Merge "Align SAD output array to be 16-byte aligned"
diff --git a/build/make/gen_msvs_proj.sh b/build/make/gen_msvs_proj.sh
index becd958..c2ef44a 100755
--- a/build/make/gen_msvs_proj.sh
+++ b/build/make/gen_msvs_proj.sh
@@ -33,6 +33,7 @@
--proj-guid=GUID GUID to use for the project
--module-def=filename File containing export definitions (for DLLs)
--ver=version Version (7,8,9) of visual studio to generate for
+ --src-path-bare=dir Path to root of source tree
-Ipath/to/include Additional include directories
-DFLAG[=value] Preprocessor macros to define
-Lpath/to/lib Additional library search paths
@@ -191,6 +192,8 @@
;;
--lib) proj_kind="lib"
;;
+ --src-path-bare=*) src_path_bare="$optval"
+ ;;
--static-crt) use_static_runtime=true
;;
--ver=*)
@@ -335,6 +338,35 @@
case "$target" in
x86*)
case "$name" in
+ obj_int_extract)
+ tag Tool \
+ Name="VCCLCompilerTool" \
+ Optimization="0" \
+ AdditionalIncludeDirectories="$incs" \
+ PreprocessorDefinitions="WIN32;DEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE" \
+ RuntimeLibrary="$debug_runtime" \
+ WarningLevel="3" \
+ Detect64BitPortabilityProblems="true" \
+ DebugInformationFormat="1" \
+ ;;
+ vpx)
+ tag Tool \
+ Name="VCPreBuildEventTool" \
+ CommandLine="call obj_int_extract.bat $src_path_bare" \
+
+ tag Tool \
+ Name="VCCLCompilerTool" \
+ Optimization="0" \
+ AdditionalIncludeDirectories="$incs" \
+ PreprocessorDefinitions="WIN32;_DEBUG;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE;$defines" \
+ RuntimeLibrary="$debug_runtime" \
+ UsePrecompiledHeader="0" \
+ WarningLevel="3" \
+ DebugInformationFormat="1" \
+ Detect64BitPortabilityProblems="true" \
+
+ $uses_asm && tag Tool Name="YASM" IncludePaths="$incs" Debug="1"
+ ;;
*)
tag Tool \
Name="VCCLCompilerTool" \
@@ -358,6 +390,12 @@
case "$target" in
x86*)
case "$name" in
+ obj_int_extract)
+ tag Tool \
+ Name="VCLinkerTool" \
+ OutputFile="${name}.exe" \
+ GenerateDebugInformation="true" \
+ ;;
*)
tag Tool \
Name="VCLinkerTool" \
@@ -406,6 +444,34 @@
case "$target" in
x86*)
case "$name" in
+ obj_int_extract)
+ tag Tool \
+ Name="VCCLCompilerTool" \
+ AdditionalIncludeDirectories="$incs" \
+ PreprocessorDefinitions="WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE" \
+ RuntimeLibrary="$release_runtime" \
+ UsePrecompiledHeader="0" \
+ WarningLevel="3" \
+ Detect64BitPortabilityProblems="true" \
+ DebugInformationFormat="0" \
+ ;;
+ vpx)
+ tag Tool \
+ Name="VCPreBuildEventTool" \
+ CommandLine="call obj_int_extract.bat $src_path_bare" \
+
+ tag Tool \
+ Name="VCCLCompilerTool" \
+ AdditionalIncludeDirectories="$incs" \
+ PreprocessorDefinitions="WIN32;NDEBUG;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE;$defines" \
+ RuntimeLibrary="$release_runtime" \
+ UsePrecompiledHeader="0" \
+ WarningLevel="3" \
+ DebugInformationFormat="0" \
+ Detect64BitPortabilityProblems="true" \
+
+ $uses_asm && tag Tool Name="YASM" IncludePaths="$incs"
+ ;;
*)
tag Tool \
Name="VCCLCompilerTool" \
@@ -428,6 +494,12 @@
case "$target" in
x86*)
case "$name" in
+ obj_int_extract)
+ tag Tool \
+ Name="VCLinkerTool" \
+ OutputFile="${name}.exe" \
