Merge "Speed up tm_predictor_16x16"
diff --git a/.mailmap b/.mailmap
index 42f3617..4672e5c 100644
--- a/.mailmap
+++ b/.mailmap
@@ -1,5 +1,4 @@
Adrian Grange <agrange@google.com>
-Adrian Grange <agrange@google.com> <agrange@agrange-macbookpro.roam.corp.google.com>
Aℓex Converse <aconverse@google.com>
Aℓex Converse <aconverse@google.com> <alex.converse@gmail.com>
Alexis Ballier <aballier@gentoo.org> <alexis.ballier@gmail.com>
@@ -8,13 +7,11 @@
Erik Niemeyer <erik.a.niemeyer@intel.com> <erik.a.niemeyer@gmail.com>
Guillaume Martres <gmartres@google.com> <smarter3@gmail.com>
Hangyu Kuang <hkuang@google.com>
-Hangyu Kuang <hkuang@google.com> <hkuang@hkuang-macbookpro.roam.corp.google.com>
Hui Su <huisu@google.com>
Jacky Chen <jackychen@google.com>
Jim Bankoski <jimbankoski@google.com>
Johann Koenig <johannkoenig@google.com>
Johann Koenig <johannkoenig@google.com> <johann.koenig@duck.com>
-Johann Koenig <johannkoenig@google.com> <johannkoenig@dhcp-172-19-7-52.mtv.corp.google.com>
Johann Koenig <johannkoenig@google.com> <johann.koenig@gmail.com>
John Koleszar <jkoleszar@google.com>
Joshua Litt <joshualitt@google.com> <joshualitt@chromium.org>
@@ -33,4 +30,3 @@
Tom Finegan <tomfinegan@google.com>
Tom Finegan <tomfinegan@google.com> <tomfinegan@chromium.org>
Yaowu Xu <yaowu@google.com> <yaowu@xuyaowu.com>
-Yaowu Xu <yaowu@google.com> <yaowu@YAOWU2-W.ad.corp.google.com>
diff --git a/build/make/iosbuild.sh b/build/make/iosbuild.sh
index 6f7180d..ae5ba18 100755
--- a/build/make/iosbuild.sh
+++ b/build/make/iosbuild.sh
@@ -29,11 +29,14 @@
LIBVPX_SOURCE_DIR=$(cd ${SCRIPT_DIR}/../..; pwd)
LIPO=$(xcrun -sdk iphoneos${SDK} -find lipo)
ORIG_PWD="$(pwd)"
-TARGETS="arm64-darwin-gcc
- armv7-darwin-gcc
- armv7s-darwin-gcc
- x86-iphonesimulator-gcc
- x86_64-iphonesimulator-gcc"
+ARM_TARGETS="arm64-darwin-gcc
+ armv7-darwin-gcc
+ armv7s-darwin-gcc"
+SIM_TARGETS="x86-iphonesimulator-gcc
+ x86_64-iphonesimulator-gcc"
+OSX_TARGETS="x86-darwin15-gcc
+ x86_64-darwin15-gcc"
+TARGETS="${ARM_TARGETS} ${SIM_TARGETS}"
# Configures for the target specified by $1, and invokes make with the dist
# target using $DIST_DIR as the distribution output directory.
@@ -197,15 +200,27 @@
fi
}
+print_list() {
+ local indent="$1"
+ shift
+ local list="$@"
+ for entry in ${list}; do
+ echo "${indent}${entry}"
+ done
+}
+
iosbuild_usage() {
cat << EOF
Usage: ${0##*/} [arguments]
--help: Display this message and exit.
--extra-configure-args <args>: Extra args to pass when configuring libvpx.
+ --macosx: Uses darwin15 targets instead of iphonesimulator targets for x86
+ and x86_64. Allows linking to framework when builds target MacOSX
+ instead of iOS.
--preserve-build-output: Do not delete the build directory.
--show-build-output: Show output from each library build.
--targets <targets>: Override default target list. Defaults:
- ${TARGETS}
+$(print_list " " ${TARGETS})
--test-link: Confirms all targets can be linked. Functionally identical to
passing --enable-examples via --extra-configure-args.
--verbose: Output information about the environment and each stage of the
@@ -249,6 +264,9 @@
TARGETS="$2"
shift
;;
+ --macosx)
+ TARGETS="${ARM_TARGETS} ${OSX_TARGETS}"
+ ;;
--verbose)
VERBOSE=yes
;;
@@ -273,10 +291,12 @@
MAKEFLAGS=${MAKEFLAGS}
ORIG_PWD=${ORIG_PWD}
PRESERVE_BUILD_OUTPUT=${PRESERVE_BUILD_OUTPUT}
- TARGETS="${TARGETS}"
+ TARGETS="$(print_list "" ${TARGETS})"
+ OSX_TARGETS="${OSX_TARGETS}"
+ SIM_TARGETS="${SIM_TARGETS}"
EOF
fi
build_framework "${TARGETS}"
echo "Successfully built '${FRAMEWORK_DIR}' for:"
-echo " ${TARGETS}"
+print_list "" ${TARGETS}
diff --git a/test/test.mk b/test/test.mk
index 8d66244..2487bd2 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -92,10 +92,9 @@
## shared library builds don't make these functions accessible.
##
ifeq ($(CONFIG_SHARED),)
-LIBVPX_TEST_SRCS-$(CONFIG_VP9) += lpf_8_test.cc
## VP8
-ifneq ($(CONFIG_VP8_ENCODER)$(CONFIG_VP8_DECODER),)
+ifeq ($(CONFIG_VP8),yes)
# These tests require both the encoder and decoder to be built.
ifeq ($(CONFIG_VP8_ENCODER)$(CONFIG_VP8_DECODER),yesyes)
@@ -105,10 +104,10 @@
LIBVPX_TEST_SRCS-$(CONFIG_POSTPROC) += pp_filter_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP8_DECODER) += vp8_decrypt_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += quantize_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += set_roi.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += variance_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += vp8_fdct4x4_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += quantize_test.cc
LIBVPX_TEST_SRCS-yes += idct_test.cc
LIBVPX_TEST_SRCS-yes += sixtap_predict_test.cc
@@ -121,7 +120,7 @@
endif # VP8
## VP9
-ifneq ($(CONFIG_VP9_ENCODER)$(CONFIG_VP9_DECODER),)
+ifeq ($(CONFIG_VP9),yes)
# These tests require both the encoder and decoder to be built.
ifeq ($(CONFIG_VP9_ENCODER)$(CONFIG_VP9_DECODER),yesyes)
@@ -134,25 +133,25 @@
LIBVPX_TEST_SRCS-yes += vp9_encoder_parms_get_to_decoder.cc
endif
-LIBVPX_TEST_SRCS-$(CONFIG_VP9) += convolve_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += vp9_thread_test.cc
+LIBVPX_TEST_SRCS-yes += convolve_test.cc
+LIBVPX_TEST_SRCS-yes += lpf_8_test.cc
+LIBVPX_TEST_SRCS-yes += vp9_intrapred_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += vp9_decrypt_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += vp9_thread_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct16x16_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct32x32_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct4x4_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_avg_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_error_block_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_quantize_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP9) += vp9_intrapred_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc
ifeq ($(CONFIG_VP9_ENCODER),yes)
LIBVPX_TEST_SRCS-$(CONFIG_SPATIAL_SVC) += svc_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_INTERNAL_STATS) += blockiness_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_INTERNAL_STATS) += consistency_test.cc
-
endif
ifeq ($(CONFIG_VP9_ENCODER)$(CONFIG_VP9_TEMPORAL_DENOISING),yesyes)
@@ -162,14 +161,19 @@
endif # VP9
-LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += sad_test.cc
-
-TEST_INTRA_PRED_SPEED_SRCS-$(CONFIG_VP9) := test_intra_pred_speed.cc
-TEST_INTRA_PRED_SPEED_SRCS-$(CONFIG_VP9) += ../md5_utils.h ../md5_utils.c
-
## VP10
+ifeq ($(CONFIG_VP10),yes)
+
+LIBVPX_TEST_SRCS-yes += vp10_inv_txfm_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP10_ENCODER) += vp10_dct_test.cc
-LIBVPX_TEST_SRCS-$(CONFIG_VP10) += vp10_inv_txfm_test.cc
+
+endif # VP10
+
+## Multi-codec / unconditional whitebox tests.
+LIBVPX_TEST_SRCS-$(CONFIG_ENCODERS) += sad_test.cc
+
+TEST_INTRA_PRED_SPEED_SRCS-yes := test_intra_pred_speed.cc
+TEST_INTRA_PRED_SPEED_SRCS-yes += ../md5_utils.h ../md5_utils.c
endif # CONFIG_SHARED
diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc
index d44a64a..4064ea6 100644
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -187,23 +187,19 @@
vpx_d153_predictor_4x4_c, vpx_d207_predictor_4x4_c,
vpx_d63_predictor_4x4_c, vpx_tm_predictor_4x4_c)
-#if HAVE_SSE && CONFIG_USE_X86INC
-INTRA_PRED_TEST(SSE, TestIntraPred4, vpx_dc_predictor_4x4_sse,
- vpx_dc_left_predictor_4x4_sse, vpx_dc_top_predictor_4x4_sse,
- vpx_dc_128_predictor_4x4_sse, vpx_v_predictor_4x4_sse, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL)
-#endif // HAVE_SSE && CONFIG_USE_X86INC
-
#if HAVE_SSE2 && CONFIG_USE_X86INC
-INTRA_PRED_TEST(SSE2, TestIntraPred4, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_4x4_sse2)
+INTRA_PRED_TEST(SSE2, TestIntraPred4, vpx_dc_predictor_4x4_sse2,
+ vpx_dc_left_predictor_4x4_sse2, vpx_dc_top_predictor_4x4_sse2,
+ vpx_dc_128_predictor_4x4_sse2, vpx_v_predictor_4x4_sse2,
+ vpx_h_predictor_4x4_sse2, NULL, NULL, NULL, NULL, NULL, NULL,
+ vpx_tm_predictor_4x4_sse2)
#endif // HAVE_SSE2 && CONFIG_USE_X86INC
#if HAVE_SSSE3 && CONFIG_USE_X86INC
INTRA_PRED_TEST(SSSE3, TestIntraPred4, NULL, NULL, NULL, NULL, NULL,
- vpx_h_predictor_4x4_ssse3, vpx_d45_predictor_4x4_ssse3, NULL,
- NULL, vpx_d153_predictor_4x4_ssse3,
- vpx_d207_predictor_4x4_ssse3, vpx_d63_predictor_4x4_ssse3, NULL)
+ NULL, vpx_d45_predictor_4x4_ssse3, NULL, NULL,
+ vpx_d153_predictor_4x4_ssse3, vpx_d207_predictor_4x4_ssse3,
+ vpx_d63_predictor_4x4_ssse3, NULL)
#endif // HAVE_SSSE3 && CONFIG_USE_X86INC
#if HAVE_DSPR2
@@ -240,23 +236,19 @@
vpx_d153_predictor_8x8_c, vpx_d207_predictor_8x8_c,
vpx_d63_predictor_8x8_c, vpx_tm_predictor_8x8_c)
-#if HAVE_SSE && CONFIG_USE_X86INC
-INTRA_PRED_TEST(SSE, TestIntraPred8, vpx_dc_predictor_8x8_sse,
- vpx_dc_left_predictor_8x8_sse, vpx_dc_top_predictor_8x8_sse,
- vpx_dc_128_predictor_8x8_sse, vpx_v_predictor_8x8_sse, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL)
-#endif // HAVE_SSE && CONFIG_USE_X86INC
-
#if HAVE_SSE2 && CONFIG_USE_X86INC
-INTRA_PRED_TEST(SSE2, TestIntraPred8, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_8x8_sse2)
+INTRA_PRED_TEST(SSE2, TestIntraPred8, vpx_dc_predictor_8x8_sse2,
+ vpx_dc_left_predictor_8x8_sse2, vpx_dc_top_predictor_8x8_sse2,
+ vpx_dc_128_predictor_8x8_sse2, vpx_v_predictor_8x8_sse2,
+ vpx_h_predictor_8x8_sse2, NULL, NULL, NULL, NULL, NULL,
+ NULL, vpx_tm_predictor_8x8_sse2)
#endif // HAVE_SSE2 && CONFIG_USE_X86INC
#if HAVE_SSSE3 && CONFIG_USE_X86INC
INTRA_PRED_TEST(SSSE3, TestIntraPred8, NULL, NULL, NULL, NULL, NULL,
- vpx_h_predictor_8x8_ssse3, vpx_d45_predictor_8x8_ssse3, NULL,
- NULL, vpx_d153_predictor_8x8_ssse3,
- vpx_d207_predictor_8x8_ssse3, vpx_d63_predictor_8x8_ssse3, NULL)
+ NULL, vpx_d45_predictor_8x8_ssse3, NULL, NULL,
+ vpx_d153_predictor_8x8_ssse3, vpx_d207_predictor_8x8_ssse3,
+ vpx_d63_predictor_8x8_ssse3, NULL)
#endif // HAVE_SSSE3 && CONFIG_USE_X86INC
#if HAVE_DSPR2
@@ -298,13 +290,13 @@
vpx_dc_left_predictor_16x16_sse2,
vpx_dc_top_predictor_16x16_sse2,
vpx_dc_128_predictor_16x16_sse2, vpx_v_predictor_16x16_sse2,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL,
+ vpx_h_predictor_16x16_sse2, NULL, NULL, NULL, NULL, NULL, NULL,
vpx_tm_predictor_16x16_sse2)
#endif // HAVE_SSE2 && CONFIG_USE_X86INC
#if HAVE_SSSE3 && CONFIG_USE_X86INC
INTRA_PRED_TEST(SSSE3, TestIntraPred16, NULL, NULL, NULL, NULL, NULL,
- vpx_h_predictor_16x16_ssse3, vpx_d45_predictor_16x16_ssse3,
+ NULL, vpx_d45_predictor_16x16_ssse3,
NULL, NULL, vpx_d153_predictor_16x16_ssse3,
vpx_d207_predictor_16x16_ssse3, vpx_d63_predictor_16x16_ssse3,
NULL)
@@ -350,23 +342,23 @@
vpx_dc_left_predictor_32x32_sse2,
vpx_dc_top_predictor_32x32_sse2,
vpx_dc_128_predictor_32x32_sse2, vpx_v_predictor_32x32_sse2,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- vpx_tm_predictor_32x32_sse2)
+ vpx_h_predictor_32x32_sse2, NULL, NULL, NULL, NULL, NULL,
+ NULL, vpx_tm_predictor_32x32_sse2)
#else
INTRA_PRED_TEST(SSE2, TestIntraPred32, vpx_dc_predictor_32x32_sse2,
vpx_dc_left_predictor_32x32_sse2,
vpx_dc_top_predictor_32x32_sse2,
vpx_dc_128_predictor_32x32_sse2, vpx_v_predictor_32x32_sse2,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL)
+ vpx_h_predictor_32x32_sse2, NULL, NULL, NULL, NULL, NULL,
+ NULL, NULL)
#endif // ARCH_X86_64
#endif // HAVE_SSE2 && CONFIG_USE_X86INC
#if HAVE_SSSE3 && CONFIG_USE_X86INC
INTRA_PRED_TEST(SSSE3, TestIntraPred32, NULL, NULL, NULL, NULL, NULL,
- vpx_h_predictor_32x32_ssse3, vpx_d45_predictor_32x32_ssse3,
- NULL, NULL, vpx_d153_predictor_32x32_ssse3,
- vpx_d207_predictor_32x32_ssse3, vpx_d63_predictor_32x32_ssse3,
- NULL)
+ NULL, vpx_d45_predictor_32x32_ssse3, NULL, NULL,
+ vpx_d153_predictor_32x32_ssse3, vpx_d207_predictor_32x32_ssse3,
+ vpx_d63_predictor_32x32_ssse3, NULL)
#endif // HAVE_SSSE3 && CONFIG_USE_X86INC
#if HAVE_NEON
diff --git a/test/vp9_avg_test.cc b/test/vp9_avg_test.cc
index 290bdc7..cbc667e 100644
--- a/test/vp9_avg_test.cc
+++ b/test/vp9_avg_test.cc
@@ -372,7 +372,10 @@
::testing::Values(
make_tuple(16, 16, 0, 8, &vp9_avg_8x8_neon),
make_tuple(16, 16, 5, 8, &vp9_avg_8x8_neon),
- make_tuple(32, 32, 15, 8, &vp9_avg_8x8_neon)));
+ make_tuple(32, 32, 15, 8, &vp9_avg_8x8_neon),
+ make_tuple(16, 16, 0, 4, &vp9_avg_4x4_neon),
+ make_tuple(16, 16, 5, 4, &vp9_avg_4x4_neon),
+ make_tuple(32, 32, 15, 4, &vp9_avg_4x4_neon)));
INSTANTIATE_TEST_CASE_P(
NEON, IntProRowTest, ::testing::Values(
diff --git a/third_party/x86inc/x86inc.asm b/third_party/x86inc/x86inc.asm
index be59de3..e7d3fa5 100644
--- a/third_party/x86inc/x86inc.asm
+++ b/third_party/x86inc/x86inc.asm
@@ -119,7 +119,7 @@
%if ABI_IS_32BIT
%if CONFIG_PIC=1
%ifidn __OUTPUT_FORMAT__,elf32
- %define GET_GOT_SAVE_ARG 1
+ %define GET_GOT_DEFINED 1
%define WRT_PLT wrt ..plt
%macro GET_GOT 1
extern _GLOBAL_OFFSET_TABLE_
@@ -138,7 +138,7 @@
%define RESTORE_GOT pop %1
%endmacro
%elifidn __OUTPUT_FORMAT__,macho32
- %define GET_GOT_SAVE_ARG 1
+ %define GET_GOT_DEFINED 1
%macro GET_GOT 1
push %1
call %%get_got
@@ -149,6 +149,8 @@
%undef RESTORE_GOT
%define RESTORE_GOT pop %1
%endmacro
+ %else
+ %define GET_GOT_DEFINED 0
%endif
%endif
diff --git a/tools/gen_authors.sh b/tools/gen_authors.sh
index e1246f0..4cfd81e 100755
--- a/tools/gen_authors.sh
+++ b/tools/gen_authors.sh
@@ -6,7 +6,7 @@
# This file is automatically generated from the git commit history
# by tools/gen_authors.sh.
-$(git log --pretty=format:"%aN <%aE>" | sort | uniq)
+$(git log --pretty=format:"%aN <%aE>" | sort | uniq | grep -v corp.google)
Google Inc.
The Mozilla Foundation
The Xiph.Org Foundation
diff --git a/vp10/common/blockd.h b/vp10/common/blockd.h
index b89d791..2f10378 100644
--- a/vp10/common/blockd.h
+++ b/vp10/common/blockd.h
@@ -63,20 +63,6 @@
#define MAX_REF_FRAMES 4
typedef int8_t MV_REFERENCE_FRAME;
-typedef struct {
- // Number of base colors for Y (0) and UV (1)
- uint8_t palette_size[2];
- // Value of base colors for Y, U, and V
-#if CONFIG_VP9_HIGHBITDEPTH
- uint16_t palette_colors[3 * PALETTE_MAX_SIZE];
-#else
- uint8_t palette_colors[3 * PALETTE_MAX_SIZE];
-#endif // CONFIG_VP9_HIGHBITDEPTH
- // Only used by encoder to store the color index of the top left pixel.
- // TODO(huisu): move this to encoder
- uint8_t palette_first_color_idx[2];
-} PALETTE_MODE_INFO;
-
// This structure now relates to 8x8 block regions.
