Merge "vp9/10 encoder: prevent NULL access on failure"
diff --git a/build/make/configure.sh b/build/make/configure.sh
index 9c30441..c592b63 100644
--- a/build/make/configure.sh
+++ b/build/make/configure.sh
@@ -201,6 +201,10 @@
eval test "x\$$1" = "xno"
}
+# Iterates through positional parameters, checks to confirm the parameter has
+# not been explicitly (force) disabled, and enables the setting controlled by
+# the parameter when the setting is not disabled.
+# Note: Does NOT alter RTCD generation options ($RTCD_OPTIONS).
soft_enable() {
for var in $*; do
if ! disabled $var; then
@@ -210,6 +214,10 @@
done
}
+# Iterates through positional parameters, checks to confirm the parameter has
+# not been explicitly (force) enabled, and disables the setting controlled by
+# the parameter when the setting is not enabled.
+# Note: Does NOT alter RTCD generation options ($RTCD_OPTIONS).
soft_disable() {
for var in $*; do
if ! enabled $var; then
@@ -625,6 +633,11 @@
xcodebuild -sdk $1 -version Path 2>/dev/null
}
+# Print the major version number of the Darwin SDK specified by $1.
+show_darwin_sdk_major_version() {
+ xcrun --sdk $1 --show-sdk-version 2>/dev/null | cut -d. -f1
+}
+
process_common_toolchain() {
if [ -z "$toolchain" ]; then
gcctarget="${CHOST:-$(gcc -dumpmachine 2> /dev/null)}"
@@ -736,7 +749,15 @@
# Handle darwin variants. Newer SDKs allow targeting older
# platforms, so use the newest one available.
case ${toolchain} in
- *-darwin*)
+ arm*-darwin*)
+ add_cflags "-miphoneos-version-min=${IOS_VERSION_MIN}"
+ iphoneos_sdk_dir="$(show_darwin_sdk_path iphoneos)"
+ if [ -d "${iphoneos_sdk_dir}" ]; then
+ add_cflags "-isysroot ${iphoneos_sdk_dir}"
+ add_ldflags "-isysroot ${iphoneos_sdk_dir}"
+ fi
+ ;;
+ x86*-darwin*)
osx_sdk_dir="$(show_darwin_sdk_path macosx)"
if [ -d "${osx_sdk_dir}" ]; then
add_cflags "-isysroot ${osx_sdk_dir}"
@@ -811,10 +832,36 @@
if disabled neon && enabled neon_asm; then
die "Disabling neon while keeping neon-asm is not supported"
fi
- soft_enable media
+ case ${toolchain} in
+ # Apple iOS SDKs no longer support armv6 as of the version 9
+ # release (coincides with release of Xcode 7). Only enable media
+ # when using earlier SDK releases.
+ *-darwin*)
+ if [ "$(show_darwin_sdk_major_version iphoneos)" -lt 9 ]; then
+ soft_enable media
+ else
+ soft_disable media
+ RTCD_OPTIONS="${RTCD_OPTIONS}--disable-media "
+ fi
+ ;;
+ *)
+ soft_enable media
+ ;;
+ esac
;;
armv6)
- soft_enable media
+ case ${toolchain} in
+ *-darwin*)
+ if [ "$(show_darwin_sdk_major_version iphoneos)" -lt 9 ]; then
+ soft_enable media
+ else
+ die "Your iOS SDK does not support armv6."
+ fi
+ ;;
+ *)
+ soft_enable media
+ ;;
+ esac
;;
esac
@@ -997,6 +1044,12 @@
done
asm_conversion_cmd="${source_path}/build/make/ads2gas_apple.pl"
+
+ if [ "$(show_darwin_sdk_major_version iphoneos)" -gt 8 ]; then
+ check_add_cflags -fembed-bitcode
+ check_add_asflags -fembed-bitcode
+ check_add_ldflags -fembed-bitcode
+ fi
;;
linux*)
@@ -1167,7 +1220,8 @@
&& AS=""
fi
[ "${AS}" = auto ] || [ -z "${AS}" ] \
- && die "Neither yasm nor nasm have been found"
+ && die "Neither yasm nor nasm have been found." \
+ "See the prerequisites section in the README for more info."
;;
esac
log_echo " using $AS"
@@ -1206,6 +1260,13 @@
enabled x86 && sim_arch="-arch i386" || sim_arch="-arch x86_64"
add_cflags ${sim_arch}
add_ldflags ${sim_arch}
+
+ if [ "$(show_darwin_sdk_major_version iphonesimulator)" -gt 8 ]; then
+ # yasm v1.3.0 doesn't know what -fembed-bitcode means, so turning it
+ # on is pointless (unless building a C-only lib). Warn the user, but
+ # do nothing here.
+ log "Warning: Bitcode embed disabled for simulator targets."
