26, 63 }, // a split, l not split
{ 52, 79, 25 }, // l split, a not split
{ 17, 14, 12 }, // a/l both split
// 64x64 -> 32x32
{ 222, 34, 30 }, // a/l both not split
{ 72, 16, 44 }, // a split, l not split
{ 58, 32, 12 }, // l split, a not split
{ 10, 7, 6 }, // a/l both split
};
static const vpx_prob default_inter_mode_probs[INTER_MODE_CONTEXTS]
[INTER_MODES - 1] = {
{2, 173, 34}, // 0 = both zero mv
{7, 145, 85}, // 1 = one zero mv + one a predicted mv
{7, 166, 63}, // 2 = two predicted mvs
{7, 94, 66}, // 3 = one predicted/zero and one new mv
{8, 64, 46}, // 4 = two new mvs
{17, 81, 31}, // 5 = one intra neighbour + x
{25, 29, 30}, // 6 = two intra neighbours
};
/* Array indices are identical to previously-existing INTRAMODECONTEXTNODES. */
const vpx_tree_index vp10_intra_mode_tree[TREE_SIZE(INTRA_MODES)] = {
-DC_PRED, 2, /* 0 = DC_NODE */
-TM_PRED, 4, /* 1 = TM_NODE */
-V_PRED, 6, /* 2 = V_NODE */
8, 12, /* 3 = COM_NODE */
-H_PRED, 10, /* 4 = H_NODE */
-D135_PRED, -D117_PRED, /* 5 = D135_NODE */
-D45_PRED, 14, /* 6 = D45_NODE */
-D63_PRED, 16, /* 7 = D63_NODE */
-D153_PRED, -D207_PRED /* 8 = D153_NODE */
};
const vpx_tree_index vp10_inter_mode_tree[TREE_SIZE(INTER_MODES)] = {
-INTER_OFFSET(ZEROMV), 2,
-INTER_OFFSET(NEARESTMV), 4,
-INTER_OFFSET(NEARMV), -INTER_OFFSET(NEWMV)
};
const vpx_tree_index vp10_partition_tree[TREE_SIZE(PARTITION_TYPES)] = {
-PARTITION_NONE, 2,
-PARTITION_HORZ, 4,
-PARTITION_VERT, -PARTITION_SPLIT
};
static const vpx_prob default_intra_inter_p[INTRA_INTER_CONTEXTS] = {
9, 102, 187, 225
};
static const vpx_prob default_comp_inter_p[COMP_INTER_CONTEXTS] = {
239, 183, 119, 96, 41
};
static const vpx_prob default_comp_ref_p[REF_CONTEXTS] = {
50, 126, 123, 221, 226
};
static const vpx_prob default_single_ref_p[REF_CONTEXTS][2] = {
{ 33, 16 },
{ 77, 74 },
{ 142, 142 },
{ 172, 170 },
{ 238, 247 }
};
static const struct tx_probs default_tx_probs = {
{ { 3, 136, 37 },
{ 5, 52, 13 } },
{ { 20, 152 },
{ 15, 101 } },
{ { 100 },
{ 66 } }
};
void vp10_tx_counts_to_branch_counts_32x32(const unsigned int *tx_count_32x32p,
unsigned int (*ct_32x32p)[2]) {
ct_32x32p[0][0] = tx_count_32x32p[TX_4X4];
ct_32x32p[0][1] = tx_count_32x32p[TX_8X8] +
tx_count_32x32p[TX_16X16] +
tx_count_32x32p[TX_32X32];
ct_32x32p[1][0] = tx_count_32x32p[TX_8X8];
ct_32x32p[1][1] = tx_count_32x32p[TX_16X16] +
tx_count_32x32p[TX_32X32];
ct_32x32p[2][0] = tx_count_32x32p[TX_16X16];
ct_32x32p[2][1] = tx_count_32x32p[TX_32X32];
}
void vp10_tx_counts_to_branch_counts_16x16(const unsigned int *tx_count_16x16p,
unsigned int (*ct_16x16p)[2]) {
ct_16x16p[0][0] = tx_count_16x16p[TX_4X4];
ct_16x16p[0][1] = tx_count_16x16p[TX_8X8] + tx_count_16x16p[TX_16X16];
ct_16x16p[1][0] = tx_count_16x16p[TX_8X8];
ct_16x16p[1][1] = tx_count_16x16p[TX_16X16];
}
void vp10_tx_counts_to_branch_counts_8x8(const unsigned int *tx_count_8x8p,
unsigned int (*ct_8x8p)[2]) {
ct_8x8p[0][0] = tx_count_8x8p[TX_4X4];
ct_8x8p[0][1] = tx_count_8x8p[TX_8X8];
}
static const vpx_prob default_skip_probs[SKIP_CONTEXTS] = {
192, 128, 64
};
static const vpx_prob default_switchable_interp_prob[SWITCHABLE_FILTER_CONTEXTS]
[SWITCHABLE_FILTERS - 1] = {
{ 235, 162, },
{ 36, 255, },
{ 34, 3, },
{ 149, 144, },
};
static void init_mode_probs(FRAME_CONTEXT *fc) {
vp10_copy(fc->uv_mode_prob, default_if_uv_probs);
vp10_copy(fc->y_mode_prob, default_if_y_probs);
vp10_copy(fc->switchable_interp_prob, default_switchable_interp_prob);
vp10_copy(fc->partition_prob, default_partition_probs);
vp10_copy(fc->intra_inter_prob, default_intra_inter_p);
vp10_copy(fc->comp_inter_prob, default_comp_inter_p);
vp10_copy(fc->comp_ref_prob, default_comp_ref_p);
vp10_copy(fc->single_ref_prob, default_single_ref_p);
fc->tx_probs = default_tx_probs;
vp10_copy(fc->skip_probs, default_skip_probs);
vp10_copy(fc->inter_mode_probs, default_inter_mode_probs);
}
const vpx_tree_index vp10_switchable_interp_tree
[TREE_SIZE(SWITCHABLE_FILTERS)] = {
-EIGHTTAP, 2,
-EIGHTTAP_SMOOTH, -EIGHTTAP_SHARP
};
void vp10_adapt_mode_probs(VP9_COMMON *cm) {
int i, j;
FRAME_CONTEXT *fc = cm->fc;
const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
const FRAME_COUNTS *counts = &cm->counts;
for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
fc->intra_inter_prob[i] = mode_mv_merge_probs(pre_fc->intra_inter_prob[i],
counts->intra_inter[i]);
for (i = 0; i < COMP_INTER_CONTEXTS; i++)
fc->comp_inter_prob[i] = mode_mv_merge_probs(pre_fc->comp_inter_prob[i],
counts->comp_inter[i]);
for (i = 0; i < REF_CONTEXTS; i++)
fc->comp_ref_prob[i] = mode_mv_merge_probs(pre_fc->comp_ref_prob[i],
counts->comp_ref[i]);
for (i = 0; i < REF_CONTEXTS; i++)
for (j = 0; j < 2; j++)
fc->single_ref_prob[i][j] = mode_mv_merge_probs(
pre_fc->single_ref_prob[i][j], counts->single_ref[i][j]);
for (i = 0; i < INTER_MODE_CONTEXTS; i++)
vpx_tree_merge_probs(vp10_inter_mode_tree, pre_fc->inter_mode_probs[i],
counts->inter_mode[i], fc->inter_mode_probs[i]);
for (i = 0; i < BLOCK_SIZE_GROUPS; i++)
vpx_tree_merge_probs(vp10_intra_mode_tree, pre_fc->y_mode_prob[i],
counts->y_mode[i], fc->y_mode_prob[i]);
for (i = 0; i < INTRA_MODES; ++i)
vpx_tree_merge_probs(vp10_intra_mode_tree, pre_fc->uv_mode_prob[i],
counts->uv_mode[i], fc->uv_mode_prob[i]);
for (i = 0; i < PARTITION_CONTEXTS; i++)
vpx_tree_merge_probs(vp10_partition_tree, pre_fc->partition_prob[i],
counts->partition[i], fc->partition_prob[i]);
if (cm->interp_filter == SWITCHABLE) {
for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
vpx_tree_merge_probs(vp10_switchable_interp_tree,
pre_fc->switchable_interp_prob[i],
counts->switchable_interp[i],
fc->switchable_interp_prob[i]);
}
if (cm->tx_mode == TX_MODE_SELECT) {
int j;
unsigned int branch_ct_8x8p[TX_SIZES - 3][2];
unsigned int branch_ct_16x16p[TX_SIZES - 2][2];
unsigned int branch_ct_32x32p[TX_SIZES - 1][2];
for (i = 0; i < TX_SIZE_CONTEXTS; ++i) {
vp10_tx_counts_to_branch_counts_8x8(counts->tx.p8x8[i], branch_ct_8x8p);
for (j = 0; j < TX_SIZES - 3; ++j)
fc->tx_probs.p8x8[i][j] = mode_mv_merge_probs(
pre_fc->tx_probs.p8x8[i][j], branch_ct_8x8p[j]);
vp10_tx_counts_to_branch_counts_16x16(counts->tx.p16x16[i], branch_ct_16x16p);
for (j = 0; j < TX_SIZES - 2; ++j)
fc->tx_probs.p16x16[i][j] = mode_mv_merge_probs(
pre_fc->tx_probs.p16x16[i][j], branch_ct_16x16p[j]);
vp10_tx_counts_to_branch_counts_32x32(counts->tx.p32x32[i], branch_ct_32x32p);
for (j = 0; j < TX_SIZES - 1; ++j)
fc->tx_probs.p32x32[i][j] = mode_mv_merge_probs(
pre_fc->tx_probs.p32x32[i][j], branch_ct_32x32p[j]);
}
}
for (i = 0; i < SKIP_CONTEXTS; ++i)
fc->skip_probs[i] = mode_mv_merge_probs(
pre_fc->skip_probs[i], counts->skip[i]);
}
static void set_default_lf_deltas(struct loopfilter *lf) {
lf->mode_ref_delta_enabled = 1;
lf->mode_ref_delta_update = 1;
lf->ref_deltas[INTRA_FRAME] = 1;
lf->ref_deltas[LAST_FRAME] = 0;
lf->ref_deltas[GOLDEN_FRAME] = -1;
lf->ref_deltas[ALTREF_FRAME] = -1;
lf->mode_deltas[0] = 0;
lf->mode_deltas[1] = 0;
}
void vp10_setup_past_independence(VP9_COMMON *cm) {
// Reset the segment feature data to the default stats:
// Features disabled, 0, with delta coding (Default state).
struct loopfilter *const lf = &cm->lf;
int i;
vp10_clearall_segfeatures(&cm->seg);
cm->seg.abs_delta = SEGMENT_DELTADATA;
if (cm->last_frame_seg_map && !cm->frame_parallel_decode)
memset(cm->last_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols));
if (cm->current_frame_seg_map)
memset(cm->current_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols));
// Reset the mode ref deltas for loop filter
vp10_zero(lf->last_ref_deltas);
vp10_zero(lf->last_mode_deltas);
set_default_lf_deltas(lf);
// To force update of the sharpness
lf->last_sharpness_level = -1;
vp10_default_coef_probs(cm);
init_mode_probs(cm->fc);
vp10_init_mv_probs(cm);
cm->fc->initialized = 1;
if (cm->frame_type == KEY_FRAME ||
cm->error_resilient_mode || cm->reset_frame_context == 3) {
// Reset all frame contexts.
for (i = 0; i < FRAME_CONTEXTS; ++i)
cm->frame_contexts[i] = *cm->fc;
} else if (cm->reset_frame_context == 2) {
// Reset only the frame context specified in the frame header.
cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
}
// prev_mip will only be allocated in encoder.
if (frame_is_intra_only(cm) && cm->prev_mip && !cm->frame_parallel_decode)
memset(cm->prev_mip, 0,
cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->prev_mip));
vp10_zero(cm->ref_frame_sign_bias);
cm->frame_context_idx = 0;
}

107
vp10/common/vp9_entropymode.h Normal file
View File

@ -0,0 +1,107 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_ENTROPYMODE_H_
#define VP9_COMMON_VP9_ENTROPYMODE_H_
#include "vp10/common/vp9_entropy.h"
#include "vp10/common/vp9_entropymv.h"
#include "vp10/common/vp9_filter.h"
#include "vpx_dsp/vpx_filter.h"
#ifdef __cplusplus
extern "C" {
#endif
#define BLOCK_SIZE_GROUPS 4
#define TX_SIZE_CONTEXTS 2
#define INTER_OFFSET(mode) ((mode) - NEARESTMV)
struct VP9Common;
struct tx_probs {
vpx_prob p32x32[TX_SIZE_CONTEXTS][TX_SIZES - 1];
vpx_prob p16x16[TX_SIZE_CONTEXTS][TX_SIZES - 2];
vpx_prob p8x8[TX_SIZE_CONTEXTS][TX_SIZES - 3];
};
struct tx_counts {
unsigned int p32x32[TX_SIZE_CONTEXTS][TX_SIZES];
unsigned int p16x16[TX_SIZE_CONTEXTS][TX_SIZES - 1];
unsigned int p8x8[TX_SIZE_CONTEXTS][TX_SIZES - 2];
unsigned int tx_totals[TX_SIZES];
};
typedef struct frame_contexts {
vpx_prob y_mode_prob[BLOCK_SIZE_GROUPS][INTRA_MODES - 1];
vpx_prob uv_mode_prob[INTRA_MODES][INTRA_MODES - 1];
vpx_prob partition_prob[PARTITION_CONTEXTS][PARTITION_TYPES - 1];
vp10_coeff_probs_model coef_probs[TX_SIZES][PLANE_TYPES];
vpx_prob switchable_interp_prob[SWITCHABLE_FILTER_CONTEXTS]
[SWITCHABLE_FILTERS - 1];
vpx_prob inter_mode_probs[INTER_MODE_CONTEXTS][INTER_MODES - 1];
vpx_prob intra_inter_prob[INTRA_INTER_CONTEXTS];
vpx_prob comp_inter_prob[COMP_INTER_CONTEXTS];
vpx_prob single_ref_prob[REF_CONTEXTS][2];
vpx_prob comp_ref_prob[REF_CONTEXTS];
struct tx_probs tx_probs;
vpx_prob skip_probs[SKIP_CONTEXTS];
nmv_context nmvc;
int initialized;
} FRAME_CONTEXT;
typedef struct FRAME_COUNTS {
unsigned int y_mode[BLOCK_SIZE_GROUPS][INTRA_MODES];
unsigned int uv_mode[INTRA_MODES][INTRA_MODES];
unsigned int partition[PARTITION_CONTEXTS][PARTITION_TYPES];
vp10_coeff_count_model coef[TX_SIZES][PLANE_TYPES];
unsigned int eob_branch[TX_SIZES][PLANE_TYPES][REF_TYPES]
[COEF_BANDS][COEFF_CONTEXTS];
unsigned int switchable_interp[SWITCHABLE_FILTER_CONTEXTS]
[SWITCHABLE_FILTERS];
unsigned int inter_mode[INTER_MODE_CONTEXTS][INTER_MODES];
unsigned int intra_inter[INTRA_INTER_CONTEXTS][2];
unsigned int comp_inter[COMP_INTER_CONTEXTS][2];
unsigned int single_ref[REF_CONTEXTS][2][2];
unsigned int comp_ref[REF_CONTEXTS][2];
struct tx_counts tx;
unsigned int skip[SKIP_CONTEXTS][2];
nmv_context_counts mv;
} FRAME_COUNTS;
extern const vpx_prob vp10_kf_uv_mode_prob[INTRA_MODES][INTRA_MODES - 1];
extern const vpx_prob vp10_kf_y_mode_prob[INTRA_MODES][INTRA_MODES]
[INTRA_MODES - 1];
extern const vpx_prob vp10_kf_partition_probs[PARTITION_CONTEXTS]
[PARTITION_TYPES - 1];
extern const vpx_tree_index vp10_intra_mode_tree[TREE_SIZE(INTRA_MODES)];
extern const vpx_tree_index vp10_inter_mode_tree[TREE_SIZE(INTER_MODES)];
extern const vpx_tree_index vp10_partition_tree[TREE_SIZE(PARTITION_TYPES)];
extern const vpx_tree_index vp10_switchable_interp_tree
[TREE_SIZE(SWITCHABLE_FILTERS)];
void vp10_setup_past_independence(struct VP9Common *cm);
void vp10_adapt_mode_probs(struct VP9Common *cm);
void vp10_tx_counts_to_branch_counts_32x32(const unsigned int *tx_count_32x32p,
unsigned int (*ct_32x32p)[2]);
void vp10_tx_counts_to_branch_counts_16x16(const unsigned int *tx_count_16x16p,
unsigned int (*ct_16x16p)[2]);
void vp10_tx_counts_to_branch_counts_8x8(const unsigned int *tx_count_8x8p,
unsigned int (*ct_8x8p)[2]);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_ENTROPYMODE_H_

218
vp10/common/vp9_entropymv.c Normal file
View File

@ -0,0 +1,218 @@
/*
* Copyright (c) 2010 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/common/vp9_onyxc_int.h"
#include "vp10/common/vp9_entropymv.h"
// Integer pel reference mv threshold for use of high-precision 1/8 mv
#define COMPANDED_MVREF_THRESH 8
const vpx_tree_index vp10_mv_joint_tree[TREE_SIZE(MV_JOINTS)] = {
-MV_JOINT_ZERO, 2,
-MV_JOINT_HNZVZ, 4,
-MV_JOINT_HZVNZ, -MV_JOINT_HNZVNZ
};
const vpx_tree_index vp10_mv_class_tree[TREE_SIZE(MV_CLASSES)] = {
-MV_CLASS_0, 2,
-MV_CLASS_1, 4,
6, 8,
-MV_CLASS_2, -MV_CLASS_3,
10, 12,
-MV_CLASS_4, -MV_CLASS_5,
-MV_CLASS_6, 14,
16, 18,
-MV_CLASS_7, -MV_CLASS_8,
-MV_CLASS_9, -MV_CLASS_10,
};
const vpx_tree_index vp10_mv_class0_tree[TREE_SIZE(CLASS0_SIZE)] = {
-0, -1,
};
const vpx_tree_index vp10_mv_fp_tree[TREE_SIZE(MV_FP_SIZE)] = {
-0, 2,
-1, 4,
-2, -3
};
static const nmv_context default_nmv_context = {
{32, 64, 96},
{
{ // Vertical component
128, // sign
{224, 144, 192, 168, 192, 176, 192, 198, 198, 245}, // class
{216}, // class0
{136, 140, 148, 160, 176, 192, 224, 234, 234, 240}, // bits
{{128, 128, 64}, {96, 112, 64}}, // class0_fp
{64, 96, 64}, // fp
160, // class0_hp bit
128, // hp
},
{ // Horizontal component
128, // sign
{216, 128, 176, 160, 176, 176, 192, 198, 198, 208}, // class
{208}, // class0
{136, 140, 148, 160, 176, 192, 224, 234, 234, 240}, // bits
{{128, 128, 64}, {96, 112, 64}}, // class0_fp
{64, 96, 64}, // fp
160, // class0_hp bit
128, // hp
}
},
};
static const uint8_t log_in_base_2[] = {
0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,
4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 10
};
static INLINE int mv_class_base(MV_CLASS_TYPE c) {
return c ? CLASS0_SIZE << (c + 2) : 0;
}
MV_CLASS_TYPE vp10_get_mv_class(int z, int *offset) {
const MV_CLASS_TYPE c = (z >= CLASS0_SIZE * 4096) ?
MV_CLASS_10 : (MV_CLASS_TYPE)log_in_base_2[z >> 3];
if (offset)
*offset = z - mv_class_base(c);
return c;
}
int vp10_use_mv_hp(const MV *ref) {
return (abs(ref->row) >> 3) < COMPANDED_MVREF_THRESH &&
(abs(ref->col) >> 3) < COMPANDED_MVREF_THRESH;
}
static void inc_mv_component(int v, nmv_component_counts *comp_counts,
int incr, int usehp) {
int s, z, c, o, d, e, f;
assert(v != 0); /* should not be zero */
s = v < 0;
comp_counts->sign[s] += incr;
z = (s ? -v : v) - 1; /* magnitude - 1 */
c = vp10_get_mv_class(z, &o);
comp_counts->classes[c] += incr;
d = (o >> 3); /* int mv data */
f = (o >> 1) & 3; /* fractional pel mv data */
e = (o & 1); /* high precision mv data */
if (c == MV_CLASS_0) {
comp_counts->class0[d] += incr;
comp_counts->class0_fp[d][f] += incr;
comp_counts->class0_hp[e] += usehp * incr;
} else {
int i;
int b = c + CLASS0_BITS - 1; // number of bits
for (i = 0; i < b; ++i)
comp_counts->bits[i][((d >> i) & 1)] += incr;
comp_counts->fp[f] += incr;
comp_counts->hp[e] += usehp * incr;
}
}
void vp10_inc_mv(const MV *mv, nmv_context_counts *counts) {
if (counts != NULL) {
const MV_JOINT_TYPE j = vp10_get_mv_joint(mv);
++counts->joints[j];
if (mv_joint_vertical(j)) {
inc_mv_component(mv->row, &counts->comps[0], 1, 1);
}
if (mv_joint_horizontal(j)) {
inc_mv_component(mv->col, &counts->comps[1], 1, 1);
}
}
}
void vp10_adapt_mv_probs(VP9_COMMON *cm, int allow_hp) {
int i, j;
nmv_context *fc = &cm->fc->nmvc;
const nmv_context *pre_fc = &cm->frame_contexts[cm->frame_context_idx].nmvc;
const nmv_context_counts *counts = &cm->counts.mv;
vpx_tree_merge_probs(vp10_mv_joint_tree, pre_fc->joints, counts->joints,
fc->joints);
for (i = 0; i < 2; ++i) {
nmv_component *comp = &fc->comps[i];
const nmv_component *pre_comp = &pre_fc->comps[i];
const nmv_component_counts *c = &counts->comps[i];
comp->sign = mode_mv_merge_probs(pre_comp->sign, c->sign);
vpx_tree_merge_probs(vp10_mv_class_tree, pre_comp->classes, c->classes,
comp->classes);
vpx_tree_merge_probs(vp10_mv_class0_tree, pre_comp->class0, c->class0,
comp->class0);
for (j = 0; j < MV_OFFSET_BITS; ++j)
comp->bits[j] = mode_mv_merge_probs(pre_comp->bits[j], c->bits[j]);
for (j = 0; j < CLASS0_SIZE; ++j)
vpx_tree_merge_probs(vp10_mv_fp_tree, pre_comp->class0_fp[j],
c->class0_fp[j], comp->class0_fp[j]);
vpx_tree_merge_probs(vp10_mv_fp_tree, pre_comp->fp, c->fp, comp->fp);
if (allow_hp) {
comp->class0_hp = mode_mv_merge_probs(pre_comp->class0_hp, c->class0_hp);
comp->hp = mode_mv_merge_probs(pre_comp->hp, c->hp);
}
}
}
void vp10_init_mv_probs(VP9_COMMON *cm) {
cm->fc->nmvc = default_nmv_context;
}

133
vp10/common/vp9_entropymv.h Normal file
View File

@ -0,0 +1,133 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_ENTROPYMV_H_
#define VP9_COMMON_VP9_ENTROPYMV_H_
#include "./vpx_config.h"
#include "vpx_dsp/prob.h"
#include "vp10/common/vp9_mv.h"
#ifdef __cplusplus
extern "C" {
#endif
struct VP9Common;
void vp10_init_mv_probs(struct VP9Common *cm);
void vp10_adapt_mv_probs(struct VP9Common *cm, int usehp);
int vp10_use_mv_hp(const MV *ref);
#define MV_UPDATE_PROB 252
/* Symbols for coding which components are zero jointly */
#define MV_JOINTS 4
typedef enum {
MV_JOINT_ZERO = 0, /* Zero vector */
MV_JOINT_HNZVZ = 1, /* Vert zero, hor nonzero */
MV_JOINT_HZVNZ = 2, /* Hor zero, vert nonzero */
MV_JOINT_HNZVNZ = 3, /* Both components nonzero */
} MV_JOINT_TYPE;
static INLINE int mv_joint_vertical(MV_JOINT_TYPE type) {
return type == MV_JOINT_HZVNZ || type == MV_JOINT_HNZVNZ;
}
static INLINE int mv_joint_horizontal(MV_JOINT_TYPE type) {
return type == MV_JOINT_HNZVZ || type == MV_JOINT_HNZVNZ;
}
/* Symbols for coding magnitude class of nonzero components */
#define MV_CLASSES 11
typedef enum {
MV_CLASS_0 = 0, /* (0, 2] integer pel */
MV_CLASS_1 = 1, /* (2, 4] integer pel */
MV_CLASS_2 = 2, /* (4, 8] integer pel */
MV_CLASS_3 = 3, /* (8, 16] integer pel */
MV_CLASS_4 = 4, /* (16, 32] integer pel */
MV_CLASS_5 = 5, /* (32, 64] integer pel */
MV_CLASS_6 = 6, /* (64, 128] integer pel */
MV_CLASS_7 = 7, /* (128, 256] integer pel */
MV_CLASS_8 = 8, /* (256, 512] integer pel */
MV_CLASS_9 = 9, /* (512, 1024] integer pel */
MV_CLASS_10 = 10, /* (1024,2048] integer pel */
} MV_CLASS_TYPE;
#define CLASS0_BITS 1 /* bits at integer precision for class 0 */
#define CLASS0_SIZE (1 << CLASS0_BITS)
#define MV_OFFSET_BITS (MV_CLASSES + CLASS0_BITS - 2)
#define MV_FP_SIZE 4
#define MV_MAX_BITS (MV_CLASSES + CLASS0_BITS + 2)
#define MV_MAX ((1 << MV_MAX_BITS) - 1)
#define MV_VALS ((MV_MAX << 1) + 1)
#define MV_IN_USE_BITS 14
#define MV_UPP ((1 << MV_IN_USE_BITS) - 1)
#define MV_LOW (-(1 << MV_IN_USE_BITS))
extern const vpx_tree_index vp10_mv_joint_tree[];
extern const vpx_tree_index vp10_mv_class_tree[];
extern const vpx_tree_index vp10_mv_class0_tree[];
extern const vpx_tree_index vp10_mv_fp_tree[];
typedef struct {
vpx_prob sign;
vpx_prob classes[MV_CLASSES - 1];
vpx_prob class0[CLASS0_SIZE - 1];
vpx_prob bits[MV_OFFSET_BITS];
vpx_prob class0_fp[CLASS0_SIZE][MV_FP_SIZE - 1];
vpx_prob fp[MV_FP_SIZE - 1];
vpx_prob class0_hp;
vpx_prob hp;
} nmv_component;
typedef struct {
vpx_prob joints[MV_JOINTS - 1];
nmv_component comps[2];
} nmv_context;
static INLINE MV_JOINT_TYPE vp10_get_mv_joint(const MV *mv) {
if (mv->row == 0) {
return mv->col == 0 ? MV_JOINT_ZERO : MV_JOINT_HNZVZ;
} else {
return mv->col == 0 ? MV_JOINT_HZVNZ : MV_JOINT_HNZVNZ;
}
}
MV_CLASS_TYPE vp10_get_mv_class(int z, int *offset);
typedef struct {
unsigned int sign[2];
unsigned int classes[MV_CLASSES];
unsigned int class0[CLASS0_SIZE];
unsigned int bits[MV_OFFSET_BITS][2];
unsigned int class0_fp[CLASS0_SIZE][MV_FP_SIZE];
unsigned int fp[MV_FP_SIZE];
unsigned int class0_hp[2];
unsigned int hp[2];
} nmv_component_counts;
typedef struct {
unsigned int joints[MV_JOINTS];
nmv_component_counts comps[2];
} nmv_context_counts;
void vp10_inc_mv(const MV *mv, nmv_context_counts *mvctx);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_ENTROPYMV_H_

147
vp10/common/vp9_enums.h Normal file
View File

@ -0,0 +1,147 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_ENUMS_H_
#define VP9_COMMON_VP9_ENUMS_H_
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#ifdef __cplusplus
extern "C" {
#endif
#define MI_SIZE_LOG2 3
#define MI_BLOCK_SIZE_LOG2 (6 - MI_SIZE_LOG2) // 64 = 2^6
#define MI_SIZE (1 << MI_SIZE_LOG2) // pixels per mi-unit
#define MI_BLOCK_SIZE (1 << MI_BLOCK_SIZE_LOG2) // mi-units per max block
#define MI_MASK (MI_BLOCK_SIZE - 1)
// Bitstream profiles indicated by 2-3 bits in the uncompressed header.
// 00: Profile 0. 8-bit 4:2:0 only.
// 10: Profile 1. 8-bit 4:4:4, 4:2:2, and 4:4:0.
// 01: Profile 2. 10-bit and 12-bit color only, with 4:2:0 sampling.
// 110: Profile 3. 10-bit and 12-bit color only, with 4:2:2/4:4:4/4:4:0
// sampling.
// 111: Undefined profile.
typedef enum BITSTREAM_PROFILE {
PROFILE_0,
PROFILE_1,
PROFILE_2,
PROFILE_3,
MAX_PROFILES
} BITSTREAM_PROFILE;
#define BLOCK_4X4 0
#define BLOCK_4X8 1
#define BLOCK_8X4 2
#define BLOCK_8X8 3
#define BLOCK_8X16 4
#define BLOCK_16X8 5
#define BLOCK_16X16 6
#define BLOCK_16X32 7
#define BLOCK_32X16 8
#define BLOCK_32X32 9
#define BLOCK_32X64 10
#define BLOCK_64X32 11
#define BLOCK_64X64 12
#define BLOCK_SIZES 13
#define BLOCK_INVALID BLOCK_SIZES
typedef uint8_t BLOCK_SIZE;
typedef enum PARTITION_TYPE {
PARTITION_NONE,
PARTITION_HORZ,
PARTITION_VERT,
PARTITION_SPLIT,
PARTITION_TYPES,
PARTITION_INVALID = PARTITION_TYPES
} PARTITION_TYPE;
typedef char PARTITION_CONTEXT;
#define PARTITION_PLOFFSET 4 // number of probability models per block size
#define PARTITION_CONTEXTS (4 * PARTITION_PLOFFSET)
// block transform size
typedef uint8_t TX_SIZE;
#define TX_4X4 ((TX_SIZE)0) // 4x4 transform
#define TX_8X8 ((TX_SIZE)1) // 8x8 transform
#define TX_16X16 ((TX_SIZE)2) // 16x16 transform
#define TX_32X32 ((TX_SIZE)3) // 32x32 transform
#define TX_SIZES ((TX_SIZE)4)
// frame transform mode
typedef enum {
ONLY_4X4 = 0, // only 4x4 transform used
ALLOW_8X8 = 1, // allow block transform size up to 8x8
ALLOW_16X16 = 2, // allow block transform size up to 16x16
ALLOW_32X32 = 3, // allow block transform size up to 32x32
TX_MODE_SELECT = 4, // transform specified for each block
TX_MODES = 5,
} TX_MODE;
typedef enum {
DCT_DCT = 0, // DCT in both horizontal and vertical
ADST_DCT = 1, // ADST in vertical, DCT in horizontal
DCT_ADST = 2, // DCT in vertical, ADST in horizontal
ADST_ADST = 3, // ADST in both directions
TX_TYPES = 4
} TX_TYPE;
typedef enum {
VP9_LAST_FLAG = 1 << 0,
VP9_GOLD_FLAG = 1 << 1,
VP9_ALT_FLAG = 1 << 2,
} VP9_REFFRAME;
typedef enum {
PLANE_TYPE_Y = 0,
PLANE_TYPE_UV = 1,
PLANE_TYPES
} PLANE_TYPE;
#define DC_PRED 0 // Average of above and left pixels
#define V_PRED 1 // Vertical
#define H_PRED 2 // Horizontal
#define D45_PRED 3 // Directional 45 deg = round(arctan(1/1) * 180/pi)
#define D135_PRED 4 // Directional 135 deg = 180 - 45
#define D117_PRED 5 // Directional 117 deg = 180 - 63
#define D153_PRED 6 // Directional 153 deg = 180 - 27
#define D207_PRED 7 // Directional 207 deg = 180 + 27
#define D63_PRED 8 // Directional 63 deg = round(arctan(2/1) * 180/pi)
#define TM_PRED 9 // True-motion
#define NEARESTMV 10
#define NEARMV 11
#define ZEROMV 12
#define NEWMV 13
#define MB_MODE_COUNT 14
typedef uint8_t PREDICTION_MODE;
#define INTRA_MODES (TM_PRED + 1)
#define INTER_MODES (1 + NEWMV - NEARESTMV)
#define SKIP_CONTEXTS 3
#define INTER_MODE_CONTEXTS 7
/* Segment Feature Masks */
#define MAX_MV_REF_CANDIDATES 2
#define INTRA_INTER_CONTEXTS 4
#define COMP_INTER_CONTEXTS 5
#define REF_CONTEXTS 5
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_ENUMS_H_

104
vp10/common/vp9_filter.c Normal file
View File

@ -0,0 +1,104 @@
/*
* Copyright (c) 2010 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 <assert.h>
#include "vp10/common/vp9_filter.h"
DECLARE_ALIGNED(256, static const InterpKernel,
bilinear_filters[SUBPEL_SHIFTS]) = {
{ 0, 0, 0, 128, 0, 0, 0, 0 },
{ 0, 0, 0, 120, 8, 0, 0, 0 },
{ 0, 0, 0, 112, 16, 0, 0, 0 },
{ 0, 0, 0, 104, 24, 0, 0, 0 },
{ 0, 0, 0, 96, 32, 0, 0, 0 },
{ 0, 0, 0, 88, 40, 0, 0, 0 },
{ 0, 0, 0, 80, 48, 0, 0, 0 },
{ 0, 0, 0, 72, 56, 0, 0, 0 },
{ 0, 0, 0, 64, 64, 0, 0, 0 },
{ 0, 0, 0, 56, 72, 0, 0, 0 },
{ 0, 0, 0, 48, 80, 0, 0, 0 },
{ 0, 0, 0, 40, 88, 0, 0, 0 },
{ 0, 0, 0, 32, 96, 0, 0, 0 },
{ 0, 0, 0, 24, 104, 0, 0, 0 },
{ 0, 0, 0, 16, 112, 0, 0, 0 },
{ 0, 0, 0, 8, 120, 0, 0, 0 }
};
// Lagrangian interpolation filter
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8[SUBPEL_SHIFTS]) = {
{ 0, 0, 0, 128, 0, 0, 0, 0},
{ 0, 1, -5, 126, 8, -3, 1, 0},
{ -1, 3, -10, 122, 18, -6, 2, 0},
{ -1, 4, -13, 118, 27, -9, 3, -1},
{ -1, 4, -16, 112, 37, -11, 4, -1},
{ -1, 5, -18, 105, 48, -14, 4, -1},
{ -1, 5, -19, 97, 58, -16, 5, -1},
{ -1, 6, -19, 88, 68, -18, 5, -1},
{ -1, 6, -19, 78, 78, -19, 6, -1},
{ -1, 5, -18, 68, 88, -19, 6, -1},
{ -1, 5, -16, 58, 97, -19, 5, -1},
{ -1, 4, -14, 48, 105, -18, 5, -1},
{ -1, 4, -11, 37, 112, -16, 4, -1},
{ -1, 3, -9, 27, 118, -13, 4, -1},
{ 0, 2, -6, 18, 122, -10, 3, -1},
{ 0, 1, -3, 8, 126, -5, 1, 0}
};
// DCT based filter
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8s[SUBPEL_SHIFTS]) = {
{0, 0, 0, 128, 0, 0, 0, 0},
{-1, 3, -7, 127, 8, -3, 1, 0},
{-2, 5, -13, 125, 17, -6, 3, -1},
{-3, 7, -17, 121, 27, -10, 5, -2},
{-4, 9, -20, 115, 37, -13, 6, -2},
{-4, 10, -23, 108, 48, -16, 8, -3},
{-4, 10, -24, 100, 59, -19, 9, -3},
{-4, 11, -24, 90, 70, -21, 10, -4},
{-4, 11, -23, 80, 80, -23, 11, -4},
{-4, 10, -21, 70, 90, -24, 11, -4},
{-3, 9, -19, 59, 100, -24, 10, -4},
{-3, 8, -16, 48, 108, -23, 10, -4},
{-2, 6, -13, 37, 115, -20, 9, -4},
{-2, 5, -10, 27, 121, -17, 7, -3},
{-1, 3, -6, 17, 125, -13, 5, -2},
{0, 1, -3, 8, 127, -7, 3, -1}
};
// freqmultiplier = 0.5
DECLARE_ALIGNED(256, static const InterpKernel,
sub_pel_filters_8lp[SUBPEL_SHIFTS]) = {
{ 0, 0, 0, 128, 0, 0, 0, 0},
{-3, -1, 32, 64, 38, 1, -3, 0},
{-2, -2, 29, 63, 41, 2, -3, 0},
{-2, -2, 26, 63, 43, 4, -4, 0},
{-2, -3, 24, 62, 46, 5, -4, 0},
{-2, -3, 21, 60, 49, 7, -4, 0},
{-1, -4, 18, 59, 51, 9, -4, 0},
{-1, -4, 16, 57, 53, 12, -4, -1},
{-1, -4, 14, 55, 55, 14, -4, -1},
{-1, -4, 12, 53, 57, 16, -4, -1},
{ 0, -4, 9, 51, 59, 18, -4, -1},
{ 0, -4, 7, 49, 60, 21, -3, -2},
{ 0, -4, 5, 46, 62, 24, -3, -2},
{ 0, -4, 4, 43, 63, 26, -2, -2},
{ 0, -3, 2, 41, 63, 29, -2, -2},
{ 0, -3, 1, 38, 64, 32, -1, -3}
};
const InterpKernel *vp10_filter_kernels[4] = {
sub_pel_filters_8,
sub_pel_filters_8lp,
sub_pel_filters_8s,
bilinear_filters
};

42
vp10/common/vp9_filter.h Normal file
View File

@ -0,0 +1,42 @@
/*
* Copyright (c) 2011 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 VP9_COMMON_VP9_FILTER_H_
#define VP9_COMMON_VP9_FILTER_H_
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/vpx_filter.h"
#include "vpx_ports/mem.h"
#ifdef __cplusplus
extern "C" {
#endif
#define EIGHTTAP 0
#define EIGHTTAP_SMOOTH 1
#define EIGHTTAP_SHARP 2
#define SWITCHABLE_FILTERS 3 /* Number of switchable filters */
#define BILINEAR 3
// The codec can operate in four possible inter prediction filter mode:
// 8-tap, 8-tap-smooth, 8-tap-sharp, and switching between the three.
#define SWITCHABLE_FILTER_CONTEXTS (SWITCHABLE_FILTERS + 1)
#define SWITCHABLE 4 /* should be the last one */
typedef uint8_t INTERP_FILTER;
extern const InterpKernel *vp10_filter_kernels[4];
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_FILTER_H_

86
vp10/common/vp9_frame_buffers.c Normal file
View File

@ -0,0 +1,86 @@
/*
* Copyright (c) 2014 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 <assert.h>
#include "vp10/common/vp9_frame_buffers.h"
#include "vpx_mem/vpx_mem.h"
int vp10_alloc_internal_frame_buffers(InternalFrameBufferList *list) {
assert(list != NULL);
vp10_free_internal_frame_buffers(list);
list->num_internal_frame_buffers =
VP9_MAXIMUM_REF_BUFFERS + VPX_MAXIMUM_WORK_BUFFERS;
list->int_fb =
(InternalFrameBuffer *)vpx_calloc(list->num_internal_frame_buffers,
sizeof(*list->int_fb));
return (list->int_fb == NULL);
}
void vp10_free_internal_frame_buffers(InternalFrameBufferList *list) {
int i;
assert(list != NULL);
for (i = 0; i < list->num_internal_frame_buffers; ++i) {
vpx_free(list->int_fb[i].data);
list->int_fb[i].data = NULL;
}
vpx_free(list->int_fb);
list->int_fb = NULL;
}
int vp10_get_frame_buffer(void *cb_priv, size_t min_size,
vpx_codec_frame_buffer_t *fb) {
int i;
InternalFrameBufferList *const int_fb_list =
(InternalFrameBufferList *)cb_priv;
if (int_fb_list == NULL)
return -1;
// Find a free frame buffer.
for (i = 0; i < int_fb_list->num_internal_frame_buffers; ++i) {
if (!int_fb_list->int_fb[i].in_use)
break;
}
if (i == int_fb_list->num_internal_frame_buffers)
return -1;
if (int_fb_list->int_fb[i].size < min_size) {
int_fb_list->int_fb[i].data =
(uint8_t *)vpx_realloc(int_fb_list->int_fb[i].data, min_size);
if (!int_fb_list->int_fb[i].data)
return -1;
// This memset is needed for fixing valgrind error from C loop filter
// due to access uninitialized memory in frame border. It could be
// removed if border is totally removed.
memset(int_fb_list->int_fb[i].data, 0, min_size);
int_fb_list->int_fb[i].size = min_size;
}
fb->data = int_fb_list->int_fb[i].data;
fb->size = int_fb_list->int_fb[i].size;
int_fb_list->int_fb[i].in_use = 1;
// Set the frame buffer's private data to point at the internal frame buffer.
fb->priv = &int_fb_list->int_fb[i];
return 0;
}
int vp10_release_frame_buffer(void *cb_priv, vpx_codec_frame_buffer_t *fb) {
InternalFrameBuffer *const int_fb = (InternalFrameBuffer *)fb->priv;
(void)cb_priv;
if (int_fb)
int_fb->in_use = 0;
return 0;
}

53
vp10/common/vp9_frame_buffers.h Normal file
View File

@ -0,0 +1,53 @@
/*
* Copyright (c) 2014 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 VP9_COMMON_VP9_FRAME_BUFFERS_H_
#define VP9_COMMON_VP9_FRAME_BUFFERS_H_
#include "vpx/vpx_frame_buffer.h"
#include "vpx/vpx_integer.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef struct InternalFrameBuffer {
uint8_t *data;
size_t size;
int in_use;
} InternalFrameBuffer;
typedef struct InternalFrameBufferList {
int num_internal_frame_buffers;
InternalFrameBuffer *int_fb;
} InternalFrameBufferList;
// Initializes |list|. Returns 0 on success.
int vp10_alloc_internal_frame_buffers(InternalFrameBufferList *list);
// Free any data allocated to the frame buffers.
void vp10_free_internal_frame_buffers(InternalFrameBufferList *list);
// Callback used by libvpx to request an external frame buffer. |cb_priv|
// Callback private data, which points to an InternalFrameBufferList.
// |min_size| is the minimum size in bytes needed to decode the next frame.
// |fb| pointer to the frame buffer.
int vp10_get_frame_buffer(void *cb_priv, size_t min_size,
vpx_codec_frame_buffer_t *fb);
// Callback used by libvpx when there are no references to the frame buffer.
// |cb_priv| is not used. |fb| pointer to the frame buffer.
int vp10_release_frame_buffer(void *cb_priv, vpx_codec_frame_buffer_t *fb);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_FRAME_BUFFERS_H_

403
vp10/common/vp9_idct.c Normal file
View File

@ -0,0 +1,403 @@
/*
* Copyright (c) 2010 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 <math.h>
#include "./vp10_rtcd.h"
#include "./vpx_dsp_rtcd.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_idct.h"
#include "vp10/common/vp9_systemdependent.h"
#include "vpx_dsp/inv_txfm.h"
#include "vpx_ports/mem.h"
void vp10_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride,
int tx_type) {
const transform_2d IHT_4[] = {
{ idct4_c, idct4_c }, // DCT_DCT = 0
{ iadst4_c, idct4_c }, // ADST_DCT = 1
{ idct4_c, iadst4_c }, // DCT_ADST = 2
{ iadst4_c, iadst4_c } // ADST_ADST = 3
};
int i, j;
tran_low_t out[4 * 4];
tran_low_t *outptr = out;
tran_low_t temp_in[4], temp_out[4];
// inverse transform row vectors
for (i = 0; i < 4; ++i) {
IHT_4[tx_type].rows(input, outptr);
input += 4;
outptr += 4;
}
// inverse transform column vectors
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j)
temp_in[j] = out[j * 4 + i];
IHT_4[tx_type].cols(temp_in, temp_out);
for (j = 0; j < 4; ++j) {
dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
ROUND_POWER_OF_TWO(temp_out[j], 4));
}
}
}
static const transform_2d IHT_8[] = {
{ idct8_c, idct8_c }, // DCT_DCT = 0
{ iadst8_c, idct8_c }, // ADST_DCT = 1
{ idct8_c, iadst8_c }, // DCT_ADST = 2
{ iadst8_c, iadst8_c } // ADST_ADST = 3
};
void vp10_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int stride,
int tx_type) {
int i, j;
tran_low_t out[8 * 8];
tran_low_t *outptr = out;
tran_low_t temp_in[8], temp_out[8];
const transform_2d ht = IHT_8[tx_type];
// inverse transform row vectors
for (i = 0; i < 8; ++i) {
ht.rows(input, outptr);
input += 8;
outptr += 8;
}
// inverse transform column vectors
for (i = 0; i < 8; ++i) {
for (j = 0; j < 8; ++j)
temp_in[j] = out[j * 8 + i];
ht.cols(temp_in, temp_out);
for (j = 0; j < 8; ++j) {
dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
ROUND_POWER_OF_TWO(temp_out[j], 5));
}
}
}
static const transform_2d IHT_16[] = {
{ idct16_c, idct16_c }, // DCT_DCT = 0
{ iadst16_c, idct16_c }, // ADST_DCT = 1
{ idct16_c, iadst16_c }, // DCT_ADST = 2
{ iadst16_c, iadst16_c } // ADST_ADST = 3
};
void vp10_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest, int stride,
int tx_type) {
int i, j;
tran_low_t out[16 * 16];
tran_low_t *outptr = out;
tran_low_t temp_in[16], temp_out[16];
const transform_2d ht = IHT_16[tx_type];
// Rows
for (i = 0; i < 16; ++i) {
ht.rows(input, outptr);
input += 16;
outptr += 16;
}
// Columns
for (i = 0; i < 16; ++i) {
for (j = 0; j < 16; ++j)
temp_in[j] = out[j * 16 + i];
ht.cols(temp_in, temp_out);
for (j = 0; j < 16; ++j) {
dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
ROUND_POWER_OF_TWO(temp_out[j], 6));
}
}
}
// idct
void vp10_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob) {
if (eob > 1)
vpx_idct4x4_16_add(input, dest, stride);
else
vpx_idct4x4_1_add(input, dest, stride);
}
void vp10_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob) {
if (eob > 1)
vpx_iwht4x4_16_add(input, dest, stride);
else
vpx_iwht4x4_1_add(input, dest, stride);
}
void vp10_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob) {
// If dc is 1, then input[0] is the reconstructed value, do not need
// dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
// The calculation can be simplified if there are not many non-zero dct
// coefficients. Use eobs to decide what to do.
// TODO(yunqingwang): "eobs = 1" case is also handled in vp10_short_idct8x8_c.
// Combine that with code here.
if (eob == 1)
// DC only DCT coefficient
vpx_idct8x8_1_add(input, dest, stride);
else if (eob <= 12)
vpx_idct8x8_12_add(input, dest, stride);
else
vpx_idct8x8_64_add(input, dest, stride);
}
void vp10_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob) {
/* The calculation can be simplified if there are not many non-zero dct
* coefficients. Use eobs to separate different cases. */
if (eob == 1)
/* DC only DCT coefficient. */
vpx_idct16x16_1_add(input, dest, stride);
else if (eob <= 10)
vpx_idct16x16_10_add(input, dest, stride);
else
vpx_idct16x16_256_add(input, dest, stride);
}
void vp10_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob) {
if (eob == 1)
vpx_idct32x32_1_add(input, dest, stride);
else if (eob <= 34)
// non-zero coeff only in upper-left 8x8
vpx_idct32x32_34_add(input, dest, stride);
else
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_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_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);
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
int stride, int tx_type, int bd) {
const highbd_transform_2d IHT_4[] = {
{ vpx_highbd_idct4_c, vpx_highbd_idct4_c }, // DCT_DCT = 0
{ vpx_highbd_iadst4_c, vpx_highbd_idct4_c }, // ADST_DCT = 1
{ vpx_highbd_idct4_c, vpx_highbd_iadst4_c }, // DCT_ADST = 2
{ vpx_highbd_iadst4_c, vpx_highbd_iadst4_c } // ADST_ADST = 3
};
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
int i, j;
tran_low_t out[4 * 4];
tran_low_t *outptr = out;
tran_low_t temp_in[4], temp_out[4];
// Inverse transform row vectors.
for (i = 0; i < 4; ++i) {
IHT_4[tx_type].rows(input, outptr, bd);
input += 4;
outptr += 4;
}
// Inverse transform column vectors.
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j)
temp_in[j] = out[j * 4 + i];
IHT_4[tx_type].cols(temp_in, temp_out, bd);
for (j = 0; j < 4; ++j) {
dest[j * stride + i] = highbd_clip_pixel_add(
dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);
}
}
}
static const highbd_transform_2d HIGH_IHT_8[] = {
{ vpx_highbd_idct8_c, vpx_highbd_idct8_c }, // DCT_DCT = 0
{ vpx_highbd_iadst8_c, vpx_highbd_idct8_c }, // ADST_DCT = 1
{ vpx_highbd_idct8_c, vpx_highbd_iadst8_c }, // DCT_ADST = 2
{ vpx_highbd_iadst8_c, vpx_highbd_iadst8_c } // ADST_ADST = 3
};
void vp10_highbd_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest8,
int stride, int tx_type, int bd) {
int i, j;
tran_low_t out[8 * 8];
tran_low_t *outptr = out;
tran_low_t temp_in[8], temp_out[8];
const highbd_transform_2d ht = HIGH_IHT_8[tx_type];
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
// Inverse transform row vectors.
for (i = 0; i < 8; ++i) {
ht.rows(input, outptr, bd);
input += 8;
outptr += 8;
}
// Inverse transform column vectors.
for (i = 0; i < 8; ++i) {
for (j = 0; j < 8; ++j)
temp_in[j] = out[j * 8 + i];
ht.cols(temp_in, temp_out, bd);
for (j = 0; j < 8; ++j) {
dest[j * stride + i] = highbd_clip_pixel_add(
dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd);
}
}
}
static const highbd_transform_2d HIGH_IHT_16[] = {
{ vpx_highbd_idct16_c, vpx_highbd_idct16_c }, // DCT_DCT = 0
{ vpx_highbd_iadst16_c, vpx_highbd_idct16_c }, // ADST_DCT = 1
{ vpx_highbd_idct16_c, vpx_highbd_iadst16_c }, // DCT_ADST = 2
{ vpx_highbd_iadst16_c, vpx_highbd_iadst16_c } // ADST_ADST = 3
};
void vp10_highbd_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest8,
int stride, int tx_type, int bd) {
int i, j;
tran_low_t out[16 * 16];
tran_low_t *outptr = out;
tran_low_t temp_in[16], temp_out[16];
const highbd_transform_2d ht = HIGH_IHT_16[tx_type];
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
// Rows
for (i = 0; i < 16; ++i) {
ht.rows(input, outptr, bd);
input += 16;
outptr += 16;
}
// Columns
for (i = 0; i < 16; ++i) {
for (j = 0; j < 16; ++j)
temp_in[j] = out[j * 16 + i];
ht.cols(temp_in, temp_out, bd);
for (j = 0; j < 16; ++j) {
dest[j * stride + i] = highbd_clip_pixel_add(
dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
}
}
}
// idct
void vp10_highbd_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd) {
if (eob > 1)
vpx_highbd_idct4x4_16_add(input, dest, stride, bd);
else
vpx_highbd_idct4x4_1_add(input, dest, stride, bd);
}
void vp10_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd) {
if (eob > 1)
vpx_highbd_iwht4x4_16_add(input, dest, stride, bd);
else
vpx_highbd_iwht4x4_1_add(input, dest, stride, bd);
}
void vp10_highbd_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd) {
// If dc is 1, then input[0] is the reconstructed value, do not need
// dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
// The calculation can be simplified if there are not many non-zero dct
// coefficients. Use eobs to decide what to do.
// TODO(yunqingwang): "eobs = 1" case is also handled in vp10_short_idct8x8_c.
// Combine that with code here.
// DC only DCT coefficient
if (eob == 1) {
vpx_highbd_idct8x8_1_add(input, dest, stride, bd);
} else if (eob <= 10) {
vpx_highbd_idct8x8_10_add(input, dest, stride, bd);
} else {
vpx_highbd_idct8x8_64_add(input, dest, stride, bd);
}
}
void vp10_highbd_idct16x16_add(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd) {
// The calculation can be simplified if there are not many non-zero dct
// coefficients. Use eobs to separate different cases.
// DC only DCT coefficient.
if (eob == 1) {
vpx_highbd_idct16x16_1_add(input, dest, stride, bd);
} else if (eob <= 10) {
vpx_highbd_idct16x16_10_add(input, dest, stride, bd);
} else {
vpx_highbd_idct16x16_256_add(input, dest, stride, bd);
}
}
void vp10_highbd_idct32x32_add(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd) {
// Non-zero coeff only in upper-left 8x8
if (eob == 1) {
vpx_highbd_idct32x32_1_add(input, dest, stride, bd);
} else if (eob <= 34) {
vpx_highbd_idct32x32_34_add(input, dest, stride, bd);
} else {
vpx_highbd_idct32x32_1024_add(input, dest, stride, bd);
}
}
// 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_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_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);
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH

81
vp10/common/vp9_idct.h Normal file
View File

@ -0,0 +1,81 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_IDCT_H_
#define VP9_COMMON_VP9_IDCT_H_
#include <assert.h>
#include "./vpx_config.h"
#include "vp10/common/vp9_common.h"
#include "vp10/common/vp9_enums.h"
#include "vpx_dsp/inv_txfm.h"
#include "vpx_dsp/txfm_common.h"
#include "vpx_ports/mem.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef void (*transform_1d)(const tran_low_t*, tran_low_t*);
typedef struct {
transform_1d cols, rows; // vertical and horizontal
} transform_2d;
#if CONFIG_VP9_HIGHBITDEPTH
typedef void (*highbd_transform_1d)(const tran_low_t*, tran_low_t*, int bd);
typedef struct {
highbd_transform_1d cols, rows; // vertical and horizontal
} highbd_transform_2d;
#endif // CONFIG_VP9_HIGHBITDEPTH
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);
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd);
void vp10_highbd_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd);
void vp10_highbd_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob, int bd);
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);
#endif // CONFIG_VP9_HIGHBITDEPTH
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_IDCT_H_

1616
vp10/common/vp9_loopfilter.c Normal file
View File

File diff suppressed because it is too large Load Diff

155
vp10/common/vp9_loopfilter.h Normal file
View File

@ -0,0 +1,155 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_LOOPFILTER_H_
#define VP9_COMMON_VP9_LOOPFILTER_H_
#include "vpx_ports/mem.h"
#include "./vpx_config.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_seg_common.h"
#ifdef __cplusplus
extern "C" {
#endif
#define MAX_LOOP_FILTER 63
#define MAX_SHARPNESS 7
#define SIMD_WIDTH 16
#define MAX_REF_LF_DELTAS 4
#define MAX_MODE_LF_DELTAS 2
enum lf_path {
LF_PATH_420,
LF_PATH_444,
LF_PATH_SLOW,
};
struct loopfilter {
int filter_level;
int sharpness_level;
int last_sharpness_level;
uint8_t mode_ref_delta_enabled;
uint8_t mode_ref_delta_update;
// 0 = Intra, Last, GF, ARF
signed char ref_deltas[MAX_REF_LF_DELTAS];
signed char last_ref_deltas[MAX_REF_LF_DELTAS];
// 0 = ZERO_MV, MV
signed char mode_deltas[MAX_MODE_LF_DELTAS];
signed char last_mode_deltas[MAX_MODE_LF_DELTAS];
};
// Need to align this structure so when it is declared and
// passed it can be loaded into vector registers.
typedef struct {
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, mblim[SIMD_WIDTH]);
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, lim[SIMD_WIDTH]);
DECLARE_ALIGNED(SIMD_WIDTH, uint8_t, hev_thr[SIMD_WIDTH]);
} loop_filter_thresh;
typedef struct {
loop_filter_thresh lfthr[MAX_LOOP_FILTER + 1];
uint8_t lvl[MAX_SEGMENTS][MAX_REF_FRAMES][MAX_MODE_LF_DELTAS];
} loop_filter_info_n;
// This structure holds bit masks for all 8x8 blocks in a 64x64 region.
// Each 1 bit represents a position in which we want to apply the loop filter.
// Left_ entries refer to whether we apply a filter on the border to the
// left of the block. Above_ entries refer to whether or not to apply a
// filter on the above border. Int_ entries refer to whether or not to
// apply borders on the 4x4 edges within the 8x8 block that each bit
// represents.
// Since each transform is accompanied by a potentially different type of
// loop filter there is a different entry in the array for each transform size.
typedef struct {
uint64_t left_y[TX_SIZES];
uint64_t above_y[TX_SIZES];
uint64_t int_4x4_y;
uint16_t left_uv[TX_SIZES];
uint16_t above_uv[TX_SIZES];
uint16_t int_4x4_uv;
uint8_t lfl_y[64];
uint8_t lfl_uv[16];
} LOOP_FILTER_MASK;
/* assorted loopfilter functions which get used elsewhere */
struct VP9Common;
struct macroblockd;
struct VP9LfSyncData;
// This function sets up the bit masks for the entire 64x64 region represented
// by mi_row, mi_col.
void vp10_setup_mask(struct VP9Common *const cm,
const int mi_row, const int mi_col,
MODE_INFO **mi_8x8, const int mode_info_stride,
LOOP_FILTER_MASK *lfm);
void vp10_filter_block_plane_ss00(struct VP9Common *const cm,
struct macroblockd_plane *const plane,
int mi_row,
LOOP_FILTER_MASK *lfm);
void vp10_filter_block_plane_ss11(struct VP9Common *const cm,
struct macroblockd_plane *const plane,
int mi_row,
LOOP_FILTER_MASK *lfm);
void vp10_filter_block_plane_non420(struct VP9Common *cm,
struct macroblockd_plane *plane,
MODE_INFO **mi_8x8,
int mi_row, int mi_col);
void vp10_loop_filter_init(struct VP9Common *cm);
// Update the loop filter for the current frame.
// This should be called before vp10_loop_filter_rows(), vp10_loop_filter_frame()
// calls this function directly.
void vp10_loop_filter_frame_init(struct VP9Common *cm, int default_filt_lvl);
void vp10_loop_filter_frame(YV12_BUFFER_CONFIG *frame,
struct VP9Common *cm,
struct macroblockd *mbd,
int filter_level,
int y_only, int partial_frame);
// Apply the loop filter to [start, stop) macro block rows in frame_buffer.
void vp10_loop_filter_rows(YV12_BUFFER_CONFIG *frame_buffer,
struct VP9Common *cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int start, int stop, int y_only);
typedef struct LoopFilterWorkerData {
YV12_BUFFER_CONFIG *frame_buffer;
struct VP9Common *cm;
struct macroblockd_plane planes[MAX_MB_PLANE];
int start;
int stop;
int y_only;
} LFWorkerData;
void vp10_loop_filter_data_reset(
LFWorkerData *lf_data, YV12_BUFFER_CONFIG *frame_buffer,
struct VP9Common *cm, const struct macroblockd_plane planes[MAX_MB_PLANE]);
// Operates on the rows described by 'lf_data'.
int vp10_loop_filter_worker(LFWorkerData *const lf_data, void *unused);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_LOOPFILTER_H_