typedef struct {
// Common for both INTER and INTRA blocks
@@ -92,7 +78,6 @@
// Only for INTRA blocks
PREDICTION_MODE uv_mode;
- PALETTE_MODE_INFO palette_mode_info;
// Only for INTER blocks
INTERP_FILTER interp_filter;
diff --git a/vp10/common/entropymode.c b/vp10/common/entropymode.c
index 8098305..2bb292a 100644
--- a/vp10/common/entropymode.c
+++ b/vp10/common/entropymode.c
@@ -279,442 +279,6 @@
{ 66 } }
};
-const vpx_tree_index vp10_palette_size_tree[TREE_SIZE(PALETTE_SIZES)] = {
- -TWO_COLORS, 2,
- -THREE_COLORS, 4,
- -FOUR_COLORS, 6,
- -FIVE_COLORS, 8,
- -SIX_COLORS, 10,
- -SEVEN_COLORS, -EIGHT_COLORS,
-};
-
-// TODO(huisu): tune these probs
-const vpx_prob
-vp10_default_palette_y_size_prob[PALETTE_BLOCK_SIZES][PALETTE_SIZES - 1] = {
- { 96, 89, 100, 64, 77, 130},
- { 22, 15, 44, 16, 34, 82},
- { 30, 19, 57, 18, 38, 86},
- { 94, 36, 104, 23, 43, 92},
- { 116, 76, 107, 46, 65, 105},
- { 112, 82, 94, 40, 70, 112},
- { 147, 124, 123, 58, 69, 103},
- { 180, 113, 136, 49, 45, 114},
- { 107, 70, 87, 49, 154, 156},
- { 98, 105, 142, 63, 64, 152},
-};
-
-const vpx_prob
-vp10_default_palette_uv_size_prob[PALETTE_BLOCK_SIZES][PALETTE_SIZES - 1] = {
- { 160, 196, 228, 213, 175, 230},
- { 87, 148, 208, 141, 166, 163},
- { 72, 151, 204, 139, 155, 161},
- { 78, 135, 171, 104, 120, 173},
- { 59, 92, 131, 78, 92, 142},
- { 75, 118, 149, 84, 90, 128},
- { 89, 87, 92, 66, 66, 128},
- { 67, 53, 54, 55, 66, 93},
- { 120, 130, 83, 171, 75, 214},
- { 72, 55, 66, 68, 79, 107},
-};
-
-const vpx_prob
-vp10_default_palette_y_mode_prob[PALETTE_BLOCK_SIZES][PALETTE_Y_MODE_CONTEXTS]
- = {
- { 240, 180, 100, },
- { 240, 180, 100, },
- { 240, 180, 100, },
- { 240, 180, 100, },
- { 240, 180, 100, },
- { 240, 180, 100, },
- { 240, 180, 100, },
- { 240, 180, 100, },
- { 240, 180, 100, },
- { 240, 180, 100, },
-};
-
-
-const vpx_prob default_uv_palette_mode_prob[2] = {
- 253, 229
-};
-
-const vpx_tree_index
-vp10_palette_color_tree[PALETTE_MAX_SIZE - 1][TREE_SIZE(PALETTE_COLORS)] = {
- { // 2 colors
- -PALETTE_COLOR_ONE, -PALETTE_COLOR_TWO,
- },
- { // 3 colors
- -PALETTE_COLOR_ONE, 2,
- -PALETTE_COLOR_TWO, -PALETTE_COLOR_THREE,
- },
- { // 4 colors
- -PALETTE_COLOR_ONE, 2,
- -PALETTE_COLOR_TWO, 4,
- -PALETTE_COLOR_THREE, -PALETTE_COLOR_FOUR,
- },
- { // 5 colors
- -PALETTE_COLOR_ONE, 2,
- -PALETTE_COLOR_TWO, 4,
- -PALETTE_COLOR_THREE, 6,
- -PALETTE_COLOR_FOUR, -PALETTE_COLOR_FIVE,
- },
- { // 6 colors
- -PALETTE_COLOR_ONE, 2,
- -PALETTE_COLOR_TWO, 4,
- -PALETTE_COLOR_THREE, 6,
- -PALETTE_COLOR_FOUR, 8,
- -PALETTE_COLOR_FIVE, -PALETTE_COLOR_SIX,
- },
- { // 7 colors
- -PALETTE_COLOR_ONE, 2,
- -PALETTE_COLOR_TWO, 4,
- -PALETTE_COLOR_THREE, 6,
- -PALETTE_COLOR_FOUR, 8,
- -PALETTE_COLOR_FIVE, 10,
- -PALETTE_COLOR_SIX, -PALETTE_COLOR_SEVEN,
- },
- { // 8 colors
- -PALETTE_COLOR_ONE, 2,
- -PALETTE_COLOR_TWO, 4,
- -PALETTE_COLOR_THREE, 6,
- -PALETTE_COLOR_FOUR, 8,
- -PALETTE_COLOR_FIVE, 10,
- -PALETTE_COLOR_SIX, 12,
- -PALETTE_COLOR_SEVEN, -PALETTE_COLOR_EIGHT,
- },
-};
-
-const vpx_prob vp10_default_palette_y_color_prob
-[PALETTE_MAX_SIZE - 1][PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 1] = {
- { // 2 colors
- { 230, 255, 128, 128, 128, 128, 128 },
- { 214, 255, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 240, 255, 128, 128, 128, 128, 128 },
- { 73, 255, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 130, 255, 128, 128, 128, 128, 128 },
- { 227, 255, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 188, 255, 128, 128, 128, 128, 128 },
- { 75, 255, 128, 128, 128, 128, 128 },
- { 250, 255, 128, 128, 128, 128, 128 },
- { 223, 255, 128, 128, 128, 128, 128 },
- { 252, 255, 128, 128, 128, 128, 128 },
- }, { // 3 colors
- { 229, 137, 255, 128, 128, 128, 128 },
- { 197, 120, 255, 128, 128, 128, 128 },
- { 107, 195, 255, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 27, 151, 255, 128, 128, 128, 128 },
- { 230, 130, 255, 128, 128, 128, 128 },
- { 37, 230, 255, 128, 128, 128, 128 },
- { 67, 221, 255, 128, 128, 128, 128 },
- { 124, 230, 255, 128, 128, 128, 128 },
- { 195, 109, 255, 128, 128, 128, 128 },
- { 99, 122, 255, 128, 128, 128, 128 },
- { 205, 208, 255, 128, 128, 128, 128 },
- { 40, 235, 255, 128, 128, 128, 128 },
- { 251, 132, 255, 128, 128, 128, 128 },
- { 237, 186, 255, 128, 128, 128, 128 },
- { 253, 112, 255, 128, 128, 128, 128 },
- }, { // 4 colors
- { 195, 87, 128, 255, 128, 128, 128 },
- { 143, 100, 123, 255, 128, 128, 128 },
- { 94, 124, 119, 255, 128, 128, 128 },
- { 77, 91, 130, 255, 128, 128, 128 },
- { 39, 114, 178, 255, 128, 128, 128 },
- { 222, 94, 125, 255, 128, 128, 128 },
- { 44, 203, 132, 255, 128, 128, 128 },
- { 68, 175, 122, 255, 128, 128, 128 },
- { 110, 187, 124, 255, 128, 128, 128 },
- { 152, 91, 128, 255, 128, 128, 128 },
- { 70, 109, 181, 255, 128, 128, 128 },
- { 133, 113, 164, 255, 128, 128, 128 },
- { 47, 205, 133, 255, 128, 128, 128 },
- { 247, 94, 136, 255, 128, 128, 128 },
- { 205, 122, 146, 255, 128, 128, 128 },
- { 251, 100, 141, 255, 128, 128, 128 },
- }, { // 5 colors
- { 195, 65, 84, 125, 255, 128, 128 },
- { 150, 76, 84, 121, 255, 128, 128 },
- { 94, 110, 81, 117, 255, 128, 128 },
- { 79, 85, 91, 139, 255, 128, 128 },
- { 26, 102, 139, 127, 255, 128, 128 },
- { 220, 73, 91, 119, 255, 128, 128 },
- { 38, 203, 86, 127, 255, 128, 128 },
- { 61, 186, 72, 124, 255, 128, 128 },
- { 132, 199, 84, 128, 255, 128, 128 },
- { 172, 52, 62, 120, 255, 128, 128 },
- { 102, 89, 121, 122, 255, 128, 128 },
- { 182, 48, 69, 186, 255, 128, 128 },
- { 36, 206, 87, 126, 255, 128, 128 },
- { 249, 55, 67, 122, 255, 128, 128 },
- { 218, 88, 75, 122, 255, 128, 128 },
- { 253, 64, 80, 119, 255, 128, 128 },
- }, { // 6 colors
- { 182, 54, 64, 75, 118, 255, 128 },
- { 126, 67, 70, 76, 116, 255, 128 },
- { 79, 92, 67, 85, 120, 255, 128 },
- { 63, 61, 81, 118, 132, 255, 128 },
- { 21, 80, 105, 83, 119, 255, 128 },
- { 215, 72, 74, 74, 111, 255, 128 },
- { 50, 176, 63, 79, 120, 255, 128 },
- { 72, 148, 66, 77, 120, 255, 128 },
- { 105, 177, 57, 78, 130, 255, 128 },
- { 150, 66, 66, 80, 127, 255, 128 },
- { 81, 76, 109, 85, 116, 255, 128 },
- { 113, 81, 62, 96, 148, 255, 128 },
- { 54, 179, 69, 82, 121, 255, 128 },
- { 244, 47, 48, 67, 118, 255, 128 },
- { 198, 83, 53, 65, 121, 255, 128 },
- { 250, 42, 51, 69, 110, 255, 128 },
- }, { // 7 colors
- { 182, 45, 54, 62, 74, 113, 255 },
- { 124, 63, 57, 62, 77, 114, 255 },
- { 77, 80, 56, 66, 76, 117, 255 },
- { 63, 57, 69, 98, 85, 131, 255 },
- { 19, 81, 98, 63, 80, 116, 255 },
- { 215, 56, 60, 63, 68, 105, 255 },
- { 50, 174, 50, 60, 79, 118, 255 },
- { 68, 151, 50, 58, 73, 117, 255 },
- { 104, 182, 53, 57, 79, 127, 255 },
- { 156, 50, 51, 63, 77, 111, 255 },
- { 88, 67, 97, 59, 82, 120, 255 },
- { 114, 81, 46, 65, 103, 132, 255 },
- { 55, 166, 57, 66, 82, 120, 255 },
- { 245, 34, 38, 43, 63, 114, 255 },
- { 203, 68, 45, 47, 60, 118, 255 },
- { 250, 35, 37, 47, 66, 110, 255 },
- }, { // 8 colors
- { 180, 43, 46, 50, 56, 69, 109 },
- { 116, 53, 51, 49, 57, 73, 115 },
- { 79, 70, 49, 50, 59, 74, 117 },
- { 60, 54, 57, 70, 62, 83, 129 },
- { 20, 73, 85, 52, 66, 81, 119 },
- { 213, 56, 52, 49, 53, 62, 104 },
- { 48, 161, 41, 45, 56, 77, 116 },
- { 68, 139, 40, 47, 54, 71, 116 },
- { 123, 166, 42, 43, 52, 76, 130 },
- { 153, 44, 44, 47, 54, 79, 129 },
- { 87, 64, 83, 49, 60, 75, 127 },
- { 131, 68, 43, 48, 73, 96, 130 },
- { 55, 152, 45, 51, 64, 77, 113 },
- { 243, 30, 28, 33, 41, 65, 114 },
- { 202, 56, 35, 36, 42, 63, 123 },
- { 249, 31, 29, 32, 45, 68, 111 },
- }
-};
-
-const vpx_prob vp10_default_palette_uv_color_prob
-[PALETTE_MAX_SIZE - 1][PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 1] = {
- { // 2 colors
- { 228, 255, 128, 128, 128, 128, 128 },
- { 195, 255, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 228, 255, 128, 128, 128, 128, 128 },
- { 71, 255, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 129, 255, 128, 128, 128, 128, 128 },
- { 206, 255, 128, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 136, 255, 128, 128, 128, 128, 128 },
- { 98, 255, 128, 128, 128, 128, 128 },
- { 236, 255, 128, 128, 128, 128, 128 },
- { 222, 255, 128, 128, 128, 128, 128 },
- { 249, 255, 128, 128, 128, 128, 128 },
- }, { // 3 colors
- { 198, 136, 255, 128, 128, 128, 128 },
- { 178, 105, 255, 128, 128, 128, 128 },
- { 100, 206, 255, 128, 128, 128, 128 },
- { 128, 128, 128, 128, 128, 128, 128 },
- { 12, 136, 255, 128, 128, 128, 128 },
- { 219, 134, 255, 128, 128, 128, 128 },
- { 50, 198, 255, 128, 128, 128, 128 },
- { 61, 231, 255, 128, 128, 128, 128 },
- { 110, 209, 255, 128, 128, 128, 128 },
- { 173, 106, 255, 128, 128, 128, 128 },
- { 145, 166, 255, 128, 128, 128, 128 },
- { 156, 175, 255, 128, 128, 128, 128 },
- { 69, 183, 255, 128, 128, 128, 128 },
- { 241, 163, 255, 128, 128, 128, 128 },
- { 224, 160, 255, 128, 128, 128, 128 },
- { 246, 154, 255, 128, 128, 128, 128 },
- }, { // 4 colors
- { 173, 88, 143, 255, 128, 128, 128 },
- { 146, 81, 127, 255, 128, 128, 128 },
- { 84, 134, 102, 255, 128, 128, 128 },
- { 69, 138, 140, 255, 128, 128, 128 },
- { 31, 103, 200, 255, 128, 128, 128 },
- { 217, 101, 139, 255, 128, 128, 128 },
- { 51, 174, 121, 255, 128, 128, 128 },
- { 64, 177, 109, 255, 128, 128, 128 },
- { 96, 179, 145, 255, 128, 128, 128 },
- { 164, 77, 114, 255, 128, 128, 128 },
- { 87, 94, 156, 255, 128, 128, 128 },
- { 105, 57, 173, 255, 128, 128, 128 },
- { 63, 158, 137, 255, 128, 128, 128 },
- { 236, 102, 156, 255, 128, 128, 128 },
- { 197, 115, 153, 255, 128, 128, 128 },
- { 245, 106, 154, 255, 128, 128, 128 },
- }, { // 5 colors
- { 179, 64, 97, 129, 255, 128, 128 },
- { 137, 56, 88, 125, 255, 128, 128 },
- { 82, 107, 61, 118, 255, 128, 128 },
- { 59, 113, 86, 115, 255, 128, 128 },
- { 23, 88, 118, 130, 255, 128, 128 },
- { 213, 66, 90, 125, 255, 128, 128 },
- { 37, 181, 103, 121, 255, 128, 128 },
- { 47, 188, 61, 131, 255, 128, 128 },
- { 104, 185, 103, 144, 255, 128, 128 },
- { 163, 39, 76, 112, 255, 128, 128 },
- { 94, 74, 131, 126, 255, 128, 128 },
- { 142, 42, 103, 163, 255, 128, 128 },
- { 53, 162, 99, 149, 255, 128, 128 },
- { 239, 54, 84, 108, 255, 128, 128 },
- { 203, 84, 110, 147, 255, 128, 128 },
- { 248, 70, 105, 151, 255, 128, 128 },
- }, { // 6 colors
- { 189, 50, 67, 90, 130, 255, 128 },
- { 114, 50, 55, 90, 123, 255, 128 },
- { 66, 76, 54, 82, 128, 255, 128 },
- { 43, 69, 69, 80, 129, 255, 128 },
- { 22, 59, 87, 88, 141, 255, 128 },
- { 203, 49, 68, 87, 122, 255, 128 },
- { 43, 157, 74, 104, 146, 255, 128 },
- { 54, 138, 51, 95, 138, 255, 128 },
- { 82, 171, 58, 102, 146, 255, 128 },
- { 129, 38, 59, 64, 168, 255, 128 },
- { 56, 67, 119, 92, 112, 255, 128 },
- { 96, 62, 53, 132, 82, 255, 128 },
- { 60, 147, 77, 108, 145, 255, 128 },
- { 238, 76, 73, 93, 148, 255, 128 },
- { 189, 86, 73, 103, 157, 255, 128 },
- { 246, 62, 75, 83, 167, 255, 128 },
- }, { // 7 colors
- { 179, 42, 51, 73, 99, 134, 255 },
- { 119, 52, 52, 61, 64, 114, 255 },
- { 53, 77, 35, 65, 71, 131, 255 },
- { 38, 70, 51, 68, 89, 144, 255 },
- { 23, 65, 128, 73, 97, 131, 255 },
- { 210, 47, 52, 63, 81, 143, 255 },
- { 42, 159, 57, 68, 98, 143, 255 },
- { 49, 153, 45, 82, 93, 143, 255 },
- { 81, 169, 52, 72, 113, 151, 255 },
- { 136, 46, 35, 56, 75, 96, 255 },
- { 57, 84, 109, 47, 107, 131, 255 },
- { 128, 78, 57, 36, 128, 85, 255 },
- { 54, 149, 68, 77, 94, 153, 255 },
- { 243, 58, 50, 71, 81, 167, 255 },
- { 189, 92, 64, 70, 121, 173, 255 },
- { 248, 35, 38, 51, 82, 201, 255 },
- }, { // 8 colors
- { 201, 40, 36, 42, 64, 92, 123 },
- { 116, 43, 33, 43, 73, 102, 128 },
- { 46, 77, 37, 69, 62, 78, 150 },
- { 40, 65, 52, 50, 76, 89, 133 },
- { 28, 48, 91, 17, 64, 77, 133 },
- { 218, 43, 43, 37, 56, 72, 163 },
- { 41, 155, 44, 83, 82, 129, 180 },
- { 44, 141, 29, 55, 64, 89, 147 },
- { 92, 166, 48, 45, 59, 126, 179 },
- { 169, 35, 49, 41, 36, 99, 139 },
- { 55, 77, 77, 56, 60, 75, 156 },
- { 155, 81, 51, 64, 57, 182, 255 },
- { 60, 134, 49, 49, 93, 128, 174 },
- { 244, 98, 51, 46, 22, 73, 238 },
- { 189, 70, 40, 87, 93, 79, 201 },
- { 248, 54, 49, 40, 29, 42, 227 },
- }
-};
-
-static const int palette_color_context_lookup[PALETTE_COLOR_CONTEXTS] = {
- // (3, 0, 0, 0), (3, 2, 0, 0), (3, 3, 2, 0), (3, 3, 2, 2),
- 3993, 4235, 4378, 4380,
- // (4, 3, 3, 0), (5, 0, 0, 0), (5, 3, 0, 0), (5, 3, 2, 0),
- 5720, 6655, 7018, 7040,
- // (5, 5, 0, 0), (6, 2, 0, 0), (6, 2, 2, 0), (6, 4, 0, 0),
- 7260, 8228, 8250, 8470,
- // (7, 3, 0, 0), (8, 0, 0, 0), (8, 2, 0, 0), (10, 0, 0, 0)
- 9680, 10648, 10890, 13310
-};
-
-int vp10_get_palette_color_context(const uint8_t *color_map, int cols,
- int r, int c, int n, int *color_order) {
- int i, j, max, max_idx, temp;
- int scores[PALETTE_MAX_SIZE + 10];
- int weights[4] = {3, 2, 3, 2};
- int color_ctx = 0;
- int color_neighbors[4];
-
- assert(n <= PALETTE_MAX_SIZE);
-
- if (c - 1 >= 0)
- color_neighbors[0] = color_map[r * cols + c - 1];
- else
- color_neighbors[0] = -1;
- if (c - 1 >= 0 && r - 1 >= 0)
- color_neighbors[1] = color_map[(r - 1) * cols + c - 1];
- else
- color_neighbors[1] = -1;
- if (r - 1 >= 0)
- color_neighbors[2] = color_map[(r - 1) * cols + c];
- else
- color_neighbors[2] = -1;
- if (r - 1 >= 0 && c + 1 <= cols - 1)
- color_neighbors[3] = color_map[(r - 1) * cols + c + 1];
- else
- color_neighbors[3] = -1;
-
- for (i = 0; i < PALETTE_MAX_SIZE; ++i)
- color_order[i] = i;
- memset(scores, 0, PALETTE_MAX_SIZE * sizeof(scores[0]));
- for (i = 0; i < 4; ++i) {
- if (color_neighbors[i] >= 0)
- scores[color_neighbors[i]] += weights[i];
- }
-
- for (i = 0; i < 4; ++i) {
- max = scores[i];
- max_idx = i;
- j = i + 1;
- while (j < n) {
- if (scores[j] > max) {
- max = scores[j];
- max_idx = j;
- }
- ++j;
- }
-
- if (max_idx != i) {
- temp = scores[i];
- scores[i] = scores[max_idx];
- scores[max_idx] = temp;
-
- temp = color_order[i];
- color_order[i] = color_order[max_idx];
- color_order[max_idx] = temp;
- }
- }
-
- for (i = 0; i < 4; ++i)
- color_ctx = color_ctx * 11 + scores[i];
-
- for (i = 0; i < PALETTE_COLOR_CONTEXTS; ++i)
- if (color_ctx == palette_color_context_lookup[i]) {
- color_ctx = i;
- break;
- }
-
- if (color_ctx >= PALETTE_COLOR_CONTEXTS)
- color_ctx = 0;
-
- return color_ctx;
-}
-
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];
diff --git a/vp10/common/entropymode.h b/vp10/common/entropymode.h
index c9b667b..42fd920 100644
--- a/vp10/common/entropymode.h
+++ b/vp10/common/entropymode.h
@@ -27,12 +27,6 @@
#define INTER_OFFSET(mode) ((mode) - NEARESTMV)
-#define PALETTE_COLOR_CONTEXTS 16
-#define PALETTE_MAX_SIZE 8
-#define PALETTE_BLOCK_SIZES (BLOCK_64X64 - BLOCK_8X8 + 1)
-#define PALETTE_Y_MODE_CONTEXTS 3
-
-
struct VP10Common;
struct tx_probs {
@@ -105,25 +99,12 @@
extern const vpx_prob vp10_kf_partition_probs[PARTITION_CONTEXTS]
[PARTITION_TYPES - 1];
#endif
-extern const vpx_prob
-vp10_default_palette_y_mode_prob[PALETTE_BLOCK_SIZES][PALETTE_Y_MODE_CONTEXTS];
-extern const vpx_prob
-vp10_default_palette_y_size_prob[PALETTE_BLOCK_SIZES][PALETTE_SIZES - 1];
-extern const vpx_prob
-vp10_default_palette_uv_size_prob[PALETTE_BLOCK_SIZES][PALETTE_SIZES - 1];
-extern const vpx_prob vp10_default_palette_y_color_prob
-[PALETTE_MAX_SIZE - 1][PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 1];
-extern const vpx_prob vp10_default_palette_uv_color_prob
-[PALETTE_MAX_SIZE - 1][PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 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)];
-extern const vpx_tree_index vp10_palette_size_tree[TREE_SIZE(PALETTE_SIZES)];
-extern const vpx_tree_index
-vp10_palette_color_tree[PALETTE_MAX_SIZE - 1][TREE_SIZE(PALETTE_COLORS)];
void vp10_setup_past_independence(struct VP10Common *cm);
@@ -147,9 +128,6 @@
return i;
}
-int vp10_get_palette_color_context(const uint8_t *color_map, int cols,
- int r, int c, int n, int *color_order);
-
#ifdef __cplusplus
} // extern "C"
#endif
diff --git a/vp10/common/enums.h b/vp10/common/enums.h
index 5f67e30..a226a2d 100644
--- a/vp10/common/enums.h
+++ b/vp10/common/enums.h
@@ -109,29 +109,6 @@
PLANE_TYPES
} PLANE_TYPE;
-typedef enum {
- TWO_COLORS,
- THREE_COLORS,
- FOUR_COLORS,
- FIVE_COLORS,
- SIX_COLORS,
- SEVEN_COLORS,
- EIGHT_COLORS,
- PALETTE_SIZES
-} PALETTE_SIZE;
-
-typedef enum {
- PALETTE_COLOR_ONE,
- PALETTE_COLOR_TWO,
- PALETTE_COLOR_THREE,
- PALETTE_COLOR_FOUR,
- PALETTE_COLOR_FIVE,
- PALETTE_COLOR_SIX,
- PALETTE_COLOR_SEVEN,
- PALETTE_COLOR_EIGHT,
- PALETTE_COLORS
-} PALETTE_COLOR;
-
#define DC_PRED 0 // Average of above and left pixels
#define V_PRED 1 // Vertical
#define H_PRED 2 // Horizontal
diff --git a/vp10/common/onyxc_int.h b/vp10/common/onyxc_int.h
index 6814133..ffef733 100644
--- a/vp10/common/onyxc_int.h
+++ b/vp10/common/onyxc_int.h
@@ -185,8 +185,6 @@
int allow_high_precision_mv;
- int allow_screen_content_tools;
-
// Flag signaling which frame contexts should be reset to default values.
RESET_FRAME_CONTEXT_MODE reset_frame_context;
diff --git a/vp10/common/reconintra.c b/vp10/common/reconintra.c
index 14af7eb..e9e3949 100644
--- a/vp10/common/reconintra.c
+++ b/vp10/common/reconintra.c
@@ -262,7 +262,7 @@
}
#if CONFIG_MISC_FIXES
-static inline void memset16(uint16_t *dst, int val, int n) {
+static INLINE void memset16(uint16_t *dst, int val, int n) {
while (n--)
*dst++ = val;
}
@@ -753,40 +753,6 @@
const int have_right = (aoff + txw) < bw;
#endif // CONFIG_MISC_FIXES
- if (xd->mi[0]->mbmi.palette_mode_info.palette_size[plane != 0] > 0) {
- const int bs = 4 * (1 << tx_size);
- const int stride = 4 * (1 << bwl_in);
- int r, c;
- uint8_t *map = NULL;
-#if CONFIG_VP9_HIGHBITDEPTH
- uint16_t *palette = xd->mi[0]->mbmi.palette_mode_info.palette_colors +
- plane * PALETTE_MAX_SIZE;
-#else
- uint8_t *palette = xd->mi[0]->mbmi.palette_mode_info.palette_colors +
- plane * PALETTE_MAX_SIZE;
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
- map = xd->plane[plane != 0].color_index_map;
-
-#if CONFIG_VP9_HIGHBITDEPTH
- if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
- uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst);
- for (r = 0; r < bs; ++r)
- for (c = 0; c < bs; ++c)
- dst16[r * dst_stride + c] =
- palette[map[(r + y) * stride + c + x]];
- } else {
-#endif // CONFIG_VP9_HIGHBITDEPTH
- for (r = 0; r < bs; ++r)
- for (c = 0; c < bs; ++c)
- dst[r * dst_stride + c] = palette[map[(r + y) * stride + c + x]];
-#if CONFIG_VP9_HIGHBITDEPTH
- }
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
- return;
- }
-
#if CONFIG_MISC_FIXES
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
diff --git a/vp10/decoder/decodeframe.c b/vp10/decoder/decodeframe.c
index 5c95e16..70d012b 100644
--- a/vp10/decoder/decodeframe.c
+++ b/vp10/decoder/decodeframe.c
@@ -851,9 +851,6 @@
const int max_blocks_high = num_4x4_h + (xd->mb_to_bottom_edge >= 0 ?
0 : xd->mb_to_bottom_edge >> (5 + pd->subsampling_y));
- if (plane <= 1 && mbmi->palette_mode_info.palette_size[plane])
- vp10_decode_palette_tokens(xd, plane, r);
-
for (row = 0; row < max_blocks_high; row += step)
for (col = 0; col < max_blocks_wide; col += step)
predict_and_reconstruct_intra_block(xd, r, mbmi, plane,
@@ -1155,16 +1152,12 @@
cm->uv_dc_delta_q = read_delta_q(rb);
cm->uv_ac_delta_q = read_delta_q(rb);
cm->dequant_bit_depth = cm->bit_depth;
- for (i = 0; i < (cm->seg.enabled ? MAX_SEGMENTS : 1); ++i) {
-#if CONFIG_MISC_FIXES
- const int qindex = vp10_get_qindex(&cm->seg, i, cm->base_qindex);
-#endif
- xd->lossless[i] = cm->y_dc_delta_q == 0 &&
-#if CONFIG_MISC_FIXES
- qindex == 0 &&
-#else
- cm->base_qindex == 0 &&
-#endif
+ for (i = 0; i < MAX_SEGMENTS; ++i) {
+ const int qindex = CONFIG_MISC_FIXES && cm->seg.enabled ?