+ fi
;;
os2)
add_asflags -f aout
diff --git a/test/resize_test.cc b/test/resize_test.cc
index a86c9d1..98b6f87 100644
--- a/test/resize_test.cc
+++ b/test/resize_test.cc
@@ -196,13 +196,27 @@
virtual void PreEncodeFrameHook(libvpx_test::VideoSource *video,
libvpx_test::Encoder *encoder) {
- if (video->frame() == kStepDownFrame) {
- struct vpx_scaling_mode mode = {VP8E_FOURFIVE, VP8E_THREEFIVE};
- encoder->Control(VP8E_SET_SCALEMODE, &mode);
- }
- if (video->frame() == kStepUpFrame) {
- struct vpx_scaling_mode mode = {VP8E_NORMAL, VP8E_NORMAL};
- encoder->Control(VP8E_SET_SCALEMODE, &mode);
+ if (change_config_) {
+ int new_q = 60;
+ if (video->frame() == 0) {
+ struct vpx_scaling_mode mode = {VP8E_ONETWO, VP8E_ONETWO};
+ encoder->Control(VP8E_SET_SCALEMODE, &mode);
+ }
+ if (video->frame() == 1) {
+ struct vpx_scaling_mode mode = {VP8E_NORMAL, VP8E_NORMAL};
+ encoder->Control(VP8E_SET_SCALEMODE, &mode);
+ cfg_.rc_min_quantizer = cfg_.rc_max_quantizer = new_q;
+ encoder->Config(&cfg_);
+ }
+ } else {
+ if (video->frame() == kStepDownFrame) {
+ struct vpx_scaling_mode mode = {VP8E_FOURFIVE, VP8E_THREEFIVE};
+ encoder->Control(VP8E_SET_SCALEMODE, &mode);
+ }
+ if (video->frame() == kStepUpFrame) {
+ struct vpx_scaling_mode mode = {VP8E_NORMAL, VP8E_NORMAL};
+ encoder->Control(VP8E_SET_SCALEMODE, &mode);
+ }
}
}
@@ -227,6 +241,7 @@
#endif
double frame0_psnr_;
+ bool change_config_;
#if WRITE_COMPRESSED_STREAM
FILE *outfile_;
unsigned int out_frames_;
@@ -237,6 +252,7 @@
::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
30, 1, 0, 10);
init_flags_ = VPX_CODEC_USE_PSNR;
+ change_config_ = false;
// q picked such that initial keyframe on this clip is ~30dB PSNR
cfg_.rc_min_quantizer = cfg_.rc_max_quantizer = 48;
@@ -261,6 +277,15 @@
}
}
+TEST_P(ResizeInternalTest, TestInternalResizeChangeConfig) {
+ ::libvpx_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+ 30, 1, 0, 10);
+ cfg_.g_w = 352;
+ cfg_.g_h = 288;
+ change_config_ = true;
+ ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
class ResizeInternalRealtimeTest : public ::libvpx_test::EncoderTest,
public ::libvpx_test::CodecTestWith2Params<libvpx_test::TestMode, int> {
protected:
diff --git a/test/test.mk b/test/test.mk
index fde9703..bb5186b 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -168,7 +168,7 @@
TEST_INTRA_PRED_SPEED_SRCS-$(CONFIG_VP9) += ../md5_utils.h ../md5_utils.c
## VP10
-LIBVPX_TEST_SRCS-$(CONFIG_VP10) += vp10_dct_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP10_ENCODER) += vp10_dct_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP10) += vp10_inv_txfm_test.cc
endif # CONFIG_SHARED
diff --git a/test/vp10_dct_test.cc b/test/vp10_dct_test.cc
index 8fb5f4f..8e49609 100644
--- a/test/vp10_dct_test.cc
+++ b/test/vp10_dct_test.cc
@@ -17,7 +17,10 @@
#include "test/util.h"
#include "./vpx_config.h"
#include "vpx_ports/msvc.h"
-#include "vp10/encoder/dct.h"
+
+#undef CONFIG_COEFFICIENT_RANGE_CHECKING
+#define CONFIG_COEFFICIENT_RANGE_CHECKING 1
+#include "vp10/encoder/dct.c"
using libvpx_test::ACMRandom;
@@ -102,8 +105,8 @@
INSTANTIATE_TEST_CASE_P(
C, Vp10FwdTxfm,
::testing::Values(
- FdctParam(&vp10_fdct4, &reference_dct_1d, 4, 1),
- FdctParam(&vp10_fdct8, &reference_dct_1d, 8, 1),
- FdctParam(&vp10_fdct16, &reference_dct_1d, 16, 2),
- FdctParam(&vp10_fdct32_local, &reference_dct_1d, 32, 4)));
+ FdctParam(&fdct4, &reference_dct_1d, 4, 1),
+ FdctParam(&fdct8, &reference_dct_1d, 8, 1),
+ FdctParam(&fdct16, &reference_dct_1d, 16, 2),
+ FdctParam(&fdct32, &reference_dct_1d, 32, 4)));
} // namespace
diff --git a/vp10/common/idct.c b/vp10/common/idct.c
index 144afc3..5ee15c8 100644
--- a/vp10/common/idct.c
+++ b/vp10/common/idct.c
@@ -179,21 +179,24 @@
}
void vp10_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
- int stride, int eob, TX_TYPE tx_type,
- void (*itxm_add_4x4)(const tran_low_t *input,
- uint8_t *dest, int stride, int eob)) {
- switch (tx_type) {
- case DCT_DCT:
- itxm_add_4x4(input, dest, stride, eob);
- break;
- case ADST_DCT:
- case DCT_ADST:
- case ADST_ADST:
- vp10_iht4x4_16_add(input, dest, stride, tx_type);
- break;
- default:
- assert(0);
- break;
+ int stride, int eob, TX_TYPE tx_type, int lossless) {