+ GenerateDebugInformation="true" \
+ ;;
*)
tag Tool \
Name="VCLinkerTool" \
diff --git a/build/make/obj_int_extract.c b/build/make/obj_int_extract.c
index 22c5cf2..01b3129 100644
--- a/build/make/obj_int_extract.c
+++ b/build/make/obj_int_extract.c
@@ -840,9 +840,18 @@
strtab_ptr = symtab_ptr + symtab_sz * 18;
if (nsections > 96)
- goto bail;
+ {
+ log_msg("Too many sections\n");
+ return 1;
+ }
- sectionlist = malloc(nsections * sizeof * sectionlist);
+ sectionlist = malloc(nsections * sizeof(sectionlist));
+
+ if (sectionlist == NULL)
+ {
+ log_msg("Allocating first level of section list failed\n");
+ return 1;
+ }
//log_msg("COFF: Found %u symbols in %u sections.\n", symtab_sz, nsections);
@@ -860,6 +869,12 @@
//log_msg("COFF: Parsing section %s\n",sectionname);
sectionlist[i] = malloc(strlen(sectionname) + 1);
+
+ if (sectionlist[i] == NULL)
+ {
+ log_msg("Allocating storage for %s failed\n", sectionname);
+ goto bail;
+ }
strcpy(sectionlist[i], sectionname);
if (!strcmp(sectionname, ".data")) sectionrawdata_ptr = get_le32(ptr + 20);
@@ -903,12 +918,14 @@
char name[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
strncpy(name, ptr, 8);
//log_msg("COFF: Parsing symbol %s\n",name);
+ /* +1 to avoid printing leading underscore */
printf("%-40s EQU ", name + 1);
}
else
{
//log_msg("COFF: Parsing symbol %s\n",
// buf + strtab_ptr + get_le32(ptr+4));
+ /* +1 to avoid printing leading underscore */
printf("%-40s EQU ", buf + strtab_ptr + get_le32(ptr + 4) + 1);
}
diff --git a/build/x86-msvs/obj_int_extract.bat b/build/x86-msvs/obj_int_extract.bat
new file mode 100644
index 0000000..1bb8653
--- /dev/null
+++ b/build/x86-msvs/obj_int_extract.bat
@@ -0,0 +1,15 @@
+REM Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+REM
+REM Use of this source code is governed by a BSD-style license
+REM that can be found in the LICENSE file in the root of the source
+REM tree. An additional intellectual property rights grant can be found
+REM in the file PATENTS. All contributing project authors may
+REM be found in the AUTHORS file in the root of the source tree.
+echo on
+
+cl /I "./" /I "%1" /nologo /c "%1/vp8/common/asm_com_offsets.c"
+cl /I "./" /I "%1" /nologo /c "%1/vp8/decoder/asm_dec_offsets.c"
+cl /I "./" /I "%1" /nologo /c "%1/vp8/encoder/asm_enc_offsets.c"
+obj_int_extract.exe rvds "asm_com_offsets.obj" > "asm_com_offsets.asm"
+obj_int_extract.exe rvds "asm_dec_offsets.obj" > "asm_dec_offsets.asm"
+obj_int_extract.exe rvds "asm_enc_offsets.obj" > "asm_enc_offsets.asm"
diff --git a/libs.mk b/libs.mk
index 37ce0b1..350b310 100644
--- a/libs.mk
+++ b/libs.mk
@@ -9,7 +9,13 @@
##
-ASM:=$(if $(filter yes,$(CONFIG_GCC)),.asm.s,.asm)
+# ARM assembly files are written in RVCT-style. We use some make magic to
+# filter those files to allow GCC compilation
+ifeq ($(ARCH_ARM),yes)
+ ASM:=$(if $(filter yes,$(CONFIG_GCC)),.asm.s,.asm)
+else
+ ASM:=.asm
+endif
CODEC_SRCS-yes += libs.mk
@@ -126,6 +132,23 @@
ifeq ($(CONFIG_EXTERNAL_BUILD),yes)
ifeq ($(CONFIG_MSVS),yes)
+obj_int_extract.vcproj: $(SRC_PATH_BARE)/build/make/obj_int_extract.c
+ @cp $(SRC_PATH_BARE)/build/x86-msvs/obj_int_extract.bat .