+ vp10_get_qindex(&cm->seg, i, cm->base_qindex) :
+ cm->base_qindex;
+ xd->lossless[i] = qindex == 0 &&
+ cm->y_dc_delta_q == 0 &&
cm->uv_dc_delta_q == 0 &&
cm->uv_ac_delta_q == 0;
}
@@ -1954,8 +1947,6 @@
memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
pbi->need_resync = 0;
}
- if (frame_is_intra_only(cm))
- cm->allow_screen_content_tools = vpx_rb_read_bit(rb);
} else {
cm->intra_only = cm->show_frame ? 0 : vpx_rb_read_bit(rb);
diff --git a/vp10/decoder/decodemv.c b/vp10/decoder/decodemv.c
index 38ea073..b516333 100644
--- a/vp10/decoder/decodemv.c
+++ b/vp10/decoder/decodemv.c
@@ -246,38 +246,6 @@
}
}
-static void read_palette_mode_info(VP10_COMMON *const cm,
- MACROBLOCKD *const xd,
- 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, palette_ctx = 0;
-
- if (above_mi)
- palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- if (left_mi)
- palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- if (vpx_read(r, vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
- [palette_ctx])) {
- int n;
- PALETTE_MODE_INFO *pmi = &mbmi->palette_mode_info;
-
- pmi->palette_size[0] =
- vpx_read_tree(r, vp10_palette_size_tree,
- vp10_default_palette_y_size_prob[bsize - BLOCK_8X8]) + 2;
- n = pmi->palette_size[0];
-
- for (i = 0; i < n; ++i)
- pmi->palette_colors[i] = vpx_read_literal(r, cm->bit_depth);
-
- xd->plane[0].color_index_map[0] = read_uniform(r, n);
- assert(xd->plane[0].color_index_map[0] < n);
- }
-}
-
static void read_intra_frame_mode_info(VP10_COMMON *const cm,
MACROBLOCKD *const xd,
int mi_row, int mi_col, vpx_reader *r) {
@@ -326,12 +294,6 @@
}
mbmi->uv_mode = read_intra_mode_uv(cm, xd, r, mbmi->mode);
-
- mbmi->palette_mode_info.palette_size[0] = 0;
- mbmi->palette_mode_info.palette_size[1] = 0;
- if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools &&
- mbmi->mode == DC_PRED)
- read_palette_mode_info(cm, xd, r);
}
static int read_mv_component(vpx_reader *r,
@@ -495,9 +457,6 @@
}
mbmi->uv_mode = read_intra_mode_uv(cm, xd, r, mbmi->mode);
-
- mbmi->palette_mode_info.palette_size[0] = 0;
- mbmi->palette_mode_info.palette_size[1] = 0;
}
static INLINE int is_mv_valid(const MV *mv) {
diff --git a/vp10/decoder/detokenize.c b/vp10/decoder/detokenize.c
index cda9888..d39e3dc 100644
--- a/vp10/decoder/detokenize.c
+++ b/vp10/decoder/detokenize.c
@@ -257,33 +257,6 @@
}
}
-void vp10_decode_palette_tokens(MACROBLOCKD *const xd, int plane,
- vpx_reader *r) {
- MODE_INFO *const mi = xd->mi[0];
- MB_MODE_INFO *const mbmi = &mi->mbmi;
- const BLOCK_SIZE bsize = mbmi->sb_type;
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
- int color_idx, color_ctx, color_order[PALETTE_MAX_SIZE];
- int n = mbmi->palette_mode_info.palette_size[plane != 0];
- int i, j;
- uint8_t *color_map = xd->plane[plane].color_index_map;
- const vpx_prob (* prob)[PALETTE_COLOR_CONTEXTS][PALETTE_COLORS - 1] =
- plane ? vp10_default_palette_uv_color_prob :
- vp10_default_palette_y_color_prob;
-
- for (i = 0; i < rows; ++i) {
- for (j = (i == 0 ? 1 : 0); j < cols; ++j) {
- color_ctx = vp10_get_palette_color_context(color_map, cols, i, j, n,
- color_order);
- color_idx = vpx_read_tree(r, vp10_palette_color_tree[n - 2],
- prob[n - 2][color_ctx]);
- assert(color_idx >= 0 && color_idx < n);
- color_map[i * cols + j] = color_order[color_idx];
- }
- }
-}
-
int vp10_decode_block_tokens(MACROBLOCKD *xd,
int plane, const scan_order *sc,
int x, int y,
diff --git a/vp10/decoder/detokenize.h b/vp10/decoder/detokenize.h
index d2677f6..c3fd90a 100644
--- a/vp10/decoder/detokenize.h
+++ b/vp10/decoder/detokenize.h
@@ -20,8 +20,6 @@
extern "C" {
#endif
-void vp10_decode_palette_tokens(MACROBLOCKD *const xd, int plane,
- vpx_reader *r);
int vp10_decode_block_tokens(MACROBLOCKD *xd,
int plane, const scan_order *sc,
int x, int y,
diff --git a/vp10/encoder/bitstream.c b/vp10/encoder/bitstream.c
index d9b4be4..5f6d9d3 100644
--- a/vp10/encoder/bitstream.c
+++ b/vp10/encoder/bitstream.c
@@ -44,20 +44,6 @@
{{0, 1}, {2, 2}, {6, 3}, {7, 3}};
static const struct vp10_token inter_mode_encodings[INTER_MODES] =
{{2, 2}, {6, 3}, {0, 1}, {7, 3}};
-static const struct vp10_token palette_size_encodings[] = {
- {0, 1}, {2, 2}, {6, 3}, {14, 4}, {30, 5}, {62, 6}, {63, 6},
-};
-static const struct vp10_token
-palette_color_encodings[PALETTE_MAX_SIZE - 1][PALETTE_MAX_SIZE] = {
- {{0, 1}, {1, 1}}, // 2 colors
- {{0, 1}, {2, 2}, {3, 2}}, // 3 colors
- {{0, 1}, {2, 2}, {6, 3}, {7, 3}}, // 4 colors
- {{0, 1}, {2, 2}, {6, 3}, {14, 4}, {15, 4}}, // 5 colors
- {{0, 1}, {2, 2}, {6, 3}, {14, 4}, {30, 5}, {31, 5}}, // 6 colors
- {{0, 1}, {2, 2}, {6, 3}, {14, 4}, {30, 5}, {62, 6}, {63, 6}}, // 7 colors
- {{0, 1}, {2, 2}, {6, 3}, {14, 4},
- {30, 5}, {62, 6}, {126, 7}, {127, 7}}, // 8 colors
-};
static INLINE void write_uniform(vpx_writer *w, int n, int v) {
int l = get_unsigned_bits(n);
@@ -147,22 +133,6 @@
counts->switchable_interp[j], SWITCHABLE_FILTERS, w);
}
-static void pack_palette_tokens(vpx_writer *w, TOKENEXTRA **tp,
- BLOCK_SIZE bsize, int n) {
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
- int i;
- TOKENEXTRA *p = *tp;
-
- for (i = 0; i < rows * cols -1; ++i) {
- vp10_write_token(w, vp10_palette_color_tree[n - 2], p->context_tree,
- &palette_color_encodings[n - 2][p->token]);
- ++p;
- }
-
- *tp = p;
-}
-
static void pack_mb_tokens(vpx_writer *w,
TOKENEXTRA **tp, const TOKENEXTRA *const stop,
vpx_bit_depth_t bit_depth, const TX_SIZE tx) {
@@ -402,36 +372,6 @@
}
}
-static void write_palette_mode_info(const VP10_COMMON *cm,
- const MACROBLOCKD *xd,
- const MODE_INFO *const mi,
- vpx_writer *w) {
- const MB_MODE_INFO *const mbmi = &mi->mbmi;
- const MODE_INFO *const above_mi = xd->above_mi;
- const MODE_INFO *const left_mi = xd->left_mi;
- const BLOCK_SIZE bsize = mbmi->sb_type;
- const PALETTE_MODE_INFO *pmi = &mbmi->palette_mode_info;
- int palette_ctx = 0;
- int n, i;
-
- n = pmi->palette_size[0];
- if (above_mi)
- palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- if (left_mi)
- palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- vpx_write(w, n > 0,
- vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8][palette_ctx]);
- if (n > 0) {
- vp10_write_token(w, vp10_palette_size_tree,
- vp10_default_palette_y_size_prob[bsize - BLOCK_8X8],
- &palette_size_encodings[n - 2]);
- for (i = 0; i < n; ++i)
- vpx_write_literal(w, pmi->palette_colors[i],
- cm->bit_depth);
- write_uniform(w, n, pmi->palette_first_color_idx[0]);
- }
-}
-
static void write_mb_modes_kf(const VP10_COMMON *cm, const MACROBLOCKD *xd,
MODE_INFO **mi_8x8, vpx_writer *w) {
const struct segmentation *const seg = &cm->seg;
@@ -472,10 +412,6 @@
}
write_intra_mode(w, mbmi->uv_mode, cm->fc->uv_mode_prob[mbmi->mode]);
-
- if (bsize >= BLOCK_8X8 && cm->allow_screen_content_tools &&
- mbmi->mode == DC_PRED)
- write_palette_mode_info(cm, xd, mi, w);
}
static void write_modes_b(VP10_COMP *cpi, const TileInfo *const tile,
@@ -502,13 +438,6 @@
pack_inter_mode_mvs(cpi, m, w);
}
- if (m->mbmi.palette_mode_info.palette_size[0] > 0) {
- assert(*tok < tok_end);
- pack_palette_tokens(w, tok, m->mbmi.sb_type,
- m->mbmi.palette_mode_info.palette_size[0]);
- assert(*tok < tok_end);
- }
-
if (!m->mbmi.skip) {
assert(*tok < tok_end);
for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
@@ -1259,8 +1188,6 @@
write_sync_code(wb);
write_bitdepth_colorspace_sampling(cm, wb);
write_frame_size(cm, wb);
- if (frame_is_intra_only(cm))
- vpx_wb_write_bit(wb, cm->allow_screen_content_tools);
} else {
if (!cm->show_frame)
vpx_wb_write_bit(wb, cm->intra_only);
@@ -1546,7 +1473,7 @@
assert(n_log2_tiles > 0);
vpx_wb_write_literal(&saved_wb, mag, 2);
if (mag < 3)
- data_sz = remux_tiles(data, data_sz, 1 << n_log2_tiles, mag);
+ data_sz = remux_tiles(data, (int)data_sz, 1 << n_log2_tiles, mag);
} else {
assert(n_log2_tiles == 0);
}
diff --git a/vp10/encoder/block.h b/vp10/encoder/block.h
index df3830c..ab0252b 100644
--- a/vp10/encoder/block.h
+++ b/vp10/encoder/block.h
@@ -52,13 +52,6 @@
uint8_t mode_context[MAX_REF_FRAMES];
} MB_MODE_INFO_EXT;
-typedef struct {
- uint8_t best_palette_color_map[4096];
- double kmeans_data_buf[4096];
- uint8_t kmeans_indices_buf[4096];
- uint8_t kmeans_pre_indices_buf[4096];
-} PALETTE_BUFFER;
-
typedef struct macroblock MACROBLOCK;
struct macroblock {
struct macroblock_plane plane[MAX_MB_PLANE];
@@ -77,6 +70,8 @@
int rddiv;
int rdmult;
int mb_energy;
+ int * m_search_count_ptr;
+ int * ex_search_count_ptr;
// These are set to their default values at the beginning, and then adjusted
// further in the encoding process.
@@ -99,8 +94,6 @@
int *nmvsadcost_hp[2];
int **mvsadcost;
- PALETTE_BUFFER *palette_buffer;
-
// These define limits to motion vector components to prevent them
// from extending outside the UMV borders
int mv_col_min;
diff --git a/vp10/encoder/dct.c b/vp10/encoder/dct.c
index c61babe..132a141 100644
--- a/vp10/encoder/dct.c
+++ b/vp10/encoder/dct.c
@@ -22,7 +22,10 @@
static INLINE void range_check(const tran_low_t *input, const int size,
const int bit) {
-#if CONFIG_COEFFICIENT_RANGE_CHECKING
+#if 0 // CONFIG_COEFFICIENT_RANGE_CHECKING
+// TODO(angiebird): the range_check is not used because the bit range
+// in fdct# is not correct. Since we are going to merge in a new version
+// of fdct# from nextgenv2, we won't fix the incorrect bit range now.
int i;
for (i = 0; i < size; ++i) {
assert(abs(input[i]) < (1 << bit));
@@ -39,7 +42,7 @@
tran_low_t step[4];
// stage 0
- range_check(input, 4, 11);
+ range_check(input, 4, 14);
// stage 1
output[0] = input[0] + input[3];
@@ -47,7 +50,7 @@
output[2] = input[1] - input[2];
output[3] = input[0] - input[3];
- range_check(output, 4, 12);
+ range_check(output, 4, 15);
// stage 2
temp = output[0] * cospi_16_64 + output[1] * cospi_16_64;
@@ -59,7 +62,7 @@
temp = output[3] * cospi_24_64 + output[2] * -cospi_8_64;
step[3] = (tran_low_t)fdct_round_shift(temp);
- range_check(step, 4, 13);
+ range_check(step, 4, 16);
// stage 3
output[0] = step[0];
@@ -67,7 +70,7 @@
output[2] = step[1];
output[3] = step[3];
- range_check(output, 4, 13);
+ range_check(output, 4, 16);
}
static void fdct8(const tran_low_t *input, tran_low_t *output) {
@@ -75,7 +78,7 @@
tran_low_t step[8];
// stage 0
- range_check(input, 8, 12);
+ range_check(input, 8, 13);
// stage 1
output[0] = input[0] + input[7];
@@ -87,7 +90,7 @@
output[6] = input[1] - input[6];
output[7] = input[0] - input[7];
- range_check(output, 8, 13);
+ range_check(output, 8, 14);
// stage 2
step[0] = output[0] + output[3];
@@ -101,7 +104,7 @@
step[6] = (tran_low_t)fdct_round_shift(temp);
step[7] = output[7];
- range_check(step, 8, 14);
+ range_check(step, 8, 15);
// stage 3
temp = step[0] * cospi_16_64 + step[1] * cospi_16_64;
@@ -117,7 +120,7 @@
output[6] = step[7] - step[6];
output[7] = step[7] + step[6];
- range_check(output, 8, 14);
+ range_check(output, 8, 16);
// stage 4
step[0] = output[0];
@@ -133,7 +136,7 @@
temp = output[7] * cospi_28_64 + output[4] * -cospi_4_64;
step[7] = (tran_low_t)fdct_round_shift(temp);
- range_check(step, 8, 14);
+ range_check(step, 8, 16);
// stage 5
output[0] = step[0];
@@ -145,7 +148,7 @@
output[6] = step[3];
output[7] = step[7];
- range_check(output, 8, 14);
+ range_check(output, 8, 16);
}
static void fdct16(const tran_low_t *input, tran_low_t *output) {
@@ -322,7 +325,7 @@
range_check(output, 16, 16);
}
-/* #TODO(angiebird): Unify this with vp10_fwd_txfm.c: vp10_fdct32
+/* TODO(angiebird): Unify this with vp10_fwd_txfm.c: vp10_fdct32
static void fdct32(const tran_low_t *input, tran_low_t *output) {
tran_high_t temp;
tran_low_t step[32];
diff --git a/vp10/encoder/encodeframe.c b/vp10/encoder/encodeframe.c
index f303b01..44ca276 100644
--- a/vp10/encoder/encodeframe.c
+++ b/vp10/encoder/encodeframe.c
@@ -1140,15 +1140,6 @@
p[i].eobs = ctx->eobs_pbuf[i][0];
}
- if (cm->current_video_frame == 0 && cm->allow_screen_content_tools) {
- for (i = 0; i < 2; ++i) {
- if (ctx->color_index_map[i] == 0) {
- CHECK_MEM_ERROR(cm, ctx->color_index_map[i],
- vpx_memalign(16, (ctx->num_4x4_blk << 4) *
- sizeof(*ctx->color_index_map[i])));
- }
- }
- }
for (i = 0; i < 2; ++i)
pd[i].color_index_map = ctx->color_index_map[i];
@@ -1164,7 +1155,7 @@
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
x->source_variance =
vp10_high_get_sby_perpixel_variance(cpi, &x->plane[0].src,
- bsize, xd->bd);
+ bsize, xd->bd);
} else {
x->source_variance =
vp10_get_sby_perpixel_variance(cpi, &x->plane[0].src, bsize);
@@ -2588,7 +2579,7 @@
}
static TX_MODE select_tx_mode(const VP10_COMP *cpi, MACROBLOCKD *const xd) {
- if (!cpi->common.seg.enabled && xd->lossless[0])
+ if (xd->lossless[0])
return ONLY_4X4;
if (cpi->sf.tx_size_search_method == USE_LARGESTALL)
return ALLOW_32X32;
@@ -2651,6 +2642,10 @@
TOKENEXTRA *tok = cpi->tile_tok[tile_row][tile_col];
int mi_row;
+ // Set up pointers to per thread motion search counters.
+ td->mb.m_search_count_ptr = &td->rd_counts.m_search_count;
+ td->mb.ex_search_count_ptr = &td->rd_counts.ex_search_count;
+
for (mi_row = tile_info->mi_row_start; mi_row < tile_info->mi_row_end;
mi_row += MI_BLOCK_SIZE) {
encode_rd_sb_row(cpi, td, this_tile, mi_row, &tok);
@@ -2704,17 +2699,15 @@
vp10_zero(rdc->coef_counts);
vp10_zero(rdc->comp_pred_diff);
vp10_zero(rdc->filter_diff);
+ rdc->m_search_count = 0; // Count of motion search hits.
+ rdc->ex_search_count = 0; // Exhaustive mesh search hits.
- for (i = 0; i < (cm->seg.enabled ? MAX_SEGMENTS : 1); ++i) {
-#if CONFIG_MISC_FIXES
- const int qindex = vp10_get_qindex(&cm->seg, i, cm->base_qindex);
-#endif
- xd->lossless[i] = cm->y_dc_delta_q == 0 &&
-#if CONFIG_MISC_FIXES
- qindex == 0 &&
-#else
- cm->base_qindex == 0 &&
-#endif
+ for (i = 0; i < MAX_SEGMENTS; ++i) {
+ const int qindex = CONFIG_MISC_FIXES && cm->seg.enabled ?
+ vp10_get_qindex(&cm->seg, i, cm->base_qindex) :
+ cm->base_qindex;
+ xd->lossless[i] = qindex == 0 &&
+ cm->y_dc_delta_q == 0 &&
cm->uv_dc_delta_q == 0 &&
cm->uv_ac_delta_q == 0;
}
@@ -2984,16 +2977,6 @@
if (output_enabled)
sum_intra_stats(td->counts, mi, xd->above_mi, xd->left_mi,
frame_is_intra_only(cm));
-
- if (bsize >= BLOCK_8X8 && output_enabled) {
- if (mbmi->palette_mode_info.palette_size[0] > 0) {
- mbmi->palette_mode_info.palette_first_color_idx[0] =
- xd->plane[0].color_index_map[0];
- // TODO(huisu): this increases the use of token buffer. Needs stretch
- // test to verify.
- vp10_tokenize_palette_sb(td, bsize, 0, t);
- }
- }
vp10_tokenize_sb(cpi, td, t, !output_enabled, VPXMAX(bsize, BLOCK_8X8));
} else {
int ref;
diff --git a/vp10/encoder/encoder.c b/vp10/encoder/encoder.c
index 5b646a2..6bba848 100644
--- a/vp10/encoder/encoder.c
+++ b/vp10/encoder/encoder.c
@@ -390,9 +390,6 @@
vp10_free_pc_tree(&cpi->td);
- if (cpi->common.allow_screen_content_tools)
- vpx_free(cpi->td.mb.palette_buffer);
-
if (cpi->source_diff_var != NULL) {
vpx_free(cpi->source_diff_var);
cpi->source_diff_var = NULL;
@@ -1435,15 +1432,6 @@
: REFRESH_FRAME_CONTEXT_BACKWARD;
cm->reset_frame_context = RESET_FRAME_CONTEXT_NONE;
- cm->allow_screen_content_tools = (cpi->oxcf.content == VP9E_CONTENT_SCREEN);
- if (cm->allow_screen_content_tools) {
- MACROBLOCK *x = &cpi->td.mb;
- if (x->palette_buffer == 0) {
- CHECK_MEM_ERROR(cm, x->palette_buffer,
- vpx_memalign(16, sizeof(*x->palette_buffer)));
- }
- }
-
vp10_reset_segment_features(cm);
vp10_set_high_precision_mv(cpi, 0);
@@ -1949,8 +1937,6 @@
// Deallocate allocated thread data.
if (t < cpi->num_workers - 1) {
- if (cpi->common.allow_screen_content_tools)
- vpx_free(thread_data->td->mb.palette_buffer);
vpx_free(thread_data->td->counts);
vp10_free_pc_tree(thread_data->td);
vpx_free(thread_data->td);
@@ -2857,7 +2843,7 @@
recon_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
if (cpi->twopass.total_left_stats.coded_error != 0.0)
- fprintf(f, "%10u %dx%d %d %d %10d %10d %10d %10d"
+ fprintf(f, "%10u %dx%d %10d %10d %d %d %10d %10d %10d %10d"
"%10"PRId64" %10"PRId64" %5d %5d %10"PRId64" "
"%10"PRId64" %10"PRId64" %10d "
"%7.2lf %7.2lf %7.2lf %7.2lf %7.2lf"
@@ -2866,6 +2852,8 @@
"%10lf %8u %10"PRId64" %10d %10d %10d\n",
cpi->common.current_video_frame,
cm->width, cm->height,
+ cpi->td.rd_counts.m_search_count,
+ cpi->td.rd_counts.ex_search_count,
cpi->rc.source_alt_ref_pending,
cpi->rc.source_alt_ref_active,
cpi->rc.this_frame_target,
diff --git a/vp10/encoder/encoder.h b/vp10/encoder/encoder.h
index 7028803..2a44e47 100644
--- a/vp10/encoder/encoder.h
+++ b/vp10/encoder/encoder.h
@@ -250,6 +250,8 @@
vp10_coeff_count coef_counts[TX_SIZES][PLANE_TYPES];
int64_t comp_pred_diff[REFERENCE_MODES];
int64_t filter_diff[SWITCHABLE_FILTER_CONTEXTS];
+ int m_search_count;
+ int ex_search_count;
} RD_COUNTS;
typedef struct ThreadData {
@@ -460,12 +462,6 @@
int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
int partition_cost[PARTITION_CONTEXTS][PARTITION_TYPES];
- int palette_y_size_cost[PALETTE_BLOCK_SIZES][PALETTE_SIZES];
- int palette_uv_size_cost[PALETTE_BLOCK_SIZES][PALETTE_SIZES];
- int palette_y_color_cost[PALETTE_MAX_SIZE - 1][PALETTE_COLOR_CONTEXTS]
- [PALETTE_COLORS];
- int palette_uv_color_cost[PALETTE_MAX_SIZE - 1][PALETTE_COLOR_CONTEXTS]
- [PALETTE_COLORS];
int multi_arf_allowed;
int multi_arf_enabled;
diff --git a/vp10/encoder/ethread.c b/vp10/encoder/ethread.c
index e20d532..ad47ccf 100644
--- a/vp10/encoder/ethread.c
+++ b/vp10/encoder/ethread.c
@@ -30,6 +30,11 @@
for (n = 0; n < ENTROPY_TOKENS; n++)
td->rd_counts.coef_counts[i][j][k][l][m][n] +=
td_t->rd_counts.coef_counts[i][j][k][l][m][n];
+
+
+ // Counts of all motion searches and exhuastive mesh searches.
+ td->rd_counts.m_search_count += td_t->rd_counts.m_search_count;
+ td->rd_counts.ex_search_count += td_t->rd_counts.ex_search_count;
}
static int enc_worker_hook(EncWorkerData *const thread_data, void *unused) {
@@ -128,13 +133,6 @@
memcpy(thread_data->td->counts, &cpi->common.counts,
sizeof(cpi->common.counts));
}
-
- // Allocate buffers used by palette coding mode.
- if (cpi->common.allow_screen_content_tools && i < num_workers - 1) {
- MACROBLOCK *x = &thread_data->td->mb;
- CHECK_MEM_ERROR(cm, x->palette_buffer,
- vpx_memalign(16, sizeof(*x->palette_buffer)));
- }
}
// Encode a frame
diff --git a/vp10/encoder/mcomp.c b/vp10/encoder/mcomp.c
index d6ab00f..04e1daf 100644
--- a/vp10/encoder/mcomp.c
+++ b/vp10/encoder/mcomp.c
@@ -1523,69 +1523,83 @@
#undef CHECK_BETTER
-int vp10_full_range_search_c(const MACROBLOCK *x,
- const search_site_config *cfg,
- MV *ref_mv, MV *best_mv,
- int search_param, int sad_per_bit, int *num00,
- const vp9_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv) {
+// Exhuastive motion search around a given centre position with a given
+// step size.
+static int exhuastive_mesh_search(const MACROBLOCK *x,
+ MV *ref_mv, MV *best_mv,
+ int range, int step, int sad_per_bit,
+ const vp9_variance_fn_ptr_t *fn_ptr,
+ const MV *center_mv) {
const MACROBLOCKD *const xd = &x->e_mbd;
const struct buf_2d *const what = &x->plane[0].src;
const struct buf_2d *const in_what = &xd->plane[0].pre[0];
- const int range = 64;
- const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
+ MV fcenter_mv = {center_mv->row, center_mv->col};
unsigned int best_sad = INT_MAX;
int r, c, i;
int start_col, end_col, start_row, end_row;
+ int col_step = (step > 1) ? step : 4;
- // The cfg and search_param parameters are not used in this search variant
- (void)cfg;
- (void)search_param;
+ assert(step >= 1);
- clamp_mv(ref_mv, x->mv_col_min, x->mv_col_max, x->mv_row_min, x->mv_row_max);
- *best_mv = *ref_mv;
- *num00 = 11;
+ clamp_mv(&fcenter_mv, x->mv_col_min, x->mv_col_max,
+ x->mv_row_min, x->mv_row_max);
+ *best_mv = fcenter_mv;
best_sad = fn_ptr->sdf(what->buf, what->stride,
- get_buf_from_mv(in_what, ref_mv), in_what->stride) +
- mvsad_err_cost(x, ref_mv, &fcenter_mv, sad_per_bit);
- start_row = VPXMAX(-range, x->mv_row_min - ref_mv->row);
- start_col = VPXMAX(-range, x->mv_col_min - ref_mv->col);
- end_row = VPXMIN(range, x->mv_row_max - ref_mv->row);
- end_col = VPXMIN(range, x->mv_col_max - ref_mv->col);
+ get_buf_from_mv(in_what, &fcenter_mv), in_what->stride) +
+ mvsad_err_cost(x, &fcenter_mv, ref_mv, sad_per_bit);
+ start_row = VPXMAX(-range, x->mv_row_min - fcenter_mv.row);
+ start_col = VPXMAX(-range, x->mv_col_min - fcenter_mv.col);
+ end_row = VPXMIN(range, x->mv_row_max - fcenter_mv.row);
+ end_col = VPXMIN(range, x->mv_col_max - fcenter_mv.col);
- for (r = start_row; r <= end_row; ++r) {
- for (c = start_col; c <= end_col; c += 4) {
- if (c + 3 <= end_col) {
- unsigned int sads[4];
- const uint8_t *addrs[4];
- for (i = 0; i < 4; ++i) {
- const MV mv = {ref_mv->row + r, ref_mv->col + c + i};
- addrs[i] = get_buf_from_mv(in_what, &mv);
- }
-
- fn_ptr->sdx4df(what->buf, what->stride, addrs, in_what->stride, sads);
-
- for (i = 0; i < 4; ++i) {
- if (sads[i] < best_sad) {
- const MV mv = {ref_mv->row + r, ref_mv->col + c + i};
- const unsigned int sad = sads[i] +
- mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
- if (sad < best_sad) {
- best_sad = sad;
- *best_mv = mv;
- }
+ for (r = start_row; r <= end_row; r += step) {
+ for (c = start_col; c <= end_col; c += col_step) {
+ // Step > 1 means we are not checking every location in this pass.
+ if (step > 1) {
+ const MV mv = {fcenter_mv.row + r, fcenter_mv.col + c};
+ unsigned int sad = fn_ptr->sdf(what->buf, what->stride,
+ get_buf_from_mv(in_what, &mv), in_what->stride);
+ if (sad < best_sad) {
+ sad += mvsad_err_cost(x, &mv, ref_mv, sad_per_bit);
+ if (sad < best_sad) {
+ best_sad = sad;
+ *best_mv = mv;
}
}
} else {
- for (i = 0; i < end_col - c; ++i) {
- const MV mv = {ref_mv->row + r, ref_mv->col + c + i};
- unsigned int sad = fn_ptr->sdf(what->buf, what->stride,
- get_buf_from_mv(in_what, &mv), in_what->stride);
- if (sad < best_sad) {
- sad += mvsad_err_cost(x, &mv, &fcenter_mv, sad_per_bit);
+ // 4 sads in a single call if we are checking every location
+ if (c + 3 <= end_col) {
+ unsigned int sads[4];
+ const uint8_t *addrs[4];
+ for (i = 0; i < 4; ++i) {
+ const MV mv = {fcenter_mv.row + r, fcenter_mv.col + c + i};
+ addrs[i] = get_buf_from_mv(in_what, &mv);
+ }
+ fn_ptr->sdx4df(what->buf, what->stride, addrs,
+ in_what->stride, sads);
+
+ for (i = 0; i < 4; ++i) {
+ if (sads[i] < best_sad) {
+ const MV mv = {fcenter_mv.row + r, fcenter_mv.col + c + i};
+ const unsigned int sad = sads[i] +
+ mvsad_err_cost(x, &mv, ref_mv, sad_per_bit);
+ if (sad < best_sad) {
+ best_sad = sad;
+ *best_mv = mv;
+ }
+ }
+ }
+ } else {
+ for (i = 0; i < end_col - c; ++i) {
+ const MV mv = {fcenter_mv.row + r, fcenter_mv.col + c + i};
+ unsigned int sad = fn_ptr->sdf(what->buf, what->stride,
+ get_buf_from_mv(in_what, &mv), in_what->stride);
if (sad < best_sad) {
- best_sad = sad;
- *best_mv = mv;
+ sad += mvsad_err_cost(x, &mv, ref_mv, sad_per_bit);
+ if (sad < best_sad) {
+ best_sad = sad;
+ *best_mv = mv;
+ }
}
}
}
@@ -2014,6 +2028,70 @@
return bestsme;
}
+#define MIN_RANGE 7
+#define MAX_RANGE 256
+#define MIN_INTERVAL 1
+// Runs an limited range exhaustive mesh search using a pattern set
+// according to the encode speed profile.