+ if (lossless) {
+ assert(tx_type == DCT_DCT);
+ vp10_iwht4x4_add(input, dest, stride, eob);
+ } else {
+ switch (tx_type) {
+ case DCT_DCT:
+ vp10_idct4x4_add(input, dest, stride, eob);
+ break;
+ case ADST_DCT:
+ case DCT_ADST:
+ case ADST_ADST:
+ vp10_iht4x4_16_add(input, dest, stride, tx_type);
+ break;
+ default:
+ assert(0);
+ break;
+ }
}
}
@@ -418,21 +421,24 @@
void vp10_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type,
- void (*highbd_itxm_add_4x4)
- (const tran_low_t *input, uint8_t *dest,
- int stride, int eob, int bd)) {
- switch (tx_type) {
- case DCT_DCT:
- highbd_itxm_add_4x4(input, dest, stride, eob, bd);
- break;
- case ADST_DCT:
- case DCT_ADST:
- case ADST_ADST:
- vp10_highbd_iht4x4_16_add(input, dest, stride, tx_type, bd);
- break;
- default:
- assert(0);
- break;
+ int lossless) {
+ if (lossless) {
+ assert(tx_type == DCT_DCT);
+ vp10_highbd_iwht4x4_add(input, dest, stride, eob, bd);
+ } else {
+ switch (tx_type) {
+ case DCT_DCT:
+ vp10_highbd_idct4x4_add(input, dest, stride, eob, bd);
+ break;
+ case ADST_DCT:
+ case DCT_ADST:
+ case ADST_ADST:
+ vp10_highbd_iht4x4_16_add(input, dest, stride, tx_type, bd);
+ break;
+ default:
+ assert(0);
+ break;
+ }
}
}
diff --git a/vp10/common/idct.h b/vp10/common/idct.h
index 2e00052..0883398 100644
--- a/vp10/common/idct.h
+++ b/vp10/common/idct.h
@@ -44,9 +44,7 @@
int eob);
void vp10_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
- int stride, int eob, TX_TYPE tx_type,
- void (*itxm_add_4x4)(const tran_low_t *input,
- uint8_t *dest, int stride, int eob));
+ int stride, int eob, TX_TYPE tx_type, int lossless);
void vp10_inv_txfm_add_8x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type);
void vp10_inv_txfm_add_16x16(const tran_low_t *input, uint8_t *dest,
@@ -67,9 +65,7 @@
int stride, int eob, int bd);
void vp10_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type,
- void (*highbd_itxm_add_4x4)
- (const tran_low_t *input, uint8_t *dest,
- int stride, int eob, int bd));
+ int lossless);
void vp10_highbd_inv_txfm_add_8x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd, TX_TYPE tx_type);
void vp10_highbd_inv_txfm_add_16x16(const tran_low_t *input, uint8_t *dest,
diff --git a/vp10/decoder/decodeframe.c b/vp10/decoder/decodeframe.c
index 50d50f8..bbea6c7 100644
--- a/vp10/decoder/decodeframe.c
+++ b/vp10/decoder/decodeframe.c
@@ -203,9 +203,7 @@
switch (tx_size) {
case TX_4X4:
vp10_highbd_inv_txfm_add_4x4(dqcoeff, dst, stride, eob, xd->bd,
- tx_type, xd->lossless ?
- vp10_highbd_iwht4x4_add :
- vp10_highbd_idct4x4_add);
+ tx_type, xd->lossless);
break;
case TX_8X8:
vp10_highbd_inv_txfm_add_8x8(dqcoeff, dst, stride, eob, xd->bd,
@@ -228,8 +226,7 @@
switch (tx_size) {
case TX_4X4:
vp10_inv_txfm_add_4x4(dqcoeff, dst, stride, eob, tx_type,
- xd->lossless ? vp10_iwht4x4_add :
- vp10_idct4x4_add);
+ xd->lossless);
break;
case TX_8X8:
vp10_inv_txfm_add_8x8(dqcoeff, dst, stride, eob, tx_type);
@@ -274,9 +271,7 @@
switch (tx_size) {
case TX_4X4:
vp10_highbd_inv_txfm_add_4x4(dqcoeff, dst, stride, eob, xd->bd,
- tx_type, xd->lossless ?
- vp10_highbd_iwht4x4_add :
- vp10_highbd_idct4x4_add);
+ tx_type, xd->lossless);
break;
case TX_8X8:
vp10_highbd_inv_txfm_add_8x8(dqcoeff, dst, stride, eob, xd->bd,
@@ -299,8 +294,7 @@
switch (tx_size) {
case TX_4X4:
vp10_inv_txfm_add_4x4(dqcoeff, dst, stride, eob, tx_type,
- xd->lossless ? vp10_iwht4x4_add :
- vp10_idct4x4_add);
+ xd->lossless);
break;
case TX_8X8:
vp10_inv_txfm_add_8x8(dqcoeff, dst, stride, eob, tx_type);
diff --git a/vp10/decoder/decoder.c b/vp10/decoder/decoder.c
index 81bd357..23851af 100644
--- a/vp10/decoder/decoder.c
+++ b/vp10/decoder/decoder.c
@@ -126,6 +126,9 @@
void vp10_decoder_remove(VP10Decoder *pbi) {
int i;
+ if (!pbi)
+ return;
+
vpx_get_worker_interface()->end(&pbi->lf_worker);
vpx_free(pbi->lf_worker.data1);
vpx_free(pbi->tile_data);
diff --git a/vp10/encoder/block.h b/vp10/encoder/block.h
index 9eff31a..cb2a234 100644
--- a/vp10/encoder/block.h
+++ b/vp10/encoder/block.h
@@ -132,13 +132,6 @@
// Strong color activity detection. Used in RTC coding mode to enhance
// the visual quality at the boundary of moving color objects.