394
vp10/common/vp9_mfqe.c Normal file
View File

@ -0,0 +1,394 @@
/*
* Copyright (c) 2014 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 "./vpx_config.h"
#include "./vp10_rtcd.h"
#include "./vpx_dsp_rtcd.h"
#include "./vpx_scale_rtcd.h"
#include "vp10/common/vp9_onyxc_int.h"
#include "vp10/common/vp9_postproc.h"
// TODO(jackychen): Replace this function with SSE2 code. There is
// one SSE2 implementation in vp8, so will consider how to share it
// between vp8 and vp9.
static void filter_by_weight(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
int block_size, int src_weight) {
const int dst_weight = (1 << MFQE_PRECISION) - src_weight;
const int rounding_bit = 1 << (MFQE_PRECISION - 1);
int r, c;
for (r = 0; r < block_size; r++) {
for (c = 0; c < block_size; c++) {
dst[c] = (src[c] * src_weight + dst[c] * dst_weight + rounding_bit)
>> MFQE_PRECISION;
}
src += src_stride;
dst += dst_stride;
}
}
void vp10_filter_by_weight8x8_c(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride, int src_weight) {
filter_by_weight(src, src_stride, dst, dst_stride, 8, src_weight);
}
void vp10_filter_by_weight16x16_c(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
int src_weight) {
filter_by_weight(src, src_stride, dst, dst_stride, 16, src_weight);
}
static void filter_by_weight32x32(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride, int weight) {
vp10_filter_by_weight16x16(src, src_stride, dst, dst_stride, weight);
vp10_filter_by_weight16x16(src + 16, src_stride, dst + 16, dst_stride,
weight);
vp10_filter_by_weight16x16(src + src_stride * 16, src_stride,
dst + dst_stride * 16, dst_stride, weight);
vp10_filter_by_weight16x16(src + src_stride * 16 + 16, src_stride,
dst + dst_stride * 16 + 16, dst_stride, weight);
}
static void filter_by_weight64x64(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride, int weight) {
filter_by_weight32x32(src, src_stride, dst, dst_stride, weight);
filter_by_weight32x32(src + 32, src_stride, dst + 32,
dst_stride, weight);
filter_by_weight32x32(src + src_stride * 32, src_stride,
dst + dst_stride * 32, dst_stride, weight);
filter_by_weight32x32(src + src_stride * 32 + 32, src_stride,
dst + dst_stride * 32 + 32, dst_stride, weight);
}
static void apply_ifactor(const uint8_t *y, int y_stride, uint8_t *yd,
int yd_stride, const uint8_t *u, const uint8_t *v,
int uv_stride, uint8_t *ud, uint8_t *vd,
int uvd_stride, BLOCK_SIZE block_size,
int weight) {
if (block_size == BLOCK_16X16) {
vp10_filter_by_weight16x16(y, y_stride, yd, yd_stride, weight);
vp10_filter_by_weight8x8(u, uv_stride, ud, uvd_stride, weight);
vp10_filter_by_weight8x8(v, uv_stride, vd, uvd_stride, weight);
} else if (block_size == BLOCK_32X32) {
filter_by_weight32x32(y, y_stride, yd, yd_stride, weight);
vp10_filter_by_weight16x16(u, uv_stride, ud, uvd_stride, weight);
vp10_filter_by_weight16x16(v, uv_stride, vd, uvd_stride, weight);
} else if (block_size == BLOCK_64X64) {
filter_by_weight64x64(y, y_stride, yd, yd_stride, weight);
filter_by_weight32x32(u, uv_stride, ud, uvd_stride, weight);
filter_by_weight32x32(v, uv_stride, vd, uvd_stride, weight);
}
}
// TODO(jackychen): Determine whether replace it with assembly code.
static void copy_mem8x8(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride) {
int r;
for (r = 0; r < 8; r++) {
memcpy(dst, src, 8);
src += src_stride;
dst += dst_stride;
}
}
static void copy_mem16x16(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride) {
int r;
for (r = 0; r < 16; r++) {
memcpy(dst, src, 16);
src += src_stride;
dst += dst_stride;
}
}
static void copy_mem32x32(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride) {
copy_mem16x16(src, src_stride, dst, dst_stride);
copy_mem16x16(src + 16, src_stride, dst + 16, dst_stride);
copy_mem16x16(src + src_stride * 16, src_stride,
dst + dst_stride * 16, dst_stride);
copy_mem16x16(src + src_stride * 16 + 16, src_stride,
dst + dst_stride * 16 + 16, dst_stride);
}
void copy_mem64x64(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride) {
copy_mem32x32(src, src_stride, dst, dst_stride);
copy_mem32x32(src + 32, src_stride, dst + 32, dst_stride);
copy_mem32x32(src + src_stride * 32, src_stride,
dst + src_stride * 32, dst_stride);
copy_mem32x32(src + src_stride * 32 + 32, src_stride,
dst + src_stride * 32 + 32, dst_stride);
}
static void copy_block(const uint8_t *y, const uint8_t *u, const uint8_t *v,
int y_stride, int uv_stride, uint8_t *yd, uint8_t *ud,
uint8_t *vd, int yd_stride, int uvd_stride,
BLOCK_SIZE bs) {
if (bs == BLOCK_16X16) {
copy_mem16x16(y, y_stride, yd, yd_stride);
copy_mem8x8(u, uv_stride, ud, uvd_stride);
copy_mem8x8(v, uv_stride, vd, uvd_stride);
} else if (bs == BLOCK_32X32) {
copy_mem32x32(y, y_stride, yd, yd_stride);
copy_mem16x16(u, uv_stride, ud, uvd_stride);
copy_mem16x16(v, uv_stride, vd, uvd_stride);
} else {
copy_mem64x64(y, y_stride, yd, yd_stride);
copy_mem32x32(u, uv_stride, ud, uvd_stride);
copy_mem32x32(v, uv_stride, vd, uvd_stride);
}
}
static void get_thr(BLOCK_SIZE bs, int qdiff, int *sad_thr, int *vdiff_thr) {
const int adj = qdiff >> MFQE_PRECISION;
if (bs == BLOCK_16X16) {
*sad_thr = 7 + adj;
} else if (bs == BLOCK_32X32) {
*sad_thr = 6 + adj;
} else { // BLOCK_64X64
*sad_thr = 5 + adj;
}
*vdiff_thr = 125 + qdiff;
}
static void mfqe_block(BLOCK_SIZE bs, const uint8_t *y, const uint8_t *u,
const uint8_t *v, int y_stride, int uv_stride,
uint8_t *yd, uint8_t *ud, uint8_t *vd, int yd_stride,
int uvd_stride, int qdiff) {
int sad, sad_thr, vdiff, vdiff_thr;
uint32_t sse;
get_thr(bs, qdiff, &sad_thr, &vdiff_thr);
if (bs == BLOCK_16X16) {
vdiff = (vpx_variance16x16(y, y_stride, yd, yd_stride, &sse) + 128) >> 8;
sad = (vpx_sad16x16(y, y_stride, yd, yd_stride) + 128) >> 8;
} else if (bs == BLOCK_32X32) {
vdiff = (vpx_variance32x32(y, y_stride, yd, yd_stride, &sse) + 512) >> 10;
sad = (vpx_sad32x32(y, y_stride, yd, yd_stride) + 512) >> 10;
} else /* if (bs == BLOCK_64X64) */ {
vdiff = (vpx_variance64x64(y, y_stride, yd, yd_stride, &sse) + 2048) >> 12;
sad = (vpx_sad64x64(y, y_stride, yd, yd_stride) + 2048) >> 12;
}
// vdiff > sad * 3 means vdiff should not be too small, otherwise,
// it might be a lighting change in smooth area. When there is a
// lighting change in smooth area, it is dangerous to do MFQE.
if (sad > 1 && vdiff > sad * 3) {
const int weight = 1 << MFQE_PRECISION;
int ifactor = weight * sad * vdiff / (sad_thr * vdiff_thr);
// When ifactor equals weight, no MFQE is done.
if (ifactor > weight) {
ifactor = weight;
}
apply_ifactor(y, y_stride, yd, yd_stride, u, v, uv_stride, ud, vd,
uvd_stride, bs, ifactor);
} else {
// Copy the block from current frame (i.e., no mfqe is done).
copy_block(y, u, v, y_stride, uv_stride, yd, ud, vd,
yd_stride, uvd_stride, bs);
}
}
static int mfqe_decision(MODE_INFO *mi, BLOCK_SIZE cur_bs) {
// Check the motion in current block(for inter frame),
// or check the motion in the correlated block in last frame (for keyframe).
const int mv_len_square = mi->mbmi.mv[0].as_mv.row *
mi->mbmi.mv[0].as_mv.row +
mi->mbmi.mv[0].as_mv.col *
mi->mbmi.mv[0].as_mv.col;
const int mv_threshold = 100;
return mi->mbmi.mode >= NEARESTMV && // Not an intra block
cur_bs >= BLOCK_16X16 &&
mv_len_square <= mv_threshold;
}
// Process each partiton in a super block, recursively.
static void mfqe_partition(VP9_COMMON *cm, MODE_INFO *mi, BLOCK_SIZE bs,
const uint8_t *y, const uint8_t *u,
const uint8_t *v, int y_stride, int uv_stride,
uint8_t *yd, uint8_t *ud, uint8_t *vd,
int yd_stride, int uvd_stride) {
int mi_offset, y_offset, uv_offset;
const BLOCK_SIZE cur_bs = mi->mbmi.sb_type;
const int qdiff = cm->base_qindex - cm->postproc_state.last_base_qindex;
const int bsl = b_width_log2_lookup[bs];
PARTITION_TYPE partition = partition_lookup[bsl][cur_bs];
const BLOCK_SIZE subsize = get_subsize(bs, partition);
if (cur_bs < BLOCK_8X8) {
// If there are blocks smaller than 8x8, it must be on the boundary.
return;
}
// No MFQE on blocks smaller than 16x16
if (bs == BLOCK_16X16) {
partition = PARTITION_NONE;
}
if (bs == BLOCK_64X64) {
mi_offset = 4;
y_offset = 32;
uv_offset = 16;
} else {
mi_offset = 2;
y_offset = 16;
uv_offset = 8;
}
switch (partition) {
BLOCK_SIZE mfqe_bs, bs_tmp;
case PARTITION_HORZ:
if (bs == BLOCK_64X64) {
mfqe_bs = BLOCK_64X32;
bs_tmp = BLOCK_32X32;
} else {
mfqe_bs = BLOCK_32X16;
bs_tmp = BLOCK_16X16;
}
if (mfqe_decision(mi, mfqe_bs)) {
// Do mfqe on the first square partition.
mfqe_block(bs_tmp, y, u, v, y_stride, uv_stride,
yd, ud, vd, yd_stride, uvd_stride, qdiff);
// Do mfqe on the second square partition.
mfqe_block(bs_tmp, y + y_offset, u + uv_offset, v + uv_offset,
y_stride, uv_stride, yd + y_offset, ud + uv_offset,
vd + uv_offset, yd_stride, uvd_stride, qdiff);
}
if (mfqe_decision(mi + mi_offset * cm->mi_stride, mfqe_bs)) {
// Do mfqe on the first square partition.
mfqe_block(bs_tmp, y + y_offset * y_stride, u + uv_offset * uv_stride,
v + uv_offset * uv_stride, y_stride, uv_stride,
yd + y_offset * yd_stride, ud + uv_offset * uvd_stride,
vd + uv_offset * uvd_stride, yd_stride, uvd_stride, qdiff);
// Do mfqe on the second square partition.
mfqe_block(bs_tmp, y + y_offset * y_stride + y_offset,
u + uv_offset * uv_stride + uv_offset,
v + uv_offset * uv_stride + uv_offset, y_stride,
uv_stride, yd + y_offset * yd_stride + y_offset,
ud + uv_offset * uvd_stride + uv_offset,
vd + uv_offset * uvd_stride + uv_offset,
yd_stride, uvd_stride, qdiff);
}
break;
case PARTITION_VERT:
if (bs == BLOCK_64X64) {
mfqe_bs = BLOCK_32X64;
bs_tmp = BLOCK_32X32;
} else {
mfqe_bs = BLOCK_16X32;
bs_tmp = BLOCK_16X16;
}
if (mfqe_decision(mi, mfqe_bs)) {
// Do mfqe on the first square partition.
mfqe_block(bs_tmp, y, u, v, y_stride, uv_stride,
yd, ud, vd, yd_stride, uvd_stride, qdiff);
// Do mfqe on the second square partition.
mfqe_block(bs_tmp, y + y_offset * y_stride, u + uv_offset * uv_stride,
v + uv_offset * uv_stride, y_stride, uv_stride,
yd + y_offset * yd_stride, ud + uv_offset * uvd_stride,
vd + uv_offset * uvd_stride, yd_stride, uvd_stride, qdiff);
}
if (mfqe_decision(mi + mi_offset, mfqe_bs)) {
// Do mfqe on the first square partition.
mfqe_block(bs_tmp, y + y_offset, u + uv_offset, v + uv_offset,
y_stride, uv_stride, yd + y_offset, ud + uv_offset,
vd + uv_offset, yd_stride, uvd_stride, qdiff);
// Do mfqe on the second square partition.
mfqe_block(bs_tmp, y + y_offset * y_stride + y_offset,
u + uv_offset * uv_stride + uv_offset,
v + uv_offset * uv_stride + uv_offset, y_stride,
uv_stride, yd + y_offset * yd_stride + y_offset,
ud + uv_offset * uvd_stride + uv_offset,
vd + uv_offset * uvd_stride + uv_offset,
yd_stride, uvd_stride, qdiff);
}
break;
case PARTITION_NONE:
if (mfqe_decision(mi, cur_bs)) {
// Do mfqe on this partition.
mfqe_block(cur_bs, y, u, v, y_stride, uv_stride,
yd, ud, vd, yd_stride, uvd_stride, qdiff);
} else {
// Copy the block from current frame(i.e., no mfqe is done).
copy_block(y, u, v, y_stride, uv_stride, yd, ud, vd,
yd_stride, uvd_stride, bs);
}
break;
case PARTITION_SPLIT:
// Recursion on four square partitions, e.g. if bs is 64X64,
// then look into four 32X32 blocks in it.
mfqe_partition(cm, mi, subsize, y, u, v, y_stride, uv_stride, yd, ud, vd,
yd_stride, uvd_stride);
mfqe_partition(cm, mi + mi_offset, subsize, y + y_offset, u + uv_offset,
v + uv_offset, y_stride, uv_stride, yd + y_offset,
ud + uv_offset, vd + uv_offset, yd_stride, uvd_stride);
mfqe_partition(cm, mi + mi_offset * cm->mi_stride, subsize,
y + y_offset * y_stride, u + uv_offset * uv_stride,
v + uv_offset * uv_stride, y_stride, uv_stride,
yd + y_offset * yd_stride, ud + uv_offset * uvd_stride,
vd + uv_offset * uvd_stride, yd_stride, uvd_stride);
mfqe_partition(cm, mi + mi_offset * cm->mi_stride + mi_offset,
subsize, y + y_offset * y_stride + y_offset,
u + uv_offset * uv_stride + uv_offset,
v + uv_offset * uv_stride + uv_offset, y_stride,
uv_stride, yd + y_offset * yd_stride + y_offset,
ud + uv_offset * uvd_stride + uv_offset,
vd + uv_offset * uvd_stride + uv_offset,
yd_stride, uvd_stride);
break;
default:
assert(0);
}
}
void vp10_mfqe(VP9_COMMON *cm) {
int mi_row, mi_col;
// Current decoded frame.
const YV12_BUFFER_CONFIG *show = cm->frame_to_show;
// Last decoded frame and will store the MFQE result.
YV12_BUFFER_CONFIG *dest = &cm->post_proc_buffer;
// Loop through each super block.
for (mi_row = 0; mi_row < cm->mi_rows; mi_row += MI_BLOCK_SIZE) {
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE) {
MODE_INFO *mi;
MODE_INFO *mi_local = cm->mi + (mi_row * cm->mi_stride + mi_col);
// Motion Info in last frame.
MODE_INFO *mi_prev = cm->postproc_state.prev_mi +
(mi_row * cm->mi_stride + mi_col);
const uint32_t y_stride = show->y_stride;
const uint32_t uv_stride = show->uv_stride;
const uint32_t yd_stride = dest->y_stride;
const uint32_t uvd_stride = dest->uv_stride;
const uint32_t row_offset_y = mi_row << 3;
const uint32_t row_offset_uv = mi_row << 2;
const uint32_t col_offset_y = mi_col << 3;
const uint32_t col_offset_uv = mi_col << 2;
const uint8_t *y = show->y_buffer + row_offset_y * y_stride +
col_offset_y;
const uint8_t *u = show->u_buffer + row_offset_uv * uv_stride +
col_offset_uv;
const uint8_t *v = show->v_buffer + row_offset_uv * uv_stride +
col_offset_uv;
uint8_t *yd = dest->y_buffer + row_offset_y * yd_stride + col_offset_y;
uint8_t *ud = dest->u_buffer + row_offset_uv * uvd_stride +
col_offset_uv;
uint8_t *vd = dest->v_buffer + row_offset_uv * uvd_stride +
col_offset_uv;
if (frame_is_intra_only(cm)) {
mi = mi_prev;
} else {
mi = mi_local;
}
mfqe_partition(cm, mi, BLOCK_64X64, y, u, v, y_stride, uv_stride, yd, ud,
vd, yd_stride, uvd_stride);
}
}
}

31
vp10/common/vp9_mfqe.h Normal file
View File

@ -0,0 +1,31 @@
/*
* Copyright (c) 2014 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 VP9_COMMON_VP9_MFQE_H_
#define VP9_COMMON_VP9_MFQE_H_
#ifdef __cplusplus
extern "C" {
#endif
// Multiframe Quality Enhancement.
// The aim for MFQE is to replace pixel blocks in the current frame with
// the correlated pixel blocks (with higher quality) in the last frame.
// The replacement can only be taken in stationary blocks by checking
// the motion of the blocks and other conditions such as the SAD of
// the current block and correlated block, the variance of the block
// difference, etc.
void vp10_mfqe(struct VP9Common *cm);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_MFQE_H_

55
vp10/common/vp9_mv.h Normal file
View File

@ -0,0 +1,55 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_MV_H_
#define VP9_COMMON_VP9_MV_H_
#include "vpx/vpx_integer.h"
#include "vp10/common/vp9_common.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef struct mv {
int16_t row;
int16_t col;
} MV;
typedef union int_mv {
uint32_t as_int;
MV as_mv;
} int_mv; /* facilitates faster equality tests and copies */
typedef struct mv32 {
int32_t row;
int32_t col;
} MV32;
static INLINE int is_zero_mv(const MV *mv) {
return *((const uint32_t *)mv) == 0;
}
static INLINE int is_equal_mv(const MV *a, const MV *b) {
return *((const uint32_t *)a) == *((const uint32_t *)b);
}
static INLINE void clamp_mv(MV *mv, int min_col, int max_col,
int min_row, int max_row) {
mv->col = clamp(mv->col, min_col, max_col);
mv->row = clamp(mv->row, min_row, max_row);
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_MV_H_

229
vp10/common/vp9_mvref_common.c Normal file
View File

@ -0,0 +1,229 @@
/*
* Copyright (c) 2012 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/common/vp9_mvref_common.h"
// This function searches the neighbourhood of a given MB/SB
// to try and find candidate reference vectors.
static void find_mv_refs_idx(const VP9_COMMON *cm, const MACROBLOCKD *xd,
MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
int_mv *mv_ref_list,
int block, int mi_row, int mi_col,
find_mv_refs_sync sync, void *const data,
uint8_t *mode_context) {
const int *ref_sign_bias = cm->ref_frame_sign_bias;
int i, refmv_count = 0;
const POSITION *const mv_ref_search = mv_ref_blocks[mi->mbmi.sb_type];
int different_ref_found = 0;
int context_counter = 0;
const MV_REF *const prev_frame_mvs = cm->use_prev_frame_mvs ?
cm->prev_frame->mvs + mi_row * cm->mi_cols + mi_col : NULL;
const TileInfo *const tile = &xd->tile;
// Blank the reference vector list
memset(mv_ref_list, 0, sizeof(*mv_ref_list) * MAX_MV_REF_CANDIDATES);
// The nearest 2 blocks are treated differently
// if the size < 8x8 we get the mv from the bmi substructure,
// and we also need to keep a mode count.
for (i = 0; i < 2; ++i) {
const POSITION *const mv_ref = &mv_ref_search[i];
if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
const MODE_INFO *const candidate_mi = xd->mi[mv_ref->col + mv_ref->row *
xd->mi_stride];
const MB_MODE_INFO *const candidate = &candidate_mi->mbmi;
// Keep counts for entropy encoding.
context_counter += mode_2_counter[candidate->mode];
different_ref_found = 1;
if (candidate->ref_frame[0] == ref_frame)
ADD_MV_REF_LIST(get_sub_block_mv(candidate_mi, 0, mv_ref->col, block),
refmv_count, mv_ref_list, Done);
else if (candidate->ref_frame[1] == ref_frame)
ADD_MV_REF_LIST(get_sub_block_mv(candidate_mi, 1, mv_ref->col, block),
refmv_count, mv_ref_list, Done);
}
}
// Check the rest of the neighbors in much the same way
// as before except we don't need to keep track of sub blocks or
// mode counts.
for (; i < MVREF_NEIGHBOURS; ++i) {
const POSITION *const mv_ref = &mv_ref_search[i];
if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row *
xd->mi_stride]->mbmi;
different_ref_found = 1;
if (candidate->ref_frame[0] == ref_frame)
ADD_MV_REF_LIST(candidate->mv[0], refmv_count, mv_ref_list, Done);
else if (candidate->ref_frame[1] == ref_frame)
ADD_MV_REF_LIST(candidate->mv[1], refmv_count, mv_ref_list, Done);
}
}
// TODO(hkuang): Remove this sync after fixing pthread_cond_broadcast
// on windows platform. The sync here is unncessary if use_perv_frame_mvs
// is 0. But after removing it, there will be hang in the unit test on windows
// due to several threads waiting for a thread's signal.
#if defined(_WIN32) && !HAVE_PTHREAD_H
if (cm->frame_parallel_decode && sync != NULL) {
sync(data, mi_row);
}
#endif
// Check the last frame's mode and mv info.
if (cm->use_prev_frame_mvs) {
// Synchronize here for frame parallel decode if sync function is provided.
if (cm->frame_parallel_decode && sync != NULL) {
sync(data, mi_row);
}
if (prev_frame_mvs->ref_frame[0] == ref_frame) {
ADD_MV_REF_LIST(prev_frame_mvs->mv[0], refmv_count, mv_ref_list, Done);
} else if (prev_frame_mvs->ref_frame[1] == ref_frame) {
ADD_MV_REF_LIST(prev_frame_mvs->mv[1], refmv_count, mv_ref_list, Done);
}
}
// Since we couldn't find 2 mvs from the same reference frame
// go back through the neighbors and find motion vectors from
// different reference frames.
if (different_ref_found) {
for (i = 0; i < MVREF_NEIGHBOURS; ++i) {
const POSITION *mv_ref = &mv_ref_search[i];
if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {
const MB_MODE_INFO *const candidate = &xd->mi[mv_ref->col + mv_ref->row
* xd->mi_stride]->mbmi;
// If the candidate is INTRA we don't want to consider its mv.
IF_DIFF_REF_FRAME_ADD_MV(candidate, ref_frame, ref_sign_bias,
refmv_count, mv_ref_list, Done);
}
}
}
// Since we still don't have a candidate we'll try the last frame.
if (cm->use_prev_frame_mvs) {
if (prev_frame_mvs->ref_frame[0] != ref_frame &&
prev_frame_mvs->ref_frame[0] > INTRA_FRAME) {
int_mv mv = prev_frame_mvs->mv[0];
if (ref_sign_bias[prev_frame_mvs->ref_frame[0]] !=
ref_sign_bias[ref_frame]) {
mv.as_mv.row *= -1;
mv.as_mv.col *= -1;
}
ADD_MV_REF_LIST(mv, refmv_count, mv_ref_list, Done);
}
if (prev_frame_mvs->ref_frame[1] > INTRA_FRAME &&
prev_frame_mvs->ref_frame[1] != ref_frame &&
prev_frame_mvs->mv[1].as_int != prev_frame_mvs->mv[0].as_int) {
int_mv mv = prev_frame_mvs->mv[1];
if (ref_sign_bias[prev_frame_mvs->ref_frame[1]] !=
ref_sign_bias[ref_frame]) {
mv.as_mv.row *= -1;
mv.as_mv.col *= -1;
}
ADD_MV_REF_LIST(mv, refmv_count, mv_ref_list, Done);
}
}
Done:
mode_context[ref_frame] = counter_to_context[context_counter];
// Clamp vectors
for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i)
clamp_mv_ref(&mv_ref_list[i].as_mv, xd);
}
void vp10_find_mv_refs(const VP9_COMMON *cm, const MACROBLOCKD *xd,
MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
int_mv *mv_ref_list,
int mi_row, int mi_col,
find_mv_refs_sync sync, void *const data,
uint8_t *mode_context) {
find_mv_refs_idx(cm, xd, mi, ref_frame, mv_ref_list, -1,
mi_row, mi_col, sync, data, mode_context);
}
static void lower_mv_precision(MV *mv, int allow_hp) {
const int use_hp = allow_hp && vp10_use_mv_hp(mv);
if (!use_hp) {
if (mv->row & 1)
mv->row += (mv->row > 0 ? -1 : 1);
if (mv->col & 1)
mv->col += (mv->col > 0 ? -1 : 1);
}
}
void vp10_find_best_ref_mvs(MACROBLOCKD *xd, int allow_hp,
int_mv *mvlist, int_mv *nearest_mv,
int_mv *near_mv) {
int i;
// Make sure all the candidates are properly clamped etc
for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i) {
lower_mv_precision(&mvlist[i].as_mv, allow_hp);
clamp_mv2(&mvlist[i].as_mv, xd);
}
*nearest_mv = mvlist[0];
*near_mv = mvlist[1];
}
void vp10_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
int block, int ref, int mi_row, int mi_col,
int_mv *nearest_mv, int_mv *near_mv,
uint8_t *mode_context) {
int_mv mv_list[MAX_MV_REF_CANDIDATES];
MODE_INFO *const mi = xd->mi[0];
b_mode_info *bmi = mi->bmi;
int n;
assert(MAX_MV_REF_CANDIDATES == 2);
find_mv_refs_idx(cm, xd, mi, mi->mbmi.ref_frame[ref], mv_list, block,
mi_row, mi_col, NULL, NULL, mode_context);
near_mv->as_int = 0;
switch (block) {
case 0:
nearest_mv->as_int = mv_list[0].as_int;
near_mv->as_int = mv_list[1].as_int;
break;
case 1:
case 2:
nearest_mv->as_int = bmi[0].as_mv[ref].as_int;
for (n = 0; n < MAX_MV_REF_CANDIDATES; ++n)
if (nearest_mv->as_int != mv_list[n].as_int) {
near_mv->as_int = mv_list[n].as_int;
break;
}
break;
case 3: {
int_mv candidates[2 + MAX_MV_REF_CANDIDATES];
candidates[0] = bmi[1].as_mv[ref];
candidates[1] = bmi[0].as_mv[ref];
candidates[2] = mv_list[0];
candidates[3] = mv_list[1];
nearest_mv->as_int = bmi[2].as_mv[ref].as_int;
for (n = 0; n < 2 + MAX_MV_REF_CANDIDATES; ++n)
if (nearest_mv->as_int != candidates[n].as_int) {
near_mv->as_int = candidates[n].as_int;
break;
}
break;
}
default:
assert(0 && "Invalid block index.");
}
}

232
vp10/common/vp9_mvref_common.h Normal file
View File

@ -0,0 +1,232 @@
/*
* Copyright (c) 2012 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 VP9_COMMON_VP9_MVREF_COMMON_H_
#define VP9_COMMON_VP9_MVREF_COMMON_H_
#include "vp10/common/vp9_onyxc_int.h"
#include "vp10/common/vp9_blockd.h"
#ifdef __cplusplus
extern "C" {
#endif
#define LEFT_TOP_MARGIN ((VP9_ENC_BORDER_IN_PIXELS - VP9_INTERP_EXTEND) << 3)
#define RIGHT_BOTTOM_MARGIN ((VP9_ENC_BORDER_IN_PIXELS -\
VP9_INTERP_EXTEND) << 3)
#define MVREF_NEIGHBOURS 8
typedef struct position {
int row;
int col;
} POSITION;
typedef enum {
BOTH_ZERO = 0,
ZERO_PLUS_PREDICTED = 1,
BOTH_PREDICTED = 2,
NEW_PLUS_NON_INTRA = 3,
BOTH_NEW = 4,
INTRA_PLUS_NON_INTRA = 5,
BOTH_INTRA = 6,
INVALID_CASE = 9
} motion_vector_context;
// This is used to figure out a context for the ref blocks. The code flattens
// an array that would have 3 possible counts (0, 1 & 2) for 3 choices by
// adding 9 for each intra block, 3 for each zero mv and 1 for each new
// motion vector. This single number is then converted into a context
// with a single lookup ( counter_to_context ).
static const int mode_2_counter[MB_MODE_COUNT] = {
9, // DC_PRED
9, // V_PRED
9, // H_PRED
9, // D45_PRED
9, // D135_PRED
9, // D117_PRED
9, // D153_PRED
9, // D207_PRED
9, // D63_PRED
9, // TM_PRED
0, // NEARESTMV
0, // NEARMV
3, // ZEROMV
1, // NEWMV
};
// There are 3^3 different combinations of 3 counts that can be either 0,1 or
// 2. However the actual count can never be greater than 2 so the highest
// counter we need is 18. 9 is an invalid counter that's never used.
static const int counter_to_context[19] = {
BOTH_PREDICTED, // 0
NEW_PLUS_NON_INTRA, // 1
BOTH_NEW, // 2
ZERO_PLUS_PREDICTED, // 3
NEW_PLUS_NON_INTRA, // 4
INVALID_CASE, // 5
BOTH_ZERO, // 6
INVALID_CASE, // 7
INVALID_CASE, // 8
INTRA_PLUS_NON_INTRA, // 9
INTRA_PLUS_NON_INTRA, // 10
INVALID_CASE, // 11
INTRA_PLUS_NON_INTRA, // 12
INVALID_CASE, // 13
INVALID_CASE, // 14
INVALID_CASE, // 15
INVALID_CASE, // 16
INVALID_CASE, // 17
BOTH_INTRA // 18
};
static const POSITION mv_ref_blocks[BLOCK_SIZES][MVREF_NEIGHBOURS] = {
// 4X4
{{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
// 4X8
{{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
// 8X4
{{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
// 8X8
{{-1, 0}, {0, -1}, {-1, -1}, {-2, 0}, {0, -2}, {-2, -1}, {-1, -2}, {-2, -2}},
// 8X16
{{0, -1}, {-1, 0}, {1, -1}, {-1, -1}, {0, -2}, {-2, 0}, {-2, -1}, {-1, -2}},
// 16X8
{{-1, 0}, {0, -1}, {-1, 1}, {-1, -1}, {-2, 0}, {0, -2}, {-1, -2}, {-2, -1}},
// 16X16
{{-1, 0}, {0, -1}, {-1, 1}, {1, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
// 16X32
{{0, -1}, {-1, 0}, {2, -1}, {-1, -1}, {-1, 1}, {0, -3}, {-3, 0}, {-3, -3}},
// 32X16
{{-1, 0}, {0, -1}, {-1, 2}, {-1, -1}, {1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
// 32X32
{{-1, 1}, {1, -1}, {-1, 2}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-3, -3}},
// 32X64
{{0, -1}, {-1, 0}, {4, -1}, {-1, 2}, {-1, -1}, {0, -3}, {-3, 0}, {2, -1}},
// 64X32
{{-1, 0}, {0, -1}, {-1, 4}, {2, -1}, {-1, -1}, {-3, 0}, {0, -3}, {-1, 2}},
// 64X64
{{-1, 3}, {3, -1}, {-1, 4}, {4, -1}, {-1, -1}, {-1, 0}, {0, -1}, {-1, 6}}
};
static const int idx_n_column_to_subblock[4][2] = {
{1, 2},
{1, 3},
{3, 2},
{3, 3}
};
// clamp_mv_ref
#define MV_BORDER (16 << 3) // Allow 16 pels in 1/8th pel units
static INLINE void clamp_mv_ref(MV *mv, const MACROBLOCKD *xd) {
clamp_mv(mv, xd->mb_to_left_edge - MV_BORDER,
xd->mb_to_right_edge + MV_BORDER,
xd->mb_to_top_edge - MV_BORDER,
xd->mb_to_bottom_edge + MV_BORDER);
}
// This function returns either the appropriate sub block or block's mv
// on whether the block_size < 8x8 and we have check_sub_blocks set.
static INLINE int_mv get_sub_block_mv(const MODE_INFO *candidate, int which_mv,
int search_col, int block_idx) {
return block_idx >= 0 && candidate->mbmi.sb_type < BLOCK_8X8
? candidate->bmi[idx_n_column_to_subblock[block_idx][search_col == 0]]
.as_mv[which_mv]
: candidate->mbmi.mv[which_mv];
}
// Performs mv sign inversion if indicated by the reference frame combination.
static INLINE int_mv scale_mv(const MB_MODE_INFO *mbmi, int ref,
const MV_REFERENCE_FRAME this_ref_frame,
const int *ref_sign_bias) {
int_mv mv = mbmi->mv[ref];
if (ref_sign_bias[mbmi->ref_frame[ref]] != ref_sign_bias[this_ref_frame]) {
mv.as_mv.row *= -1;
mv.as_mv.col *= -1;
}
return mv;
}
// This macro is used to add a motion vector mv_ref list if it isn't
// already in the list. If it's the second motion vector it will also
// skip all additional processing and jump to done!
#define ADD_MV_REF_LIST(mv, refmv_count, mv_ref_list, Done) \
do { \
if (refmv_count) { \
if ((mv).as_int != (mv_ref_list)[0].as_int) { \
(mv_ref_list)[(refmv_count)] = (mv); \
goto Done; \
} \
} else { \
(mv_ref_list)[(refmv_count)++] = (mv); \
} \
} while (0)
// If either reference frame is different, not INTRA, and they
// are different from each other scale and add the mv to our list.
#define IF_DIFF_REF_FRAME_ADD_MV(mbmi, ref_frame, ref_sign_bias, refmv_count, \
mv_ref_list, Done) \
do { \
if (is_inter_block(mbmi)) { \
if ((mbmi)->ref_frame[0] != ref_frame) \
ADD_MV_REF_LIST(scale_mv((mbmi), 0, ref_frame, ref_sign_bias), \
refmv_count, mv_ref_list, Done); \
if (has_second_ref(mbmi) && \
(mbmi)->ref_frame[1] != ref_frame && \
(mbmi)->mv[1].as_int != (mbmi)->mv[0].as_int) \
ADD_MV_REF_LIST(scale_mv((mbmi), 1, ref_frame, ref_sign_bias), \
refmv_count, mv_ref_list, Done); \
} \
} while (0)
// Checks that the given mi_row, mi_col and search point
// are inside the borders of the tile.
static INLINE int is_inside(const TileInfo *const tile,
int mi_col, int mi_row, int mi_rows,
const POSITION *mi_pos) {
return !(mi_row + mi_pos->row < 0 ||
mi_col + mi_pos->col < tile->mi_col_start ||
mi_row + mi_pos->row >= mi_rows ||
mi_col + mi_pos->col >= tile->mi_col_end);
}
// TODO(jingning): this mv clamping function should be block size dependent.
static INLINE void clamp_mv2(MV *mv, const MACROBLOCKD *xd) {
clamp_mv(mv, xd->mb_to_left_edge - LEFT_TOP_MARGIN,
xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN,
xd->mb_to_top_edge - LEFT_TOP_MARGIN,
xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN);
}
typedef void (*find_mv_refs_sync)(void *const data, int mi_row);
void vp10_find_mv_refs(const VP9_COMMON *cm, const MACROBLOCKD *xd,
MODE_INFO *mi, MV_REFERENCE_FRAME ref_frame,
int_mv *mv_ref_list, int mi_row, int mi_col,
find_mv_refs_sync sync, void *const data,
uint8_t *mode_context);
// check a list of motion vectors by sad score using a number rows of pixels
// above and a number cols of pixels in the left to select the one with best
// score to use as ref motion vector
void vp10_find_best_ref_mvs(MACROBLOCKD *xd, int allow_hp,
int_mv *mvlist, int_mv *nearest_mv, int_mv *near_mv);
void vp10_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
int block, int ref, int mi_row, int mi_col,
int_mv *nearest_mv, int_mv *near_mv,
uint8_t *mode_context);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_MVREF_COMMON_H_

470
vp10/common/vp9_onyxc_int.h Normal file
View File

@ -0,0 +1,470 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_ONYXC_INT_H_
#define VP9_COMMON_VP9_ONYXC_INT_H_
#include "./vpx_config.h"
#include "vpx/internal/vpx_codec_internal.h"
#include "vpx_util/vpx_thread.h"
#include "./vp10_rtcd.h"
#include "vp10/common/vp9_alloccommon.h"
#include "vp10/common/vp9_loopfilter.h"
#include "vp10/common/vp9_entropymv.h"
#include "vp10/common/vp9_entropy.h"
#include "vp10/common/vp9_entropymode.h"
#include "vp10/common/vp9_frame_buffers.h"
#include "vp10/common/vp9_quant_common.h"
#include "vp10/common/vp9_tile_common.h"
#if CONFIG_VP9_POSTPROC
#include "vp10/common/vp9_postproc.h"
#endif
#ifdef __cplusplus
extern "C" {
#endif
#define REFS_PER_FRAME 3
#define REF_FRAMES_LOG2 3
#define REF_FRAMES (1 << REF_FRAMES_LOG2)
// 4 scratch frames for the new frames to support a maximum of 4 cores decoding
// in parallel, 3 for scaled references on the encoder.
// TODO(hkuang): Add ondemand frame buffers instead of hardcoding the number
// of framebuffers.
// TODO(jkoleszar): These 3 extra references could probably come from the
// normal reference pool.
#define FRAME_BUFFERS (REF_FRAMES + 7)
#define FRAME_CONTEXTS_LOG2 2
#define FRAME_CONTEXTS (1 << FRAME_CONTEXTS_LOG2)
#define NUM_PING_PONG_BUFFERS 2
typedef enum {
SINGLE_REFERENCE = 0,
COMPOUND_REFERENCE = 1,
REFERENCE_MODE_SELECT = 2,
REFERENCE_MODES = 3,
} REFERENCE_MODE;
typedef struct {
int_mv mv[2];
MV_REFERENCE_FRAME ref_frame[2];
} MV_REF;
typedef struct {
int ref_count;
MV_REF *mvs;
int mi_rows;
int mi_cols;
vpx_codec_frame_buffer_t raw_frame_buffer;
YV12_BUFFER_CONFIG buf;
// The Following variables will only be used in frame parallel decode.
// frame_worker_owner indicates which FrameWorker owns this buffer. NULL means
// that no FrameWorker owns, or is decoding, this buffer.
VPxWorker *frame_worker_owner;
// row and col indicate which position frame has been decoded to in real
// pixel unit. They are reset to -1 when decoding begins and set to INT_MAX
// when the frame is fully decoded.
int row;
int col;
} RefCntBuffer;
typedef struct BufferPool {
// Protect BufferPool from being accessed by several FrameWorkers at
// the same time during frame parallel decode.
// TODO(hkuang): Try to use atomic variable instead of locking the whole pool.
#if CONFIG_MULTITHREAD
pthread_mutex_t pool_mutex;
#endif
// Private data associated with the frame buffer callbacks.
void *cb_priv;
vpx_get_frame_buffer_cb_fn_t get_fb_cb;
vpx_release_frame_buffer_cb_fn_t release_fb_cb;
RefCntBuffer frame_bufs[FRAME_BUFFERS];
// Frame buffers allocated internally by the codec.
InternalFrameBufferList int_frame_buffers;
} BufferPool;
typedef struct VP9Common {
struct vpx_internal_error_info error;
vpx_color_space_t color_space;
int width;
int height;
int display_width;
int display_height;
int last_width;
int last_height;
// TODO(jkoleszar): this implies chroma ss right now, but could vary per
// plane. Revisit as part of the future change to YV12_BUFFER_CONFIG to
// support additional planes.
int subsampling_x;
int subsampling_y;
#if CONFIG_VP9_HIGHBITDEPTH
int use_highbitdepth; // Marks if we need to use 16bit frame buffers.
#endif
YV12_BUFFER_CONFIG *frame_to_show;
RefCntBuffer *prev_frame;
// TODO(hkuang): Combine this with cur_buf in macroblockd.
RefCntBuffer *cur_frame;
int ref_frame_map[REF_FRAMES]; /* maps fb_idx to reference slot */
// Prepare ref_frame_map for the next frame.
// Only used in frame parallel decode.
int next_ref_frame_map[REF_FRAMES];
// TODO(jkoleszar): could expand active_ref_idx to 4, with 0 as intra, and
// roll new_fb_idx into it.
// Each frame can reference REFS_PER_FRAME buffers
RefBuffer frame_refs[REFS_PER_FRAME];
int new_fb_idx;
#if CONFIG_VP9_POSTPROC
YV12_BUFFER_CONFIG post_proc_buffer;
YV12_BUFFER_CONFIG post_proc_buffer_int;
#endif
FRAME_TYPE last_frame_type; /* last frame's frame type for motion search.*/
FRAME_TYPE frame_type;
int show_frame;
int last_show_frame;
int show_existing_frame;
// Flag signaling that the frame is encoded using only INTRA modes.
uint8_t intra_only;
uint8_t last_intra_only;
int allow_high_precision_mv;
// Flag signaling that the frame context should be reset to default values.
// 0 or 1 implies don't reset, 2 reset just the context specified in the
// frame header, 3 reset all contexts.
int reset_frame_context;
// MBs, mb_rows/cols is in 16-pixel units; mi_rows/cols is in
// MODE_INFO (8-pixel) units.
int MBs;
int mb_rows, mi_rows;
int mb_cols, mi_cols;
int mi_stride;
/* profile settings */
TX_MODE tx_mode;
int base_qindex;
int y_dc_delta_q;
int uv_dc_delta_q;
int uv_ac_delta_q;
int16_t y_dequant[MAX_SEGMENTS][2];
int16_t uv_dequant[MAX_SEGMENTS][2];
/* We allocate a MODE_INFO struct for each macroblock, together with
an extra row on top and column on the left to simplify prediction. */
int mi_alloc_size;
MODE_INFO *mip; /* Base of allocated array */
MODE_INFO *mi; /* Corresponds to upper left visible macroblock */
// TODO(agrange): Move prev_mi into encoder structure.
// prev_mip and prev_mi will only be allocated in VP9 encoder.
MODE_INFO *prev_mip; /* MODE_INFO array 'mip' from last decoded frame */
MODE_INFO *prev_mi; /* 'mi' from last frame (points into prev_mip) */
// Separate mi functions between encoder and decoder.
int (*alloc_mi)(struct VP9Common *cm, int mi_size);
void (*free_mi)(struct VP9Common *cm);
void (*setup_mi)(struct VP9Common *cm);
// Grid of pointers to 8x8 MODE_INFO structs. Any 8x8 not in the visible
// area will be NULL.
MODE_INFO **mi_grid_base;
MODE_INFO **mi_grid_visible;
MODE_INFO **prev_mi_grid_base;
MODE_INFO **prev_mi_grid_visible;
// Whether to use previous frame's motion vectors for prediction.
int use_prev_frame_mvs;
// Persistent mb segment id map used in prediction.
int seg_map_idx;
int prev_seg_map_idx;
uint8_t *seg_map_array[NUM_PING_PONG_BUFFERS];
uint8_t *last_frame_seg_map;
uint8_t *current_frame_seg_map;
int seg_map_alloc_size;
INTERP_FILTER interp_filter;
loop_filter_info_n lf_info;
int refresh_frame_context; /* Two state 0 = NO, 1 = YES */
int ref_frame_sign_bias[MAX_REF_FRAMES]; /* Two state 0, 1 */
struct loopfilter lf;
struct segmentation seg;
// TODO(hkuang): Remove this as it is the same as frame_parallel_decode
// in pbi.
int frame_parallel_decode; // frame-based threading.
// Context probabilities for reference frame prediction
MV_REFERENCE_FRAME comp_fixed_ref;
MV_REFERENCE_FRAME comp_var_ref[2];
REFERENCE_MODE reference_mode;
FRAME_CONTEXT *fc; /* this frame entropy */
FRAME_CONTEXT *frame_contexts; // FRAME_CONTEXTS
unsigned int frame_context_idx; /* Context to use/update */
FRAME_COUNTS counts;
unsigned int current_video_frame;
BITSTREAM_PROFILE profile;
// VPX_BITS_8 in profile 0 or 1, VPX_BITS_10 or VPX_BITS_12 in profile 2 or 3.
vpx_bit_depth_t bit_depth;
vpx_bit_depth_t dequant_bit_depth; // bit_depth of current dequantizer
#if CONFIG_VP9_POSTPROC
struct postproc_state postproc_state;
#endif
int error_resilient_mode;
int frame_parallel_decoding_mode;
int log2_tile_cols, log2_tile_rows;
int byte_alignment;
int skip_loop_filter;
// Private data associated with the frame buffer callbacks.
void *cb_priv;
vpx_get_frame_buffer_cb_fn_t get_fb_cb;
vpx_release_frame_buffer_cb_fn_t release_fb_cb;
// Handles memory for the codec.
InternalFrameBufferList int_frame_buffers;
// External BufferPool passed from outside.
BufferPool *buffer_pool;
PARTITION_CONTEXT *above_seg_context;
ENTROPY_CONTEXT *above_context;
int above_context_alloc_cols;
} VP9_COMMON;
// TODO(hkuang): Don't need to lock the whole pool after implementing atomic
// frame reference count.
static void lock_buffer_pool(BufferPool *const pool) {
#if CONFIG_MULTITHREAD
pthread_mutex_lock(&pool->pool_mutex);
#else
(void)pool;
#endif
}
static void unlock_buffer_pool(BufferPool *const pool) {
#if CONFIG_MULTITHREAD
pthread_mutex_unlock(&pool->pool_mutex);
#else
(void)pool;
#endif
}
static INLINE YV12_BUFFER_CONFIG *get_ref_frame(VP9_COMMON *cm, int index) {
if (index < 0 || index >= REF_FRAMES)
return NULL;
if (cm->ref_frame_map[index] < 0)
return NULL;
assert(cm->ref_frame_map[index] < FRAME_BUFFERS);
return &cm->buffer_pool->frame_bufs[cm->ref_frame_map[index]].buf;
}
static INLINE YV12_BUFFER_CONFIG *get_frame_new_buffer(VP9_COMMON *cm) {
return &cm->buffer_pool->frame_bufs[cm->new_fb_idx].buf;
}
static INLINE int get_free_fb(VP9_COMMON *cm) {
RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
int i;
lock_buffer_pool(cm->buffer_pool);
for (i = 0; i < FRAME_BUFFERS; ++i)
if (frame_bufs[i].ref_count == 0)
break;
if (i != FRAME_BUFFERS) {
frame_bufs[i].ref_count = 1;
} else {
// Reset i to be INVALID_IDX to indicate no free buffer found.
i = INVALID_IDX;
}
unlock_buffer_pool(cm->buffer_pool);
return i;
}
static INLINE void ref_cnt_fb(RefCntBuffer *bufs, int *idx, int new_idx) {
const int ref_index = *idx;
if (ref_index >= 0 && bufs[ref_index].ref_count > 0)
bufs[ref_index].ref_count--;
*idx = new_idx;
bufs[new_idx].ref_count++;
}
static INLINE int mi_cols_aligned_to_sb(int n_mis) {
return ALIGN_POWER_OF_TWO(n_mis, MI_BLOCK_SIZE_LOG2);
}
static INLINE int frame_is_intra_only(const VP9_COMMON *const cm) {
return cm->frame_type == KEY_FRAME || cm->intra_only;
}
static INLINE void set_partition_probs(const VP9_COMMON *const cm,
MACROBLOCKD *const xd) {
xd->partition_probs =
frame_is_intra_only(cm) ?
&vp10_kf_partition_probs[0] :
(const vpx_prob (*)[PARTITION_TYPES - 1])cm->fc->partition_prob;
}
static INLINE void vp10_init_macroblockd(VP9_COMMON *cm, MACROBLOCKD *xd,
tran_low_t *dqcoeff) {
int i;
for (i = 0; i < MAX_MB_PLANE; ++i) {
xd->plane[i].dqcoeff = dqcoeff;
xd->above_context[i] = cm->above_context +
i * sizeof(*cm->above_context) * 2 * mi_cols_aligned_to_sb(cm->mi_cols);
if (xd->plane[i].plane_type == PLANE_TYPE_Y) {
memcpy(xd->plane[i].seg_dequant, cm->y_dequant, sizeof(cm->y_dequant));
} else {
memcpy(xd->plane[i].seg_dequant, cm->uv_dequant, sizeof(cm->uv_dequant));
}
xd->fc = cm->fc;
xd->frame_parallel_decoding_mode = cm->frame_parallel_decoding_mode;
}
xd->above_seg_context = cm->above_seg_context;
xd->mi_stride = cm->mi_stride;
xd->error_info = &cm->error;
set_partition_probs(cm, xd);
}
static INLINE const vpx_prob* get_partition_probs(const MACROBLOCKD *xd,
int ctx) {
return xd->partition_probs[ctx];
}
static INLINE void set_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col) {
const int above_idx = mi_col * 2;
const int left_idx = (mi_row * 2) & 15;
int i;
for (i = 0; i < MAX_MB_PLANE; ++i) {
struct macroblockd_plane *const pd = &xd->plane[i];
pd->above_context = &xd->above_context[i][above_idx >> pd->subsampling_x];
pd->left_context = &xd->left_context[i][left_idx >> pd->subsampling_y];
}
}
static INLINE int calc_mi_size(int len) {
// len is in mi units.
return len + MI_BLOCK_SIZE;
}
static INLINE void set_mi_row_col(MACROBLOCKD *xd, const TileInfo *const tile,
int mi_row, int bh,
int mi_col, int bw,
int mi_rows, int mi_cols) {
xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8);
xd->mb_to_bottom_edge = ((mi_rows - bh - mi_row) * MI_SIZE) * 8;
xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8);
xd->mb_to_right_edge = ((mi_cols - bw - mi_col) * MI_SIZE) * 8;
// Are edges available for intra prediction?
xd->up_available = (mi_row != 0);
xd->left_available = (mi_col > tile->mi_col_start);
if (xd->up_available) {
xd->above_mi = xd->mi[-xd->mi_stride];
// above_mi may be NULL in VP9 encoder's first pass.
xd->above_mbmi = xd->above_mi ? &xd->above_mi->mbmi : NULL;
} else {
xd->above_mi = NULL;
xd->above_mbmi = NULL;
}
if (xd->left_available) {
xd->left_mi = xd->mi[-1];
// left_mi may be NULL in VP9 encoder's first pass.
xd->left_mbmi = xd->left_mi ? &xd->left_mi->mbmi : NULL;
} else {
xd->left_mi = NULL;
xd->left_mbmi = NULL;
}
}
static INLINE void update_partition_context(MACROBLOCKD *xd,
int mi_row, int mi_col,
BLOCK_SIZE subsize,
BLOCK_SIZE bsize) {
PARTITION_CONTEXT *const above_ctx = xd->above_seg_context + mi_col;
PARTITION_CONTEXT *const left_ctx = xd->left_seg_context + (mi_row & MI_MASK);
// num_4x4_blocks_wide_lookup[bsize] / 2
const int bs = num_8x8_blocks_wide_lookup[bsize];
// update the partition context at the end notes. set partition bits
// of block sizes larger than the current one to be one, and partition
// bits of smaller block sizes to be zero.
memset(above_ctx, partition_context_lookup[subsize].above, bs);
memset(left_ctx, partition_context_lookup[subsize].left, bs);
}
static INLINE int partition_plane_context(const MACROBLOCKD *xd,
int mi_row, int mi_col,
BLOCK_SIZE bsize) {
const PARTITION_CONTEXT *above_ctx = xd->above_seg_context + mi_col;
const PARTITION_CONTEXT *left_ctx = xd->left_seg_context + (mi_row & MI_MASK);
const int bsl = mi_width_log2_lookup[bsize];
int above = (*above_ctx >> bsl) & 1 , left = (*left_ctx >> bsl) & 1;
assert(b_width_log2_lookup[bsize] == b_height_log2_lookup[bsize]);
assert(bsl >= 0);
return (left * 2 + above) + bsl * PARTITION_PLOFFSET;
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_ONYXC_INT_H_