+static int full_pixel_exhaustive(VP10_COMP *cpi, MACROBLOCK *x,
+ MV *centre_mv_full, int sadpb, int *cost_list,
+ const vp9_variance_fn_ptr_t *fn_ptr,
+ const MV *ref_mv, MV *dst_mv) {
+ const SPEED_FEATURES *const sf = &cpi->sf;
+ MV temp_mv = {centre_mv_full->row, centre_mv_full->col};
+ MV f_ref_mv = {ref_mv->row >> 3, ref_mv->col >> 3};
+ int bestsme;
+ int i;
+ int interval = sf->mesh_patterns[0].interval;
+ int range = sf->mesh_patterns[0].range;
+ int baseline_interval_divisor;
+
+ // Keep track of number of exhaustive calls (this frame in this thread).
+ ++(*x->ex_search_count_ptr);
+
+ // Trap illegal values for interval and range for this function.
+ if ((range < MIN_RANGE) || (range > MAX_RANGE) ||
+ (interval < MIN_INTERVAL) || (interval > range))
+ return INT_MAX;
+
+ baseline_interval_divisor = range / interval;
+
+ // Check size of proposed first range against magnitude of the centre
+ // value used as a starting point.
+ range = VPXMAX(range, (5 * VPXMAX(abs(temp_mv.row), abs(temp_mv.col))) / 4);
+ range = VPXMIN(range, MAX_RANGE);
+ interval = VPXMAX(interval, range / baseline_interval_divisor);
+
+ // initial search
+ bestsme = exhuastive_mesh_search(x, &f_ref_mv, &temp_mv, range,
+ interval, sadpb, fn_ptr, &temp_mv);
+
+ if ((interval > MIN_INTERVAL) && (range > MIN_RANGE)) {
+ // Progressive searches with range and step size decreasing each time
+ // till we reach a step size of 1. Then break out.
+ for (i = 1; i < MAX_MESH_STEP; ++i) {
+ // First pass with coarser step and longer range
+ bestsme = exhuastive_mesh_search(x, &f_ref_mv, &temp_mv,
+ sf->mesh_patterns[i].range,
+ sf->mesh_patterns[i].interval,
+ sadpb, fn_ptr, &temp_mv);
+
+ if (sf->mesh_patterns[i].interval == 1)
+ break;
+ }
+ }
+
+ if (bestsme < INT_MAX)
+ bestsme = vp10_get_mvpred_var(x, &temp_mv, ref_mv, fn_ptr, 1);
+ *dst_mv = temp_mv;
+
+ // Return cost list.
+ if (cost_list) {
+ calc_int_cost_list(x, ref_mv, sadpb, fn_ptr, dst_mv, cost_list);
+ }
+ return bestsme;
+}
+
int vp10_full_search_sad_c(const MACROBLOCK *x, const MV *ref_mv,
int sad_per_bit, int distance,
const vp9_variance_fn_ptr_t *fn_ptr,
@@ -2327,6 +2405,18 @@
return best_sad;
}
+#define MIN_EX_SEARCH_LIMIT 128
+static int is_exhaustive_allowed(VP10_COMP *cpi, MACROBLOCK *x) {
+ const SPEED_FEATURES *const sf = &cpi->sf;
+ const int max_ex = VPXMAX(MIN_EX_SEARCH_LIMIT,
+ (*x->m_search_count_ptr * sf->max_exaustive_pct) / 100);
+
+ return sf->allow_exhaustive_searches &&
+ (sf->exhaustive_searches_thresh < INT_MAX) &&
+ (*x->ex_search_count_ptr <= max_ex) &&
+ !cpi->rc.is_src_frame_alt_ref;
+}
+
int vp10_full_pixel_search(VP10_COMP *cpi, MACROBLOCK *x,
BLOCK_SIZE bsize, MV *mvp_full,
int step_param, int error_per_bit,
@@ -2345,6 +2435,9 @@
cost_list[4] = INT_MAX;
}
+ // Keep track of number of searches (this frame in this thread).
+ ++(*x->m_search_count_ptr);
+
switch (method) {
case FAST_DIAMOND:
var = vp10_fast_dia_search(x, mvp_full, step_param, error_per_bit, 0,
@@ -2370,6 +2463,29 @@
var = vp10_full_pixel_diamond(cpi, x, mvp_full, step_param, error_per_bit,
MAX_MVSEARCH_STEPS - 1 - step_param,
1, cost_list, fn_ptr, ref_mv, tmp_mv);
+
+ // Should we allow a follow on exhaustive search?
+ if (is_exhaustive_allowed(cpi, x)) {
+ int64_t exhuastive_thr = sf->exhaustive_searches_thresh;
+ exhuastive_thr >>= 8 - (b_width_log2_lookup[bsize] +
+ b_height_log2_lookup[bsize]);
+
+ // Threshold variance for an exhaustive full search.
+ if (var > exhuastive_thr) {
+ int var_ex;
+ MV tmp_mv_ex;
+ var_ex = full_pixel_exhaustive(cpi, x, tmp_mv,
+ error_per_bit, cost_list, fn_ptr,
+ ref_mv, &tmp_mv_ex);
+
+ if (var_ex < var) {
+ var = var_ex;
+ *tmp_mv = tmp_mv_ex;
+ }
+ }
+ }
+ break;
+
break;
default:
assert(0 && "Invalid search method.");
diff --git a/vp10/encoder/palette.c b/vp10/encoder/palette.c
deleted file mode 100644
index 522e185..0000000
--- a/vp10/encoder/palette.c
+++ /dev/null
@@ -1,194 +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.
- */
-
-#include <math.h>
-#include "vp10/encoder/palette.h"
-
-static double calc_dist(const double *p1, const double *p2, int dim) {
- double dist = 0;
- int i = 0;
-
- for (i = 0; i < dim; ++i) {
- dist = dist + (p1[i] - round(p2[i])) * (p1[i] - round(p2[i]));
- }
- return dist;
-}
-
-void vp10_calc_indices(const double *data, const double *centroids,
- uint8_t *indices, int n, int k, int dim) {
- int i, j;
- double min_dist, this_dist;
-
- for (i = 0; i < n; ++i) {
- min_dist = calc_dist(data + i * dim, centroids, dim);
- indices[i] = 0;
- for (j = 1; j < k; ++j) {
- this_dist = calc_dist(data + i * dim, centroids + j * dim, dim);
- if (this_dist < min_dist) {
- min_dist = this_dist;
- indices[i] = j;
- }
- }
- }
-}
-
-// Generate a random number in the range [0, 32768).
-static unsigned int lcg_rand16(unsigned int *state) {
- *state = *state * 1103515245 + 12345;
- return *state / 65536 % 32768;
-}
-
-static void calc_centroids(const double *data, double *centroids,
- const uint8_t *indices, int n, int k, int dim) {
- int i, j, index;
- int count[PALETTE_MAX_SIZE];
- unsigned int rand_state = (unsigned int)data[0];
-
- assert(n <= 32768);
-
- memset(count, 0, sizeof(count[0]) * k);
- memset(centroids, 0, sizeof(centroids[0]) * k * dim);
-
- for (i = 0; i < n; ++i) {
- index = indices[i];
- assert(index < k);
- ++count[index];
- for (j = 0; j < dim; ++j) {
- centroids[index * dim + j] += data[i * dim + j];
- }
- }
-
- for (i = 0; i < k; ++i) {
- if (count[i] == 0) {
- memcpy(centroids + i * dim, data + (lcg_rand16(&rand_state) % n) * dim,
- sizeof(centroids[0]) * dim);
- } else {
- const double norm = 1.0 / count[i];
- for (j = 0; j < dim; ++j)
- centroids[i * dim + j] *= norm;
- }
- }
-}
-
-static double calc_total_dist(const double *data, const double *centroids,
- const uint8_t *indices, int n, int k, int dim) {
- double dist = 0;
- int i;
- (void) k;
-
- for (i = 0; i < n; ++i)
- dist += calc_dist(data + i * dim, centroids + indices[i] * dim, dim);
-
- return dist;
-}
-
-int vp10_k_means(const double *data, double *centroids, uint8_t *indices,
- uint8_t *pre_indices, int n, int k, int dim, int max_itr) {
- int i = 0;
- double pre_dist, this_dist;
- double pre_centroids[PALETTE_MAX_SIZE];
-
- vp10_calc_indices(data, centroids, indices, n, k, dim);
- pre_dist = calc_total_dist(data, centroids, indices, n, k, dim);
- memcpy(pre_centroids, centroids, sizeof(pre_centroids[0]) * k * dim);
- memcpy(pre_indices, indices, sizeof(pre_indices[0]) * n);
- while (i < max_itr) {
- calc_centroids(data, centroids, indices, n, k, dim);
- vp10_calc_indices(data, centroids, indices, n, k, dim);
- this_dist = calc_total_dist(data, centroids, indices, n, k, dim);
-
- if (this_dist > pre_dist) {
- memcpy(centroids, pre_centroids, sizeof(pre_centroids[0]) * k * dim);
- memcpy(indices, pre_indices, sizeof(pre_indices[0]) * n);
- break;
- }
- if (!memcmp(centroids, pre_centroids, sizeof(pre_centroids[0]) * k * dim))
- break;
-
- memcpy(pre_centroids, centroids, sizeof(pre_centroids[0]) * k * dim);
- memcpy(pre_indices, indices, sizeof(pre_indices[0]) * n);
- pre_dist = this_dist;
- ++i;
- }
-
- return i;
-}
-
-void vp10_insertion_sort(double *data, int n) {
- int i, j, k;
- double val;
-
- if (n <= 1)
- return;
-
- for (i = 1; i < n; ++i) {
- val = data[i];
- j = 0;
- while (val > data[j] && j < i)
- ++j;
-
- if (j == i)
- continue;
-
- for (k = i; k > j; --k)
- data[k] = data[k - 1];
- data[j] = val;
- }
-}
-
-int vp10_count_colors(const uint8_t *src, int stride, int rows, int cols) {
- int n = 0, r, c, i, val_count[256];
- uint8_t val;
- memset(val_count, 0, sizeof(val_count));
-
- for (r = 0; r < rows; ++r) {
- for (c = 0; c < cols; ++c) {
- val = src[r * stride + c];
- ++val_count[val];
- }
- }
-
- for (i = 0; i < 256; ++i) {
- if (val_count[i]) {
- ++n;
- }
- }
-
- return n;
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-int vp10_count_colors_highbd(const uint8_t *src8, int stride, int rows,
- int cols, int bit_depth) {
- int n = 0, r, c, i;
- uint16_t val;
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- int val_count[1 << 12];
-
- assert(bit_depth <= 12);
- memset(val_count, 0, (1 << 12) * sizeof(val_count[0]));
- for (r = 0; r < rows; ++r) {
- for (c = 0; c < cols; ++c) {
- val = src[r * stride + c];
- ++val_count[val];
- }
- }
-
- for (i = 0; i < (1 << bit_depth); ++i) {
- if (val_count[i]) {
- ++n;
- }
- }
-
- return n;
-}
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
-
diff --git a/vp10/encoder/palette.h b/vp10/encoder/palette.h
deleted file mode 100644
index 124cf74..0000000
--- a/vp10/encoder/palette.h
+++ /dev/null
@@ -1,35 +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_PALETTE_H_
-#define VP10_ENCODER_PALETTE_H_
-
-#include "vp10/common/blockd.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-void vp10_insertion_sort(double *data, int n);
-void vp10_calc_indices(const double *data, const double *centroids,
- uint8_t *indices, int n, int k, int dim);
-int vp10_k_means(const double *data, double *centroids, uint8_t *indices,
- uint8_t *pre_indices, int n, int k, int dim, int max_itr);
-int vp10_count_colors(const uint8_t *src, int stride, int rows, int cols);
-#if CONFIG_VP9_HIGHBITDEPTH
-int vp10_count_colors_highbd(const uint8_t *src8, int stride, int rows,
- int cols, int bit_depth);
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
-#ifdef __cplusplus
-} // extern "C"
-#endif
-
-#endif /* VP10_ENCODER_PALETTE_H_ */
diff --git a/vp10/encoder/ratectrl.c b/vp10/encoder/ratectrl.c
index d4c25c0..3ff2476 100644
--- a/vp10/encoder/ratectrl.c
+++ b/vp10/encoder/ratectrl.c
@@ -1210,8 +1210,12 @@
rc->frames_since_golden = 0;
// If we are not using alt ref in the up and coming group clear the arf
- // active flag.
- if (!rc->source_alt_ref_pending) {
+ // active flag. In multi arf group case, if the index is not 0 then
+ // we are overlaying a mid group arf so should not reset the flag.
+ if (cpi->oxcf.pass == 2) {
+ if (!rc->source_alt_ref_pending && (cpi->twopass.gf_group.index == 0))
+ rc->source_alt_ref_active = 0;
+ } else if (!rc->source_alt_ref_pending) {
rc->source_alt_ref_active = 0;
}
diff --git a/vp10/encoder/rd.c b/vp10/encoder/rd.c
index 4ed1ae2..5623a72 100644
--- a/vp10/encoder/rd.c
+++ b/vp10/encoder/rd.c
@@ -83,25 +83,6 @@
for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i)
vp10_cost_tokens(cpi->switchable_interp_costs[i],
fc->switchable_interp_prob[i], vp10_switchable_interp_tree);
-
- for (i = 0; i < PALETTE_BLOCK_SIZES; ++i) {
- vp10_cost_tokens(cpi->palette_y_size_cost[i],
- vp10_default_palette_y_size_prob[i],
- vp10_palette_size_tree);
- vp10_cost_tokens(cpi->palette_uv_size_cost[i],
- vp10_default_palette_uv_size_prob[i],
- vp10_palette_size_tree);
- }
-
- for (i = 0; i < PALETTE_MAX_SIZE - 1; ++i)
- for (j = 0; j < PALETTE_COLOR_CONTEXTS; ++j) {
- vp10_cost_tokens(cpi->palette_y_color_cost[i][j],
- vp10_default_palette_y_color_prob[i][j],
- vp10_palette_color_tree[i]);
- vp10_cost_tokens(cpi->palette_uv_color_cost[i][j],
- vp10_default_palette_uv_color_prob[i][j],
- vp10_palette_color_tree[i]);
- }
}
static void fill_token_costs(vp10_coeff_cost *c,
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index d87d164..bba2171 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -36,7 +36,6 @@
#include "vp10/encoder/encodemv.h"
#include "vp10/encoder/encoder.h"
#include "vp10/encoder/mcomp.h"
-#include "vp10/encoder/palette.h"
#include "vp10/encoder/quantize.h"
#include "vp10/encoder/ratectrl.h"
#include "vp10/encoder/rd.h"
@@ -760,163 +759,6 @@
return 0;
}
-void rd_pick_palette_intra_sby(VP10_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
- int palette_ctx, int dc_mode_cost,
- PALETTE_MODE_INFO *palette_mode_info,
- uint8_t *best_palette_color_map,
- TX_SIZE *best_tx, PREDICTION_MODE *mode_selected,
- int64_t *best_rd) {
- MACROBLOCKD *const xd = &x->e_mbd;
- MODE_INFO *const mic = xd->mi[0];
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
- int this_rate, this_rate_tokenonly, s;
- int64_t this_distortion, this_rd;
- int colors, n;
- int src_stride = x->plane[0].src.stride;
- uint8_t *src = x->plane[0].src.buf;
-
-#if CONFIG_VP9_HIGHBITDEPTH
- if (cpi->common.use_highbitdepth)
- colors = vp10_count_colors_highbd(src, src_stride, rows, cols,
- cpi->common.bit_depth);
- else
-#endif // CONFIG_VP9_HIGHBITDEPTH
- colors = vp10_count_colors(src, src_stride, rows, cols);
- palette_mode_info->palette_size[0] = 0;
-
- if (colors > 1 && colors <= 64 && cpi->common.allow_screen_content_tools) {
- int r, c, i, j, k;
- int max_itr = 50;
- int color_ctx, color_idx = 0;
- int color_order[PALETTE_MAX_SIZE];
- double *data = x->palette_buffer->kmeans_data_buf;
- uint8_t *indices = x->palette_buffer->kmeans_indices_buf;
- uint8_t *pre_indices = x->palette_buffer->kmeans_pre_indices_buf;
- double centroids[PALETTE_MAX_SIZE];
- uint8_t *color_map;
- double lb, ub, val;
- PALETTE_MODE_INFO *pmi = &mic->mbmi.palette_mode_info;
-#if CONFIG_VP9_HIGHBITDEPTH
- uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
- if (cpi->common.use_highbitdepth)
- lb = ub = src16[0];
- else
-#endif // CONFIG_VP9_HIGHBITDEPTH
- lb = ub = src[0];
-
-#if CONFIG_VP9_HIGHBITDEPTH
- if (cpi->common.use_highbitdepth) {
- for (r = 0; r < rows; ++r) {
- for (c = 0; c < cols; ++c) {
- val = src16[r * src_stride + c];
- data[r * cols + c] = val;
- if (val < lb)
- lb = val;
- else if (val > ub)
- ub = val;
- }
- }
- } else {
-#endif // CONFIG_VP9_HIGHBITDEPTH
- for (r = 0; r < rows; ++r) {
- for (c = 0; c < cols; ++c) {
- val = src[r * src_stride + c];
- data[r * cols + c] = val;
- if (val < lb)
- lb = val;
- else if (val > ub)
- ub = val;
- }
- }
-#if CONFIG_VP9_HIGHBITDEPTH
- }
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
- mic->mbmi.mode = DC_PRED;
-
- for (n = colors > PALETTE_MAX_SIZE ? PALETTE_MAX_SIZE : colors;
- n >= 2; --n) {
- for (i = 0; i < n; ++i)
- centroids[i] = lb + (2 * i + 1) * (ub - lb) / n / 2;
- vp10_k_means(data, centroids, indices, pre_indices, rows * cols,
- n, 1, max_itr);
- vp10_insertion_sort(centroids, n);
- for (i = 0; i < n; ++i)
- centroids[i] = round(centroids[i]);
- // remove duplicates
- i = 1;
- k = n;
- while (i < k) {
- if (centroids[i] == centroids[i - 1]) {
- j = i;
- while (j < k - 1) {
- centroids[j] = centroids[j + 1];
- ++j;
- }
- --k;
- } else {
- ++i;
- }
- }
-
-#if CONFIG_VP9_HIGHBITDEPTH
- if (cpi->common.use_highbitdepth)
- for (i = 0; i < k; ++i)
- mic->mbmi.palette_mode_info.palette_colors[i] =
- clip_pixel_highbd(round(centroids[i]), cpi->common.bit_depth);
- else
-#endif // CONFIG_VP9_HIGHBITDEPTH
- for (i = 0; i < k; ++i)
- pmi->palette_colors[i] = clip_pixel((int)round(centroids[i]));
- pmi->palette_size[0] = k;
-
- vp10_calc_indices(data, centroids, indices, rows * cols, k, 1);
- for (r = 0; r < rows; ++r)
- for (c = 0; c < cols; ++c)
- xd->plane[0].color_index_map[r * cols + c] = indices[r * cols + c];
-
- super_block_yrd(cpi, x, &this_rate_tokenonly, &this_distortion,
- &s, NULL, bsize, *best_rd);
- if (this_rate_tokenonly == INT_MAX)
- continue;
-
- this_rate = this_rate_tokenonly + dc_mode_cost +
- cpi->common.bit_depth * k * vp10_cost_bit(128, 0) +
- cpi->palette_y_size_cost[bsize - BLOCK_8X8][k - 2];
- this_rate +=
- vp10_cost_bit(vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
- [palette_ctx], 1);
- color_map = xd->plane[0].color_index_map;
- this_rate += write_uniform_cost(k, xd->plane[0].color_index_map[0]);
- for (i = 0; i < rows; ++i) {
- for (j = (i == 0 ? 1 : 0); j < cols; ++j) {
- color_ctx = vp10_get_palette_color_context(color_map, cols, i, j,
- k, color_order);
- for (r = 0; r < k; ++r)
- if (color_map[i * cols + j] == color_order[r]) {
- color_idx = r;
- break;
- }
- assert(color_idx < k);
- this_rate +=
- cpi->palette_y_color_cost[k - 2][color_ctx][color_idx];
- }
- }
- this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
-
- if (this_rd < *best_rd) {
- *best_rd = this_rd;
- *palette_mode_info = mic->mbmi.palette_mode_info;
- memcpy(best_palette_color_map, xd->plane[0].color_index_map,
- rows * cols * sizeof(xd->plane[0].color_index_map[0]));
- *mode_selected = DC_PRED;
- *best_tx = mic->mbmi.tx_size;
- }
- }
- }
-}
-
static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
int row, int col,
PREDICTION_MODE *best_mode,
@@ -947,7 +789,6 @@
memcpy(ta, a, sizeof(ta));
memcpy(tl, l, sizeof(tl));
xd->mi[0]->mbmi.tx_size = TX_4X4;
- xd->mi[0]->mbmi.palette_mode_info.palette_size[0] = 0;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
@@ -1225,12 +1066,6 @@
int64_t this_distortion, this_rd;
TX_SIZE best_tx = TX_4X4;
int *bmode_costs;
- PALETTE_MODE_INFO palette_mode_info;
- uint8_t *best_palette_color_map = cpi->common.allow_screen_content_tools ?
- x->palette_buffer->best_palette_color_map : NULL;
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
- int palette_ctx = 0;
const MODE_INFO *above_mi = xd->above_mi;
const MODE_INFO *left_mi = xd->left_mi;
const PREDICTION_MODE A = vp10_above_block_mode(mic, above_mi, 0);
@@ -1238,12 +1073,6 @@
bmode_costs = cpi->y_mode_costs[A][L];
memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
- palette_mode_info.palette_size[0] = 0;
- mic->mbmi.palette_mode_info.palette_size[0] = 0;
- if (above_mi)
- palette_ctx += (above_mi->mbmi.palette_mode_info.palette_size[0] > 0);
- if (left_mi)
- palette_ctx += (left_mi->mbmi.palette_mode_info.palette_size[0] > 0);
/* Y Search for intra prediction mode */
for (mode = DC_PRED; mode <= TM_PRED; mode++) {
@@ -1256,10 +1085,6 @@
continue;
this_rate = this_rate_tokenonly + bmode_costs[mode];
- if (cpi->common.allow_screen_content_tools && mode == DC_PRED)
- this_rate +=
- vp10_cost_bit(vp10_default_palette_y_mode_prob[bsize - BLOCK_8X8]
- [palette_ctx], 0);
this_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_distortion);
if (this_rd < best_rd) {
@@ -1273,22 +1098,8 @@
}
}
- if (cpi->common.allow_screen_content_tools)
- rd_pick_palette_intra_sby(cpi, x, bsize, palette_ctx, bmode_costs[DC_PRED],
- &palette_mode_info, best_palette_color_map,
- &best_tx, &mode_selected, &best_rd);
-
mic->mbmi.mode = mode_selected;
mic->mbmi.tx_size = best_tx;
- mic->mbmi.palette_mode_info.palette_size[0] =
- palette_mode_info.palette_size[0];
- if (palette_mode_info.palette_size[0] > 0) {
- memcpy(mic->mbmi.palette_mode_info.palette_colors,
- palette_mode_info.palette_colors,
- PALETTE_MAX_SIZE * sizeof(palette_mode_info.palette_colors[0]));
- memcpy(xd->plane[0].color_index_map, best_palette_color_map,
- rows * cols * sizeof(best_palette_color_map[0]));
- }
return best_rd;
}
@@ -1359,7 +1170,6 @@
int64_t this_distortion, this_sse;
memset(x->skip_txfm, SKIP_TXFM_NONE, sizeof(x->skip_txfm));
- xd->mi[0]->mbmi.palette_mode_info.palette_size[1] = 0;
for (mode = DC_PRED; mode <= TM_PRED; ++mode) {
if (!(cpi->sf.intra_uv_mode_mask[max_tx_size] & (1 << mode)))
continue;
@@ -1973,7 +1783,7 @@
seg_mvs[i][mbmi->ref_frame[0]].as_int == INVALID_MV) {
MV *const new_mv = &mode_mv[NEWMV][0].as_mv;
int step_param = 0;
- int thissme, bestsme = INT_MAX;
+ int bestsme = INT_MAX;
int sadpb = x->sadperbit4;
MV mvp_full;
int max_mv;
@@ -2028,27 +1838,6 @@
&bsi->ref_mv[0]->as_mv, new_mv,
INT_MAX, 1);
- // Should we do a full search (best quality only)
- if (cpi->oxcf.mode == BEST) {
- int_mv *const best_mv = &mi->bmi[i].as_mv[0];
- /* Check if mvp_full is within the range. */
- clamp_mv(&mvp_full, x->mv_col_min, x->mv_col_max,
- x->mv_row_min, x->mv_row_max);
- thissme = cpi->full_search_sad(x, &mvp_full,
- sadpb, 16, &cpi->fn_ptr[bsize],
- &bsi->ref_mv[0]->as_mv,
- &best_mv->as_mv);
- cost_list[1] = cost_list[2] = cost_list[3] = cost_list[4] = INT_MAX;
- if (thissme < bestsme) {
- bestsme = thissme;
- *new_mv = best_mv->as_mv;
- } else {
- // The full search result is actually worse so re-instate the
- // previous best vector
- best_mv->as_mv = *new_mv;
- }
- }
-
if (bestsme < INT_MAX) {
int distortion;
cpi->find_fractional_mv_step(
@@ -3292,8 +3081,6 @@
midx = end_pos;
}
- mbmi->palette_mode_info.palette_size[0] = 0;
- mbmi->palette_mode_info.palette_size[1] = 0;
for (midx = 0; midx < MAX_MODES; ++midx) {
int mode_index = mode_map[midx];
int mode_excluded = 0;
@@ -3773,8 +3560,6 @@
assert(segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP));
- mbmi->palette_mode_info.palette_size[0] = 0;
- mbmi->palette_mode_info.palette_size[1] = 0;
mbmi->mode = ZEROMV;
mbmi->uv_mode = DC_PRED;
mbmi->ref_frame[0] = LAST_FRAME;
@@ -3922,9 +3707,6 @@
frame_mv[ZEROMV][ref_frame].as_int = 0;
}
- mbmi->palette_mode_info.palette_size[0] = 0;
- mbmi->palette_mode_info.palette_size[1] = 0;
-
for (ref_index = 0; ref_index < MAX_REFS; ++ref_index) {
int mode_excluded = 0;
int64_t this_rd = INT64_MAX;
diff --git a/vp10/encoder/speed_features.c b/vp10/encoder/speed_features.c
index d40383f..ce0aebe 100644
--- a/vp10/encoder/speed_features.c
+++ b/vp10/encoder/speed_features.c
@@ -16,6 +16,23 @@
#include "vpx_dsp/vpx_dsp_common.h"
+// Mesh search patters for various speed settings
+static MESH_PATTERN best_quality_mesh_pattern[MAX_MESH_STEP] =
+ {{64, 4}, {28, 2}, {15, 1}, {7, 1}};
+
+#define MAX_MESH_SPEED 5 // Max speed setting for mesh motion method
+static MESH_PATTERN good_quality_mesh_patterns[MAX_MESH_SPEED + 1]
+ [MAX_MESH_STEP] =
+ {{{64, 8}, {28, 4}, {15, 1}, {7, 1}},
+ {{64, 8}, {28, 4}, {15, 1}, {7, 1}},
+ {{64, 8}, {14, 2}, {7, 1}, {7, 1}},
+ {{64, 16}, {24, 8}, {12, 4}, {7, 1}},
+ {{64, 16}, {24, 8}, {12, 4}, {7, 1}},
+ {{64, 16}, {24, 8}, {12, 4}, {7, 1}},
+ };
+static unsigned char good_quality_max_mesh_pct[MAX_MESH_SPEED + 1] =
+ {50, 25, 15, 5, 1, 1};
+
// Intra only frames, golden frames (except alt ref overlays) and
// alt ref frames tend to be coded at a higher than ambient quality
static int frame_is_boosted(const VP10_COMP *cpi) {
@@ -251,6 +268,8 @@
sf->static_segmentation = 0;
sf->adaptive_rd_thresh = 1;
sf->use_fast_coef_costing = 1;
+ sf->allow_exhaustive_searches = 0;
+ sf->exhaustive_searches_thresh = INT_MAX;
if (speed >= 1) {
sf->use_square_partition_only = !frame_is_intra_only(cm);
@@ -498,8 +517,36 @@
set_good_speed_feature(cpi, cm, sf, oxcf->speed);
cpi->full_search_sad = vp10_full_search_sad;
- cpi->diamond_search_sad = oxcf->mode == BEST ? vp10_full_range_search
- : vp10_diamond_search_sad;
+ cpi->diamond_search_sad = vp10_diamond_search_sad;
+
+ sf->allow_exhaustive_searches = 1;
+ if (oxcf->mode == BEST) {
+ if (cpi->twopass.fr_content_type == FC_GRAPHICS_ANIMATION)
+ sf->exhaustive_searches_thresh = (1 << 20);
+ else
+ sf->exhaustive_searches_thresh = (1 << 21);
+ sf->max_exaustive_pct = 100;
+ for (i = 0; i < MAX_MESH_STEP; ++i) {
+ sf->mesh_patterns[i].range = best_quality_mesh_pattern[i].range;
+ sf->mesh_patterns[i].interval = best_quality_mesh_pattern[i].interval;
+ }
+ } else {
+ int speed = (oxcf->speed > MAX_MESH_SPEED) ? MAX_MESH_SPEED : oxcf->speed;
+ if (cpi->twopass.fr_content_type == FC_GRAPHICS_ANIMATION)
+ sf->exhaustive_searches_thresh = (1 << 22);
+ else
+ sf->exhaustive_searches_thresh = (1 << 23);
+ sf->max_exaustive_pct = good_quality_max_mesh_pct[speed];
+ if (speed > 0)
+ sf->exhaustive_searches_thresh = sf->exhaustive_searches_thresh << 1;
+
+ for (i = 0; i < MAX_MESH_STEP; ++i) {
+ sf->mesh_patterns[i].range =
+ good_quality_mesh_patterns[speed][i].range;
+ sf->mesh_patterns[i].interval =
+ good_quality_mesh_patterns[speed][i].interval;
+ }
+ }
// Slow quant, dct and trellis not worthwhile for first pass
// so make sure they are always turned off.