uint8_t color_sensitivity[2];
-
- void (*fwd_txm4x4)(const int16_t *input, tran_low_t *output, int stride);
- void (*itxm_add)(const tran_low_t *input, uint8_t *dest, int stride, int eob);
-#if CONFIG_VP9_HIGHBITDEPTH
- void (*highbd_itxm_add)(const tran_low_t *input, uint8_t *dest, int stride,
- int eob, int bd);
-#endif
};
#ifdef __cplusplus
diff --git a/vp10/encoder/dct.c b/vp10/encoder/dct.c
index 78f151e..80ace06 100644
--- a/vp10/encoder/dct.c
+++ b/vp10/encoder/dct.c
@@ -34,7 +34,7 @@
#endif
}
-void vp10_fdct4(const tran_low_t *input, tran_low_t *output) {
+static void fdct4(const tran_low_t *input, tran_low_t *output) {
tran_high_t temp;
tran_low_t step[4];
@@ -70,7 +70,7 @@
range_check(output, 4, 13);
}
-void vp10_fdct8(const tran_low_t *input, tran_low_t *output) {
+static void fdct8(const tran_low_t *input, tran_low_t *output) {
tran_high_t temp;
tran_low_t step[8];
@@ -148,7 +148,7 @@
range_check(output, 8, 14);
}
-void vp10_fdct16(const tran_low_t *input, tran_low_t *output) {
+static void fdct16(const tran_low_t *input, tran_low_t *output) {
tran_high_t temp;
tran_low_t step[16];
@@ -322,8 +322,7 @@
range_check(output, 16, 16);
}
-// TODO(angiebird): Unify this with vp10_fwd_txfm.c: vp10_fdct32
-void vp10_fdct32_local(const tran_low_t *input, tran_low_t *output) {
+static void fdct32(const tran_low_t *input, tran_low_t *output) {
tran_high_t temp;
tran_low_t step[32];
@@ -996,24 +995,24 @@
}
static const transform_2d FHT_4[] = {
- { vp10_fdct4, vp10_fdct4 }, // DCT_DCT = 0
- { fadst4, vp10_fdct4 }, // ADST_DCT = 1
- { vp10_fdct4, fadst4 }, // DCT_ADST = 2
- { fadst4, fadst4 } // ADST_ADST = 3
+ { fdct4, fdct4 }, // DCT_DCT = 0
+ { fadst4, fdct4 }, // ADST_DCT = 1
+ { fdct4, fadst4 }, // DCT_ADST = 2
+ { fadst4, fadst4 } // ADST_ADST = 3
};
static const transform_2d FHT_8[] = {
- { vp10_fdct8, vp10_fdct8 }, // DCT_DCT = 0
- { fadst8, vp10_fdct8 }, // ADST_DCT = 1
- { vp10_fdct8, fadst8 }, // DCT_ADST = 2
- { fadst8, fadst8 } // ADST_ADST = 3
+ { fdct8, fdct8 }, // DCT_DCT = 0
+ { fadst8, fdct8 }, // ADST_DCT = 1
+ { fdct8, fadst8 }, // DCT_ADST = 2
+ { fadst8, fadst8 } // ADST_ADST = 3
};
static const transform_2d FHT_16[] = {
- { vp10_fdct16, vp10_fdct16 }, // DCT_DCT = 0
- { fadst16, vp10_fdct16 }, // ADST_DCT = 1
- { vp10_fdct16, fadst16 }, // DCT_ADST = 2
- { fadst16, fadst16 } // ADST_ADST = 3
+ { fdct16, fdct16 }, // DCT_DCT = 0
+ { fadst16, fdct16 }, // ADST_DCT = 1
+ { fdct16, fadst16 }, // DCT_ADST = 2
+ { fadst16, fadst16 } // ADST_ADST = 3
};
void vp10_fht4x4_c(const int16_t *input, tran_low_t *output,
@@ -1124,7 +1123,7 @@
// Rows
for (i = 0; i < 8; ++i) {
- vp10_fdct8(&intermediate[i * 8], &coeff_ptr[i * 8]);
+ fdct8(&intermediate[i * 8], &coeff_ptr[i * 8]);
for (j = 0; j < 8; ++j)
coeff_ptr[j + i * 8] /= 2;
}
diff --git a/vp10/encoder/dct.h b/vp10/encoder/dct.h
deleted file mode 100644
index ab0db93..0000000
--- a/vp10/encoder/dct.h
+++ /dev/null
@@ -1,28 +0,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-#ifndef VP10_ENCODER_DCT_H_
-#define VP10_ENCODER_DCT_H_
-
-#include "vpx_dsp/vpx_dsp_common.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-void vp10_fdct4(const tran_low_t *input, tran_low_t *output);
-void vp10_fdct8(const tran_low_t *input, tran_low_t *output);
-void vp10_fdct16(const tran_low_t *input, tran_low_t *output);
-void vp10_fdct32_local(const tran_low_t *input, tran_low_t *output);
-
-#ifdef __cplusplus
-} // extern "C"
-#endif
-
-#endif // VP10_ENCODER_DCT_H_
diff --git a/vp10/encoder/encodeframe.c b/vp10/encoder/encodeframe.c
index c3b6265..019e5b1 100644
--- a/vp10/encoder/encodeframe.c
+++ b/vp10/encoder/encodeframe.c
@@ -2696,18 +2696,6 @@
cm->uv_dc_delta_q == 0 &&
cm->uv_ac_delta_q == 0;
-#if CONFIG_VP9_HIGHBITDEPTH
- if (cm->use_highbitdepth)
- x->fwd_txm4x4 = xd->lossless ? vp10_highbd_fwht4x4 : vpx_highbd_fdct4x4;
- else
- x->fwd_txm4x4 = xd->lossless ? vp10_fwht4x4 : vpx_fdct4x4;
- x->highbd_itxm_add = xd->lossless ? vp10_highbd_iwht4x4_add :
- vp10_highbd_idct4x4_add;
-#else
- x->fwd_txm4x4 = xd->lossless ? vp10_fwht4x4 : vpx_fdct4x4;
-#endif // CONFIG_VP9_HIGHBITDEPTH
- x->itxm_add = xd->lossless ? vp10_iwht4x4_add : vp10_idct4x4_add;
-
if (xd->lossless)
x->optimize = 0;
diff --git a/vp10/encoder/encodemb.c b/vp10/encoder/encodemb.c
index 65692dd..3307720 100644
--- a/vp10/encoder/encodemb.c
+++ b/vp10/encoder/encodemb.c
@@ -367,7 +367,11 @@
scan_order->scan, scan_order->iscan);
break;
case TX_4X4:
- x->fwd_txm4x4(src_diff, coeff, diff_stride);
+ if (xd->lossless) {
+ vp10_highbd_fwht4x4(src_diff, coeff, diff_stride);
+ } else {
+ vpx_highbd_fdct4x4(src_diff, coeff, diff_stride);
+ }
vp10_highbd_quantize_fp(coeff, 16, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
@@ -403,7 +407,11 @@
scan_order->scan, scan_order->iscan);