745
vp10/common/vp9_postproc.c Normal file
View File

@ -0,0 +1,745 @@
/*
* Copyright (c) 2010 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 <math.h>
#include <stdlib.h>
#include <stdio.h>
#include "./vpx_config.h"
#include "./vpx_scale_rtcd.h"
#include "./vp10_rtcd.h"
#include "vpx_ports/mem.h"
#include "vpx_scale/vpx_scale.h"
#include "vpx_scale/yv12config.h"
#include "vp10/common/vp9_onyxc_int.h"
#include "vp10/common/vp9_postproc.h"
#include "vp10/common/vp9_systemdependent.h"
#include "vp10/common/vp9_textblit.h"
#if CONFIG_VP9_POSTPROC
static const short kernel5[] = {
1, 1, 4, 1, 1
};
const short vp10_rv[] = {
8, 5, 2, 2, 8, 12, 4, 9, 8, 3,
0, 3, 9, 0, 0, 0, 8, 3, 14, 4,
10, 1, 11, 14, 1, 14, 9, 6, 12, 11,
8, 6, 10, 0, 0, 8, 9, 0, 3, 14,
8, 11, 13, 4, 2, 9, 0, 3, 9, 6,
1, 2, 3, 14, 13, 1, 8, 2, 9, 7,
3, 3, 1, 13, 13, 6, 6, 5, 2, 7,
11, 9, 11, 8, 7, 3, 2, 0, 13, 13,
14, 4, 12, 5, 12, 10, 8, 10, 13, 10,
4, 14, 4, 10, 0, 8, 11, 1, 13, 7,
7, 14, 6, 14, 13, 2, 13, 5, 4, 4,
0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
0, 10, 0, 5, 13, 2, 12, 7, 11, 13,
8, 0, 4, 10, 7, 2, 7, 2, 2, 5,
3, 4, 7, 3, 3, 14, 14, 5, 9, 13,
3, 14, 3, 6, 3, 0, 11, 8, 13, 1,
13, 1, 12, 0, 10, 9, 7, 6, 2, 8,
5, 2, 13, 7, 1, 13, 14, 7, 6, 7,
9, 6, 10, 11, 7, 8, 7, 5, 14, 8,
4, 4, 0, 8, 7, 10, 0, 8, 14, 11,
3, 12, 5, 7, 14, 3, 14, 5, 2, 6,
11, 12, 12, 8, 0, 11, 13, 1, 2, 0,
5, 10, 14, 7, 8, 0, 4, 11, 0, 8,
0, 3, 10, 5, 8, 0, 11, 6, 7, 8,
10, 7, 13, 9, 2, 5, 1, 5, 10, 2,
4, 3, 5, 6, 10, 8, 9, 4, 11, 14,
3, 8, 3, 7, 8, 5, 11, 4, 12, 3,
11, 9, 14, 8, 14, 13, 4, 3, 1, 2,
14, 6, 5, 4, 4, 11, 4, 6, 2, 1,
5, 8, 8, 12, 13, 5, 14, 10, 12, 13,
0, 9, 5, 5, 11, 10, 13, 9, 10, 13,
};
static const uint8_t q_diff_thresh = 20;
static const uint8_t last_q_thresh = 170;
void vp10_post_proc_down_and_across_c(const uint8_t *src_ptr,
uint8_t *dst_ptr,
int src_pixels_per_line,
int dst_pixels_per_line,
int rows,
int cols,
int flimit) {
uint8_t const *p_src;
uint8_t *p_dst;
int row, col, i, v, kernel;
int pitch = src_pixels_per_line;
uint8_t d[8];
(void)dst_pixels_per_line;
for (row = 0; row < rows; row++) {
/* post_proc_down for one row */
p_src = src_ptr;
p_dst = dst_ptr;
for (col = 0; col < cols; col++) {
kernel = 4;
v = p_src[col];
for (i = -2; i <= 2; i++) {
if (abs(v - p_src[col + i * pitch]) > flimit)
goto down_skip_convolve;
kernel += kernel5[2 + i] * p_src[col + i * pitch];
}
v = (kernel >> 3);
down_skip_convolve:
p_dst[col] = v;
}
/* now post_proc_across */
p_src = dst_ptr;
p_dst = dst_ptr;
for (i = 0; i < 8; i++)
d[i] = p_src[i];
for (col = 0; col < cols; col++) {
kernel = 4;
v = p_src[col];
d[col & 7] = v;
for (i = -2; i <= 2; i++) {
if (abs(v - p_src[col + i]) > flimit)
goto across_skip_convolve;
kernel += kernel5[2 + i] * p_src[col + i];
}
d[col & 7] = (kernel >> 3);
across_skip_convolve:
if (col >= 2)
p_dst[col - 2] = d[(col - 2) & 7];
}
/* handle the last two pixels */
p_dst[col - 2] = d[(col - 2) & 7];
p_dst[col - 1] = d[(col - 1) & 7];
/* next row */
src_ptr += pitch;
dst_ptr += pitch;
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_post_proc_down_and_across_c(const uint16_t *src_ptr,
uint16_t *dst_ptr,
int src_pixels_per_line,
int dst_pixels_per_line,
int rows,
int cols,
int flimit) {
uint16_t const *p_src;
uint16_t *p_dst;
int row, col, i, v, kernel;
int pitch = src_pixels_per_line;
uint16_t d[8];
for (row = 0; row < rows; row++) {
// post_proc_down for one row.
p_src = src_ptr;
p_dst = dst_ptr;
for (col = 0; col < cols; col++) {
kernel = 4;
v = p_src[col];
for (i = -2; i <= 2; i++) {
if (abs(v - p_src[col + i * pitch]) > flimit)
goto down_skip_convolve;
kernel += kernel5[2 + i] * p_src[col + i * pitch];
}
v = (kernel >> 3);
down_skip_convolve:
p_dst[col] = v;
}
/* now post_proc_across */
p_src = dst_ptr;
p_dst = dst_ptr;
for (i = 0; i < 8; i++)
d[i] = p_src[i];
for (col = 0; col < cols; col++) {
kernel = 4;
v = p_src[col];
d[col & 7] = v;
for (i = -2; i <= 2; i++) {
if (abs(v - p_src[col + i]) > flimit)
goto across_skip_convolve;
kernel += kernel5[2 + i] * p_src[col + i];
}
d[col & 7] = (kernel >> 3);
across_skip_convolve:
if (col >= 2)
p_dst[col - 2] = d[(col - 2) & 7];
}
/* handle the last two pixels */
p_dst[col - 2] = d[(col - 2) & 7];
p_dst[col - 1] = d[(col - 1) & 7];
/* next row */
src_ptr += pitch;
dst_ptr += dst_pixels_per_line;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
static int q2mbl(int x) {
if (x < 20) x = 20;
x = 50 + (x - 50) * 10 / 8;
return x * x / 3;
}
void vp10_mbpost_proc_across_ip_c(uint8_t *src, int pitch,
int rows, int cols, int flimit) {
int r, c, i;
uint8_t *s = src;
uint8_t d[16];
for (r = 0; r < rows; r++) {
int sumsq = 0;
int sum = 0;
for (i = -8; i <= 6; i++) {
sumsq += s[i] * s[i];
sum += s[i];
d[i + 8] = 0;
}
for (c = 0; c < cols + 8; c++) {
int x = s[c + 7] - s[c - 8];
int y = s[c + 7] + s[c - 8];
sum += x;
sumsq += x * y;
d[c & 15] = s[c];
if (sumsq * 15 - sum * sum < flimit) {
d[c & 15] = (8 + sum + s[c]) >> 4;
}
s[c - 8] = d[(c - 8) & 15];
}
s += pitch;
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_mbpost_proc_across_ip_c(uint16_t *src, int pitch,
int rows, int cols, int flimit) {
int r, c, i;
uint16_t *s = src;
uint16_t d[16];
for (r = 0; r < rows; r++) {
int sumsq = 0;
int sum = 0;
for (i = -8; i <= 6; i++) {
sumsq += s[i] * s[i];
sum += s[i];
d[i + 8] = 0;
}
for (c = 0; c < cols + 8; c++) {
int x = s[c + 7] - s[c - 8];
int y = s[c + 7] + s[c - 8];
sum += x;
sumsq += x * y;
d[c & 15] = s[c];
if (sumsq * 15 - sum * sum < flimit) {
d[c & 15] = (8 + sum + s[c]) >> 4;
}
s[c - 8] = d[(c - 8) & 15];
}
s += pitch;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
void vp10_mbpost_proc_down_c(uint8_t *dst, int pitch,
int rows, int cols, int flimit) {
int r, c, i;
const short *rv3 = &vp10_rv[63 & rand()]; // NOLINT
for (c = 0; c < cols; c++) {
uint8_t *s = &dst[c];
int sumsq = 0;
int sum = 0;
uint8_t d[16];
const short *rv2 = rv3 + ((c * 17) & 127);
for (i = -8; i <= 6; i++) {
sumsq += s[i * pitch] * s[i * pitch];
sum += s[i * pitch];
}
for (r = 0; r < rows + 8; r++) {
sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
sum += s[7 * pitch] - s[-8 * pitch];
d[r & 15] = s[0];
if (sumsq * 15 - sum * sum < flimit) {
d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
}
s[-8 * pitch] = d[(r - 8) & 15];
s += pitch;
}
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_mbpost_proc_down_c(uint16_t *dst, int pitch,
int rows, int cols, int flimit) {
int r, c, i;
const int16_t *rv3 = &vp10_rv[63 & rand()]; // NOLINT
for (c = 0; c < cols; c++) {
uint16_t *s = &dst[c];
int sumsq = 0;
int sum = 0;
uint16_t d[16];
const int16_t *rv2 = rv3 + ((c * 17) & 127);
for (i = -8; i <= 6; i++) {
sumsq += s[i * pitch] * s[i * pitch];
sum += s[i * pitch];
}
for (r = 0; r < rows + 8; r++) {
sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
sum += s[7 * pitch] - s[-8 * pitch];
d[r & 15] = s[0];
if (sumsq * 15 - sum * sum < flimit) {
d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
}
s[-8 * pitch] = d[(r - 8) & 15];
s += pitch;
}
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG *source,
YV12_BUFFER_CONFIG *post,
int q,
int low_var_thresh,
int flag) {
double level = 6.0e-05 * q * q * q - .0067 * q * q + .306 * q + .0065;
int ppl = (int)(level + .5);
(void) low_var_thresh;
(void) flag;
#if CONFIG_VP9_HIGHBITDEPTH
if (source->flags & YV12_FLAG_HIGHBITDEPTH) {
vp10_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->y_buffer),
CONVERT_TO_SHORTPTR(post->y_buffer),
source->y_stride, post->y_stride,
source->y_height, source->y_width,
ppl);
vp10_highbd_mbpost_proc_across_ip(CONVERT_TO_SHORTPTR(post->y_buffer),
post->y_stride, post->y_height,
post->y_width, q2mbl(q));
vp10_highbd_mbpost_proc_down(CONVERT_TO_SHORTPTR(post->y_buffer),
post->y_stride, post->y_height,
post->y_width, q2mbl(q));
vp10_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->u_buffer),
CONVERT_TO_SHORTPTR(post->u_buffer),
source->uv_stride, post->uv_stride,
source->uv_height, source->uv_width,
ppl);
vp10_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->v_buffer),
CONVERT_TO_SHORTPTR(post->v_buffer),
source->uv_stride, post->uv_stride,
source->uv_height, source->uv_width,
ppl);
} else {
vp10_post_proc_down_and_across(source->y_buffer, post->y_buffer,
source->y_stride, post->y_stride,
source->y_height, source->y_width, ppl);
vp10_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
post->y_width, q2mbl(q));
vp10_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
post->y_width, q2mbl(q));
vp10_post_proc_down_and_across(source->u_buffer, post->u_buffer,
source->uv_stride, post->uv_stride,
source->uv_height, source->uv_width, ppl);
vp10_post_proc_down_and_across(source->v_buffer, post->v_buffer,
source->uv_stride, post->uv_stride,
source->uv_height, source->uv_width, ppl);
}
#else
vp10_post_proc_down_and_across(source->y_buffer, post->y_buffer,
source->y_stride, post->y_stride,
source->y_height, source->y_width, ppl);
vp10_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
post->y_width, q2mbl(q));
vp10_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
post->y_width, q2mbl(q));
vp10_post_proc_down_and_across(source->u_buffer, post->u_buffer,
source->uv_stride, post->uv_stride,
source->uv_height, source->uv_width, ppl);
vp10_post_proc_down_and_across(source->v_buffer, post->v_buffer,
source->uv_stride, post->uv_stride,
source->uv_height, source->uv_width, ppl);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
void vp10_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
int q) {
const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
+ 0.0065 + 0.5);
int i;
const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
const int src_widths[3] = {src->y_width, src->uv_width, src->uv_width};
const int src_heights[3] = {src->y_height, src->uv_height, src->uv_height};
uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
for (i = 0; i < MAX_MB_PLANE; ++i) {
#if CONFIG_VP9_HIGHBITDEPTH
assert((src->flags & YV12_FLAG_HIGHBITDEPTH) ==
(dst->flags & YV12_FLAG_HIGHBITDEPTH));
if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
vp10_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(srcs[i]),
CONVERT_TO_SHORTPTR(dsts[i]),
src_strides[i], dst_strides[i],
src_heights[i], src_widths[i], ppl);
} else {
vp10_post_proc_down_and_across(srcs[i], dsts[i],
src_strides[i], dst_strides[i],
src_heights[i], src_widths[i], ppl);
}
#else
vp10_post_proc_down_and_across(srcs[i], dsts[i],
src_strides[i], dst_strides[i],
src_heights[i], src_widths[i], ppl);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}
void vp10_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
int q) {
const int ppl = (int)(6.0e-05 * q * q * q - 0.0067 * q * q + 0.306 * q
+ 0.0065 + 0.5);
int i;
const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
const int src_widths[3] = {src->y_width, src->uv_width, src->uv_width};
const int src_heights[3] = {src->y_height, src->uv_height, src->uv_height};
uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
for (i = 0; i < MAX_MB_PLANE; ++i) {
const int src_stride = src_strides[i];
const int src_width = src_widths[i] - 4;
const int src_height = src_heights[i] - 4;
const int dst_stride = dst_strides[i];
#if CONFIG_VP9_HIGHBITDEPTH
assert((src->flags & YV12_FLAG_HIGHBITDEPTH) ==
(dst->flags & YV12_FLAG_HIGHBITDEPTH));
if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
const uint16_t *const src_plane = CONVERT_TO_SHORTPTR(
srcs[i] + 2 * src_stride + 2);
uint16_t *const dst_plane = CONVERT_TO_SHORTPTR(
dsts[i] + 2 * dst_stride + 2);
vp10_highbd_post_proc_down_and_across(src_plane, dst_plane, src_stride,
dst_stride, src_height, src_width,
ppl);
} else {
const uint8_t *const src_plane = srcs[i] + 2 * src_stride + 2;
uint8_t *const dst_plane = dsts[i] + 2 * dst_stride + 2;
vp10_post_proc_down_and_across(src_plane, dst_plane, src_stride,
dst_stride, src_height, src_width, ppl);
}
#else
const uint8_t *const src_plane = srcs[i] + 2 * src_stride + 2;
uint8_t *const dst_plane = dsts[i] + 2 * dst_stride + 2;
vp10_post_proc_down_and_across(src_plane, dst_plane, src_stride, dst_stride,
src_height, src_width, ppl);
#endif
}
}
static double gaussian(double sigma, double mu, double x) {
return 1 / (sigma * sqrt(2.0 * 3.14159265)) *
(exp(-(x - mu) * (x - mu) / (2 * sigma * sigma)));
}
static void fillrd(struct postproc_state *state, int q, int a) {
char char_dist[300];
double sigma;
int ai = a, qi = q, i;
vpx_clear_system_state();
sigma = ai + .5 + .6 * (63 - qi) / 63.0;
/* set up a lookup table of 256 entries that matches
* a gaussian distribution with sigma determined by q.
*/
{
int next, j;
next = 0;
for (i = -32; i < 32; i++) {
int a_i = (int)(0.5 + 256 * gaussian(sigma, 0, i));
if (a_i) {
for (j = 0; j < a_i; j++) {
char_dist[next + j] = (char) i;
}
next = next + j;
}
}
for (; next < 256; next++)
char_dist[next] = 0;
}
for (i = 0; i < 3072; i++) {
state->noise[i] = char_dist[rand() & 0xff]; // NOLINT
}
for (i = 0; i < 16; i++) {
state->blackclamp[i] = -char_dist[0];
state->whiteclamp[i] = -char_dist[0];
state->bothclamp[i] = -2 * char_dist[0];
}
state->last_q = q;
state->last_noise = a;
}
void vp10_plane_add_noise_c(uint8_t *start, char *noise,
char blackclamp[16],
char whiteclamp[16],
char bothclamp[16],
unsigned int width, unsigned int height, int pitch) {
unsigned int i, j;
// TODO(jbb): why does simd code use both but c doesn't, normalize and
// fix..
(void) bothclamp;
for (i = 0; i < height; i++) {
uint8_t *pos = start + i * pitch;
char *ref = (char *)(noise + (rand() & 0xff)); // NOLINT
for (j = 0; j < width; j++) {
if (pos[j] < blackclamp[0])
pos[j] = blackclamp[0];
if (pos[j] > 255 + whiteclamp[0])
pos[j] = 255 + whiteclamp[0];
pos[j] += ref[j];
}
}
}
static void swap_mi_and_prev_mi(VP9_COMMON *cm) {
// Current mip will be the prev_mip for the next frame.
MODE_INFO *temp = cm->postproc_state.prev_mip;
cm->postproc_state.prev_mip = cm->mip;
cm->mip = temp;
// Update the upper left visible macroblock ptrs.
cm->mi = cm->mip + cm->mi_stride + 1;
cm->postproc_state.prev_mi = cm->postproc_state.prev_mip + cm->mi_stride + 1;
}
int vp10_post_proc_frame(struct VP9Common *cm,
YV12_BUFFER_CONFIG *dest, vp10_ppflags_t *ppflags) {
const int q = MIN(105, cm->lf.filter_level * 2);
const int flags = ppflags->post_proc_flag;
YV12_BUFFER_CONFIG *const ppbuf = &cm->post_proc_buffer;
struct postproc_state *const ppstate = &cm->postproc_state;
if (!cm->frame_to_show)
return -1;
if (!flags) {
*dest = *cm->frame_to_show;
return 0;
}
vpx_clear_system_state();
// Alloc memory for prev_mip in the first frame.
if (cm->current_video_frame == 1) {
cm->postproc_state.last_base_qindex = cm->base_qindex;
cm->postproc_state.last_frame_valid = 1;
ppstate->prev_mip = vpx_calloc(cm->mi_alloc_size, sizeof(*cm->mip));
if (!ppstate->prev_mip) {
return 1;
}
ppstate->prev_mi = ppstate->prev_mip + cm->mi_stride + 1;
memset(ppstate->prev_mip, 0,
cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mip));
}
// Allocate post_proc_buffer_int if needed.
if ((flags & VP9D_MFQE) && !cm->post_proc_buffer_int.buffer_alloc) {
if ((flags & VP9D_DEMACROBLOCK) || (flags & VP9D_DEBLOCK)) {
const int width = ALIGN_POWER_OF_TWO(cm->width, 4);
const int height = ALIGN_POWER_OF_TWO(cm->height, 4);
if (vp9_alloc_frame_buffer(&cm->post_proc_buffer_int, width, height,
cm->subsampling_x, cm->subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth,
#endif // CONFIG_VP9_HIGHBITDEPTH
VP9_ENC_BORDER_IN_PIXELS,
cm->byte_alignment) < 0) {
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate MFQE framebuffer");
}
// Ensure that postproc is set to all 0s so that post proc
// doesn't pull random data in from edge.
memset(cm->post_proc_buffer_int.buffer_alloc, 128,
cm->post_proc_buffer.frame_size);
}
}
if (vp9_realloc_frame_buffer(&cm->post_proc_buffer, cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth,
#endif
VP9_DEC_BORDER_IN_PIXELS, cm->byte_alignment,
NULL, NULL, NULL) < 0)
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate post-processing buffer");
if ((flags & VP9D_MFQE) && cm->current_video_frame >= 2 &&
cm->postproc_state.last_frame_valid && cm->bit_depth == 8 &&
cm->postproc_state.last_base_qindex <= last_q_thresh &&
cm->base_qindex - cm->postproc_state.last_base_qindex >= q_diff_thresh) {
vp10_mfqe(cm);
// TODO(jackychen): Consider whether enable deblocking by default
// if mfqe is enabled. Need to take both the quality and the speed
// into consideration.
if ((flags & VP9D_DEMACROBLOCK) || (flags & VP9D_DEBLOCK)) {
vp8_yv12_copy_frame(ppbuf, &cm->post_proc_buffer_int);
}
if ((flags & VP9D_DEMACROBLOCK) && cm->post_proc_buffer_int.buffer_alloc) {
deblock_and_de_macro_block(&cm->post_proc_buffer_int, ppbuf,
q + (ppflags->deblocking_level - 5) * 10,
1, 0);
} else if (flags & VP9D_DEBLOCK) {
vp10_deblock(&cm->post_proc_buffer_int, ppbuf, q);
} else {
vp8_yv12_copy_frame(&cm->post_proc_buffer_int, ppbuf);
}
} else if (flags & VP9D_DEMACROBLOCK) {
deblock_and_de_macro_block(cm->frame_to_show, ppbuf,
q + (ppflags->deblocking_level - 5) * 10, 1, 0);
} else if (flags & VP9D_DEBLOCK) {
vp10_deblock(cm->frame_to_show, ppbuf, q);
} else {
vp8_yv12_copy_frame(cm->frame_to_show, ppbuf);
}
cm->postproc_state.last_base_qindex = cm->base_qindex;
cm->postproc_state.last_frame_valid = 1;
if (flags & VP9D_ADDNOISE) {
const int noise_level = ppflags->noise_level;
if (ppstate->last_q != q ||
ppstate->last_noise != noise_level) {
fillrd(ppstate, 63 - q, noise_level);
}
vp10_plane_add_noise(ppbuf->y_buffer, ppstate->noise, ppstate->blackclamp,
ppstate->whiteclamp, ppstate->bothclamp,
ppbuf->y_width, ppbuf->y_height, ppbuf->y_stride);
}
*dest = *ppbuf;
/* handle problem with extending borders */
dest->y_width = cm->width;
dest->y_height = cm->height;
dest->uv_width = dest->y_width >> cm->subsampling_x;
dest->uv_height = dest->y_height >> cm->subsampling_y;
swap_mi_and_prev_mi(cm);
return 0;
}
#endif // CONFIG_VP9_POSTPROC

53
vp10/common/vp9_postproc.h Normal file
View File

@ -0,0 +1,53 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_POSTPROC_H_
#define VP9_COMMON_VP9_POSTPROC_H_
#include "vpx_ports/mem.h"
#include "vpx_scale/yv12config.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_mfqe.h"
#include "vp10/common/vp9_ppflags.h"
#ifdef __cplusplus
extern "C" {
#endif
struct postproc_state {
int last_q;
int last_noise;
char noise[3072];
int last_base_qindex;
int last_frame_valid;
MODE_INFO *prev_mip;
MODE_INFO *prev_mi;
DECLARE_ALIGNED(16, char, blackclamp[16]);
DECLARE_ALIGNED(16, char, whiteclamp[16]);
DECLARE_ALIGNED(16, char, bothclamp[16]);
};
struct VP9Common;
#define MFQE_PRECISION 4
int vp10_post_proc_frame(struct VP9Common *cm,
YV12_BUFFER_CONFIG *dest, vp10_ppflags_t *flags);
void vp10_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst, int q);
void vp10_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst, int q);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_POSTPROC_H_

43
vp10/common/vp9_ppflags.h Normal file
View File

@ -0,0 +1,43 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_PPFLAGS_H_
#define VP9_COMMON_VP9_PPFLAGS_H_
#ifdef __cplusplus
extern "C" {
#endif
enum {
VP9D_NOFILTERING = 0,
VP9D_DEBLOCK = 1 << 0,
VP9D_DEMACROBLOCK = 1 << 1,
VP9D_ADDNOISE = 1 << 2,
VP9D_DEBUG_TXT_FRAME_INFO = 1 << 3,
VP9D_DEBUG_TXT_MBLK_MODES = 1 << 4,
VP9D_DEBUG_TXT_DC_DIFF = 1 << 5,
VP9D_DEBUG_TXT_RATE_INFO = 1 << 6,
VP9D_DEBUG_DRAW_MV = 1 << 7,
VP9D_DEBUG_CLR_BLK_MODES = 1 << 8,
VP9D_DEBUG_CLR_FRM_REF_BLKS = 1 << 9,
VP9D_MFQE = 1 << 10
};
typedef struct {
int post_proc_flag;
int deblocking_level;
int noise_level;
} vp10_ppflags_t;
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_PPFLAGS_H_

339
vp10/common/vp9_pred_common.c Normal file
View File

@ -0,0 +1,339 @@
/*
* Copyright (c) 2012 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/common/vp9_common.h"
#include "vp10/common/vp9_pred_common.h"
#include "vp10/common/vp9_seg_common.h"
// Returns a context number for the given MB prediction signal
int vp10_get_pred_context_switchable_interp(const MACROBLOCKD *xd) {
// Note:
// The mode info data structure has a one element border above and to the
// left of the entries correpsonding to real macroblocks.
// The prediction flags in these dummy entries are initialised to 0.
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int left_type = xd->left_available && is_inter_block(left_mbmi) ?
left_mbmi->interp_filter : SWITCHABLE_FILTERS;
const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
const int above_type = xd->up_available && is_inter_block(above_mbmi) ?
above_mbmi->interp_filter : SWITCHABLE_FILTERS;
if (left_type == above_type)
return left_type;
else if (left_type == SWITCHABLE_FILTERS && above_type != SWITCHABLE_FILTERS)
return above_type;
else if (left_type != SWITCHABLE_FILTERS && above_type == SWITCHABLE_FILTERS)
return left_type;
else
return SWITCHABLE_FILTERS;
}
// The mode info data structure has a one element border above and to the
// left of the entries corresponding to real macroblocks.
// The prediction flags in these dummy entries are initialized to 0.
// 0 - inter/inter, inter/--, --/inter, --/--
// 1 - intra/inter, inter/intra
// 2 - intra/--, --/intra
// 3 - intra/intra
int vp10_get_intra_inter_context(const MACROBLOCKD *xd) {
const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int has_above = xd->up_available;
const int has_left = xd->left_available;
if (has_above && has_left) { // both edges available
const int above_intra = !is_inter_block(above_mbmi);
const int left_intra = !is_inter_block(left_mbmi);
return left_intra && above_intra ? 3
: left_intra || above_intra;
} else if (has_above || has_left) { // one edge available
return 2 * !is_inter_block(has_above ? above_mbmi : left_mbmi);
} else {
return 0;
}
}
int vp10_get_reference_mode_context(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
int ctx;
const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int has_above = xd->up_available;
const int has_left = xd->left_available;
// Note:
// The mode info data structure has a one element border above and to the
// left of the entries correpsonding to real macroblocks.
// The prediction flags in these dummy entries are initialised to 0.
if (has_above && has_left) { // both edges available
if (!has_second_ref(above_mbmi) && !has_second_ref(left_mbmi))
// neither edge uses comp pred (0/1)
ctx = (above_mbmi->ref_frame[0] == cm->comp_fixed_ref) ^
(left_mbmi->ref_frame[0] == cm->comp_fixed_ref);
else if (!has_second_ref(above_mbmi))
// one of two edges uses comp pred (2/3)
ctx = 2 + (above_mbmi->ref_frame[0] == cm->comp_fixed_ref ||
!is_inter_block(above_mbmi));
else if (!has_second_ref(left_mbmi))
// one of two edges uses comp pred (2/3)
ctx = 2 + (left_mbmi->ref_frame[0] == cm->comp_fixed_ref ||
!is_inter_block(left_mbmi));
else // both edges use comp pred (4)
ctx = 4;
} else if (has_above || has_left) { // one edge available
const MB_MODE_INFO *edge_mbmi = has_above ? above_mbmi : left_mbmi;
if (!has_second_ref(edge_mbmi))
// edge does not use comp pred (0/1)
ctx = edge_mbmi->ref_frame[0] == cm->comp_fixed_ref;
else
// edge uses comp pred (3)
ctx = 3;
} else { // no edges available (1)
ctx = 1;
}
assert(ctx >= 0 && ctx < COMP_INTER_CONTEXTS);
return ctx;
}
// Returns a context number for the given MB prediction signal
int vp10_get_pred_context_comp_ref_p(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
int pred_context;
const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int above_in_image = xd->up_available;
const int left_in_image = xd->left_available;
// Note:
// The mode info data structure has a one element border above and to the
// left of the entries correpsonding to real macroblocks.
// The prediction flags in these dummy entries are initialised to 0.
const int fix_ref_idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
const int var_ref_idx = !fix_ref_idx;
if (above_in_image && left_in_image) { // both edges available
const int above_intra = !is_inter_block(above_mbmi);
const int left_intra = !is_inter_block(left_mbmi);
if (above_intra && left_intra) { // intra/intra (2)
pred_context = 2;
} else if (above_intra || left_intra) { // intra/inter
const MB_MODE_INFO *edge_mbmi = above_intra ? left_mbmi : above_mbmi;
if (!has_second_ref(edge_mbmi)) // single pred (1/3)
pred_context = 1 + 2 * (edge_mbmi->ref_frame[0] != cm->comp_var_ref[1]);
else // comp pred (1/3)
pred_context = 1 + 2 * (edge_mbmi->ref_frame[var_ref_idx]
!= cm->comp_var_ref[1]);
} else { // inter/inter
const int l_sg = !has_second_ref(left_mbmi);
const int a_sg = !has_second_ref(above_mbmi);
const MV_REFERENCE_FRAME vrfa = a_sg ? above_mbmi->ref_frame[0]
: above_mbmi->ref_frame[var_ref_idx];
const MV_REFERENCE_FRAME vrfl = l_sg ? left_mbmi->ref_frame[0]
: left_mbmi->ref_frame[var_ref_idx];
if (vrfa == vrfl && cm->comp_var_ref[1] == vrfa) {
pred_context = 0;
} else if (l_sg && a_sg) { // single/single
if ((vrfa == cm->comp_fixed_ref && vrfl == cm->comp_var_ref[0]) ||
(vrfl == cm->comp_fixed_ref && vrfa == cm->comp_var_ref[0]))
pred_context = 4;
else if (vrfa == vrfl)
pred_context = 3;
else
pred_context = 1;
} else if (l_sg || a_sg) { // single/comp
const MV_REFERENCE_FRAME vrfc = l_sg ? vrfa : vrfl;
const MV_REFERENCE_FRAME rfs = a_sg ? vrfa : vrfl;
if (vrfc == cm->comp_var_ref[1] && rfs != cm->comp_var_ref[1])
pred_context = 1;
else if (rfs == cm->comp_var_ref[1] && vrfc != cm->comp_var_ref[1])
pred_context = 2;
else
pred_context = 4;
} else if (vrfa == vrfl) { // comp/comp
pred_context = 4;
} else {
pred_context = 2;
}
}
} else if (above_in_image || left_in_image) { // one edge available
const MB_MODE_INFO *edge_mbmi = above_in_image ? above_mbmi : left_mbmi;
if (!is_inter_block(edge_mbmi)) {
pred_context = 2;
} else {
if (has_second_ref(edge_mbmi))
pred_context = 4 * (edge_mbmi->ref_frame[var_ref_idx]
!= cm->comp_var_ref[1]);
else
pred_context = 3 * (edge_mbmi->ref_frame[0] != cm->comp_var_ref[1]);
}
} else { // no edges available (2)
pred_context = 2;
}
assert(pred_context >= 0 && pred_context < REF_CONTEXTS);
return pred_context;
}
int vp10_get_pred_context_single_ref_p1(const MACROBLOCKD *xd) {
int pred_context;
const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int has_above = xd->up_available;
const int has_left = xd->left_available;
// Note:
// The mode info data structure has a one element border above and to the
// left of the entries correpsonding to real macroblocks.
// The prediction flags in these dummy entries are initialised to 0.
if (has_above && has_left) { // both edges available
const int above_intra = !is_inter_block(above_mbmi);
const int left_intra = !is_inter_block(left_mbmi);
if (above_intra && left_intra) { // intra/intra
pred_context = 2;
} else if (above_intra || left_intra) { // intra/inter or inter/intra
const MB_MODE_INFO *edge_mbmi = above_intra ? left_mbmi : above_mbmi;
if (!has_second_ref(edge_mbmi))
pred_context = 4 * (edge_mbmi->ref_frame[0] == LAST_FRAME);
else
pred_context = 1 + (edge_mbmi->ref_frame[0] == LAST_FRAME ||
edge_mbmi->ref_frame[1] == LAST_FRAME);
} else { // inter/inter
const int above_has_second = has_second_ref(above_mbmi);
const int left_has_second = has_second_ref(left_mbmi);
const MV_REFERENCE_FRAME above0 = above_mbmi->ref_frame[0];
const MV_REFERENCE_FRAME above1 = above_mbmi->ref_frame[1];
const MV_REFERENCE_FRAME left0 = left_mbmi->ref_frame[0];
const MV_REFERENCE_FRAME left1 = left_mbmi->ref_frame[1];
if (above_has_second && left_has_second) {
pred_context = 1 + (above0 == LAST_FRAME || above1 == LAST_FRAME ||
left0 == LAST_FRAME || left1 == LAST_FRAME);
} else if (above_has_second || left_has_second) {
const MV_REFERENCE_FRAME rfs = !above_has_second ? above0 : left0;
const MV_REFERENCE_FRAME crf1 = above_has_second ? above0 : left0;
const MV_REFERENCE_FRAME crf2 = above_has_second ? above1 : left1;
if (rfs == LAST_FRAME)
pred_context = 3 + (crf1 == LAST_FRAME || crf2 == LAST_FRAME);
else
pred_context = (crf1 == LAST_FRAME || crf2 == LAST_FRAME);
} else {
pred_context = 2 * (above0 == LAST_FRAME) + 2 * (left0 == LAST_FRAME);
}
}
} else if (has_above || has_left) { // one edge available
const MB_MODE_INFO *edge_mbmi = has_above ? above_mbmi : left_mbmi;
if (!is_inter_block(edge_mbmi)) { // intra
pred_context = 2;
} else { // inter
if (!has_second_ref(edge_mbmi))
pred_context = 4 * (edge_mbmi->ref_frame[0] == LAST_FRAME);
else
pred_context = 1 + (edge_mbmi->ref_frame[0] == LAST_FRAME ||
edge_mbmi->ref_frame[1] == LAST_FRAME);
}
} else { // no edges available
pred_context = 2;
}
assert(pred_context >= 0 && pred_context < REF_CONTEXTS);
return pred_context;
}
int vp10_get_pred_context_single_ref_p2(const MACROBLOCKD *xd) {
int pred_context;
const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int has_above = xd->up_available;
const int has_left = xd->left_available;
// Note:
// The mode info data structure has a one element border above and to the
// left of the entries correpsonding to real macroblocks.
// The prediction flags in these dummy entries are initialised to 0.
if (has_above && has_left) { // both edges available
const int above_intra = !is_inter_block(above_mbmi);
const int left_intra = !is_inter_block(left_mbmi);
if (above_intra && left_intra) { // intra/intra
pred_context = 2;
} else if (above_intra || left_intra) { // intra/inter or inter/intra
const MB_MODE_INFO *edge_mbmi = above_intra ? left_mbmi : above_mbmi;
if (!has_second_ref(edge_mbmi)) {
if (edge_mbmi->ref_frame[0] == LAST_FRAME)
pred_context = 3;
else
pred_context = 4 * (edge_mbmi->ref_frame[0] == GOLDEN_FRAME);
} else {
pred_context = 1 + 2 * (edge_mbmi->ref_frame[0] == GOLDEN_FRAME ||
edge_mbmi->ref_frame[1] == GOLDEN_FRAME);
}
} else { // inter/inter
const int above_has_second = has_second_ref(above_mbmi);
const int left_has_second = has_second_ref(left_mbmi);
const MV_REFERENCE_FRAME above0 = above_mbmi->ref_frame[0];
const MV_REFERENCE_FRAME above1 = above_mbmi->ref_frame[1];
const MV_REFERENCE_FRAME left0 = left_mbmi->ref_frame[0];
const MV_REFERENCE_FRAME left1 = left_mbmi->ref_frame[1];
if (above_has_second && left_has_second) {
if (above0 == left0 && above1 == left1)
pred_context = 3 * (above0 == GOLDEN_FRAME ||
above1 == GOLDEN_FRAME ||
left0 == GOLDEN_FRAME ||
left1 == GOLDEN_FRAME);
else
pred_context = 2;
} else if (above_has_second || left_has_second) {
const MV_REFERENCE_FRAME rfs = !above_has_second ? above0 : left0;
const MV_REFERENCE_FRAME crf1 = above_has_second ? above0 : left0;
const MV_REFERENCE_FRAME crf2 = above_has_second ? above1 : left1;
if (rfs == GOLDEN_FRAME)
pred_context = 3 + (crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME);
else if (rfs == ALTREF_FRAME)
pred_context = crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME;
else
pred_context = 1 + 2 * (crf1 == GOLDEN_FRAME || crf2 == GOLDEN_FRAME);
} else {
if (above0 == LAST_FRAME && left0 == LAST_FRAME) {
pred_context = 3;
} else if (above0 == LAST_FRAME || left0 == LAST_FRAME) {
const MV_REFERENCE_FRAME edge0 = (above0 == LAST_FRAME) ? left0
: above0;
pred_context = 4 * (edge0 == GOLDEN_FRAME);
} else {
pred_context = 2 * (above0 == GOLDEN_FRAME) +
2 * (left0 == GOLDEN_FRAME);
}
}
}
} else if (has_above || has_left) { // one edge available
const MB_MODE_INFO *edge_mbmi = has_above ? above_mbmi : left_mbmi;
if (!is_inter_block(edge_mbmi) ||
(edge_mbmi->ref_frame[0] == LAST_FRAME && !has_second_ref(edge_mbmi)))
pred_context = 2;
else if (!has_second_ref(edge_mbmi))
pred_context = 4 * (edge_mbmi->ref_frame[0] == GOLDEN_FRAME);
else
pred_context = 3 * (edge_mbmi->ref_frame[0] == GOLDEN_FRAME ||
edge_mbmi->ref_frame[1] == GOLDEN_FRAME);
} else { // no edges available (2)
pred_context = 2;
}
assert(pred_context >= 0 && pred_context < REF_CONTEXTS);
return pred_context;
}

170
vp10/common/vp9_pred_common.h Normal file
View File

@ -0,0 +1,170 @@
/*
* Copyright (c) 2012 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 VP9_COMMON_VP9_PRED_COMMON_H_
#define VP9_COMMON_VP9_PRED_COMMON_H_
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_onyxc_int.h"
#ifdef __cplusplus
extern "C" {
#endif
static INLINE int get_segment_id(const VP9_COMMON *cm,
const uint8_t *segment_ids,
BLOCK_SIZE bsize, int mi_row, int mi_col) {
const int mi_offset = mi_row * cm->mi_cols + mi_col;
const int bw = num_8x8_blocks_wide_lookup[bsize];
const int bh = num_8x8_blocks_high_lookup[bsize];
const int xmis = MIN(cm->mi_cols - mi_col, bw);
const int ymis = MIN(cm->mi_rows - mi_row, bh);
int x, y, segment_id = MAX_SEGMENTS;
for (y = 0; y < ymis; ++y)
for (x = 0; x < xmis; ++x)
segment_id = MIN(segment_id,
segment_ids[mi_offset + y * cm->mi_cols + x]);
assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
return segment_id;
}
static INLINE int vp10_get_pred_context_seg_id(const MACROBLOCKD *xd) {
const MODE_INFO *const above_mi = xd->above_mi;
const MODE_INFO *const left_mi = xd->left_mi;
const int above_sip = (above_mi != NULL) ?
above_mi->mbmi.seg_id_predicted : 0;
const int left_sip = (left_mi != NULL) ? left_mi->mbmi.seg_id_predicted : 0;
return above_sip + left_sip;
}
static INLINE vpx_prob vp10_get_pred_prob_seg_id(const struct segmentation *seg,
const MACROBLOCKD *xd) {
return seg->pred_probs[vp10_get_pred_context_seg_id(xd)];
}
static INLINE int vp10_get_skip_context(const MACROBLOCKD *xd) {
const MODE_INFO *const above_mi = xd->above_mi;
const MODE_INFO *const left_mi = xd->left_mi;
const int above_skip = (above_mi != NULL) ? above_mi->mbmi.skip : 0;
const int left_skip = (left_mi != NULL) ? left_mi->mbmi.skip : 0;
return above_skip + left_skip;
}
static INLINE vpx_prob vp10_get_skip_prob(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->skip_probs[vp10_get_skip_context(xd)];
}
int vp10_get_pred_context_switchable_interp(const MACROBLOCKD *xd);
int vp10_get_intra_inter_context(const MACROBLOCKD *xd);
static INLINE vpx_prob vp10_get_intra_inter_prob(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->intra_inter_prob[vp10_get_intra_inter_context(xd)];
}
int vp10_get_reference_mode_context(const VP9_COMMON *cm, const MACROBLOCKD *xd);
static INLINE vpx_prob vp10_get_reference_mode_prob(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->comp_inter_prob[vp10_get_reference_mode_context(cm, xd)];
}
int vp10_get_pred_context_comp_ref_p(const VP9_COMMON *cm,
const MACROBLOCKD *xd);
static INLINE vpx_prob vp10_get_pred_prob_comp_ref_p(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
const int pred_context = vp10_get_pred_context_comp_ref_p(cm, xd);
return cm->fc->comp_ref_prob[pred_context];
}
int vp10_get_pred_context_single_ref_p1(const MACROBLOCKD *xd);
static INLINE vpx_prob vp10_get_pred_prob_single_ref_p1(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->single_ref_prob[vp10_get_pred_context_single_ref_p1(xd)][0];
}
int vp10_get_pred_context_single_ref_p2(const MACROBLOCKD *xd);
static INLINE vpx_prob vp10_get_pred_prob_single_ref_p2(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
return cm->fc->single_ref_prob[vp10_get_pred_context_single_ref_p2(xd)][1];
}
// Returns a context number for the given MB prediction signal
// The mode info data structure has a one element border above and to the
// left of the entries corresponding to real blocks.
// The prediction flags in these dummy entries are initialized to 0.
static INLINE int get_tx_size_context(const MACROBLOCKD *xd) {
const int max_tx_size = max_txsize_lookup[xd->mi[0]->mbmi.sb_type];
const MB_MODE_INFO *const above_mbmi = xd->above_mbmi;
const MB_MODE_INFO *const left_mbmi = xd->left_mbmi;
const int has_above = xd->up_available;
const int has_left = xd->left_available;
int above_ctx = (has_above && !above_mbmi->skip) ? (int)above_mbmi->tx_size
: max_tx_size;
int left_ctx = (has_left && !left_mbmi->skip) ? (int)left_mbmi->tx_size
: max_tx_size;
if (!has_left)
left_ctx = above_ctx;
if (!has_above)
above_ctx = left_ctx;
return (above_ctx + left_ctx) > max_tx_size;
}
static INLINE const vpx_prob *get_tx_probs(TX_SIZE max_tx_size, int ctx,
const struct tx_probs *tx_probs) {
switch (max_tx_size) {
case TX_8X8:
return tx_probs->p8x8[ctx];
case TX_16X16:
return tx_probs->p16x16[ctx];
case TX_32X32:
return tx_probs->p32x32[ctx];
default:
assert(0 && "Invalid max_tx_size.");
return NULL;
}
}
static INLINE const vpx_prob *get_tx_probs2(TX_SIZE max_tx_size,
const MACROBLOCKD *xd,
const struct tx_probs *tx_probs) {
return get_tx_probs(max_tx_size, get_tx_size_context(xd), tx_probs);
}
static INLINE unsigned int *get_tx_counts(TX_SIZE max_tx_size, int ctx,
struct tx_counts *tx_counts) {
switch (max_tx_size) {
case TX_8X8:
return tx_counts->p8x8[ctx];
case TX_16X16:
return tx_counts->p16x16[ctx];
case TX_32X32:
return tx_counts->p32x32[ctx];
default:
assert(0 && "Invalid max_tx_size.");
return NULL;
}
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_PRED_COMMON_H_

278
vp10/common/vp9_quant_common.c Normal file
View File

@ -0,0 +1,278 @@
/*
* Copyright (c) 2010 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/common/vp9_common.h"
#include "vp10/common/vp9_quant_common.h"
#include "vp10/common/vp9_seg_common.h"
static const int16_t dc_qlookup[QINDEX_RANGE] = {
4, 8, 8, 9, 10, 11, 12, 12,
13, 14, 15, 16, 17, 18, 19, 19,
20, 21, 22, 23, 24, 25, 26, 26,
27, 28, 29, 30, 31, 32, 32, 33,
34, 35, 36, 37, 38, 38, 39, 40,
41, 42, 43, 43, 44, 45, 46, 47,
48, 48, 49, 50, 51, 52, 53, 53,
54, 55, 56, 57, 57, 58, 59, 60,
61, 62, 62, 63, 64, 65, 66, 66,
67, 68, 69, 70, 70, 71, 72, 73,
74, 74, 75, 76, 77, 78, 78, 79,
80, 81, 81, 82, 83, 84, 85, 85,
87, 88, 90, 92, 93, 95, 96, 98,
99, 101, 102, 104, 105, 107, 108, 110,
111, 113, 114, 116, 117, 118, 120, 121,
123, 125, 127, 129, 131, 134, 136, 138,
140, 142, 144, 146, 148, 150, 152, 154,
156, 158, 161, 164, 166, 169, 172, 174,
177, 180, 182, 185, 187, 190, 192, 195,
199, 202, 205, 208, 211, 214, 217, 220,
223, 226, 230, 233, 237, 240, 243, 247,
250, 253, 257, 261, 265, 269, 272, 276,
280, 284, 288, 292, 296, 300, 304, 309,
313, 317, 322, 326, 330, 335, 340, 344,
349, 354, 359, 364, 369, 374, 379, 384,
389, 395, 400, 406, 411, 417, 423, 429,
435, 441, 447, 454, 461, 467, 475, 482,
489, 497, 505, 513, 522, 530, 539, 549,
559, 569, 579, 590, 602, 614, 626, 640,
654, 668, 684, 700, 717, 736, 755, 775,
796, 819, 843, 869, 896, 925, 955, 988,
1022, 1058, 1098, 1139, 1184, 1232, 1282, 1336,
};
#if CONFIG_VP9_HIGHBITDEPTH
static const int16_t dc_qlookup_10[QINDEX_RANGE] = {
4, 9, 10, 13, 15, 17, 20, 22,
25, 28, 31, 34, 37, 40, 43, 47,
50, 53, 57, 60, 64, 68, 71, 75,
78, 82, 86, 90, 93, 97, 101, 105,
109, 113, 116, 120, 124, 128, 132, 136,
140, 143, 147, 151, 155, 159, 163, 166,
170, 174, 178, 182, 185, 189, 193, 197,
200, 204, 208, 212, 215, 219, 223, 226,
230, 233, 237, 241, 244, 248, 251, 255,
259, 262, 266, 269, 273, 276, 280, 283,
287, 290, 293, 297, 300, 304, 307, 310,
314, 317, 321, 324, 327, 331, 334, 337,
343, 350, 356, 362, 369, 375, 381, 387,
394, 400, 406, 412, 418, 424, 430, 436,
442, 448, 454, 460, 466, 472, 478, 484,
490, 499, 507, 516, 525, 533, 542, 550,
559, 567, 576, 584, 592, 601, 609, 617,
625, 634, 644, 655, 666, 676, 687, 698,
708, 718, 729, 739, 749, 759, 770, 782,
795, 807, 819, 831, 844, 856, 868, 880,
891, 906, 920, 933, 947, 961, 975, 988,
1001, 1015, 1030, 1045, 1061, 1076, 1090, 1105,
1120, 1137, 1153, 1170, 1186, 1202, 1218, 1236,
1253, 1271, 1288, 1306, 1323, 1342, 1361, 1379,
1398, 1416, 1436, 1456, 1476, 1496, 1516, 1537,
1559, 1580, 1601, 1624, 1647, 1670, 1692, 1717,
1741, 1766, 1791, 1817, 1844, 1871, 1900, 1929,
1958, 1990, 2021, 2054, 2088, 2123, 2159, 2197,
2236, 2276, 2319, 2363, 2410, 2458, 2508, 2561,
2616, 2675, 2737, 2802, 2871, 2944, 3020, 3102,
3188, 3280, 3375, 3478, 3586, 3702, 3823, 3953,
4089, 4236, 4394, 4559, 4737, 4929, 5130, 5347,
};
static const int16_t dc_qlookup_12[QINDEX_RANGE] = {
4, 12, 18, 25, 33, 41, 50, 60,
70, 80, 91, 103, 115, 127, 140, 153,
166, 180, 194, 208, 222, 237, 251, 266,
281, 296, 312, 327, 343, 358, 374, 390,
405, 421, 437, 453, 469, 484, 500, 516,
532, 548, 564, 580, 596, 611, 627, 643,
659, 674, 690, 706, 721, 737, 752, 768,
783, 798, 814, 829, 844, 859, 874, 889,
904, 919, 934, 949, 964, 978, 993, 1008,
1022, 1037, 1051, 1065, 1080, 1094, 1108, 1122,
1136, 1151, 1165, 1179, 1192, 1206, 1220, 1234,
1248, 1261, 1275, 1288, 1302, 1315, 1329, 1342,
1368, 1393, 1419, 1444, 1469, 1494, 1519, 1544,
1569, 1594, 1618, 1643, 1668, 1692, 1717, 1741,
1765, 1789, 1814, 1838, 1862, 1885, 1909, 1933,
1957, 1992, 2027, 2061, 2096, 2130, 2165, 2199,
2233, 2267, 2300, 2334, 2367, 2400, 2434, 2467,
2499, 2532, 2575, 2618, 2661, 2704, 2746, 2788,
2830, 2872, 2913, 2954, 2995, 3036, 3076, 3127,
3177, 3226, 3275, 3324, 3373, 3421, 3469, 3517,
3565, 3621, 3677, 3733, 3788, 3843, 3897, 3951,
4005, 4058, 4119, 4181, 4241, 4301, 4361, 4420,
4479, 4546, 4612, 4677, 4742, 4807, 4871, 4942,
5013, 5083, 5153, 5222, 5291, 5367, 5442, 5517,
5591, 5665, 5745, 5825, 5905, 5984, 6063, 6149,
6234, 6319, 6404, 6495, 6587, 6678, 6769, 6867,
6966, 7064, 7163, 7269, 7376, 7483, 7599, 7715,
7832, 7958, 8085, 8214, 8352, 8492, 8635, 8788,
8945, 9104, 9275, 9450, 9639, 9832, 10031, 10245,
10465, 10702, 10946, 11210, 11482, 11776, 12081, 12409,
12750, 13118, 13501, 13913, 14343, 14807, 15290, 15812,
16356, 16943, 17575, 18237, 18949, 19718, 20521, 21387,
};
#endif
static const int16_t ac_qlookup[QINDEX_RANGE] = {
4, 8, 9, 10, 11, 12, 13, 14,
15, 16, 17, 18, 19, 20, 21, 22,
23, 24, 25, 26, 27, 28, 29, 30,
31, 32, 33, 34, 35, 36, 37, 38,
39, 40, 41, 42, 43, 44, 45, 46,
47, 48, 49, 50, 51, 52, 53, 54,
55, 56, 57, 58, 59, 60, 61, 62,
63, 64, 65, 66, 67, 68, 69, 70,
71, 72, 73, 74, 75, 76, 77, 78,
79, 80, 81, 82, 83, 84, 85, 86,
87, 88, 89, 90, 91, 92, 93, 94,
95, 96, 97, 98, 99, 100, 101, 102,
104, 106, 108, 110, 112, 114, 116, 118,
120, 122, 124, 126, 128, 130, 132, 134,
136, 138, 140, 142, 144, 146, 148, 150,
152, 155, 158, 161, 164, 167, 170, 173,
176, 179, 182, 185, 188, 191, 194, 197,
200, 203, 207, 211, 215, 219, 223, 227,
231, 235, 239, 243, 247, 251, 255, 260,
265, 270, 275, 280, 285, 290, 295, 300,
305, 311, 317, 323, 329, 335, 341, 347,
353, 359, 366, 373, 380, 387, 394, 401,
408, 416, 424, 432, 440, 448, 456, 465,
474, 483, 492, 501, 510, 520, 530, 540,
550, 560, 571, 582, 593, 604, 615, 627,
639, 651, 663, 676, 689, 702, 715, 729,
743, 757, 771, 786, 801, 816, 832, 848,
864, 881, 898, 915, 933, 951, 969, 988,
1007, 1026, 1046, 1066, 1087, 1108, 1129, 1151,
1173, 1196, 1219, 1243, 1267, 1292, 1317, 1343,
1369, 1396, 1423, 1451, 1479, 1508, 1537, 1567,
1597, 1628, 1660, 1692, 1725, 1759, 1793, 1828,
};
#if CONFIG_VP9_HIGHBITDEPTH
static const int16_t ac_qlookup_10[QINDEX_RANGE] = {
4, 9, 11, 13, 16, 18, 21, 24,
27, 30, 33, 37, 40, 44, 48, 51,
55, 59, 63, 67, 71, 75, 79, 83,
88, 92, 96, 100, 105, 109, 114, 118,
122, 127, 131, 136, 140, 145, 149, 154,
158, 163, 168, 172, 177, 181, 186, 190,
195, 199, 204, 208, 213, 217, 222, 226,
231, 235, 240, 244, 249, 253, 258, 262,
267, 271, 275, 280, 284, 289, 293, 297,
302, 306, 311, 315, 319, 324, 328, 332,
337, 341, 345, 349, 354, 358, 362, 367,
371, 375, 379, 384, 388, 392, 396, 401,
409, 417, 425, 433, 441, 449, 458, 466,
474, 482, 490, 498, 506, 514, 523, 531,
539, 547, 555, 563, 571, 579, 588, 596,
604, 616, 628, 640, 652, 664, 676, 688,
700, 713, 725, 737, 749, 761, 773, 785,
797, 809, 825, 841, 857, 873, 889, 905,
922, 938, 954, 970, 986, 1002, 1018, 1038,
1058, 1078, 1098, 1118, 1138, 1158, 1178, 1198,
1218, 1242, 1266, 1290, 1314, 1338, 1362, 1386,
1411, 1435, 1463, 1491, 1519, 1547, 1575, 1603,
1631, 1663, 1695, 1727, 1759, 1791, 1823, 1859,
1895, 1931, 1967, 2003, 2039, 2079, 2119, 2159,
2199, 2239, 2283, 2327, 2371, 2415, 2459, 2507,
2555, 2603, 2651, 2703, 2755, 2807, 2859, 2915,
2971, 3027, 3083, 3143, 3203, 3263, 3327, 3391,
3455, 3523, 3591, 3659, 3731, 3803, 3876, 3952,
4028, 4104, 4184, 4264, 4348, 4432, 4516, 4604,
4692, 4784, 4876, 4972, 5068, 5168, 5268, 5372,
5476, 5584, 5692, 5804, 5916, 6032, 6148, 6268,
6388, 6512, 6640, 6768, 6900, 7036, 7172, 7312,
};
static const int16_t ac_qlookup_12[QINDEX_RANGE] = {
4, 13, 19, 27, 35, 44, 54, 64,
75, 87, 99, 112, 126, 139, 154, 168,
183, 199, 214, 230, 247, 263, 280, 297,
314, 331, 349, 366, 384, 402, 420, 438,
456, 475, 493, 511, 530, 548, 567, 586,
604, 623, 642, 660, 679, 698, 716, 735,
753, 772, 791, 809, 828, 846, 865, 884,
902, 920, 939, 957, 976, 994, 1012, 1030,
1049, 1067, 1085, 1103, 1121, 1139, 1157, 1175,
1193, 1211, 1229, 1246, 1264, 1282, 1299, 1317,
1335, 1352, 1370, 1387, 1405, 1422, 1440, 1457,
1474, 1491, 1509, 1526, 1543, 1560, 1577, 1595,
1627, 1660, 1693, 1725, 1758, 1791, 1824, 1856,
1889, 1922, 1954, 1987, 2020, 2052, 2085, 2118,
2150, 2183, 2216, 2248, 2281, 2313, 2346, 2378,
2411, 2459, 2508, 2556, 2605, 2653, 2701, 2750,
2798, 2847, 2895, 2943, 2992, 3040, 3088, 3137,
3185, 3234, 3298, 3362, 3426, 3491, 3555, 3619,
3684, 3748, 3812, 3876, 3941, 4005, 4069, 4149,
4230, 4310, 4390, 4470, 4550, 4631, 4711, 4791,
4871, 4967, 5064, 5160, 5256, 5352, 5448, 5544,
5641, 5737, 5849, 5961, 6073, 6185, 6297, 6410,
6522, 6650, 6778, 6906, 7034, 7162, 7290, 7435,
7579, 7723, 7867, 8011, 8155, 8315, 8475, 8635,
8795, 8956, 9132, 9308, 9484, 9660, 9836, 10028,
10220, 10412, 10604, 10812, 11020, 11228, 11437, 11661,
11885, 12109, 12333, 12573, 12813, 13053, 13309, 13565,
13821, 14093, 14365, 14637, 14925, 15213, 15502, 15806,
16110, 16414, 16734, 17054, 17390, 17726, 18062, 18414,
18766, 19134, 19502, 19886, 20270, 20670, 21070, 21486,
21902, 22334, 22766, 23214, 23662, 24126, 24590, 25070,
25551, 26047, 26559, 27071, 27599, 28143, 28687, 29247,
};
#endif
int16_t vp10_dc_quant(int qindex, int delta, vpx_bit_depth_t bit_depth) {
#if CONFIG_VP9_HIGHBITDEPTH
switch (bit_depth) {
case VPX_BITS_8:
return dc_qlookup[clamp(qindex + delta, 0, MAXQ)];
case VPX_BITS_10:
return dc_qlookup_10[clamp(qindex + delta, 0, MAXQ)];
case VPX_BITS_12:
return dc_qlookup_12[clamp(qindex + delta, 0, MAXQ)];
default:
assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
return -1;
}
#else
(void) bit_depth;
return dc_qlookup[clamp(qindex + delta, 0, MAXQ)];
#endif
}
int16_t vp10_ac_quant(int qindex, int delta, vpx_bit_depth_t bit_depth) {
#if CONFIG_VP9_HIGHBITDEPTH
switch (bit_depth) {
case VPX_BITS_8:
return ac_qlookup[clamp(qindex + delta, 0, MAXQ)];
case VPX_BITS_10:
return ac_qlookup_10[clamp(qindex + delta, 0, MAXQ)];
case VPX_BITS_12:
return ac_qlookup_12[clamp(qindex + delta, 0, MAXQ)];
default:
assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
return -1;
}
#else
(void) bit_depth;
return ac_qlookup[clamp(qindex + delta, 0, MAXQ)];
#endif
}
int vp10_get_qindex(const struct segmentation *seg, int segment_id,
int base_qindex) {
if (segfeature_active(seg, segment_id, SEG_LVL_ALT_Q)) {
const int data = get_segdata(seg, segment_id, SEG_LVL_ALT_Q);
const int seg_qindex = seg->abs_delta == SEGMENT_ABSDATA ?
data : base_qindex + data;
return clamp(seg_qindex, 0, MAXQ);
} else {
return base_qindex;
}
}

36
vp10/common/vp9_quant_common.h Normal file
View File

@ -0,0 +1,36 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_QUANT_COMMON_H_
#define VP9_COMMON_VP9_QUANT_COMMON_H_
#include "vpx/vpx_codec.h"
#include "vp10/common/vp9_seg_common.h"
#ifdef __cplusplus
extern "C" {
#endif
#define MINQ 0
#define MAXQ 255
#define QINDEX_RANGE (MAXQ - MINQ + 1)
#define QINDEX_BITS 8
int16_t vp10_dc_quant(int qindex, int delta, vpx_bit_depth_t bit_depth);
int16_t vp10_ac_quant(int qindex, int delta, vpx_bit_depth_t bit_depth);
int vp10_get_qindex(const struct segmentation *seg, int segment_id,
int base_qindex);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_QUANT_COMMON_H_

213
vp10/common/vp9_reconinter.c Normal file
View File

@ -0,0 +1,213 @@
/*
* Copyright (c) 2010 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 <assert.h>
#include "./vpx_scale_rtcd.h"
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_reconinter.h"
#include "vp10/common/vp9_reconintra.h"
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_build_inter_predictor(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
const MV *src_mv,
const struct scale_factors *sf,
int w, int h, int ref,
const InterpKernel *kernel,
enum mv_precision precision,
int x, int y, int bd) {
const int is_q4 = precision == MV_PRECISION_Q4;
const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
is_q4 ? src_mv->col : src_mv->col * 2 };
MV32 mv = vp10_scale_mv(&mv_q4, x, y, sf);
const int subpel_x = mv.col & SUBPEL_MASK;
const int subpel_y = mv.row & SUBPEL_MASK;
src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);
high_inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4, bd);
}
#endif // CONFIG_VP9_HIGHBITDEPTH
void vp10_build_inter_predictor(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
const MV *src_mv,
const struct scale_factors *sf,
int w, int h, int ref,
const InterpKernel *kernel,
enum mv_precision precision,
int x, int y) {
const int is_q4 = precision == MV_PRECISION_Q4;
const MV mv_q4 = { is_q4 ? src_mv->row : src_mv->row * 2,
is_q4 ? src_mv->col : src_mv->col * 2 };
MV32 mv = vp10_scale_mv(&mv_q4, x, y, sf);
const int subpel_x = mv.col & SUBPEL_MASK;
const int subpel_y = mv.row & SUBPEL_MASK;
src += (mv.row >> SUBPEL_BITS) * src_stride + (mv.col >> SUBPEL_BITS);
inter_predictor(src, src_stride, dst, dst_stride, subpel_x, subpel_y,
sf, w, h, ref, kernel, sf->x_step_q4, sf->y_step_q4);
}
void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
int bw, int bh,
int x, int y, int w, int h,
int mi_x, int mi_y) {
struct macroblockd_plane *const pd = &xd->plane[plane];
const MODE_INFO *mi = xd->mi[0];
const int is_compound = has_second_ref(&mi->mbmi);
const InterpKernel *kernel = vp10_filter_kernels[mi->mbmi.interp_filter];
int ref;
for (ref = 0; ref < 1 + is_compound; ++ref) {
const struct scale_factors *const sf = &xd->block_refs[ref]->sf;
struct buf_2d *const pre_buf = &pd->pre[ref];
struct buf_2d *const dst_buf = &pd->dst;
uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
const MV mv = mi->mbmi.sb_type < BLOCK_8X8
? average_split_mvs(pd, mi, ref, block)
: mi->mbmi.mv[ref].as_mv;
// TODO(jkoleszar): This clamping is done in the incorrect place for the
// scaling case. It needs to be done on the scaled MV, not the pre-scaling
// MV. Note however that it performs the subsampling aware scaling so
// that the result is always q4.
// mv_precision precision is MV_PRECISION_Q4.
const MV mv_q4 = clamp_mv_to_umv_border_sb(xd, &mv, bw, bh,
pd->subsampling_x,
pd->subsampling_y);
uint8_t *pre;
MV32 scaled_mv;
int xs, ys, subpel_x, subpel_y;
const int is_scaled = vp10_is_scaled(sf);
if (is_scaled) {
pre = pre_buf->buf + scaled_buffer_offset(x, y, pre_buf->stride, sf);
scaled_mv = vp10_scale_mv(&mv_q4, mi_x + x, mi_y + y, sf);
xs = sf->x_step_q4;
ys = sf->y_step_q4;
} else {
pre = pre_buf->buf + (y * pre_buf->stride + x);
scaled_mv.row = mv_q4.row;
scaled_mv.col = mv_q4.col;
xs = ys = 16;
}
subpel_x = scaled_mv.col & SUBPEL_MASK;
subpel_y = scaled_mv.row & SUBPEL_MASK;
pre += (scaled_mv.row >> SUBPEL_BITS) * pre_buf->stride
+ (scaled_mv.col >> SUBPEL_BITS);
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
high_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys,
xd->bd);
} else {
inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
}
#else
inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
subpel_x, subpel_y, sf, w, h, ref, kernel, xs, ys);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}
static void build_inter_predictors_for_planes(MACROBLOCKD *xd, BLOCK_SIZE bsize,
int mi_row, int mi_col,
int plane_from, int plane_to) {
int plane;
const int mi_x = mi_col * MI_SIZE;
const int mi_y = mi_row * MI_SIZE;
for (plane = plane_from; plane <= plane_to; ++plane) {
const BLOCK_SIZE plane_bsize = get_plane_block_size(bsize,
&xd->plane[plane]);
const int num_4x4_w = num_4x4_blocks_wide_lookup[plane_bsize];
const int num_4x4_h = num_4x4_blocks_high_lookup[plane_bsize];
const int bw = 4 * num_4x4_w;
const int bh = 4 * num_4x4_h;
if (xd->mi[0]->mbmi.sb_type < BLOCK_8X8) {
int i = 0, x, y;
assert(bsize == BLOCK_8X8);
for (y = 0; y < num_4x4_h; ++y)
for (x = 0; x < num_4x4_w; ++x)
build_inter_predictors(xd, plane, i++, bw, bh,
4 * x, 4 * y, 4, 4, mi_x, mi_y);
} else {
build_inter_predictors(xd, plane, 0, bw, bh,
0, 0, bw, bh, mi_x, mi_y);
}
}
}
void vp10_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0, 0);
}
void vp10_build_inter_predictors_sbp(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize, int plane) {
build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, plane, plane);
}
void vp10_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 1,
MAX_MB_PLANE - 1);
}
void vp10_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize) {
build_inter_predictors_for_planes(xd, bsize, mi_row, mi_col, 0,
MAX_MB_PLANE - 1);
}
void vp10_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],
const YV12_BUFFER_CONFIG *src,
int mi_row, int mi_col) {
uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer,
src->v_buffer};
const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride,
src->uv_stride};
int i;
for (i = 0; i < MAX_MB_PLANE; ++i) {
struct macroblockd_plane *const pd = &planes[i];
setup_pred_plane(&pd->dst, buffers[i], strides[i], mi_row, mi_col, NULL,
pd->subsampling_x, pd->subsampling_y);
}
}
void vp10_setup_pre_planes(MACROBLOCKD *xd, int idx,
const YV12_BUFFER_CONFIG *src,
int mi_row, int mi_col,
const struct scale_factors *sf) {
if (src != NULL) {
int i;
uint8_t *const buffers[MAX_MB_PLANE] = { src->y_buffer, src->u_buffer,
src->v_buffer};
const int strides[MAX_MB_PLANE] = { src->y_stride, src->uv_stride,
src->uv_stride};
for (i = 0; i < MAX_MB_PLANE; ++i) {
struct macroblockd_plane *const pd = &xd->plane[i];
setup_pred_plane(&pd->pre[idx], buffers[i], strides[i], mi_row, mi_col,
sf, pd->subsampling_x, pd->subsampling_y);
}
}
}

195
vp10/common/vp9_reconinter.h Normal file
View File

@ -0,0 +1,195 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_RECONINTER_H_
#define VP9_COMMON_VP9_RECONINTER_H_
#include "vp10/common/vp9_filter.h"
#include "vp10/common/vp9_onyxc_int.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/vpx_filter.h"
#ifdef __cplusplus
extern "C" {
#endif
static INLINE void inter_predictor(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
const int subpel_x,
const int subpel_y,
const struct scale_factors *sf,
int w, int h, int ref,
const InterpKernel *kernel,
int xs, int ys) {
sf->predict[subpel_x != 0][subpel_y != 0][ref](
src, src_stride, dst, dst_stride,
kernel[subpel_x], xs, kernel[subpel_y], ys, w, h);
}
#if CONFIG_VP9_HIGHBITDEPTH
static void high_inter_predictor(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
const int subpel_x,
const int subpel_y,
const struct scale_factors *sf,
int w, int h, int ref,
const InterpKernel *kernel,
int xs, int ys, int bd) {
sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref](
src, src_stride, dst, dst_stride,
kernel[subpel_x], xs, kernel[subpel_y], ys, w, h, bd);
}
#endif // CONFIG_VP9_HIGHBITDEPTH
static INLINE int round_mv_comp_q4(int value) {
return (value < 0 ? value - 2 : value + 2) / 4;
}
static MV mi_mv_pred_q4(const MODE_INFO *mi, int idx) {
MV res = { round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.row +
mi->bmi[1].as_mv[idx].as_mv.row +
mi->bmi[2].as_mv[idx].as_mv.row +
mi->bmi[3].as_mv[idx].as_mv.row),
round_mv_comp_q4(mi->bmi[0].as_mv[idx].as_mv.col +
mi->bmi[1].as_mv[idx].as_mv.col +
mi->bmi[2].as_mv[idx].as_mv.col +
mi->bmi[3].as_mv[idx].as_mv.col) };
return res;
}
static INLINE int round_mv_comp_q2(int value) {
return (value < 0 ? value - 1 : value + 1) / 2;
}
static MV mi_mv_pred_q2(const MODE_INFO *mi, int idx, int block0, int block1) {
MV res = { round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.row +
mi->bmi[block1].as_mv[idx].as_mv.row),
round_mv_comp_q2(mi->bmi[block0].as_mv[idx].as_mv.col +
mi->bmi[block1].as_mv[idx].as_mv.col) };
return res;
}
// TODO(jkoleszar): yet another mv clamping function :-(
static MV clamp_mv_to_umv_border_sb(const MACROBLOCKD *xd, const MV *src_mv,
int bw, int bh, int ss_x, int ss_y) {
// If the MV points so far into the UMV border that no visible pixels
// are used for reconstruction, the subpel part of the MV can be
// discarded and the MV limited to 16 pixels with equivalent results.
const int spel_left = (VP9_INTERP_EXTEND + bw) << SUBPEL_BITS;
const int spel_right = spel_left - SUBPEL_SHIFTS;
const int spel_top = (VP9_INTERP_EXTEND + bh) << SUBPEL_BITS;
const int spel_bottom = spel_top - SUBPEL_SHIFTS;
MV clamped_mv = {
src_mv->row * (1 << (1 - ss_y)),
src_mv->col * (1 << (1 - ss_x))
};
assert(ss_x <= 1);
assert(ss_y <= 1);
clamp_mv(&clamped_mv,
xd->mb_to_left_edge * (1 << (1 - ss_x)) - spel_left,
xd->mb_to_right_edge * (1 << (1 - ss_x)) + spel_right,
xd->mb_to_top_edge * (1 << (1 - ss_y)) - spel_top,
xd->mb_to_bottom_edge * (1 << (1 - ss_y)) + spel_bottom);
return clamped_mv;
}
static MV average_split_mvs(const struct macroblockd_plane *pd,
const MODE_INFO *mi, int ref, int block) {
const int ss_idx = ((pd->subsampling_x > 0) << 1) | (pd->subsampling_y > 0);
MV res = {0, 0};
switch (ss_idx) {
case 0:
res = mi->bmi[block].as_mv[ref].as_mv;
break;
case 1:
res = mi_mv_pred_q2(mi, ref, block, block + 2);
break;
case 2:
res = mi_mv_pred_q2(mi, ref, block, block + 1);
break;
case 3:
res = mi_mv_pred_q4(mi, ref);
break;
default:
assert(ss_idx <= 3 && ss_idx >= 0);
}
return res;
}
void build_inter_predictors(MACROBLOCKD *xd, int plane, int block,
int bw, int bh,
int x, int y, int w, int h,
int mi_x, int mi_y);
void vp10_build_inter_predictors_sby(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize);
void vp10_build_inter_predictors_sbp(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize, int plane);
void vp10_build_inter_predictors_sbuv(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize);
void vp10_build_inter_predictors_sb(MACROBLOCKD *xd, int mi_row, int mi_col,
BLOCK_SIZE bsize);
void vp10_build_inter_predictor(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
const MV *mv_q3,
const struct scale_factors *sf,
int w, int h, int do_avg,
const InterpKernel *kernel,
enum mv_precision precision,
int x, int y);
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_build_inter_predictor(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_stride,
const MV *mv_q3,
const struct scale_factors *sf,
int w, int h, int do_avg,
const InterpKernel *kernel,
enum mv_precision precision,
int x, int y, int bd);
#endif
static INLINE int scaled_buffer_offset(int x_offset, int y_offset, int stride,
const struct scale_factors *sf) {
const int x = sf ? sf->scale_value_x(x_offset, sf) : x_offset;
const int y = sf ? sf->scale_value_y(y_offset, sf) : y_offset;
return y * stride + x;
}
static INLINE void setup_pred_plane(struct buf_2d *dst,
uint8_t *src, int stride,
int mi_row, int mi_col,
const struct scale_factors *scale,
int subsampling_x, int subsampling_y) {
const int x = (MI_SIZE * mi_col) >> subsampling_x;
const int y = (MI_SIZE * mi_row) >> subsampling_y;
dst->buf = src + scaled_buffer_offset(x, y, stride, scale);
dst->stride = stride;
}
void vp10_setup_dst_planes(struct macroblockd_plane planes[MAX_MB_PLANE],
const YV12_BUFFER_CONFIG *src,
int mi_row, int mi_col);
void vp10_setup_pre_planes(MACROBLOCKD *xd, int idx,
const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
const struct scale_factors *sf);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_RECONINTER_H_