+ @echo " [CREATE] $@"
+ $(SRC_PATH_BARE)/build/make/gen_msvs_proj.sh \
+ --exe \
+ --target=$(TOOLCHAIN) \
+ --name=obj_int_extract \
+ --ver=$(CONFIG_VS_VERSION) \
+ --proj-guid=E1360C65-D375-4335-8057-7ED99CC3F9B2 \
+ $(if $(CONFIG_STATIC_MSVCRT),--static-crt) \
+ --out=$@ $^ \
+ -I. \
+ -I"$(SRC_PATH_BARE)" \
+
+PROJECTS-$(BUILD_LIBVPX) += obj_int_extract.vcproj
+PROJECTS-$(BUILD_LIBVPX) += obj_int_extract.bat
+
vpx.def: $(call enabled,CODEC_EXPORTS)
@echo " [CREATE] $@"
$(SRC_PATH_BARE)/build/make/gen_msvs_def.sh\
@@ -135,15 +158,16 @@
vpx.vcproj: $(CODEC_SRCS) vpx.def
@echo " [CREATE] $@"
- $(SRC_PATH_BARE)/build/make/gen_msvs_proj.sh\
- --lib\
- --target=$(TOOLCHAIN)\
+ $(SRC_PATH_BARE)/build/make/gen_msvs_proj.sh \
+ --lib \
+ --target=$(TOOLCHAIN) \
$(if $(CONFIG_STATIC_MSVCRT),--static-crt) \
- --name=vpx\
- --proj-guid=DCE19DAF-69AC-46DB-B14A-39F0FAA5DB74\
- --module-def=vpx.def\
- --ver=$(CONFIG_VS_VERSION)\
- --out=$@ $(CFLAGS) $^\
+ --name=vpx \
+ --proj-guid=DCE19DAF-69AC-46DB-B14A-39F0FAA5DB74 \
+ --module-def=vpx.def \
+ --ver=$(CONFIG_VS_VERSION) \
+ --out=$@ $(CFLAGS) $^ \
+ --src-path-bare="$(SRC_PATH_BARE)" \
PROJECTS-$(BUILD_LIBVPX) += vpx.vcproj
@@ -207,36 +231,38 @@
#
# Add assembler dependencies for configuration and offsets
#
-$(filter %.s.o,$(OBJS-yes)): $(BUILD_PFX)vpx_config.asm
-$(filter %.asm.o,$(OBJS-yes)): $(BUILD_PFX)vpx_config.asm
+$(filter %.s.o,$(OBJS-yes)): $(BUILD_PFX)vpx_config.asm
+$(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)vpx_config.asm
#
# Calculate platform- and compiler-specific offsets for hand coded assembly
#
-ifeq ($(ARCH_ARM), yes)
- asm_com_offsets.asm: obj_int_extract
- asm_com_offsets.asm: $(VP8_PREFIX)common/asm_com_offsets.c.o
+ifeq ($(CONFIG_EXTERNAL_BUILD),) # Visual Studio uses obj_int_extract.bat
+ ifeq ($(ARCH_ARM), yes)
+ asm_com_offsets.asm: obj_int_extract
+ asm_com_offsets.asm: $(VP8_PREFIX)common/asm_com_offsets.c.o
./obj_int_extract rvds $< $(ADS2GAS) > $@
- OBJS-yes += $(VP8_PREFIX)common/asm_com_offsets.c.o
- CLEAN-OBJS += asm_com_offsets.asm
- $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_com_offsets.asm
+ OBJS-yes += $(VP8_PREFIX)common/asm_com_offsets.c.o
+ CLEAN-OBJS += asm_com_offsets.asm
+ $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_com_offsets.asm
- ifeq ($(CONFIG_VP8_ENCODER), yes)
- asm_enc_offsets.asm: obj_int_extract
- asm_enc_offsets.asm: $(VP8_PREFIX)encoder/asm_enc_offsets.c.o
+ ifeq ($(CONFIG_VP8_ENCODER), yes)
+ asm_enc_offsets.asm: obj_int_extract
+ asm_enc_offsets.asm: $(VP8_PREFIX)encoder/asm_enc_offsets.c.o
./obj_int_extract rvds $< $(ADS2GAS) > $@
- OBJS-yes += $(VP8_PREFIX)encoder/asm_enc_offsets.c.o
- CLEAN-OBJS += asm_enc_offsets.asm
- $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_enc_offsets.asm
- endif
+ OBJS-yes += $(VP8_PREFIX)encoder/asm_enc_offsets.c.o
+ CLEAN-OBJS += asm_enc_offsets.asm
+ $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_enc_offsets.asm
+ endif
- ifeq ($(CONFIG_VP8_DECODER), yes)
- asm_dec_offsets.asm: obj_int_extract
- asm_dec_offsets.asm: $(VP8_PREFIX)decoder/asm_dec_offsets.c.o
+ ifeq ($(CONFIG_VP8_DECODER), yes)
+ asm_dec_offsets.asm: obj_int_extract
+ asm_dec_offsets.asm: $(VP8_PREFIX)decoder/asm_dec_offsets.c.o
./obj_int_extract rvds $< $(ADS2GAS) > $@
- OBJS-yes += $(VP8_PREFIX)decoder/asm_dec_offsets.c.o
- CLEAN-OBJS += asm_dec_offsets.asm
- $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_dec_offsets.asm
+ OBJS-yes += $(VP8_PREFIX)decoder/asm_dec_offsets.c.o
+ CLEAN-OBJS += asm_dec_offsets.asm
+ $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_dec_offsets.asm
+ endif
endif
endif
diff --git a/solution.mk b/solution.mk
index bef0088..782150f 100644
--- a/solution.mk
+++ b/solution.mk
@@ -13,8 +13,9 @@
@echo " [CREATE] $@"
$(SRC_PATH_BARE)/build/make/gen_msvs_sln.sh \
$(if $(filter %vpx.vcproj,$^),\