break;
case TX_4X4:
- x->fwd_txm4x4(src_diff, coeff, diff_stride);
+ if (xd->lossless) {
+ vp10_fwht4x4(src_diff, coeff, diff_stride);
+ } else {
+ vpx_fdct4x4(src_diff, coeff, diff_stride);
+ }
vp10_quantize_fp(coeff, 16, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
@@ -453,7 +461,11 @@
pd->dequant[0], eob);
break;
case TX_4X4:
- x->fwd_txm4x4(src_diff, coeff, diff_stride);
+ if (xd->lossless) {
+ vp10_highbd_fwht4x4(src_diff, coeff, diff_stride);
+ } else {
+ vpx_highbd_fdct4x4(src_diff, coeff, diff_stride);
+ }
vpx_highbd_quantize_dc(coeff, 16, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
@@ -485,7 +497,11 @@
pd->dequant[0], eob);
break;
case TX_4X4:
- x->fwd_txm4x4(src_diff, coeff, diff_stride);
+ if (xd->lossless) {
+ vp10_fwht4x4(src_diff, coeff, diff_stride);
+ } else {
+ vpx_fdct4x4(src_diff, coeff, diff_stride);
+ }
vpx_quantize_dc(coeff, 16, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
@@ -496,22 +512,24 @@
}
}
-void vp10_fwd_txfm_4x4(const int16_t *src_diff,
- tran_low_t *coeff, int diff_stride, TX_TYPE tx_type,
- void (*fwd_txm4x4)(const int16_t *input,
- tran_low_t *output, int stride)) {
- switch (tx_type) {
- case DCT_DCT:
- fwd_txm4x4(src_diff, coeff, diff_stride);
- break;
- case ADST_DCT:
- case DCT_ADST:
- case ADST_ADST:
- vp10_fht4x4(src_diff, coeff, diff_stride, tx_type);
- break;
- default:
- assert(0);
- break;
+void vp10_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
+ int diff_stride, TX_TYPE tx_type, int lossless) {
+ if (lossless) {
+ vp10_fwht4x4(src_diff, coeff, diff_stride);
+ } else {
+ switch (tx_type) {
+ case DCT_DCT:
+ vpx_fdct4x4(src_diff, coeff, diff_stride);
+ break;
+ case ADST_DCT:
+ case DCT_ADST:
+ case ADST_ADST:
+ vp10_fht4x4(src_diff, coeff, diff_stride, tx_type);
+ break;
+ default:
+ assert(0);
+ break;
+ }
}
}
@@ -565,21 +583,24 @@
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
- int diff_stride, TX_TYPE tx_type,
- void (*highbd_fwd_txm4x4)(const int16_t *input,
- tran_low_t *output, int stride)) {
- switch (tx_type) {
- case DCT_DCT:
- highbd_fwd_txm4x4(src_diff, coeff, diff_stride);
- break;
- case ADST_DCT:
- case DCT_ADST:
- case ADST_ADST:
- vp10_highbd_fht4x4(src_diff, coeff, diff_stride, tx_type);
- break;
- default:
- assert(0);
- break;
+ int diff_stride, TX_TYPE tx_type, int lossless) {
+ if (lossless) {
+ assert(tx_type == DCT_DCT);
+ vp10_highbd_fwht4x4(src_diff, coeff, diff_stride);
+ } else {
+ switch (tx_type) {
+ case DCT_DCT:
+ vpx_highbd_fdct4x4(src_diff, coeff, diff_stride);
+ break;
+ case ADST_DCT:
+ case DCT_ADST:
+ case ADST_ADST:
+ vp10_highbd_fht4x4(src_diff, coeff, diff_stride, tx_type);
+ break;
+ default:
+ assert(0);
+ break;
+ }
}
}
@@ -681,7 +702,7 @@
break;
case TX_4X4:
vp10_highbd_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type,
- x->fwd_txm4x4);
+ xd->lossless);
vpx_highbd_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
@@ -717,7 +738,7 @@
scan_order->scan, scan_order->iscan);
break;
case TX_4X4:
- vp10_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type, x->fwd_txm4x4);
+ vp10_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type, xd->lossless);
vpx_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
@@ -820,7 +841,7 @@
// case.
vp10_highbd_inv_txfm_add_4x4(dqcoeff, dst, pd->dst.stride,
p->eobs[block], xd->bd, tx_type,
- x->highbd_itxm_add);
+ xd->lossless);
break;
default:
assert(0 && "Invalid transform size");
@@ -849,7 +870,7 @@
// which is significant (not just an optimization) for the lossless
// case.
vp10_inv_txfm_add_4x4(dqcoeff, dst, pd->dst.stride, p->eobs[block],
- tx_type, x->itxm_add);
+ tx_type, xd->lossless);
break;
default:
assert(0 && "Invalid transform size");
@@ -874,11 +895,21 @@
if (p->eobs[block] > 0) {
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
- x->highbd_itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block], xd->bd);
- return;
+ if (xd->lossless) {
+ vp10_highbd_iwht4x4_add(dqcoeff, dst, pd->dst.stride,
+ p->eobs[block], xd->bd);
+ } else {
+ vp10_highbd_idct4x4_add(dqcoeff, dst, pd->dst.stride,
+ p->eobs[block], xd->bd);
+ }
+ return;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
- x->itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
+ if (xd->lossless) {
+ vp10_iwht4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
+ } else {
+ vp10_idct4x4_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
+ }
}
}
@@ -999,7 +1030,7 @@
vpx_highbd_subtract_block(4, 4, src_diff, diff_stride,
src, src_stride, dst, dst_stride, xd->bd);
vp10_highbd_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type,
- x->fwd_txm4x4);
+ xd->lossless);
vpx_highbd_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
@@ -1011,7 +1042,7 @@
// eob<=1 which is significant (not just an optimization) for the
// lossless case.