402
vp10/common/vp9_reconintra.c Normal file
View File

@ -0,0 +1,402 @@
/*
* Copyright (c) 2010 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 "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
#if CONFIG_VP9_HIGHBITDEPTH
#include "vpx_dsp/vpx_dsp_common.h"
#endif // CONFIG_VP9_HIGHBITDEPTH
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
#include "vpx_ports/vpx_once.h"
#include "vp10/common/vp9_reconintra.h"
#include "vp10/common/vp9_onyxc_int.h"
enum {
NEED_LEFT = 1 << 1,
NEED_ABOVE = 1 << 2,
NEED_ABOVERIGHT = 1 << 3,
};
static const uint8_t extend_modes[INTRA_MODES] = {
NEED_ABOVE | NEED_LEFT, // DC
NEED_ABOVE, // V
NEED_LEFT, // H
NEED_ABOVERIGHT, // D45
NEED_LEFT | NEED_ABOVE, // D135
NEED_LEFT | NEED_ABOVE, // D117
NEED_LEFT | NEED_ABOVE, // D153
NEED_LEFT, // D207
NEED_ABOVERIGHT, // D63
NEED_LEFT | NEED_ABOVE, // TM
};
typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride,
const uint8_t *above, const uint8_t *left);
static intra_pred_fn pred[INTRA_MODES][TX_SIZES];
static intra_pred_fn dc_pred[2][2][TX_SIZES];
#if CONFIG_VP9_HIGHBITDEPTH
typedef void (*intra_high_pred_fn)(uint16_t *dst, ptrdiff_t stride,
const uint16_t *above, const uint16_t *left,
int bd);
static intra_high_pred_fn pred_high[INTRA_MODES][4];
static intra_high_pred_fn dc_pred_high[2][2][4];
#endif // CONFIG_VP9_HIGHBITDEPTH
static void vp10_init_intra_predictors_internal(void) {
#define INIT_ALL_SIZES(p, type) \
p[TX_4X4] = vpx_##type##_predictor_4x4; \
p[TX_8X8] = vpx_##type##_predictor_8x8; \
p[TX_16X16] = vpx_##type##_predictor_16x16; \
p[TX_32X32] = vpx_##type##_predictor_32x32
INIT_ALL_SIZES(pred[V_PRED], v);
INIT_ALL_SIZES(pred[H_PRED], h);
INIT_ALL_SIZES(pred[D207_PRED], d207);
INIT_ALL_SIZES(pred[D45_PRED], d45);
INIT_ALL_SIZES(pred[D63_PRED], d63);
INIT_ALL_SIZES(pred[D117_PRED], d117);
INIT_ALL_SIZES(pred[D135_PRED], d135);
INIT_ALL_SIZES(pred[D153_PRED], d153);
INIT_ALL_SIZES(pred[TM_PRED], tm);
INIT_ALL_SIZES(dc_pred[0][0], dc_128);
INIT_ALL_SIZES(dc_pred[0][1], dc_top);
INIT_ALL_SIZES(dc_pred[1][0], dc_left);
INIT_ALL_SIZES(dc_pred[1][1], dc);
#if CONFIG_VP9_HIGHBITDEPTH
INIT_ALL_SIZES(pred_high[V_PRED], highbd_v);
INIT_ALL_SIZES(pred_high[H_PRED], highbd_h);
INIT_ALL_SIZES(pred_high[D207_PRED], highbd_d207);
INIT_ALL_SIZES(pred_high[D45_PRED], highbd_d45);
INIT_ALL_SIZES(pred_high[D63_PRED], highbd_d63);
INIT_ALL_SIZES(pred_high[D117_PRED], highbd_d117);
INIT_ALL_SIZES(pred_high[D135_PRED], highbd_d135);
INIT_ALL_SIZES(pred_high[D153_PRED], highbd_d153);
INIT_ALL_SIZES(pred_high[TM_PRED], highbd_tm);
INIT_ALL_SIZES(dc_pred_high[0][0], highbd_dc_128);
INIT_ALL_SIZES(dc_pred_high[0][1], highbd_dc_top);
INIT_ALL_SIZES(dc_pred_high[1][0], highbd_dc_left);
INIT_ALL_SIZES(dc_pred_high[1][1], highbd_dc);
#endif // CONFIG_VP9_HIGHBITDEPTH
#undef intra_pred_allsizes
}
#if CONFIG_VP9_HIGHBITDEPTH
static void build_intra_predictors_high(const MACROBLOCKD *xd,
const uint8_t *ref8,
int ref_stride,
uint8_t *dst8,
int dst_stride,
PREDICTION_MODE mode,
TX_SIZE tx_size,
int up_available,
int left_available,
int right_available,
int x, int y,
int plane, int bd) {
int i;
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
DECLARE_ALIGNED(16, uint16_t, left_col[32]);
DECLARE_ALIGNED(16, uint16_t, above_data[64 + 16]);
uint16_t *above_row = above_data + 16;
const uint16_t *const_above_row = above_row;
const int bs = 4 << tx_size;
int frame_width, frame_height;
int x0, y0;
const struct macroblockd_plane *const pd = &xd->plane[plane];
// int base=128;
int base = 128 << (bd - 8);
// 127 127 127 .. 127 127 127 127 127 127
// 129 A B .. Y Z
// 129 C D .. W X
// 129 E F .. U V
// 129 G H .. S T T T T T
// Get current frame pointer, width and height.
if (plane == 0) {
frame_width = xd->cur_buf->y_width;
frame_height = xd->cur_buf->y_height;
} else {
frame_width = xd->cur_buf->uv_width;
frame_height = xd->cur_buf->uv_height;
}
// Get block position in current frame.
x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
// left
if (left_available) {
if (xd->mb_to_bottom_edge < 0) {
/* slower path if the block needs border extension */
if (y0 + bs <= frame_height) {
for (i = 0; i < bs; ++i)
left_col[i] = ref[i * ref_stride - 1];
} else {
const int extend_bottom = frame_height - y0;
for (i = 0; i < extend_bottom; ++i)
left_col[i] = ref[i * ref_stride - 1];
for (; i < bs; ++i)
left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
}
} else {
/* faster path if the block does not need extension */
for (i = 0; i < bs; ++i)
left_col[i] = ref[i * ref_stride - 1];
}
} else {
// TODO(Peter): this value should probably change for high bitdepth
vpx_memset16(left_col, base + 1, bs);
}
// TODO(hkuang) do not extend 2*bs pixels for all modes.
// above
if (up_available) {
const uint16_t *above_ref = ref - ref_stride;
if (xd->mb_to_right_edge < 0) {
/* slower path if the block needs border extension */
if (x0 + 2 * bs <= frame_width) {
if (right_available && bs == 4) {
memcpy(above_row, above_ref, 2 * bs * sizeof(uint16_t));
} else {
memcpy(above_row, above_ref, bs * sizeof(uint16_t));
vpx_memset16(above_row + bs, above_row[bs - 1], bs);
}
} else if (x0 + bs <= frame_width) {
const int r = frame_width - x0;
if (right_available && bs == 4) {
memcpy(above_row, above_ref, r * sizeof(uint16_t));
vpx_memset16(above_row + r, above_row[r - 1],
x0 + 2 * bs - frame_width);
} else {
memcpy(above_row, above_ref, bs * sizeof(uint16_t));
vpx_memset16(above_row + bs, above_row[bs - 1], bs);
}
} else if (x0 <= frame_width) {
const int r = frame_width - x0;
memcpy(above_row, above_ref, r * sizeof(uint16_t));
vpx_memset16(above_row + r, above_row[r - 1],
x0 + 2 * bs - frame_width);
}
// TODO(Peter) this value should probably change for high bitdepth
above_row[-1] = left_available ? above_ref[-1] : (base+1);
} else {
/* faster path if the block does not need extension */
if (bs == 4 && right_available && left_available) {
const_above_row = above_ref;
} else {
memcpy(above_row, above_ref, bs * sizeof(uint16_t));
if (bs == 4 && right_available)
memcpy(above_row + bs, above_ref + bs, bs * sizeof(uint16_t));
else
vpx_memset16(above_row + bs, above_row[bs - 1], bs);
// TODO(Peter): this value should probably change for high bitdepth
above_row[-1] = left_available ? above_ref[-1] : (base+1);
}
}
} else {
vpx_memset16(above_row, base - 1, bs * 2);
// TODO(Peter): this value should probably change for high bitdepth
above_row[-1] = base - 1;
}
// predict
if (mode == DC_PRED) {
dc_pred_high[left_available][up_available][tx_size](dst, dst_stride,
const_above_row,
left_col, xd->bd);
} else {
pred_high[mode][tx_size](dst, dst_stride, const_above_row, left_col,
xd->bd);
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
int ref_stride, uint8_t *dst, int dst_stride,
PREDICTION_MODE mode, TX_SIZE tx_size,
int up_available, int left_available,
int right_available, int x, int y,
int plane) {
int i;
DECLARE_ALIGNED(16, uint8_t, left_col[32]);
DECLARE_ALIGNED(16, uint8_t, above_data[64 + 16]);
uint8_t *above_row = above_data + 16;
const uint8_t *const_above_row = above_row;
const int bs = 4 << tx_size;
int frame_width, frame_height;
int x0, y0;
const struct macroblockd_plane *const pd = &xd->plane[plane];
// 127 127 127 .. 127 127 127 127 127 127
// 129 A B .. Y Z
// 129 C D .. W X
// 129 E F .. U V
// 129 G H .. S T T T T T
// ..
// Get current frame pointer, width and height.
if (plane == 0) {
frame_width = xd->cur_buf->y_width;
frame_height = xd->cur_buf->y_height;
} else {
frame_width = xd->cur_buf->uv_width;
frame_height = xd->cur_buf->uv_height;
}
// Get block position in current frame.
x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
// NEED_LEFT
if (extend_modes[mode] & NEED_LEFT) {
if (left_available) {
if (xd->mb_to_bottom_edge < 0) {
/* slower path if the block needs border extension */
if (y0 + bs <= frame_height) {
for (i = 0; i < bs; ++i)
left_col[i] = ref[i * ref_stride - 1];
} else {
const int extend_bottom = frame_height - y0;
for (i = 0; i < extend_bottom; ++i)
left_col[i] = ref[i * ref_stride - 1];
for (; i < bs; ++i)
left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
}
} else {
/* faster path if the block does not need extension */
for (i = 0; i < bs; ++i)
left_col[i] = ref[i * ref_stride - 1];
}
} else {
memset(left_col, 129, bs);
}
}
// NEED_ABOVE
if (extend_modes[mode] & NEED_ABOVE) {
if (up_available) {
const uint8_t *above_ref = ref - ref_stride;
if (xd->mb_to_right_edge < 0) {
/* slower path if the block needs border extension */
if (x0 + bs <= frame_width) {
memcpy(above_row, above_ref, bs);
} else if (x0 <= frame_width) {
const int r = frame_width - x0;
memcpy(above_row, above_ref, r);
memset(above_row + r, above_row[r - 1], x0 + bs - frame_width);
}
} else {
/* faster path if the block does not need extension */
if (bs == 4 && right_available && left_available) {
const_above_row = above_ref;
} else {
memcpy(above_row, above_ref, bs);
}
}
above_row[-1] = left_available ? above_ref[-1] : 129;
} else {
memset(above_row, 127, bs);
above_row[-1] = 127;
}
}
// NEED_ABOVERIGHT
if (extend_modes[mode] & NEED_ABOVERIGHT) {
if (up_available) {
const uint8_t *above_ref = ref - ref_stride;
if (xd->mb_to_right_edge < 0) {
/* slower path if the block needs border extension */
if (x0 + 2 * bs <= frame_width) {
if (right_available && bs == 4) {
memcpy(above_row, above_ref, 2 * bs);
} else {
memcpy(above_row, above_ref, bs);
memset(above_row + bs, above_row[bs - 1], bs);
}
} else if (x0 + bs <= frame_width) {
const int r = frame_width - x0;
if (right_available && bs == 4) {
memcpy(above_row, above_ref, r);
memset(above_row + r, above_row[r - 1], x0 + 2 * bs - frame_width);
} else {
memcpy(above_row, above_ref, bs);
memset(above_row + bs, above_row[bs - 1], bs);
}
} else if (x0 <= frame_width) {
const int r = frame_width - x0;
memcpy(above_row, above_ref, r);
memset(above_row + r, above_row[r - 1], x0 + 2 * bs - frame_width);
}
} else {
/* faster path if the block does not need extension */
if (bs == 4 && right_available && left_available) {
const_above_row = above_ref;
} else {
memcpy(above_row, above_ref, bs);
if (bs == 4 && right_available)
memcpy(above_row + bs, above_ref + bs, bs);
else
memset(above_row + bs, above_row[bs - 1], bs);
}
}
above_row[-1] = left_available ? above_ref[-1] : 129;
} else {
memset(above_row, 127, bs * 2);
above_row[-1] = 127;
}
}
// predict
if (mode == DC_PRED) {
dc_pred[left_available][up_available][tx_size](dst, dst_stride,
const_above_row, left_col);
} else {
pred[mode][tx_size](dst, dst_stride, const_above_row, left_col);
}
}
void vp10_predict_intra_block(const MACROBLOCKD *xd, int bwl_in,
TX_SIZE tx_size, PREDICTION_MODE mode,
const uint8_t *ref, int ref_stride,
uint8_t *dst, int dst_stride,
int aoff, int loff, int plane) {
const int bw = (1 << bwl_in);
const int txw = (1 << tx_size);
const int have_top = loff || xd->up_available;
const int have_left = aoff || xd->left_available;
const int have_right = (aoff + txw) < bw;
const int x = aoff * 4;
const int y = loff * 4;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
build_intra_predictors_high(xd, ref, ref_stride, dst, dst_stride, mode,
tx_size, have_top, have_left, have_right,
x, y, plane, xd->bd);
return;
}
#endif
build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size,
have_top, have_left, have_right, x, y, plane);
}
void vp10_init_intra_predictors(void) {
once(vp10_init_intra_predictors_internal);
}

32
vp10/common/vp9_reconintra.h Normal file
View File

@ -0,0 +1,32 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_RECONINTRA_H_
#define VP9_COMMON_VP9_RECONINTRA_H_
#include "vpx/vpx_integer.h"
#include "vp10/common/vp9_blockd.h"
#ifdef __cplusplus
extern "C" {
#endif
void vp10_init_intra_predictors(void);
void vp10_predict_intra_block(const MACROBLOCKD *xd, int bwl_in,
TX_SIZE tx_size, PREDICTION_MODE mode,
const uint8_t *ref, int ref_stride,
uint8_t *dst, int dst_stride,
int aoff, int loff, int plane);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_RECONINTRA_H_

166
vp10/common/vp9_scale.c Normal file
View File

@ -0,0 +1,166 @@
/*
* Copyright (c) 2013 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 "./vpx_dsp_rtcd.h"
#include "vp10/common/vp9_filter.h"
#include "vp10/common/vp9_scale.h"
#include "vpx_dsp/vpx_filter.h"
static INLINE int scaled_x(int val, const struct scale_factors *sf) {
return (int)((int64_t)val * sf->x_scale_fp >> REF_SCALE_SHIFT);
}
static INLINE int scaled_y(int val, const struct scale_factors *sf) {
return (int)((int64_t)val * sf->y_scale_fp >> REF_SCALE_SHIFT);
}
static int unscaled_value(int val, const struct scale_factors *sf) {
(void) sf;
return val;
}
static int get_fixed_point_scale_factor(int other_size, int this_size) {
// Calculate scaling factor once for each reference frame
// and use fixed point scaling factors in decoding and encoding routines.
// Hardware implementations can calculate scale factor in device driver
// and use multiplication and shifting on hardware instead of division.
return (other_size << REF_SCALE_SHIFT) / this_size;
}
MV32 vp10_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf) {
const int x_off_q4 = scaled_x(x << SUBPEL_BITS, sf) & SUBPEL_MASK;
const int y_off_q4 = scaled_y(y << SUBPEL_BITS, sf) & SUBPEL_MASK;
const MV32 res = {
scaled_y(mv->row, sf) + y_off_q4,
scaled_x(mv->col, sf) + x_off_q4
};
return res;
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_setup_scale_factors_for_frame(struct scale_factors *sf,
int other_w, int other_h,
int this_w, int this_h,
int use_highbd) {
#else
void vp10_setup_scale_factors_for_frame(struct scale_factors *sf,
int other_w, int other_h,
int this_w, int this_h) {
#endif
if (!valid_ref_frame_size(other_w, other_h, this_w, this_h)) {
sf->x_scale_fp = REF_INVALID_SCALE;
sf->y_scale_fp = REF_INVALID_SCALE;
return;
}
sf->x_scale_fp = get_fixed_point_scale_factor(other_w, this_w);
sf->y_scale_fp = get_fixed_point_scale_factor(other_h, this_h);
sf->x_step_q4 = scaled_x(16, sf);
sf->y_step_q4 = scaled_y(16, sf);
if (vp10_is_scaled(sf)) {
sf->scale_value_x = scaled_x;
sf->scale_value_y = scaled_y;
} else {
sf->scale_value_x = unscaled_value;
sf->scale_value_y = unscaled_value;
}
// TODO(agrange): Investigate the best choice of functions to use here
// for EIGHTTAP_SMOOTH. Since it is not interpolating, need to choose what
// to do at full-pel offsets. The current selection, where the filter is
// applied in one direction only, and not at all for 0,0, seems to give the
// best quality, but it may be worth trying an additional mode that does
// do the filtering on full-pel.
if (sf->x_step_q4 == 16) {
if (sf->y_step_q4 == 16) {
// No scaling in either direction.
sf->predict[0][0][0] = vpx_convolve_copy;
sf->predict[0][0][1] = vpx_convolve_avg;
sf->predict[0][1][0] = vpx_convolve8_vert;
sf->predict[0][1][1] = vpx_convolve8_avg_vert;
sf->predict[1][0][0] = vpx_convolve8_horiz;
sf->predict[1][0][1] = vpx_convolve8_avg_horiz;
} else {
// No scaling in x direction. Must always scale in the y direction.
sf->predict[0][0][0] = vpx_convolve8_vert;
sf->predict[0][0][1] = vpx_convolve8_avg_vert;
sf->predict[0][1][0] = vpx_convolve8_vert;
sf->predict[0][1][1] = vpx_convolve8_avg_vert;
sf->predict[1][0][0] = vpx_convolve8;
sf->predict[1][0][1] = vpx_convolve8_avg;
}
} else {
if (sf->y_step_q4 == 16) {
// No scaling in the y direction. Must always scale in the x direction.
sf->predict[0][0][0] = vpx_convolve8_horiz;
sf->predict[0][0][1] = vpx_convolve8_avg_horiz;
sf->predict[0][1][0] = vpx_convolve8;
sf->predict[0][1][1] = vpx_convolve8_avg;
sf->predict[1][0][0] = vpx_convolve8_horiz;
sf->predict[1][0][1] = vpx_convolve8_avg_horiz;
} else {
// Must always scale in both directions.
sf->predict[0][0][0] = vpx_convolve8;
sf->predict[0][0][1] = vpx_convolve8_avg;
sf->predict[0][1][0] = vpx_convolve8;
sf->predict[0][1][1] = vpx_convolve8_avg;
sf->predict[1][0][0] = vpx_convolve8;
sf->predict[1][0][1] = vpx_convolve8_avg;
}
}
// 2D subpel motion always gets filtered in both directions
sf->predict[1][1][0] = vpx_convolve8;
sf->predict[1][1][1] = vpx_convolve8_avg;
#if CONFIG_VP9_HIGHBITDEPTH
if (use_highbd) {
if (sf->x_step_q4 == 16) {
if (sf->y_step_q4 == 16) {
// No scaling in either direction.
sf->highbd_predict[0][0][0] = vpx_highbd_convolve_copy;
sf->highbd_predict[0][0][1] = vpx_highbd_convolve_avg;
sf->highbd_predict[0][1][0] = vpx_highbd_convolve8_vert;
sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg_vert;
sf->highbd_predict[1][0][0] = vpx_highbd_convolve8_horiz;
sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg_horiz;
} else {
// No scaling in x direction. Must always scale in the y direction.
sf->highbd_predict[0][0][0] = vpx_highbd_convolve8_vert;
sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg_vert;
sf->highbd_predict[0][1][0] = vpx_highbd_convolve8_vert;
sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg_vert;
sf->highbd_predict[1][0][0] = vpx_highbd_convolve8;
sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg;
}
} else {
if (sf->y_step_q4 == 16) {
// No scaling in the y direction. Must always scale in the x direction.
sf->highbd_predict[0][0][0] = vpx_highbd_convolve8_horiz;
sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg_horiz;
sf->highbd_predict[0][1][0] = vpx_highbd_convolve8;
sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg;
sf->highbd_predict[1][0][0] = vpx_highbd_convolve8_horiz;
sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg_horiz;
} else {
// Must always scale in both directions.
sf->highbd_predict[0][0][0] = vpx_highbd_convolve8;
sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg;
sf->highbd_predict[0][1][0] = vpx_highbd_convolve8;
sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg;
sf->highbd_predict[1][0][0] = vpx_highbd_convolve8;
sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg;
}
}
// 2D subpel motion always gets filtered in both directions.
sf->highbd_predict[1][1][0] = vpx_highbd_convolve8;
sf->highbd_predict[1][1][1] = vpx_highbd_convolve8_avg;
}
#endif
}

75
vp10/common/vp9_scale.h Normal file
View File

@ -0,0 +1,75 @@
/*
* Copyright (c) 2013 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 VP9_COMMON_VP9_SCALE_H_
#define VP9_COMMON_VP9_SCALE_H_
#include "vp10/common/vp9_mv.h"
#include "vpx_dsp/vpx_convolve.h"
#ifdef __cplusplus
extern "C" {
#endif
#define REF_SCALE_SHIFT 14
#define REF_NO_SCALE (1 << REF_SCALE_SHIFT)
#define REF_INVALID_SCALE -1
struct scale_factors {
int x_scale_fp; // horizontal fixed point scale factor
int y_scale_fp; // vertical fixed point scale factor
int x_step_q4;
int y_step_q4;
int (*scale_value_x)(int val, const struct scale_factors *sf);
int (*scale_value_y)(int val, const struct scale_factors *sf);
convolve_fn_t predict[2][2][2]; // horiz, vert, avg
#if CONFIG_VP9_HIGHBITDEPTH
highbd_convolve_fn_t highbd_predict[2][2][2]; // horiz, vert, avg
#endif
};
MV32 vp10_scale_mv(const MV *mv, int x, int y, const struct scale_factors *sf);
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_setup_scale_factors_for_frame(struct scale_factors *sf,
int other_w, int other_h,
int this_w, int this_h,
int use_high);
#else
void vp10_setup_scale_factors_for_frame(struct scale_factors *sf,
int other_w, int other_h,
int this_w, int this_h);
#endif
static INLINE int vp10_is_valid_scale(const struct scale_factors *sf) {
return sf->x_scale_fp != REF_INVALID_SCALE &&
sf->y_scale_fp != REF_INVALID_SCALE;
}
static INLINE int vp10_is_scaled(const struct scale_factors *sf) {
return vp10_is_valid_scale(sf) &&
(sf->x_scale_fp != REF_NO_SCALE || sf->y_scale_fp != REF_NO_SCALE);
}
static INLINE int valid_ref_frame_size(int ref_width, int ref_height,
int this_width, int this_height) {
return 2 * this_width >= ref_width &&
2 * this_height >= ref_height &&
this_width <= 16 * ref_width &&
this_height <= 16 * ref_height;
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_SCALE_H_

727
vp10/common/vp9_scan.c Normal file
View File

@ -0,0 +1,727 @@
/*
* Copyright (c) 2010 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 <assert.h>
#include "vp10/common/vp9_scan.h"
DECLARE_ALIGNED(16, static const int16_t, default_scan_4x4[16]) = {
0, 4, 1, 5,
8, 2, 12, 9,
3, 6, 13, 10,
7, 14, 11, 15,
};
DECLARE_ALIGNED(16, static const int16_t, col_scan_4x4[16]) = {
0, 4, 8, 1,
12, 5, 9, 2,
13, 6, 10, 3,
7, 14, 11, 15,
};
DECLARE_ALIGNED(16, static const int16_t, row_scan_4x4[16]) = {
0, 1, 4, 2,
5, 3, 6, 8,
9, 7, 12, 10,
13, 11, 14, 15,
};
DECLARE_ALIGNED(16, static const int16_t, default_scan_8x8[64]) = {
0, 8, 1, 16, 9, 2, 17, 24,
10, 3, 18, 25, 32, 11, 4, 26,
33, 19, 40, 12, 34, 27, 5, 41,
20, 48, 13, 35, 42, 28, 21, 6,
49, 56, 36, 43, 29, 7, 14, 50,
57, 44, 22, 37, 15, 51, 58, 30,
45, 23, 52, 59, 38, 31, 60, 53,
46, 39, 61, 54, 47, 62, 55, 63,
};
DECLARE_ALIGNED(16, static const int16_t, col_scan_8x8[64]) = {
0, 8, 16, 1, 24, 9, 32, 17,
2, 40, 25, 10, 33, 18, 48, 3,
26, 41, 11, 56, 19, 34, 4, 49,
27, 42, 12, 35, 20, 57, 50, 28,
5, 43, 13, 36, 58, 51, 21, 44,
6, 29, 59, 37, 14, 52, 22, 7,
45, 60, 30, 15, 38, 53, 23, 46,
31, 61, 39, 54, 47, 62, 55, 63,
};
DECLARE_ALIGNED(16, static const int16_t, row_scan_8x8[64]) = {
0, 1, 2, 8, 9, 3, 16, 10,
4, 17, 11, 24, 5, 18, 25, 12,
19, 26, 32, 6, 13, 20, 33, 27,
7, 34, 40, 21, 28, 41, 14, 35,
48, 42, 29, 36, 49, 22, 43, 15,
56, 37, 50, 44, 30, 57, 23, 51,
58, 45, 38, 52, 31, 59, 53, 46,
60, 39, 61, 47, 54, 55, 62, 63,
};
DECLARE_ALIGNED(16, static const int16_t, default_scan_16x16[256]) = {
0, 16, 1, 32, 17, 2, 48, 33, 18, 3, 64, 34, 49, 19, 65, 80,
50, 4, 35, 66, 20, 81, 96, 51, 5, 36, 82, 97, 67, 112, 21, 52,
98, 37, 83, 113, 6, 68, 128, 53, 22, 99, 114, 84, 7, 129, 38, 69,
100, 115, 144, 130, 85, 54, 23, 8, 145, 39, 70, 116, 101, 131, 160, 146,
55, 86, 24, 71, 132, 117, 161, 40, 9, 102, 147, 176, 162, 87, 56, 25,
133, 118, 177, 148, 72, 103, 41, 163, 10, 192, 178, 88, 57, 134, 149, 119,
26, 164, 73, 104, 193, 42, 179, 208, 11, 135, 89, 165, 120, 150, 58, 194,
180, 27, 74, 209, 105, 151, 136, 43, 90, 224, 166, 195, 181, 121, 210, 59,
12, 152, 106, 167, 196, 75, 137, 225, 211, 240, 182, 122, 91, 28, 197, 13,
226, 168, 183, 153, 44, 212, 138, 107, 241, 60, 29, 123, 198, 184, 227, 169,
242, 76, 213, 154, 45, 92, 14, 199, 139, 61, 228, 214, 170, 185, 243, 108,
77, 155, 30, 15, 200, 229, 124, 215, 244, 93, 46, 186, 171, 201, 109, 140,
230, 62, 216, 245, 31, 125, 78, 156, 231, 47, 187, 202, 217, 94, 246, 141,
63, 232, 172, 110, 247, 157, 79, 218, 203, 126, 233, 188, 248, 95, 173, 142,
219, 111, 249, 234, 158, 127, 189, 204, 250, 235, 143, 174, 220, 205, 159,
251,
190, 221, 175, 236, 237, 191, 206, 252, 222, 253, 207, 238, 223, 254, 239,
255,
};
DECLARE_ALIGNED(16, static const int16_t, col_scan_16x16[256]) = {
0, 16, 32, 48, 1, 64, 17, 80, 33, 96, 49, 2, 65, 112, 18, 81,
34, 128, 50, 97, 3, 66, 144, 19, 113, 35, 82, 160, 98, 51, 129, 4,
67, 176, 20, 114, 145, 83, 36, 99, 130, 52, 192, 5, 161, 68, 115, 21,
146, 84, 208, 177, 37, 131, 100, 53, 162, 224, 69, 6, 116, 193, 147, 85,
22, 240, 132, 38, 178, 101, 163, 54, 209, 117, 70, 7, 148, 194, 86, 179,
225, 23, 133, 39, 164, 8, 102, 210, 241, 55, 195, 118, 149, 71, 180, 24,
87, 226, 134, 165, 211, 40, 103, 56, 72, 150, 196, 242, 119, 9, 181, 227,
88, 166, 25, 135, 41, 104, 212, 57, 151, 197, 120, 73, 243, 182, 136, 167,
213, 89, 10, 228, 105, 152, 198, 26, 42, 121, 183, 244, 168, 58, 137, 229,
74, 214, 90, 153, 199, 184, 11, 106, 245, 27, 122, 230, 169, 43, 215, 59,
200, 138, 185, 246, 75, 12, 91, 154, 216, 231, 107, 28, 44, 201, 123, 170,
60, 247, 232, 76, 139, 13, 92, 217, 186, 248, 155, 108, 29, 124, 45, 202,
233, 171, 61, 14, 77, 140, 15, 249, 93, 30, 187, 156, 218, 46, 109, 125,
62, 172, 78, 203, 31, 141, 234, 94, 47, 188, 63, 157, 110, 250, 219, 79,
126, 204, 173, 142, 95, 189, 111, 235, 158, 220, 251, 127, 174, 143, 205,
236,
159, 190, 221, 252, 175, 206, 237, 191, 253, 222, 238, 207, 254, 223, 239,
255,
};
DECLARE_ALIGNED(16, static const int16_t, row_scan_16x16[256]) = {
0, 1, 2, 16, 3, 17, 4, 18, 32, 5, 33, 19, 6, 34, 48, 20,
49, 7, 35, 21, 50, 64, 8, 36, 65, 22, 51, 37, 80, 9, 66, 52,
23, 38, 81, 67, 10, 53, 24, 82, 68, 96, 39, 11, 54, 83, 97, 69,
25, 98, 84, 40, 112, 55, 12, 70, 99, 113, 85, 26, 41, 56, 114, 100,
13, 71, 128, 86, 27, 115, 101, 129, 42, 57, 72, 116, 14, 87, 130, 102,
144, 73, 131, 117, 28, 58, 15, 88, 43, 145, 103, 132, 146, 118, 74, 160,
89, 133, 104, 29, 59, 147, 119, 44, 161, 148, 90, 105, 134, 162, 120, 176,
75, 135, 149, 30, 60, 163, 177, 45, 121, 91, 106, 164, 178, 150, 192, 136,
165, 179, 31, 151, 193, 76, 122, 61, 137, 194, 107, 152, 180, 208, 46, 166,
167, 195, 92, 181, 138, 209, 123, 153, 224, 196, 77, 168, 210, 182, 240, 108,
197, 62, 154, 225, 183, 169, 211, 47, 139, 93, 184, 226, 212, 241, 198, 170,
124, 155, 199, 78, 213, 185, 109, 227, 200, 63, 228, 242, 140, 214, 171, 186,
156, 229, 243, 125, 94, 201, 244, 215, 216, 230, 141, 187, 202, 79, 172, 110,
157, 245, 217, 231, 95, 246, 232, 126, 203, 247, 233, 173, 218, 142, 111,
158,
188, 248, 127, 234, 219, 249, 189, 204, 143, 174, 159, 250, 235, 205, 220,
175,
190, 251, 221, 191, 206, 236, 207, 237, 252, 222, 253, 223, 238, 239, 254,
255,
};
DECLARE_ALIGNED(16, static const int16_t, default_scan_32x32[1024]) = {
0, 32, 1, 64, 33, 2, 96, 65, 34, 128, 3, 97, 66, 160,
129, 35, 98, 4, 67, 130, 161, 192, 36, 99, 224, 5, 162, 193,
68, 131, 37, 100,
225, 194, 256, 163, 69, 132, 6, 226, 257, 288, 195, 101, 164, 38,
258, 7, 227, 289, 133, 320, 70, 196, 165, 290, 259, 228, 39, 321,
102, 352, 8, 197,
71, 134, 322, 291, 260, 353, 384, 229, 166, 103, 40, 354, 323, 292,
135, 385, 198, 261, 72, 9, 416, 167, 386, 355, 230, 324, 104, 293,
41, 417, 199, 136,
262, 387, 448, 325, 356, 10, 73, 418, 231, 168, 449, 294, 388, 105,
419, 263, 42, 200, 357, 450, 137, 480, 74, 326, 232, 11, 389, 169,
295, 420, 106, 451,
481, 358, 264, 327, 201, 43, 138, 512, 482, 390, 296, 233, 170, 421,
75, 452, 359, 12, 513, 265, 483, 328, 107, 202, 514, 544, 422, 391,
453, 139, 44, 234,
484, 297, 360, 171, 76, 515, 545, 266, 329, 454, 13, 423, 203, 108,
546, 485, 576, 298, 235, 140, 361, 330, 172, 547, 45, 455, 267, 577,
486, 77, 204, 362,
608, 14, 299, 578, 109, 236, 487, 609, 331, 141, 579, 46, 15, 173,
610, 363, 78, 205, 16, 110, 237, 611, 142, 47, 174, 79, 206, 17,
111, 238, 48, 143,
80, 175, 112, 207, 49, 18, 239, 81, 113, 19, 50, 82, 114, 51,
83, 115, 640, 516, 392, 268, 144, 20, 672, 641, 548, 517, 424,
393, 300, 269, 176, 145,
52, 21, 704, 673, 642, 580, 549, 518, 456, 425, 394, 332, 301,
270, 208, 177, 146, 84, 53, 22, 736, 705, 674, 643, 612, 581,
550, 519, 488, 457, 426, 395,
364, 333, 302, 271, 240, 209, 178, 147, 116, 85, 54, 23, 737,
706, 675, 613, 582, 551, 489, 458, 427, 365, 334, 303, 241,
210, 179, 117, 86, 55, 738, 707,
614, 583, 490, 459, 366, 335, 242, 211, 118, 87, 739, 615, 491,
367, 243, 119, 768, 644, 520, 396, 272, 148, 24, 800, 769, 676,
645, 552, 521, 428, 397, 304,
273, 180, 149, 56, 25, 832, 801, 770, 708, 677, 646, 584, 553,
522, 460, 429, 398, 336, 305, 274, 212, 181, 150, 88, 57, 26,
864, 833, 802, 771, 740, 709,
678, 647, 616, 585, 554, 523, 492, 461, 430, 399, 368, 337, 306,
275, 244, 213, 182, 151, 120, 89, 58, 27, 865, 834, 803, 741,
710, 679, 617, 586, 555, 493,
462, 431, 369, 338, 307, 245, 214, 183, 121, 90, 59, 866, 835,
742, 711, 618, 587, 494, 463, 370, 339, 246, 215, 122, 91, 867,
743, 619, 495, 371, 247, 123,
896, 772, 648, 524, 400, 276, 152, 28, 928, 897, 804, 773, 680,
649, 556, 525, 432, 401, 308, 277, 184, 153, 60, 29, 960, 929,
898, 836, 805, 774, 712, 681,
650, 588, 557, 526, 464, 433, 402, 340, 309, 278, 216, 185, 154,
92, 61, 30, 992, 961, 930, 899, 868, 837, 806, 775, 744, 713, 682,
651, 620, 589, 558, 527,
496, 465, 434, 403, 372, 341, 310, 279, 248, 217, 186, 155, 124,
93, 62, 31, 993, 962, 931, 869, 838, 807, 745, 714, 683, 621, 590,
559, 497, 466, 435, 373,
342, 311, 249, 218, 187, 125, 94, 63, 994, 963, 870, 839, 746, 715,
622, 591, 498, 467, 374, 343, 250, 219, 126, 95, 995, 871, 747, 623,
499, 375, 251, 127,
900, 776, 652, 528, 404, 280, 156, 932, 901, 808, 777, 684, 653, 560,
529, 436, 405, 312, 281, 188, 157, 964, 933, 902, 840, 809, 778, 716,
685, 654, 592, 561,
530, 468, 437, 406, 344, 313, 282, 220, 189, 158, 996, 965, 934, 903,
872, 841, 810, 779, 748, 717, 686, 655, 624, 593, 562, 531, 500, 469,
438, 407, 376, 345,
314, 283, 252, 221, 190, 159, 997, 966, 935, 873, 842, 811, 749, 718,
687, 625, 594, 563, 501, 470, 439, 377, 346, 315, 253, 222, 191, 998,
967, 874, 843, 750,
719, 626, 595, 502, 471, 378, 347, 254, 223, 999, 875, 751, 627, 503,
379, 255, 904, 780, 656, 532, 408, 284, 936, 905, 812, 781, 688, 657,
564, 533, 440, 409,
316, 285, 968, 937, 906, 844, 813, 782, 720, 689, 658, 596, 565, 534,
472, 441, 410, 348, 317, 286, 1000, 969, 938, 907, 876, 845, 814, 783,
752, 721, 690, 659,
628, 597, 566, 535, 504, 473, 442, 411, 380, 349, 318, 287, 1001, 970,
939, 877, 846, 815, 753, 722, 691, 629, 598, 567, 505, 474, 443, 381,
350, 319, 1002, 971,
878, 847, 754, 723, 630, 599, 506, 475, 382, 351, 1003, 879, 755, 631,
507, 383, 908, 784, 660, 536, 412, 940, 909, 816, 785, 692, 661, 568,
537, 444, 413, 972,
941, 910, 848, 817, 786, 724, 693, 662, 600, 569, 538, 476, 445, 414,
1004, 973, 942, 911, 880, 849, 818, 787, 756, 725, 694, 663, 632, 601,
570, 539, 508, 477,
446, 415, 1005, 974, 943, 881, 850, 819, 757, 726, 695, 633, 602, 571,
509, 478, 447, 1006, 975, 882, 851, 758, 727, 634, 603, 510, 479,
1007, 883, 759, 635, 511,
912, 788, 664, 540, 944, 913, 820, 789, 696, 665, 572, 541, 976, 945,
914, 852, 821, 790, 728, 697, 666, 604, 573, 542, 1008, 977, 946, 915,
884, 853, 822, 791,
760, 729, 698, 667, 636, 605, 574, 543, 1009, 978, 947, 885, 854, 823,
761, 730, 699, 637, 606, 575, 1010, 979, 886, 855, 762, 731, 638, 607,
1011, 887, 763, 639,
916, 792, 668, 948, 917, 824, 793, 700, 669, 980, 949, 918, 856, 825,
794, 732, 701, 670, 1012, 981, 950, 919, 888, 857, 826, 795, 764, 733,
702, 671, 1013, 982,
951, 889, 858, 827, 765, 734, 703, 1014, 983, 890, 859, 766, 735, 1015,
891, 767, 920, 796, 952, 921, 828, 797, 984, 953, 922, 860, 829, 798,
1016, 985, 954, 923,
892, 861, 830, 799, 1017, 986, 955, 893, 862, 831, 1018, 987, 894, 863,
1019, 895, 924, 956, 925, 988, 957, 926, 1020, 989, 958, 927, 1021,
990, 959, 1022, 991, 1023,
};
// Neighborhood 5-tuples for various scans and blocksizes,
// in {top, left, topleft, topright, bottomleft} order
// for each position in raster scan order.
// -1 indicates the neighbor does not exist.
DECLARE_ALIGNED(16, static const int16_t,
default_scan_4x4_neighbors[17 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 0, 1, 4, 4, 4, 1, 1, 8, 8, 5, 8, 2, 2, 2, 5, 9, 12, 6, 9,
3, 6, 10, 13, 7, 10, 11, 14, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
col_scan_4x4_neighbors[17 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 4, 4, 0, 0, 8, 8, 1, 1, 5, 5, 1, 1, 9, 9, 2, 2, 6, 6, 2, 2, 3,
3, 10, 10, 7, 7, 11, 11, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
row_scan_4x4_neighbors[17 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 0, 1, 1, 4, 4, 2, 2, 5, 5, 4, 4, 8, 8, 6, 6, 8, 8, 9, 9, 12,
12, 10, 10, 13, 13, 14, 14, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
col_scan_8x8_neighbors[65 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 8, 8, 0, 0, 16, 16, 1, 1, 24, 24, 9, 9, 1, 1, 32, 32, 17, 17, 2,
2, 25, 25, 10, 10, 40, 40, 2, 2, 18, 18, 33, 33, 3, 3, 48, 48, 11, 11, 26,
26, 3, 3, 41, 41, 19, 19, 34, 34, 4, 4, 27, 27, 12, 12, 49, 49, 42, 42, 20,
20, 4, 4, 35, 35, 5, 5, 28, 28, 50, 50, 43, 43, 13, 13, 36, 36, 5, 5, 21, 21,
51, 51, 29, 29, 6, 6, 44, 44, 14, 14, 6, 6, 37, 37, 52, 52, 22, 22, 7, 7, 30,
30, 45, 45, 15, 15, 38, 38, 23, 23, 53, 53, 31, 31, 46, 46, 39, 39, 54, 54,
47, 47, 55, 55, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
row_scan_8x8_neighbors[65 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 1, 1, 0, 0, 8, 8, 2, 2, 8, 8, 9, 9, 3, 3, 16, 16, 10, 10, 16, 16,
4, 4, 17, 17, 24, 24, 11, 11, 18, 18, 25, 25, 24, 24, 5, 5, 12, 12, 19, 19,
32, 32, 26, 26, 6, 6, 33, 33, 32, 32, 20, 20, 27, 27, 40, 40, 13, 13, 34, 34,
40, 40, 41, 41, 28, 28, 35, 35, 48, 48, 21, 21, 42, 42, 14, 14, 48, 48, 36,
36, 49, 49, 43, 43, 29, 29, 56, 56, 22, 22, 50, 50, 57, 57, 44, 44, 37, 37,
51, 51, 30, 30, 58, 58, 52, 52, 45, 45, 59, 59, 38, 38, 60, 60, 46, 46, 53,
53, 54, 54, 61, 61, 62, 62, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
default_scan_8x8_neighbors[65 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 0, 8, 8, 1, 8, 1, 1, 9, 16, 16, 16, 2, 9, 2, 2, 10, 17, 17,
24, 24, 24, 3, 10, 3, 3, 18, 25, 25, 32, 11, 18, 32, 32, 4, 11, 26, 33, 19,
26, 4, 4, 33, 40, 12, 19, 40, 40, 5, 12, 27, 34, 34, 41, 20, 27, 13, 20, 5,
5, 41, 48, 48, 48, 28, 35, 35, 42, 21, 28, 6, 6, 6, 13, 42, 49, 49, 56, 36,
43, 14, 21, 29, 36, 7, 14, 43, 50, 50, 57, 22, 29, 37, 44, 15, 22, 44, 51,
51, 58, 30, 37, 23, 30, 52, 59, 45, 52, 38, 45, 31, 38, 53, 60, 46, 53, 39,
46, 54, 61, 47, 54, 55, 62, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
col_scan_16x16_neighbors[257 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 16, 16, 32, 32, 0, 0, 48, 48, 1, 1, 64, 64,
17, 17, 80, 80, 33, 33, 1, 1, 49, 49, 96, 96, 2, 2, 65, 65,
18, 18, 112, 112, 34, 34, 81, 81, 2, 2, 50, 50, 128, 128, 3, 3,
97, 97, 19, 19, 66, 66, 144, 144, 82, 82, 35, 35, 113, 113, 3, 3,
51, 51, 160, 160, 4, 4, 98, 98, 129, 129, 67, 67, 20, 20, 83, 83,
114, 114, 36, 36, 176, 176, 4, 4, 145, 145, 52, 52, 99, 99, 5, 5,
130, 130, 68, 68, 192, 192, 161, 161, 21, 21, 115, 115, 84, 84, 37, 37,
146, 146, 208, 208, 53, 53, 5, 5, 100, 100, 177, 177, 131, 131, 69, 69,
6, 6, 224, 224, 116, 116, 22, 22, 162, 162, 85, 85, 147, 147, 38, 38,
193, 193, 101, 101, 54, 54, 6, 6, 132, 132, 178, 178, 70, 70, 163, 163,
209, 209, 7, 7, 117, 117, 23, 23, 148, 148, 7, 7, 86, 86, 194, 194,
225, 225, 39, 39, 179, 179, 102, 102, 133, 133, 55, 55, 164, 164, 8, 8,
71, 71, 210, 210, 118, 118, 149, 149, 195, 195, 24, 24, 87, 87, 40, 40,
56, 56, 134, 134, 180, 180, 226, 226, 103, 103, 8, 8, 165, 165, 211, 211,
72, 72, 150, 150, 9, 9, 119, 119, 25, 25, 88, 88, 196, 196, 41, 41,
135, 135, 181, 181, 104, 104, 57, 57, 227, 227, 166, 166, 120, 120, 151, 151,
197, 197, 73, 73, 9, 9, 212, 212, 89, 89, 136, 136, 182, 182, 10, 10,
26, 26, 105, 105, 167, 167, 228, 228, 152, 152, 42, 42, 121, 121, 213, 213,
58, 58, 198, 198, 74, 74, 137, 137, 183, 183, 168, 168, 10, 10, 90, 90,
229, 229, 11, 11, 106, 106, 214, 214, 153, 153, 27, 27, 199, 199, 43, 43,
184, 184, 122, 122, 169, 169, 230, 230, 59, 59, 11, 11, 75, 75, 138, 138,
200, 200, 215, 215, 91, 91, 12, 12, 28, 28, 185, 185, 107, 107, 154, 154,
44, 44, 231, 231, 216, 216, 60, 60, 123, 123, 12, 12, 76, 76, 201, 201,
170, 170, 232, 232, 139, 139, 92, 92, 13, 13, 108, 108, 29, 29, 186, 186,
217, 217, 155, 155, 45, 45, 13, 13, 61, 61, 124, 124, 14, 14, 233, 233,
77, 77, 14, 14, 171, 171, 140, 140, 202, 202, 30, 30, 93, 93, 109, 109,
46, 46, 156, 156, 62, 62, 187, 187, 15, 15, 125, 125, 218, 218, 78, 78,
31, 31, 172, 172, 47, 47, 141, 141, 94, 94, 234, 234, 203, 203, 63, 63,
110, 110, 188, 188, 157, 157, 126, 126, 79, 79, 173, 173, 95, 95, 219, 219,
142, 142, 204, 204, 235, 235, 111, 111, 158, 158, 127, 127, 189, 189, 220,
220, 143, 143, 174, 174, 205, 205, 236, 236, 159, 159, 190, 190, 221, 221,
175, 175, 237, 237, 206, 206, 222, 222, 191, 191, 238, 238, 207, 207, 223,
223, 239, 239, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
row_scan_16x16_neighbors[257 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 1, 1, 0, 0, 2, 2, 16, 16, 3, 3, 17, 17,
16, 16, 4, 4, 32, 32, 18, 18, 5, 5, 33, 33, 32, 32, 19, 19,
48, 48, 6, 6, 34, 34, 20, 20, 49, 49, 48, 48, 7, 7, 35, 35,
64, 64, 21, 21, 50, 50, 36, 36, 64, 64, 8, 8, 65, 65, 51, 51,
22, 22, 37, 37, 80, 80, 66, 66, 9, 9, 52, 52, 23, 23, 81, 81,
67, 67, 80, 80, 38, 38, 10, 10, 53, 53, 82, 82, 96, 96, 68, 68,
24, 24, 97, 97, 83, 83, 39, 39, 96, 96, 54, 54, 11, 11, 69, 69,
98, 98, 112, 112, 84, 84, 25, 25, 40, 40, 55, 55, 113, 113, 99, 99,
12, 12, 70, 70, 112, 112, 85, 85, 26, 26, 114, 114, 100, 100, 128, 128,
41, 41, 56, 56, 71, 71, 115, 115, 13, 13, 86, 86, 129, 129, 101, 101,
128, 128, 72, 72, 130, 130, 116, 116, 27, 27, 57, 57, 14, 14, 87, 87,
42, 42, 144, 144, 102, 102, 131, 131, 145, 145, 117, 117, 73, 73, 144, 144,
88, 88, 132, 132, 103, 103, 28, 28, 58, 58, 146, 146, 118, 118, 43, 43,
160, 160, 147, 147, 89, 89, 104, 104, 133, 133, 161, 161, 119, 119, 160, 160,
74, 74, 134, 134, 148, 148, 29, 29, 59, 59, 162, 162, 176, 176, 44, 44,
120, 120, 90, 90, 105, 105, 163, 163, 177, 177, 149, 149, 176, 176, 135, 135,
164, 164, 178, 178, 30, 30, 150, 150, 192, 192, 75, 75, 121, 121, 60, 60,
136, 136, 193, 193, 106, 106, 151, 151, 179, 179, 192, 192, 45, 45, 165, 165,
166, 166, 194, 194, 91, 91, 180, 180, 137, 137, 208, 208, 122, 122, 152, 152,
208, 208, 195, 195, 76, 76, 167, 167, 209, 209, 181, 181, 224, 224, 107, 107,
196, 196, 61, 61, 153, 153, 224, 224, 182, 182, 168, 168, 210, 210, 46, 46,
138, 138, 92, 92, 183, 183, 225, 225, 211, 211, 240, 240, 197, 197, 169, 169,
123, 123, 154, 154, 198, 198, 77, 77, 212, 212, 184, 184, 108, 108, 226, 226,
199, 199, 62, 62, 227, 227, 241, 241, 139, 139, 213, 213, 170, 170, 185, 185,
155, 155, 228, 228, 242, 242, 124, 124, 93, 93, 200, 200, 243, 243, 214, 214,
215, 215, 229, 229, 140, 140, 186, 186, 201, 201, 78, 78, 171, 171, 109, 109,
156, 156, 244, 244, 216, 216, 230, 230, 94, 94, 245, 245, 231, 231, 125, 125,
202, 202, 246, 246, 232, 232, 172, 172, 217, 217, 141, 141, 110, 110, 157,
157, 187, 187, 247, 247, 126, 126, 233, 233, 218, 218, 248, 248, 188, 188,
203, 203, 142, 142, 173, 173, 158, 158, 249, 249, 234, 234, 204, 204, 219,
219, 174, 174, 189, 189, 250, 250, 220, 220, 190, 190, 205, 205, 235, 235,
206, 206, 236, 236, 251, 251, 221, 221, 252, 252, 222, 222, 237, 237, 238,
238, 253, 253, 254, 254, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
default_scan_16x16_neighbors[257 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 0, 16, 16, 1, 16, 1, 1, 32, 32, 17, 32,
2, 17, 2, 2, 48, 48, 18, 33, 33, 48, 3, 18, 49, 64, 64, 64,
34, 49, 3, 3, 19, 34, 50, 65, 4, 19, 65, 80, 80, 80, 35, 50,
4, 4, 20, 35, 66, 81, 81, 96, 51, 66, 96, 96, 5, 20, 36, 51,
82, 97, 21, 36, 67, 82, 97, 112, 5, 5, 52, 67, 112, 112, 37, 52,
6, 21, 83, 98, 98, 113, 68, 83, 6, 6, 113, 128, 22, 37, 53, 68,
84, 99, 99, 114, 128, 128, 114, 129, 69, 84, 38, 53, 7, 22, 7, 7,
129, 144, 23, 38, 54, 69, 100, 115, 85, 100, 115, 130, 144, 144, 130, 145,
39, 54, 70, 85, 8, 23, 55, 70, 116, 131, 101, 116, 145, 160, 24, 39,
8, 8, 86, 101, 131, 146, 160, 160, 146, 161, 71, 86, 40, 55, 9, 24,
117, 132, 102, 117, 161, 176, 132, 147, 56, 71, 87, 102, 25, 40, 147, 162,
9, 9, 176, 176, 162, 177, 72, 87, 41, 56, 118, 133, 133, 148, 103, 118,
10, 25, 148, 163, 57, 72, 88, 103, 177, 192, 26, 41, 163, 178, 192, 192,
10, 10, 119, 134, 73, 88, 149, 164, 104, 119, 134, 149, 42, 57, 178, 193,
164, 179, 11, 26, 58, 73, 193, 208, 89, 104, 135, 150, 120, 135, 27, 42,
74, 89, 208, 208, 150, 165, 179, 194, 165, 180, 105, 120, 194, 209, 43, 58,
11, 11, 136, 151, 90, 105, 151, 166, 180, 195, 59, 74, 121, 136, 209, 224,
195, 210, 224, 224, 166, 181, 106, 121, 75, 90, 12, 27, 181, 196, 12, 12,
210, 225, 152, 167, 167, 182, 137, 152, 28, 43, 196, 211, 122, 137, 91, 106,
225, 240, 44, 59, 13, 28, 107, 122, 182, 197, 168, 183, 211, 226, 153, 168,
226, 241, 60, 75, 197, 212, 138, 153, 29, 44, 76, 91, 13, 13, 183, 198,
123, 138, 45, 60, 212, 227, 198, 213, 154, 169, 169, 184, 227, 242, 92, 107,
61, 76, 139, 154, 14, 29, 14, 14, 184, 199, 213, 228, 108, 123, 199, 214,
228, 243, 77, 92, 30, 45, 170, 185, 155, 170, 185, 200, 93, 108, 124, 139,
214, 229, 46, 61, 200, 215, 229, 244, 15, 30, 109, 124, 62, 77, 140, 155,
215, 230, 31, 46, 171, 186, 186, 201, 201, 216, 78, 93, 230, 245, 125, 140,
47, 62, 216, 231, 156, 171, 94, 109, 231, 246, 141, 156, 63, 78, 202, 217,
187, 202, 110, 125, 217, 232, 172, 187, 232, 247, 79, 94, 157, 172, 126, 141,
203, 218, 95, 110, 233, 248, 218, 233, 142, 157, 111, 126, 173, 188, 188, 203,
234, 249, 219, 234, 127, 142, 158, 173, 204, 219, 189, 204, 143, 158, 235,
250, 174, 189, 205, 220, 159, 174, 220, 235, 221, 236, 175, 190, 190, 205,
236, 251, 206, 221, 237, 252, 191, 206, 222, 237, 207, 222, 238, 253, 223,
238, 239, 254, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t,
default_scan_32x32_neighbors[1025 * MAX_NEIGHBORS]) = {
0, 0, 0, 0, 0, 0, 32, 32, 1, 32, 1, 1, 64, 64, 33, 64,
2, 33, 96, 96, 2, 2, 65, 96, 34, 65, 128, 128, 97, 128, 3, 34,
66, 97, 3, 3, 35, 66, 98, 129, 129, 160, 160, 160, 4, 35, 67, 98,
192, 192, 4, 4, 130, 161, 161, 192, 36, 67, 99, 130, 5, 36, 68, 99,
193, 224, 162, 193, 224, 224, 131, 162, 37, 68, 100, 131, 5, 5, 194, 225,
225, 256, 256, 256, 163, 194, 69, 100, 132, 163, 6, 37, 226, 257, 6, 6,
195, 226, 257, 288, 101, 132, 288, 288, 38, 69, 164, 195, 133, 164, 258, 289,
227, 258, 196, 227, 7, 38, 289, 320, 70, 101, 320, 320, 7, 7, 165, 196,
39, 70, 102, 133, 290, 321, 259, 290, 228, 259, 321, 352, 352, 352, 197, 228,
134, 165, 71, 102, 8, 39, 322, 353, 291, 322, 260, 291, 103, 134, 353, 384,
166, 197, 229, 260, 40, 71, 8, 8, 384, 384, 135, 166, 354, 385, 323, 354,
198, 229, 292, 323, 72, 103, 261, 292, 9, 40, 385, 416, 167, 198, 104, 135,
230, 261, 355, 386, 416, 416, 293, 324, 324, 355, 9, 9, 41, 72, 386, 417,
199, 230, 136, 167, 417, 448, 262, 293, 356, 387, 73, 104, 387, 418, 231, 262,
10, 41, 168, 199, 325, 356, 418, 449, 105, 136, 448, 448, 42, 73, 294, 325,
200, 231, 10, 10, 357, 388, 137, 168, 263, 294, 388, 419, 74, 105, 419, 450,
449, 480, 326, 357, 232, 263, 295, 326, 169, 200, 11, 42, 106, 137, 480, 480,
450, 481, 358, 389, 264, 295, 201, 232, 138, 169, 389, 420, 43, 74, 420, 451,
327, 358, 11, 11, 481, 512, 233, 264, 451, 482, 296, 327, 75, 106, 170, 201,
482, 513, 512, 512, 390, 421, 359, 390, 421, 452, 107, 138, 12, 43, 202, 233,
452, 483, 265, 296, 328, 359, 139, 170, 44, 75, 483, 514, 513, 544, 234, 265,
297, 328, 422, 453, 12, 12, 391, 422, 171, 202, 76, 107, 514, 545, 453, 484,
544, 544, 266, 297, 203, 234, 108, 139, 329, 360, 298, 329, 140, 171, 515,
546, 13, 44, 423, 454, 235, 266, 545, 576, 454, 485, 45, 76, 172, 203, 330,
361, 576, 576, 13, 13, 267, 298, 546, 577, 77, 108, 204, 235, 455, 486, 577,
608, 299, 330, 109, 140, 547, 578, 14, 45, 14, 14, 141, 172, 578, 609, 331,
362, 46, 77, 173, 204, 15, 15, 78, 109, 205, 236, 579, 610, 110, 141, 15, 46,
142, 173, 47, 78, 174, 205, 16, 16, 79, 110, 206, 237, 16, 47, 111, 142,
48, 79, 143, 174, 80, 111, 175, 206, 17, 48, 17, 17, 207, 238, 49, 80,
81, 112, 18, 18, 18, 49, 50, 81, 82, 113, 19, 50, 51, 82, 83, 114, 608, 608,
484, 515, 360, 391, 236, 267, 112, 143, 19, 19, 640, 640, 609, 640, 516, 547,
485, 516, 392, 423, 361, 392, 268, 299, 237, 268, 144, 175, 113, 144, 20, 51,
20, 20, 672, 672, 641, 672, 610, 641, 548, 579, 517, 548, 486, 517, 424, 455,
393, 424, 362, 393, 300, 331, 269, 300, 238, 269, 176, 207, 145, 176, 114,
145, 52, 83, 21, 52, 21, 21, 704, 704, 673, 704, 642, 673, 611, 642, 580,
611, 549, 580, 518, 549, 487, 518, 456, 487, 425, 456, 394, 425, 363, 394,
332, 363, 301, 332, 270, 301, 239, 270, 208, 239, 177, 208, 146, 177, 115,
146, 84, 115, 53, 84, 22, 53, 22, 22, 705, 736, 674, 705, 643, 674, 581, 612,
550, 581, 519, 550, 457, 488, 426, 457, 395, 426, 333, 364, 302, 333, 271,
302, 209, 240, 178, 209, 147, 178, 85, 116, 54, 85, 23, 54, 706, 737, 675,
706, 582, 613, 551, 582, 458, 489, 427, 458, 334, 365, 303, 334, 210, 241,
179, 210, 86, 117, 55, 86, 707, 738, 583, 614, 459, 490, 335, 366, 211, 242,
87, 118, 736, 736, 612, 643, 488, 519, 364, 395, 240, 271, 116, 147, 23, 23,
768, 768, 737, 768, 644, 675, 613, 644, 520, 551, 489, 520, 396, 427, 365,
396, 272, 303, 241, 272, 148, 179, 117, 148, 24, 55, 24, 24, 800, 800, 769,
800, 738, 769, 676, 707, 645, 676, 614, 645, 552, 583, 521, 552, 490, 521,
428, 459, 397, 428, 366, 397, 304, 335, 273, 304, 242, 273, 180, 211, 149,
180, 118, 149, 56, 87, 25, 56, 25, 25, 832, 832, 801, 832, 770, 801, 739,
770, 708, 739, 677, 708, 646, 677, 615, 646, 584, 615, 553, 584, 522, 553,
491, 522, 460, 491, 429, 460, 398, 429, 367, 398, 336, 367, 305, 336, 274,
305, 243, 274, 212, 243, 181, 212, 150, 181, 119, 150, 88, 119, 57, 88, 26,
57, 26, 26, 833, 864, 802, 833, 771, 802, 709, 740, 678, 709, 647, 678, 585,
616, 554, 585, 523, 554, 461, 492, 430, 461, 399, 430, 337, 368, 306, 337,
275, 306, 213, 244, 182, 213, 151, 182, 89, 120, 58, 89, 27, 58, 834, 865,
803, 834, 710, 741, 679, 710, 586, 617, 555, 586, 462, 493, 431, 462, 338,
369, 307, 338, 214, 245, 183, 214, 90, 121, 59, 90, 835, 866, 711, 742, 587,
618, 463, 494, 339, 370, 215, 246, 91, 122, 864, 864, 740, 771, 616, 647,
492, 523, 368, 399, 244, 275, 120, 151, 27, 27, 896, 896, 865, 896, 772, 803,
741, 772, 648, 679, 617, 648, 524, 555, 493, 524, 400, 431, 369, 400, 276,
307, 245, 276, 152, 183, 121, 152, 28, 59, 28, 28, 928, 928, 897, 928, 866,
897, 804, 835, 773, 804, 742, 773, 680, 711, 649, 680, 618, 649, 556, 587,
525, 556, 494, 525, 432, 463, 401, 432, 370, 401, 308, 339, 277, 308, 246,
277, 184, 215, 153, 184, 122, 153, 60, 91, 29, 60, 29, 29, 960, 960, 929,
960, 898, 929, 867, 898, 836, 867, 805, 836, 774, 805, 743, 774, 712, 743,
681, 712, 650, 681, 619, 650, 588, 619, 557, 588, 526, 557, 495, 526, 464,
495, 433, 464, 402, 433, 371, 402, 340, 371, 309, 340, 278, 309, 247, 278,
216, 247, 185, 216, 154, 185, 123, 154, 92, 123, 61, 92, 30, 61, 30, 30,
961, 992, 930, 961, 899, 930, 837, 868, 806, 837, 775, 806, 713, 744, 682,
713, 651, 682, 589, 620, 558, 589, 527, 558, 465, 496, 434, 465, 403, 434,
341, 372, 310, 341, 279, 310, 217, 248, 186, 217, 155, 186, 93, 124, 62, 93,
31, 62, 962, 993, 931, 962, 838, 869, 807, 838, 714, 745, 683, 714, 590, 621,
559, 590, 466, 497, 435, 466, 342, 373, 311, 342, 218, 249, 187, 218, 94,
125, 63, 94, 963, 994, 839, 870, 715, 746, 591, 622, 467, 498, 343, 374, 219,
250, 95, 126, 868, 899, 744, 775, 620, 651, 496, 527, 372, 403, 248, 279,
124, 155, 900, 931, 869, 900, 776, 807, 745, 776, 652, 683, 621, 652, 528,
559, 497, 528, 404, 435, 373, 404, 280, 311, 249, 280, 156, 187, 125, 156,
932, 963, 901, 932, 870, 901, 808, 839, 777, 808, 746, 777, 684, 715, 653,
684, 622, 653, 560, 591, 529, 560, 498, 529, 436, 467, 405, 436, 374, 405,
312, 343, 281, 312, 250, 281, 188, 219, 157, 188, 126, 157, 964, 995, 933,
964, 902, 933, 871, 902, 840, 871, 809, 840, 778, 809, 747, 778, 716, 747,
685, 716, 654, 685, 623, 654, 592, 623, 561, 592, 530, 561, 499, 530, 468,
499, 437, 468, 406, 437, 375, 406, 344, 375, 313, 344, 282, 313, 251, 282,
220, 251, 189, 220, 158, 189, 127, 158, 965, 996, 934, 965, 903, 934, 841,
872, 810, 841, 779, 810, 717, 748, 686, 717, 655, 686, 593, 624, 562, 593,
531, 562, 469, 500, 438, 469, 407, 438, 345, 376, 314, 345, 283, 314, 221,
252, 190, 221, 159, 190, 966, 997, 935, 966, 842, 873, 811, 842, 718, 749,
687, 718, 594, 625, 563, 594, 470, 501, 439, 470, 346, 377, 315, 346, 222,
253, 191, 222, 967, 998, 843, 874, 719, 750, 595, 626, 471, 502, 347, 378,
223, 254, 872, 903, 748, 779, 624, 655, 500, 531, 376, 407, 252, 283, 904,
935, 873, 904, 780, 811, 749, 780, 656, 687, 625, 656, 532, 563, 501, 532,
408, 439, 377, 408, 284, 315, 253, 284, 936, 967, 905, 936, 874, 905, 812,
843, 781, 812, 750, 781, 688, 719, 657, 688, 626, 657, 564, 595, 533, 564,
502, 533, 440, 471, 409, 440, 378, 409, 316, 347, 285, 316, 254, 285, 968,
999, 937, 968, 906, 937, 875, 906, 844, 875, 813, 844, 782, 813, 751, 782,
720, 751, 689, 720, 658, 689, 627, 658, 596, 627, 565, 596, 534, 565, 503,
534, 472, 503, 441, 472, 410, 441, 379, 410, 348, 379, 317, 348, 286, 317,
255, 286, 969, 1000, 938, 969, 907, 938, 845, 876, 814, 845, 783, 814, 721,
752, 690, 721, 659, 690, 597, 628, 566, 597, 535, 566, 473, 504, 442, 473,
411, 442, 349, 380, 318, 349, 287, 318, 970, 1001, 939, 970, 846, 877, 815,
846, 722, 753, 691, 722, 598, 629, 567, 598, 474, 505, 443, 474, 350, 381,
319, 350, 971, 1002, 847, 878, 723, 754, 599, 630, 475, 506, 351, 382, 876,
907, 752, 783, 628, 659, 504, 535, 380, 411, 908, 939, 877, 908, 784, 815,
753, 784, 660, 691, 629, 660, 536, 567, 505, 536, 412, 443, 381, 412, 940,
971, 909, 940, 878, 909, 816, 847, 785, 816, 754, 785, 692, 723, 661, 692,
630, 661, 568, 599, 537, 568, 506, 537, 444, 475, 413, 444, 382, 413, 972,
1003, 941, 972, 910, 941, 879, 910, 848, 879, 817, 848, 786, 817, 755, 786,
724, 755, 693, 724, 662, 693, 631, 662, 600, 631, 569, 600, 538, 569, 507,
538, 476, 507, 445, 476, 414, 445, 383, 414, 973, 1004, 942, 973, 911, 942,
849, 880, 818, 849, 787, 818, 725, 756, 694, 725, 663, 694, 601, 632, 570,
601, 539, 570, 477, 508, 446, 477, 415, 446, 974, 1005, 943, 974, 850, 881,
819, 850, 726, 757, 695, 726, 602, 633, 571, 602, 478, 509, 447, 478, 975,
1006, 851, 882, 727, 758, 603, 634, 479, 510, 880, 911, 756, 787, 632, 663,
508, 539, 912, 943, 881, 912, 788, 819, 757, 788, 664, 695, 633, 664, 540,
571, 509, 540, 944, 975, 913, 944, 882, 913, 820, 851, 789, 820, 758, 789,
696, 727, 665, 696, 634, 665, 572, 603, 541, 572, 510, 541, 976, 1007, 945,
976, 914, 945, 883, 914, 852, 883, 821, 852, 790, 821, 759, 790, 728, 759,
697, 728, 666, 697, 635, 666, 604, 635, 573, 604, 542, 573, 511, 542, 977,
1008, 946, 977, 915, 946, 853, 884, 822, 853, 791, 822, 729, 760, 698, 729,
667, 698, 605, 636, 574, 605, 543, 574, 978, 1009, 947, 978, 854, 885, 823,
854, 730, 761, 699, 730, 606, 637, 575, 606, 979, 1010, 855, 886, 731, 762,
607, 638, 884, 915, 760, 791, 636, 667, 916, 947, 885, 916, 792, 823, 761,
792, 668, 699, 637, 668, 948, 979, 917, 948, 886, 917, 824, 855, 793, 824,
762, 793, 700, 731, 669, 700, 638, 669, 980, 1011, 949, 980, 918, 949, 887,
918, 856, 887, 825, 856, 794, 825, 763, 794, 732, 763, 701, 732, 670, 701,
639, 670, 981, 1012, 950, 981, 919, 950, 857, 888, 826, 857, 795, 826, 733,
764, 702, 733, 671, 702, 982, 1013, 951, 982, 858, 889, 827, 858, 734, 765,
703, 734, 983, 1014, 859, 890, 735, 766, 888, 919, 764, 795, 920, 951, 889,
920, 796, 827, 765, 796, 952, 983, 921, 952, 890, 921, 828, 859, 797, 828,
766, 797, 984, 1015, 953, 984, 922, 953, 891, 922, 860, 891, 829, 860, 798,
829, 767, 798, 985, 1016, 954, 985, 923, 954, 861, 892, 830, 861, 799, 830,
986, 1017, 955, 986, 862, 893, 831, 862, 987, 1018, 863, 894, 892, 923, 924,
955, 893, 924, 956, 987, 925, 956, 894, 925, 988, 1019, 957, 988, 926, 957,
895, 926, 989, 1020, 958, 989, 927, 958, 990, 1021, 959, 990, 991, 1022, 0, 0,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_default_iscan_4x4[16]) = {
0, 2, 5, 8, 1, 3, 9, 12, 4, 7, 11, 14, 6, 10, 13, 15,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_col_iscan_4x4[16]) = {
0, 3, 7, 11, 1, 5, 9, 12, 2, 6, 10, 14, 4, 8, 13, 15,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_row_iscan_4x4[16]) = {
0, 1, 3, 5, 2, 4, 6, 9, 7, 8, 11, 13, 10, 12, 14, 15,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_col_iscan_8x8[64]) = {
0, 3, 8, 15, 22, 32, 40, 47, 1, 5, 11, 18, 26, 34, 44, 51,
2, 7, 13, 20, 28, 38, 46, 54, 4, 10, 16, 24, 31, 41, 50, 56,
6, 12, 21, 27, 35, 43, 52, 58, 9, 17, 25, 33, 39, 48, 55, 60,
14, 23, 30, 37, 45, 53, 59, 62, 19, 29, 36, 42, 49, 57, 61, 63,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_row_iscan_8x8[64]) = {
0, 1, 2, 5, 8, 12, 19, 24, 3, 4, 7, 10, 15, 20, 30, 39,
6, 9, 13, 16, 21, 27, 37, 46, 11, 14, 17, 23, 28, 34, 44, 52,
18, 22, 25, 31, 35, 41, 50, 57, 26, 29, 33, 38, 43, 49, 55, 59,
32, 36, 42, 47, 51, 54, 60, 61, 40, 45, 48, 53, 56, 58, 62, 63,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_default_iscan_8x8[64]) = {
0, 2, 5, 9, 14, 22, 31, 37, 1, 4, 8, 13, 19, 26, 38, 44,
3, 6, 10, 17, 24, 30, 42, 49, 7, 11, 15, 21, 29, 36, 47, 53,
12, 16, 20, 27, 34, 43, 52, 57, 18, 23, 28, 35, 41, 48, 56, 60,
25, 32, 39, 45, 50, 55, 59, 62, 33, 40, 46, 51, 54, 58, 61, 63,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_col_iscan_16x16[256]) = {
0, 4, 11, 20, 31, 43, 59, 75, 85, 109, 130, 150, 165, 181, 195, 198,
1, 6, 14, 23, 34, 47, 64, 81, 95, 114, 135, 153, 171, 188, 201, 212,
2, 8, 16, 25, 38, 52, 67, 83, 101, 116, 136, 157, 172, 190, 205, 216,
3, 10, 18, 29, 41, 55, 71, 89, 103, 119, 141, 159, 176, 194, 208, 218,
5, 12, 21, 32, 45, 58, 74, 93, 104, 123, 144, 164, 179, 196, 210, 223,
7, 15, 26, 37, 49, 63, 78, 96, 112, 129, 146, 166, 182, 200, 215, 228,
9, 19, 28, 39, 54, 69, 86, 102, 117, 132, 151, 170, 187, 206, 220, 230,
13, 24, 35, 46, 60, 73, 91, 108, 122, 137, 154, 174, 189, 207, 224, 235,
17, 30, 40, 53, 66, 82, 98, 115, 126, 142, 161, 180, 197, 213, 227, 237,
22, 36, 48, 62, 76, 92, 105, 120, 133, 147, 167, 186, 203, 219, 232, 240,
27, 44, 56, 70, 84, 99, 113, 127, 140, 156, 175, 193, 209, 226, 236, 244,
33, 51, 68, 79, 94, 110, 125, 138, 149, 162, 184, 202, 217, 229, 241, 247,
42, 61, 77, 90, 106, 121, 134, 148, 160, 173, 191, 211, 225, 238, 245, 251,
50, 72, 87, 100, 118, 128, 145, 158, 168, 183, 204, 222, 233, 242, 249, 253,
57, 80, 97, 111, 131, 143, 155, 169, 178, 192, 214, 231, 239, 246, 250, 254,
65, 88, 107, 124, 139, 152, 163, 177, 185, 199, 221, 234, 243, 248, 252, 255,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_row_iscan_16x16[256]) = {
0, 1, 2, 4, 6, 9, 12, 17, 22, 29, 36, 43, 54, 64, 76, 86,
3, 5, 7, 11, 15, 19, 25, 32, 38, 48, 59, 68, 84, 99, 115, 130,
8, 10, 13, 18, 23, 27, 33, 42, 51, 60, 72, 88, 103, 119, 142, 167,
14, 16, 20, 26, 31, 37, 44, 53, 61, 73, 85, 100, 116, 135, 161, 185,
21, 24, 30, 35, 40, 47, 55, 65, 74, 81, 94, 112, 133, 154, 179, 205,
28, 34, 39, 45, 50, 58, 67, 77, 87, 96, 106, 121, 146, 169, 196, 212,
41, 46, 49, 56, 63, 70, 79, 90, 98, 107, 122, 138, 159, 182, 207, 222,
52, 57, 62, 69, 75, 83, 93, 102, 110, 120, 134, 150, 176, 195, 215, 226,
66, 71, 78, 82, 91, 97, 108, 113, 127, 136, 148, 168, 188, 202, 221, 232,
80, 89, 92, 101, 105, 114, 125, 131, 139, 151, 162, 177, 192, 208, 223, 234,
95, 104, 109, 117, 123, 128, 143, 144, 155, 165, 175, 190, 206, 219, 233, 239,
111, 118, 124, 129, 140, 147, 157, 164, 170, 181, 191, 203, 224, 230, 240,
243, 126, 132, 137, 145, 153, 160, 174, 178, 184, 197, 204, 216, 231, 237,
244, 246, 141, 149, 156, 166, 172, 180, 189, 199, 200, 210, 220, 228, 238,
242, 249, 251, 152, 163, 171, 183, 186, 193, 201, 211, 214, 218, 227, 236,
245, 247, 252, 253, 158, 173, 187, 194, 198, 209, 213, 217, 225, 229, 235,
241, 248, 250, 254, 255,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_default_iscan_16x16[256]) = {
0, 2, 5, 9, 17, 24, 36, 44, 55, 72, 88, 104, 128, 143, 166, 179,
1, 4, 8, 13, 20, 30, 40, 54, 66, 79, 96, 113, 141, 154, 178, 196,
3, 7, 11, 18, 25, 33, 46, 57, 71, 86, 101, 119, 148, 164, 186, 201,
6, 12, 16, 23, 31, 39, 53, 64, 78, 92, 110, 127, 153, 169, 193, 208,
10, 14, 19, 28, 37, 47, 58, 67, 84, 98, 114, 133, 161, 176, 198, 214,
15, 21, 26, 34, 43, 52, 65, 77, 91, 106, 120, 140, 165, 185, 205, 221,
22, 27, 32, 41, 48, 60, 73, 85, 99, 116, 130, 151, 175, 190, 211, 225,
29, 35, 42, 49, 59, 69, 81, 95, 108, 125, 139, 155, 182, 197, 217, 229,
38, 45, 51, 61, 68, 80, 93, 105, 118, 134, 150, 168, 191, 207, 223, 234,
50, 56, 63, 74, 83, 94, 109, 117, 129, 147, 163, 177, 199, 213, 228, 238,
62, 70, 76, 87, 97, 107, 122, 131, 145, 159, 172, 188, 210, 222, 235, 242,
75, 82, 90, 102, 112, 124, 138, 146, 157, 173, 187, 202, 219, 230, 240, 245,
89, 100, 111, 123, 132, 142, 156, 167, 180, 189, 203, 216, 231, 237, 246, 250,
103, 115, 126, 136, 149, 162, 171, 183, 194, 204, 215, 224, 236, 241, 248,
252, 121, 135, 144, 158, 170, 181, 192, 200, 209, 218, 227, 233, 243, 244,
251, 254, 137, 152, 160, 174, 184, 195, 206, 212, 220, 226, 232, 239, 247,
249, 253, 255,
};
DECLARE_ALIGNED(16, static const int16_t, vp10_default_iscan_32x32[1024]) = {
0, 2, 5, 10, 17, 25, 38, 47, 62, 83, 101, 121, 145, 170, 193, 204,
210, 219, 229, 233, 245, 257, 275, 299, 342, 356, 377, 405, 455, 471, 495,
527, 1, 4, 8, 15, 22, 30, 45, 58, 74, 92, 112, 133, 158, 184, 203, 215, 222,
228, 234, 237, 256, 274, 298, 317, 355, 376, 404, 426, 470, 494, 526, 551,
3, 7, 12, 18, 28, 36, 52, 64, 82, 102, 118, 142, 164, 189, 208, 217, 224,
231, 235, 238, 273, 297, 316, 329, 375, 403, 425, 440, 493, 525, 550, 567,
6, 11, 16, 23, 31, 43, 60, 73, 90, 109, 126, 150, 173, 196, 211, 220, 226,
232, 236, 239, 296, 315, 328, 335, 402, 424, 439, 447, 524, 549, 566, 575,
9, 14, 19, 29, 37, 50, 65, 78, 95, 116, 134, 157, 179, 201, 214, 223, 244,
255, 272, 295, 341, 354, 374, 401, 454, 469, 492, 523, 582, 596, 617, 645,
13, 20, 26, 35, 44, 54, 72, 85, 105, 123, 140, 163, 182, 205, 216, 225,
254, 271, 294, 314, 353, 373, 400, 423, 468, 491, 522, 548, 595, 616, 644,
666, 21, 27, 33, 42, 53, 63, 80, 94, 113, 132, 151, 172, 190, 209, 218, 227,
270, 293, 313, 327, 372, 399, 422, 438, 490, 521, 547, 565, 615, 643, 665,
680, 24, 32, 39, 48, 57, 71, 88, 104, 120, 139, 159, 178, 197, 212, 221, 230,
292, 312, 326, 334, 398, 421, 437, 446, 520, 546, 564, 574, 642, 664, 679,
687, 34, 40, 46, 56, 68, 81, 96, 111, 130, 147, 167, 186, 243, 253, 269, 291,
340, 352, 371, 397, 453, 467, 489, 519, 581, 594, 614, 641, 693, 705, 723,
747, 41, 49, 55, 67, 77, 91, 107, 124, 138, 161, 177, 194, 252, 268, 290,
311, 351, 370, 396, 420, 466, 488, 518, 545, 593, 613, 640, 663, 704, 722,
746, 765, 51, 59, 66, 76, 89, 99, 119, 131, 149, 168, 181, 200, 267, 289,
310, 325, 369, 395, 419, 436, 487, 517, 544, 563, 612, 639, 662, 678, 721,
745, 764, 777, 61, 69, 75, 87, 100, 114, 129, 144, 162, 180, 191, 207, 288,
309, 324, 333, 394, 418, 435, 445, 516, 543, 562, 573, 638, 661, 677, 686,
744, 763, 776, 783, 70, 79, 86, 97, 108, 122, 137, 155, 242, 251, 266, 287,
339, 350, 368, 393, 452, 465, 486, 515, 580, 592, 611, 637, 692, 703, 720,
743, 788, 798, 813, 833, 84, 93, 103, 110, 125, 141, 154, 171, 250, 265, 286,
308, 349, 367, 392, 417, 464, 485, 514, 542, 591, 610, 636, 660, 702, 719,
742, 762, 797, 812, 832, 848, 98, 106, 115, 127, 143, 156, 169, 185, 264,
285, 307, 323, 366, 391, 416, 434, 484, 513, 541, 561, 609, 635, 659, 676,
718, 741, 761, 775, 811, 831, 847, 858, 117, 128, 136, 148, 160, 175, 188,
198, 284, 306, 322, 332, 390, 415, 433, 444, 512, 540, 560, 572, 634, 658,
675, 685, 740, 760, 774, 782, 830, 846, 857, 863, 135, 146, 152, 165, 241,
249, 263, 283, 338, 348, 365, 389, 451, 463, 483, 511, 579, 590, 608, 633,
691, 701, 717, 739, 787, 796, 810, 829, 867, 875, 887, 903, 153, 166, 174,
183, 248, 262, 282, 305, 347, 364, 388, 414, 462, 482, 510, 539, 589, 607,
632, 657, 700, 716, 738, 759, 795, 809, 828, 845, 874, 886, 902, 915, 176,
187, 195, 202, 261, 281, 304, 321, 363, 387, 413, 432, 481, 509, 538, 559,
606, 631, 656, 674, 715, 737, 758, 773, 808, 827, 844, 856, 885, 901, 914,
923, 192, 199, 206, 213, 280, 303, 320, 331, 386, 412, 431, 443, 508, 537,
558, 571, 630, 655, 673, 684, 736, 757, 772, 781, 826, 843, 855, 862, 900,
913, 922, 927, 240, 247, 260, 279, 337, 346, 362, 385, 450, 461, 480, 507,
578, 588, 605, 629, 690, 699, 714, 735, 786, 794, 807, 825, 866, 873, 884,
899, 930, 936, 945, 957, 246, 259, 278, 302, 345, 361, 384, 411, 460, 479,
506, 536, 587, 604, 628, 654, 698, 713, 734, 756, 793, 806, 824, 842, 872,
883, 898, 912, 935, 944, 956, 966, 258, 277, 301, 319, 360, 383, 410, 430,
478, 505, 535, 557, 603, 627, 653, 672, 712, 733, 755, 771, 805, 823, 841,
854, 882, 897, 911, 921, 943, 955, 965, 972, 276, 300, 318, 330, 382, 409,
429, 442, 504, 534, 556, 570, 626, 652, 671, 683, 732, 754, 770, 780, 822,
840, 853, 861, 896, 910, 920, 926, 954, 964, 971, 975, 336, 344, 359, 381,
449, 459, 477, 503, 577, 586, 602, 625, 689, 697, 711, 731, 785, 792, 804,
821, 865, 871, 881, 895, 929, 934, 942, 953, 977, 981, 987, 995, 343, 358,
380, 408, 458, 476, 502, 533, 585, 601, 624, 651, 696, 710, 730, 753, 791,
803, 820, 839, 870, 880, 894, 909, 933, 941, 952, 963, 980, 986, 994, 1001,
357, 379, 407, 428, 475, 501, 532, 555, 600, 623, 650, 670, 709, 729, 752,
769, 802, 819, 838, 852, 879, 893, 908, 919, 940, 951, 962, 970, 985, 993,
1000, 1005, 378, 406, 427, 441, 500, 531, 554, 569, 622, 649, 669, 682, 728,
751, 768, 779, 818, 837, 851, 860, 892, 907, 918, 925, 950, 961, 969, 974,
992, 999, 1004, 1007, 448, 457, 474, 499, 576, 584, 599, 621, 688, 695, 708,
727, 784, 790, 801, 817, 864, 869, 878, 891, 928, 932, 939, 949, 976, 979,
984, 991, 1008, 1010, 1013, 1017, 456, 473, 498, 530, 583, 598, 620, 648,
694, 707, 726, 750, 789, 800, 816, 836, 868, 877, 890, 906, 931, 938, 948,
960, 978, 983, 990, 998, 1009, 1012, 1016, 1020, 472, 497, 529, 553, 597,
619, 647, 668, 706, 725, 749, 767, 799, 815, 835, 850, 876, 889, 905, 917,
937, 947, 959, 968, 982, 989, 997, 1003, 1011, 1015, 1019, 1022, 496, 528,
552, 568, 618, 646, 667, 681, 724, 748, 766, 778, 814, 834, 849, 859, 888,
904, 916, 924, 946, 958, 967, 973, 988, 996, 1002, 1006, 1014, 1018, 1021,
1023,
};
const scan_order vp10_default_scan_orders[TX_SIZES] = {
{default_scan_4x4, vp10_default_iscan_4x4, default_scan_4x4_neighbors},
{default_scan_8x8, vp10_default_iscan_8x8, default_scan_8x8_neighbors},
{default_scan_16x16, vp10_default_iscan_16x16, default_scan_16x16_neighbors},
{default_scan_32x32, vp10_default_iscan_32x32, default_scan_32x32_neighbors},
};
const scan_order vp10_scan_orders[TX_SIZES][TX_TYPES] = {
{ // TX_4X4
{default_scan_4x4, vp10_default_iscan_4x4, default_scan_4x4_neighbors},
{row_scan_4x4, vp10_row_iscan_4x4, row_scan_4x4_neighbors},
{col_scan_4x4, vp10_col_iscan_4x4, col_scan_4x4_neighbors},
{default_scan_4x4, vp10_default_iscan_4x4, default_scan_4x4_neighbors}
}, { // TX_8X8
{default_scan_8x8, vp10_default_iscan_8x8, default_scan_8x8_neighbors},
{row_scan_8x8, vp10_row_iscan_8x8, row_scan_8x8_neighbors},
{col_scan_8x8, vp10_col_iscan_8x8, col_scan_8x8_neighbors},
{default_scan_8x8, vp10_default_iscan_8x8, default_scan_8x8_neighbors}
}, { // TX_16X16
{default_scan_16x16, vp10_default_iscan_16x16, default_scan_16x16_neighbors},
{row_scan_16x16, vp10_row_iscan_16x16, row_scan_16x16_neighbors},
{col_scan_16x16, vp10_col_iscan_16x16, col_scan_16x16_neighbors},
{default_scan_16x16, vp10_default_iscan_16x16, default_scan_16x16_neighbors}
}, { // TX_32X32
{default_scan_32x32, vp10_default_iscan_32x32, default_scan_32x32_neighbors},
{default_scan_32x32, vp10_default_iscan_32x32, default_scan_32x32_neighbors},
{default_scan_32x32, vp10_default_iscan_32x32, default_scan_32x32_neighbors},
{default_scan_32x32, vp10_default_iscan_32x32, default_scan_32x32_neighbors},
}
};