- $(foreach vcp,$(filter-out %vpx.vcproj,$^),\
+ $(foreach vcp,$(filter-out %vpx.vcproj %obj_int_extract.vcproj,$^),\
--dep=$(vcp:.vcproj=):vpx)) \
+ --dep=vpx:obj_int_extract \
--ver=$(CONFIG_VS_VERSION)\
--out=$@ $^
vpx.sln.mk: vpx.sln
diff --git a/vp8/encoder/ethreading.c b/vp8/encoder/ethreading.c
index 12d5f66..5c607a0 100644
--- a/vp8/encoder/ethreading.c
+++ b/vp8/encoder/ethreading.c
@@ -24,6 +24,35 @@
extern void vp8_build_block_offsets(MACROBLOCK *x);
extern void vp8_setup_block_ptrs(MACROBLOCK *x);
+#if CONFIG_MULTITHREAD
+
+extern void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm);
+
+static THREAD_FUNCTION loopfilter_thread(void *p_data)
+{
+ VP8_COMP *cpi = (VP8_COMP *)(((LPFTHREAD_DATA *)p_data)->ptr1);
+ VP8_COMMON *cm = &cpi->common;
+
+ while (1)
+ {
+ if (cpi->b_multi_threaded == 0)
+ break;
+
+ if (sem_wait(&cpi->h_event_start_lpf) == 0)
+ {
+ if (cpi->b_multi_threaded == FALSE) // we're shutting down
+ break;
+
+ loopfilter_frame(cpi, cm);
+
+ sem_post(&cpi->h_event_end_lpf);
+ }
+ }
+
+ return 0;
+}
+#endif
+
static
THREAD_FUNCTION thread_encoding_proc(void *p_data)
{
@@ -479,6 +508,15 @@
pthread_create(&cpi->h_encoding_thread[ithread], 0, thread_encoding_proc, ethd);
}
+ {
+ LPFTHREAD_DATA * lpfthd = &cpi->lpf_thread_data;
+
+ sem_init(&cpi->h_event_start_lpf, 0, 0);
+ sem_init(&cpi->h_event_end_lpf, 0, 0);
+
+ lpfthd->ptr1 = (void *)cpi;
+ pthread_create(&cpi->h_filter_thread, 0, loopfilter_thread, lpfthd);
+ }
}
}
@@ -500,9 +538,14 @@
sem_destroy(&cpi->h_event_start_encoding[i]);
}
+
+ sem_post(&cpi->h_event_start_lpf);
+ pthread_join(cpi->h_filter_thread, 0);
}
sem_destroy(&cpi->h_event_end_encoding);
+ sem_destroy(&cpi->h_event_end_lpf);
+ sem_destroy(&cpi->h_event_start_lpf);
//free thread related resources
vpx_free(cpi->h_event_start_encoding);
diff --git a/vp8/encoder/generic/csystemdependent.c b/vp8/encoder/generic/csystemdependent.c
index fc0580d..81108fe 100644
--- a/vp8/encoder/generic/csystemdependent.c
+++ b/vp8/encoder/generic/csystemdependent.c
@@ -103,6 +103,10 @@
// Pure C:
vp8_yv12_copy_partial_frame_ptr = vp8_yv12_copy_partial_frame;
+#if CONFIG_PSNR
+ cpi->rtcd.variance.ssimpf_8x8 = ssim_parms_8x8_c;
+ cpi->rtcd.variance.ssimpf = ssim_parms_c;
+#endif
#if ARCH_X86 || ARCH_X86_64
vp8_arch_x86_encoder_init(cpi);
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index 8c477ef..c028db1 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -86,9 +86,11 @@
YV12_BUFFER_CONFIG *source,
YV12_BUFFER_CONFIG *dest,
int lumamask,
- double *weight
+ double *weight,
+ const vp8_variance_rtcd_vtable_t *rtcd
);
+
extern double vp8_calc_ssimg
(
YV12_BUFFER_CONFIG *source,
@@ -3269,6 +3271,89 @@
return force_recode;
}
+void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
+{
+ if (cm->no_lpf)
+ {
+ cm->filter_level = 0;
+ }
+ else
+ {
+ struct vpx_usec_timer timer;
+
+ vp8_clear_system_state();
+
+ vpx_usec_timer_start(&timer);
+ if (cpi->sf.auto_filter == 0)
+ vp8cx_pick_filter_level_fast(cpi->Source, cpi);
+
+ else
+ vp8cx_pick_filter_level(cpi->Source, cpi);
+
+ vpx_usec_timer_mark(&timer);
+ cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
+ }
+
+#if CONFIG_MULTITHREAD
+ sem_post(&cpi->h_event_end_lpf); /* signal that we have set filter_level */
+#endif
+
+ if (cm->filter_level > 0)
+ {
+ vp8cx_set_alt_lf_level(cpi, cm->filter_level);
+ vp8_loop_filter_frame(cm, &cpi->mb.e_mbd, cm->filter_level);
+ cm->last_filter_type = cm->filter_type;
+ cm->last_sharpness_level = cm->sharpness_level;
+ }
+
+ vp8_yv12_extend_frame_borders_ptr(cm->frame_to_show);
+
+ {
+ YV12_BUFFER_CONFIG *lst_yv12 = &cm->yv12_fb[cm->lst_fb_idx];
+ YV12_BUFFER_CONFIG *new_yv12 = &cm->yv12_fb[cm->new_fb_idx];
+ YV12_BUFFER_CONFIG *gld_yv12 = &cm->yv12_fb[cm->gld_fb_idx];
+ YV12_BUFFER_CONFIG *alt_yv12 = &cm->yv12_fb[cm->alt_fb_idx];
+ // At this point the new frame has been encoded.