vp10_highbd_inv_txfm_add_4x4(dqcoeff, dst, dst_stride, *eob, xd->bd,
- tx_type, x->highbd_itxm_add);
+ tx_type, xd->lossless);
break;
default:
assert(0);
@@ -1068,7 +1099,7 @@
if (!x->skip_recode) {
vpx_subtract_block(4, 4, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
- vp10_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type, x->fwd_txm4x4);
+ vp10_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type, xd->lossless);
vpx_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob, scan_order->scan,
@@ -1080,7 +1111,7 @@
// which is significant (not just an optimization) for the lossless
// case.
vp10_inv_txfm_add_4x4(dqcoeff, dst, dst_stride, *eob, tx_type,
- x->itxm_add);
+ xd->lossless);
}
break;
default:
diff --git a/vp10/encoder/encodemb.h b/vp10/encoder/encodemb.h
index 928be6c..62a7db4 100644
--- a/vp10/encoder/encodemb.h
+++ b/vp10/encoder/encodemb.h
@@ -39,16 +39,12 @@
void vp10_encode_intra_block_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane);
-void vp10_fwd_txfm_4x4(const int16_t *src_diff,
- tran_low_t *coeff, int diff_stride, TX_TYPE tx_type,
- void (*fwd_txm4x4)(const int16_t *input,
- tran_low_t *output, int stride));
+void vp10_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
+ int diff_stride, TX_TYPE tx_type, int lossless);
#if CONFIG_VP9_HIGHBITDEPTH
void vp10_highbd_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
- int diff_stride, TX_TYPE tx_type,
- void (*highbd_fwd_txm4x4)(const int16_t *input,
- tran_low_t *output, int stride));
+ int diff_stride, TX_TYPE tx_type, int lossless);
#endif // CONFIG_VP9_HIGHBITDEPTH
#ifdef __cplusplus
diff --git a/vp10/encoder/encoder.c b/vp10/encoder/encoder.c
index d5545a0..1d95d7d 100644
--- a/vp10/encoder/encoder.c
+++ b/vp10/encoder/encoder.c
@@ -4087,19 +4087,7 @@
}
if (oxcf->pass == 1) {
- const int lossless = is_lossless_requested(oxcf);
-#if CONFIG_VP9_HIGHBITDEPTH
- if (cpi->oxcf.use_highbitdepth)
- cpi->td.mb.fwd_txm4x4 = lossless ?
- vp10_highbd_fwht4x4 : vpx_highbd_fdct4x4;
- else
- cpi->td.mb.fwd_txm4x4 = lossless ? vp10_fwht4x4 : vpx_fdct4x4;
- cpi->td.mb.highbd_itxm_add = lossless ? vp10_highbd_iwht4x4_add :
- vp10_highbd_idct4x4_add;
-#else
- cpi->td.mb.fwd_txm4x4 = lossless ? vp10_fwht4x4 : vpx_fdct4x4;
-#endif // CONFIG_VP9_HIGHBITDEPTH
- cpi->td.mb.itxm_add = lossless ? vp10_iwht4x4_add : vp10_idct4x4_add;
+ cpi->td.mb.e_mbd.lossless = is_lossless_requested(oxcf);
vp10_first_pass(cpi, source);
} else if (oxcf->pass == 2) {
Pass2Encode(cpi, size, dest, frame_flags);
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index 20b7d50..e3bbdd3 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -798,8 +798,7 @@
if (xd->lossless) {
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block);
const scan_order *so = get_scan(TX_4X4, tx_type);
- vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT,
- vp10_highbd_fwht4x4);
+ vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, 1);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
so->scan, so->neighbors,
@@ -808,14 +807,12 @@
goto next_highbd;
vp10_highbd_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
dst, dst_stride, p->eobs[block],
- xd->bd, DCT_DCT,
- vp10_highbd_iwht4x4_add);
+ xd->bd, DCT_DCT, 1);
} else {
int64_t unused;
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block);
const scan_order *so = get_scan(TX_4X4, tx_type);
- vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, tx_type,
- vpx_highbd_fdct4x4);
+ vp10_highbd_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, 0);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
so->scan, so->neighbors,
@@ -827,8 +824,7 @@
goto next_highbd;
vp10_highbd_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
dst, dst_stride, p->eobs[block],
- xd->bd, tx_type,
- vp10_highbd_idct4x4_add);
+ xd->bd, tx_type, 0);
}
}
}
@@ -901,7 +897,7 @@
if (xd->lossless) {
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block);
const scan_order *so = get_scan(TX_4X4, tx_type);
- vp10_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, vp10_fwht4x4);
+ vp10_fwd_txfm_4x4(src_diff, coeff, 8, DCT_DCT, 1);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
so->scan, so->neighbors,
@@ -909,13 +905,12 @@
if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
goto next;
vp10_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
- dst, dst_stride, p->eobs[block], DCT_DCT,
- vp10_iwht4x4_add);
+ dst, dst_stride, p->eobs[block], DCT_DCT, 1);
} else {
int64_t unused;
TX_TYPE tx_type = get_tx_type(PLANE_TYPE_Y, xd, block);
const scan_order *so = get_scan(TX_4X4, tx_type);
- vp10_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, vpx_fdct4x4);
+ vp10_fwd_txfm_4x4(src_diff, coeff, 8, tx_type, 0);
vp10_regular_quantize_b_4x4(x, 0, block, so->scan, so->iscan);
ratey += cost_coeffs(x, 0, block, tempa + idx, templ + idy, TX_4X4,
so->scan, so->neighbors,
@@ -925,8 +920,7 @@
if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)
goto next;
vp10_inv_txfm_add_4x4(BLOCK_OFFSET(pd->dqcoeff, block),
- dst, dst_stride, p->eobs[block], tx_type,
- vp10_idct4x4_add);
+ dst, dst_stride, p->eobs[block], tx_type, 0);