57
vp10/common/vp9_scan.h Normal file
View File

@ -0,0 +1,57 @@
/*
* Copyright (c) 2013 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 VP9_COMMON_VP9_SCAN_H_
#define VP9_COMMON_VP9_SCAN_H_
#include "vpx/vpx_integer.h"
#include "vpx_ports/mem.h"
#include "vp10/common/vp9_enums.h"
#include "vp10/common/vp9_blockd.h"
#ifdef __cplusplus
extern "C" {
#endif
#define MAX_NEIGHBORS 2
typedef struct {
const int16_t *scan;
const int16_t *iscan;
const int16_t *neighbors;
} scan_order;
extern const scan_order vp10_default_scan_orders[TX_SIZES];
extern const scan_order vp10_scan_orders[TX_SIZES][TX_TYPES];
static INLINE int get_coef_context(const int16_t *neighbors,
const uint8_t *token_cache, int c) {
return (1 + token_cache[neighbors[MAX_NEIGHBORS * c + 0]] +
token_cache[neighbors[MAX_NEIGHBORS * c + 1]]) >> 1;
}
static INLINE const scan_order *get_scan(const MACROBLOCKD *xd, TX_SIZE tx_size,
PLANE_TYPE type, int block_idx) {
const MODE_INFO *const mi = xd->mi[0];
if (is_inter_block(&mi->mbmi) || type != PLANE_TYPE_Y || xd->lossless) {
return &vp10_default_scan_orders[tx_size];
} else {
const PREDICTION_MODE mode = get_y_mode(mi, block_idx);
return &vp10_scan_orders[tx_size][intra_mode_to_tx_type_lookup[mode]];
}
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_SCAN_H_

63
vp10/common/vp9_seg_common.c Normal file
View File

@ -0,0 +1,63 @@
/*
* Copyright (c) 2010 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 <assert.h>
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_loopfilter.h"
#include "vp10/common/vp9_seg_common.h"
#include "vp10/common/vp9_quant_common.h"
static const int seg_feature_data_signed[SEG_LVL_MAX] = { 1, 1, 0, 0 };
static const int seg_feature_data_max[SEG_LVL_MAX] = {
MAXQ, MAX_LOOP_FILTER, 3, 0 };
// These functions provide access to new segment level features.
// Eventually these function may be "optimized out" but for the moment,
// the coding mechanism is still subject to change so these provide a
// convenient single point of change.
void vp10_clearall_segfeatures(struct segmentation *seg) {
vp10_zero(seg->feature_data);
vp10_zero(seg->feature_mask);
}
void vp10_enable_segfeature(struct segmentation *seg, int segment_id,
SEG_LVL_FEATURES feature_id) {
seg->feature_mask[segment_id] |= 1 << feature_id;
}
int vp10_seg_feature_data_max(SEG_LVL_FEATURES feature_id) {
return seg_feature_data_max[feature_id];
}
int vp10_is_segfeature_signed(SEG_LVL_FEATURES feature_id) {
return seg_feature_data_signed[feature_id];
}
void vp10_set_segdata(struct segmentation *seg, int segment_id,
SEG_LVL_FEATURES feature_id, int seg_data) {
assert(seg_data <= seg_feature_data_max[feature_id]);
if (seg_data < 0) {
assert(seg_feature_data_signed[feature_id]);
assert(-seg_data <= seg_feature_data_max[feature_id]);
}
seg->feature_data[segment_id][feature_id] = seg_data;
}
const vpx_tree_index vp10_segment_tree[TREE_SIZE(MAX_SEGMENTS)] = {
2, 4, 6, 8, 10, 12,
0, -1, -2, -3, -4, -5, -6, -7
};
// TBD? Functions to read and write segment data with range / validity checking

86
vp10/common/vp9_seg_common.h Normal file
View File

@ -0,0 +1,86 @@
/*
* Copyright (c) 2012 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 VP9_COMMON_VP9_SEG_COMMON_H_
#define VP9_COMMON_VP9_SEG_COMMON_H_
#include "vpx_dsp/prob.h"
#ifdef __cplusplus
extern "C" {
#endif
#define SEGMENT_DELTADATA 0
#define SEGMENT_ABSDATA 1
#define MAX_SEGMENTS 8
#define SEG_TREE_PROBS (MAX_SEGMENTS-1)
#define PREDICTION_PROBS 3
// Segment level features.
typedef enum {
SEG_LVL_ALT_Q = 0, // Use alternate Quantizer ....
SEG_LVL_ALT_LF = 1, // Use alternate loop filter value...
SEG_LVL_REF_FRAME = 2, // Optional Segment reference frame
SEG_LVL_SKIP = 3, // Optional Segment (0,0) + skip mode
SEG_LVL_MAX = 4 // Number of features supported
} SEG_LVL_FEATURES;
struct segmentation {
uint8_t enabled;
uint8_t update_map;
uint8_t update_data;
uint8_t abs_delta;
uint8_t temporal_update;
vpx_prob tree_probs[SEG_TREE_PROBS];
vpx_prob pred_probs[PREDICTION_PROBS];
int16_t feature_data[MAX_SEGMENTS][SEG_LVL_MAX];
unsigned int feature_mask[MAX_SEGMENTS];
};
static INLINE int segfeature_active(const struct segmentation *seg,
int segment_id,
SEG_LVL_FEATURES feature_id) {
return seg->enabled &&
(seg->feature_mask[segment_id] & (1 << feature_id));
}
void vp10_clearall_segfeatures(struct segmentation *seg);
void vp10_enable_segfeature(struct segmentation *seg,
int segment_id,
SEG_LVL_FEATURES feature_id);
int vp10_seg_feature_data_max(SEG_LVL_FEATURES feature_id);
int vp10_is_segfeature_signed(SEG_LVL_FEATURES feature_id);
void vp10_set_segdata(struct segmentation *seg,
int segment_id,
SEG_LVL_FEATURES feature_id,
int seg_data);
static INLINE int get_segdata(const struct segmentation *seg, int segment_id,
SEG_LVL_FEATURES feature_id) {
return seg->feature_data[segment_id][feature_id];
}
extern const vpx_tree_index vp10_segment_tree[TREE_SIZE(MAX_SEGMENTS)];
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_SEG_COMMON_H_

84
vp10/common/vp9_systemdependent.h Normal file
View File

@ -0,0 +1,84 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_SYSTEMDEPENDENT_H_
#define VP9_COMMON_VP9_SYSTEMDEPENDENT_H_
#include "vpx_ports/msvc.h"
#ifdef _MSC_VER
# include <math.h> // the ceil() definition must precede intrin.h
# if _MSC_VER > 1310 && (defined(_M_X64) || defined(_M_IX86))
# include <intrin.h>
# define USE_MSC_INTRINSICS
# endif
#endif
#ifdef __cplusplus
extern "C" {
#endif
#include "./vpx_config.h"
#if ARCH_X86 || ARCH_X86_64
void vpx_reset_mmx_state(void);
#define vpx_clear_system_state() vpx_reset_mmx_state()
#else
#define vpx_clear_system_state()
#endif
#if defined(_MSC_VER) && _MSC_VER < 1800
// round is not defined in MSVC before VS2013.
static INLINE int round(double x) {
if (x < 0)
return (int)ceil(x - 0.5);
else
return (int)floor(x + 0.5);
}
#endif
// use GNU builtins where available.
#if defined(__GNUC__) && \
((__GNUC__ == 3 && __GNUC_MINOR__ >= 4) || __GNUC__ >= 4)
static INLINE int get_msb(unsigned int n) {
return 31 ^ __builtin_clz(n);
}
#elif defined(USE_MSC_INTRINSICS)
#pragma intrinsic(_BitScanReverse)
static INLINE int get_msb(unsigned int n) {
unsigned long first_set_bit;
_BitScanReverse(&first_set_bit, n);
return first_set_bit;
}
#undef USE_MSC_INTRINSICS
#else
// Returns (int)floor(log2(n)). n must be > 0.
static INLINE int get_msb(unsigned int n) {
int log = 0;
unsigned int value = n;
int i;
for (i = 4; i >= 0; --i) {
const int shift = (1 << i);
const unsigned int x = value >> shift;
if (x != 0) {
value = x;
log += shift;
}
}
return log;
}
#endif
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_SYSTEMDEPENDENT_H_

120
vp10/common/vp9_textblit.c Normal file
View File

@ -0,0 +1,120 @@
/*
* Copyright (c) 2010 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 <stdlib.h>
#include "vp10/common/vp9_textblit.h"
static const int font[] = {
0x0, 0x5C00, 0x8020, 0xAFABEA, 0xD7EC0, 0x1111111, 0x1855740, 0x18000,
0x45C0, 0x74400, 0x51140, 0x23880, 0xC4000, 0x21080, 0x80000, 0x111110,
0xE9D72E, 0x87E40, 0x12AD732, 0xAAD62A, 0x4F94C4, 0x4D6B7, 0x456AA,
0x3E8423, 0xAAD6AA, 0xAAD6A2, 0x2800, 0x2A00, 0x8A880, 0x52940, 0x22A20,
0x15422, 0x6AD62E, 0x1E4A53E, 0xAAD6BF, 0x8C62E, 0xE8C63F, 0x118D6BF,
0x1094BF, 0xCAC62E, 0x1F2109F, 0x118FE31, 0xF8C628, 0x8A89F, 0x108421F,
0x1F1105F, 0x1F4105F, 0xE8C62E, 0x2294BF, 0x164C62E, 0x12694BF, 0x8AD6A2,
0x10FC21, 0x1F8421F, 0x744107, 0xF8220F, 0x1151151, 0x117041, 0x119D731,
0x47E0, 0x1041041, 0xFC400, 0x10440, 0x1084210, 0x820
};
static void plot(int x, int y, unsigned char *image, int pitch) {
image[x + y * pitch] ^= 255;
}
void vp10_blit_text(const char *msg, unsigned char *address, const int pitch) {
int letter_bitmap;
unsigned char *output_pos = address;
int colpos = 0;
while (msg[colpos] != 0) {
char letter = msg[colpos];
int fontcol, fontrow;
if (letter <= 'Z' && letter >= ' ')
letter_bitmap = font[letter - ' '];
else if (letter <= 'z' && letter >= 'a')
letter_bitmap = font[letter - 'a' + 'A' - ' '];
else
letter_bitmap = font[0];
for (fontcol = 6; fontcol >= 0; fontcol--)
for (fontrow = 0; fontrow < 5; fontrow++)
output_pos[fontrow * pitch + fontcol] =
((letter_bitmap >> (fontcol * 5)) & (1 << fontrow) ? 255 : 0);
output_pos += 7;
colpos++;
}
}
/* Bresenham line algorithm */
void vp10_blit_line(int x0, int x1, int y0, int y1, unsigned char *image,
int pitch) {
int steep = abs(y1 - y0) > abs(x1 - x0);
int deltax, deltay;
int error, ystep, y, x;
if (steep) {
int t;
t = x0;
x0 = y0;
y0 = t;
t = x1;
x1 = y1;
y1 = t;
}
if (x0 > x1) {
int t;
t = x0;
x0 = x1;
x1 = t;
t = y0;
y0 = y1;
y1 = t;
}
deltax = x1 - x0;
deltay = abs(y1 - y0);
error = deltax / 2;
y = y0;
if (y0 < y1)
ystep = 1;
else
ystep = -1;
if (steep) {
for (x = x0; x <= x1; x++) {
plot(y, x, image, pitch);
error = error - deltay;
if (error < 0) {
y = y + ystep;
error = error + deltax;
}
}
} else {
for (x = x0; x <= x1; x++) {
plot(x, y, image, pitch);
error = error - deltay;
if (error < 0) {
y = y + ystep;
error = error + deltax;
}
}
}
}

27
vp10/common/vp9_textblit.h Normal file
View File

@ -0,0 +1,27 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_TEXTBLIT_H_
#define VP9_COMMON_VP9_TEXTBLIT_H_
#ifdef __cplusplus
extern "C" {
#endif
void vp10_blit_text(const char *msg, unsigned char *address, int pitch);
void vp10_blit_line(int x0, int x1, int y0, int y1, unsigned char *image,
int pitch);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_TEXTBLIT_H_

436
vp10/common/vp9_thread_common.c Normal file
View File

@ -0,0 +1,436 @@
/*
* Copyright (c) 2014 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 "./vpx_config.h"
#include "vpx_mem/vpx_mem.h"
#include "vp10/common/vp9_entropymode.h"
#include "vp10/common/vp9_thread_common.h"
#include "vp10/common/vp9_reconinter.h"
#include "vp10/common/vp9_loopfilter.h"
#if CONFIG_MULTITHREAD
static INLINE void mutex_lock(pthread_mutex_t *const mutex) {
const int kMaxTryLocks = 4000;
int locked = 0;
int i;
for (i = 0; i < kMaxTryLocks; ++i) {
if (!pthread_mutex_trylock(mutex)) {
locked = 1;
break;
}
}
if (!locked)
pthread_mutex_lock(mutex);
}
#endif // CONFIG_MULTITHREAD
static INLINE void sync_read(VP9LfSync *const lf_sync, int r, int c) {
#if CONFIG_MULTITHREAD
const int nsync = lf_sync->sync_range;
if (r && !(c & (nsync - 1))) {
pthread_mutex_t *const mutex = &lf_sync->mutex_[r - 1];
mutex_lock(mutex);
while (c > lf_sync->cur_sb_col[r - 1] - nsync) {
pthread_cond_wait(&lf_sync->cond_[r - 1], mutex);
}
pthread_mutex_unlock(mutex);
}
#else
(void)lf_sync;
(void)r;
(void)c;
#endif // CONFIG_MULTITHREAD
}
static INLINE void sync_write(VP9LfSync *const lf_sync, int r, int c,
const int sb_cols) {
#if CONFIG_MULTITHREAD
const int nsync = lf_sync->sync_range;
int cur;
// Only signal when there are enough filtered SB for next row to run.
int sig = 1;
if (c < sb_cols - 1) {
cur = c;
if (c % nsync)
sig = 0;
} else {
cur = sb_cols + nsync;
}
if (sig) {
mutex_lock(&lf_sync->mutex_[r]);
lf_sync->cur_sb_col[r] = cur;
pthread_cond_signal(&lf_sync->cond_[r]);
pthread_mutex_unlock(&lf_sync->mutex_[r]);
}
#else
(void)lf_sync;
(void)r;
(void)c;
(void)sb_cols;
#endif // CONFIG_MULTITHREAD
}
// Implement row loopfiltering for each thread.
static INLINE
void thread_loop_filter_rows(const YV12_BUFFER_CONFIG *const frame_buffer,
VP9_COMMON *const cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int start, int stop, int y_only,
VP9LfSync *const lf_sync) {
const int num_planes = y_only ? 1 : MAX_MB_PLANE;
const int sb_cols = mi_cols_aligned_to_sb(cm->mi_cols) >> MI_BLOCK_SIZE_LOG2;
int mi_row, mi_col;
enum lf_path path;
if (y_only)
path = LF_PATH_444;
else if (planes[1].subsampling_y == 1 && planes[1].subsampling_x == 1)
path = LF_PATH_420;
else if (planes[1].subsampling_y == 0 && planes[1].subsampling_x == 0)
path = LF_PATH_444;
else
path = LF_PATH_SLOW;
for (mi_row = start; mi_row < stop;
mi_row += lf_sync->num_workers * MI_BLOCK_SIZE) {
MODE_INFO **const mi = cm->mi_grid_visible + mi_row * cm->mi_stride;
for (mi_col = 0; mi_col < cm->mi_cols; mi_col += MI_BLOCK_SIZE) {
const int r = mi_row >> MI_BLOCK_SIZE_LOG2;
const int c = mi_col >> MI_BLOCK_SIZE_LOG2;
LOOP_FILTER_MASK lfm;
int plane;
sync_read(lf_sync, r, c);
vp10_setup_dst_planes(planes, frame_buffer, mi_row, mi_col);
// TODO(JBB): Make setup_mask work for non 420.
vp10_setup_mask(cm, mi_row, mi_col, mi + mi_col, cm->mi_stride,
&lfm);
vp10_filter_block_plane_ss00(cm, &planes[0], mi_row, &lfm);
for (plane = 1; plane < num_planes; ++plane) {
switch (path) {
case LF_PATH_420:
vp10_filter_block_plane_ss11(cm, &planes[plane], mi_row, &lfm);
break;
case LF_PATH_444:
vp10_filter_block_plane_ss00(cm, &planes[plane], mi_row, &lfm);
break;
case LF_PATH_SLOW:
vp10_filter_block_plane_non420(cm, &planes[plane], mi + mi_col,
mi_row, mi_col);
break;
}
}
sync_write(lf_sync, r, c, sb_cols);
}
}
}
// Row-based multi-threaded loopfilter hook
static int loop_filter_row_worker(VP9LfSync *const lf_sync,
LFWorkerData *const lf_data) {
thread_loop_filter_rows(lf_data->frame_buffer, lf_data->cm, lf_data->planes,
lf_data->start, lf_data->stop, lf_data->y_only,
lf_sync);
return 1;
}
static void loop_filter_rows_mt(YV12_BUFFER_CONFIG *frame,
VP9_COMMON *cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int start, int stop, int y_only,
VPxWorker *workers, int nworkers,
VP9LfSync *lf_sync) {
const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
// Number of superblock rows and cols
const int sb_rows = mi_cols_aligned_to_sb(cm->mi_rows) >> MI_BLOCK_SIZE_LOG2;
// Decoder may allocate more threads than number of tiles based on user's
// input.
const int tile_cols = 1 << cm->log2_tile_cols;
const int num_workers = MIN(nworkers, tile_cols);
int i;
if (!lf_sync->sync_range || sb_rows != lf_sync->rows ||
num_workers > lf_sync->num_workers) {
vp10_loop_filter_dealloc(lf_sync);
vp10_loop_filter_alloc(lf_sync, cm, sb_rows, cm->width, num_workers);
}
// Initialize cur_sb_col to -1 for all SB rows.
memset(lf_sync->cur_sb_col, -1, sizeof(*lf_sync->cur_sb_col) * sb_rows);
// Set up loopfilter thread data.
// The decoder is capping num_workers because it has been observed that using
// more threads on the loopfilter than there are cores will hurt performance
// on Android. This is because the system will only schedule the tile decode
// workers on cores equal to the number of tile columns. Then if the decoder
// tries to use more threads for the loopfilter, it will hurt performance
// because of contention. If the multithreading code changes in the future
// then the number of workers used by the loopfilter should be revisited.
for (i = 0; i < num_workers; ++i) {
VPxWorker *const worker = &workers[i];
LFWorkerData *const lf_data = &lf_sync->lfdata[i];
worker->hook = (VPxWorkerHook)loop_filter_row_worker;
worker->data1 = lf_sync;
worker->data2 = lf_data;
// Loopfilter data
vp10_loop_filter_data_reset(lf_data, frame, cm, planes);
lf_data->start = start + i * MI_BLOCK_SIZE;
lf_data->stop = stop;
lf_data->y_only = y_only;
// Start loopfiltering
if (i == num_workers - 1) {
winterface->execute(worker);
} else {
winterface->launch(worker);
}
}
// Wait till all rows are finished
for (i = 0; i < num_workers; ++i) {
winterface->sync(&workers[i]);
}
}
void vp10_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame,
VP9_COMMON *cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int frame_filter_level,
int y_only, int partial_frame,
VPxWorker *workers, int num_workers,
VP9LfSync *lf_sync) {
int start_mi_row, end_mi_row, mi_rows_to_filter;
if (!frame_filter_level) return;
start_mi_row = 0;
mi_rows_to_filter = cm->mi_rows;
if (partial_frame && cm->mi_rows > 8) {
start_mi_row = cm->mi_rows >> 1;
start_mi_row &= 0xfffffff8;
mi_rows_to_filter = MAX(cm->mi_rows / 8, 8);
}
end_mi_row = start_mi_row + mi_rows_to_filter;
vp10_loop_filter_frame_init(cm, frame_filter_level);
loop_filter_rows_mt(frame, cm, planes, start_mi_row, end_mi_row,
y_only, workers, num_workers, lf_sync);
}
// Set up nsync by width.
static INLINE int get_sync_range(int width) {
// nsync numbers are picked by testing. For example, for 4k
// video, using 4 gives best performance.
if (width < 640)
return 1;
else if (width <= 1280)
return 2;
else if (width <= 4096)
return 4;
else
return 8;
}
// Allocate memory for lf row synchronization
void vp10_loop_filter_alloc(VP9LfSync *lf_sync, VP9_COMMON *cm, int rows,
int width, int num_workers) {
lf_sync->rows = rows;
#if CONFIG_MULTITHREAD
{
int i;
CHECK_MEM_ERROR(cm, lf_sync->mutex_,
vpx_malloc(sizeof(*lf_sync->mutex_) * rows));
if (lf_sync->mutex_) {
for (i = 0; i < rows; ++i) {
pthread_mutex_init(&lf_sync->mutex_[i], NULL);
}
}
CHECK_MEM_ERROR(cm, lf_sync->cond_,
vpx_malloc(sizeof(*lf_sync->cond_) * rows));
if (lf_sync->cond_) {
for (i = 0; i < rows; ++i) {
pthread_cond_init(&lf_sync->cond_[i], NULL);
}
}
}
#endif // CONFIG_MULTITHREAD
CHECK_MEM_ERROR(cm, lf_sync->lfdata,
vpx_malloc(num_workers * sizeof(*lf_sync->lfdata)));
lf_sync->num_workers = num_workers;
CHECK_MEM_ERROR(cm, lf_sync->cur_sb_col,
vpx_malloc(sizeof(*lf_sync->cur_sb_col) * rows));
// Set up nsync.
lf_sync->sync_range = get_sync_range(width);
}
// Deallocate lf synchronization related mutex and data
void vp10_loop_filter_dealloc(VP9LfSync *lf_sync) {
if (lf_sync != NULL) {
#if CONFIG_MULTITHREAD
int i;
if (lf_sync->mutex_ != NULL) {
for (i = 0; i < lf_sync->rows; ++i) {
pthread_mutex_destroy(&lf_sync->mutex_[i]);
}
vpx_free(lf_sync->mutex_);
}
if (lf_sync->cond_ != NULL) {
for (i = 0; i < lf_sync->rows; ++i) {
pthread_cond_destroy(&lf_sync->cond_[i]);
}
vpx_free(lf_sync->cond_);
}
#endif // CONFIG_MULTITHREAD
vpx_free(lf_sync->lfdata);
vpx_free(lf_sync->cur_sb_col);
// clear the structure as the source of this call may be a resize in which
// case this call will be followed by an _alloc() which may fail.
vp10_zero(*lf_sync);
}
}
// Accumulate frame counts.
void vp10_accumulate_frame_counts(VP9_COMMON *cm, FRAME_COUNTS *counts,
int is_dec) {
int i, j, k, l, m;
for (i = 0; i < BLOCK_SIZE_GROUPS; i++)
for (j = 0; j < INTRA_MODES; j++)
cm->counts.y_mode[i][j] += counts->y_mode[i][j];
for (i = 0; i < INTRA_MODES; i++)
for (j = 0; j < INTRA_MODES; j++)
cm->counts.uv_mode[i][j] += counts->uv_mode[i][j];
for (i = 0; i < PARTITION_CONTEXTS; i++)
for (j = 0; j < PARTITION_TYPES; j++)
cm->counts.partition[i][j] += counts->partition[i][j];
if (is_dec) {
int n;
for (i = 0; i < TX_SIZES; i++)
for (j = 0; j < PLANE_TYPES; j++)
for (k = 0; k < REF_TYPES; k++)
for (l = 0; l < COEF_BANDS; l++)
for (m = 0; m < COEFF_CONTEXTS; m++) {
cm->counts.eob_branch[i][j][k][l][m] +=
counts->eob_branch[i][j][k][l][m];
for (n = 0; n < UNCONSTRAINED_NODES + 1; n++)
cm->counts.coef[i][j][k][l][m][n] +=
counts->coef[i][j][k][l][m][n];
}
} else {
for (i = 0; i < TX_SIZES; i++)
for (j = 0; j < PLANE_TYPES; j++)
for (k = 0; k < REF_TYPES; k++)
for (l = 0; l < COEF_BANDS; l++)
for (m = 0; m < COEFF_CONTEXTS; m++)
cm->counts.eob_branch[i][j][k][l][m] +=
counts->eob_branch[i][j][k][l][m];
// In the encoder, cm->counts.coef is only updated at frame
// level, so not need to accumulate it here.
// for (n = 0; n < UNCONSTRAINED_NODES + 1; n++)
// cm->counts.coef[i][j][k][l][m][n] +=
// counts->coef[i][j][k][l][m][n];
}
for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
for (j = 0; j < SWITCHABLE_FILTERS; j++)
cm->counts.switchable_interp[i][j] += counts->switchable_interp[i][j];
for (i = 0; i < INTER_MODE_CONTEXTS; i++)
for (j = 0; j < INTER_MODES; j++)
cm->counts.inter_mode[i][j] += counts->inter_mode[i][j];
for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
for (j = 0; j < 2; j++)
cm->counts.intra_inter[i][j] += counts->intra_inter[i][j];
for (i = 0; i < COMP_INTER_CONTEXTS; i++)
for (j = 0; j < 2; j++)
cm->counts.comp_inter[i][j] += counts->comp_inter[i][j];
for (i = 0; i < REF_CONTEXTS; i++)
for (j = 0; j < 2; j++)
for (k = 0; k < 2; k++)
cm->counts.single_ref[i][j][k] += counts->single_ref[i][j][k];
for (i = 0; i < REF_CONTEXTS; i++)
for (j = 0; j < 2; j++)
cm->counts.comp_ref[i][j] += counts->comp_ref[i][j];
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
for (j = 0; j < TX_SIZES; j++)
cm->counts.tx.p32x32[i][j] += counts->tx.p32x32[i][j];
for (j = 0; j < TX_SIZES - 1; j++)
cm->counts.tx.p16x16[i][j] += counts->tx.p16x16[i][j];
for (j = 0; j < TX_SIZES - 2; j++)
cm->counts.tx.p8x8[i][j] += counts->tx.p8x8[i][j];
}
for (i = 0; i < TX_SIZES; i++)
cm->counts.tx.tx_totals[i] += counts->tx.tx_totals[i];
for (i = 0; i < SKIP_CONTEXTS; i++)
for (j = 0; j < 2; j++)
cm->counts.skip[i][j] += counts->skip[i][j];
for (i = 0; i < MV_JOINTS; i++)
cm->counts.mv.joints[i] += counts->mv.joints[i];
for (k = 0; k < 2; k++) {
nmv_component_counts *comps = &cm->counts.mv.comps[k];
nmv_component_counts *comps_t = &counts->mv.comps[k];
for (i = 0; i < 2; i++) {
comps->sign[i] += comps_t->sign[i];
comps->class0_hp[i] += comps_t->class0_hp[i];
comps->hp[i] += comps_t->hp[i];
}
for (i = 0; i < MV_CLASSES; i++)
comps->classes[i] += comps_t->classes[i];
for (i = 0; i < CLASS0_SIZE; i++) {
comps->class0[i] += comps_t->class0[i];
for (j = 0; j < MV_FP_SIZE; j++)
comps->class0_fp[i][j] += comps_t->class0_fp[i][j];
}
for (i = 0; i < MV_OFFSET_BITS; i++)
for (j = 0; j < 2; j++)
comps->bits[i][j] += comps_t->bits[i][j];
for (i = 0; i < MV_FP_SIZE; i++)
comps->fp[i] += comps_t->fp[i];
}
}

57
vp10/common/vp9_thread_common.h Normal file
View File

@ -0,0 +1,57 @@
/*
* Copyright (c) 2014 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 VP9_COMMON_VP9_LOOPFILTER_THREAD_H_
#define VP9_COMMON_VP9_LOOPFILTER_THREAD_H_
#include "./vpx_config.h"
#include "vp10/common/vp9_loopfilter.h"
#include "vpx_util/vpx_thread.h"
struct VP9Common;
struct FRAME_COUNTS;
// Loopfilter row synchronization
typedef struct VP9LfSyncData {
#if CONFIG_MULTITHREAD
pthread_mutex_t *mutex_;
pthread_cond_t *cond_;
#endif
// Allocate memory to store the loop-filtered superblock index in each row.
int *cur_sb_col;
// The optimal sync_range for different resolution and platform should be
// determined by testing. Currently, it is chosen to be a power-of-2 number.
int sync_range;
int rows;
// Row-based parallel loopfilter data
LFWorkerData *lfdata;
int num_workers;
} VP9LfSync;
// Allocate memory for loopfilter row synchronization.
void vp10_loop_filter_alloc(VP9LfSync *lf_sync, struct VP9Common *cm, int rows,
int width, int num_workers);
// Deallocate loopfilter synchronization related mutex and data.
void vp10_loop_filter_dealloc(VP9LfSync *lf_sync);
// Multi-threaded loopfilter that uses the tile threads.
void vp10_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame,
struct VP9Common *cm,
struct macroblockd_plane planes[MAX_MB_PLANE],
int frame_filter_level,
int y_only, int partial_frame,
VPxWorker *workers, int num_workers,
VP9LfSync *lf_sync);
void vp10_accumulate_frame_counts(struct VP9Common *cm,
struct FRAME_COUNTS *counts, int is_dec);
#endif // VP9_COMMON_VP9_LOOPFILTER_THREAD_H_

59
vp10/common/vp9_tile_common.c Normal file
View File

@ -0,0 +1,59 @@
/*
* Copyright (c) 2010 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/common/vp9_tile_common.h"
#include "vp10/common/vp9_onyxc_int.h"
#define MIN_TILE_WIDTH_B64 4
#define MAX_TILE_WIDTH_B64 64
static int get_tile_offset(int idx, int mis, int log2) {
const int sb_cols = mi_cols_aligned_to_sb(mis) >> MI_BLOCK_SIZE_LOG2;
const int offset = ((idx * sb_cols) >> log2) << MI_BLOCK_SIZE_LOG2;
return MIN(offset, mis);
}
void vp10_tile_set_row(TileInfo *tile, const VP9_COMMON *cm, int row) {
tile->mi_row_start = get_tile_offset(row, cm->mi_rows, cm->log2_tile_rows);
tile->mi_row_end = get_tile_offset(row + 1, cm->mi_rows, cm->log2_tile_rows);
}
void vp10_tile_set_col(TileInfo *tile, const VP9_COMMON *cm, int col) {
tile->mi_col_start = get_tile_offset(col, cm->mi_cols, cm->log2_tile_cols);
tile->mi_col_end = get_tile_offset(col + 1, cm->mi_cols, cm->log2_tile_cols);
}
void vp10_tile_init(TileInfo *tile, const VP9_COMMON *cm, int row, int col) {
vp10_tile_set_row(tile, cm, row);
vp10_tile_set_col(tile, cm, col);
}
static int get_min_log2_tile_cols(const int sb64_cols) {
int min_log2 = 0;
while ((MAX_TILE_WIDTH_B64 << min_log2) < sb64_cols)
++min_log2;
return min_log2;
}
static int get_max_log2_tile_cols(const int sb64_cols) {
int max_log2 = 1;
while ((sb64_cols >> max_log2) >= MIN_TILE_WIDTH_B64)
++max_log2;
return max_log2 - 1;
}
void vp10_get_tile_n_bits(int mi_cols,
int *min_log2_tile_cols, int *max_log2_tile_cols) {
const int sb64_cols = mi_cols_aligned_to_sb(mi_cols) >> MI_BLOCK_SIZE_LOG2;
*min_log2_tile_cols = get_min_log2_tile_cols(sb64_cols);
*max_log2_tile_cols = get_max_log2_tile_cols(sb64_cols);
assert(*min_log2_tile_cols <= *max_log2_tile_cols);
}

40
vp10/common/vp9_tile_common.h Normal file
View File

@ -0,0 +1,40 @@
/*
* Copyright (c) 2010 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 VP9_COMMON_VP9_TILE_COMMON_H_
#define VP9_COMMON_VP9_TILE_COMMON_H_
#ifdef __cplusplus
extern "C" {
#endif
struct VP9Common;
typedef struct TileInfo {
int mi_row_start, mi_row_end;
int mi_col_start, mi_col_end;
} TileInfo;
// initializes 'tile->mi_(row|col)_(start|end)' for (row, col) based on
// 'cm->log2_tile_(rows|cols)' & 'cm->mi_(rows|cols)'
void vp10_tile_init(TileInfo *tile, const struct VP9Common *cm,
int row, int col);
void vp10_tile_set_row(TileInfo *tile, const struct VP9Common *cm, int row);
void vp10_tile_set_col(TileInfo *tile, const struct VP9Common *cm, int col);
void vp10_get_tile_n_bits(int mi_cols,
int *min_log2_tile_cols, int *max_log2_tile_cols);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_COMMON_VP9_TILE_COMMON_H_