diff --git a/vp10/encoder/speed_features.h b/vp10/encoder/speed_features.h
index 3969a2f..3b91999 100644
--- a/vp10/encoder/speed_features.h
+++ b/vp10/encoder/speed_features.h
@@ -195,6 +195,13 @@
int fullpel_search_step_param;
} MV_SPEED_FEATURES;
+#define MAX_MESH_STEP 4
+
+typedef struct MESH_PATTERN {
+ int range;
+ int interval;
+} MESH_PATTERN;
+
typedef struct SPEED_FEATURES {
MV_SPEED_FEATURES mv;
@@ -290,6 +297,18 @@
// point for this motion search and limits the search range around it.
int adaptive_motion_search;
+ // Flag for allowing some use of exhaustive searches;
+ int allow_exhaustive_searches;
+
+ // Threshold for allowing exhaistive motion search.
+ int exhaustive_searches_thresh;
+
+ // Maximum number of exhaustive searches for a frame.
+ int max_exaustive_pct;
+
+ // Pattern to be used for any exhaustive mesh searches.
+ MESH_PATTERN mesh_patterns[MAX_MESH_STEP];
+
int schedule_mode_search;
// Allows sub 8x8 modes to use the prediction filter that was determined
diff --git a/vp10/encoder/tokenize.c b/vp10/encoder/tokenize.c
index e568c0b..1b94190 100644
--- a/vp10/encoder/tokenize.c
+++ b/vp10/encoder/tokenize.c
@@ -487,39 +487,6 @@
return segfeature_active(seg, segment_id, SEG_LVL_SKIP) ? 0 : eob_max;
}
-void vp10_tokenize_palette_sb(struct ThreadData *const td,
- BLOCK_SIZE bsize, int plane,
- TOKENEXTRA **t) {
- MACROBLOCK *const x = &td->mb;
- MACROBLOCKD *const xd = &x->e_mbd;
- MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
- uint8_t *color_map = xd->plane[0].color_index_map;
- PALETTE_MODE_INFO *pmi = &mbmi->palette_mode_info;
- int n = pmi->palette_size[plane != 0];
- int i, j, k;
- int color_new_idx = -1, color_ctx, color_order[PALETTE_MAX_SIZE];
- int rows = 4 * num_4x4_blocks_high_lookup[bsize];
- int cols = 4 * num_4x4_blocks_wide_lookup[bsize];
-
- for (i = 0; i < rows; ++i) {
- for (j = (i == 0 ? 1 : 0); j < cols; ++j) {
- color_ctx = vp10_get_palette_color_context(color_map, cols, i, j, n,
- color_order);
- for (k = 0; k < n; ++k)
- if (color_map[i * cols + j] == color_order[k]) {
- color_new_idx = k;
- break;
- }
- assert(color_new_idx >= 0 && color_new_idx < n);
-
- (*t)->token = color_new_idx;
- (*t)->context_tree = vp10_default_palette_y_color_prob[n - 2][color_ctx];
- (*t)->skip_eob_node = 0;
- ++(*t);
- }
- }
-}
-
static void tokenize_b(int plane, int block, int blk_row, int blk_col,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
diff --git a/vp10/encoder/tokenize.h b/vp10/encoder/tokenize.h
index 1614add..5bad415 100644
--- a/vp10/encoder/tokenize.h
+++ b/vp10/encoder/tokenize.h
@@ -51,9 +51,6 @@
struct VP10_COMP;
struct ThreadData;
-void vp10_tokenize_palette_sb(struct ThreadData *const td,
- BLOCK_SIZE bsize, int plane,
- TOKENEXTRA **t);
void vp10_tokenize_sb(struct VP10_COMP *cpi, struct ThreadData *td,
TOKENEXTRA **t, int dry_run, BLOCK_SIZE bsize);
diff --git a/vp10/vp10cx.mk b/vp10/vp10cx.mk
index 7b77f31..ead993a 100644
--- a/vp10/vp10cx.mk
+++ b/vp10/vp10cx.mk
@@ -52,8 +52,6 @@
VP10_CX_SRCS-yes += encoder/treewriter.h
VP10_CX_SRCS-yes += encoder/mcomp.c
VP10_CX_SRCS-yes += encoder/encoder.c
-VP10_CX_SRCS-yes += encoder/palette.h
-VP10_CX_SRCS-yes += encoder/palette.c
VP10_CX_SRCS-yes += encoder/picklpf.c
VP10_CX_SRCS-yes += encoder/picklpf.h
VP10_CX_SRCS-yes += encoder/quantize.c
diff --git a/vp8/common/vp8_loopfilter.c b/vp8/common/vp8_loopfilter.c
index 8b55dff..756ad48 100644
--- a/vp8/common/vp8_loopfilter.c
+++ b/vp8/common/vp8_loopfilter.c
@@ -141,8 +141,8 @@
else /* Delta Value */
{
lvl_seg += mbd->segment_feature_data[MB_LVL_ALT_LF][seg];
- lvl_seg = (lvl_seg > 0) ? ((lvl_seg > 63) ? 63: lvl_seg) : 0;
}
+ lvl_seg = (lvl_seg > 0) ? ((lvl_seg > 63) ? 63: lvl_seg) : 0;
}
if (!mbd->mode_ref_lf_delta_enabled)
diff --git a/vp8/decoder/decodeframe.c b/vp8/decoder/decodeframe.c
index f0d7603..4bc87eb 100644
--- a/vp8/decoder/decodeframe.c
+++ b/vp8/decoder/decodeframe.c
@@ -73,10 +73,9 @@
/* Delta Value */
else
- {
QIndex = pc->base_qindex + xd->segment_feature_data[MB_LVL_ALT_Q][mbmi->segment_id];
- QIndex = (QIndex >= 0) ? ((QIndex <= MAXQ) ? QIndex : MAXQ) : 0; /* Clamp to valid range */
- }
+
+ QIndex = (QIndex >= 0) ? ((QIndex <= MAXQ) ? QIndex : MAXQ) : 0; /* Clamp to valid range */
}
else
QIndex = pc->base_qindex;
diff --git a/vp9/common/vp9_idct.c b/vp9/common/vp9_idct.c
index d12cd76..1b42014 100644
--- a/vp9/common/vp9_idct.c
+++ b/vp9/common/vp9_idct.c
@@ -174,6 +174,9 @@
else if (eob <= 34)
// non-zero coeff only in upper-left 8x8
vpx_idct32x32_34_add(input, dest, stride);
+ else if (eob <= 135)
+ // non-zero coeff only in upper-left 16x16
+ vpx_idct32x32_135_add(input, dest, stride);
else
vpx_idct32x32_1024_add(input, dest, stride);
}
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 8fe6503..d166bbf 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -198,7 +198,7 @@
specialize qw/vp9_avg_8x8 sse2 neon msa/;
add_proto qw/unsigned int vp9_avg_4x4/, "const uint8_t *, int p";
-specialize qw/vp9_avg_4x4 sse2 msa/;
+specialize qw/vp9_avg_4x4 sse2 neon msa/;
add_proto qw/void vp9_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
specialize qw/vp9_minmax_8x8 sse2/;
diff --git a/vp9/encoder/arm/neon/vp9_avg_neon.c b/vp9/encoder/arm/neon/vp9_avg_neon.c
index 5996bd4..78467ce 100644
--- a/vp9/encoder/arm/neon/vp9_avg_neon.c
+++ b/vp9/encoder/arm/neon/vp9_avg_neon.c
@@ -24,6 +24,18 @@
return vget_lane_u32(c, 0);
}
+unsigned int vp9_avg_4x4_neon(const uint8_t *s, int p) {
+ uint16x8_t v_sum;
+ uint32x2_t v_s0 = vdup_n_u32(0);
+ uint32x2_t v_s1 = vdup_n_u32(0);
+ v_s0 = vld1_lane_u32((const uint32_t *)s, v_s0, 0);
+ v_s0 = vld1_lane_u32((const uint32_t *)(s + p), v_s0, 1);
+ v_s1 = vld1_lane_u32((const uint32_t *)(s + 2 * p), v_s1, 0);
+ v_s1 = vld1_lane_u32((const uint32_t *)(s + 3 * p), v_s1, 1);
+ v_sum = vaddl_u8(vreinterpret_u8_u32(v_s0), vreinterpret_u8_u32(v_s1));
+ return (horizontal_add_u16x8(v_sum) + 8) >> 4;
+}
+
unsigned int vp9_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);
diff --git a/vp9/encoder/vp9_denoiser.c b/vp9/encoder/vp9_denoiser.c
index fc76c11..93aa40a 100644
--- a/vp9/encoder/vp9_denoiser.c
+++ b/vp9/encoder/vp9_denoiser.c
@@ -194,7 +194,8 @@
int mi_col,
PICK_MODE_CONTEXT *ctx,
int motion_magnitude,
- int is_skin) {
+ int is_skin,
+ int *zeromv_filter) {
int mv_col, mv_row;
int sse_diff = ctx->zeromv_sse - ctx->newmv_sse;
MV_REFERENCE_FRAME frame;
@@ -237,6 +238,7 @@
mbmi->mv[0].as_int = 0;
ctx->best_sse_inter_mode = ZEROMV;
ctx->best_sse_mv.as_int = 0;
+ *zeromv_filter = 1;
}
if (ctx->newmv_sse > sse_thresh(bs, increase_denoising)) {
@@ -320,6 +322,7 @@
VP9_DENOISER_DECISION *denoiser_decision) {
int mv_col, mv_row;
int motion_magnitude = 0;
+ int zeromv_filter = 0;
VP9_DENOISER_DECISION decision = COPY_BLOCK;
YV12_BUFFER_CONFIG avg = denoiser->running_avg_y[INTRA_FRAME];
YV12_BUFFER_CONFIG mc_avg = denoiser->mc_running_avg_y;
@@ -360,7 +363,8 @@
denoiser->increase_denoising,
mi_row, mi_col, ctx,
motion_magnitude,
- is_skin);
+ is_skin,
+ &zeromv_filter);
if (decision == FILTER_BLOCK) {
decision = vp9_denoiser_filter(src.buf, src.stride,
@@ -382,6 +386,8 @@
num_4x4_blocks_high_lookup[bs] << 2);
}
*denoiser_decision = decision;
+ if (decision == FILTER_BLOCK && zeromv_filter == 1)
+ *denoiser_decision = FILTER_ZEROMV_BLOCK;
}
static void copy_frame(YV12_BUFFER_CONFIG * const dest,
diff --git a/vp9/encoder/vp9_denoiser.h b/vp9/encoder/vp9_denoiser.h
index c8c9352..d07056b 100644
--- a/vp9/encoder/vp9_denoiser.h
+++ b/vp9/encoder/vp9_denoiser.h
@@ -23,7 +23,8 @@
typedef enum vp9_denoiser_decision {
COPY_BLOCK,
- FILTER_BLOCK
+ FILTER_BLOCK,
+ FILTER_ZEROMV_BLOCK
} VP9_DENOISER_DECISION;
typedef enum vp9_denoiser_level {
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 7e56989..c643b18 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -401,7 +401,6 @@
variance_node vt;
const int block_width = num_8x8_blocks_wide_lookup[bsize];
const int block_height = num_8x8_blocks_high_lookup[bsize];
- const int low_res = (cm->width <= 352 && cm->height <= 288);
assert(block_height == block_width);
tree_to_node(data, bsize, &vt);
@@ -414,7 +413,7 @@
// No check for vert/horiz split as too few samples for variance.
if (bsize == bsize_min) {
// Variance already computed to set the force_split.
- if (low_res || cm->frame_type == KEY_FRAME)
+ if (cm->frame_type == KEY_FRAME)
get_variance(&vt.part_variances->none);
if (mi_col + block_width / 2 < cm->mi_cols &&
mi_row + block_height / 2 < cm->mi_rows &&
@@ -425,7 +424,7 @@
return 0;
} else if (bsize > bsize_min) {
// Variance already computed to set the force_split.
- if (low_res || cm->frame_type == KEY_FRAME)
+ if (cm->frame_type == KEY_FRAME)
get_variance(&vt.part_variances->none);
// For key frame: take split for bsize above 32X32 or very high variance.
if (cm->frame_type == KEY_FRAME &&
@@ -489,13 +488,14 @@
thresholds[2] = threshold_base >> 2;
thresholds[3] = threshold_base << 2;
} else {
- // Increase base variance threshold if estimated noise level is high.
+ // Increase base variance threshold based on estimated noise level.
if (cpi->noise_estimate.enabled) {
- if (cpi->noise_estimate.level == kHigh)
+ NOISE_LEVEL noise_level = vp9_noise_estimate_extract_level(
+ &cpi->noise_estimate);
+ if (noise_level == kHigh)
threshold_base = 3 * threshold_base;
- else
- if (cpi->noise_estimate.level == kMedium)
- threshold_base = threshold_base << 1;
+ else if (noise_level == kMedium)
+ threshold_base = threshold_base << 1;
}
if (cm->width <= 352 && cm->height <= 288) {
thresholds[0] = threshold_base >> 3;
@@ -843,7 +843,8 @@
force_split[split_index] = 1;
force_split[i + 1] = 1;
force_split[0] = 1;
- } else if (vt.split[i].split[j].part_variances.none.variance >
+ } else if (cpi->oxcf.speed < 8 &&
+ vt.split[i].split[j].part_variances.none.variance >
thresholds[1] &&
!cyclic_refresh_segment_id_boosted(segment_id)) {
// We have some nominal amount of 16x16 variance (based on average),
@@ -896,6 +897,14 @@
for (m = 0; m < 4; m++)
fill_variance_tree(&vtemp->split[m], BLOCK_8X8);
fill_variance_tree(vtemp, BLOCK_16X16);
+ // If variance of this 16x16 block is above the threshold, force block
+ // to split. This also forces a split on the upper levels.
+ get_variance(&vtemp->part_variances.none);
+ if (vtemp->part_variances.none.variance > thresholds[2]) {
+ force_split[5 + i2 + j] = 1;
+ force_split[i + 1] = 1;
+ force_split[0] = 1;
+ }
}
}
fill_variance_tree(&vt.split[i], BLOCK_32X32);
@@ -2422,8 +2431,15 @@
if (cpi->sf.use_square_partition_only &&
bsize > cpi->sf.use_square_only_threshold) {
+ if (cpi->use_svc) {
+ if (!vp9_active_h_edge(cpi, mi_row, mi_step) || x->e_mbd.lossless)
+ partition_horz_allowed &= force_horz_split;
+ if (!vp9_active_v_edge(cpi, mi_row, mi_step) || x->e_mbd.lossless)
+ partition_vert_allowed &= force_vert_split;
+ } else {
partition_horz_allowed &= force_horz_split;
partition_vert_allowed &= force_vert_split;
+ }
}
save_context(x, mi_row, mi_col, a, l, sa, sl, bsize);
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index a57cf87..72fa828 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -3682,12 +3682,16 @@
if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
cm->mi_rows * MI_SIZE != unscaled->y_height) {
#if CONFIG_VP9_HIGHBITDEPTH
- if (use_normative_scaler)
+ if (use_normative_scaler &&
+ unscaled->y_width <= (scaled->y_width << 1) &&
+ unscaled->y_height <= (scaled->y_height << 1))
scale_and_extend_frame(unscaled, scaled, (int)cm->bit_depth);
else
scale_and_extend_frame_nonnormative(unscaled, scaled, (int)cm->bit_depth);
#else
- if (use_normative_scaler)
+ if (use_normative_scaler &&
+ unscaled->y_width <= (scaled->y_width << 1) &&
+ unscaled->y_height <= (scaled->y_height << 1))
scale_and_extend_frame(unscaled, scaled);
else
scale_and_extend_frame_nonnormative(unscaled, scaled);
diff --git a/vp9/encoder/vp9_noise_estimate.c b/vp9/encoder/vp9_noise_estimate.c
index b41ffd0..4befbb0 100644
--- a/vp9/encoder/vp9_noise_estimate.c
+++ b/vp9/encoder/vp9_noise_estimate.c
@@ -25,7 +25,7 @@
int width,
int height) {
ne->enabled = 0;
- ne->level = kLow;
+ ne->level = kLowLow;
ne->value = 0;
ne->count = 0;
ne->thresh = 90;
@@ -82,6 +82,21 @@
}
}
+NOISE_LEVEL vp9_noise_estimate_extract_level(NOISE_ESTIMATE *const ne) {
+ int noise_level = kLowLow;
+ if (ne->value > (ne->thresh << 1)) {
+ noise_level = kHigh;
+ } else {
+ if (ne->value > ne->thresh)
+ noise_level = kMedium;
+ else if (ne->value > (ne->thresh >> 1))
+ noise_level = kLow;
+ else
+ noise_level = kLowLow;
+ }
+ return noise_level;
+}
+
void vp9_update_noise_estimate(VP9_COMP *const cpi) {
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
@@ -220,22 +235,16 @@
// Reset counter and check noise level condition.
ne->num_frames_estimate = 30;
ne->count = 0;
- if (ne->value > (ne->thresh << 1))
- ne->level = kHigh;
- else
- if (ne->value > ne->thresh)
- ne->level = kMedium;
- else if (ne->value > (ne->thresh >> 1))
- ne->level = kLow;
- else
- ne->level = kLowLow;
+ ne->level = vp9_noise_estimate_extract_level(ne);
+#if CONFIG_VP9_TEMPORAL_DENOISING
+ if (cpi->oxcf.noise_sensitivity > 0)
+ vp9_denoiser_set_noise_level(&cpi->denoiser, ne->level);
+#endif
}
}
}
#if CONFIG_VP9_TEMPORAL_DENOISING
- if (cpi->oxcf.noise_sensitivity > 0) {
+ if (cpi->oxcf.noise_sensitivity > 0)
copy_frame(&cpi->denoiser.last_source, cpi->Source);
- vp9_denoiser_set_noise_level(&cpi->denoiser, ne->level);
- }
#endif
}
diff --git a/vp9/encoder/vp9_noise_estimate.h b/vp9/encoder/vp9_noise_estimate.h
index 0d22ef0..826d125 100644
--- a/vp9/encoder/vp9_noise_estimate.h
+++ b/vp9/encoder/vp9_noise_estimate.h
@@ -47,6 +47,8 @@
int width,
int height);
+NOISE_LEVEL vp9_noise_estimate_extract_level(NOISE_ESTIMATE *const ne);
+
void vp9_update_noise_estimate(struct VP9_COMP *const cpi);
#ifdef __cplusplus
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 02e9674..90650db 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -1696,15 +1696,14 @@
VP9_DENOISER_DECISION decision = COPY_BLOCK;
vp9_denoiser_denoise(&cpi->denoiser, x, mi_row, mi_col,
VPXMAX(BLOCK_8X8, bsize), ctx, &decision);
- // If INTRA mode was selected, re-evaluate ZEROMV on denoised result.
- // Only do this under noise conditions, and if rdcost of ZEROMV on
- // original source is not significantly higher than rdcost of INTRA MODE.
- if (best_ref_frame == INTRA_FRAME &&
- decision == FILTER_BLOCK &&
+ // If INTRA or GOLDEN reference was selected, re-evaluate ZEROMV on denoised
+ // result. Only do this under noise conditions, and if rdcost of ZEROMV on
+ // original source is not significantly higher than rdcost of best mode.
+ if (((best_ref_frame == INTRA_FRAME && decision >= FILTER_BLOCK) ||
+ (best_ref_frame == GOLDEN_FRAME && decision == FILTER_ZEROMV_BLOCK)) &&
cpi->noise_estimate.enabled &&
cpi->noise_estimate.level > kLow &&
- zero_last_cost_orig < (best_rdc.rdcost << 2) &&
- !reuse_inter_pred) {
+ zero_last_cost_orig < (best_rdc.rdcost << 3)) {
// Check if we should pick ZEROMV on denoised signal.
int rate = 0;
int64_t dist = 0;
@@ -1713,6 +1712,7 @@
mbmi->ref_frame[1] = NONE;
mbmi->mv[0].as_int = 0;
mbmi->interp_filter = EIGHTTAP;
+ xd->plane[0].pre[0] = yv12_mb[LAST_FRAME][0];
vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
model_rd_for_sb_y(cpi, bsize, x, xd, &rate, &dist, &var_y, &sse_y);
this_rdc.rate = rate + ref_frame_cost[LAST_FRAME] +
@@ -1721,13 +1721,21 @@
this_rdc.dist = dist;
this_rdc.rdcost = RDCOST(x->rdmult, x->rddiv, rate, dist);
// Switch to ZEROMV if the rdcost for ZEROMV on denoised source
- // is lower than INTRA (on original source).
+ // is lower than best_ref mode (on original source).
if (this_rdc.rdcost > best_rdc.rdcost) {
this_rdc = best_rdc;
mbmi->mode = best_mode;
mbmi->ref_frame[0] = best_ref_frame;
- mbmi->mv[0].as_int = INVALID_MV;
mbmi->interp_filter = best_pred_filter;
+ if (best_ref_frame == INTRA_FRAME)
+ mbmi->mv[0].as_int = INVALID_MV;
+ else if (best_ref_frame == GOLDEN_FRAME) {
+ mbmi->mv[0].as_int = frame_mv[best_mode][best_ref_frame].as_int;
+ if (reuse_inter_pred) {
+ xd->plane[0].pre[0] = yv12_mb[GOLDEN_FRAME][0];
+ vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
+ }
+ }
mbmi->tx_size = best_tx_size;
x->skip_txfm[0] = best_mode_skip_txfm;
} else {
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index fdff363..8ab51cd 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -1075,7 +1075,7 @@
if (!cpi->refresh_alt_ref_frame) {
active_best_quality = cq_level;
} else {
- active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
+ active_best_quality = get_gf_active_quality(rc, q, cm->bit_depth);
// Modify best quality for second level arfs. For mode VPX_Q this
// becomes the baseline frame q.