}
}
}
@@ -1290,6 +1284,8 @@
const int width = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
const int height = 4 * num_4x4_blocks_high_lookup[plane_bsize];
int idx, idy;
+ void (*fwd_txm4x4)(const int16_t *input, tran_low_t *output, int stride);
+
const uint8_t *const src =
&p->src.buf[vp10_raster_block_offset(BLOCK_8X8, i, p->src.stride)];
uint8_t *const dst = &pd->dst.buf[vp10_raster_block_offset(BLOCK_8X8, i,
@@ -1303,6 +1299,16 @@
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+ fwd_txm4x4 = xd->lossless ? vp10_highbd_fwht4x4 : vpx_highbd_fdct4x4;
+ } else {
+ fwd_txm4x4 = xd->lossless ? vp10_fwht4x4 : vpx_fdct4x4;
+ }
+#else
+ fwd_txm4x4 = xd->lossless ? vp10_fwht4x4 : vpx_fdct4x4;
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
+#if CONFIG_VP9_HIGHBITDEPTH
+ if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
vpx_highbd_subtract_block(
height, width, vp10_raster_block_offset_int16(BLOCK_8X8, i, p->src_diff),
8, src, p->src.stride, dst, pd->dst.stride, xd->bd);
@@ -1325,8 +1331,8 @@
k += (idy * 2 + idx);
coeff = BLOCK_OFFSET(p->coeff, k);
- x->fwd_txm4x4(vp10_raster_block_offset_int16(BLOCK_8X8, k, p->src_diff),
- coeff, 8);
+ fwd_txm4x4(vp10_raster_block_offset_int16(BLOCK_8X8, k, p->src_diff),
+ coeff, 8);
vp10_regular_quantize_b_4x4(x, 0, k, so->scan, so->iscan);
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
diff --git a/vp10/vp10cx.mk b/vp10/vp10cx.mk
index 7393a4e..ead993a 100644
--- a/vp10/vp10cx.mk
+++ b/vp10/vp10cx.mk
@@ -23,7 +23,6 @@
VP10_CX_SRCS-yes += encoder/context_tree.h
VP10_CX_SRCS-yes += encoder/cost.h
VP10_CX_SRCS-yes += encoder/cost.c
-VP10_CX_SRCS-yes += encoder/dct.h
VP10_CX_SRCS-yes += encoder/dct.c
VP10_CX_SRCS-$(CONFIG_VP9_TEMPORAL_DENOISING) += encoder/denoiser.c
VP10_CX_SRCS-$(CONFIG_VP9_TEMPORAL_DENOISING) += encoder/denoiser.h
diff --git a/vp8/encoder/ratectrl.c b/vp8/encoder/ratectrl.c
index f264222..63c8e68 100644
--- a/vp8/encoder/ratectrl.c
+++ b/vp8/encoder/ratectrl.c
@@ -1596,6 +1596,10 @@
// Drop this frame: advance frame counters, and set force_maxqp flag.
cpi->common.current_video_frame++;
cpi->frames_since_key++;
+ // Adjust rate correction factor upwards.
+ cpi->rate_correction_factor *= 2.0;
+ if (cpi->rate_correction_factor > MAX_BPB_FACTOR)
+ cpi->rate_correction_factor = MAX_BPB_FACTOR;
// Flag to indicate we will force next frame to be encoded at max QP.
cpi->force_maxqp = 1;
return 1;
diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index 61077cd..4e88819 100644
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -126,6 +126,9 @@
void vp9_decoder_remove(VP9Decoder *pbi) {
int i;
+ if (!pbi)
+ return;
+
vpx_get_worker_interface()->end(&pbi->lf_worker);
vpx_free(pbi->lf_worker.data1);
vpx_free(pbi->tile_data);
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.c b/vp9/encoder/vp9_aq_cyclicrefresh.c
index bb1e179..968dad2 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.c
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.c
@@ -23,6 +23,7 @@
CYCLIC_REFRESH *vp9_cyclic_refresh_alloc(int mi_rows, int mi_cols) {
size_t last_coded_q_map_size;
+ size_t consec_zero_mv_size;
CYCLIC_REFRESH *const cr = vpx_calloc(1, sizeof(*cr));
if (cr == NULL)
return NULL;
@@ -41,12 +42,20 @@
assert(MAXQ <= 255);
memset(cr->last_coded_q_map, MAXQ, last_coded_q_map_size);
+ consec_zero_mv_size = mi_rows * mi_cols * sizeof(*cr->consec_zero_mv);
+ cr->consec_zero_mv = vpx_malloc(consec_zero_mv_size);
+ if (cr->consec_zero_mv == NULL) {
+ vpx_free(cr);
+ return NULL;
+ }
+ memset(cr->consec_zero_mv, 0, consec_zero_mv_size);
return cr;
}
void vp9_cyclic_refresh_free(CYCLIC_REFRESH *cr) {
vpx_free(cr->map);
vpx_free(cr->last_coded_q_map);
+ vpx_free(cr->consec_zero_mv);
vpx_free(cr);
}
@@ -228,22 +237,48 @@
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;
+ }
+}
+
+void vp9_cyclic_refresh_update_sb_postencode(VP9_COMP *const cpi,
+ const MB_MODE_INFO *const mbmi,
+ int mi_row, int mi_col,
+ BLOCK_SIZE bsize) {
+ const VP9_COMMON *const cm = &cpi->common;
+ CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
+ MV mv = mbmi->mv[0].as_mv;
+ const int bw = num_8x8_blocks_wide_lookup[bsize];
+ const int bh = num_8x8_blocks_high_lookup[bsize];
+ const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
+ const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
+ const int block_index = mi_row * cm->mi_cols + mi_col;
+ int x, y;
+ for (y = 0; y < ymis; y++)
+ for (x = 0; x < xmis; x++) {
+ int map_offset = block_index + y * cm->mi_cols + x;
// 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) &&
+ if ((!is_inter_block(mbmi) || !mbmi->skip) &&
mbmi->segment_id <= CR_SEGMENT_ID_BOOST2) {
cr->last_coded_q_map[map_offset] = clamp(
cm->base_qindex + cr->qindex_delta[mbmi->segment_id], 0, MAXQ);
- } else if (is_inter_block(mbmi) && skip &&
+ } else if (is_inter_block(mbmi) && mbmi->skip &&
mbmi->segment_id <= CR_SEGMENT_ID_BOOST2) {
cr->last_coded_q_map[map_offset] = VPXMIN(
clamp(cm->base_qindex + cr->qindex_delta[mbmi->segment_id],
0, MAXQ),
cr->last_coded_q_map[map_offset]);
+ // Update the consecutive zero/low_mv count.