180
vp10/common/x86/vp9_idct_intrin_sse2.c Normal file
View File

@ -0,0 +1,180 @@
/*
* Copyright (c) 2012 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 "vpx_dsp/x86/inv_txfm_sse2.h"
#include "vpx_dsp/x86/txfm_common_sse2.h"
#include "vpx_ports/mem.h"
void vp10_iht4x4_16_add_sse2(const int16_t *input, uint8_t *dest, int stride,
int tx_type) {
__m128i in[2];
const __m128i zero = _mm_setzero_si128();
const __m128i eight = _mm_set1_epi16(8);
in[0] = _mm_loadu_si128((const __m128i *)(input));
in[1] = _mm_loadu_si128((const __m128i *)(input + 8));
switch (tx_type) {
case 0: // DCT_DCT
idct4_sse2(in);
idct4_sse2(in);
break;
case 1: // ADST_DCT
idct4_sse2(in);
iadst4_sse2(in);
break;
case 2: // DCT_ADST
iadst4_sse2(in);
idct4_sse2(in);
break;
case 3: // ADST_ADST
iadst4_sse2(in);
iadst4_sse2(in);
break;
default:
assert(0);
break;
}
// Final round and shift
in[0] = _mm_add_epi16(in[0], eight);
in[1] = _mm_add_epi16(in[1], eight);
in[0] = _mm_srai_epi16(in[0], 4);
in[1] = _mm_srai_epi16(in[1], 4);
// Reconstruction and Store
{
__m128i d0 = _mm_cvtsi32_si128(*(const int *)(dest));
__m128i d2 = _mm_cvtsi32_si128(*(const int *)(dest + stride * 2));
d0 = _mm_unpacklo_epi32(d0,
_mm_cvtsi32_si128(*(const int *)(dest + stride)));
d2 = _mm_unpacklo_epi32(
d2, _mm_cvtsi32_si128(*(const int *)(dest + stride * 3)));
d0 = _mm_unpacklo_epi8(d0, zero);
d2 = _mm_unpacklo_epi8(d2, zero);
d0 = _mm_add_epi16(d0, in[0]);
d2 = _mm_add_epi16(d2, in[1]);
d0 = _mm_packus_epi16(d0, d2);
// store result[0]
*(int *)dest = _mm_cvtsi128_si32(d0);
// store result[1]
d0 = _mm_srli_si128(d0, 4);
*(int *)(dest + stride) = _mm_cvtsi128_si32(d0);
// store result[2]
d0 = _mm_srli_si128(d0, 4);
*(int *)(dest + stride * 2) = _mm_cvtsi128_si32(d0);
// store result[3]
d0 = _mm_srli_si128(d0, 4);
*(int *)(dest + stride * 3) = _mm_cvtsi128_si32(d0);
}
}
void vp10_iht8x8_64_add_sse2(const int16_t *input, uint8_t *dest, int stride,
int tx_type) {
__m128i in[8];
const __m128i zero = _mm_setzero_si128();
const __m128i final_rounding = _mm_set1_epi16(1 << 4);
// load input data
in[0] = _mm_load_si128((const __m128i *)input);
in[1] = _mm_load_si128((const __m128i *)(input + 8 * 1));
in[2] = _mm_load_si128((const __m128i *)(input + 8 * 2));
in[3] = _mm_load_si128((const __m128i *)(input + 8 * 3));
in[4] = _mm_load_si128((const __m128i *)(input + 8 * 4));
in[5] = _mm_load_si128((const __m128i *)(input + 8 * 5));
in[6] = _mm_load_si128((const __m128i *)(input + 8 * 6));
in[7] = _mm_load_si128((const __m128i *)(input + 8 * 7));
switch (tx_type) {
case 0: // DCT_DCT
idct8_sse2(in);
idct8_sse2(in);
break;
case 1: // ADST_DCT
idct8_sse2(in);
iadst8_sse2(in);
break;
case 2: // DCT_ADST
iadst8_sse2(in);
idct8_sse2(in);
break;
case 3: // ADST_ADST
iadst8_sse2(in);
iadst8_sse2(in);
break;
default:
assert(0);
break;
}
// Final rounding and shift
in[0] = _mm_adds_epi16(in[0], final_rounding);
in[1] = _mm_adds_epi16(in[1], final_rounding);
in[2] = _mm_adds_epi16(in[2], final_rounding);
in[3] = _mm_adds_epi16(in[3], final_rounding);
in[4] = _mm_adds_epi16(in[4], final_rounding);
in[5] = _mm_adds_epi16(in[5], final_rounding);
in[6] = _mm_adds_epi16(in[6], final_rounding);
in[7] = _mm_adds_epi16(in[7], final_rounding);
in[0] = _mm_srai_epi16(in[0], 5);
in[1] = _mm_srai_epi16(in[1], 5);
in[2] = _mm_srai_epi16(in[2], 5);
in[3] = _mm_srai_epi16(in[3], 5);
in[4] = _mm_srai_epi16(in[4], 5);
in[5] = _mm_srai_epi16(in[5], 5);
in[6] = _mm_srai_epi16(in[6], 5);
in[7] = _mm_srai_epi16(in[7], 5);
RECON_AND_STORE(dest + 0 * stride, in[0]);
RECON_AND_STORE(dest + 1 * stride, in[1]);
RECON_AND_STORE(dest + 2 * stride, in[2]);
RECON_AND_STORE(dest + 3 * stride, in[3]);
RECON_AND_STORE(dest + 4 * stride, in[4]);
RECON_AND_STORE(dest + 5 * stride, in[5]);
RECON_AND_STORE(dest + 6 * stride, in[6]);
RECON_AND_STORE(dest + 7 * stride, in[7]);
}
void vp10_iht16x16_256_add_sse2(const int16_t *input, uint8_t *dest, int stride,
int tx_type) {
__m128i in0[16], in1[16];
load_buffer_8x16(input, in0);
input += 8;
load_buffer_8x16(input, in1);
switch (tx_type) {
case 0: // DCT_DCT
idct16_sse2(in0, in1);
idct16_sse2(in0, in1);
break;
case 1: // ADST_DCT
idct16_sse2(in0, in1);
iadst16_sse2(in0, in1);
break;
case 2: // DCT_ADST
iadst16_sse2(in0, in1);
idct16_sse2(in0, in1);
break;
case 3: // ADST_ADST
iadst16_sse2(in0, in1);
iadst16_sse2(in0, in1);
break;
default:
assert(0);
break;
}
write_buffer_8x16(dest, in0, stride);
dest += 8;
write_buffer_8x16(dest, in1, stride);
}

287
vp10/common/x86/vp9_mfqe_sse2.asm Normal file
View File

@ -0,0 +1,287 @@
;
; 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.
;
; This file is a duplicate of mfqe_sse2.asm in VP8.
; TODO(jackychen): Find a way to fix the duplicate.
%include "vpx_ports/x86_abi_support.asm"
;void vp10_filter_by_weight16x16_sse2
;(
; unsigned char *src,
; int src_stride,
; unsigned char *dst,
; int dst_stride,
; int src_weight
;)
global sym(vp10_filter_by_weight16x16_sse2) PRIVATE
sym(vp10_filter_by_weight16x16_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
SAVE_XMM 6
GET_GOT rbx
push rsi
push rdi
; end prolog
movd xmm0, arg(4) ; src_weight
pshuflw xmm0, xmm0, 0x0 ; replicate to all low words
punpcklqdq xmm0, xmm0 ; replicate to all hi words
movdqa xmm1, [GLOBAL(tMFQE)]
psubw xmm1, xmm0 ; dst_weight
mov rax, arg(0) ; src
mov rsi, arg(1) ; src_stride
mov rdx, arg(2) ; dst
mov rdi, arg(3) ; dst_stride
mov rcx, 16 ; loop count
pxor xmm6, xmm6
.combine
movdqa xmm2, [rax]
movdqa xmm4, [rdx]
add rax, rsi
; src * src_weight
movdqa xmm3, xmm2
punpcklbw xmm2, xmm6
punpckhbw xmm3, xmm6
pmullw xmm2, xmm0
pmullw xmm3, xmm0
; dst * dst_weight
movdqa xmm5, xmm4
punpcklbw xmm4, xmm6
punpckhbw xmm5, xmm6
pmullw xmm4, xmm1
pmullw xmm5, xmm1
; sum, round and shift
paddw xmm2, xmm4
paddw xmm3, xmm5
paddw xmm2, [GLOBAL(tMFQE_round)]
paddw xmm3, [GLOBAL(tMFQE_round)]
psrlw xmm2, 4
psrlw xmm3, 4
packuswb xmm2, xmm3
movdqa [rdx], xmm2
add rdx, rdi
dec rcx
jnz .combine
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
RESTORE_XMM
UNSHADOW_ARGS
pop rbp
ret
;void vp10_filter_by_weight8x8_sse2
;(
; unsigned char *src,
; int src_stride,
; unsigned char *dst,
; int dst_stride,
; int src_weight
;)
global sym(vp10_filter_by_weight8x8_sse2) PRIVATE
sym(vp10_filter_by_weight8x8_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
GET_GOT rbx
push rsi
push rdi
; end prolog
movd xmm0, arg(4) ; src_weight
pshuflw xmm0, xmm0, 0x0 ; replicate to all low words
punpcklqdq xmm0, xmm0 ; replicate to all hi words
movdqa xmm1, [GLOBAL(tMFQE)]
psubw xmm1, xmm0 ; dst_weight
mov rax, arg(0) ; src
mov rsi, arg(1) ; src_stride
mov rdx, arg(2) ; dst
mov rdi, arg(3) ; dst_stride
mov rcx, 8 ; loop count
pxor xmm4, xmm4
.combine
movq xmm2, [rax]
movq xmm3, [rdx]
add rax, rsi
; src * src_weight
punpcklbw xmm2, xmm4
pmullw xmm2, xmm0
; dst * dst_weight
punpcklbw xmm3, xmm4
pmullw xmm3, xmm1
; sum, round and shift
paddw xmm2, xmm3
paddw xmm2, [GLOBAL(tMFQE_round)]
psrlw xmm2, 4
packuswb xmm2, xmm4
movq [rdx], xmm2
add rdx, rdi
dec rcx
jnz .combine
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
;void vp10_variance_and_sad_16x16_sse2 | arg
;(
; unsigned char *src1, 0
; int stride1, 1
; unsigned char *src2, 2
; int stride2, 3
; unsigned int *variance, 4
; unsigned int *sad, 5
;)
global sym(vp10_variance_and_sad_16x16_sse2) PRIVATE
sym(vp10_variance_and_sad_16x16_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 6
GET_GOT rbx
push rsi
push rdi
; end prolog
mov rax, arg(0) ; src1
mov rcx, arg(1) ; stride1
mov rdx, arg(2) ; src2
mov rdi, arg(3) ; stride2
mov rsi, 16 ; block height
; Prep accumulator registers
pxor xmm3, xmm3 ; SAD
pxor xmm4, xmm4 ; sum of src2
pxor xmm5, xmm5 ; sum of src2^2
; Because we're working with the actual output frames
; we can't depend on any kind of data alignment.
.accumulate
movdqa xmm0, [rax] ; src1
movdqa xmm1, [rdx] ; src2
add rax, rcx ; src1 + stride1
add rdx, rdi ; src2 + stride2
; SAD(src1, src2)
psadbw xmm0, xmm1
paddusw xmm3, xmm0
; SUM(src2)
pxor xmm2, xmm2
psadbw xmm2, xmm1 ; sum src2 by misusing SAD against 0
paddusw xmm4, xmm2
; pmaddubsw would be ideal if it took two unsigned values. instead,
; it expects a signed and an unsigned value. so instead we zero extend
; and operate on words.
pxor xmm2, xmm2
movdqa xmm0, xmm1
punpcklbw xmm0, xmm2
punpckhbw xmm1, xmm2
pmaddwd xmm0, xmm0
pmaddwd xmm1, xmm1
paddd xmm5, xmm0
paddd xmm5, xmm1
sub rsi, 1
jnz .accumulate
; phaddd only operates on adjacent double words.
; Finalize SAD and store
movdqa xmm0, xmm3
psrldq xmm0, 8
paddusw xmm0, xmm3
paddd xmm0, [GLOBAL(t128)]
psrld xmm0, 8
mov rax, arg(5)
movd [rax], xmm0
; Accumulate sum of src2
movdqa xmm0, xmm4
psrldq xmm0, 8
paddusw xmm0, xmm4
; Square src2. Ignore high value
pmuludq xmm0, xmm0
psrld xmm0, 8
; phaddw could be used to sum adjacent values but we want
; all the values summed. promote to doubles, accumulate,
; shift and sum
pxor xmm2, xmm2
movdqa xmm1, xmm5
punpckldq xmm1, xmm2
punpckhdq xmm5, xmm2
paddd xmm1, xmm5
movdqa xmm2, xmm1
psrldq xmm1, 8
paddd xmm1, xmm2
psubd xmm1, xmm0
; (variance + 128) >> 8
paddd xmm1, [GLOBAL(t128)]
psrld xmm1, 8
mov rax, arg(4)
movd [rax], xmm1
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
SECTION_RODATA
align 16
t128:
%ifndef __NASM_VER__
ddq 128
%elif CONFIG_BIG_ENDIAN
dq 0, 128
%else
dq 128, 0
%endif
align 16
tMFQE: ; 1 << MFQE_PRECISION
times 8 dw 0x10
align 16
tMFQE_round: ; 1 << (MFQE_PRECISION - 1)
times 8 dw 0x08

694
vp10/common/x86/vp9_postproc_sse2.asm Normal file
View File

@ -0,0 +1,694 @@
;
; Copyright (c) 2010 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 "vpx_ports/x86_abi_support.asm"
;void vp10_post_proc_down_and_across_xmm
;(
; unsigned char *src_ptr,
; unsigned char *dst_ptr,
; int src_pixels_per_line,
; int dst_pixels_per_line,
; int rows,
; int cols,
; int flimit
;)
global sym(vp10_post_proc_down_and_across_xmm) PRIVATE
sym(vp10_post_proc_down_and_across_xmm):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 7
SAVE_XMM 7
GET_GOT rbx
push rsi
push rdi
; end prolog
%if ABI_IS_32BIT=1 && CONFIG_PIC=1
ALIGN_STACK 16, rax
; move the global rd onto the stack, since we don't have enough registers
; to do PIC addressing
movdqa xmm0, [GLOBAL(rd42)]
sub rsp, 16
movdqa [rsp], xmm0
%define RD42 [rsp]
%else
%define RD42 [GLOBAL(rd42)]
%endif
movd xmm2, dword ptr arg(6) ;flimit
punpcklwd xmm2, xmm2
punpckldq xmm2, xmm2
punpcklqdq xmm2, xmm2
mov rsi, arg(0) ;src_ptr
mov rdi, arg(1) ;dst_ptr
movsxd rcx, DWORD PTR arg(4) ;rows
movsxd rax, DWORD PTR arg(2) ;src_pixels_per_line ; destination pitch?
pxor xmm0, xmm0 ; mm0 = 00000000
.nextrow:
xor rdx, rdx ; clear out rdx for use as loop counter
.nextcol:
movq xmm3, QWORD PTR [rsi] ; mm4 = r0 p0..p7
punpcklbw xmm3, xmm0 ; mm3 = p0..p3
movdqa xmm1, xmm3 ; mm1 = p0..p3
psllw xmm3, 2 ;
movq xmm5, QWORD PTR [rsi + rax] ; mm4 = r1 p0..p7
punpcklbw xmm5, xmm0 ; mm5 = r1 p0..p3
paddusw xmm3, xmm5 ; mm3 += mm6
; thresholding
movdqa xmm7, xmm1 ; mm7 = r0 p0..p3
psubusw xmm7, xmm5 ; mm7 = r0 p0..p3 - r1 p0..p3
psubusw xmm5, xmm1 ; mm5 = r1 p0..p3 - r0 p0..p3
paddusw xmm7, xmm5 ; mm7 = abs(r0 p0..p3 - r1 p0..p3)
pcmpgtw xmm7, xmm2
movq xmm5, QWORD PTR [rsi + 2*rax] ; mm4 = r2 p0..p7
punpcklbw xmm5, xmm0 ; mm5 = r2 p0..p3
paddusw xmm3, xmm5 ; mm3 += mm5
; thresholding
movdqa xmm6, xmm1 ; mm6 = r0 p0..p3
psubusw xmm6, xmm5 ; mm6 = r0 p0..p3 - r2 p0..p3
psubusw xmm5, xmm1 ; mm5 = r2 p0..p3 - r2 p0..p3
paddusw xmm6, xmm5 ; mm6 = abs(r0 p0..p3 - r2 p0..p3)
pcmpgtw xmm6, xmm2
por xmm7, xmm6 ; accumulate thresholds
neg rax
movq xmm5, QWORD PTR [rsi+2*rax] ; mm4 = r-2 p0..p7
punpcklbw xmm5, xmm0 ; mm5 = r-2 p0..p3
paddusw xmm3, xmm5 ; mm3 += mm5
; thresholding
movdqa xmm6, xmm1 ; mm6 = r0 p0..p3
psubusw xmm6, xmm5 ; mm6 = p0..p3 - r-2 p0..p3
psubusw xmm5, xmm1 ; mm5 = r-2 p0..p3 - p0..p3
paddusw xmm6, xmm5 ; mm6 = abs(r0 p0..p3 - r-2 p0..p3)
pcmpgtw xmm6, xmm2
por xmm7, xmm6 ; accumulate thresholds
movq xmm4, QWORD PTR [rsi+rax] ; mm4 = r-1 p0..p7
punpcklbw xmm4, xmm0 ; mm4 = r-1 p0..p3
paddusw xmm3, xmm4 ; mm3 += mm5
; thresholding
movdqa xmm6, xmm1 ; mm6 = r0 p0..p3
psubusw xmm6, xmm4 ; mm6 = p0..p3 - r-2 p0..p3
psubusw xmm4, xmm1 ; mm5 = r-1 p0..p3 - p0..p3
paddusw xmm6, xmm4 ; mm6 = abs(r0 p0..p3 - r-1 p0..p3)
pcmpgtw xmm6, xmm2
por xmm7, xmm6 ; accumulate thresholds
paddusw xmm3, RD42 ; mm3 += round value
psraw xmm3, 3 ; mm3 /= 8
pand xmm1, xmm7 ; mm1 select vals > thresh from source
pandn xmm7, xmm3 ; mm7 select vals < thresh from blurred result
paddusw xmm1, xmm7 ; combination
packuswb xmm1, xmm0 ; pack to bytes
movq QWORD PTR [rdi], xmm1 ;
neg rax ; pitch is positive
add rsi, 8
add rdi, 8
add rdx, 8
cmp edx, dword arg(5) ;cols
jl .nextcol
; done with the all cols, start the across filtering in place
sub rsi, rdx
sub rdi, rdx
xor rdx, rdx
movq mm0, QWORD PTR [rdi-8];
.acrossnextcol:
movq xmm7, QWORD PTR [rdi +rdx -2]
movd xmm4, DWORD PTR [rdi +rdx +6]
pslldq xmm4, 8
por xmm4, xmm7
movdqa xmm3, xmm4
psrldq xmm3, 2
punpcklbw xmm3, xmm0 ; mm3 = p0..p3
movdqa xmm1, xmm3 ; mm1 = p0..p3
psllw xmm3, 2
movdqa xmm5, xmm4
psrldq xmm5, 3
punpcklbw xmm5, xmm0 ; mm5 = p1..p4
paddusw xmm3, xmm5 ; mm3 += mm6
; thresholding
movdqa xmm7, xmm1 ; mm7 = p0..p3
psubusw xmm7, xmm5 ; mm7 = p0..p3 - p1..p4
psubusw xmm5, xmm1 ; mm5 = p1..p4 - p0..p3
paddusw xmm7, xmm5 ; mm7 = abs(p0..p3 - p1..p4)
pcmpgtw xmm7, xmm2
movdqa xmm5, xmm4
psrldq xmm5, 4
punpcklbw xmm5, xmm0 ; mm5 = p2..p5
paddusw xmm3, xmm5 ; mm3 += mm5
; thresholding
movdqa xmm6, xmm1 ; mm6 = p0..p3
psubusw xmm6, xmm5 ; mm6 = p0..p3 - p1..p4
psubusw xmm5, xmm1 ; mm5 = p1..p4 - p0..p3
paddusw xmm6, xmm5 ; mm6 = abs(p0..p3 - p1..p4)
pcmpgtw xmm6, xmm2
por xmm7, xmm6 ; accumulate thresholds
movdqa xmm5, xmm4 ; mm5 = p-2..p5
punpcklbw xmm5, xmm0 ; mm5 = p-2..p1
paddusw xmm3, xmm5 ; mm3 += mm5
; thresholding
movdqa xmm6, xmm1 ; mm6 = p0..p3
psubusw xmm6, xmm5 ; mm6 = p0..p3 - p1..p4
psubusw xmm5, xmm1 ; mm5 = p1..p4 - p0..p3
paddusw xmm6, xmm5 ; mm6 = abs(p0..p3 - p1..p4)
pcmpgtw xmm6, xmm2
por xmm7, xmm6 ; accumulate thresholds
psrldq xmm4, 1 ; mm4 = p-1..p5
punpcklbw xmm4, xmm0 ; mm4 = p-1..p2
paddusw xmm3, xmm4 ; mm3 += mm5
; thresholding
movdqa xmm6, xmm1 ; mm6 = p0..p3
psubusw xmm6, xmm4 ; mm6 = p0..p3 - p1..p4
psubusw xmm4, xmm1 ; mm5 = p1..p4 - p0..p3
paddusw xmm6, xmm4 ; mm6 = abs(p0..p3 - p1..p4)
pcmpgtw xmm6, xmm2
por xmm7, xmm6 ; accumulate thresholds
paddusw xmm3, RD42 ; mm3 += round value
psraw xmm3, 3 ; mm3 /= 8
pand xmm1, xmm7 ; mm1 select vals > thresh from source
pandn xmm7, xmm3 ; mm7 select vals < thresh from blurred result
paddusw xmm1, xmm7 ; combination
packuswb xmm1, xmm0 ; pack to bytes
movq QWORD PTR [rdi+rdx-8], mm0 ; store previous four bytes
movdq2q mm0, xmm1
add rdx, 8
cmp edx, dword arg(5) ;cols
jl .acrossnextcol;
; last 8 pixels
movq QWORD PTR [rdi+rdx-8], mm0
; done with this rwo
add rsi,rax ; next line
mov eax, dword arg(3) ;dst_pixels_per_line ; destination pitch?
add rdi,rax ; next destination
mov eax, dword arg(2) ;src_pixels_per_line ; destination pitch?
dec rcx ; decrement count
jnz .nextrow ; next row
%if ABI_IS_32BIT=1 && CONFIG_PIC=1
add rsp,16
pop rsp
%endif
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
RESTORE_XMM
UNSHADOW_ARGS
pop rbp
ret
%undef RD42
;void vp10_mbpost_proc_down_xmm(unsigned char *dst,
; int pitch, int rows, int cols,int flimit)
extern sym(vp10_rv)
global sym(vp10_mbpost_proc_down_xmm) PRIVATE
sym(vp10_mbpost_proc_down_xmm):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
SAVE_XMM 7
GET_GOT rbx
push rsi
push rdi
; end prolog
ALIGN_STACK 16, rax
sub rsp, 128+16
; unsigned char d[16][8] at [rsp]
; create flimit2 at [rsp+128]
mov eax, dword ptr arg(4) ;flimit
mov [rsp+128], eax
mov [rsp+128+4], eax
mov [rsp+128+8], eax
mov [rsp+128+12], eax
%define flimit4 [rsp+128]
%if ABI_IS_32BIT=0
lea r8, [GLOBAL(sym(vp10_rv))]
%endif
;rows +=8;
add dword arg(2), 8
;for(c=0; c<cols; c+=8)
.loop_col:
mov rsi, arg(0) ; s
pxor xmm0, xmm0 ;
movsxd rax, dword ptr arg(1) ;pitch ;
neg rax ; rax = -pitch
lea rsi, [rsi + rax*8]; ; rdi = s[-pitch*8]
neg rax
pxor xmm5, xmm5
pxor xmm6, xmm6 ;
pxor xmm7, xmm7 ;
mov rdi, rsi
mov rcx, 15 ;
.loop_initvar:
movq xmm1, QWORD PTR [rdi];
punpcklbw xmm1, xmm0 ;
paddw xmm5, xmm1 ;
pmullw xmm1, xmm1 ;
movdqa xmm2, xmm1 ;
punpcklwd xmm1, xmm0 ;
punpckhwd xmm2, xmm0 ;
paddd xmm6, xmm1 ;
paddd xmm7, xmm2 ;
lea rdi, [rdi+rax] ;
dec rcx
jne .loop_initvar
;save the var and sum
xor rdx, rdx
.loop_row:
movq xmm1, QWORD PTR [rsi] ; [s-pitch*8]
movq xmm2, QWORD PTR [rdi] ; [s+pitch*7]
punpcklbw xmm1, xmm0
punpcklbw xmm2, xmm0
paddw xmm5, xmm2
psubw xmm5, xmm1
pmullw xmm2, xmm2
movdqa xmm4, xmm2
punpcklwd xmm2, xmm0
punpckhwd xmm4, xmm0
paddd xmm6, xmm2
paddd xmm7, xmm4
pmullw xmm1, xmm1
movdqa xmm2, xmm1
punpcklwd xmm1, xmm0
psubd xmm6, xmm1
punpckhwd xmm2, xmm0
psubd xmm7, xmm2
movdqa xmm3, xmm6
pslld xmm3, 4
psubd xmm3, xmm6
movdqa xmm1, xmm5
movdqa xmm4, xmm5
pmullw xmm1, xmm1
pmulhw xmm4, xmm4
movdqa xmm2, xmm1
punpcklwd xmm1, xmm4
punpckhwd xmm2, xmm4
movdqa xmm4, xmm7
pslld xmm4, 4
psubd xmm4, xmm7
psubd xmm3, xmm1
psubd xmm4, xmm2
psubd xmm3, flimit4
psubd xmm4, flimit4
psrad xmm3, 31
psrad xmm4, 31
packssdw xmm3, xmm4
packsswb xmm3, xmm0
movq xmm1, QWORD PTR [rsi+rax*8]
movq xmm2, xmm1
punpcklbw xmm1, xmm0
paddw xmm1, xmm5
mov rcx, rdx
and rcx, 127
%if ABI_IS_32BIT=1 && CONFIG_PIC=1
push rax
lea rax, [GLOBAL(sym(vp10_rv))]
movdqu xmm4, [rax + rcx*2] ;vp10_rv[rcx*2]
pop rax
%elif ABI_IS_32BIT=0
movdqu xmm4, [r8 + rcx*2] ;vp10_rv[rcx*2]
%else
movdqu xmm4, [sym(vp10_rv) + rcx*2]
%endif
paddw xmm1, xmm4
;paddw xmm1, eight8s
psraw xmm1, 4
packuswb xmm1, xmm0
pand xmm1, xmm3
pandn xmm3, xmm2
por xmm1, xmm3
and rcx, 15
movq QWORD PTR [rsp + rcx*8], xmm1 ;d[rcx*8]
mov rcx, rdx
sub rcx, 8
and rcx, 15
movq mm0, [rsp + rcx*8] ;d[rcx*8]
movq [rsi], mm0
lea rsi, [rsi+rax]
lea rdi, [rdi+rax]
add rdx, 1
cmp edx, dword arg(2) ;rows
jl .loop_row
add dword arg(0), 8 ; s += 8
sub dword arg(3), 8 ; cols -= 8
cmp dword arg(3), 0
jg .loop_col
add rsp, 128+16
pop rsp
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
RESTORE_XMM
UNSHADOW_ARGS
pop rbp
ret
%undef flimit4
;void vp10_mbpost_proc_across_ip_xmm(unsigned char *src,
; int pitch, int rows, int cols,int flimit)
global sym(vp10_mbpost_proc_across_ip_xmm) PRIVATE
sym(vp10_mbpost_proc_across_ip_xmm):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
SAVE_XMM 7
GET_GOT rbx
push rsi
push rdi
; end prolog
ALIGN_STACK 16, rax
sub rsp, 16
; create flimit4 at [rsp]
mov eax, dword ptr arg(4) ;flimit
mov [rsp], eax
mov [rsp+4], eax
mov [rsp+8], eax
mov [rsp+12], eax
%define flimit4 [rsp]
;for(r=0;r<rows;r++)
.ip_row_loop:
xor rdx, rdx ;sumsq=0;
xor rcx, rcx ;sum=0;
mov rsi, arg(0); s
mov rdi, -8
.ip_var_loop:
;for(i=-8;i<=6;i++)
;{
; sumsq += s[i]*s[i];
; sum += s[i];
;}
movzx eax, byte [rsi+rdi]
add ecx, eax
mul al
add edx, eax
add rdi, 1
cmp rdi, 6
jle .ip_var_loop
;mov rax, sumsq
;movd xmm7, rax
movd xmm7, edx
;mov rax, sum
;movd xmm6, rax
movd xmm6, ecx
mov rsi, arg(0) ;s
xor rcx, rcx
movsxd rdx, dword arg(3) ;cols
add rdx, 8
pxor mm0, mm0
pxor mm1, mm1
pxor xmm0, xmm0
.nextcol4:
movd xmm1, DWORD PTR [rsi+rcx-8] ; -8 -7 -6 -5
movd xmm2, DWORD PTR [rsi+rcx+7] ; +7 +8 +9 +10
punpcklbw xmm1, xmm0 ; expanding
punpcklbw xmm2, xmm0 ; expanding
punpcklwd xmm1, xmm0 ; expanding to dwords
punpcklwd xmm2, xmm0 ; expanding to dwords
psubd xmm2, xmm1 ; 7--8 8--7 9--6 10--5
paddd xmm1, xmm1 ; -8*2 -7*2 -6*2 -5*2
paddd xmm1, xmm2 ; 7+-8 8+-7 9+-6 10+-5
pmaddwd xmm1, xmm2 ; squared of 7+-8 8+-7 9+-6 10+-5
paddd xmm6, xmm2
paddd xmm7, xmm1
pshufd xmm6, xmm6, 0 ; duplicate the last ones
pshufd xmm7, xmm7, 0 ; duplicate the last ones
psrldq xmm1, 4 ; 8--7 9--6 10--5 0000
psrldq xmm2, 4 ; 8--7 9--6 10--5 0000
pshufd xmm3, xmm1, 3 ; 0000 8--7 8--7 8--7 squared
pshufd xmm4, xmm2, 3 ; 0000 8--7 8--7 8--7 squared
paddd xmm6, xmm4
paddd xmm7, xmm3
pshufd xmm3, xmm1, 01011111b ; 0000 0000 9--6 9--6 squared
pshufd xmm4, xmm2, 01011111b ; 0000 0000 9--6 9--6 squared
paddd xmm7, xmm3
paddd xmm6, xmm4
pshufd xmm3, xmm1, 10111111b ; 0000 0000 8--7 8--7 squared
pshufd xmm4, xmm2, 10111111b ; 0000 0000 8--7 8--7 squared
paddd xmm7, xmm3
paddd xmm6, xmm4
movdqa xmm3, xmm6
pmaddwd xmm3, xmm3
movdqa xmm5, xmm7
pslld xmm5, 4
psubd xmm5, xmm7
psubd xmm5, xmm3
psubd xmm5, flimit4
psrad xmm5, 31
packssdw xmm5, xmm0
packsswb xmm5, xmm0
movd xmm1, DWORD PTR [rsi+rcx]
movq xmm2, xmm1
punpcklbw xmm1, xmm0
punpcklwd xmm1, xmm0
paddd xmm1, xmm6
paddd xmm1, [GLOBAL(four8s)]
psrad xmm1, 4
packssdw xmm1, xmm0
packuswb xmm1, xmm0
pand xmm1, xmm5
pandn xmm5, xmm2
por xmm5, xmm1
movd [rsi+rcx-8], mm0
movq mm0, mm1
movdq2q mm1, xmm5
psrldq xmm7, 12
psrldq xmm6, 12
add rcx, 4
cmp rcx, rdx
jl .nextcol4
;s+=pitch;
movsxd rax, dword arg(1)
add arg(0), rax
sub dword arg(2), 1 ;rows-=1
cmp dword arg(2), 0
jg .ip_row_loop
add rsp, 16
pop rsp
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
RESTORE_XMM
UNSHADOW_ARGS
pop rbp
ret
%undef flimit4
;void vp10_plane_add_noise_wmt (unsigned char *start, unsigned char *noise,
; unsigned char blackclamp[16],
; unsigned char whiteclamp[16],
; unsigned char bothclamp[16],
; unsigned int width, unsigned int height, int pitch)
global sym(vp10_plane_add_noise_wmt) PRIVATE
sym(vp10_plane_add_noise_wmt):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 8
GET_GOT rbx
push rsi
push rdi
; end prolog
.addnoise_loop:
call sym(LIBVPX_RAND) WRT_PLT
mov rcx, arg(1) ;noise
and rax, 0xff
add rcx, rax
; we rely on the fact that the clamping vectors are stored contiguously
; in black/white/both order. Note that we have to reload this here because
; rdx could be trashed by rand()
mov rdx, arg(2) ; blackclamp
mov rdi, rcx
movsxd rcx, dword arg(5) ;[Width]
mov rsi, arg(0) ;Pos
xor rax,rax
.addnoise_nextset:
movdqu xmm1,[rsi+rax] ; get the source
psubusb xmm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
paddusb xmm1, [rdx+32] ;bothclamp
psubusb xmm1, [rdx+16] ;whiteclamp
movdqu xmm2,[rdi+rax] ; get the noise for this line
paddb xmm1,xmm2 ; add it in
movdqu [rsi+rax],xmm1 ; store the result
add rax,16 ; move to the next line
cmp rax, rcx
jl .addnoise_nextset
movsxd rax, dword arg(7) ; Pitch
add arg(0), rax ; Start += Pitch
sub dword arg(6), 1 ; Height -= 1
jg .addnoise_loop
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
SECTION_RODATA
align 16
rd42:
times 8 dw 0x04
four8s:
times 4 dd 8

2247
vp10/decoder/vp9_decodeframe.c Normal file
View File

File diff suppressed because it is too large Load Diff

35
vp10/decoder/vp9_decodeframe.h Normal file
View File

@ -0,0 +1,35 @@
/*
* Copyright (c) 2010 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 VP9_DECODER_VP9_DECODEFRAME_H_
#define VP9_DECODER_VP9_DECODEFRAME_H_
#ifdef __cplusplus
extern "C" {
#endif
struct VP9Decoder;
struct vpx_read_bit_buffer;
int vp10_read_sync_code(struct vpx_read_bit_buffer *const rb);
void vp10_read_frame_size(struct vpx_read_bit_buffer *rb,
int *width, int *height);
BITSTREAM_PROFILE vp10_read_profile(struct vpx_read_bit_buffer *rb);
void vp10_decode_frame(struct VP9Decoder *pbi,
const uint8_t *data, const uint8_t *data_end,
const uint8_t **p_data_end);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_DECODER_VP9_DECODEFRAME_H_

631
vp10/decoder/vp9_decodemv.c Normal file
View File

@ -0,0 +1,631 @@
/*
Copyright (c) 2010 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 <assert.h>
#include "vp10/common/vp9_common.h"
#include "vp10/common/vp9_entropy.h"
#include "vp10/common/vp9_entropymode.h"
#include "vp10/common/vp9_entropymv.h"
#include "vp10/common/vp9_mvref_common.h"
#include "vp10/common/vp9_pred_common.h"
#include "vp10/common/vp9_reconinter.h"
#include "vp10/common/vp9_seg_common.h"
#include "vp10/decoder/vp9_decodemv.h"
#include "vp10/decoder/vp9_decodeframe.h"
static PREDICTION_MODE read_intra_mode(vpx_reader *r, const vpx_prob *p) {
return (PREDICTION_MODE)vpx_read_tree(r, vp10_intra_mode_tree, p);
}
static PREDICTION_MODE read_intra_mode_y(VP9_COMMON *cm, MACROBLOCKD *xd,
vpx_reader *r, int size_group) {
const PREDICTION_MODE y_mode =
read_intra_mode(r, cm->fc->y_mode_prob[size_group]);
FRAME_COUNTS *counts = xd->counts;
if (counts)
++counts->y_mode[size_group][y_mode];
return y_mode;
}
static PREDICTION_MODE read_intra_mode_uv(VP9_COMMON *cm, MACROBLOCKD *xd,
vpx_reader *r,
PREDICTION_MODE y_mode) {
const PREDICTION_MODE uv_mode = read_intra_mode(r,
cm->fc->uv_mode_prob[y_mode]);
FRAME_COUNTS *counts = xd->counts;
if (counts)
++counts->uv_mode[y_mode][uv_mode];
return uv_mode;
}
static PREDICTION_MODE read_inter_mode(VP9_COMMON *cm, MACROBLOCKD *xd,
vpx_reader *r, int ctx) {
const int mode = vpx_read_tree(r, vp10_inter_mode_tree,
cm->fc->inter_mode_probs[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts)
++counts->inter_mode[ctx][mode];
return NEARESTMV + mode;
}
static int read_segment_id(vpx_reader *r, const struct segmentation *seg) {
return vpx_read_tree(r, vp10_segment_tree, seg->tree_probs);
}
static TX_SIZE read_selected_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd,
TX_SIZE max_tx_size, vpx_reader *r) {
FRAME_COUNTS *counts = xd->counts;
const int ctx = get_tx_size_context(xd);
const vpx_prob *tx_probs = get_tx_probs(max_tx_size, ctx, &cm->fc->tx_probs);
int tx_size = vpx_read(r, tx_probs[0]);
if (tx_size != TX_4X4 && max_tx_size >= TX_16X16) {
tx_size += vpx_read(r, tx_probs[1]);
if (tx_size != TX_8X8 && max_tx_size >= TX_32X32)
tx_size += vpx_read(r, tx_probs[2]);
}
if (counts)
++get_tx_counts(max_tx_size, ctx, &counts->tx)[tx_size];
return (TX_SIZE)tx_size;
}
static TX_SIZE read_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd,
int allow_select, vpx_reader *r) {
TX_MODE tx_mode = cm->tx_mode;
BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
const TX_SIZE max_tx_size = max_txsize_lookup[bsize];
if (allow_select && tx_mode == TX_MODE_SELECT && bsize >= BLOCK_8X8)
return read_selected_tx_size(cm, xd, max_tx_size, r);
else
return MIN(max_tx_size, tx_mode_to_biggest_tx_size[tx_mode]);
}
static int dec_get_segment_id(const VP9_COMMON *cm, const uint8_t *segment_ids,
int mi_offset, int x_mis, int y_mis) {
int x, y, segment_id = INT_MAX;
for (y = 0; y < y_mis; y++)
for (x = 0; x < x_mis; x++)
segment_id = MIN(segment_id,
segment_ids[mi_offset + y * cm->mi_cols + x]);
assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
return segment_id;
}
static void set_segment_id(VP9_COMMON *cm, int mi_offset,
int x_mis, int y_mis, int segment_id) {
int x, y;
assert(segment_id >= 0 && segment_id < MAX_SEGMENTS);
for (y = 0; y < y_mis; y++)
for (x = 0; x < x_mis; x++)
cm->current_frame_seg_map[mi_offset + y * cm->mi_cols + x] = segment_id;
}
static void copy_segment_id(const VP9_COMMON *cm,
const uint8_t *last_segment_ids,
uint8_t *current_segment_ids,
int mi_offset, int x_mis, int y_mis) {
int x, y;
for (y = 0; y < y_mis; y++)
for (x = 0; x < x_mis; x++)
current_segment_ids[mi_offset + y * cm->mi_cols + x] = last_segment_ids ?
last_segment_ids[mi_offset + y * cm->mi_cols + x] : 0;
}
static int read_intra_segment_id(VP9_COMMON *const cm, int mi_offset,
int x_mis, int y_mis,
vpx_reader *r) {
struct segmentation *const seg = &cm->seg;
int segment_id;
if (!seg->enabled)
return 0; // Default for disabled segmentation
if (!seg->update_map) {
copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map,
mi_offset, x_mis, y_mis);
return 0;
}
segment_id = read_segment_id(r, seg);
set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id);
return segment_id;
}
static int read_inter_segment_id(VP9_COMMON *const cm, MACROBLOCKD *const xd,
int mi_row, int mi_col, vpx_reader *r) {
struct segmentation *const seg = &cm->seg;
MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
int predicted_segment_id, segment_id;
const int mi_offset = mi_row * cm->mi_cols + mi_col;
const int bw = xd->plane[0].n4_w >> 1;
const int bh = xd->plane[0].n4_h >> 1;
// TODO(slavarnway): move x_mis, y_mis into xd ?????
const int x_mis = MIN(cm->mi_cols - mi_col, bw);
const int y_mis = MIN(cm->mi_rows - mi_row, bh);
if (!seg->enabled)
return 0; // Default for disabled segmentation
predicted_segment_id = cm->last_frame_seg_map ?
dec_get_segment_id(cm, cm->last_frame_seg_map, mi_offset, x_mis, y_mis) :
0;
if (!seg->update_map) {
copy_segment_id(cm, cm->last_frame_seg_map, cm->current_frame_seg_map,
mi_offset, x_mis, y_mis);
return predicted_segment_id;
}
if (seg->temporal_update) {
const vpx_prob pred_prob = vp10_get_pred_prob_seg_id(seg, xd);
mbmi->seg_id_predicted = vpx_read(r, pred_prob);
segment_id = mbmi->seg_id_predicted ? predicted_segment_id
: read_segment_id(r, seg);
} else {
segment_id = read_segment_id(r, seg);
}
set_segment_id(cm, mi_offset, x_mis, y_mis, segment_id);
return segment_id;
}
static int read_skip(VP9_COMMON *cm, const MACROBLOCKD *xd,
int segment_id, vpx_reader *r) {
if (segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP)) {
return 1;
} else {
const int ctx = vp10_get_skip_context(xd);
const int skip = vpx_read(r, cm->fc->skip_probs[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts)
++counts->skip[ctx][skip];
return skip;
}
}
static void read_intra_frame_mode_info(VP9_COMMON *const cm,
MACROBLOCKD *const xd,
int mi_row, int mi_col, vpx_reader *r) {
MODE_INFO *const mi = xd->mi[0];
MB_MODE_INFO *const mbmi = &mi->mbmi;
const MODE_INFO *above_mi = xd->above_mi;
const MODE_INFO *left_mi = xd->left_mi;
const BLOCK_SIZE bsize = mbmi->sb_type;
int i;
const int mi_offset = mi_row * cm->mi_cols + mi_col;
const int bw = xd->plane[0].n4_w >> 1;
const int bh = xd->plane[0].n4_h >> 1;
// TODO(slavarnway): move x_mis, y_mis into xd ?????
const int x_mis = MIN(cm->mi_cols - mi_col, bw);
const int y_mis = MIN(cm->mi_rows - mi_row, bh);
mbmi->segment_id = read_intra_segment_id(cm, mi_offset, x_mis, y_mis, r);
mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r);
mbmi->tx_size = read_tx_size(cm, xd, 1, r);
mbmi->ref_frame[0] = INTRA_FRAME;
mbmi->ref_frame[1] = NONE;
switch (bsize) {
case BLOCK_4X4:
for (i = 0; i < 4; ++i)
mi->bmi[i].as_mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, i));
mbmi->mode = mi->bmi[3].as_mode;
break;
case BLOCK_4X8:
mi->bmi[0].as_mode = mi->bmi[2].as_mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0));
mi->bmi[1].as_mode = mi->bmi[3].as_mode = mbmi->mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 1));
break;
case BLOCK_8X4:
mi->bmi[0].as_mode = mi->bmi[1].as_mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 0));
mi->bmi[2].as_mode = mi->bmi[3].as_mode = mbmi->mode =
read_intra_mode(r, get_y_mode_probs(mi, above_mi, left_mi, 2));
break;
default:
mbmi->mode = read_intra_mode(r,
get_y_mode_probs(mi, above_mi, left_mi, 0));
}
mbmi->uv_mode = read_intra_mode(r, vp10_kf_uv_mode_prob[mbmi->mode]);
}
static int read_mv_component(vpx_reader *r,
const nmv_component *mvcomp, int usehp) {
int mag, d, fr, hp;
const int sign = vpx_read(r, mvcomp->sign);
const int mv_class = vpx_read_tree(r, vp10_mv_class_tree, mvcomp->classes);
const int class0 = mv_class == MV_CLASS_0;
// Integer part
if (class0) {
d = vpx_read_tree(r, vp10_mv_class0_tree, mvcomp->class0);
mag = 0;
} else {
int i;
const int n = mv_class + CLASS0_BITS - 1; // number of bits
d = 0;
for (i = 0; i < n; ++i)
d |= vpx_read(r, mvcomp->bits[i]) << i;
mag = CLASS0_SIZE << (mv_class + 2);
}
// Fractional part
fr = vpx_read_tree(r, vp10_mv_fp_tree, class0 ? mvcomp->class0_fp[d]
: mvcomp->fp);
// High precision part (if hp is not used, the default value of the hp is 1)
hp = usehp ? vpx_read(r, class0 ? mvcomp->class0_hp : mvcomp->hp)
: 1;
// Result
mag += ((d << 3) | (fr << 1) | hp) + 1;
return sign ? -mag : mag;
}
static INLINE void read_mv(vpx_reader *r, MV *mv, const MV *ref,
const nmv_context *ctx,
nmv_context_counts *counts, int allow_hp) {
const MV_JOINT_TYPE joint_type =
(MV_JOINT_TYPE)vpx_read_tree(r, vp10_mv_joint_tree, ctx->joints);
const int use_hp = allow_hp && vp10_use_mv_hp(ref);
MV diff = {0, 0};
if (mv_joint_vertical(joint_type))
diff.row = read_mv_component(r, &ctx->comps[0], use_hp);
if (mv_joint_horizontal(joint_type))
diff.col = read_mv_component(r, &ctx->comps[1], use_hp);
vp10_inc_mv(&diff, counts);
mv->row = ref->row + diff.row;
mv->col = ref->col + diff.col;
}
static REFERENCE_MODE read_block_reference_mode(VP9_COMMON *cm,
const MACROBLOCKD *xd,
vpx_reader *r) {
if (cm->reference_mode == REFERENCE_MODE_SELECT) {
const int ctx = vp10_get_reference_mode_context(cm, xd);
const REFERENCE_MODE mode =
(REFERENCE_MODE)vpx_read(r, cm->fc->comp_inter_prob[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts)
++counts->comp_inter[ctx][mode];
return mode; // SINGLE_REFERENCE or COMPOUND_REFERENCE
} else {
return cm->reference_mode;
}
}
// Read the referncence frame
static void read_ref_frames(VP9_COMMON *const cm, MACROBLOCKD *const xd,
vpx_reader *r,
int segment_id, MV_REFERENCE_FRAME ref_frame[2]) {
FRAME_CONTEXT *const fc = cm->fc;
FRAME_COUNTS *counts = xd->counts;
if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
ref_frame[0] = (MV_REFERENCE_FRAME)get_segdata(&cm->seg, segment_id,
SEG_LVL_REF_FRAME);
ref_frame[1] = NONE;
} else {
const REFERENCE_MODE mode = read_block_reference_mode(cm, xd, r);
// FIXME(rbultje) I'm pretty sure this breaks segmentation ref frame coding
if (mode == COMPOUND_REFERENCE) {
const int idx = cm->ref_frame_sign_bias[cm->comp_fixed_ref];
const int ctx = vp10_get_pred_context_comp_ref_p(cm, xd);
const int bit = vpx_read(r, fc->comp_ref_prob[ctx]);
if (counts)
++counts->comp_ref[ctx][bit];
ref_frame[idx] = cm->comp_fixed_ref;
ref_frame[!idx] = cm->comp_var_ref[bit];
} else if (mode == SINGLE_REFERENCE) {
const int ctx0 = vp10_get_pred_context_single_ref_p1(xd);
const int bit0 = vpx_read(r, fc->single_ref_prob[ctx0][0]);
if (counts)
++counts->single_ref[ctx0][0][bit0];
if (bit0) {
const int ctx1 = vp10_get_pred_context_single_ref_p2(xd);
const int bit1 = vpx_read(r, fc->single_ref_prob[ctx1][1]);
if (counts)
++counts->single_ref[ctx1][1][bit1];
ref_frame[0] = bit1 ? ALTREF_FRAME : GOLDEN_FRAME;
} else {
ref_frame[0] = LAST_FRAME;
}
ref_frame[1] = NONE;
} else {
assert(0 && "Invalid prediction mode.");
}
}
}
static INLINE INTERP_FILTER read_switchable_interp_filter(
VP9_COMMON *const cm, MACROBLOCKD *const xd,
vpx_reader *r) {
const int ctx = vp10_get_pred_context_switchable_interp(xd);
const INTERP_FILTER type =
(INTERP_FILTER)vpx_read_tree(r, vp10_switchable_interp_tree,
cm->fc->switchable_interp_prob[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts)
++counts->switchable_interp[ctx][type];
return type;
}
static void read_intra_block_mode_info(VP9_COMMON *const cm,
MACROBLOCKD *const xd, MODE_INFO *mi,
vpx_reader *r) {
MB_MODE_INFO *const mbmi = &mi->mbmi;
const BLOCK_SIZE bsize = mi->mbmi.sb_type;
int i;
mbmi->ref_frame[0] = INTRA_FRAME;
mbmi->ref_frame[1] = NONE;
switch (bsize) {
case BLOCK_4X4:
for (i = 0; i < 4; ++i)
mi->bmi[i].as_mode = read_intra_mode_y(cm, xd, r, 0);
mbmi->mode = mi->bmi[3].as_mode;
break;
case BLOCK_4X8:
mi->bmi[0].as_mode = mi->bmi[2].as_mode = read_intra_mode_y(cm, xd,
r, 0);
mi->bmi[1].as_mode = mi->bmi[3].as_mode = mbmi->mode =
read_intra_mode_y(cm, xd, r, 0);
break;
case BLOCK_8X4:
mi->bmi[0].as_mode = mi->bmi[1].as_mode = read_intra_mode_y(cm, xd,
r, 0);
mi->bmi[2].as_mode = mi->bmi[3].as_mode = mbmi->mode =
read_intra_mode_y(cm, xd, r, 0);
break;
default:
mbmi->mode = read_intra_mode_y(cm, xd, r, size_group_lookup[bsize]);
}
mbmi->uv_mode = read_intra_mode_uv(cm, xd, r, mbmi->mode);
}
static INLINE int is_mv_valid(const MV *mv) {
return mv->row > MV_LOW && mv->row < MV_UPP &&
mv->col > MV_LOW && mv->col < MV_UPP;
}
static INLINE int assign_mv(VP9_COMMON *cm, MACROBLOCKD *xd,
PREDICTION_MODE mode,
int_mv mv[2], int_mv ref_mv[2],
int_mv nearest_mv[2], int_mv near_mv[2],
int is_compound, int allow_hp, vpx_reader *r) {
int i;
int ret = 1;
switch (mode) {
case NEWMV: {
FRAME_COUNTS *counts = xd->counts;
nmv_context_counts *const mv_counts = counts ? &counts->mv : NULL;
for (i = 0; i < 1 + is_compound; ++i) {
read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, &cm->fc->nmvc, mv_counts,
allow_hp);
ret = ret && is_mv_valid(&mv[i].as_mv);
}
break;
}
case NEARESTMV: {
mv[0].as_int = nearest_mv[0].as_int;
if (is_compound)
mv[1].as_int = nearest_mv[1].as_int;
break;
}
case NEARMV: {
mv[0].as_int = near_mv[0].as_int;
if (is_compound)
mv[1].as_int = near_mv[1].as_int;
break;
}
case ZEROMV: {
mv[0].as_int = 0;
if (is_compound)
mv[1].as_int = 0;
break;
}
default: {
return 0;
}
}
return ret;
}
static int read_is_inter_block(VP9_COMMON *const cm, MACROBLOCKD *const xd,
int segment_id, vpx_reader *r) {
if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
return get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME) != INTRA_FRAME;
} else {
const int ctx = vp10_get_intra_inter_context(xd);
const int is_inter = vpx_read(r, cm->fc->intra_inter_prob[ctx]);
FRAME_COUNTS *counts = xd->counts;
if (counts)
++counts->intra_inter[ctx][is_inter];
return is_inter;
}
}
static void fpm_sync(void *const data, int mi_row) {
VP9Decoder *const pbi = (VP9Decoder *)data;
vp10_frameworker_wait(pbi->frame_worker_owner, pbi->common.prev_frame,
mi_row << MI_BLOCK_SIZE_LOG2);
}
static void read_inter_block_mode_info(VP9Decoder *const pbi,
MACROBLOCKD *const xd,
MODE_INFO *const mi,
int mi_row, int mi_col, vpx_reader *r) {
VP9_COMMON *const cm = &pbi->common;
MB_MODE_INFO *const mbmi = &mi->mbmi;
const BLOCK_SIZE bsize = mbmi->sb_type;
const int allow_hp = cm->allow_high_precision_mv;
int_mv nearestmv[2], nearmv[2];
int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
int ref, is_compound;
uint8_t inter_mode_ctx[MAX_REF_FRAMES];
read_ref_frames(cm, xd, r, mbmi->segment_id, mbmi->ref_frame);
is_compound = has_second_ref(mbmi);
for (ref = 0; ref < 1 + is_compound; ++ref) {
const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];
xd->block_refs[ref] = ref_buf;
if ((!vp10_is_valid_scale(&ref_buf->sf)))
vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM,
"Reference frame has invalid dimensions");
vp10_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col,
&ref_buf->sf);
vp10_find_mv_refs(cm, xd, mi, frame, ref_mvs[frame],
mi_row, mi_col, fpm_sync, (void *)pbi, inter_mode_ctx);
}
if (segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
mbmi->mode = ZEROMV;
if (bsize < BLOCK_8X8) {
vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM,
"Invalid usage of segement feature on small blocks");
return;
}
} else {
if (bsize >= BLOCK_8X8)
mbmi->mode = read_inter_mode(cm, xd, r,
inter_mode_ctx[mbmi->ref_frame[0]]);
}
if (bsize < BLOCK_8X8 || mbmi->mode != ZEROMV) {
for (ref = 0; ref < 1 + is_compound; ++ref) {
vp10_find_best_ref_mvs(xd, allow_hp, ref_mvs[mbmi->ref_frame[ref]],
&nearestmv[ref], &nearmv[ref]);
}
}
mbmi->interp_filter = (cm->interp_filter == SWITCHABLE)
? read_switchable_interp_filter(cm, xd, r)
: cm->interp_filter;
if (bsize < BLOCK_8X8) {
const int num_4x4_w = 1 << xd->bmode_blocks_wl;
const int num_4x4_h = 1 << xd->bmode_blocks_hl;
int idx, idy;
PREDICTION_MODE b_mode;
int_mv nearest_sub8x8[2], near_sub8x8[2];
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
int_mv block[2];
const int j = idy * 2 + idx;
b_mode = read_inter_mode(cm, xd, r, inter_mode_ctx[mbmi->ref_frame[0]]);
if (b_mode == NEARESTMV || b_mode == NEARMV) {
uint8_t dummy_mode_ctx[MAX_REF_FRAMES];
for (ref = 0; ref < 1 + is_compound; ++ref)
vp10_append_sub8x8_mvs_for_idx(cm, xd, j, ref, mi_row, mi_col,
&nearest_sub8x8[ref],
&near_sub8x8[ref],
dummy_mode_ctx);
}
if (!assign_mv(cm, xd, b_mode, block, nearestmv,
nearest_sub8x8, near_sub8x8,
is_compound, allow_hp, r)) {
xd->corrupted |= 1;
break;
};
mi->bmi[j].as_mv[0].as_int = block[0].as_int;
if (is_compound)
mi->bmi[j].as_mv[1].as_int = block[1].as_int;
if (num_4x4_h == 2)
mi->bmi[j + 2] = mi->bmi[j];
if (num_4x4_w == 2)
mi->bmi[j + 1] = mi->bmi[j];
}
}
mi->mbmi.mode = b_mode;
mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
} else {
xd->corrupted |= !assign_mv(cm, xd, mbmi->mode, mbmi->mv, nearestmv,
nearestmv, nearmv, is_compound, allow_hp, r);
}
}
static void read_inter_frame_mode_info(VP9Decoder *const pbi,
MACROBLOCKD *const xd,
int mi_row, int mi_col, vpx_reader *r) {
VP9_COMMON *const cm = &pbi->common;
MODE_INFO *const mi = xd->mi[0];
MB_MODE_INFO *const mbmi = &mi->mbmi;
int inter_block;
mbmi->mv[0].as_int = 0;
mbmi->mv[1].as_int = 0;
mbmi->segment_id = read_inter_segment_id(cm, xd, mi_row, mi_col, r);
mbmi->skip = read_skip(cm, xd, mbmi->segment_id, r);
inter_block = read_is_inter_block(cm, xd, mbmi->segment_id, r);
mbmi->tx_size = read_tx_size(cm, xd, !mbmi->skip || !inter_block, r);
if (inter_block)
read_inter_block_mode_info(pbi, xd, mi, mi_row, mi_col, r);
else
read_intra_block_mode_info(cm, xd, mi, r);
}
void vp10_read_mode_info(VP9Decoder *const pbi, MACROBLOCKD *xd,
int mi_row, int mi_col, vpx_reader *r,
int x_mis, int y_mis) {
VP9_COMMON *const cm = &pbi->common;
MODE_INFO *const mi = xd->mi[0];
MV_REF* frame_mvs = cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col;
int w, h;
if (frame_is_intra_only(cm)) {
read_intra_frame_mode_info(cm, xd, mi_row, mi_col, r);
} else {
read_inter_frame_mode_info(pbi, xd, mi_row, mi_col, r);
for (h = 0; h < y_mis; ++h) {
MV_REF *const frame_mv = frame_mvs + h * cm->mi_cols;
for (w = 0; w < x_mis; ++w) {
MV_REF *const mv = frame_mv + w;
mv->ref_frame[0] = mi->mbmi.ref_frame[0];
mv->ref_frame[1] = mi->mbmi.ref_frame[1];
mv->mv[0].as_int = mi->mbmi.mv[0].as_int;
mv->mv[1].as_int = mi->mbmi.mv[1].as_int;
}
}
}
}

30
vp10/decoder/vp9_decodemv.h Normal file
View File

@ -0,0 +1,30 @@
/*
* Copyright (c) 2010 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 VP9_DECODER_VP9_DECODEMV_H_
#define VP9_DECODER_VP9_DECODEMV_H_
#include "vpx_dsp/bitreader.h"
#include "vp10/decoder/vp9_decoder.h"
#ifdef __cplusplus
extern "C" {
#endif
void vp10_read_mode_info(VP9Decoder *const pbi, MACROBLOCKD *xd,
int mi_row, int mi_col, vpx_reader *r,
int x_mis, int y_mis);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_DECODER_VP9_DECODEMV_H_