+ // If any buffer copy / swapping is signaled it should be done here.
+ if (cm->frame_type == KEY_FRAME)
+ {
+ vp8_yv12_copy_frame_ptr(cm->frame_to_show, gld_yv12);
+ vp8_yv12_copy_frame_ptr(cm->frame_to_show, alt_yv12);
+ }
+ else // For non key frames
+ {
+ // Code to copy between reference buffers
+ if (cm->copy_buffer_to_arf)
+ {
+ if (cm->copy_buffer_to_arf == 1)
+ {
+ if (cm->refresh_last_frame)
+ // We copy new_frame here because last and new buffers will already have been swapped if cm->refresh_last_frame is set.
+ vp8_yv12_copy_frame_ptr(new_yv12, alt_yv12);
+ else
+ vp8_yv12_copy_frame_ptr(lst_yv12, alt_yv12);
+ }
+ else if (cm->copy_buffer_to_arf == 2)
+ vp8_yv12_copy_frame_ptr(gld_yv12, alt_yv12);
+ }
+
+ if (cm->copy_buffer_to_gf)
+ {
+ if (cm->copy_buffer_to_gf == 1)
+ {
+ if (cm->refresh_last_frame)
+ // We copy new_frame here because last and new buffers will already have been swapped if cm->refresh_last_frame is set.
+ vp8_yv12_copy_frame_ptr(new_yv12, gld_yv12);
+ else
+ vp8_yv12_copy_frame_ptr(lst_yv12, gld_yv12);
+ }
+ else if (cm->copy_buffer_to_gf == 2)
+ vp8_yv12_copy_frame_ptr(alt_yv12, gld_yv12);
+ }
+ }
+ }
+}
+
static void encode_frame_to_data_rate
(
VP8_COMP *cpi,
@@ -3819,8 +3904,8 @@
vp8_setup_key_frame(cpi);
// transform / motion compensation build reconstruction frame
-
vp8_encode_frame(cpi);
+
cpi->projected_frame_size -= vp8_estimate_entropy_savings(cpi);
cpi->projected_frame_size = (cpi->projected_frame_size > 0) ? cpi->projected_frame_size : 0;
@@ -4169,92 +4254,43 @@
else
cm->frame_to_show = &cm->yv12_fb[cm->new_fb_idx];
- if (cm->no_lpf)
+
+#if CONFIG_MULTITHREAD
+ if (cpi->b_multi_threaded)
{
- cm->filter_level = 0;
+ sem_post(&cpi->h_event_start_lpf); /* start loopfilter in separate thread */
}
else
+#endif
{
- struct vpx_usec_timer timer;
-
- vpx_usec_timer_start(&timer);
-
- if (cpi->sf.auto_filter == 0)
- vp8cx_pick_filter_level_fast(cpi->Source, cpi);
- else
- vp8cx_pick_filter_level(cpi->Source, cpi);
-
- vpx_usec_timer_mark(&timer);
-
- cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
+ loopfilter_frame(cpi, cm);
}
- if (cm->filter_level > 0)
- {
- vp8cx_set_alt_lf_level(cpi, cm->filter_level);
- vp8_loop_filter_frame(cm, &cpi->mb.e_mbd, cm->filter_level);
- cm->last_filter_type = cm->filter_type;
- cm->last_sharpness_level = cm->sharpness_level;
- }
-
- /* Move storing frame_type out of the above loop since it is also
- * needed in motion search besides loopfilter */
- cm->last_frame_type = cm->frame_type;
-
- vp8_yv12_extend_frame_borders_ptr(cm->frame_to_show);
-
if (cpi->oxcf.error_resilient_mode == 1)
{
cm->refresh_entropy_probs = 0;
}
+#if CONFIG_MULTITHREAD
+ /* wait that filter_level is picked so that we can continue with stream packing */
+ if (cpi->b_multi_threaded)
+ sem_wait(&cpi->h_event_end_lpf);
+#endif
+
// build the bitstream
vp8_pack_bitstream(cpi, dest, size);
+#if CONFIG_MULTITHREAD
+ /* wait for loopfilter thread done */
+ if (cpi->b_multi_threaded)
{
- YV12_BUFFER_CONFIG *lst_yv12 = &cm->yv12_fb[cm->lst_fb_idx];
- YV12_BUFFER_CONFIG *new_yv12 = &cm->yv12_fb[cm->new_fb_idx];
- YV12_BUFFER_CONFIG *gld_yv12 = &cm->yv12_fb[cm->gld_fb_idx];
- YV12_BUFFER_CONFIG *alt_yv12 = &cm->yv12_fb[cm->alt_fb_idx];
- // At this point the new frame has been encoded coded.