+ if (is_inter_block(mbmi) && (abs(mv.row) < 8 && abs(mv.col) < 8)) {
+ if (cr->consec_zero_mv[map_offset] < 255)
+ cr->consec_zero_mv[map_offset]++;
+ } else {
+ cr->consec_zero_mv[map_offset] = 0;
}
}
+ }
}
// Update the actual number of blocks that were applied the segment delta q.
@@ -380,9 +415,10 @@
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
- ? vp9_get_qindex(&cm->seg, CR_SEGMENT_ID_BOOST2, cm->base_qindex)
- : 0;
+ vp9_get_qindex(&cm->seg, CR_SEGMENT_ID_BOOST2, cm->base_qindex);
+ int consec_zero_mv_thresh =
+ cpi->oxcf.content == VP9E_CONTENT_SCREEN ? 0
+ : 10 * (100 / cr->percent_refresh);
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;
@@ -398,7 +434,8 @@
// 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)
+ if (cr->last_coded_q_map[bl_index2] > qindex_thresh ||
+ cr->consec_zero_mv[bl_index2] < consec_zero_mv_thresh)
sum_map++;
} else if (cr->map[bl_index2] < 0) {
cr->map[bl_index2]++;
@@ -475,6 +512,8 @@
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));
+ memset(cr->consec_zero_mv, 0,
+ cm->mi_rows * cm->mi_cols * sizeof(*cr->consec_zero_mv));
cr->sb_index = 0;
}
return;
@@ -544,6 +583,8 @@
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
memset(cr->map, 0, cm->mi_rows * cm->mi_cols);
+ memset(cr->last_coded_q_map, MAXQ, cm->mi_rows * cm->mi_cols);
+ memset(cr->consec_zero_mv, 0, cm->mi_rows * cm->mi_cols);
cr->sb_index = 0;
cpi->refresh_golden_frame = 1;
}
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.h b/vp9/encoder/vp9_aq_cyclicrefresh.h
index 7da1f94..839ce6d 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.h
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.h
@@ -51,6 +51,8 @@
signed char *map;
// Map of the last q a block was coded at.
uint8_t *last_coded_q_map;
+ // Count on how many consecutive times a block uses ZER0MV for encoding.
+ uint8_t *consec_zero_mv;
// Thresholds applied to the projected rate/distortion of the coding block,
// when deciding whether block should be refreshed.
int64_t thresh_rate_sb;
@@ -92,6 +94,11 @@
int mi_row, int mi_col, BLOCK_SIZE bsize,
int64_t rate, int64_t dist, int skip);
+void vp9_cyclic_refresh_update_sb_postencode(struct VP9_COMP *const cpi,
+ const MB_MODE_INFO *const mbmi,
+ int mi_row, int mi_col,
+ BLOCK_SIZE bsize);
+
// Update the segmentation map, and related quantities: cyclic refresh map,
// refresh sb_index, and target number of blocks to be refreshed.
void vp9_cyclic_refresh_update__map(struct VP9_COMP *const cpi);
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 4a4301e..1c4f35a 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -4213,5 +4213,7 @@
}
++td->counts->tx.tx_totals[mbmi->tx_size];
++td->counts->tx.tx_totals[get_uv_tx_size(mbmi, &xd->plane[1])];
+ if (cm->seg.enabled && cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ)
+ vp9_cyclic_refresh_update_sb_postencode(cpi, mbmi, mi_row, mi_col, bsize);
}
}
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index aca289f..b2e800d 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -1517,7 +1517,10 @@
}
if (cpi->initial_width) {
- if (cm->width > cpi->initial_width || cm->height > cpi->initial_height) {
+ int new_mi_size = 0;
+ vp9_set_mb_mi(cm, cm->width, cm->height);
+ new_mi_size = cm->mi_stride * calc_mi_size(cm->mi_rows);
+ if (cm->mi_alloc_size < new_mi_size) {
vp9_free_context_buffers(cm);
alloc_compressor_data(cpi);
realloc_segmentation_maps(cpi);
@@ -4643,8 +4646,10 @@
// always go to the next whole number
cm->width = (hs - 1 + cpi->oxcf.width * hr) / hs;
cm->height = (vs - 1 + cpi->oxcf.height * vr) / vs;
- assert(cm->width <= cpi->initial_width);
- assert(cm->height <= cpi->initial_height);
+ if (cm->current_video_frame) {
+ assert(cm->width <= cpi->initial_width);
+ assert(cm->height <= cpi->initial_height);
+ }
update_frame_size(cpi);