@@ -1257,8 +1257,12 @@
rc->frames_since_golden = 0;
// If we are not using alt ref in the up and coming group clear the arf
- // active flag.
- if (!rc->source_alt_ref_pending) {
+ // active flag. In multi arf group case, if the index is not 0 then
+ // we are overlaying a mid group arf so should not reset the flag.
+ if (cpi->oxcf.pass == 2) {
+ if (!rc->source_alt_ref_pending && (cpi->twopass.gf_group.index == 0))
+ rc->source_alt_ref_active = 0;
+ } else if (!rc->source_alt_ref_pending) {
rc->source_alt_ref_active = 0;
}
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index 20516a0..318d810 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -380,16 +380,6 @@
sf->adaptive_rd_thresh = 2;
// This feature is only enabled when partition search is disabled.
sf->reuse_inter_pred_sby = 1;
- // TODO(marpan): When denoising, we may re-evaluate the mode selection and
- // this seems to cause problems when reuse_inter_pred_sby is enabled.
- // Disabling reuse_inter_pred_sby for now (under denoising conditions), and
- // will look into re-enabling it.
-#if CONFIG_VP9_TEMPORAL_DENOISING
- if (cpi->oxcf.noise_sensitivity > 0 &&
- cpi->noise_estimate.enabled &&
- cpi->noise_estimate.level > kLow)
- sf->reuse_inter_pred_sby = 0;
-#endif
sf->partition_search_breakout_rate_thr = 200;
sf->coeff_prob_appx_step = 4;
sf->use_fast_coef_updates = is_keyframe ? TWO_LOOP : ONE_LOOP_REDUCED;
diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index 13da155..b0617c1 100644
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -279,7 +279,7 @@
// Reset the frames_since_key and frames_to_key counters to their values
// before the layer restore. Keep these defined for the stream (not layer).
if (cpi->svc.number_temporal_layers > 1 ||
- cpi->svc.number_spatial_layers > 1) {
+ (cpi->svc.number_spatial_layers > 1 && !is_two_pass_svc(cpi))) {
cpi->rc.frames_since_key = old_frame_since_key;
cpi->rc.frames_to_key = old_frame_to_key;
}
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index 16f9c85..015dbc0 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -135,15 +135,38 @@
for (i = 0, k = 0; i < block_height; i++) {
for (j = 0; j < block_width; j++, k++) {
- int src_byte = frame1[byte];
- int pixel_value = *frame2++;
+ int pixel_value = *frame2;
- modifier = src_byte - pixel_value;
- // This is an integer approximation of:
- // float coeff = (3.0 * modifer * modifier) / pow(2, strength);
- // modifier = (int)roundf(coeff > 16 ? 0 : 16-coeff);
- modifier *= modifier;
- modifier *= 3;
+ // non-local mean approach
+ int diff_sse[9] = { 0 };
+ int idx, idy, index = 0;
+
+ for (idy = -1; idy <= 1; ++idy) {
+ for (idx = -1; idx <= 1; ++idx) {
+ int row = i + idy;
+ int col = j + idx;
+
+ if (row >= 0 && row < (int)block_height &&
+ col >= 0 && col < (int)block_width) {
+ int diff = frame1[byte + idy * (int)stride + idx] -
+ frame2[idy * (int)block_width + idx];
+ diff_sse[index] = diff * diff;
+ ++index;
+ }
+ }
+ }
+
+ assert(index > 0);
+
+ modifier = 0;
+ for (idx = 0; idx < 9; ++idx)
+ modifier += diff_sse[idx];
+
+ modifier *= 3;
+ modifier /= index;
+
+ ++frame2;
+
modifier += rounding;
modifier >>= strength;
@@ -182,15 +205,34 @@
for (i = 0, k = 0; i < block_height; i++) {
for (j = 0; j < block_width; j++, k++) {
- int src_byte = frame1[byte];
- int pixel_value = *frame2++;
+ int pixel_value = *frame2;
+ int diff_sse[9] = { 0 };
+ int idx, idy, index = 0;
- modifier = src_byte - pixel_value;
- // This is an integer approximation of:
- // float coeff = (3.0 * modifer * modifier) / pow(2, strength);
- // modifier = (int)roundf(coeff > 16 ? 0 : 16-coeff);
- modifier *= modifier;
+ for (idy = -1; idy <= 1; ++idy) {
+ for (idx = -1; idx <= 1; ++idx) {
+ int row = i + idy;
+ int col = j + idx;
+
+ if (row >= 0 && row < (int)block_height &&
+ col >= 0 && col < (int)block_width) {
+ int diff = frame1[byte + idy * (int)stride + idx] -
+ frame2[idy * (int)block_width + idx];
+ diff_sse[index] = diff * diff;
+ ++index;
+ }
+ }
+ }
+ assert(index > 0);
+
+ modifier = 0;
+ for (idx = 0; idx < 9; ++idx)
+ modifier += diff_sse[idx];
+
modifier *= 3;
+ modifier /= index;
+
+ ++frame2;
modifier += rounding;
modifier >>= strength;
@@ -383,55 +425,58 @@
if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
int adj_strength = strength + 2 * (mbd->bd - 8);
// Apply the filter (YUV)
- vp9_highbd_temporal_filter_apply(f->y_buffer + mb_y_offset,
- f->y_stride,
- predictor, 16, 16, adj_strength,
- filter_weight,
- accumulator, count);
- vp9_highbd_temporal_filter_apply(f->u_buffer + mb_uv_offset,
- f->uv_stride, predictor + 256,
- mb_uv_width, mb_uv_height,
- adj_strength,
- filter_weight, accumulator + 256,
- count + 256);
- vp9_highbd_temporal_filter_apply(f->v_buffer + mb_uv_offset,
- f->uv_stride, predictor + 512,
- mb_uv_width, mb_uv_height,
- adj_strength, filter_weight,
- accumulator + 512, count + 512);
+ vp9_highbd_temporal_filter_apply_c(f->y_buffer + mb_y_offset,
+ f->y_stride,
+ predictor, 16, 16, adj_strength,
+ filter_weight,
+ accumulator, count);
+ vp9_highbd_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
+ f->uv_stride, predictor + 256,
+ mb_uv_width, mb_uv_height,
+ adj_strength,
+ filter_weight, accumulator + 256,
+ count + 256);
+ vp9_highbd_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
+ f->uv_stride, predictor + 512,
+ mb_uv_width, mb_uv_height,
+ adj_strength, filter_weight,
+ accumulator + 512, count + 512);
} else {
// Apply the filter (YUV)
- vp9_temporal_filter_apply(f->y_buffer + mb_y_offset, f->y_stride,
+ vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
+ predictor, 16, 16,
+ strength, filter_weight,
+ accumulator, count);
+ vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
+ f->uv_stride,
+ predictor + 256,
+ mb_uv_width, mb_uv_height, strength,
+ filter_weight, accumulator + 256,
+ count + 256);
+ vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
+ f->uv_stride,
+ predictor + 512,
+ mb_uv_width, mb_uv_height, strength,
+ filter_weight, accumulator + 512,
+ count + 512);
+ }
+#else
+ // Apply the filter (YUV)
+ // TODO(jingning): Need SIMD optimization for this.
+ vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
predictor, 16, 16,
strength, filter_weight,
accumulator, count);
- vp9_temporal_filter_apply(f->u_buffer + mb_uv_offset, f->uv_stride,
+ vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset, f->uv_stride,
predictor + 256,
mb_uv_width, mb_uv_height, strength,
filter_weight, accumulator + 256,
count + 256);
- vp9_temporal_filter_apply(f->v_buffer + mb_uv_offset, f->uv_stride,
+ vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset, f->uv_stride,
predictor + 512,
mb_uv_width, mb_uv_height, strength,
filter_weight, accumulator + 512,
count + 512);
- }
-#else
- // Apply the filter (YUV)
- vp9_temporal_filter_apply(f->y_buffer + mb_y_offset, f->y_stride,
- predictor, 16, 16,
- strength, filter_weight,
- accumulator, count);
- vp9_temporal_filter_apply(f->u_buffer + mb_uv_offset, f->uv_stride,
- predictor + 256,
- mb_uv_width, mb_uv_height, strength,
- filter_weight, accumulator + 256,
- count + 256);
- vp9_temporal_filter_apply(f->v_buffer + mb_uv_offset, f->uv_stride,
- predictor + 512,
- mb_uv_width, mb_uv_height, strength,
- filter_weight, accumulator + 512,
- count + 512);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}
diff --git a/vpx/src/svc_encodeframe.c b/vpx/src/svc_encodeframe.c
index 68d1d8d..5c3fe93 100644
--- a/vpx/src/svc_encodeframe.c
+++ b/vpx/src/svc_encodeframe.c
@@ -423,13 +423,15 @@
svc_ctx->temporal_layers = 2;
}
+ for (sl = 0; sl < VPX_SS_MAX_LAYERS; ++sl) {
+ si->svc_params.scaling_factor_num[sl] = DEFAULT_SCALE_FACTORS_NUM[sl];
+ si->svc_params.scaling_factor_den[sl] = DEFAULT_SCALE_FACTORS_DEN[sl];
+ }
for (tl = 0; tl < svc_ctx->temporal_layers; ++tl) {
for (sl = 0; sl < svc_ctx->spatial_layers; ++sl) {
i = sl * svc_ctx->temporal_layers + tl;
si->svc_params.max_quantizers[i] = MAX_QUANTIZER;
si->svc_params.min_quantizers[i] = 0;
- si->svc_params.scaling_factor_num[sl] = DEFAULT_SCALE_FACTORS_NUM[sl];
- si->svc_params.scaling_factor_den[sl] = DEFAULT_SCALE_FACTORS_DEN[sl];
}
}
diff --git a/vpx_dsp/inv_txfm.c b/vpx_dsp/inv_txfm.c
index 5f3cfdd..a0f59bf 100644
--- a/vpx_dsp/inv_txfm.c
+++ b/vpx_dsp/inv_txfm.c
@@ -1194,6 +1194,33 @@
}
}
+void vpx_idct32x32_135_add_c(const tran_low_t *input, uint8_t *dest,
+ int stride) {
+ tran_low_t out[32 * 32] = {0};
+ tran_low_t *outptr = out;
+ int i, j;
+ tran_low_t temp_in[32], temp_out[32];
+
+ // Rows
+ // only upper-left 16x16 has non-zero coeff
+ for (i = 0; i < 16; ++i) {
+ idct32_c(input, outptr);
+ input += 32;
+ outptr += 32;
+ }
+
+ // Columns
+ for (i = 0; i < 32; ++i) {
+ for (j = 0; j < 32; ++j)
+ temp_in[j] = out[j * 32 + i];
+ idct32_c(temp_in, temp_out);
+ for (j = 0; j < 32; ++j) {
+ dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
+ ROUND_POWER_OF_TWO(temp_out[j], 6));
+ }
+ }
+}
+
void vpx_idct32x32_34_add_c(const tran_low_t *input, uint8_t *dest,
int stride) {
tran_low_t out[32 * 32] = {0};
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 4de85a4..4d36e27 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -76,7 +76,7 @@
specialize qw/vpx_d63f_predictor_4x4/;
add_proto qw/void vpx_h_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_4x4 neon dspr2 msa/, "$ssse3_x86inc";
+specialize qw/vpx_h_predictor_4x4 neon dspr2 msa/, "$sse2_x86inc";
add_proto qw/void vpx_he_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_he_predictor_4x4/;
@@ -91,7 +91,7 @@
specialize qw/vpx_d153_predictor_4x4/, "$ssse3_x86inc";
add_proto qw/void vpx_v_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_v_predictor_4x4 neon msa/, "$sse_x86inc";
+specialize qw/vpx_v_predictor_4x4 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_ve_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_ve_predictor_4x4/;
@@ -100,16 +100,16 @@
specialize qw/vpx_tm_predictor_4x4 neon dspr2 msa/, "$sse2_x86inc";
add_proto qw/void vpx_dc_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_predictor_4x4 dspr2 msa neon/, "$sse_x86inc";
+specialize qw/vpx_dc_predictor_4x4 dspr2 msa neon/, "$sse2_x86inc";
add_proto qw/void vpx_dc_top_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_top_predictor_4x4 msa neon/, "$sse_x86inc";
+specialize qw/vpx_dc_top_predictor_4x4 msa neon/, "$sse2_x86inc";
add_proto qw/void vpx_dc_left_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_left_predictor_4x4 msa neon/, "$sse_x86inc";
+specialize qw/vpx_dc_left_predictor_4x4 msa neon/, "$sse2_x86inc";
add_proto qw/void vpx_dc_128_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_128_predictor_4x4 msa neon/, "$sse_x86inc";
+specialize qw/vpx_dc_128_predictor_4x4 msa neon/, "$sse2_x86inc";
add_proto qw/void vpx_d207_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_d207_predictor_8x8/, "$ssse3_x86inc";
@@ -130,7 +130,7 @@
specialize qw/vpx_d63e_predictor_8x8/;
add_proto qw/void vpx_h_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_8x8 neon dspr2 msa/, "$ssse3_x86inc";
+specialize qw/vpx_h_predictor_8x8 neon dspr2 msa/, "$sse2_x86inc";
add_proto qw/void vpx_d117_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_d117_predictor_8x8/;
@@ -142,22 +142,22 @@
specialize qw/vpx_d153_predictor_8x8/, "$ssse3_x86inc";
add_proto qw/void vpx_v_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_v_predictor_8x8 neon msa/, "$sse_x86inc";
+specialize qw/vpx_v_predictor_8x8 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_tm_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_tm_predictor_8x8 neon dspr2 msa/, "$sse2_x86inc";
add_proto qw/void vpx_dc_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_predictor_8x8 dspr2 neon msa/, "$sse_x86inc";
+specialize qw/vpx_dc_predictor_8x8 dspr2 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_dc_top_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_top_predictor_8x8 neon msa/, "$sse_x86inc";
+specialize qw/vpx_dc_top_predictor_8x8 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_dc_left_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_left_predictor_8x8 neon msa/, "$sse_x86inc";
+specialize qw/vpx_dc_left_predictor_8x8 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_dc_128_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_128_predictor_8x8 neon msa/, "$sse_x86inc";
+specialize qw/vpx_dc_128_predictor_8x8 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_d207_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_d207_predictor_16x16/, "$ssse3_x86inc";
@@ -178,7 +178,7 @@
specialize qw/vpx_d63e_predictor_16x16/;
add_proto qw/void vpx_h_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_16x16 neon dspr2 msa/, "$ssse3_x86inc";
+specialize qw/vpx_h_predictor_16x16 neon dspr2 msa/, "$sse2_x86inc";
add_proto qw/void vpx_d117_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_d117_predictor_16x16/;
@@ -226,7 +226,7 @@
specialize qw/vpx_d63e_predictor_32x32/;
add_proto qw/void vpx_h_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_32x32 neon msa/, "$ssse3_x86inc";
+specialize qw/vpx_h_predictor_32x32 neon msa/, "$sse2_x86inc";
add_proto qw/void vpx_d117_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_d117_predictor_32x32/;
@@ -754,6 +754,9 @@
add_proto qw/void vpx_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_1024_add/;
+ add_proto qw/void vpx_idct32x32_135_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+ specialize qw/vpx_idct32x32_135_add/;
+
add_proto qw/void vpx_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_34_add/;
@@ -802,6 +805,10 @@
add_proto qw/void vpx_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_1024_add sse2/;
+ add_proto qw/void vpx_idct32x32_135_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+ specialize qw/vpx_idct32x32_135_add sse2/;
+ $vpx_idct32x32_135_add_sse2=vpx_idct32x32_1024_add_sse2;
+
add_proto qw/void vpx_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_34_add sse2/;
@@ -853,6 +860,9 @@
add_proto qw/void vpx_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_1024_add/;
+ add_proto qw/void vpx_idct32x32_135_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+ specialize qw/vpx_idct32x32_135_add/;
+
add_proto qw/void vpx_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_34_add/;
@@ -890,7 +900,15 @@
specialize qw/vpx_idct16x16_10_add sse2 neon dspr2 msa/;
add_proto qw/void vpx_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
- specialize qw/vpx_idct32x32_1024_add sse2 neon dspr2 msa/;
+ specialize qw/vpx_idct32x32_1024_add sse2 neon dspr2 msa/, "$ssse3_x86_64_x86inc";
+
+ add_proto qw/void vpx_idct32x32_135_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+ specialize qw/vpx_idct32x32_135_add sse2 neon dspr2 msa/, "$ssse3_x86_64_x86inc";
+ # Need to add 135 eob idct32x32 implementations.
+ $vpx_idct32x32_135_add_sse2=vpx_idct32x32_1024_add_sse2;
+ $vpx_idct32x32_135_add_neon=vpx_idct32x32_1024_add_neon;
+ $vpx_idct32x32_135_add_dspr2=vpx_idct32x32_1024_add_dspr2;
+ $vpx_idct32x32_135_add_msa=vpx_idct32x32_1024_add_msa;
add_proto qw/void vpx_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vpx_idct32x32_34_add sse2 neon_asm dspr2 msa/, "$ssse3_x86_64_x86inc";
diff --git a/vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm b/vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm
index 93df92a..22d52a2 100644
--- a/vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm
+++ b/vpx_dsp/x86/highbd_subpel_variance_impl_sse2.asm
@@ -123,8 +123,10 @@
%define sec_str sec_stridemp
; Store bilin_filter and pw_8 location in stack
- GET_GOT eax
- add esp, 4 ; restore esp
+ %if GET_GOT_DEFINED == 1
+ GET_GOT eax
+ add esp, 4 ; restore esp
+ %endif
lea ecx, [GLOBAL(bilin_filter_m)]
mov g_bilin_filterm, ecx
@@ -140,8 +142,10 @@
%define block_height heightd
; Store bilin_filter and pw_8 location in stack
- GET_GOT eax
- add esp, 4 ; restore esp
+ %if GET_GOT_DEFINED == 1
+ GET_GOT eax
+ add esp, 4 ; restore esp
+ %endif
lea ecx, [GLOBAL(bilin_filter_m)]
mov g_bilin_filterm, ecx
diff --git a/vpx_dsp/x86/intrapred_sse2.asm b/vpx_dsp/x86/intrapred_sse2.asm
index a8c1bc6..6f924a7 100644
--- a/vpx_dsp/x86/intrapred_sse2.asm
+++ b/vpx_dsp/x86/intrapred_sse2.asm
@@ -23,17 +23,18 @@
SECTION .text
-INIT_MMX sse
-cglobal dc_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
+INIT_XMM sse2
+cglobal dc_predictor_4x4, 4, 5, 3, dst, stride, above, left, goffset
GET_GOT goffsetq
- pxor m1, m1
+ movd m2, [leftq]
movd m0, [aboveq]
- punpckldq m0, [leftq]
+ pxor m1, m1
+ punpckldq m0, m2
psadbw m0, m1
paddw m0, [GLOBAL(pw_4)]
psraw m0, 3
- pshufw m0, m0, 0x0
+ pshuflw m0, m0, 0x0
packuswb m0, m0
movd [dstq ], m0
movd [dstq+strideq], m0
@@ -44,8 +45,9 @@
RESTORE_GOT
RET
-INIT_MMX sse
-cglobal dc_left_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
+INIT_XMM sse2
+cglobal dc_left_predictor_4x4, 2, 5, 2, dst, stride, above, left, goffset
+ movifnidn leftq, leftmp
GET_GOT goffsetq
pxor m1, m1
@@ -53,7 +55,7 @@
psadbw m0, m1
paddw m0, [GLOBAL(pw2_4)]
psraw m0, 2
- pshufw m0, m0, 0x0
+ pshuflw m0, m0, 0x0
packuswb m0, m0
movd [dstq ], m0
movd [dstq+strideq], m0
@@ -64,8 +66,8 @@
RESTORE_GOT
RET
-INIT_MMX sse
-cglobal dc_top_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
+INIT_XMM sse2
+cglobal dc_top_predictor_4x4, 3, 5, 2, dst, stride, above, left, goffset
GET_GOT goffsetq
pxor m1, m1
@@ -73,7 +75,7 @@
psadbw m0, m1
paddw m0, [GLOBAL(pw2_4)]
psraw m0, 2
- pshufw m0, m0, 0x0
+ pshuflw m0, m0, 0x0
packuswb m0, m0
movd [dstq ], m0
movd [dstq+strideq], m0
@@ -84,7 +86,7 @@
RESTORE_GOT
RET
-INIT_MMX sse
+INIT_XMM sse2
cglobal dc_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
GET_GOT goffsetq
@@ -98,8 +100,8 @@
paddw m0, m2
paddw m0, [GLOBAL(pw_8)]
psraw m0, 4
- pshufw m0, m0, 0x0
- packuswb m0, m0
+ punpcklbw m0, m0
+ pshuflw m0, m0, 0x0
movq [dstq ], m0
movq [dstq+strideq ], m0
movq [dstq+strideq*2], m0
@@ -113,8 +115,8 @@
RESTORE_GOT
RET
-INIT_MMX sse
-cglobal dc_top_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
+INIT_XMM sse2
+cglobal dc_top_predictor_8x8, 3, 5, 2, dst, stride, above, left, goffset
GET_GOT goffsetq
pxor m1, m1
@@ -124,8 +126,8 @@
psadbw m0, m1
paddw m0, [GLOBAL(pw2_8)]
psraw m0, 3
- pshufw m0, m0, 0x0
- packuswb m0, m0
+ punpcklbw m0, m0
+ pshuflw m0, m0, 0x0
movq [dstq ], m0
movq [dstq+strideq ], m0
movq [dstq+strideq*2], m0
@@ -139,8 +141,9 @@
RESTORE_GOT
RET
-INIT_MMX sse
-cglobal dc_left_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
+INIT_XMM sse2
+cglobal dc_left_predictor_8x8, 2, 5, 2, dst, stride, above, left, goffset
+ movifnidn leftq, leftmp
GET_GOT goffsetq
pxor m1, m1
@@ -150,8 +153,8 @@
psadbw m0, m1
paddw m0, [GLOBAL(pw2_8)]
psraw m0, 3
- pshufw m0, m0, 0x0
- packuswb m0, m0
+ punpcklbw m0, m0
+ pshuflw m0, m0, 0x0
movq [dstq ], m0
movq [dstq+strideq ], m0
movq [dstq+strideq*2], m0
@@ -165,8 +168,8 @@
RESTORE_GOT
RET
-INIT_MMX sse
-cglobal dc_128_predictor_4x4, 4, 5, 3, dst, stride, above, left, goffset
+INIT_XMM sse2
+cglobal dc_128_predictor_4x4, 2, 5, 1, dst, stride, above, left, goffset
GET_GOT goffsetq
DEFINE_ARGS dst, stride, stride3
@@ -179,8 +182,8 @@
RESTORE_GOT
RET
-INIT_MMX sse
-cglobal dc_128_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
+INIT_XMM sse2
+cglobal dc_128_predictor_8x8, 2, 5, 1, dst, stride, above, left, goffset
GET_GOT goffsetq
DEFINE_ARGS dst, stride, stride3
@@ -236,14 +239,11 @@
GET_GOT goffsetq
pxor m1, m1
- pxor m2, m2
mova m0, [aboveq]
DEFINE_ARGS dst, stride, stride3, lines4
lea stride3q, [strideq*3]
mov lines4d, 4
psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
movhlps m2, m0
paddw m0, m2
paddw m0, [GLOBAL(pw2_16)]
@@ -268,14 +268,11 @@
GET_GOT goffsetq
pxor m1, m1
- pxor m2, m2
mova m0, [leftq]
DEFINE_ARGS dst, stride, stride3, lines4
lea stride3q, [strideq*3]
mov lines4d, 4
psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
movhlps m2, m0
paddw m0, m2
paddw m0, [GLOBAL(pw2_16)]
@@ -452,7 +449,7 @@
RESTORE_GOT
RET
-INIT_MMX sse
+INIT_XMM sse2
cglobal v_predictor_4x4, 3, 3, 1, dst, stride, above
movd m0, [aboveq]
movd [dstq ], m0
@@ -462,7 +459,7 @@
movd [dstq+strideq], m0
RET
-INIT_MMX sse
+INIT_XMM sse2
cglobal v_predictor_8x8, 3, 3, 1, dst, stride, above
movq m0, [aboveq]
DEFINE_ARGS dst, stride, stride3
@@ -516,6 +513,97 @@
REP_RET
INIT_XMM sse2
+cglobal h_predictor_4x4, 2, 4, 4, dst, stride, line, left
+ movifnidn leftq, leftmp
+ movd m0, [leftq]
+ punpcklbw m0, m0
+ punpcklbw m0, m0
+ pshufd m1, m0, 0x1
+ movd [dstq ], m0
+ movd [dstq+strideq], m1
+ pshufd m2, m0, 0x2
+ lea dstq, [dstq+strideq*2]
+ pshufd m3, m0, 0x3
+ movd [dstq ], m2
+ movd [dstq+strideq], m3
+ RET
+
+INIT_XMM sse2
+cglobal h_predictor_8x8, 2, 5, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ mov lineq, -2
+ DEFINE_ARGS dst, stride, line, left, stride3
+ lea stride3q, [strideq*3]
+ movq m0, [leftq ]
+ punpcklbw m0, m0 ; l1 l1 l2 l2 ... l8 l8
+.loop:
+ pshuflw m1, m0, 0x0 ; l1 l1 l1 l1 l1 l1 l1 l1
+ pshuflw m2, m0, 0x55 ; l2 l2 l2 l2 l2 l2 l2 l2
+ movq [dstq ], m1
+ movq [dstq+strideq], m2
+ pshuflw m1, m0, 0xaa
+ pshuflw m2, m0, 0xff
+ movq [dstq+strideq*2], m1
+ movq [dstq+stride3q ], m2
+ pshufd m0, m0, 0xe ; [63:0] l5 l5 l6 l6 l7 l7 l8 l8
+ inc lineq
+ lea dstq, [dstq+strideq*4]
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal h_predictor_16x16, 2, 5, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ mov lineq, -4
+ DEFINE_ARGS dst, stride, line, left, stride3
+ lea stride3q, [strideq*3]
+.loop:
+ movd m0, [leftq]
+ punpcklbw m0, m0
+ punpcklbw m0, m0 ; l1 to l4 each repeated 4 times
+ pshufd m1, m0, 0x0 ; l1 repeated 16 times
+ pshufd m2, m0, 0x55 ; l2 repeated 16 times
+ mova [dstq ], m1
+ mova [dstq+strideq ], m2
+ pshufd m1, m0, 0xaa
+ pshufd m2, m0, 0xff
+ mova [dstq+strideq*2], m1
+ mova [dstq+stride3q ], m2
+ inc lineq
+ lea leftq, [leftq+4 ]
+ lea dstq, [dstq+strideq*4]
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal h_predictor_32x32, 2, 5, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ mov lineq, -8
+ DEFINE_ARGS dst, stride, line, left, stride3
+ lea stride3q, [strideq*3]
+.loop:
+ movd m0, [leftq]
+ punpcklbw m0, m0
+ punpcklbw m0, m0 ; l1 to l4 each repeated 4 times
+ pshufd m1, m0, 0x0 ; l1 repeated 16 times
+ pshufd m2, m0, 0x55 ; l2 repeated 16 times
+ mova [dstq ], m1
+ mova [dstq+16 ], m1
+ mova [dstq+strideq ], m2
+ mova [dstq+strideq+16 ], m2
+ pshufd m1, m0, 0xaa
+ pshufd m2, m0, 0xff
+ mova [dstq+strideq*2 ], m1
+ mova [dstq+strideq*2+16], m1
+ mova [dstq+stride3q ], m2
+ mova [dstq+stride3q+16 ], m2
+ inc lineq
+ lea leftq, [leftq+4 ]
+ lea dstq, [dstq+strideq*4]
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
cglobal tm_predictor_4x4, 4, 4, 5, dst, stride, above, left
pxor m1, m1
movq m0, [aboveq-1]; [63:0] tl t1 t2 t3 t4 x x x
diff --git a/vpx_dsp/x86/intrapred_ssse3.asm b/vpx_dsp/x86/intrapred_ssse3.asm
index 88df9b2..d061278 100644
--- a/vpx_dsp/x86/intrapred_ssse3.asm
+++ b/vpx_dsp/x86/intrapred_ssse3.asm
@@ -34,80 +34,6 @@
SECTION .text
INIT_MMX ssse3
-cglobal h_predictor_4x4, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 4
- mov lineq, -2
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- movd [dstq ], m1
- movd [dstq+strideq], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_MMX ssse3
-cglobal h_predictor_8x8, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 8
- mov lineq, -4
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- movq [dstq ], m1
- movq [dstq+strideq], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_XMM ssse3
-cglobal h_predictor_16x16, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 16
- mov lineq, -8
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- mova [dstq ], m1
- mova [dstq+strideq], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_XMM ssse3
-cglobal h_predictor_32x32, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 32
- mov lineq, -16
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- mova [dstq ], m1
- mova [dstq +16], m1
- mova [dstq+strideq ], m2
- mova [dstq+strideq+16], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_MMX ssse3
cglobal d45_predictor_4x4, 3, 4, 4, dst, stride, above, goffset
GET_GOT goffsetq
diff --git a/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm b/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
index 2675eab..80a330b 100644
--- a/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
+++ b/vpx_dsp/x86/inv_txfm_ssse3_x86_64.asm
@@ -18,9 +18,13 @@
pw_11585x2: times 8 dw 23170
-pw_m2404x2: times 8 dw -2404*2
-pw_m4756x2: times 8 dw -4756*2
-pw_m5520x2: times 8 dw -5520*2
+pw_m2404x2: times 8 dw -2404*2
+pw_m4756x2: times 8 dw -4756*2
+pw_m5520x2: times 8 dw -5520*2
+pw_m8423x2: times 8 dw -8423*2
+pw_m9102x2: times 8 dw -9102*2
+pw_m10394x2: times 8 dw -10394*2
+pw_m11003x2: times 8 dw -11003*2
pw_16364x2: times 8 dw 16364*2
pw_16305x2: times 8 dw 16305*2
@@ -29,6 +33,18 @@
pw_15893x2: times 8 dw 15893*2
pw_15679x2: times 8 dw 15679*2
pw_15426x2: times 8 dw 15426*2
+pw_15137x2: times 8 dw 15137*2
+pw_14811x2: times 8 dw 14811*2
+pw_14449x2: times 8 dw 14449*2
+pw_14053x2: times 8 dw 14053*2
+pw_13623x2: times 8 dw 13623*2
+pw_13160x2: times 8 dw 13160*2
+pw_12665x2: times 8 dw 12665*2
+pw_12140x2: times 8 dw 12140*2
+pw__9760x2: times 8 dw 9760*2
+pw__7723x2: times 8 dw 7723*2
+pw__7005x2: times 8 dw 7005*2
+pw__6270x2: times 8 dw 6270*2
pw__3981x2: times 8 dw 3981*2
pw__3196x2: times 8 dw 3196*2
pw__1606x2: times 8 dw 1606*2
@@ -57,6 +73,14 @@
TRANSFORM_COEFFS 15679, 4756
TRANSFORM_COEFFS 11585, 11585
+; constants for 32x32_1024
+TRANSFORM_COEFFS 12140, 11003
+TRANSFORM_COEFFS 7005, 14811
+TRANSFORM_COEFFS 14053, 8423
+TRANSFORM_COEFFS 9760, 13160
+TRANSFORM_COEFFS 12665, 10394
+TRANSFORM_COEFFS 7723, 14449
+
%macro PAIR_PP_COEFFS 2
dpw_%1_%2: dw %1, %1, %1, %1, %2, %2, %2, %2
%endmacro
@@ -368,23 +392,24 @@
%define idx30 16 * 6
%define idx31 16 * 7
+; FROM idct32x32_add_neon.asm
+;
+; Instead of doing the transforms stage by stage, it is done by loading
+; some input values and doing as many stages as possible to minimize the
+; storing/loading of intermediate results. To fit within registers, the
+; final coefficients are cut into four blocks:
+; BLOCK A: 16-19,28-31
+; BLOCK B: 20-23,24-27
+; BLOCK C: 8-11,12-15
+; BLOCK D: 0-3,4-7
+; Blocks A and C are straight calculation through the various stages. In
+; block B, further calculations are performed using the results from
+; block A. In block D, further calculations are performed using the results
+; from block C and then the final calculations are done using results from
+; block A and B which have been combined at the end of block B.