- // If any buffer copy / swaping is signalled it should be done here.
- if (cm->frame_type == KEY_FRAME)
- {
- vp8_yv12_copy_frame_ptr(cm->frame_to_show, gld_yv12);
- vp8_yv12_copy_frame_ptr(cm->frame_to_show, alt_yv12);
- }
- else // For non key frames
- {
- // Code to copy between reference buffers
- if (cm->copy_buffer_to_arf)
- {
- if (cm->copy_buffer_to_arf == 1)
- {
- if (cm->refresh_last_frame)
- // We copy new_frame here because last and new buffers will already have been swapped if cm->refresh_last_frame is set.
- vp8_yv12_copy_frame_ptr(new_yv12, alt_yv12);
- else
- vp8_yv12_copy_frame_ptr(lst_yv12, alt_yv12);
- }
- else if (cm->copy_buffer_to_arf == 2)
- vp8_yv12_copy_frame_ptr(gld_yv12, alt_yv12);
- }
-
- if (cm->copy_buffer_to_gf)
- {
- if (cm->copy_buffer_to_gf == 1)
- {
- if (cm->refresh_last_frame)
- // We copy new_frame here because last and new buffers will already have been swapped if cm->refresh_last_frame is set.
- vp8_yv12_copy_frame_ptr(new_yv12, gld_yv12);
- else
- vp8_yv12_copy_frame_ptr(lst_yv12, gld_yv12);
- }
- else if (cm->copy_buffer_to_gf == 2)
- vp8_yv12_copy_frame_ptr(alt_yv12, gld_yv12);
- }
- }
+ sem_wait(&cpi->h_event_end_lpf);
}
+#endif
+
+ /* Move storing frame_type out of the above loop since it is also
+ * needed in motion search besides loopfilter */
+ cm->last_frame_type = cm->frame_type;
// Update rate control heuristics
cpi->total_byte_count += (*size);
@@ -5089,7 +5125,9 @@
cpi->time_compress_data += vpx_usec_timer_elapsed(&cmptimer);
if (cpi->b_calculate_psnr && cpi->pass != 1 && cm->show_frame)
+ {
generate_psnr_packet(cpi);
+ }
#if CONFIG_PSNR
@@ -5105,12 +5143,35 @@
if (cpi->b_calculate_psnr)
{
double y, u, v;
- double sq_error;
- double frame_psnr = vp8_calc_psnr(cpi->Source, cm->frame_to_show, &y, &u, &v, &sq_error);
+ double ye,ue,ve;
+ double frame_psnr;
+ YV12_BUFFER_CONFIG *orig = cpi->Source;
+ YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
+ YV12_BUFFER_CONFIG *pp = &cm->post_proc_buffer;
+ int y_samples = orig->y_height * orig->y_width ;
+ int uv_samples = orig->uv_height * orig->uv_width ;
+ int t_samples = y_samples + 2 * uv_samples;
+ long long sq_error;
- cpi->total_y += y;
- cpi->total_u += u;
- cpi->total_v += v;
+ ye = calc_plane_error(orig->y_buffer, orig->y_stride,
+ recon->y_buffer, recon->y_stride, orig->y_width, orig->y_height,
+ IF_RTCD(&cpi->rtcd.variance));
+
+ ue = calc_plane_error(orig->u_buffer, orig->uv_stride,
+ recon->u_buffer, recon->uv_stride, orig->uv_width, orig->uv_height,
+ IF_RTCD(&cpi->rtcd.variance));
+
+ ve = calc_plane_error(orig->v_buffer, orig->uv_stride,
+ recon->v_buffer, recon->uv_stride, orig->uv_width, orig->uv_height,
+ IF_RTCD(&cpi->rtcd.variance));
+
+ sq_error = ye + ue + ve;
+
+ frame_psnr = vp8_mse2psnr(t_samples, 255.0, sq_error);
+
+ cpi->total_y += vp8_mse2psnr(y_samples, 255.0, ye);
+ cpi->total_u += vp8_mse2psnr(uv_samples, 255.0, ue);