509
vp10/decoder/vp9_decoder.c Normal file
View File

@ -0,0 +1,509 @@
/*
* Copyright (c) 2010 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 <assert.h>
#include <limits.h>
#include <stdio.h>
#include "./vp10_rtcd.h"
#include "./vpx_dsp_rtcd.h"
#include "./vpx_scale_rtcd.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/vpx_once.h"
#include "vpx_ports/vpx_timer.h"
#include "vpx_scale/vpx_scale.h"
#include "vpx_util/vpx_thread.h"
#include "vp10/common/vp9_alloccommon.h"
#include "vp10/common/vp9_loopfilter.h"
#include "vp10/common/vp9_onyxc_int.h"
#if CONFIG_VP9_POSTPROC
#include "vp10/common/vp9_postproc.h"
#endif
#include "vp10/common/vp9_quant_common.h"
#include "vp10/common/vp9_reconintra.h"
#include "vp10/common/vp9_systemdependent.h"
#include "vp10/decoder/vp9_decodeframe.h"
#include "vp10/decoder/vp9_decoder.h"
#include "vp10/decoder/vp9_detokenize.h"
static void initialize_dec(void) {
static volatile int init_done = 0;
if (!init_done) {
vp10_rtcd();
vpx_dsp_rtcd();
vpx_scale_rtcd();
vp10_init_intra_predictors();
init_done = 1;
}
}
static void vp10_dec_setup_mi(VP9_COMMON *cm) {
cm->mi = cm->mip + cm->mi_stride + 1;
cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1;
memset(cm->mi_grid_base, 0,
cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mi_grid_base));
}
static int vp10_dec_alloc_mi(VP9_COMMON *cm, int mi_size) {
cm->mip = vpx_calloc(mi_size, sizeof(*cm->mip));
if (!cm->mip)
return 1;
cm->mi_alloc_size = mi_size;
cm->mi_grid_base = (MODE_INFO **)vpx_calloc(mi_size, sizeof(MODE_INFO*));
if (!cm->mi_grid_base)
return 1;
return 0;
}
static void vp10_dec_free_mi(VP9_COMMON *cm) {
vpx_free(cm->mip);
cm->mip = NULL;
vpx_free(cm->mi_grid_base);
cm->mi_grid_base = NULL;
}
VP9Decoder *vp10_decoder_create(BufferPool *const pool) {
VP9Decoder *volatile const pbi = vpx_memalign(32, sizeof(*pbi));
VP9_COMMON *volatile const cm = pbi ? &pbi->common : NULL;
if (!cm)
return NULL;
vp10_zero(*pbi);
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
vp10_decoder_remove(pbi);
return NULL;
}
cm->error.setjmp = 1;
CHECK_MEM_ERROR(cm, cm->fc,
(FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
CHECK_MEM_ERROR(cm, cm->frame_contexts,
(FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
sizeof(*cm->frame_contexts)));
pbi->need_resync = 1;
once(initialize_dec);
// Initialize the references to not point to any frame buffers.
memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map));
cm->current_video_frame = 0;
pbi->ready_for_new_data = 1;
pbi->common.buffer_pool = pool;
cm->bit_depth = VPX_BITS_8;
cm->dequant_bit_depth = VPX_BITS_8;
cm->alloc_mi = vp10_dec_alloc_mi;
cm->free_mi = vp10_dec_free_mi;
cm->setup_mi = vp10_dec_setup_mi;
vp10_loop_filter_init(cm);
cm->error.setjmp = 0;
vpx_get_worker_interface()->init(&pbi->lf_worker);
return pbi;
}
void vp10_decoder_remove(VP9Decoder *pbi) {
int i;
vpx_get_worker_interface()->end(&pbi->lf_worker);
vpx_free(pbi->lf_worker.data1);
vpx_free(pbi->tile_data);
for (i = 0; i < pbi->num_tile_workers; ++i) {
VPxWorker *const worker = &pbi->tile_workers[i];
vpx_get_worker_interface()->end(worker);
}
vpx_free(pbi->tile_worker_data);
vpx_free(pbi->tile_worker_info);
vpx_free(pbi->tile_workers);
if (pbi->num_tile_workers > 0) {
vp10_loop_filter_dealloc(&pbi->lf_row_sync);
}
vpx_free(pbi);
}
static int equal_dimensions(const YV12_BUFFER_CONFIG *a,
const YV12_BUFFER_CONFIG *b) {
return a->y_height == b->y_height && a->y_width == b->y_width &&
a->uv_height == b->uv_height && a->uv_width == b->uv_width;
}
vpx_codec_err_t vp10_copy_reference_dec(VP9Decoder *pbi,
VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
VP9_COMMON *cm = &pbi->common;
/* TODO(jkoleszar): The decoder doesn't have any real knowledge of what the
* encoder is using the frame buffers for. This is just a stub to keep the
* vpxenc --test-decode functionality working, and will be replaced in a
* later commit that adds VP9-specific controls for this functionality.
*/
if (ref_frame_flag == VP9_LAST_FLAG) {
const YV12_BUFFER_CONFIG *const cfg = get_ref_frame(cm, 0);
if (cfg == NULL) {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"No 'last' reference frame");
return VPX_CODEC_ERROR;
}
if (!equal_dimensions(cfg, sd))
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Incorrect buffer dimensions");
else
vp8_yv12_copy_frame(cfg, sd);
} else {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Invalid reference frame");
}
return cm->error.error_code;
}
vpx_codec_err_t vp10_set_reference_dec(VP9_COMMON *cm,
VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
RefBuffer *ref_buf = NULL;
RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
// TODO(jkoleszar): The decoder doesn't have any real knowledge of what the
// encoder is using the frame buffers for. This is just a stub to keep the
// vpxenc --test-decode functionality working, and will be replaced in a
// later commit that adds VP9-specific controls for this functionality.
if (ref_frame_flag == VP9_LAST_FLAG) {
ref_buf = &cm->frame_refs[0];
} else if (ref_frame_flag == VP9_GOLD_FLAG) {
ref_buf = &cm->frame_refs[1];
} else if (ref_frame_flag == VP9_ALT_FLAG) {
ref_buf = &cm->frame_refs[2];
} else {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Invalid reference frame");
return cm->error.error_code;
}
if (!equal_dimensions(ref_buf->buf, sd)) {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Incorrect buffer dimensions");
} else {
int *ref_fb_ptr = &ref_buf->idx;
// Find an empty frame buffer.
const int free_fb = get_free_fb(cm);
if (cm->new_fb_idx == INVALID_IDX)
return VPX_CODEC_MEM_ERROR;
// Decrease ref_count since it will be increased again in
// ref_cnt_fb() below.
--frame_bufs[free_fb].ref_count;
// Manage the reference counters and copy image.
ref_cnt_fb(frame_bufs, ref_fb_ptr, free_fb);
ref_buf->buf = &frame_bufs[*ref_fb_ptr].buf;
vp8_yv12_copy_frame(sd, ref_buf->buf);
}
return cm->error.error_code;
}
/* If any buffer updating is signaled it should be done here. */
static void swap_frame_buffers(VP9Decoder *pbi) {
int ref_index = 0, mask;
VP9_COMMON *const cm = &pbi->common;
BufferPool *const pool = cm->buffer_pool;
RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
lock_buffer_pool(pool);
for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
const int old_idx = cm->ref_frame_map[ref_index];
// Current thread releases the holding of reference frame.
decrease_ref_count(old_idx, frame_bufs, pool);
// Release the reference frame in reference map.
if ((mask & 1) && old_idx >= 0) {
decrease_ref_count(old_idx, frame_bufs, pool);
}
cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index];
++ref_index;
}
// Current thread releases the holding of reference frame.
for (; ref_index < REF_FRAMES && !cm->show_existing_frame; ++ref_index) {
const int old_idx = cm->ref_frame_map[ref_index];
decrease_ref_count(old_idx, frame_bufs, pool);
cm->ref_frame_map[ref_index] = cm->next_ref_frame_map[ref_index];
}
unlock_buffer_pool(pool);
pbi->hold_ref_buf = 0;
cm->frame_to_show = get_frame_new_buffer(cm);
if (!pbi->frame_parallel_decode || !cm->show_frame) {
lock_buffer_pool(pool);
--frame_bufs[cm->new_fb_idx].ref_count;
unlock_buffer_pool(pool);
}
// Invalidate these references until the next frame starts.
for (ref_index = 0; ref_index < 3; ref_index++)
cm->frame_refs[ref_index].idx = -1;
}
int vp10_receive_compressed_data(VP9Decoder *pbi,
size_t size, const uint8_t **psource) {
VP9_COMMON *volatile const cm = &pbi->common;
BufferPool *volatile const pool = cm->buffer_pool;
RefCntBuffer *volatile const frame_bufs = cm->buffer_pool->frame_bufs;
const uint8_t *source = *psource;
int retcode = 0;
cm->error.error_code = VPX_CODEC_OK;
if (size == 0) {
// This is used to signal that we are missing frames.
// We do not know if the missing frame(s) was supposed to update
// any of the reference buffers, but we act conservative and
// mark only the last buffer as corrupted.
//
// TODO(jkoleszar): Error concealment is undefined and non-normative
// at this point, but if it becomes so, [0] may not always be the correct
// thing to do here.
if (cm->frame_refs[0].idx > 0) {
assert(cm->frame_refs[0].buf != NULL);
cm->frame_refs[0].buf->corrupted = 1;
}
}
pbi->ready_for_new_data = 0;
// Check if the previous frame was a frame without any references to it.
// Release frame buffer if not decoding in frame parallel mode.
if (!pbi->frame_parallel_decode && cm->new_fb_idx >= 0
&& frame_bufs[cm->new_fb_idx].ref_count == 0)
pool->release_fb_cb(pool->cb_priv,
&frame_bufs[cm->new_fb_idx].raw_frame_buffer);
// Find a free frame buffer. Return error if can not find any.
cm->new_fb_idx = get_free_fb(cm);
if (cm->new_fb_idx == INVALID_IDX)
return VPX_CODEC_MEM_ERROR;
// Assign a MV array to the frame buffer.
cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx];
pbi->hold_ref_buf = 0;
if (pbi->frame_parallel_decode) {
VPxWorker *const worker = pbi->frame_worker_owner;
vp10_frameworker_lock_stats(worker);
frame_bufs[cm->new_fb_idx].frame_worker_owner = worker;
// Reset decoding progress.
pbi->cur_buf = &frame_bufs[cm->new_fb_idx];
pbi->cur_buf->row = -1;
pbi->cur_buf->col = -1;
vp10_frameworker_unlock_stats(worker);
} else {
pbi->cur_buf = &frame_bufs[cm->new_fb_idx];
}
if (setjmp(cm->error.jmp)) {
const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
int i;
cm->error.setjmp = 0;
pbi->ready_for_new_data = 1;
// Synchronize all threads immediately as a subsequent decode call may
// cause a resize invalidating some allocations.
winterface->sync(&pbi->lf_worker);
for (i = 0; i < pbi->num_tile_workers; ++i) {
winterface->sync(&pbi->tile_workers[i]);
}
lock_buffer_pool(pool);
// Release all the reference buffers if worker thread is holding them.
if (pbi->hold_ref_buf == 1) {
int ref_index = 0, mask;
for (mask = pbi->refresh_frame_flags; mask; mask >>= 1) {
const int old_idx = cm->ref_frame_map[ref_index];
// Current thread releases the holding of reference frame.
decrease_ref_count(old_idx, frame_bufs, pool);
// Release the reference frame in reference map.
if ((mask & 1) && old_idx >= 0) {
decrease_ref_count(old_idx, frame_bufs, pool);
}
++ref_index;
}
// Current thread releases the holding of reference frame.
for (; ref_index < REF_FRAMES && !cm->show_existing_frame; ++ref_index) {
const int old_idx = cm->ref_frame_map[ref_index];
decrease_ref_count(old_idx, frame_bufs, pool);
}
pbi->hold_ref_buf = 0;
}
// Release current frame.
decrease_ref_count(cm->new_fb_idx, frame_bufs, pool);
unlock_buffer_pool(pool);
vpx_clear_system_state();
return -1;
}
cm->error.setjmp = 1;
vp10_decode_frame(pbi, source, source + size, psource);
swap_frame_buffers(pbi);
vpx_clear_system_state();
if (!cm->show_existing_frame) {
cm->last_show_frame = cm->show_frame;
cm->prev_frame = cm->cur_frame;
if (cm->seg.enabled && !pbi->frame_parallel_decode)
vp10_swap_current_and_last_seg_map(cm);
}
// Update progress in frame parallel decode.
if (pbi->frame_parallel_decode) {
// Need to lock the mutex here as another thread may
// be accessing this buffer.
VPxWorker *const worker = pbi->frame_worker_owner;
FrameWorkerData *const frame_worker_data = worker->data1;
vp10_frameworker_lock_stats(worker);
if (cm->show_frame) {
cm->current_video_frame++;
}
frame_worker_data->frame_decoded = 1;
frame_worker_data->frame_context_ready = 1;
vp10_frameworker_signal_stats(worker);
vp10_frameworker_unlock_stats(worker);
} else {
cm->last_width = cm->width;
cm->last_height = cm->height;
if (cm->show_frame) {
cm->current_video_frame++;
}
}
cm->error.setjmp = 0;
return retcode;
}
int vp10_get_raw_frame(VP9Decoder *pbi, YV12_BUFFER_CONFIG *sd,
vp10_ppflags_t *flags) {
VP9_COMMON *const cm = &pbi->common;
int ret = -1;
#if !CONFIG_VP9_POSTPROC
(void)*flags;
#endif
if (pbi->ready_for_new_data == 1)
return ret;
pbi->ready_for_new_data = 1;
/* no raw frame to show!!! */
if (!cm->show_frame)
return ret;
pbi->ready_for_new_data = 1;
#if CONFIG_VP9_POSTPROC
if (!cm->show_existing_frame) {
ret = vp10_post_proc_frame(cm, sd, flags);
} else {
*sd = *cm->frame_to_show;
ret = 0;
}
#else
*sd = *cm->frame_to_show;
ret = 0;
#endif /*!CONFIG_POSTPROC*/
vpx_clear_system_state();
return ret;
}
vpx_codec_err_t vp10_parse_superframe_index(const uint8_t *data,
size_t data_sz,
uint32_t sizes[8], int *count,
vpx_decrypt_cb decrypt_cb,
void *decrypt_state) {
// A chunk ending with a byte matching 0xc0 is an invalid chunk unless
// it is a super frame index. If the last byte of real video compression
// data is 0xc0 the encoder must add a 0 byte. If we have the marker but
// not the associated matching marker byte at the front of the index we have
// an invalid bitstream and need to return an error.
uint8_t marker;
assert(data_sz);
marker = read_marker(decrypt_cb, decrypt_state, data + data_sz - 1);
*count = 0;
if ((marker & 0xe0) == 0xc0) {
const uint32_t frames = (marker & 0x7) + 1;
const uint32_t mag = ((marker >> 3) & 0x3) + 1;
const size_t index_sz = 2 + mag * frames;
// This chunk is marked as having a superframe index but doesn't have
// enough data for it, thus it's an invalid superframe index.
if (data_sz < index_sz)
return VPX_CODEC_CORRUPT_FRAME;
{
const uint8_t marker2 = read_marker(decrypt_cb, decrypt_state,
data + data_sz - index_sz);
// This chunk is marked as having a superframe index but doesn't have
// the matching marker byte at the front of the index therefore it's an
// invalid chunk.
if (marker != marker2)
return VPX_CODEC_CORRUPT_FRAME;
}
{
// Found a valid superframe index.
uint32_t i, j;
const uint8_t *x = &data[data_sz - index_sz + 1];
// Frames has a maximum of 8 and mag has a maximum of 4.
uint8_t clear_buffer[32];
assert(sizeof(clear_buffer) >= frames * mag);
if (decrypt_cb) {
decrypt_cb(decrypt_state, x, clear_buffer, frames * mag);
x = clear_buffer;
}
for (i = 0; i < frames; ++i) {
uint32_t this_sz = 0;
for (j = 0; j < mag; ++j)
this_sz |= (*x++) << (j * 8);
sizes[i] = this_sz;
}
*count = frames;
}
}
return VPX_CODEC_OK;
}

141
vp10/decoder/vp9_decoder.h Normal file
View File

@ -0,0 +1,141 @@
/*
* Copyright (c) 2010 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 VP9_DECODER_VP9_DECODER_H_
#define VP9_DECODER_VP9_DECODER_H_
#include "./vpx_config.h"
#include "vpx/vpx_codec.h"
#include "vpx_dsp/bitreader.h"
#include "vpx_scale/yv12config.h"
#include "vpx_util/vpx_thread.h"
#include "vp10/common/vp9_thread_common.h"
#include "vp10/common/vp9_onyxc_int.h"
#include "vp10/common/vp9_ppflags.h"
#include "vp10/decoder/vp9_dthread.h"
#ifdef __cplusplus
extern "C" {
#endif
// TODO(hkuang): combine this with TileWorkerData.
typedef struct TileData {
VP9_COMMON *cm;
vpx_reader bit_reader;
DECLARE_ALIGNED(16, MACROBLOCKD, xd);
/* dqcoeff are shared by all the planes. So planes must be decoded serially */
DECLARE_ALIGNED(16, tran_low_t, dqcoeff[32 * 32]);
} TileData;
typedef struct TileWorkerData {
struct VP9Decoder *pbi;
vpx_reader bit_reader;
FRAME_COUNTS counts;
DECLARE_ALIGNED(16, MACROBLOCKD, xd);
/* dqcoeff are shared by all the planes. So planes must be decoded serially */
DECLARE_ALIGNED(16, tran_low_t, dqcoeff[32 * 32]);
struct vpx_internal_error_info error_info;
} TileWorkerData;
typedef struct VP9Decoder {
DECLARE_ALIGNED(16, MACROBLOCKD, mb);
DECLARE_ALIGNED(16, VP9_COMMON, common);
int ready_for_new_data;
int refresh_frame_flags;
int frame_parallel_decode; // frame-based threading.
// TODO(hkuang): Combine this with cur_buf in macroblockd as they are
// the same.
RefCntBuffer *cur_buf; // Current decoding frame buffer.
VPxWorker *frame_worker_owner; // frame_worker that owns this pbi.
VPxWorker lf_worker;
VPxWorker *tile_workers;
TileWorkerData *tile_worker_data;
TileInfo *tile_worker_info;
int num_tile_workers;
TileData *tile_data;
int total_tiles;
VP9LfSync lf_row_sync;
vpx_decrypt_cb decrypt_cb;
void *decrypt_state;
int max_threads;
int inv_tile_order;
int need_resync; // wait for key/intra-only frame.
int hold_ref_buf; // hold the reference buffer.
} VP9Decoder;
int vp10_receive_compressed_data(struct VP9Decoder *pbi,
size_t size, const uint8_t **dest);
int vp10_get_raw_frame(struct VP9Decoder *pbi, YV12_BUFFER_CONFIG *sd,
vp10_ppflags_t *flags);
vpx_codec_err_t vp10_copy_reference_dec(struct VP9Decoder *pbi,
VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd);
vpx_codec_err_t vp10_set_reference_dec(VP9_COMMON *cm,
VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd);
static INLINE uint8_t read_marker(vpx_decrypt_cb decrypt_cb,
void *decrypt_state,
const uint8_t *data) {
if (decrypt_cb) {
uint8_t marker;
decrypt_cb(decrypt_state, data, &marker, 1);
return marker;
}
return *data;
}
// This function is exposed for use in tests, as well as the inlined function
// "read_marker".
vpx_codec_err_t vp10_parse_superframe_index(const uint8_t *data,
size_t data_sz,
uint32_t sizes[8], int *count,
vpx_decrypt_cb decrypt_cb,
void *decrypt_state);
struct VP9Decoder *vp10_decoder_create(BufferPool *const pool);
void vp10_decoder_remove(struct VP9Decoder *pbi);
static INLINE void decrease_ref_count(int idx, RefCntBuffer *const frame_bufs,
BufferPool *const pool) {
if (idx >= 0) {
--frame_bufs[idx].ref_count;
// A worker may only get a free framebuffer index when calling get_free_fb.
// But the private buffer is not set up until finish decoding header.
// So any error happens during decoding header, the frame_bufs will not
// have valid priv buffer.
if (frame_bufs[idx].ref_count == 0 &&
frame_bufs[idx].raw_frame_buffer.priv) {
pool->release_fb_cb(pool->cb_priv, &frame_bufs[idx].raw_frame_buffer);
}
}
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_DECODER_VP9_DECODER_H_

269
vp10/decoder/vp9_detokenize.c Normal file
View File

@ -0,0 +1,269 @@
/*
* Copyright (c) 2010 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 "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
#include "vp10/common/vp9_blockd.h"
#include "vp10/common/vp9_common.h"
#include "vp10/common/vp9_entropy.h"
#if CONFIG_COEFFICIENT_RANGE_CHECKING
#include "vp10/common/vp9_idct.h"
#endif
#include "vp10/decoder/vp9_detokenize.h"
#define EOB_CONTEXT_NODE 0
#define ZERO_CONTEXT_NODE 1
#define ONE_CONTEXT_NODE 2
#define LOW_VAL_CONTEXT_NODE 0
#define TWO_CONTEXT_NODE 1
#define THREE_CONTEXT_NODE 2
#define HIGH_LOW_CONTEXT_NODE 3
#define CAT_ONE_CONTEXT_NODE 4
#define CAT_THREEFOUR_CONTEXT_NODE 5
#define CAT_THREE_CONTEXT_NODE 6
#define CAT_FIVE_CONTEXT_NODE 7
#define INCREMENT_COUNT(token) \
do { \
if (counts) \
++coef_counts[band][ctx][token]; \
} while (0)
static INLINE int read_coeff(const vpx_prob *probs, int n, vpx_reader *r) {
int i, val = 0;
for (i = 0; i < n; ++i)
val = (val << 1) | vpx_read(r, probs[i]);
return val;
}
static int decode_coefs(const MACROBLOCKD *xd,
PLANE_TYPE type,
tran_low_t *dqcoeff, TX_SIZE tx_size, const int16_t *dq,
int ctx, const int16_t *scan, const int16_t *nb,
vpx_reader *r) {
FRAME_COUNTS *counts = xd->counts;
const int max_eob = 16 << (tx_size << 1);
const FRAME_CONTEXT *const fc = xd->fc;
const int ref = is_inter_block(&xd->mi[0]->mbmi);
int band, c = 0;
const vpx_prob (*coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] =
fc->coef_probs[tx_size][type][ref];
const vpx_prob *prob;
unsigned int (*coef_counts)[COEFF_CONTEXTS][UNCONSTRAINED_NODES + 1];
unsigned int (*eob_branch_count)[COEFF_CONTEXTS];
uint8_t token_cache[32 * 32];
const uint8_t *band_translate = get_band_translate(tx_size);
const int dq_shift = (tx_size == TX_32X32);
int v, token;
int16_t dqv = dq[0];
const uint8_t *cat1_prob;
const uint8_t *cat2_prob;
const uint8_t *cat3_prob;
const uint8_t *cat4_prob;
const uint8_t *cat5_prob;
const uint8_t *cat6_prob;
if (counts) {
coef_counts = counts->coef[tx_size][type][ref];
eob_branch_count = counts->eob_branch[tx_size][type][ref];
}
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->bd > VPX_BITS_8) {
if (xd->bd == VPX_BITS_10) {
cat1_prob = vp10_cat1_prob_high10;
cat2_prob = vp10_cat2_prob_high10;
cat3_prob = vp10_cat3_prob_high10;
cat4_prob = vp10_cat4_prob_high10;
cat5_prob = vp10_cat5_prob_high10;
cat6_prob = vp10_cat6_prob_high10;
} else {
cat1_prob = vp10_cat1_prob_high12;
cat2_prob = vp10_cat2_prob_high12;
cat3_prob = vp10_cat3_prob_high12;
cat4_prob = vp10_cat4_prob_high12;
cat5_prob = vp10_cat5_prob_high12;
cat6_prob = vp10_cat6_prob_high12;
}
} else {
cat1_prob = vp10_cat1_prob;
cat2_prob = vp10_cat2_prob;
cat3_prob = vp10_cat3_prob;
cat4_prob = vp10_cat4_prob;
cat5_prob = vp10_cat5_prob;
cat6_prob = vp10_cat6_prob;
}
#else
cat1_prob = vp10_cat1_prob;
cat2_prob = vp10_cat2_prob;
cat3_prob = vp10_cat3_prob;
cat4_prob = vp10_cat4_prob;
cat5_prob = vp10_cat5_prob;
cat6_prob = vp10_cat6_prob;
#endif
while (c < max_eob) {
int val = -1;
band = *band_translate++;
prob = coef_probs[band][ctx];
if (counts)
++eob_branch_count[band][ctx];
if (!vpx_read(r, prob[EOB_CONTEXT_NODE])) {
INCREMENT_COUNT(EOB_MODEL_TOKEN);
break;
}
while (!vpx_read(r, prob[ZERO_CONTEXT_NODE])) {
INCREMENT_COUNT(ZERO_TOKEN);
dqv = dq[1];
token_cache[scan[c]] = 0;
++c;
if (c >= max_eob)
return c; // zero tokens at the end (no eob token)
ctx = get_coef_context(nb, token_cache, c);
band = *band_translate++;
prob = coef_probs[band][ctx];
}
if (!vpx_read(r, prob[ONE_CONTEXT_NODE])) {
INCREMENT_COUNT(ONE_TOKEN);
token = ONE_TOKEN;
val = 1;
} else {
INCREMENT_COUNT(TWO_TOKEN);
token = vpx_read_tree(r, vp10_coef_con_tree,
vp10_pareto8_full[prob[PIVOT_NODE] - 1]);
switch (token) {
case TWO_TOKEN:
case THREE_TOKEN:
case FOUR_TOKEN:
val = token;
break;
case CATEGORY1_TOKEN:
val = CAT1_MIN_VAL + read_coeff(cat1_prob, 1, r);
break;
case CATEGORY2_TOKEN:
val = CAT2_MIN_VAL + read_coeff(cat2_prob, 2, r);
break;
case CATEGORY3_TOKEN:
val = CAT3_MIN_VAL + read_coeff(cat3_prob, 3, r);
break;
case CATEGORY4_TOKEN:
val = CAT4_MIN_VAL + read_coeff(cat4_prob, 4, r);
break;
case CATEGORY5_TOKEN:
val = CAT5_MIN_VAL + read_coeff(cat5_prob, 5, r);
break;
case CATEGORY6_TOKEN:
#if CONFIG_VP9_HIGHBITDEPTH
switch (xd->bd) {
case VPX_BITS_8:
val = CAT6_MIN_VAL + read_coeff(cat6_prob, 14, r);
break;
case VPX_BITS_10:
val = CAT6_MIN_VAL + read_coeff(cat6_prob, 16, r);
break;
case VPX_BITS_12:
val = CAT6_MIN_VAL + read_coeff(cat6_prob, 18, r);
break;
default:
assert(0);
return -1;
}
#else
val = CAT6_MIN_VAL + read_coeff(cat6_prob, 14, r);
#endif
break;
}
}
v = (val * dqv) >> dq_shift;
#if CONFIG_COEFFICIENT_RANGE_CHECKING
#if CONFIG_VP9_HIGHBITDEPTH
dqcoeff[scan[c]] = highbd_check_range((vpx_read_bit(r) ? -v : v),
xd->bd);
#else
dqcoeff[scan[c]] = check_range(vpx_read_bit(r) ? -v : v);
#endif // CONFIG_VP9_HIGHBITDEPTH
#else
dqcoeff[scan[c]] = vpx_read_bit(r) ? -v : v;
#endif // CONFIG_COEFFICIENT_RANGE_CHECKING
token_cache[scan[c]] = vp10_pt_energy_class[token];
++c;
ctx = get_coef_context(nb, token_cache, c);
dqv = dq[1];
}
return c;
}
// TODO(slavarnway): Decode version of vp10_set_context. Modify vp10_set_context
// after testing is complete, then delete this version.
static
void dec_set_contexts(const MACROBLOCKD *xd, struct macroblockd_plane *pd,
TX_SIZE tx_size, int has_eob,
int aoff, int loff) {
ENTROPY_CONTEXT *const a = pd->above_context + aoff;
ENTROPY_CONTEXT *const l = pd->left_context + loff;
const int tx_size_in_blocks = 1 << tx_size;
// above
if (has_eob && xd->mb_to_right_edge < 0) {
int i;
const int blocks_wide = pd->n4_w +
(xd->mb_to_right_edge >> (5 + pd->subsampling_x));
int above_contexts = tx_size_in_blocks;
if (above_contexts + aoff > blocks_wide)
above_contexts = blocks_wide - aoff;
for (i = 0; i < above_contexts; ++i)
a[i] = has_eob;
for (i = above_contexts; i < tx_size_in_blocks; ++i)
a[i] = 0;
} else {
memset(a, has_eob, sizeof(ENTROPY_CONTEXT) * tx_size_in_blocks);
}
// left
if (has_eob && xd->mb_to_bottom_edge < 0) {
int i;
const int blocks_high = pd->n4_h +
(xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
int left_contexts = tx_size_in_blocks;
if (left_contexts + loff > blocks_high)
left_contexts = blocks_high - loff;
for (i = 0; i < left_contexts; ++i)
l[i] = has_eob;
for (i = left_contexts; i < tx_size_in_blocks; ++i)
l[i] = 0;
} else {
memset(l, has_eob, sizeof(ENTROPY_CONTEXT) * tx_size_in_blocks);
}
}
int vp10_decode_block_tokens(MACROBLOCKD *xd,
int plane, const scan_order *sc,
int x, int y,
TX_SIZE tx_size, vpx_reader *r,
int seg_id) {
struct macroblockd_plane *const pd = &xd->plane[plane];
const int16_t *const dequant = pd->seg_dequant[seg_id];
const int ctx = get_entropy_context(tx_size, pd->above_context + x,
pd->left_context + y);
const int eob = decode_coefs(xd, pd->plane_type,
pd->dqcoeff, tx_size,
dequant, ctx, sc->scan, sc->neighbors, r);
dec_set_contexts(xd, pd, tx_size, eob > 0, x, y);
return eob;
}

33
vp10/decoder/vp9_detokenize.h Normal file
View File

@ -0,0 +1,33 @@
/*
* Copyright (c) 2010 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 VP9_DECODER_VP9_DETOKENIZE_H_
#define VP9_DECODER_VP9_DETOKENIZE_H_
#include "vpx_dsp/bitreader.h"
#include "vp10/decoder/vp9_decoder.h"
#include "vp10/common/vp9_scan.h"
#ifdef __cplusplus
extern "C" {
#endif
int vp10_decode_block_tokens(MACROBLOCKD *xd,
int plane, const scan_order *sc,
int x, int y,
TX_SIZE tx_size, vpx_reader *r,
int seg_id);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_DECODER_VP9_DETOKENIZE_H_

76
vp10/decoder/vp9_dsubexp.c Normal file
View File

@ -0,0 +1,76 @@
/*
Copyright (c) 2010 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 <assert.h>
#include "vp10/common/vp9_entropy.h"
#include "vp10/decoder/vp9_dsubexp.h"
static int inv_recenter_nonneg(int v, int m) {
if (v > 2 * m)
return v;
return (v & 1) ? m - ((v + 1) >> 1) : m + (v >> 1);
}
static int decode_uniform(vpx_reader *r) {
const int l = 8;
const int m = (1 << l) - 191;
const int v = vpx_read_literal(r, l - 1);
return v < m ? v : (v << 1) - m + vpx_read_bit(r);
}
static int inv_remap_prob(int v, int m) {
static int inv_map_table[MAX_PROB] = {
7, 20, 33, 46, 59, 72, 85, 98, 111, 124, 137, 150, 163, 176, 189,
202, 215, 228, 241, 254, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11,
12, 13, 14, 15, 16, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27,
28, 29, 30, 31, 32, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43,
44, 45, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 60,
61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 73, 74, 75, 76,
77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92,
93, 94, 95, 96, 97, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108,
109, 110, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 125,
126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 138, 139, 140, 141,
142, 143, 144, 145, 146, 147, 148, 149, 151, 152, 153, 154, 155, 156, 157,
158, 159, 160, 161, 162, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173,
174, 175, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 190,
191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 204, 205, 206,
207, 208, 209, 210, 211, 212, 213, 214, 216, 217, 218, 219, 220, 221, 222,
223, 224, 225, 226, 227, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238,
239, 240, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 253
};
assert(v < (int)(sizeof(inv_map_table) / sizeof(inv_map_table[0])));
v = inv_map_table[v];
m--;
if ((m << 1) <= MAX_PROB) {
return 1 + inv_recenter_nonneg(v, m);
} else {
return MAX_PROB - inv_recenter_nonneg(v, MAX_PROB - 1 - m);
}
}
static int decode_term_subexp(vpx_reader *r) {
if (!vpx_read_bit(r))
return vpx_read_literal(r, 4);
if (!vpx_read_bit(r))
return vpx_read_literal(r, 4) + 16;
if (!vpx_read_bit(r))
return vpx_read_literal(r, 5) + 32;
return decode_uniform(r) + 64;
}
void vp10_diff_update_prob(vpx_reader *r, vpx_prob* p) {
if (vpx_read(r, DIFF_UPDATE_PROB)) {
const int delp = decode_term_subexp(r);
*p = (vpx_prob)inv_remap_prob(delp, *p);
}
}

27
vp10/decoder/vp9_dsubexp.h Normal file
View File

@ -0,0 +1,27 @@
/*
* Copyright (c) 2013 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 VP9_DECODER_VP9_DSUBEXP_H_
#define VP9_DECODER_VP9_DSUBEXP_H_
#include "vpx_dsp/bitreader.h"
#ifdef __cplusplus
extern "C" {
#endif
void vp10_diff_update_prob(vpx_reader *r, vpx_prob* p);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_DECODER_VP9_DSUBEXP_H_

189
vp10/decoder/vp9_dthread.c Normal file
View File

@ -0,0 +1,189 @@
/*
* Copyright (c) 2014 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 "./vpx_config.h"
#include "vpx_mem/vpx_mem.h"
#include "vp10/common/vp9_reconinter.h"
#include "vp10/decoder/vp9_dthread.h"
#include "vp10/decoder/vp9_decoder.h"
// #define DEBUG_THREAD
// TODO(hkuang): Clean up all the #ifdef in this file.
void vp10_frameworker_lock_stats(VPxWorker *const worker) {
#if CONFIG_MULTITHREAD
FrameWorkerData *const worker_data = worker->data1;
pthread_mutex_lock(&worker_data->stats_mutex);
#else
(void)worker;
#endif
}
void vp10_frameworker_unlock_stats(VPxWorker *const worker) {
#if CONFIG_MULTITHREAD
FrameWorkerData *const worker_data = worker->data1;
pthread_mutex_unlock(&worker_data->stats_mutex);
#else
(void)worker;
#endif
}
void vp10_frameworker_signal_stats(VPxWorker *const worker) {
#if CONFIG_MULTITHREAD
FrameWorkerData *const worker_data = worker->data1;
// TODO(hkuang): Fix the pthread_cond_broadcast in windows wrapper.
#if defined(_WIN32) && !HAVE_PTHREAD_H
pthread_cond_signal(&worker_data->stats_cond);
#else
pthread_cond_broadcast(&worker_data->stats_cond);
#endif
#else
(void)worker;
#endif
}
// This macro prevents thread_sanitizer from reporting known concurrent writes.
#if defined(__has_feature)
#if __has_feature(thread_sanitizer)
#define BUILDING_WITH_TSAN
#endif
#endif
// TODO(hkuang): Remove worker parameter as it is only used in debug code.
void vp10_frameworker_wait(VPxWorker *const worker, RefCntBuffer *const ref_buf,
int row) {
#if CONFIG_MULTITHREAD
if (!ref_buf)
return;
#ifndef BUILDING_WITH_TSAN
// The following line of code will get harmless tsan error but it is the key
// to get best performance.
if (ref_buf->row >= row && ref_buf->buf.corrupted != 1) return;
#endif
{
// Find the worker thread that owns the reference frame. If the reference
// frame has been fully decoded, it may not have owner.
VPxWorker *const ref_worker = ref_buf->frame_worker_owner;
FrameWorkerData *const ref_worker_data =
(FrameWorkerData *)ref_worker->data1;
const VP9Decoder *const pbi = ref_worker_data->pbi;
#ifdef DEBUG_THREAD
{
FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
printf("%d %p worker is waiting for %d %p worker (%d) ref %d \r\n",
worker_data->worker_id, worker, ref_worker_data->worker_id,
ref_buf->frame_worker_owner, row, ref_buf->row);
}
#endif
vp10_frameworker_lock_stats(ref_worker);
while (ref_buf->row < row && pbi->cur_buf == ref_buf &&
ref_buf->buf.corrupted != 1) {
pthread_cond_wait(&ref_worker_data->stats_cond,
&ref_worker_data->stats_mutex);
}
if (ref_buf->buf.corrupted == 1) {
FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
vp10_frameworker_unlock_stats(ref_worker);
vpx_internal_error(&worker_data->pbi->common.error,
VPX_CODEC_CORRUPT_FRAME,
"Worker %p failed to decode frame", worker);
}
vp10_frameworker_unlock_stats(ref_worker);
}
#else
(void)worker;
(void)ref_buf;
(void)row;
(void)ref_buf;
#endif // CONFIG_MULTITHREAD
}
void vp10_frameworker_broadcast(RefCntBuffer *const buf, int row) {
#if CONFIG_MULTITHREAD
VPxWorker *worker = buf->frame_worker_owner;
#ifdef DEBUG_THREAD
{
FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
printf("%d %p worker decode to (%d) \r\n", worker_data->worker_id,
buf->frame_worker_owner, row);
}
#endif
vp10_frameworker_lock_stats(worker);
buf->row = row;
vp10_frameworker_signal_stats(worker);
vp10_frameworker_unlock_stats(worker);
#else
(void)buf;
(void)row;
#endif // CONFIG_MULTITHREAD
}
void vp10_frameworker_copy_context(VPxWorker *const dst_worker,
VPxWorker *const src_worker) {
#if CONFIG_MULTITHREAD
FrameWorkerData *const src_worker_data = (FrameWorkerData *)src_worker->data1;
FrameWorkerData *const dst_worker_data = (FrameWorkerData *)dst_worker->data1;
VP9_COMMON *const src_cm = &src_worker_data->pbi->common;
VP9_COMMON *const dst_cm = &dst_worker_data->pbi->common;
int i;
// Wait until source frame's context is ready.
vp10_frameworker_lock_stats(src_worker);
while (!src_worker_data->frame_context_ready) {
pthread_cond_wait(&src_worker_data->stats_cond,
&src_worker_data->stats_mutex);
}
dst_cm->last_frame_seg_map = src_cm->seg.enabled ?
src_cm->current_frame_seg_map : src_cm->last_frame_seg_map;
dst_worker_data->pbi->need_resync = src_worker_data->pbi->need_resync;
vp10_frameworker_unlock_stats(src_worker);
dst_cm->bit_depth = src_cm->bit_depth;
#if CONFIG_VP9_HIGHBITDEPTH
dst_cm->use_highbitdepth = src_cm->use_highbitdepth;
#endif
dst_cm->prev_frame = src_cm->show_existing_frame ?
src_cm->prev_frame : src_cm->cur_frame;
dst_cm->last_width = !src_cm->show_existing_frame ?
src_cm->width : src_cm->last_width;
dst_cm->last_height = !src_cm->show_existing_frame ?
src_cm->height : src_cm->last_height;
dst_cm->subsampling_x = src_cm->subsampling_x;
dst_cm->subsampling_y = src_cm->subsampling_y;
dst_cm->frame_type = src_cm->frame_type;
dst_cm->last_show_frame = !src_cm->show_existing_frame ?
src_cm->show_frame : src_cm->last_show_frame;
for (i = 0; i < REF_FRAMES; ++i)
dst_cm->ref_frame_map[i] = src_cm->next_ref_frame_map[i];
memcpy(dst_cm->lf_info.lfthr, src_cm->lf_info.lfthr,
(MAX_LOOP_FILTER + 1) * sizeof(loop_filter_thresh));
dst_cm->lf.last_sharpness_level = src_cm->lf.sharpness_level;
dst_cm->lf.filter_level = src_cm->lf.filter_level;
memcpy(dst_cm->lf.ref_deltas, src_cm->lf.ref_deltas, MAX_REF_LF_DELTAS);
memcpy(dst_cm->lf.mode_deltas, src_cm->lf.mode_deltas, MAX_MODE_LF_DELTAS);
dst_cm->seg = src_cm->seg;
memcpy(dst_cm->frame_contexts, src_cm->frame_contexts,
FRAME_CONTEXTS * sizeof(dst_cm->frame_contexts[0]));
#else
(void) dst_worker;
(void) src_worker;
#endif // CONFIG_MULTITHREAD
}

66
vp10/decoder/vp9_dthread.h Normal file
View File

@ -0,0 +1,66 @@
/*
* Copyright (c) 2014 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 VP9_DECODER_VP9_DTHREAD_H_
#define VP9_DECODER_VP9_DTHREAD_H_
#include "./vpx_config.h"
#include "vpx_util/vpx_thread.h"
#include "vpx/internal/vpx_codec_internal.h"
struct VP9Common;
struct VP9Decoder;
// WorkerData for the FrameWorker thread. It contains all the information of
// the worker and decode structures for decoding a frame.
typedef struct FrameWorkerData {
struct VP9Decoder *pbi;
const uint8_t *data;
const uint8_t *data_end;
size_t data_size;
void *user_priv;
int result;
int worker_id;
int received_frame;
// scratch_buffer is used in frame parallel mode only.
// It is used to make a copy of the compressed data.
uint8_t *scratch_buffer;
size_t scratch_buffer_size;
#if CONFIG_MULTITHREAD
pthread_mutex_t stats_mutex;
pthread_cond_t stats_cond;
#endif
int frame_context_ready; // Current frame's context is ready to read.
int frame_decoded; // Finished decoding current frame.
} FrameWorkerData;
void vp10_frameworker_lock_stats(VPxWorker *const worker);
void vp10_frameworker_unlock_stats(VPxWorker *const worker);
void vp10_frameworker_signal_stats(VPxWorker *const worker);
// Wait until ref_buf has been decoded to row in real pixel unit.
// Note: worker may already finish decoding ref_buf and release it in order to
// start decoding next frame. So need to check whether worker is still decoding
// ref_buf.
void vp10_frameworker_wait(VPxWorker *const worker, RefCntBuffer *const ref_buf,
int row);
// FrameWorker broadcasts its decoding progress so other workers that are
// waiting on it can resume decoding.
void vp10_frameworker_broadcast(RefCntBuffer *const buf, int row);
// Copy necessary decoding context from src worker to dst worker.
void vp10_frameworker_copy_context(VPxWorker *const dst_worker,
VPxWorker *const src_worker);
#endif // VP9_DECODER_VP9_DTHREAD_H_

160
vp10/encoder/arm/neon/vp9_avg_neon.c Normal file
View File

@ -0,0 +1,160 @@
/*
* 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 <arm_neon.h>
#include <assert.h>
#include "./vp10_rtcd.h"
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
static INLINE unsigned int horizontal_add_u16x8(const uint16x8_t v_16x8) {
const uint32x4_t a = vpaddlq_u16(v_16x8);
const uint64x2_t b = vpaddlq_u32(a);
const uint32x2_t c = vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
vreinterpret_u32_u64(vget_high_u64(b)));
return vget_lane_u32(c, 0);
}
unsigned int vp10_avg_8x8_neon(const uint8_t *s, int p) {
uint8x8_t v_s0 = vld1_u8(s);
const uint8x8_t v_s1 = vld1_u8(s + p);
uint16x8_t v_sum = vaddl_u8(v_s0, v_s1);
v_s0 = vld1_u8(s + 2 * p);
v_sum = vaddw_u8(v_sum, v_s0);
v_s0 = vld1_u8(s + 3 * p);
v_sum = vaddw_u8(v_sum, v_s0);
v_s0 = vld1_u8(s + 4 * p);
v_sum = vaddw_u8(v_sum, v_s0);
v_s0 = vld1_u8(s + 5 * p);
v_sum = vaddw_u8(v_sum, v_s0);
v_s0 = vld1_u8(s + 6 * p);
v_sum = vaddw_u8(v_sum, v_s0);
v_s0 = vld1_u8(s + 7 * p);
v_sum = vaddw_u8(v_sum, v_s0);
return (horizontal_add_u16x8(v_sum) + 32) >> 6;
}
void vp10_int_pro_row_neon(int16_t hbuf[16], uint8_t const *ref,
const int ref_stride, const int height) {
int i;
uint16x8_t vec_sum_lo = vdupq_n_u16(0);
uint16x8_t vec_sum_hi = vdupq_n_u16(0);
const int shift_factor = ((height >> 5) + 3) * -1;
const int16x8_t vec_shift = vdupq_n_s16(shift_factor);
for (i = 0; i < height; i += 8) {
const uint8x16_t vec_row1 = vld1q_u8(ref);
const uint8x16_t vec_row2 = vld1q_u8(ref + ref_stride);
const uint8x16_t vec_row3 = vld1q_u8(ref + ref_stride * 2);
const uint8x16_t vec_row4 = vld1q_u8(ref + ref_stride * 3);
const uint8x16_t vec_row5 = vld1q_u8(ref + ref_stride * 4);
const uint8x16_t vec_row6 = vld1q_u8(ref + ref_stride * 5);
const uint8x16_t vec_row7 = vld1q_u8(ref + ref_stride * 6);
const uint8x16_t vec_row8 = vld1q_u8(ref + ref_stride * 7);
vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row1));
vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row1));
vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row2));
vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row2));
vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row3));
vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row3));
vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row4));
vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row4));
vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row5));
vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row5));
vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row6));
vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row6));
vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row7));
vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row7));
vec_sum_lo = vaddw_u8(vec_sum_lo, vget_low_u8(vec_row8));
vec_sum_hi = vaddw_u8(vec_sum_hi, vget_high_u8(vec_row8));
ref += ref_stride * 8;
}
vec_sum_lo = vshlq_u16(vec_sum_lo, vec_shift);
vec_sum_hi = vshlq_u16(vec_sum_hi, vec_shift);
vst1q_s16(hbuf, vreinterpretq_s16_u16(vec_sum_lo));
hbuf += 8;
vst1q_s16(hbuf, vreinterpretq_s16_u16(vec_sum_hi));
}
int16_t vp10_int_pro_col_neon(uint8_t const *ref, const int width) {
int i;
uint16x8_t vec_sum = vdupq_n_u16(0);
for (i = 0; i < width; i += 16) {
const uint8x16_t vec_row = vld1q_u8(ref);
vec_sum = vaddw_u8(vec_sum, vget_low_u8(vec_row));
vec_sum = vaddw_u8(vec_sum, vget_high_u8(vec_row));
ref += 16;
}
return horizontal_add_u16x8(vec_sum);
}
// ref, src = [0, 510] - max diff = 16-bits
// bwl = {2, 3, 4}, width = {16, 32, 64}
int vp10_vector_var_neon(int16_t const *ref, int16_t const *src, const int bwl) {
int width = 4 << bwl;
int32x4_t sse = vdupq_n_s32(0);
int16x8_t total = vdupq_n_s16(0);
assert(width >= 8);
assert((width % 8) == 0);
do {
const int16x8_t r = vld1q_s16(ref);
const int16x8_t s = vld1q_s16(src);
const int16x8_t diff = vsubq_s16(r, s); // [-510, 510], 10 bits.
const int16x4_t diff_lo = vget_low_s16(diff);
const int16x4_t diff_hi = vget_high_s16(diff);
sse = vmlal_s16(sse, diff_lo, diff_lo); // dynamic range 26 bits.
sse = vmlal_s16(sse, diff_hi, diff_hi);
total = vaddq_s16(total, diff); // dynamic range 16 bits.
ref += 8;
src += 8;
width -= 8;
} while (width != 0);
{
// Note: 'total''s pairwise addition could be implemented similarly to
// horizontal_add_u16x8(), but one less vpaddl with 'total' when paired
// with the summation of 'sse' performed better on a Cortex-A15.
const int32x4_t t0 = vpaddlq_s16(total); // cascading summation of 'total'
const int32x2_t t1 = vadd_s32(vget_low_s32(t0), vget_high_s32(t0));
const int32x2_t t2 = vpadd_s32(t1, t1);
const int t = vget_lane_s32(t2, 0);
const int64x2_t s0 = vpaddlq_s32(sse); // cascading summation of 'sse'.
const int32x2_t s1 = vadd_s32(vreinterpret_s32_s64(vget_low_s64(s0)),
vreinterpret_s32_s64(vget_high_s64(s0)));
const int s = vget_lane_s32(s1, 0);
const int shift_factor = bwl + 2;
return s - ((t * t) >> shift_factor);
}
}

36
vp10/encoder/arm/neon/vp9_dct_neon.c Normal file
View File

@ -0,0 +1,36 @@
/*
* Copyright (c) 2014 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 <arm_neon.h>
#include "./vp10_rtcd.h"
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
#include "vp10/common/vp9_blockd.h"
#include "vpx_dsp/txfm_common.h"
void vp10_fdct8x8_quant_neon(const int16_t *input, int stride,
int16_t* coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t* zbin_ptr,
const int16_t* round_ptr, const int16_t* quant_ptr,
const int16_t* quant_shift_ptr,
int16_t* qcoeff_ptr, int16_t* dqcoeff_ptr,
const int16_t* dequant_ptr, uint16_t* eob_ptr,
const int16_t* scan_ptr,
const int16_t* iscan_ptr) {
int16_t temp_buffer[64];
(void)coeff_ptr;
vpx_fdct8x8_neon(input, temp_buffer, stride);
vp10_quantize_fp_neon(temp_buffer, n_coeffs, skip_block, zbin_ptr, round_ptr,
quant_ptr, quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr,
dequant_ptr, eob_ptr, scan_ptr, iscan_ptr);
}

41
vp10/encoder/arm/neon/vp9_error_neon.c Normal file
View File

@ -0,0 +1,41 @@
/*
* 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 <arm_neon.h>
#include <assert.h>
#include "./vp10_rtcd.h"
int64_t vp10_block_error_fp_neon(const int16_t *coeff, const int16_t *dqcoeff,
int block_size) {
int64x2_t error = vdupq_n_s64(0);
assert(block_size >= 8);
assert((block_size % 8) == 0);
do {
const int16x8_t c = vld1q_s16(coeff);
const int16x8_t d = vld1q_s16(dqcoeff);
const int16x8_t diff = vsubq_s16(c, d);
const int16x4_t diff_lo = vget_low_s16(diff);
const int16x4_t diff_hi = vget_high_s16(diff);
// diff is 15-bits, the squares 30, so we can store 2 in 31-bits before
// accumulating them in 64-bits.
const int32x4_t err0 = vmull_s16(diff_lo, diff_lo);
const int32x4_t err1 = vmlal_s16(err0, diff_hi, diff_hi);
const int64x2_t err2 = vaddl_s32(vget_low_s32(err1), vget_high_s32(err1));
error = vaddq_s64(error, err2);
coeff += 8;
dqcoeff += 8;
block_size -= 8;
} while (block_size != 0);
return vgetq_lane_s64(error, 0) + vgetq_lane_s64(error, 1);
}

118
vp10/encoder/arm/neon/vp9_quantize_neon.c Normal file
View File

@ -0,0 +1,118 @@
/*
* Copyright (c) 2014 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 <arm_neon.h>
#include <math.h>
#include "vpx_mem/vpx_mem.h"
#include "vp10/common/vp9_quant_common.h"
#include "vp10/common/vp9_seg_common.h"
#include "vp10/encoder/vp9_encoder.h"
#include "vp10/encoder/vp9_quantize.h"
#include "vp10/encoder/vp9_rd.h"
void vp10_quantize_fp_neon(const int16_t *coeff_ptr, intptr_t count,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr,
int16_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
// TODO(jingning) Decide the need of these arguments after the
// quantization process is completed.
(void)zbin_ptr;
(void)quant_shift_ptr;
(void)scan;
if (!skip_block) {
// Quantization pass: All coefficients with index >= zero_flag are
// skippable. Note: zero_flag can be zero.
int i;
const int16x8_t v_zero = vdupq_n_s16(0);
const int16x8_t v_one = vdupq_n_s16(1);
int16x8_t v_eobmax_76543210 = vdupq_n_s16(-1);
int16x8_t v_round = vmovq_n_s16(round_ptr[1]);
int16x8_t v_quant = vmovq_n_s16(quant_ptr[1]);
int16x8_t v_dequant = vmovq_n_s16(dequant_ptr[1]);
// adjust for dc
v_round = vsetq_lane_s16(round_ptr[0], v_round, 0);
v_quant = vsetq_lane_s16(quant_ptr[0], v_quant, 0);
v_dequant = vsetq_lane_s16(dequant_ptr[0], v_dequant, 0);
// process dc and the first seven ac coeffs
{
const int16x8_t v_iscan = vld1q_s16(&iscan[0]);
const int16x8_t v_coeff = vld1q_s16(&coeff_ptr[0]);
const int16x8_t v_coeff_sign = vshrq_n_s16(v_coeff, 15);
const int16x8_t v_tmp = vabaq_s16(v_round, v_coeff, v_zero);
const int32x4_t v_tmp_lo = vmull_s16(vget_low_s16(v_tmp),
vget_low_s16(v_quant));
const int32x4_t v_tmp_hi = vmull_s16(vget_high_s16(v_tmp),
vget_high_s16(v_quant));
const int16x8_t v_tmp2 = vcombine_s16(vshrn_n_s32(v_tmp_lo, 16),
vshrn_n_s32(v_tmp_hi, 16));
const uint16x8_t v_nz_mask = vceqq_s16(v_tmp2, v_zero);
const int16x8_t v_iscan_plus1 = vaddq_s16(v_iscan, v_one);
const int16x8_t v_nz_iscan = vbslq_s16(v_nz_mask, v_zero, v_iscan_plus1);
const int16x8_t v_qcoeff_a = veorq_s16(v_tmp2, v_coeff_sign);
const int16x8_t v_qcoeff = vsubq_s16(v_qcoeff_a, v_coeff_sign);
const int16x8_t v_dqcoeff = vmulq_s16(v_qcoeff, v_dequant);
v_eobmax_76543210 = vmaxq_s16(v_eobmax_76543210, v_nz_iscan);
vst1q_s16(&qcoeff_ptr[0], v_qcoeff);
vst1q_s16(&dqcoeff_ptr[0], v_dqcoeff);
v_round = vmovq_n_s16(round_ptr[1]);
v_quant = vmovq_n_s16(quant_ptr[1]);
v_dequant = vmovq_n_s16(dequant_ptr[1]);
}
// now process the rest of the ac coeffs
for (i = 8; i < count; i += 8) {
const int16x8_t v_iscan = vld1q_s16(&iscan[i]);
const int16x8_t v_coeff = vld1q_s16(&coeff_ptr[i]);
const int16x8_t v_coeff_sign = vshrq_n_s16(v_coeff, 15);
const int16x8_t v_tmp = vabaq_s16(v_round, v_coeff, v_zero);
const int32x4_t v_tmp_lo = vmull_s16(vget_low_s16(v_tmp),
vget_low_s16(v_quant));
const int32x4_t v_tmp_hi = vmull_s16(vget_high_s16(v_tmp),
vget_high_s16(v_quant));
const int16x8_t v_tmp2 = vcombine_s16(vshrn_n_s32(v_tmp_lo, 16),
vshrn_n_s32(v_tmp_hi, 16));
const uint16x8_t v_nz_mask = vceqq_s16(v_tmp2, v_zero);
const int16x8_t v_iscan_plus1 = vaddq_s16(v_iscan, v_one);
const int16x8_t v_nz_iscan = vbslq_s16(v_nz_mask, v_zero, v_iscan_plus1);
const int16x8_t v_qcoeff_a = veorq_s16(v_tmp2, v_coeff_sign);
const int16x8_t v_qcoeff = vsubq_s16(v_qcoeff_a, v_coeff_sign);
const int16x8_t v_dqcoeff = vmulq_s16(v_qcoeff, v_dequant);
v_eobmax_76543210 = vmaxq_s16(v_eobmax_76543210, v_nz_iscan);
vst1q_s16(&qcoeff_ptr[i], v_qcoeff);
vst1q_s16(&dqcoeff_ptr[i], v_dqcoeff);
}
{
const int16x4_t v_eobmax_3210 =
vmax_s16(vget_low_s16(v_eobmax_76543210),
vget_high_s16(v_eobmax_76543210));
const int64x1_t v_eobmax_xx32 =
vshr_n_s64(vreinterpret_s64_s16(v_eobmax_3210), 32);
const int16x4_t v_eobmax_tmp =
vmax_s16(v_eobmax_3210, vreinterpret_s16_s64(v_eobmax_xx32));
const int64x1_t v_eobmax_xxx3 =
vshr_n_s64(vreinterpret_s64_s16(v_eobmax_tmp), 16);
const int16x4_t v_eobmax_final =
vmax_s16(v_eobmax_tmp, vreinterpret_s16_s64(v_eobmax_xxx3));
*eob_ptr = (uint16_t)vget_lane_s16(v_eobmax_final, 0);
}
} else {
memset(qcoeff_ptr, 0, count * sizeof(int16_t));
memset(dqcoeff_ptr, 0, count * sizeof(int16_t));
*eob_ptr = 0;
}
}

56
vp10/encoder/mips/msa/vp9_avg_msa.c Normal file
View File

@ -0,0 +1,56 @@
/*
* 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_dsp/mips/macros_msa.h"
uint32_t vp10_avg_8x8_msa(const uint8_t *src, int32_t src_stride) {
uint32_t sum_out;
v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
v8u16 sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7;
v4u32 sum = { 0 };
LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
HADD_UB4_UH(src0, src1, src2, src3, sum0, sum1, sum2, sum3);
HADD_UB4_UH(src4, src5, src6, src7, sum4, sum5, sum6, sum7);
ADD4(sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7, sum0, sum2, sum4, sum6);
ADD2(sum0, sum2, sum4, sum6, sum0, sum4);
sum0 += sum4;
sum = __msa_hadd_u_w(sum0, sum0);
sum0 = (v8u16)__msa_pckev_h((v8i16)sum, (v8i16)sum);
sum = __msa_hadd_u_w(sum0, sum0);
sum = (v4u32)__msa_srari_w((v4i32)sum, 6);
sum_out = __msa_copy_u_w((v4i32)sum, 0);
return sum_out;
}
uint32_t vp10_avg_4x4_msa(const uint8_t *src, int32_t src_stride) {
uint32_t sum_out;
uint32_t src0, src1, src2, src3;
v16u8 vec = { 0 };
v8u16 sum0;
v4u32 sum1;
v2u64 sum2;
LW4(src, src_stride, src0, src1, src2, src3);
INSERT_W4_UB(src0, src1, src2, src3, vec);
sum0 = __msa_hadd_u_h(vec, vec);
sum1 = __msa_hadd_u_w(sum0, sum0);
sum0 = (v8u16)__msa_pckev_h((v8i16)sum1, (v8i16)sum1);
sum1 = __msa_hadd_u_w(sum0, sum0);
sum2 = __msa_hadd_u_d(sum1, sum1);
sum1 = (v4u32)__msa_srari_w((v4i32)sum2, 4);
sum_out = __msa_copy_u_w((v4i32)sum1, 0);
return sum_out;
}

114
vp10/encoder/mips/msa/vp9_error_msa.c Normal file
View File

@ -0,0 +1,114 @@
/*
* 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_dsp/mips/macros_msa.h"
#define BLOCK_ERROR_BLOCKSIZE_MSA(BSize) \
static int64_t block_error_##BSize##size_msa(const int16_t *coeff_ptr, \
const int16_t *dq_coeff_ptr, \
int64_t *ssz) { \
int64_t err = 0; \
uint32_t loop_cnt; \
v8i16 coeff, dq_coeff, coeff_r_h, coeff_l_h; \
v4i32 diff_r, diff_l, coeff_r_w, coeff_l_w; \
v2i64 sq_coeff_r, sq_coeff_l; \
v2i64 err0, err_dup0, err1, err_dup1; \
\
coeff = LD_SH(coeff_ptr); \
dq_coeff = LD_SH(dq_coeff_ptr); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DOTP_SW2_SD(coeff_r_w, coeff_l_w, coeff_r_w, coeff_l_w, \
sq_coeff_r, sq_coeff_l); \
DOTP_SW2_SD(diff_r, diff_l, diff_r, diff_l, err0, err1); \
\
coeff = LD_SH(coeff_ptr + 8); \
dq_coeff = LD_SH(dq_coeff_ptr + 8); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff_ptr += 16; \
dq_coeff_ptr += 16; \
\
for (loop_cnt = ((BSize >> 4) - 1); loop_cnt--;) { \
coeff = LD_SH(coeff_ptr); \
dq_coeff = LD_SH(dq_coeff_ptr); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff = LD_SH(coeff_ptr + 8); \
dq_coeff = LD_SH(dq_coeff_ptr + 8); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff_ptr += 16; \
dq_coeff_ptr += 16; \
} \
\
err_dup0 = __msa_splati_d(sq_coeff_r, 1); \
err_dup1 = __msa_splati_d(sq_coeff_l, 1); \
sq_coeff_r += err_dup0; \
sq_coeff_l += err_dup1; \
*ssz = __msa_copy_s_d(sq_coeff_r, 0); \
*ssz += __msa_copy_s_d(sq_coeff_l, 0); \
\
err_dup0 = __msa_splati_d(err0, 1); \
err_dup1 = __msa_splati_d(err1, 1); \
err0 += err_dup0; \
err1 += err_dup1; \
err = __msa_copy_s_d(err0, 0); \
err += __msa_copy_s_d(err1, 0); \
\
return err; \
}
BLOCK_ERROR_BLOCKSIZE_MSA(16);
BLOCK_ERROR_BLOCKSIZE_MSA(64);
BLOCK_ERROR_BLOCKSIZE_MSA(256);
BLOCK_ERROR_BLOCKSIZE_MSA(1024);
int64_t vp10_block_error_msa(const tran_low_t *coeff_ptr,
const tran_low_t *dq_coeff_ptr,
intptr_t blk_size, int64_t *ssz) {
int64_t err;
const int16_t *coeff = (const int16_t *)coeff_ptr;
const int16_t *dq_coeff = (const int16_t *)dq_coeff_ptr;
switch (blk_size) {
case 16:
err = block_error_16size_msa(coeff, dq_coeff, ssz);
break;
case 64:
err = block_error_64size_msa(coeff, dq_coeff, ssz);
break;
case 256:
err = block_error_256size_msa(coeff, dq_coeff, ssz);
break;
case 1024:
err = block_error_1024size_msa(coeff, dq_coeff, ssz);
break;
default:
err = vp10_block_error_c(coeff_ptr, dq_coeff_ptr, blk_size, ssz);
break;
}
return err;
}