+;
+
%macro IDCT32X32_34 4
- ; FROM idct32x32_add_neon.asm
- ;
- ; Instead of doing the transforms stage by stage, it is done by loading
- ; some input values and doing as many stages as possible to minimize the
- ; storing/loading of intermediate results. To fit within registers, the
- ; final coefficients are cut into four blocks:
- ; BLOCK A: 16-19,28-31
- ; BLOCK B: 20-23,24-27
- ; BLOCK C: 8-11,12-15
- ; BLOCK D: 0-3,4-7
- ; Blocks A and C are straight calculation through the various stages. In
- ; block B, further calculations are performed using the results from
- ; block A. In block D, further calculations are performed using the results
- ; from block C and then the final calculations are done using results from
- ; block A and B which have been combined at the end of block B.
- ;
; BLOCK A STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
mova m11, m1
pmulhrsw m1, [pw___804x2] ; stp1_16
@@ -475,7 +500,7 @@
; BLOCK B STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
%if 0 ; overflow occurs in SUM_SUB when using test streams
mova m10, [pw_11585x2]
- SUM_SUB 6, 5, 9
+ SUM_SUB 6, 5, 9
pmulhrsw m6, m10 ; stp1_27
pmulhrsw m5, m10 ; stp1_20
SUM_SUB 13, 14, 9
@@ -539,10 +564,10 @@
; BLOCK C STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
%if 0 ; overflow occurs in SUM_SUB when using test streams
mova m10, [pw_11585x2]
- SUM_SUB 5, 4, 9
+ SUM_SUB 5, 4, 9
pmulhrsw m5, m10 ; stp1_13
pmulhrsw m4, m10 ; stp1_10
- SUM_SUB 6, 7, 9
+ SUM_SUB 6, 7, 9
pmulhrsw m6, m10 ; stp1_12
pmulhrsw m7, m10 ; stp1_11
%else
@@ -783,4 +808,887 @@
RECON_AND_STORE pass_two_start
RET
+
+%macro IDCT32X32_135 4
+ ; BLOCK A STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m1, [rsp + transposed_in + 16 * 1]
+ mova m11, m1
+ pmulhrsw m1, [pw___804x2] ; stp1_16
+ pmulhrsw m11, [pw_16364x2] ; stp2_31
+
+ mova m7, [rsp + transposed_in + 16 * 7]
+ mova m12, m7
+ pmulhrsw m7, [pw_15426x2] ; stp1_28
+ pmulhrsw m12, [pw_m5520x2] ; stp2_19
+
+ mova m3, [rsp + transposed_in + 16 * 9]
+ mova m4, m3
+ pmulhrsw m3, [pw__7005x2] ; stp1_18
+ pmulhrsw m4, [pw_14811x2] ; stp2_29
+
+ mova m0, [rsp + transposed_in + 16 * 15]
+ mova m2, m0
+ pmulhrsw m0, [pw_12140x2] ; stp1_30
+ pmulhrsw m2, [pw_m11003x2] ; stp2_17
+
+ ; BLOCK A STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 1, 2, 9 ; stp2_16, stp2_17
+ SUM_SUB 12, 3, 9 ; stp2_19, stp2_18
+ SUM_SUB 7, 4, 9 ; stp2_28, stp2_29
+ SUM_SUB 11, 0, 9 ; stp2_31, stp2_30
+
+ ; BLOCK A STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4X 0, 2, 3196, 16069, m8, 9, 10 ; stp1_17, stp1_30
+ BUTTERFLY_4Xmm 4, 3, 3196, 16069, m8, 9, 10 ; stp1_29, stp1_18
+
+ ; BLOCK A STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 1, 12, 9 ; stp2_16, stp2_19
+ SUM_SUB 0, 3, 9 ; stp2_17, stp2_18
+ SUM_SUB 11, 7, 9 ; stp2_31, stp2_28
+ SUM_SUB 2, 4, 9 ; stp2_30, stp2_29
+
+ ; BLOCK A STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4X 4, 3, 6270, 15137, m8, 9, 10 ; stp1_18, stp1_29
+ BUTTERFLY_4X 7, 12, 6270, 15137, m8, 9, 10 ; stp1_19, stp1_28
+
+ mova [stp + %3 + idx16], m1
+ mova [stp + %3 + idx17], m0
+ mova [stp + %3 + idx18], m4
+ mova [stp + %3 + idx19], m7
+ mova [stp + %4 + idx28], m12
+ mova [stp + %4 + idx29], m3
+ mova [stp + %4 + idx30], m2
+ mova [stp + %4 + idx31], m11
+
+ ; BLOCK B STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m2, [rsp + transposed_in + 16 * 3]
+ mova m3, m2
+ pmulhrsw m3, [pw_m2404x2] ; stp1_23
+ pmulhrsw m2, [pw_16207x2] ; stp2_24
+
+ mova m5, [rsp + transposed_in + 16 * 5]
+ mova m6, m5
+ pmulhrsw m5, [pw__3981x2] ; stp1_20
+ pmulhrsw m6, [pw_15893x2] ; stp2_27
+
+ mova m14, [rsp + transposed_in + 16 * 11]
+ mova m13, m14
+ pmulhrsw m13, [pw_m8423x2] ; stp1_21
+ pmulhrsw m14, [pw_14053x2] ; stp2_26
+
+ mova m0, [rsp + transposed_in + 16 * 13]
+ mova m1, m0
+ pmulhrsw m0, [pw__9760x2] ; stp1_22
+ pmulhrsw m1, [pw_13160x2] ; stp2_25
+
+ ; BLOCK B STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 5, 13, 9 ; stp2_20, stp2_21
+ SUM_SUB 3, 0, 9 ; stp2_23, stp2_22
+ SUM_SUB 2, 1, 9 ; stp2_24, stp2_25
+ SUM_SUB 6, 14, 9 ; stp2_27, stp2_26
+
+ ; BLOCK B STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4X 14, 13, 13623, 9102, m8, 9, 10 ; stp1_21, stp1_26
+ BUTTERFLY_4Xmm 1, 0, 13623, 9102, m8, 9, 10 ; stp1_25, stp1_22
+
+ ; BLOCK B STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 3, 5, 9 ; stp2_23, stp2_20
+ SUM_SUB 0, 14, 9 ; stp2_22, stp2_21
+ SUM_SUB 2, 6, 9 ; stp2_24, stp2_27
+ SUM_SUB 1, 13, 9 ; stp2_25, stp2_26
+
+ ; BLOCK B STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4Xmm 6, 5, 6270, 15137, m8, 9, 10 ; stp1_27, stp1_20
+ BUTTERFLY_4Xmm 13, 14, 6270, 15137, m8, 9, 10 ; stp1_26, stp1_21
+
+ ; BLOCK B STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m4, [stp + %3 + idx16]
+ mova m7, [stp + %3 + idx17]
+ mova m11, [stp + %3 + idx18]
+ mova m12, [stp + %3 + idx19]
+ SUM_SUB 4, 3, 9 ; stp2_16, stp2_23
+ SUM_SUB 7, 0, 9 ; stp2_17, stp2_22
+ SUM_SUB 11, 14, 9 ; stp2_18, stp2_21
+ SUM_SUB 12, 5, 9 ; stp2_19, stp2_20
+ mova [stp + %3 + idx16], m4
+ mova [stp + %3 + idx17], m7
+ mova [stp + %3 + idx18], m11
+ mova [stp + %3 + idx19], m12
+
+ mova m4, [stp + %4 + idx28]
+ mova m7, [stp + %4 + idx29]
+ mova m11, [stp + %4 + idx30]
+ mova m12, [stp + %4 + idx31]
+ SUM_SUB 4, 6, 9 ; stp2_28, stp2_27
+ SUM_SUB 7, 13, 9 ; stp2_29, stp2_26
+ SUM_SUB 11, 1, 9 ; stp2_30, stp2_25
+ SUM_SUB 12, 2, 9 ; stp2_31, stp2_24
+ mova [stp + %4 + idx28], m4
+ mova [stp + %4 + idx29], m7
+ mova [stp + %4 + idx30], m11
+ mova [stp + %4 + idx31], m12
+
+ ; BLOCK B STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+%if 0 ; overflow occurs in SUM_SUB when using test streams
+ mova m10, [pw_11585x2]
+ SUM_SUB 6, 5, 9
+ pmulhrsw m6, m10 ; stp1_27
+ pmulhrsw m5, m10 ; stp1_20
+ SUM_SUB 13, 14, 9
+ pmulhrsw m13, m10 ; stp1_26
+ pmulhrsw m14, m10 ; stp1_21
+ SUM_SUB 1, 0, 9
+ pmulhrsw m1, m10 ; stp1_25
+ pmulhrsw m0, m10 ; stp1_22
+ SUM_SUB 2, 3, 9
+ pmulhrsw m2, m10 ; stp1_25
+ pmulhrsw m3, m10 ; stp1_22
+%else
+ BUTTERFLY_4X 6, 5, 11585, 11585, m8, 9, 10 ; stp1_20, stp1_27
+ SWAP 6, 5
+ BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_21, stp1_26
+ SWAP 13, 14
+ BUTTERFLY_4X 1, 0, 11585, 11585, m8, 9, 10 ; stp1_22, stp1_25
+ SWAP 1, 0
+ BUTTERFLY_4X 2, 3, 11585, 11585, m8, 9, 10 ; stp1_23, stp1_24
+ SWAP 2, 3
+%endif
+ mova [stp + %3 + idx20], m5
+ mova [stp + %3 + idx21], m14
+ mova [stp + %3 + idx22], m0
+ mova [stp + %3 + idx23], m3
+ mova [stp + %4 + idx24], m2
+ mova [stp + %4 + idx25], m1
+ mova [stp + %4 + idx26], m13
+ mova [stp + %4 + idx27], m6
+
+ ; BLOCK C STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ ;
+ ; BLOCK C STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m0, [rsp + transposed_in + 16 * 2]
+ mova m1, m0
+ pmulhrsw m0, [pw__1606x2] ; stp1_8
+ pmulhrsw m1, [pw_16305x2] ; stp2_15
+
+ mova m6, [rsp + transposed_in + 16 * 6]
+ mova m7, m6
+ pmulhrsw m7, [pw_m4756x2] ; stp2_11
+ pmulhrsw m6, [pw_15679x2] ; stp1_12
+
+ mova m4, [rsp + transposed_in + 16 * 10]
+ mova m5, m4
+ pmulhrsw m4, [pw__7723x2] ; stp1_10
+ pmulhrsw m5, [pw_14449x2] ; stp2_13
+
+ mova m2, [rsp + transposed_in + 16 * 14]
+ mova m3, m2
+ pmulhrsw m3, [pw_m10394x2] ; stp1_9
+ pmulhrsw m2, [pw_12665x2] ; stp2_14
+
+ ; BLOCK C STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 0, 3, 9 ; stp1_8, stp1_9
+ SUM_SUB 7, 4, 9 ; stp1_11, stp1_10
+ SUM_SUB 6, 5, 9 ; stp1_12, stp1_13
+ SUM_SUB 1, 2, 9 ; stp1_15, stp1_14
+
+ ; BLOCK C STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4X 2, 3, 6270, 15137, m8, 9, 10 ; stp1_9, stp1_14
+ BUTTERFLY_4Xmm 5, 4, 6270, 15137, m8, 9, 10 ; stp1_13, stp1_10
+
+ ; BLOCK C STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 0, 7, 9 ; stp1_8, stp1_11
+ SUM_SUB 2, 4, 9 ; stp1_9, stp1_10
+ SUM_SUB 1, 6, 9 ; stp1_15, stp1_12
+ SUM_SUB 3, 5, 9 ; stp1_14, stp1_13
+
+ ; BLOCK C STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+%if 0 ; overflow occurs in SUM_SUB when using test streams
+ mova m10, [pw_11585x2]
+ SUM_SUB 5, 4, 9
+ pmulhrsw m5, m10 ; stp1_13
+ pmulhrsw m4, m10 ; stp1_10
+ SUM_SUB 6, 7, 9
+ pmulhrsw m6, m10 ; stp1_12
+ pmulhrsw m7, m10 ; stp1_11
+%else
+ BUTTERFLY_4X 5, 4, 11585, 11585, m8, 9, 10 ; stp1_10, stp1_13
+ SWAP 5, 4
+ BUTTERFLY_4X 6, 7, 11585, 11585, m8, 9, 10 ; stp1_11, stp1_12
+ SWAP 6, 7
+%endif
+ ; BLOCK C STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova [stp + %2 + idx8], m0
+ mova [stp + %2 + idx9], m2
+ mova [stp + %2 + idx10], m4
+ mova [stp + %2 + idx11], m7
+ mova [stp + %2 + idx12], m6
+ mova [stp + %2 + idx13], m5
+ mova [stp + %2 + idx14], m3
+ mova [stp + %2 + idx15], m1
+
+ ; BLOCK D STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ ;
+ ; BLOCK D STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ ;
+ ; BLOCK D STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m11, [rsp + transposed_in + 16 * 4]
+ mova m12, m11
+ pmulhrsw m11, [pw__3196x2] ; stp1_4
+ pmulhrsw m12, [pw_16069x2] ; stp1_7
+
+ mova m13, [rsp + transposed_in + 16 * 12]
+ mova m14, m13
+ pmulhrsw m13, [pw_13623x2] ; stp1_6
+ pmulhrsw m14, [pw_m9102x2] ; stp1_5
+
+ ; BLOCK D STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m0, [rsp + transposed_in + 16 * 0]
+ mova m2, [rsp + transposed_in + 16 * 8]
+ pmulhrsw m0, [pw_11585x2] ; stp1_1
+ mova m3, m2
+ pmulhrsw m2, [pw__6270x2] ; stp1_2
+ pmulhrsw m3, [pw_15137x2] ; stp1_3
+
+ SUM_SUB 11, 14, 9 ; stp1_4, stp1_5
+ SUM_SUB 12, 13, 9 ; stp1_7, stp1_6
+
+ ; BLOCK D STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+%if 0 ; overflow occurs in SUM_SUB when using test streams
+ mova m10, [pw_11585x2]
+ SUM_SUB 13, 14, 9
+ pmulhrsw m13, m10 ; stp1_6
+ pmulhrsw m14, m10 ; stp1_5
+%else
+ BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_5, stp1_6
+ SWAP 13, 14
+%endif
+ mova m1, m0 ; stp1_0 = stp1_1
+ SUM_SUB 0, 3, 9 ; stp1_0, stp1_3
+ SUM_SUB 1, 2, 9 ; stp1_1, stp1_2
+
+ ; BLOCK D STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 0, 12, 9 ; stp1_0, stp1_7
+ SUM_SUB 1, 13, 9 ; stp1_1, stp1_6
+ SUM_SUB 2, 14, 9 ; stp1_2, stp1_5
+ SUM_SUB 3, 11, 9 ; stp1_3, stp1_4
+
+ ; BLOCK D STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m4, [stp + %2 + idx12]
+ mova m5, [stp + %2 + idx13]
+ mova m6, [stp + %2 + idx14]
+ mova m7, [stp + %2 + idx15]
+ SUM_SUB 0, 7, 9 ; stp1_0, stp1_15
+ SUM_SUB 1, 6, 9 ; stp1_1, stp1_14
+ SUM_SUB 2, 5, 9 ; stp1_2, stp1_13
+ SUM_SUB 3, 4, 9 ; stp1_3, stp1_12
+
+ ; 0-3, 28-31 final stage
+ mova m10, [stp + %4 + idx31]
+ mova m15, [stp + %4 + idx30]
+ SUM_SUB 0, 10, 9 ; stp1_0, stp1_31
+ SUM_SUB 1, 15, 9 ; stp1_1, stp1_30
+ mova [stp + %1 + idx0], m0
+ mova [stp + %1 + idx1], m1
+ mova [stp + %4 + idx31], m10
+ mova [stp + %4 + idx30], m15
+ mova m0, [stp + %4 + idx29]
+ mova m1, [stp + %4 + idx28]
+ SUM_SUB 2, 0, 9 ; stp1_2, stp1_29
+ SUM_SUB 3, 1, 9 ; stp1_3, stp1_28
+ mova [stp + %1 + idx2], m2
+ mova [stp + %1 + idx3], m3
+ mova [stp + %4 + idx29], m0
+ mova [stp + %4 + idx28], m1
+
+ ; 12-15, 16-19 final stage
+ mova m0, [stp + %3 + idx16]
+ mova m1, [stp + %3 + idx17]
+ mova m2, [stp + %3 + idx18]
+ mova m3, [stp + %3 + idx19]
+ SUM_SUB 7, 0, 9 ; stp1_15, stp1_16
+ SUM_SUB 6, 1, 9 ; stp1_14, stp1_17
+ SUM_SUB 5, 2, 9 ; stp1_13, stp1_18
+ SUM_SUB 4, 3, 9 ; stp1_12, stp1_19
+ mova [stp + %2 + idx12], m4
+ mova [stp + %2 + idx13], m5
+ mova [stp + %2 + idx14], m6
+ mova [stp + %2 + idx15], m7
+ mova [stp + %3 + idx16], m0
+ mova [stp + %3 + idx17], m1
+ mova [stp + %3 + idx18], m2
+ mova [stp + %3 + idx19], m3
+
+ mova m4, [stp + %2 + idx8]
+ mova m5, [stp + %2 + idx9]
+ mova m6, [stp + %2 + idx10]
+ mova m7, [stp + %2 + idx11]
+ SUM_SUB 11, 7, 9 ; stp1_4, stp1_11
+ SUM_SUB 14, 6, 9 ; stp1_5, stp1_10
+ SUM_SUB 13, 5, 9 ; stp1_6, stp1_9
+ SUM_SUB 12, 4, 9 ; stp1_7, stp1_8
+
+ ; 4-7, 24-27 final stage
+ mova m3, [stp + %4 + idx24]
+ mova m2, [stp + %4 + idx25]
+ mova m1, [stp + %4 + idx26]
+ mova m0, [stp + %4 + idx27]
+ SUM_SUB 12, 3, 9 ; stp1_7, stp1_24
+ SUM_SUB 13, 2, 9 ; stp1_6, stp1_25
+ SUM_SUB 14, 1, 9 ; stp1_5, stp1_26
+ SUM_SUB 11, 0, 9 ; stp1_4, stp1_27
+ mova [stp + %4 + idx24], m3
+ mova [stp + %4 + idx25], m2
+ mova [stp + %4 + idx26], m1
+ mova [stp + %4 + idx27], m0
+ mova [stp + %1 + idx4], m11
+ mova [stp + %1 + idx5], m14
+ mova [stp + %1 + idx6], m13
+ mova [stp + %1 + idx7], m12
+
+ ; 8-11, 20-23 final stage
+ mova m0, [stp + %3 + idx20]
+ mova m1, [stp + %3 + idx21]
+ mova m2, [stp + %3 + idx22]
+ mova m3, [stp + %3 + idx23]
+ SUM_SUB 7, 0, 9 ; stp1_11, stp_20
+ SUM_SUB 6, 1, 9 ; stp1_10, stp_21
+ SUM_SUB 5, 2, 9 ; stp1_9, stp_22
+ SUM_SUB 4, 3, 9 ; stp1_8, stp_23
+ mova [stp + %2 + idx8], m4
+ mova [stp + %2 + idx9], m5
+ mova [stp + %2 + idx10], m6
+ mova [stp + %2 + idx11], m7
+ mova [stp + %3 + idx20], m0
+ mova [stp + %3 + idx21], m1
+ mova [stp + %3 + idx22], m2
+ mova [stp + %3 + idx23], m3
+%endmacro
+
+INIT_XMM ssse3
+cglobal idct32x32_135_add, 3, 11, 16, i32x32_size, input, output, stride
+ mova m8, [pd_8192]
+ mov r6, 2
+ lea stp, [rsp + pass_one_start]
+
+idct32x32_135:
+ mov r3, inputq
+ lea r4, [rsp + transposed_in]
+ mov r7, 2
+
+idct32x32_135_transpose:
+ mova m0, [r3 + 0]
+ mova m1, [r3 + 16 * 4]
+ mova m2, [r3 + 16 * 8]
+ mova m3, [r3 + 16 * 12]
+ mova m4, [r3 + 16 * 16]
+ mova m5, [r3 + 16 * 20]
+ mova m6, [r3 + 16 * 24]
+ mova m7, [r3 + 16 * 28]
+
+ TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
+
+ mova [r4 + 0], m0
+ mova [r4 + 16 * 1], m1
+ mova [r4 + 16 * 2], m2
+ mova [r4 + 16 * 3], m3
+ mova [r4 + 16 * 4], m4
+ mova [r4 + 16 * 5], m5
+ mova [r4 + 16 * 6], m6
+ mova [r4 + 16 * 7], m7
+
+ add r3, 16
+ add r4, 16 * 8
+ dec r7
+ jne idct32x32_135_transpose
+
+ IDCT32X32_135 16*0, 16*32, 16*64, 16*96
+ lea stp, [stp + 16 * 8]
+ lea inputq, [inputq + 16 * 32]
+ dec r6
+ jnz idct32x32_135
+
+ mov r6, 4
+ lea stp, [rsp + pass_one_start]
+ lea r9, [rsp + pass_one_start]
+
+idct32x32_135_2:
+ lea r4, [rsp + transposed_in]
+ mov r3, r9
+ mov r7, 2
+
+idct32x32_135_transpose_2:
+ mova m0, [r3 + 0]
+ mova m1, [r3 + 16 * 1]
+ mova m2, [r3 + 16 * 2]
+ mova m3, [r3 + 16 * 3]
+ mova m4, [r3 + 16 * 4]
+ mova m5, [r3 + 16 * 5]
+ mova m6, [r3 + 16 * 6]
+ mova m7, [r3 + 16 * 7]
+
+ TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
+
+ mova [r4 + 0], m0
+ mova [r4 + 16 * 1], m1
+ mova [r4 + 16 * 2], m2
+ mova [r4 + 16 * 3], m3
+ mova [r4 + 16 * 4], m4
+ mova [r4 + 16 * 5], m5
+ mova [r4 + 16 * 6], m6
+ mova [r4 + 16 * 7], m7
+
+ add r3, 16 * 8
+ add r4, 16 * 8
+ dec r7
+ jne idct32x32_135_transpose_2
+
+ IDCT32X32_135 16*0, 16*8, 16*16, 16*24
+
+ lea stp, [stp + 16 * 32]
+ add r9, 16 * 32
+ dec r6
+ jnz idct32x32_135_2
+
+ RECON_AND_STORE pass_two_start
+
+ RET
+
+%macro IDCT32X32_1024 4
+ ; BLOCK A STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m1, [rsp + transposed_in + 16 * 1]