+ cpi->total_v += vp8_mse2psnr(uv_samples, 255.0, ve);
cpi->total_sq_error += sq_error;
cpi->total += frame_psnr;
{
@@ -5119,18 +5180,36 @@
vp8_deblock(cm->frame_to_show, &cm->post_proc_buffer, cm->filter_level * 10 / 6, 1, 0, IF_RTCD(&cm->rtcd.postproc));
vp8_clear_system_state();
- frame_psnr2 = vp8_calc_psnr(cpi->Source, &cm->post_proc_buffer, &y2, &u2, &v2, &sq_error);
- frame_ssim2 = vp8_calc_ssim(cpi->Source, &cm->post_proc_buffer, 1, &weight);
+
+ ye = calc_plane_error(orig->y_buffer, orig->y_stride,
+ pp->y_buffer, pp->y_stride, orig->y_width, orig->y_height,
+ IF_RTCD(&cpi->rtcd.variance));
+
+ ue = calc_plane_error(orig->u_buffer, orig->uv_stride,
+ pp->u_buffer, pp->uv_stride, orig->uv_width, orig->uv_height,
+ IF_RTCD(&cpi->rtcd.variance));
+
+ ve = calc_plane_error(orig->v_buffer, orig->uv_stride,
+ pp->v_buffer, pp->uv_stride, orig->uv_width, orig->uv_height,
+ IF_RTCD(&cpi->rtcd.variance));
+
+ sq_error = ye + ue + ve;
+
+ frame_psnr2 = vp8_mse2psnr(t_samples, 255.0, sq_error);
+
+ cpi->totalp_y += vp8_mse2psnr(y_samples, 255.0, ye);
+ cpi->totalp_u += vp8_mse2psnr(uv_samples, 255.0, ue);
+ cpi->totalp_v += vp8_mse2psnr(uv_samples, 255.0, ve);
+ cpi->total_sq_error2 += sq_error;
+ cpi->totalp += frame_psnr2;
+
+ frame_ssim2 = vp8_calc_ssim(cpi->Source,
+ &cm->post_proc_buffer, 1, &weight,
+ IF_RTCD(&cpi->rtcd.variance));
cpi->summed_quality += frame_ssim2 * weight;
cpi->summed_weights += weight;
- cpi->totalp_y += y2;
- cpi->totalp_u += u2;
- cpi->totalp_v += v2;
- cpi->totalp += frame_psnr2;
- cpi->total_sq_error2 += sq_error;
-
}
}
diff --git a/vp8/encoder/onyx_int.h b/vp8/encoder/onyx_int.h
index 0547d57..0e53f68 100644
--- a/vp8/encoder/onyx_int.h
+++ b/vp8/encoder/onyx_int.h
@@ -599,12 +599,17 @@
int encoding_thread_count;
pthread_t *h_encoding_thread;
+ pthread_t h_filter_thread;
+
MB_ROW_COMP *mb_row_ei;
ENCODETHREAD_DATA *en_thread_data;
+ LPFTHREAD_DATA lpf_thread_data;
//events
sem_t *h_event_start_encoding;
sem_t h_event_end_encoding;
+ sem_t h_event_start_lpf;
+ sem_t h_event_end_lpf;
#endif
TOKENLIST *tplist;
diff --git a/vp8/encoder/ssim.c b/vp8/encoder/ssim.c
index 4ebcba1..64d67c6 100644
--- a/vp8/encoder/ssim.c
+++ b/vp8/encoder/ssim.c
@@ -11,298 +11,13 @@
#include "vpx_scale/yv12config.h"
#include "math.h"
+#include "onyx_int.h"
-#define C1 (float)(64 * 64 * 0.01*255*0.01*255)
-#define C2 (float)(64 * 64 * 0.03*255*0.03*255)
-
-static int width_y;
-static int height_y;
-static int height_uv;
-static int width_uv;
-static int stride_uv;
-static int stride;
-static int lumimask;
-static int luminance;
-static double plane_summed_weights = 0;
-
-static short img12_sum_block[8*4096*4096*2] ;
-
-static short img1_sum[8*4096*2];
-static short img2_sum[8*4096*2];
-static int img1_sq_sum[8*4096*2];
-static int img2_sq_sum[8*4096*2];
-static int img12_mul_sum[8*4096*2];
-
-
-double vp8_similarity