507
vp10/encoder/mips/msa/vp9_fdct16x16_msa.c Normal file
View File

@ -0,0 +1,507 @@
/*
* 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 <assert.h>
#include "vp10/common/vp9_enums.h"
#include "vp10/encoder/mips/msa/vp9_fdct_msa.h"
#include "vpx_dsp/mips/fwd_txfm_msa.h"
static void fadst16_cols_step1_msa(const int16_t *input, int32_t stride,
const int32_t *const0, int16_t *int_buf) {
v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3;
v4i32 k0, k1, k2, k3;
/* load input data */
r0 = LD_SH(input);
r15 = LD_SH(input + 15 * stride);
r7 = LD_SH(input + 7 * stride);
r8 = LD_SH(input + 8 * stride);
SLLI_4V(r0, r15, r7, r8, 2);
/* stage 1 */
LD_SW2(const0, 4, k0, k1);
LD_SW2(const0 + 8, 4, k2, k3);
MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
r3 = LD_SH(input + 3 * stride);
r4 = LD_SH(input + 4 * stride);
r11 = LD_SH(input + 11 * stride);
r12 = LD_SH(input + 12 * stride);
SLLI_4V(r3, r4, r11, r12, 2);
LD_SW2(const0 + 4 * 4, 4, k0, k1);
LD_SW2(const0 + 4 * 6, 4, k2, k3);
MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
/* stage 2 */
BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1);
ST_SH2(tp0, tp2, int_buf, 8);
ST_SH2(tp1, tp3, int_buf + 4 * 8, 8);
LD_SW2(const0 + 4 * 8, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 10);
MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
ST_SH2(h0, h1, int_buf + 8 * 8, 8);
ST_SH2(h3, h2, int_buf + 12 * 8, 8);
r9 = LD_SH(input + 9 * stride);
r6 = LD_SH(input + 6 * stride);
r1 = LD_SH(input + stride);
r14 = LD_SH(input + 14 * stride);
SLLI_4V(r9, r6, r1, r14, 2);
LD_SW2(const0 + 4 * 11, 4, k0, k1);
LD_SW2(const0 + 4 * 13, 4, k2, k3);
MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3);
ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8);
r13 = LD_SH(input + 13 * stride);
r2 = LD_SH(input + 2 * stride);
r5 = LD_SH(input + 5 * stride);
r10 = LD_SH(input + 10 * stride);
SLLI_4V(r13, r2, r5, r10, 2);
LD_SW2(const0 + 4 * 15, 4, k0, k1);
LD_SW2(const0 + 4 * 17, 4, k2, k3);
MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3);
ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8);
BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3);
ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8);
}
static void fadst16_cols_step2_msa(int16_t *int_buf, const int32_t *const0,
int16_t *out) {
int16_t *out_ptr = out + 128;
v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15;
v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11;
v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
v4i32 k0, k1, k2, k3;
LD_SH2(int_buf + 3 * 8, 4 * 8, g13, g15);
LD_SH2(int_buf + 11 * 8, 4 * 8, g5, g7);
LD_SW2(const0 + 4 * 19, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 21);
MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
tp0 = LD_SH(int_buf + 4 * 8);
tp1 = LD_SH(int_buf + 5 * 8);
tp3 = LD_SH(int_buf + 10 * 8);
tp2 = LD_SH(int_buf + 14 * 8);
LD_SW2(const0 + 4 * 22, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 24);
MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7);
out4 = -out4;
ST_SH(out4, (out + 3 * 16));
ST_SH(out5, (out_ptr + 4 * 16));
h1 = LD_SH(int_buf + 9 * 8);
h3 = LD_SH(int_buf + 12 * 8);
MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
out13 = -out13;
ST_SH(out12, (out + 2 * 16));
ST_SH(out13, (out_ptr + 5 * 16));
tp0 = LD_SH(int_buf);
tp1 = LD_SH(int_buf + 8);
tp2 = LD_SH(int_buf + 2 * 8);
tp3 = LD_SH(int_buf + 6 * 8);
BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10);
out1 = -out1;
ST_SH(out0, (out));
ST_SH(out1, (out_ptr + 7 * 16));
h0 = LD_SH(int_buf + 8 * 8);
h2 = LD_SH(int_buf + 13 * 8);
BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
out8 = -out8;
ST_SH(out8, (out + 16));
ST_SH(out9, (out_ptr + 6 * 16));
/* stage 4 */
LD_SW2(const0 + 4 * 25, 4, k0, k1);
LD_SW2(const0 + 4 * 27, 4, k2, k3);
MADD_SHORT(h10, h11, k1, k2, out2, out3);
ST_SH(out2, (out + 7 * 16));
ST_SH(out3, (out_ptr));
MADD_SHORT(out6, out7, k0, k3, out6, out7);
ST_SH(out6, (out + 4 * 16));
ST_SH(out7, (out_ptr + 3 * 16));
MADD_SHORT(out10, out11, k0, k3, out10, out11);
ST_SH(out10, (out + 6 * 16));
ST_SH(out11, (out_ptr + 16));
MADD_SHORT(out14, out15, k1, k2, out14, out15);
ST_SH(out14, (out + 5 * 16));
ST_SH(out15, (out_ptr + 2 * 16));
}
static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) {
v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
/* load input data */
LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
r0, r1, r2, r3, r4, r5, r6, r7);
FDCT_POSTPROC_2V_NEG_H(r0, r1);
FDCT_POSTPROC_2V_NEG_H(r2, r3);
FDCT_POSTPROC_2V_NEG_H(r4, r5);
FDCT_POSTPROC_2V_NEG_H(r6, r7);
ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8);
out += 64;
LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
r8, r9, r10, r11, r12, r13, r14, r15);
FDCT_POSTPROC_2V_NEG_H(r8, r9);
FDCT_POSTPROC_2V_NEG_H(r10, r11);
FDCT_POSTPROC_2V_NEG_H(r12, r13);
FDCT_POSTPROC_2V_NEG_H(r14, r15);
ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8);
out += 64;
/* load input data */
input += 128;
LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
r0, r1, r2, r3, r4, r5, r6, r7);
FDCT_POSTPROC_2V_NEG_H(r0, r1);
FDCT_POSTPROC_2V_NEG_H(r2, r3);
FDCT_POSTPROC_2V_NEG_H(r4, r5);
FDCT_POSTPROC_2V_NEG_H(r6, r7);
ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8);
out += 64;
LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
r8, r9, r10, r11, r12, r13, r14, r15);
FDCT_POSTPROC_2V_NEG_H(r8, r9);
FDCT_POSTPROC_2V_NEG_H(r10, r11);
FDCT_POSTPROC_2V_NEG_H(r12, r13);
FDCT_POSTPROC_2V_NEG_H(r14, r15);
ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8);
}
static void fadst16_rows_step1_msa(int16_t *input, const int32_t *const0,
int16_t *int_buf) {
v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3;
v4i32 k0, k1, k2, k3;
/* load input data */
r0 = LD_SH(input);
r7 = LD_SH(input + 7 * 8);
r8 = LD_SH(input + 8 * 8);
r15 = LD_SH(input + 15 * 8);
/* stage 1 */
LD_SW2(const0, 4, k0, k1);
LD_SW2(const0 + 4 * 2, 4, k2, k3);
MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
r3 = LD_SH(input + 3 * 8);
r4 = LD_SH(input + 4 * 8);
r11 = LD_SH(input + 11 * 8);
r12 = LD_SH(input + 12 * 8);
LD_SW2(const0 + 4 * 4, 4, k0, k1);
LD_SW2(const0 + 4 * 6, 4, k2, k3);
MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
/* stage 2 */
BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1);
ST_SH2(tp0, tp1, int_buf, 4 * 8);
ST_SH2(tp2, tp3, int_buf + 8, 4 * 8);
LD_SW2(const0 + 4 * 8, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 10);
MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
ST_SH2(h0, h3, int_buf + 8 * 8, 4 * 8);
ST_SH2(h1, h2, int_buf + 9 * 8, 4 * 8);
r1 = LD_SH(input + 8);
r6 = LD_SH(input + 6 * 8);
r9 = LD_SH(input + 9 * 8);
r14 = LD_SH(input + 14 * 8);
LD_SW2(const0 + 4 * 11, 4, k0, k1);
LD_SW2(const0 + 4 * 13, 4, k2, k3);
MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3);
ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8);
r2 = LD_SH(input + 2 * 8);
r5 = LD_SH(input + 5 * 8);
r10 = LD_SH(input + 10 * 8);
r13 = LD_SH(input + 13 * 8);
LD_SW2(const0 + 4 * 15, 4, k0, k1);
LD_SW2(const0 + 4 * 17, 4, k2, k3);
MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3);
ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8);
BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3);
ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8);
}
static void fadst16_rows_step2_msa(int16_t *int_buf, const int32_t *const0,
int16_t *out) {
int16_t *out_ptr = out + 8;
v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15;
v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11;
v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
v4i32 k0, k1, k2, k3;
g13 = LD_SH(int_buf + 3 * 8);
g15 = LD_SH(int_buf + 7 * 8);
g5 = LD_SH(int_buf + 11 * 8);
g7 = LD_SH(int_buf + 15 * 8);
LD_SW2(const0 + 4 * 19, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 21);
MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
tp0 = LD_SH(int_buf + 4 * 8);
tp1 = LD_SH(int_buf + 5 * 8);
tp3 = LD_SH(int_buf + 10 * 8);
tp2 = LD_SH(int_buf + 14 * 8);
LD_SW2(const0 + 4 * 22, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 24);
MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7);
out4 = -out4;
ST_SH(out4, (out + 3 * 16));
ST_SH(out5, (out_ptr + 4 * 16));
h1 = LD_SH(int_buf + 9 * 8);
h3 = LD_SH(int_buf + 12 * 8);
MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
out13 = -out13;
ST_SH(out12, (out + 2 * 16));
ST_SH(out13, (out_ptr + 5 * 16));
tp0 = LD_SH(int_buf);
tp1 = LD_SH(int_buf + 8);
tp2 = LD_SH(int_buf + 2 * 8);
tp3 = LD_SH(int_buf + 6 * 8);
BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10);
out1 = -out1;
ST_SH(out0, (out));
ST_SH(out1, (out_ptr + 7 * 16));
h0 = LD_SH(int_buf + 8 * 8);
h2 = LD_SH(int_buf + 13 * 8);
BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
out8 = -out8;
ST_SH(out8, (out + 16));
ST_SH(out9, (out_ptr + 6 * 16));
/* stage 4 */
LD_SW2(const0 + 4 * 25, 4, k0, k1);
LD_SW2(const0 + 4 * 27, 4, k2, k3);
MADD_SHORT(h10, h11, k1, k2, out2, out3);
ST_SH(out2, (out + 7 * 16));
ST_SH(out3, (out_ptr));
MADD_SHORT(out6, out7, k0, k3, out6, out7);
ST_SH(out6, (out + 4 * 16));
ST_SH(out7, (out_ptr + 3 * 16));
MADD_SHORT(out10, out11, k0, k3, out10, out11);
ST_SH(out10, (out + 6 * 16));
ST_SH(out11, (out_ptr + 16));
MADD_SHORT(out14, out15, k1, k2, out14, out15);
ST_SH(out14, (out + 5 * 16));
ST_SH(out15, (out_ptr + 2 * 16));
}
static void fadst16_transpose_msa(int16_t *input, int16_t *out) {
v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
/* load input data */
LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11,
l4, l12, l5, l13, l6, l14, l7, l15);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
r0, r1, r2, r3, r4, r5, r6, r7);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
r8, r9, r10, r11, r12, r13, r14, r15);
ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
out += 16 * 8;
/* load input data */
input += 128;
LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11,
l4, l12, l5, l13, l6, l14, l7, l15);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
r0, r1, r2, r3, r4, r5, r6, r7);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
r8, r9, r10, r11, r12, r13, r14, r15);
ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
}
static void postproc_fdct16x8_1d_row(int16_t *intermediate, int16_t *output) {
int16_t *temp = intermediate;
int16_t *out = output;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11;
v8i16 in12, in13, in14, in15;
LD_SH8(temp, 16, in0, in1, in2, in3, in4, in5, in6, in7);
temp = intermediate + 8;
LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
in8, in9, in10, in11, in12, in13, in14, in15);
FDCT_POSTPROC_2V_NEG_H(in0, in1);
FDCT_POSTPROC_2V_NEG_H(in2, in3);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
FDCT_POSTPROC_2V_NEG_H(in6, in7);
FDCT_POSTPROC_2V_NEG_H(in8, in9);
FDCT_POSTPROC_2V_NEG_H(in10, in11);
FDCT_POSTPROC_2V_NEG_H(in12, in13);
FDCT_POSTPROC_2V_NEG_H(in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15,
tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
in8, in9, in10, in11, in12, in13, in14, in15);
temp = intermediate;
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, temp, 16);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
temp = intermediate;
LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3,
tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3);
ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, out, 16);
TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7,
tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7);
out = output + 8;
ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, out, 16);
}
void vp10_fht16x16_msa(const int16_t *input, int16_t *output,
int32_t stride, int32_t tx_type) {
DECLARE_ALIGNED(32, int16_t, tmp[256]);
DECLARE_ALIGNED(32, int16_t, trans_buf[256]);
DECLARE_ALIGNED(32, int16_t, tmp_buf[128]);
int32_t i;
int16_t *ptmpbuf = &tmp_buf[0];
int16_t *trans = &trans_buf[0];
const int32_t const_arr[29 * 4] = {
52707308, 52707308, 52707308, 52707308,
-1072430300, -1072430300, -1072430300, -1072430300,
795618043, 795618043, 795618043, 795618043,
-721080468, -721080468, -721080468, -721080468,
459094491, 459094491, 459094491, 459094491,
-970646691, -970646691, -970646691, -970646691,
1010963856, 1010963856, 1010963856, 1010963856,
-361743294, -361743294, -361743294, -361743294,
209469125, 209469125, 209469125, 209469125,
-1053094788, -1053094788, -1053094788, -1053094788,
1053160324, 1053160324, 1053160324, 1053160324,
639644520, 639644520, 639644520, 639644520,
-862444000, -862444000, -862444000, -862444000,
1062144356, 1062144356, 1062144356, 1062144356,
-157532337, -157532337, -157532337, -157532337,
260914709, 260914709, 260914709, 260914709,
-1041559667, -1041559667, -1041559667, -1041559667,
920985831, 920985831, 920985831, 920985831,
-551995675, -551995675, -551995675, -551995675,
596522295, 596522295, 596522295, 596522295,
892853362, 892853362, 892853362, 892853362,
-892787826, -892787826, -892787826, -892787826,
410925857, 410925857, 410925857, 410925857,
-992012162, -992012162, -992012162, -992012162,
992077698, 992077698, 992077698, 992077698,
759246145, 759246145, 759246145, 759246145,
-759180609, -759180609, -759180609, -759180609,
-759222975, -759222975, -759222975, -759222975,
759288511, 759288511, 759288511, 759288511 };
switch (tx_type) {
case DCT_DCT:
/* column transform */
for (i = 0; i < 2; ++i) {
fdct8x16_1d_column(input + 8 * i, tmp + 8 * i, stride);
}
/* row transform */
for (i = 0; i < 2; ++i) {
fdct16x8_1d_row(tmp + (128 * i), output + (128 * i));
}
break;
case ADST_DCT:
/* column transform */
for (i = 0; i < 2; ++i) {
fadst16_cols_step1_msa(input + (i << 3), stride, const_arr, ptmpbuf);
fadst16_cols_step2_msa(ptmpbuf, const_arr, tmp + (i << 3));
}
/* row transform */
for (i = 0; i < 2; ++i) {
postproc_fdct16x8_1d_row(tmp + (128 * i), output + (128 * i));
}
break;
case DCT_ADST:
/* column transform */
for (i = 0; i < 2; ++i) {
fdct8x16_1d_column(input + 8 * i, tmp + 8 * i, stride);
}
fadst16_transpose_postproc_msa(tmp, trans);
/* row transform */
for (i = 0; i < 2; ++i) {
fadst16_rows_step1_msa(trans + (i << 7), const_arr, ptmpbuf);
fadst16_rows_step2_msa(ptmpbuf, const_arr, tmp + (i << 7));
}
fadst16_transpose_msa(tmp, output);
break;
case ADST_ADST:
/* column transform */
for (i = 0; i < 2; ++i) {
fadst16_cols_step1_msa(input + (i << 3), stride, const_arr, ptmpbuf);
fadst16_cols_step2_msa(ptmpbuf, const_arr, tmp + (i << 3));
}
fadst16_transpose_postproc_msa(tmp, trans);
/* row transform */
for (i = 0; i < 2; ++i) {
fadst16_rows_step1_msa(trans + (i << 7), const_arr, ptmpbuf);
fadst16_rows_step2_msa(ptmpbuf, const_arr, tmp + (i << 7));
}
fadst16_transpose_msa(tmp, output);
break;
default:
assert(0);
break;
}
}

99
vp10/encoder/mips/msa/vp9_fdct4x4_msa.c Normal file
View File

@ -0,0 +1,99 @@
/*
* 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 <assert.h>
#include "vp10/common/vp9_enums.h"
#include "vp10/encoder/mips/msa/vp9_fdct_msa.h"
void vp10_fwht4x4_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3, in4;
LD_SH4(input, src_stride, in0, in1, in2, in3);
in0 += in1;
in3 -= in2;
in4 = (in0 - in3) >> 1;
SUB2(in4, in1, in4, in2, in1, in2);
in0 -= in2;
in3 += in1;
TRANSPOSE4x4_SH_SH(in0, in2, in3, in1, in0, in2, in3, in1);
in0 += in2;
in1 -= in3;
in4 = (in0 - in1) >> 1;
SUB2(in4, in2, in4, in3, in2, in3);
in0 -= in3;
in1 += in2;
SLLI_4V(in0, in1, in2, in3, 2);
TRANSPOSE4x4_SH_SH(in0, in3, in1, in2, in0, in3, in1, in2);
ST4x2_UB(in0, output, 4);
ST4x2_UB(in3, output + 4, 4);
ST4x2_UB(in1, output + 8, 4);
ST4x2_UB(in2, output + 12, 4);
}
void vp10_fht4x4_msa(const int16_t *input, int16_t *output, int32_t stride,
int32_t tx_type) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 temp, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
temp = __msa_ceqi_h(in0, 0);
temp = (v8i16)__msa_xori_b((v16u8)temp, 255);
temp = mask & temp;
in0 += temp;
}
switch (tx_type) {
case DCT_DCT:
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_DCT:
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case DCT_ADST:
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_ADST:
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
default:
assert(0);
break;
}
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 8);
}

66
vp10/encoder/mips/msa/vp9_fdct8x8_msa.c Normal file
View File

@ -0,0 +1,66 @@
/*
* 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 <assert.h>
#include "vp10/common/vp9_enums.h"
#include "vp10/encoder/mips/msa/vp9_fdct_msa.h"
void vp10_fht8x8_msa(const int16_t *input, int16_t *output, int32_t stride,
int32_t tx_type) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
LD_SH8(input, stride, in0, in1, in2, in3, in4, in5, in6, in7);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
switch (tx_type) {
case DCT_DCT:
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_DCT:
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case DCT_ADST:
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_ADST:
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
default:
assert(0);
break;
}
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 8);
}

117
vp10/encoder/mips/msa/vp9_fdct_msa.h Normal file
View File

@ -0,0 +1,117 @@
/*
* 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 VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_
#define VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_
#include "vpx_dsp/mips/fwd_txfm_msa.h"
#include "vpx_dsp/mips/txfm_macros_msa.h"
#include "vpx_ports/mem.h"
#define VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 cnst0_m, cnst1_m, cnst2_m, cnst3_m, cnst4_m; \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m, s0_m, s1_m; \
v8i16 coeff0_m = { cospi_2_64, cospi_6_64, cospi_10_64, cospi_14_64, \
cospi_18_64, cospi_22_64, cospi_26_64, cospi_30_64 }; \
v8i16 coeff1_m = { cospi_8_64, -cospi_8_64, cospi_16_64, -cospi_16_64, \
cospi_24_64, -cospi_24_64, 0, 0 }; \
\
SPLATI_H2_SH(coeff0_m, 0, 7, cnst0_m, cnst1_m); \
cnst2_m = -cnst0_m; \
ILVEV_H2_SH(cnst0_m, cnst1_m, cnst1_m, cnst2_m, cnst0_m, cnst1_m); \
SPLATI_H2_SH(coeff0_m, 4, 3, cnst2_m, cnst3_m); \
cnst4_m = -cnst2_m; \
ILVEV_H2_SH(cnst2_m, cnst3_m, cnst3_m, cnst4_m, cnst2_m, cnst3_m); \
\
ILVRL_H2_SH(in0, in7, vec1_m, vec0_m); \
ILVRL_H2_SH(in4, in3, vec3_m, vec2_m); \
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, \
cnst1_m, cnst2_m, cnst3_m, in7, in0, \
in4, in3); \
\
SPLATI_H2_SH(coeff0_m, 2, 5, cnst0_m, cnst1_m); \
cnst2_m = -cnst0_m; \
ILVEV_H2_SH(cnst0_m, cnst1_m, cnst1_m, cnst2_m, cnst0_m, cnst1_m); \
SPLATI_H2_SH(coeff0_m, 6, 1, cnst2_m, cnst3_m); \
cnst4_m = -cnst2_m; \
ILVEV_H2_SH(cnst2_m, cnst3_m, cnst3_m, cnst4_m, cnst2_m, cnst3_m); \
\
ILVRL_H2_SH(in2, in5, vec1_m, vec0_m); \
ILVRL_H2_SH(in6, in1, vec3_m, vec2_m); \
\
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, \
cnst1_m, cnst2_m, cnst3_m, in5, in2, \
in6, in1); \
BUTTERFLY_4(in7, in0, in2, in5, s1_m, s0_m, in2, in5); \
out7 = -s0_m; \
out0 = s1_m; \
\
SPLATI_H4_SH(coeff1_m, 0, 4, 1, 5, cnst0_m, cnst1_m, cnst2_m, cnst3_m); \
\
ILVEV_H2_SH(cnst3_m, cnst0_m, cnst1_m, cnst2_m, cnst3_m, cnst2_m); \
cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
cnst1_m = cnst0_m; \
\
ILVRL_H2_SH(in4, in3, vec1_m, vec0_m); \
ILVRL_H2_SH(in6, in1, vec3_m, vec2_m); \
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, \
cnst2_m, cnst3_m, cnst1_m, out1, out6, \
s0_m, s1_m); \
\
SPLATI_H2_SH(coeff1_m, 2, 3, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
\
ILVRL_H2_SH(in2, in5, vec1_m, vec0_m); \
ILVRL_H2_SH(s0_m, s1_m, vec3_m, vec2_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst0_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst1_m); \
\
out1 = -out1; \
out3 = -out3; \
out5 = -out5; \
}
#define VP9_FADST4(in0, in1, in2, in3, out0, out1, out2, out3) { \
v4i32 s0_m, s1_m, s2_m, s3_m, constant_m; \
v4i32 in0_r_m, in1_r_m, in2_r_m, in3_r_m; \
\
UNPCK_R_SH_SW(in0, in0_r_m); \
UNPCK_R_SH_SW(in1, in1_r_m); \
UNPCK_R_SH_SW(in2, in2_r_m); \
UNPCK_R_SH_SW(in3, in3_r_m); \
\
constant_m = __msa_fill_w(sinpi_4_9); \
MUL2(in0_r_m, constant_m, in3_r_m, constant_m, s1_m, s0_m); \
\
constant_m = __msa_fill_w(sinpi_1_9); \
s0_m += in0_r_m * constant_m; \
s1_m -= in1_r_m * constant_m; \
\
constant_m = __msa_fill_w(sinpi_2_9); \
s0_m += in1_r_m * constant_m; \
s1_m += in3_r_m * constant_m; \
\
s2_m = in0_r_m + in1_r_m - in3_r_m; \
\
constant_m = __msa_fill_w(sinpi_3_9); \
MUL2(in2_r_m, constant_m, s2_m, constant_m, s3_m, in1_r_m); \
\
in0_r_m = s0_m + s3_m; \
s2_m = s1_m - s3_m; \
s3_m = s1_m - s0_m + s3_m; \
\
SRARI_W4_SW(in0_r_m, in1_r_m, s2_m, s3_m, DCT_CONST_BITS); \
PCKEV_H4_SH(in0_r_m, in0_r_m, in1_r_m, in1_r_m, s2_m, s2_m, \
s3_m, s3_m, out0, out1, out2, out3); \
}
#endif /* VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_ */

289
vp10/encoder/mips/msa/vp9_temporal_filter_msa.c Normal file
View File

@ -0,0 +1,289 @@
/*
* 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_dsp/mips/macros_msa.h"
static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
uint32_t stride,
uint8_t *frm2_ptr,
int32_t filt_sth,
int32_t filt_wgt,
uint32_t *acc,
uint16_t *cnt) {
uint32_t row;
uint64_t f0, f1, f2, f3;
v16i8 frm2, frm1 = { 0 };
v16i8 frm4, frm3 = { 0 };
v16u8 frm_r, frm_l;
v8i16 frm2_r, frm2_l;
v8i16 diff0, diff1, mod0_h, mod1_h;
v4i32 cnst3, cnst16, filt_wt, strength;
v4i32 mod0_w, mod1_w, mod2_w, mod3_w;
v4i32 diff0_r, diff0_l, diff1_r, diff1_l;
v4i32 frm2_rr, frm2_rl, frm2_lr, frm2_ll;
v4i32 acc0, acc1, acc2, acc3;
v8i16 cnt0, cnt1;
filt_wt = __msa_fill_w(filt_wgt);
strength = __msa_fill_w(filt_sth);
cnst3 = __msa_ldi_w(3);
cnst16 = __msa_ldi_w(16);
for (row = 2; row--;) {
LD4(frm1_ptr, stride, f0, f1, f2, f3);
frm1_ptr += (4 * stride);
LD_SB2(frm2_ptr, 16, frm2, frm4);
frm2_ptr += 32;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
INSERT_D2_SB(f0, f1, frm1);
INSERT_D2_SB(f2, f3, frm3);
ILVRL_B2_UB(frm1, frm2, frm_r, frm_l);
HSUB_UB2_SH(frm_r, frm_l, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < cnst16);
diff0_l = (mod1_w < cnst16);
diff1_r = (mod2_w < cnst16);
diff1_l = (mod3_w < cnst16);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filt_wt, mod1_w, filt_wt, mod2_w, filt_wt, mod3_w, filt_wt,
mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
UNPCK_UB_SH(frm2, frm2_r, frm2_l);
UNPCK_SH_SW(frm2_r, frm2_rr, frm2_rl);
UNPCK_SH_SW(frm2_l, frm2_lr, frm2_ll);
MUL4(mod0_w, frm2_rr, mod1_w, frm2_rl, mod2_w, frm2_lr, mod3_w, frm2_ll,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
acc += 8;
ST_SW2(mod2_w, mod3_w, acc, 4);
acc += 8;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frm3, frm4, frm_r, frm_l);
HSUB_UB2_SH(frm_r, frm_l, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < cnst16);
diff0_l = (mod1_w < cnst16);
diff1_r = (mod2_w < cnst16);
diff1_l = (mod3_w < cnst16);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filt_wt, mod1_w, filt_wt, mod2_w, filt_wt, mod3_w, filt_wt,
mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
UNPCK_UB_SH(frm4, frm2_r, frm2_l);
UNPCK_SH_SW(frm2_r, frm2_rr, frm2_rl);
UNPCK_SH_SW(frm2_l, frm2_lr, frm2_ll);
MUL4(mod0_w, frm2_rr, mod1_w, frm2_rl, mod2_w, frm2_lr, mod3_w, frm2_ll,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
acc += 8;
ST_SW2(mod2_w, mod3_w, acc, 4);
acc += 8;
}
}
static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr,
uint32_t stride,
uint8_t *frm2_ptr,
int32_t filt_sth,
int32_t filt_wgt,
uint32_t *acc,
uint16_t *cnt) {
uint32_t row;
v16i8 frm1, frm2, frm3, frm4;
v16u8 frm_r, frm_l;
v16i8 zero = { 0 };
v8u16 frm2_r, frm2_l;
v8i16 diff0, diff1, mod0_h, mod1_h;
v4i32 cnst3, cnst16, filt_wt, strength;
v4i32 mod0_w, mod1_w, mod2_w, mod3_w;
v4i32 diff0_r, diff0_l, diff1_r, diff1_l;
v4i32 frm2_rr, frm2_rl, frm2_lr, frm2_ll;
v4i32 acc0, acc1, acc2, acc3;
v8i16 cnt0, cnt1;
filt_wt = __msa_fill_w(filt_wgt);
strength = __msa_fill_w(filt_sth);
cnst3 = __msa_ldi_w(3);
cnst16 = __msa_ldi_w(16);
for (row = 8; row--;) {
LD_SB2(frm1_ptr, stride, frm1, frm3);
frm1_ptr += stride;
LD_SB2(frm2_ptr, 16, frm2, frm4);
frm2_ptr += 16;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frm1, frm2, frm_r, frm_l);
HSUB_UB2_SH(frm_r, frm_l, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l, diff1_l,
mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < cnst16);
diff0_l = (mod1_w < cnst16);
diff1_r = (mod2_w < cnst16);
diff1_l = (mod3_w < cnst16);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filt_wt, mod1_w, filt_wt, mod2_w, filt_wt, mod3_w, filt_wt,
mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
ILVRL_B2_UH(zero, frm2, frm2_r, frm2_l);
UNPCK_SH_SW(frm2_r, frm2_rr, frm2_rl);
UNPCK_SH_SW(frm2_l, frm2_lr, frm2_ll);
MUL4(mod0_w, frm2_rr, mod1_w, frm2_rl, mod2_w, frm2_lr, mod3_w, frm2_ll,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
acc += 8;
ST_SW2(mod2_w, mod3_w, acc, 4);
acc += 8;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frm3, frm4, frm_r, frm_l);
HSUB_UB2_SH(frm_r, frm_l, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l, diff1_l,
mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < cnst16);
diff0_l = (mod1_w < cnst16);
diff1_r = (mod2_w < cnst16);
diff1_l = (mod3_w < cnst16);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filt_wt, mod1_w, filt_wt, mod2_w, filt_wt, mod3_w, filt_wt,
mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
ILVRL_B2_UH(zero, frm4, frm2_r, frm2_l);
UNPCK_SH_SW(frm2_r, frm2_rr, frm2_rl);
UNPCK_SH_SW(frm2_l, frm2_lr, frm2_ll);
MUL4(mod0_w, frm2_rr, mod1_w, frm2_rl, mod2_w, frm2_lr, mod3_w, frm2_ll,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
acc += 8;
ST_SW2(mod2_w, mod3_w, acc, 4);
acc += 8;
frm1_ptr += stride;
frm2_ptr += 16;
}
}
void vp10_temporal_filter_apply_msa(uint8_t *frame1_ptr, uint32_t stride,
uint8_t *frame2_ptr, uint32_t blk_w,
uint32_t blk_h, int32_t strength,
int32_t filt_wgt, uint32_t *accu,
uint16_t *cnt) {
if (8 == (blk_w * blk_h)) {
temporal_filter_apply_8size_msa(frame1_ptr, stride, frame2_ptr,
strength, filt_wgt, accu, cnt);
} else if (16 == (blk_w * blk_h)) {
temporal_filter_apply_16size_msa(frame1_ptr, stride, frame2_ptr,
strength, filt_wgt, accu, cnt);
} else {
vp10_temporal_filter_apply_c(frame1_ptr, stride, frame2_ptr, blk_w, blk_h,
strength, filt_wgt, accu, cnt);
}
}

163
vp10/encoder/vp9_aq_complexity.c Normal file
View File

@ -0,0 +1,163 @@
/*
* Copyright (c) 2014 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 <limits.h>
#include <math.h>
#include "vp10/encoder/vp9_aq_complexity.h"
#include "vp10/encoder/vp9_aq_variance.h"
#include "vp10/encoder/vp9_encodeframe.h"
#include "vp10/common/vp9_seg_common.h"
#include "vp10/encoder/vp9_segmentation.h"
#define AQ_C_SEGMENTS 5
#define DEFAULT_AQ2_SEG 3 // Neutral Q segment
#define AQ_C_STRENGTHS 3
static const double aq_c_q_adj_factor[AQ_C_STRENGTHS][AQ_C_SEGMENTS] =
{ {1.75, 1.25, 1.05, 1.00, 0.90},
{2.00, 1.50, 1.15, 1.00, 0.85},
{2.50, 1.75, 1.25, 1.00, 0.80} };
static const double aq_c_transitions[AQ_C_STRENGTHS][AQ_C_SEGMENTS] =
{ {0.15, 0.30, 0.55, 2.00, 100.0},
{0.20, 0.40, 0.65, 2.00, 100.0},
{0.25, 0.50, 0.75, 2.00, 100.0} };
static const double aq_c_var_thresholds[AQ_C_STRENGTHS][AQ_C_SEGMENTS] =
{ {-4.0, -3.0, -2.0, 100.00, 100.0},
{-3.5, -2.5, -1.5, 100.00, 100.0},
{-3.0, -2.0, -1.0, 100.00, 100.0} };
#define DEFAULT_COMPLEXITY 64
static int get_aq_c_strength(int q_index, vpx_bit_depth_t bit_depth) {
// Approximate base quatizer (truncated to int)
const int base_quant = vp10_ac_quant(q_index, 0, bit_depth) / 4;
return (base_quant > 10) + (base_quant > 25);
}
void vp10_setup_in_frame_q_adj(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
struct segmentation *const seg = &cm->seg;
// Make SURE use of floating point in this function is safe.
vpx_clear_system_state();
if (cm->frame_type == KEY_FRAME ||
cpi->refresh_alt_ref_frame ||
(cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref)) {
int segment;
const int aq_strength = get_aq_c_strength(cm->base_qindex, cm->bit_depth);
// Clear down the segment map.
memset(cpi->segmentation_map, DEFAULT_AQ2_SEG, cm->mi_rows * cm->mi_cols);
vp10_clearall_segfeatures(seg);
// Segmentation only makes sense if the target bits per SB is above a
// threshold. Below this the overheads will usually outweigh any benefit.
if (cpi->rc.sb64_target_rate < 256) {
vp10_disable_segmentation(seg);
return;
}
vp10_enable_segmentation(seg);
// Select delta coding method.
seg->abs_delta = SEGMENT_DELTADATA;
// Default segment "Q" feature is disabled so it defaults to the baseline Q.
vp10_disable_segfeature(seg, DEFAULT_AQ2_SEG, SEG_LVL_ALT_Q);
// Use some of the segments for in frame Q adjustment.
for (segment = 0; segment < AQ_C_SEGMENTS; ++segment) {
int qindex_delta;
if (segment == DEFAULT_AQ2_SEG)
continue;
qindex_delta =
vp10_compute_qdelta_by_rate(&cpi->rc, cm->frame_type, cm->base_qindex,
aq_c_q_adj_factor[aq_strength][segment],
cm->bit_depth);
// For AQ complexity mode, we dont allow Q0 in a segment if the base
// Q is not 0. Q0 (lossless) implies 4x4 only and in AQ mode 2 a segment
// Q delta is sometimes applied without going back around the rd loop.
// This could lead to an illegal combination of partition size and q.
if ((cm->base_qindex != 0) && ((cm->base_qindex + qindex_delta) == 0)) {
qindex_delta = -cm->base_qindex + 1;
}
if ((cm->base_qindex + qindex_delta) > 0) {
vp10_enable_segfeature(seg, segment, SEG_LVL_ALT_Q);
vp10_set_segdata(seg, segment, SEG_LVL_ALT_Q, qindex_delta);
}
}
}
}
#define DEFAULT_LV_THRESH 10.0
#define MIN_DEFAULT_LV_THRESH 8.0
#define VAR_STRENGTH_STEP 0.25
// Select a segment for the current block.
// The choice of segment for a block depends on the ratio of the projected
// bits for the block vs a target average and its spatial complexity.
void vp10_caq_select_segment(VP9_COMP *cpi, MACROBLOCK *mb, BLOCK_SIZE bs,
int mi_row, int mi_col, int projected_rate) {
VP9_COMMON *const cm = &cpi->common;
const int mi_offset = mi_row * cm->mi_cols + mi_col;
const int bw = num_8x8_blocks_wide_lookup[BLOCK_64X64];
const int bh = num_8x8_blocks_high_lookup[BLOCK_64X64];
const int xmis = MIN(cm->mi_cols - mi_col, num_8x8_blocks_wide_lookup[bs]);
const int ymis = MIN(cm->mi_rows - mi_row, num_8x8_blocks_high_lookup[bs]);
int x, y;
int i;
unsigned char segment;
if (0) {
segment = DEFAULT_AQ2_SEG;
} else {
// Rate depends on fraction of a SB64 in frame (xmis * ymis / bw * bh).
// It is converted to bits * 256 units.
const int target_rate = (cpi->rc.sb64_target_rate * xmis * ymis * 256) /
(bw * bh);
double logvar;
double low_var_thresh;
const int aq_strength = get_aq_c_strength(cm->base_qindex, cm->bit_depth);
vpx_clear_system_state();
low_var_thresh = (cpi->oxcf.pass == 2)
? MAX(cpi->twopass.mb_av_energy, MIN_DEFAULT_LV_THRESH)
: DEFAULT_LV_THRESH;
vp10_setup_src_planes(mb, cpi->Source, mi_row, mi_col);
logvar = vp10_log_block_var(cpi, mb, bs);
segment = AQ_C_SEGMENTS - 1; // Just in case no break out below.
for (i = 0; i < AQ_C_SEGMENTS; ++i) {
// Test rate against a threshold value and variance against a threshold.
// Increasing segment number (higher variance and complexity) = higher Q.
if ((projected_rate <
target_rate * aq_c_transitions[aq_strength][i]) &&
(logvar < (low_var_thresh + aq_c_var_thresholds[aq_strength][i]))) {
segment = i;
break;
}
}
}
// Fill in the entires in the segment map corresponding to this SB64.
for (y = 0; y < ymis; y++) {
for (x = 0; x < xmis; x++) {
cpi->segmentation_map[mi_offset + y * cm->mi_cols + x] = segment;
}
}
}

37
vp10/encoder/vp9_aq_complexity.h Normal file
View File

@ -0,0 +1,37 @@
/*
* Copyright (c) 2014 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 VP9_ENCODER_VP9_AQ_COMPLEXITY_H_
#define VP9_ENCODER_VP9_AQ_COMPLEXITY_H_
#ifdef __cplusplus
extern "C" {
#endif
#include "vp10/common/vp9_enums.h"
struct VP9_COMP;
struct macroblock;
// Select a segment for the current Block.
void vp10_caq_select_segment(struct VP9_COMP *cpi, struct macroblock *,
BLOCK_SIZE bs,
int mi_row, int mi_col, int projected_rate);
// This function sets up a set of segments with delta Q values around
// the baseline frame quantizer.
void vp10_setup_in_frame_q_adj(struct VP9_COMP *cpi);
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_ENCODER_VP9_AQ_COMPLEXITY_H_

569
vp10/encoder/vp9_aq_cyclicrefresh.c Normal file
View File

@ -0,0 +1,569 @@
/*
* Copyright (c) 2014 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 <limits.h>
#include <math.h>
#include "vp10/encoder/vp9_aq_cyclicrefresh.h"
#include "vp10/common/vp9_seg_common.h"
#include "vp10/encoder/vp9_ratectrl.h"
#include "vp10/encoder/vp9_segmentation.h"
struct CYCLIC_REFRESH {
// Percentage of blocks per frame that are targeted as candidates
// for cyclic refresh.
int percent_refresh;
// Maximum q-delta as percentage of base q.
int max_qdelta_perc;
// Superblock starting index for cycling through the frame.
int sb_index;
// Controls how long block will need to wait to be refreshed again, in
// excess of the cycle time, i.e., in the case of all zero motion, block
// will be refreshed every (100/percent_refresh + time_for_refresh) frames.
int time_for_refresh;
// Target number of (8x8) blocks that are set for delta-q.
int target_num_seg_blocks;
// Actual number of (8x8) blocks that were applied delta-q.
int actual_num_seg1_blocks;
int actual_num_seg2_blocks;
// RD mult. parameters for segment 1.
int rdmult;
// Cyclic refresh map.
signed char *map;
// Map of the last q a block was coded at.
uint8_t *last_coded_q_map;
// Thresholds applied to the projected rate/distortion of the coding block,
// when deciding whether block should be refreshed.
int64_t thresh_rate_sb;
int64_t thresh_dist_sb;
// Threshold applied to the motion vector (in units of 1/8 pel) of the
// coding block, when deciding whether block should be refreshed.
int16_t motion_thresh;
// Rate target ratio to set q delta.
double rate_ratio_qdelta;
// Boost factor for rate target ratio, for segment CR_SEGMENT_ID_BOOST2.
int rate_boost_fac;
double low_content_avg;
int qindex_delta[3];
};
CYCLIC_REFRESH *vp10_cyclic_refresh_alloc(int mi_rows, int mi_cols) {
size_t last_coded_q_map_size;
CYCLIC_REFRESH *const cr = vpx_calloc(1, sizeof(*cr));
if (cr == NULL)
return NULL;
cr->map = vpx_calloc(mi_rows * mi_cols, sizeof(*cr->map));
if (cr->map == NULL) {
vpx_free(cr);
return NULL;
}
last_coded_q_map_size = mi_rows * mi_cols * sizeof(*cr->last_coded_q_map);
cr->last_coded_q_map = vpx_malloc(last_coded_q_map_size);
if (cr->last_coded_q_map == NULL) {
vpx_free(cr);
return NULL;
}
assert(MAXQ <= 255);
memset(cr->last_coded_q_map, MAXQ, last_coded_q_map_size);
return cr;
}
void vp10_cyclic_refresh_free(CYCLIC_REFRESH *cr) {
vpx_free(cr->map);
vpx_free(cr->last_coded_q_map);
vpx_free(cr);
}
// Check if we should turn off cyclic refresh based on bitrate condition.
static int apply_cyclic_refresh_bitrate(const VP9_COMMON *cm,
const RATE_CONTROL *rc) {
// Turn off cyclic refresh if bits available per frame is not sufficiently
// larger than bit cost of segmentation. Segment map bit cost should scale
// with number of seg blocks, so compare available bits to number of blocks.
// Average bits available per frame = avg_frame_bandwidth
// Number of (8x8) blocks in frame = mi_rows * mi_cols;
const float factor = 0.25;
const int number_blocks = cm->mi_rows * cm->mi_cols;
// The condition below corresponds to turning off at target bitrates:
// (at 30fps), ~12kbps for CIF, 36kbps for VGA, 100kps for HD/720p.
// Also turn off at very small frame sizes, to avoid too large fraction of
// superblocks to be refreshed per frame. Threshold below is less than QCIF.
if (rc->avg_frame_bandwidth < factor * number_blocks ||
number_blocks / 64 < 5)
return 0;
else
return 1;
}
// Check if this coding block, of size bsize, should be considered for refresh
// (lower-qp coding). Decision can be based on various factors, such as
// size of the coding block (i.e., below min_block size rejected), coding
// mode, and rate/distortion.
static int candidate_refresh_aq(const CYCLIC_REFRESH *cr,
const MB_MODE_INFO *mbmi,
int64_t rate,
int64_t dist,
int bsize) {
MV mv = mbmi->mv[0].as_mv;
// Reject the block for lower-qp coding if projected distortion
// is above the threshold, and any of the following is true:
// 1) mode uses large mv
// 2) mode is an intra-mode
// Otherwise accept for refresh.
if (dist > cr->thresh_dist_sb &&
(mv.row > cr->motion_thresh || mv.row < -cr->motion_thresh ||
mv.col > cr->motion_thresh || mv.col < -cr->motion_thresh ||
!is_inter_block(mbmi)))
return CR_SEGMENT_ID_BASE;
else if (bsize >= BLOCK_16X16 &&
rate < cr->thresh_rate_sb &&
is_inter_block(mbmi) &&
mbmi->mv[0].as_int == 0 &&
cr->rate_boost_fac > 10)
// More aggressive delta-q for bigger blocks with zero motion.
return CR_SEGMENT_ID_BOOST2;
else
return CR_SEGMENT_ID_BOOST1;
}
// Compute delta-q for the segment.
static int compute_deltaq(const VP9_COMP *cpi, int q, double rate_factor) {
const CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
const RATE_CONTROL *const rc = &cpi->rc;
int deltaq = vp10_compute_qdelta_by_rate(rc, cpi->common.frame_type,
q, rate_factor,
cpi->common.bit_depth);
if ((-deltaq) > cr->max_qdelta_perc * q / 100) {
deltaq = -cr->max_qdelta_perc * q / 100;
}
return deltaq;
}
// For the just encoded frame, estimate the bits, incorporating the delta-q
// from non-base segment. For now ignore effect of multiple segments
// (with different delta-q). Note this function is called in the postencode
// (called from rc_update_rate_correction_factors()).
int vp10_cyclic_refresh_estimate_bits_at_q(const VP9_COMP *cpi,
double correction_factor) {
const VP9_COMMON *const cm = &cpi->common;
const CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
int estimated_bits;
int mbs = cm->MBs;
int num8x8bl = mbs << 2;
// Weight for non-base segments: use actual number of blocks refreshed in
// previous/just encoded frame. Note number of blocks here is in 8x8 units.
double weight_segment1 = (double)cr->actual_num_seg1_blocks / num8x8bl;
double weight_segment2 = (double)cr->actual_num_seg2_blocks / num8x8bl;
// Take segment weighted average for estimated bits.
estimated_bits = (int)((1.0 - weight_segment1 - weight_segment2) *
vp10_estimate_bits_at_q(cm->frame_type, cm->base_qindex, mbs,
correction_factor, cm->bit_depth) +
weight_segment1 *
vp10_estimate_bits_at_q(cm->frame_type,
cm->base_qindex + cr->qindex_delta[1], mbs,
correction_factor, cm->bit_depth) +
weight_segment2 *
vp10_estimate_bits_at_q(cm->frame_type,
cm->base_qindex + cr->qindex_delta[2], mbs,
correction_factor, cm->bit_depth));
return estimated_bits;
}
// Prior to encoding the frame, estimate the bits per mb, for a given q = i and
// a corresponding delta-q (for segment 1). This function is called in the
// rc_regulate_q() to set the base qp index.
// Note: the segment map is set to either 0/CR_SEGMENT_ID_BASE (no refresh) or
// to 1/CR_SEGMENT_ID_BOOST1 (refresh) for each superblock, prior to encoding.
int vp10_cyclic_refresh_rc_bits_per_mb(const VP9_COMP *cpi, int i,
double correction_factor) {
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
int bits_per_mb;
int num8x8bl = cm->MBs << 2;
// Weight for segment prior to encoding: take the average of the target
// number for the frame to be encoded and the actual from the previous frame.
double weight_segment = (double)((cr->target_num_seg_blocks +
cr->actual_num_seg1_blocks + cr->actual_num_seg2_blocks) >> 1) /
num8x8bl;
// Compute delta-q corresponding to qindex i.
int deltaq = compute_deltaq(cpi, i, cr->rate_ratio_qdelta);
// Take segment weighted average for bits per mb.
bits_per_mb = (int)((1.0 - weight_segment) *
vp10_rc_bits_per_mb(cm->frame_type, i, correction_factor, cm->bit_depth) +
weight_segment *
vp10_rc_bits_per_mb(cm->frame_type, i + deltaq, correction_factor,
cm->bit_depth));
return bits_per_mb;
}
// Prior to coding a given prediction block, of size bsize at (mi_row, mi_col),
// check if we should reset the segment_id, and update the cyclic_refresh map
// and segmentation map.
void vp10_cyclic_refresh_update_segment(VP9_COMP *const cpi,
MB_MODE_INFO *const mbmi,
int mi_row, int mi_col,
BLOCK_SIZE bsize,
int64_t rate,
int64_t dist,
int skip) {
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
const int bw = num_8x8_blocks_wide_lookup[bsize];
const int bh = num_8x8_blocks_high_lookup[bsize];
const int xmis = MIN(cm->mi_cols - mi_col, bw);
const int ymis = MIN(cm->mi_rows - mi_row, bh);
const int block_index = mi_row * cm->mi_cols + mi_col;
const int refresh_this_block = candidate_refresh_aq(cr, mbmi, rate, dist,
bsize);
// Default is to not update the refresh map.
int new_map_value = cr->map[block_index];
int x = 0; int y = 0;
// If this block is labeled for refresh, check if we should reset the
// segment_id.
if (cyclic_refresh_segment_id_boosted(mbmi->segment_id)) {
mbmi->segment_id = refresh_this_block;
// Reset segment_id if will be skipped.
if (skip)
mbmi->segment_id = CR_SEGMENT_ID_BASE;
}
// Update the cyclic refresh map, to be used for setting segmentation map
// for the next frame. If the block will be refreshed this frame, mark it
// as clean. The magnitude of the -ve influences how long before we consider
// it for refresh again.
if (cyclic_refresh_segment_id_boosted(mbmi->segment_id)) {
new_map_value = -cr->time_for_refresh;
} else if (refresh_this_block) {
// Else if it is accepted as candidate for refresh, and has not already
// been refreshed (marked as 1) then mark it as a candidate for cleanup
// for future time (marked as 0), otherwise don't update it.
if (cr->map[block_index] == 1)
new_map_value = 0;
} else {
// Leave it marked as block that is not candidate for refresh.
new_map_value = 1;
}
// Update entries in the cyclic refresh map with new_map_value, and
// copy mbmi->segment_id into global segmentation map.
for (y = 0; y < ymis; y++)
for (x = 0; x < xmis; x++) {
int map_offset = block_index + y * cm->mi_cols + x;
cr->map[map_offset] = new_map_value;
cpi->segmentation_map[map_offset] = mbmi->segment_id;
// Inter skip blocks were clearly not coded at the current qindex, so
// don't update the map for them. For cases where motion is non-zero or
// the reference frame isn't the previous frame, the previous value in
// the map for this spatial location is not entirely correct.
if (!is_inter_block(mbmi) || !skip)
cr->last_coded_q_map[map_offset] = clamp(
cm->base_qindex + cr->qindex_delta[mbmi->segment_id], 0, MAXQ);
}
}
// Update the actual number of blocks that were applied the segment delta q.
void vp10_cyclic_refresh_postencode(VP9_COMP *const cpi) {
VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
unsigned char *const seg_map = cpi->segmentation_map;
int mi_row, mi_col;
cr->actual_num_seg1_blocks = 0;
cr->actual_num_seg2_blocks = 0;
for (mi_row = 0; mi_row < cm->mi_rows; mi_row++)
for (mi_col = 0; mi_col < cm->mi_cols; mi_col++) {
if (cyclic_refresh_segment_id(
seg_map[mi_row * cm->mi_cols + mi_col]) == CR_SEGMENT_ID_BOOST1)
cr->actual_num_seg1_blocks++;
else if (cyclic_refresh_segment_id(
seg_map[mi_row * cm->mi_cols + mi_col]) == CR_SEGMENT_ID_BOOST2)
cr->actual_num_seg2_blocks++;
}
}
// Set golden frame update interval, for non-svc 1 pass CBR mode.
void vp10_cyclic_refresh_set_golden_update(VP9_COMP *const cpi) {
RATE_CONTROL *const rc = &cpi->rc;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
// Set minimum gf_interval for GF update to a multiple (== 2) of refresh
// period. Depending on past encoding stats, GF flag may be reset and update
// may not occur until next baseline_gf_interval.
if (cr->percent_refresh > 0)
rc->baseline_gf_interval = 4 * (100 / cr->percent_refresh);
else
rc->baseline_gf_interval = 40;
}
// Update some encoding stats (from the just encoded frame). If this frame's
// background has high motion, refresh the golden frame. Otherwise, if the
// golden reference is to be updated check if we should NOT update the golden
// ref.
void vp10_cyclic_refresh_check_golden_update(VP9_COMP *const cpi) {
VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
int mi_row, mi_col;
double fraction_low = 0.0;
int low_content_frame = 0;
MODE_INFO **mi = cm->mi_grid_visible;
RATE_CONTROL *const rc = &cpi->rc;
const int rows = cm->mi_rows, cols = cm->mi_cols;
int cnt1 = 0, cnt2 = 0;
int force_gf_refresh = 0;
for (mi_row = 0; mi_row < rows; mi_row++) {
for (mi_col = 0; mi_col < cols; mi_col++) {
int16_t abs_mvr = mi[0]->mbmi.mv[0].as_mv.row >= 0 ?
mi[0]->mbmi.mv[0].as_mv.row : -1 * mi[0]->mbmi.mv[0].as_mv.row;
int16_t abs_mvc = mi[0]->mbmi.mv[0].as_mv.col >= 0 ?
mi[0]->mbmi.mv[0].as_mv.col : -1 * mi[0]->mbmi.mv[0].as_mv.col;
// Calculate the motion of the background.
if (abs_mvr <= 16 && abs_mvc <= 16) {
cnt1++;
if (abs_mvr == 0 && abs_mvc == 0)
cnt2++;
}
mi++;
// Accumulate low_content_frame.
if (cr->map[mi_row * cols + mi_col] < 1)
low_content_frame++;
}
mi += 8;
}
// For video conference clips, if the background has high motion in current
// frame because of the camera movement, set this frame as the golden frame.
// Use 70% and 5% as the thresholds for golden frame refreshing.
// Also, force this frame as a golden update frame if this frame will change
// the resolution (resize_pending != 0).
if (cpi->resize_pending != 0 ||
(cnt1 * 10 > (70 * rows * cols) && cnt2 * 20 < cnt1)) {
vp10_cyclic_refresh_set_golden_update(cpi);
rc->frames_till_gf_update_due = rc->baseline_gf_interval;
if (rc->frames_till_gf_update_due > rc->frames_to_key)
rc->frames_till_gf_update_due = rc->frames_to_key;
cpi->refresh_golden_frame = 1;
force_gf_refresh = 1;
}
fraction_low =
(double)low_content_frame / (rows * cols);
// Update average.
cr->low_content_avg = (fraction_low + 3 * cr->low_content_avg) / 4;
if (!force_gf_refresh && cpi->refresh_golden_frame == 1) {
// Don't update golden reference if the amount of low_content for the
// current encoded frame is small, or if the recursive average of the
// low_content over the update interval window falls below threshold.
if (fraction_low < 0.8 || cr->low_content_avg < 0.7)
cpi->refresh_golden_frame = 0;
// Reset for next internal.
cr->low_content_avg = fraction_low;
}
}
// Update the segmentation map, and related quantities: cyclic refresh map,
// refresh sb_index, and target number of blocks to be refreshed.
// The map is set to either 0/CR_SEGMENT_ID_BASE (no refresh) or to
// 1/CR_SEGMENT_ID_BOOST1 (refresh) for each superblock.
// Blocks labeled as BOOST1 may later get set to BOOST2 (during the
// encoding of the superblock).
static void cyclic_refresh_update_map(VP9_COMP *const cpi) {
VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
unsigned char *const seg_map = cpi->segmentation_map;
int i, block_count, bl_index, sb_rows, sb_cols, sbs_in_frame;
int xmis, ymis, x, y;
memset(seg_map, CR_SEGMENT_ID_BASE, cm->mi_rows * cm->mi_cols);
sb_cols = (cm->mi_cols + MI_BLOCK_SIZE - 1) / MI_BLOCK_SIZE;
sb_rows = (cm->mi_rows + MI_BLOCK_SIZE - 1) / MI_BLOCK_SIZE;
sbs_in_frame = sb_cols * sb_rows;
// Number of target blocks to get the q delta (segment 1).
block_count = cr->percent_refresh * cm->mi_rows * cm->mi_cols / 100;
// Set the segmentation map: cycle through the superblocks, starting at
// cr->mb_index, and stopping when either block_count blocks have been found
// to be refreshed, or we have passed through whole frame.
assert(cr->sb_index < sbs_in_frame);
i = cr->sb_index;
cr->target_num_seg_blocks = 0;
do {
int sum_map = 0;
// Get the mi_row/mi_col corresponding to superblock index i.
int sb_row_index = (i / sb_cols);
int sb_col_index = i - sb_row_index * sb_cols;
int mi_row = sb_row_index * MI_BLOCK_SIZE;
int mi_col = sb_col_index * MI_BLOCK_SIZE;
int qindex_thresh =
cpi->oxcf.content == VP9E_CONTENT_SCREEN
? vp10_get_qindex(&cm->seg, CR_SEGMENT_ID_BOOST2, cm->base_qindex)
: 0;
assert(mi_row >= 0 && mi_row < cm->mi_rows);
assert(mi_col >= 0 && mi_col < cm->mi_cols);
bl_index = mi_row * cm->mi_cols + mi_col;
// Loop through all 8x8 blocks in superblock and update map.
xmis = MIN(cm->mi_cols - mi_col,
num_8x8_blocks_wide_lookup[BLOCK_64X64]);
ymis = MIN(cm->mi_rows - mi_row,
num_8x8_blocks_high_lookup[BLOCK_64X64]);
for (y = 0; y < ymis; y++) {
for (x = 0; x < xmis; x++) {
const int bl_index2 = bl_index + y * cm->mi_cols + x;
// If the block is as a candidate for clean up then mark it
// for possible boost/refresh (segment 1). The segment id may get
// reset to 0 later if block gets coded anything other than ZEROMV.
if (cr->map[bl_index2] == 0) {
if (cr->last_coded_q_map[bl_index2] > qindex_thresh)
sum_map++;
} else if (cr->map[bl_index2] < 0) {
cr->map[bl_index2]++;
}
}
}
// Enforce constant segment over superblock.
// If segment is at least half of superblock, set to 1.
if (sum_map >= xmis * ymis / 2) {
for (y = 0; y < ymis; y++)
for (x = 0; x < xmis; x++) {
seg_map[bl_index + y * cm->mi_cols + x] = CR_SEGMENT_ID_BOOST1;
}
cr->target_num_seg_blocks += xmis * ymis;
}
i++;
if (i == sbs_in_frame) {
i = 0;
}
} while (cr->target_num_seg_blocks < block_count && i != cr->sb_index);
cr->sb_index = i;
}
// Set cyclic refresh parameters.
void vp10_cyclic_refresh_update_parameters(VP9_COMP *const cpi) {
const RATE_CONTROL *const rc = &cpi->rc;
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
cr->percent_refresh = 10;
cr->max_qdelta_perc = 50;
cr->time_for_refresh = 0;
// Use larger delta-qp (increase rate_ratio_qdelta) for first few (~4)
// periods of the refresh cycle, after a key frame.
if (rc->frames_since_key < 4 * cr->percent_refresh)
cr->rate_ratio_qdelta = 3.0;
else
cr->rate_ratio_qdelta = 2.0;
// Adjust some parameters for low resolutions at low bitrates.
if (cm->width <= 352 &&
cm->height <= 288 &&
rc->avg_frame_bandwidth < 3400) {
cr->motion_thresh = 4;
cr->rate_boost_fac = 10;
} else {
cr->motion_thresh = 32;
cr->rate_boost_fac = 17;
}
}
// Setup cyclic background refresh: set delta q and segmentation map.
void vp10_cyclic_refresh_setup(VP9_COMP *const cpi) {
VP9_COMMON *const cm = &cpi->common;
const RATE_CONTROL *const rc = &cpi->rc;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
struct segmentation *const seg = &cm->seg;
const int apply_cyclic_refresh = apply_cyclic_refresh_bitrate(cm, rc);
if (cm->current_video_frame == 0)
cr->low_content_avg = 0.0;
// Don't apply refresh on key frame or enhancement layer frames.
if (!apply_cyclic_refresh ||
(cm->frame_type == KEY_FRAME) ||
(cpi->svc.temporal_layer_id > 0) ||
(cpi->svc.spatial_layer_id > 0)) {
// Set segmentation map to 0 and disable.
unsigned char *const seg_map = cpi->segmentation_map;
memset(seg_map, 0, cm->mi_rows * cm->mi_cols);
vp10_disable_segmentation(&cm->seg);
if (cm->frame_type == KEY_FRAME) {
memset(cr->last_coded_q_map, MAXQ,
cm->mi_rows * cm->mi_cols * sizeof(*cr->last_coded_q_map));
cr->sb_index = 0;
}
return;
} else {
int qindex_delta = 0;
int qindex2;
const double q = vp10_convert_qindex_to_q(cm->base_qindex, cm->bit_depth);
vpx_clear_system_state();
// Set rate threshold to some multiple (set to 2 for now) of the target
// rate (target is given by sb64_target_rate and scaled by 256).
cr->thresh_rate_sb = ((int64_t)(rc->sb64_target_rate) << 8) << 2;
// Distortion threshold, quadratic in Q, scale factor to be adjusted.
// q will not exceed 457, so (q * q) is within 32bit; see:
// vp10_convert_qindex_to_q(), vp10_ac_quant(), ac_qlookup*[].
cr->thresh_dist_sb = ((int64_t)(q * q)) << 2;
// Set up segmentation.
// Clear down the segment map.
vp10_enable_segmentation(&cm->seg);
vp10_clearall_segfeatures(seg);
// Select delta coding method.
seg->abs_delta = SEGMENT_DELTADATA;
// Note: setting temporal_update has no effect, as the seg-map coding method
// (temporal or spatial) is determined in vp10_choose_segmap_coding_method(),
// based on the coding cost of each method. For error_resilient mode on the
// last_frame_seg_map is set to 0, so if temporal coding is used, it is
// relative to 0 previous map.
// seg->temporal_update = 0;
// Segment BASE "Q" feature is disabled so it defaults to the baseline Q.
vp10_disable_segfeature(seg, CR_SEGMENT_ID_BASE, SEG_LVL_ALT_Q);
// Use segment BOOST1 for in-frame Q adjustment.
vp10_enable_segfeature(seg, CR_SEGMENT_ID_BOOST1, SEG_LVL_ALT_Q);
// Use segment BOOST2 for more aggressive in-frame Q adjustment.
vp10_enable_segfeature(seg, CR_SEGMENT_ID_BOOST2, SEG_LVL_ALT_Q);
// Set the q delta for segment BOOST1.
qindex_delta = compute_deltaq(cpi, cm->base_qindex, cr->rate_ratio_qdelta);
cr->qindex_delta[1] = qindex_delta;
// Compute rd-mult for segment BOOST1.
qindex2 = clamp(cm->base_qindex + cm->y_dc_delta_q + qindex_delta, 0, MAXQ);
cr->rdmult = vp10_compute_rd_mult(cpi, qindex2);
vp10_set_segdata(seg, CR_SEGMENT_ID_BOOST1, SEG_LVL_ALT_Q, qindex_delta);
// Set a more aggressive (higher) q delta for segment BOOST2.
qindex_delta = compute_deltaq(
cpi, cm->base_qindex, MIN(CR_MAX_RATE_TARGET_RATIO,
0.1 * cr->rate_boost_fac * cr->rate_ratio_qdelta));
cr->qindex_delta[2] = qindex_delta;
vp10_set_segdata(seg, CR_SEGMENT_ID_BOOST2, SEG_LVL_ALT_Q, qindex_delta);
// Update the segmentation and refresh map.
cyclic_refresh_update_map(cpi);
}
}
int vp10_cyclic_refresh_get_rdmult(const CYCLIC_REFRESH *cr) {
return cr->rdmult;
}
void vp10_cyclic_refresh_reset_resize(VP9_COMP *const cpi) {
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
memset(cr->map, 0, cm->mi_rows * cm->mi_cols);
cr->sb_index = 0;
cpi->refresh_golden_frame = 1;
}

Some files were not shown because too many files have changed in this diff Show More