+ mova m11, [rsp + transposed_in + 16 * 31]
+ BUTTERFLY_4X 1, 11, 804, 16364, m8, 9, 10 ; stp1_16, stp1_31
+
+ mova m0, [rsp + transposed_in + 16 * 15]
+ mova m2, [rsp + transposed_in + 16 * 17]
+ BUTTERFLY_4X 2, 0, 12140, 11003, m8, 9, 10 ; stp1_17, stp1_30
+
+ mova m7, [rsp + transposed_in + 16 * 7]
+ mova m12, [rsp + transposed_in + 16 * 25]
+ BUTTERFLY_4X 12, 7, 15426, 5520, m8, 9, 10 ; stp1_19, stp1_28
+
+ mova m3, [rsp + transposed_in + 16 * 9]
+ mova m4, [rsp + transposed_in + 16 * 23]
+ BUTTERFLY_4X 3, 4, 7005, 14811, m8, 9, 10 ; stp1_18, stp1_29
+
+ ; BLOCK A STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 1, 2, 9 ; stp2_16, stp2_17
+ SUM_SUB 12, 3, 9 ; stp2_19, stp2_18
+ SUM_SUB 7, 4, 9 ; stp2_28, stp2_29
+ SUM_SUB 11, 0, 9 ; stp2_31, stp2_30
+
+ ; BLOCK A STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4X 0, 2, 3196, 16069, m8, 9, 10 ; stp1_17, stp1_30
+ BUTTERFLY_4Xmm 4, 3, 3196, 16069, m8, 9, 10 ; stp1_29, stp1_18
+
+ ; BLOCK A STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 1, 12, 9 ; stp2_16, stp2_19
+ SUM_SUB 0, 3, 9 ; stp2_17, stp2_18
+ SUM_SUB 11, 7, 9 ; stp2_31, stp2_28
+ SUM_SUB 2, 4, 9 ; stp2_30, stp2_29
+
+ ; BLOCK A STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4X 4, 3, 6270, 15137, m8, 9, 10 ; stp1_18, stp1_29
+ BUTTERFLY_4X 7, 12, 6270, 15137, m8, 9, 10 ; stp1_19, stp1_28
+
+ mova [stp + %3 + idx16], m1
+ mova [stp + %3 + idx17], m0
+ mova [stp + %3 + idx18], m4
+ mova [stp + %3 + idx19], m7
+ mova [stp + %4 + idx28], m12
+ mova [stp + %4 + idx29], m3
+ mova [stp + %4 + idx30], m2
+ mova [stp + %4 + idx31], m11
+
+ ; BLOCK B STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m5, [rsp + transposed_in + 16 * 5]
+ mova m6, [rsp + transposed_in + 16 * 27]
+ BUTTERFLY_4X 5, 6, 3981, 15893, m8, 9, 10 ; stp1_20, stp1_27
+
+ mova m13, [rsp + transposed_in + 16 * 21]
+ mova m14, [rsp + transposed_in + 16 * 11]
+ BUTTERFLY_4X 13, 14, 14053, 8423, m8, 9, 10 ; stp1_21, stp1_26
+
+ mova m0, [rsp + transposed_in + 16 * 13]
+ mova m1, [rsp + transposed_in + 16 * 19]
+ BUTTERFLY_4X 0, 1, 9760, 13160, m8, 9, 10 ; stp1_22, stp1_25
+
+ mova m2, [rsp + transposed_in + 16 * 3]
+ mova m3, [rsp + transposed_in + 16 * 29]
+ BUTTERFLY_4X 3, 2, 16207, 2404, m8, 9, 10 ; stp1_23, stp1_24
+
+ ; BLOCK B STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 5, 13, 9 ; stp2_20, stp2_21
+ SUM_SUB 3, 0, 9 ; stp2_23, stp2_22
+ SUM_SUB 2, 1, 9 ; stp2_24, stp2_25
+ SUM_SUB 6, 14, 9 ; stp2_27, stp2_26
+
+ ; BLOCK B STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4X 14, 13, 13623, 9102, m8, 9, 10 ; stp1_21, stp1_26
+ BUTTERFLY_4Xmm 1, 0, 13623, 9102, m8, 9, 10 ; stp1_25, stp1_22
+
+ ; BLOCK B STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 3, 5, 9 ; stp2_23, stp2_20
+ SUM_SUB 0, 14, 9 ; stp2_22, stp2_21
+ SUM_SUB 2, 6, 9 ; stp2_24, stp2_27
+ SUM_SUB 1, 13, 9 ; stp2_25, stp2_26
+
+ ; BLOCK B STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4Xmm 6, 5, 6270, 15137, m8, 9, 10 ; stp1_27, stp1_20
+ BUTTERFLY_4Xmm 13, 14, 6270, 15137, m8, 9, 10 ; stp1_26, stp1_21
+
+ ; BLOCK B STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m4, [stp + %3 + idx16]
+ mova m7, [stp + %3 + idx17]
+ mova m11, [stp + %3 + idx18]
+ mova m12, [stp + %3 + idx19]
+ SUM_SUB 4, 3, 9 ; stp2_16, stp2_23
+ SUM_SUB 7, 0, 9 ; stp2_17, stp2_22
+ SUM_SUB 11, 14, 9 ; stp2_18, stp2_21
+ SUM_SUB 12, 5, 9 ; stp2_19, stp2_20
+ mova [stp + %3 + idx16], m4
+ mova [stp + %3 + idx17], m7
+ mova [stp + %3 + idx18], m11
+ mova [stp + %3 + idx19], m12
+
+ mova m4, [stp + %4 + idx28]
+ mova m7, [stp + %4 + idx29]
+ mova m11, [stp + %4 + idx30]
+ mova m12, [stp + %4 + idx31]
+ SUM_SUB 4, 6, 9 ; stp2_28, stp2_27
+ SUM_SUB 7, 13, 9 ; stp2_29, stp2_26
+ SUM_SUB 11, 1, 9 ; stp2_30, stp2_25
+ SUM_SUB 12, 2, 9 ; stp2_31, stp2_24
+ mova [stp + %4 + idx28], m4
+ mova [stp + %4 + idx29], m7
+ mova [stp + %4 + idx30], m11
+ mova [stp + %4 + idx31], m12
+
+ ; BLOCK B STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+%if 0 ; overflow occurs in SUM_SUB when using test streams
+ mova m10, [pw_11585x2]
+ SUM_SUB 6, 5, 9
+ pmulhrsw m6, m10 ; stp1_27
+ pmulhrsw m5, m10 ; stp1_20
+ SUM_SUB 13, 14, 9
+ pmulhrsw m13, m10 ; stp1_26
+ pmulhrsw m14, m10 ; stp1_21
+ SUM_SUB 1, 0, 9
+ pmulhrsw m1, m10 ; stp1_25
+ pmulhrsw m0, m10 ; stp1_22
+ SUM_SUB 2, 3, 9
+ pmulhrsw m2, m10 ; stp1_25
+ pmulhrsw m3, m10 ; stp1_22
+%else
+ BUTTERFLY_4X 6, 5, 11585, 11585, m8, 9, 10 ; stp1_20, stp1_27
+ SWAP 6, 5
+ BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_21, stp1_26
+ SWAP 13, 14
+ BUTTERFLY_4X 1, 0, 11585, 11585, m8, 9, 10 ; stp1_22, stp1_25
+ SWAP 1, 0
+ BUTTERFLY_4X 2, 3, 11585, 11585, m8, 9, 10 ; stp1_23, stp1_24
+ SWAP 2, 3
+%endif
+ mova [stp + %3 + idx20], m5
+ mova [stp + %3 + idx21], m14
+ mova [stp + %3 + idx22], m0
+ mova [stp + %3 + idx23], m3
+ mova [stp + %4 + idx24], m2
+ mova [stp + %4 + idx25], m1
+ mova [stp + %4 + idx26], m13
+ mova [stp + %4 + idx27], m6
+
+ ; BLOCK C STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ ;
+ ; BLOCK C STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m0, [rsp + transposed_in + 16 * 2]
+ mova m1, [rsp + transposed_in + 16 * 30]
+ BUTTERFLY_4X 0, 1, 1606, 16305, m8, 9, 10 ; stp1_8, stp1_15
+
+ mova m2, [rsp + transposed_in + 16 * 14]
+ mova m3, [rsp + transposed_in + 16 * 18]
+ BUTTERFLY_4X 3, 2, 12665, 10394, m8, 9, 10 ; stp1_9, stp1_14
+
+ mova m4, [rsp + transposed_in + 16 * 10]
+ mova m5, [rsp + transposed_in + 16 * 22]
+ BUTTERFLY_4X 4, 5, 7723, 14449, m8, 9, 10 ; stp1_10, stp1_13
+
+ mova m6, [rsp + transposed_in + 16 * 6]
+ mova m7, [rsp + transposed_in + 16 * 26]
+ BUTTERFLY_4X 7, 6, 15679, 4756, m8, 9, 10 ; stp1_11, stp1_12
+
+ ; BLOCK C STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 0, 3, 9 ; stp1_8, stp1_9
+ SUM_SUB 7, 4, 9 ; stp1_11, stp1_10
+ SUM_SUB 6, 5, 9 ; stp1_12, stp1_13
+ SUM_SUB 1, 2, 9 ; stp1_15, stp1_14
+
+ ; BLOCK C STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ BUTTERFLY_4X 2, 3, 6270, 15137, m8, 9, 10 ; stp1_9, stp1_14
+ BUTTERFLY_4Xmm 5, 4, 6270, 15137, m8, 9, 10 ; stp1_13, stp1_10
+
+ ; BLOCK C STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 0, 7, 9 ; stp1_8, stp1_11
+ SUM_SUB 2, 4, 9 ; stp1_9, stp1_10
+ SUM_SUB 1, 6, 9 ; stp1_15, stp1_12
+ SUM_SUB 3, 5, 9 ; stp1_14, stp1_13
+
+ ; BLOCK C STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+%if 0 ; overflow occurs in SUM_SUB when using test streams
+ mova m10, [pw_11585x2]
+ SUM_SUB 5, 4, 9
+ pmulhrsw m5, m10 ; stp1_13
+ pmulhrsw m4, m10 ; stp1_10
+ SUM_SUB 6, 7, 9
+ pmulhrsw m6, m10 ; stp1_12
+ pmulhrsw m7, m10 ; stp1_11
+%else
+ BUTTERFLY_4X 5, 4, 11585, 11585, m8, 9, 10 ; stp1_10, stp1_13
+ SWAP 5, 4
+ BUTTERFLY_4X 6, 7, 11585, 11585, m8, 9, 10 ; stp1_11, stp1_12
+ SWAP 6, 7
+%endif
+ ; BLOCK C STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova [stp + %2 + idx8], m0
+ mova [stp + %2 + idx9], m2
+ mova [stp + %2 + idx10], m4
+ mova [stp + %2 + idx11], m7
+ mova [stp + %2 + idx12], m6
+ mova [stp + %2 + idx13], m5
+ mova [stp + %2 + idx14], m3
+ mova [stp + %2 + idx15], m1
+
+ ; BLOCK D STAGE 1 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ ;
+ ; BLOCK D STAGE 2 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ ;
+ ; BLOCK D STAGE 3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m11, [rsp + transposed_in + 16 * 4]
+ mova m12, [rsp + transposed_in + 16 * 28]
+ BUTTERFLY_4X 11, 12, 3196, 16069, m8, 9, 10 ; stp1_4, stp1_7
+
+ mova m13, [rsp + transposed_in + 16 * 12]
+ mova m14, [rsp + transposed_in + 16 * 20]
+ BUTTERFLY_4X 14, 13, 13623, 9102, m8, 9, 10 ; stp1_5, stp1_6
+
+ ; BLOCK D STAGE 4 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m0, [rsp + transposed_in + 16 * 0]
+ mova m1, [rsp + transposed_in + 16 * 16]
+
+%if 0 ; overflow occurs in SUM_SUB when using test streams
+ mova m10, [pw_11585x2]
+ SUM_SUB 0, 1, 9
+ pmulhrsw m0, m10 ; stp1_1
+ pmulhrsw m1, m10 ; stp1_0
+%else
+ BUTTERFLY_4X 0, 1, 11585, 11585, m8, 9, 10 ; stp1_1, stp1_0
+ SWAP 0, 1
+%endif
+ mova m2, [rsp + transposed_in + 16 * 8]
+ mova m3, [rsp + transposed_in + 16 * 24]
+ BUTTERFLY_4X 2, 3, 6270, 15137, m8, 9, 10 ; stp1_2, stp1_3
+
+ mova m10, [pw_11585x2]
+ SUM_SUB 11, 14, 9 ; stp1_4, stp1_5
+ SUM_SUB 12, 13, 9 ; stp1_7, stp1_6
+
+ ; BLOCK D STAGE 5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+%if 0 ; overflow occurs in SUM_SUB when using test streams
+ SUM_SUB 13, 14, 9
+ pmulhrsw m13, m10 ; stp1_6
+ pmulhrsw m14, m10 ; stp1_5
+%else
+ BUTTERFLY_4X 13, 14, 11585, 11585, m8, 9, 10 ; stp1_5, stp1_6
+ SWAP 13, 14
+%endif
+ SUM_SUB 0, 3, 9 ; stp1_0, stp1_3
+ SUM_SUB 1, 2, 9 ; stp1_1, stp1_2
+
+ ; BLOCK D STAGE 6 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ SUM_SUB 0, 12, 9 ; stp1_0, stp1_7
+ SUM_SUB 1, 13, 9 ; stp1_1, stp1_6
+ SUM_SUB 2, 14, 9 ; stp1_2, stp1_5
+ SUM_SUB 3, 11, 9 ; stp1_3, stp1_4
+
+ ; BLOCK D STAGE 7 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ mova m4, [stp + %2 + idx12]
+ mova m5, [stp + %2 + idx13]
+ mova m6, [stp + %2 + idx14]
+ mova m7, [stp + %2 + idx15]
+ SUM_SUB 0, 7, 9 ; stp1_0, stp1_15
+ SUM_SUB 1, 6, 9 ; stp1_1, stp1_14
+ SUM_SUB 2, 5, 9 ; stp1_2, stp1_13
+ SUM_SUB 3, 4, 9 ; stp1_3, stp1_12
+
+ ; 0-3, 28-31 final stage
+ mova m10, [stp + %4 + idx31]
+ mova m15, [stp + %4 + idx30]
+ SUM_SUB 0, 10, 9 ; stp1_0, stp1_31
+ SUM_SUB 1, 15, 9 ; stp1_1, stp1_30
+ mova [stp + %1 + idx0], m0
+ mova [stp + %1 + idx1], m1
+ mova [stp + %4 + idx31], m10
+ mova [stp + %4 + idx30], m15
+ mova m0, [stp + %4 + idx29]
+ mova m1, [stp + %4 + idx28]
+ SUM_SUB 2, 0, 9 ; stp1_2, stp1_29
+ SUM_SUB 3, 1, 9 ; stp1_3, stp1_28
+ mova [stp + %1 + idx2], m2
+ mova [stp + %1 + idx3], m3
+ mova [stp + %4 + idx29], m0
+ mova [stp + %4 + idx28], m1
+
+ ; 12-15, 16-19 final stage
+ mova m0, [stp + %3 + idx16]
+ mova m1, [stp + %3 + idx17]
+ mova m2, [stp + %3 + idx18]
+ mova m3, [stp + %3 + idx19]
+ SUM_SUB 7, 0, 9 ; stp1_15, stp1_16
+ SUM_SUB 6, 1, 9 ; stp1_14, stp1_17
+ SUM_SUB 5, 2, 9 ; stp1_13, stp1_18
+ SUM_SUB 4, 3, 9 ; stp1_12, stp1_19
+ mova [stp + %2 + idx12], m4
+ mova [stp + %2 + idx13], m5
+ mova [stp + %2 + idx14], m6
+ mova [stp + %2 + idx15], m7
+ mova [stp + %3 + idx16], m0
+ mova [stp + %3 + idx17], m1
+ mova [stp + %3 + idx18], m2
+ mova [stp + %3 + idx19], m3
+
+ mova m4, [stp + %2 + idx8]
+ mova m5, [stp + %2 + idx9]
+ mova m6, [stp + %2 + idx10]
+ mova m7, [stp + %2 + idx11]
+ SUM_SUB 11, 7, 9 ; stp1_4, stp1_11
+ SUM_SUB 14, 6, 9 ; stp1_5, stp1_10
+ SUM_SUB 13, 5, 9 ; stp1_6, stp1_9
+ SUM_SUB 12, 4, 9 ; stp1_7, stp1_8
+
+ ; 4-7, 24-27 final stage
+ mova m3, [stp + %4 + idx24]
+ mova m2, [stp + %4 + idx25]
+ mova m1, [stp + %4 + idx26]
+ mova m0, [stp + %4 + idx27]
+ SUM_SUB 12, 3, 9 ; stp1_7, stp1_24
+ SUM_SUB 13, 2, 9 ; stp1_6, stp1_25
+ SUM_SUB 14, 1, 9 ; stp1_5, stp1_26
+ SUM_SUB 11, 0, 9 ; stp1_4, stp1_27
+ mova [stp + %4 + idx24], m3
+ mova [stp + %4 + idx25], m2
+ mova [stp + %4 + idx26], m1
+ mova [stp + %4 + idx27], m0
+ mova [stp + %1 + idx4], m11
+ mova [stp + %1 + idx5], m14
+ mova [stp + %1 + idx6], m13
+ mova [stp + %1 + idx7], m12
+
+ ; 8-11, 20-23 final stage
+ mova m0, [stp + %3 + idx20]
+ mova m1, [stp + %3 + idx21]
+ mova m2, [stp + %3 + idx22]
+ mova m3, [stp + %3 + idx23]
+ SUM_SUB 7, 0, 9 ; stp1_11, stp_20
+ SUM_SUB 6, 1, 9 ; stp1_10, stp_21
+ SUM_SUB 5, 2, 9 ; stp1_9, stp_22
+ SUM_SUB 4, 3, 9 ; stp1_8, stp_23
+ mova [stp + %2 + idx8], m4
+ mova [stp + %2 + idx9], m5
+ mova [stp + %2 + idx10], m6
+ mova [stp + %2 + idx11], m7
+ mova [stp + %3 + idx20], m0
+ mova [stp + %3 + idx21], m1
+ mova [stp + %3 + idx22], m2
+ mova [stp + %3 + idx23], m3
+%endmacro
+
+INIT_XMM ssse3
+cglobal idct32x32_1024_add, 3, 11, 16, i32x32_size, input, output, stride
+ mova m8, [pd_8192]
+ mov r6, 4
+ lea stp, [rsp + pass_one_start]
+
+idct32x32_1024:
+ mov r3, inputq
+ lea r4, [rsp + transposed_in]
+ mov r7, 4
+
+idct32x32_1024_transpose:
+ mova m0, [r3 + 0]
+ mova m1, [r3 + 16 * 4]
+ mova m2, [r3 + 16 * 8]
+ mova m3, [r3 + 16 * 12]
+ mova m4, [r3 + 16 * 16]
+ mova m5, [r3 + 16 * 20]
+ mova m6, [r3 + 16 * 24]
+ mova m7, [r3 + 16 * 28]
+
+ TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
+
+ mova [r4 + 0], m0
+ mova [r4 + 16 * 1], m1
+ mova [r4 + 16 * 2], m2
+ mova [r4 + 16 * 3], m3
+ mova [r4 + 16 * 4], m4
+ mova [r4 + 16 * 5], m5
+ mova [r4 + 16 * 6], m6
+ mova [r4 + 16 * 7], m7
+
+ add r3, 16
+ add r4, 16 * 8
+ dec r7
+ jne idct32x32_1024_transpose
+
+ IDCT32X32_1024 16*0, 16*32, 16*64, 16*96
+
+ lea stp, [stp + 16 * 8]
+ lea inputq, [inputq + 16 * 32]
+ dec r6
+ jnz idct32x32_1024
+
+ mov r6, 4
+ lea stp, [rsp + pass_one_start]
+ lea r9, [rsp + pass_one_start]
+
+idct32x32_1024_2:
+ lea r4, [rsp + transposed_in]
+ mov r3, r9
+ mov r7, 4
+
+idct32x32_1024_transpose_2:
+ mova m0, [r3 + 0]
+ mova m1, [r3 + 16 * 1]
+ mova m2, [r3 + 16 * 2]
+ mova m3, [r3 + 16 * 3]
+ mova m4, [r3 + 16 * 4]
+ mova m5, [r3 + 16 * 5]
+ mova m6, [r3 + 16 * 6]
+ mova m7, [r3 + 16 * 7]
+
+ TRANSPOSE8X8 0, 1, 2, 3, 4, 5, 6, 7, 9
+
+ mova [r4 + 0], m0
+ mova [r4 + 16 * 1], m1
+ mova [r4 + 16 * 2], m2
+ mova [r4 + 16 * 3], m3
+ mova [r4 + 16 * 4], m4
+ mova [r4 + 16 * 5], m5
+ mova [r4 + 16 * 6], m6
+ mova [r4 + 16 * 7], m7
+
+ add r3, 16 * 8
+ add r4, 16 * 8
+ dec r7
+ jne idct32x32_1024_transpose_2
+
+ IDCT32X32_1024 16*0, 16*8, 16*16, 16*24
+
+ lea stp, [stp + 16 * 32]
+ add r9, 16 * 32
+ dec r6
+ jnz idct32x32_1024_2
+
+ RECON_AND_STORE pass_two_start
+
+ RET
%endif
diff --git a/vpx_dsp/x86/subpel_variance_sse2.asm b/vpx_dsp/x86/subpel_variance_sse2.asm
index 05dcff7..c655e4b 100644
--- a/vpx_dsp/x86/subpel_variance_sse2.asm
+++ b/vpx_dsp/x86/subpel_variance_sse2.asm
@@ -139,8 +139,10 @@
%define sec_str sec_stridemp
;Store bilin_filter and pw_8 location in stack
- GET_GOT eax
- add esp, 4 ; restore esp
+ %if GET_GOT_DEFINED == 1
+ GET_GOT eax
+ add esp, 4 ; restore esp
+ %endif
lea ecx, [GLOBAL(bilin_filter_m)]
mov g_bilin_filterm, ecx
@@ -156,8 +158,10 @@
%define block_height heightd
;Store bilin_filter and pw_8 location in stack
- GET_GOT eax
- add esp, 4 ; restore esp
+ %if GET_GOT_DEFINED == 1
+ GET_GOT eax
+ add esp, 4 ; restore esp
+ %endif
lea ecx, [GLOBAL(bilin_filter_m)]
mov g_bilin_filterm, ecx
diff --git a/vpx_ports/x86_abi_support.asm b/vpx_ports/x86_abi_support.asm
index c94b76a..708fa10 100644
--- a/vpx_ports/x86_abi_support.asm
+++ b/vpx_ports/x86_abi_support.asm
@@ -189,7 +189,6 @@
%if ABI_IS_32BIT
%if CONFIG_PIC=1
%ifidn __OUTPUT_FORMAT__,elf32
- %define GET_GOT_SAVE_ARG 1
%define WRT_PLT wrt ..plt
%macro GET_GOT 1
extern _GLOBAL_OFFSET_TABLE_
@@ -208,7 +207,6 @@
%define RESTORE_GOT pop %1
%endmacro
%elifidn __OUTPUT_FORMAT__,macho32
- %define GET_GOT_SAVE_ARG 1
%macro GET_GOT 1
push %1
call %%get_got