-(
- int mu_x,
- int mu_y,
- int pre_mu_x2,
- int pre_mu_y2,
- int pre_mu_xy2
-)
-{
- int mu_x2, mu_y2, mu_xy, theta_x2, theta_y2, theta_xy;
-
- mu_x2 = mu_x * mu_x;
- mu_y2 = mu_y * mu_y;
- mu_xy = mu_x * mu_y;
-
- theta_x2 = 64 * pre_mu_x2 - mu_x2;
- theta_y2 = 64 * pre_mu_y2 - mu_y2;
- theta_xy = 64 * pre_mu_xy2 - mu_xy;
-
- return (2 * mu_xy + C1) * (2 * theta_xy + C2) / ((mu_x2 + mu_y2 + C1) * (theta_x2 + theta_y2 + C2));
-}
-
-double vp8_ssim
-(
- const unsigned char *img1,
- const unsigned char *img2,
- int stride_img1,
- int stride_img2,
- int width,
- int height
-)
-{
- int x, y, x2, y2, img1_block, img2_block, img1_sq_block, img2_sq_block, img12_mul_block, temp;
-
- double plane_quality, weight, mean;
-
- short *img1_sum_ptr1, *img1_sum_ptr2;
- short *img2_sum_ptr1, *img2_sum_ptr2;
- int *img1_sq_sum_ptr1, *img1_sq_sum_ptr2;
- int *img2_sq_sum_ptr1, *img2_sq_sum_ptr2;
- int *img12_mul_sum_ptr1, *img12_mul_sum_ptr2;
-
- plane_quality = 0;
-
- if (lumimask)
- plane_summed_weights = 0.0f;
- else
- plane_summed_weights = (height - 7) * (width - 7);
-
- //some prologue for the main loop
- temp = 8 * width;
-
- img1_sum_ptr1 = img1_sum + temp;
- img2_sum_ptr1 = img2_sum + temp;
- img1_sq_sum_ptr1 = img1_sq_sum + temp;
- img2_sq_sum_ptr1 = img2_sq_sum + temp;
- img12_mul_sum_ptr1 = img12_mul_sum + temp;
-
- for (x = 0; x < width; x++)
- {
- img1_sum[x] = img1[x];
- img2_sum[x] = img2[x];
- img1_sq_sum[x] = img1[x] * img1[x];
- img2_sq_sum[x] = img2[x] * img2[x];
- img12_mul_sum[x] = img1[x] * img2[x];
-
- img1_sum_ptr1[x] = 0;
- img2_sum_ptr1[x] = 0;
- img1_sq_sum_ptr1[x] = 0;
- img2_sq_sum_ptr1[x] = 0;
- img12_mul_sum_ptr1[x] = 0;
- }
-
- //the main loop
- for (y = 1; y < height; y++)
- {
- img1 += stride_img1;
- img2 += stride_img2;
-
- temp = (y - 1) % 9 * width;
-
- img1_sum_ptr1 = img1_sum + temp;
- img2_sum_ptr1 = img2_sum + temp;
- img1_sq_sum_ptr1 = img1_sq_sum + temp;
- img2_sq_sum_ptr1 = img2_sq_sum + temp;
- img12_mul_sum_ptr1 = img12_mul_sum + temp;
-
- temp = y % 9 * width;
-
- img1_sum_ptr2 = img1_sum + temp;
- img2_sum_ptr2 = img2_sum + temp;
- img1_sq_sum_ptr2 = img1_sq_sum + temp;
- img2_sq_sum_ptr2 = img2_sq_sum + temp;
- img12_mul_sum_ptr2 = img12_mul_sum + temp;
-
- for (x = 0; x < width; x++)
- {
- img1_sum_ptr2[x] = img1_sum_ptr1[x] + img1[x];
- img2_sum_ptr2[x] = img2_sum_ptr1[x] + img2[x];
- img1_sq_sum_ptr2[x] = img1_sq_sum_ptr1[x] + img1[x] * img1[x];
- img2_sq_sum_ptr2[x] = img2_sq_sum_ptr1[x] + img2[x] * img2[x];
- img12_mul_sum_ptr2[x] = img12_mul_sum_ptr1[x] + img1[x] * img2[x];
- }
-
- if (y > 6)
- {
- //calculate the sum of the last 8 lines by subtracting the total sum of 8 lines back from the present sum
- temp = (y + 1) % 9 * width;
-
- img1_sum_ptr1 = img1_sum + temp;
- img2_sum_ptr1 = img2_sum + temp;
- img1_sq_sum_ptr1 = img1_sq_sum + temp;
- img2_sq_sum_ptr1 = img2_sq_sum + temp;
- img12_mul_sum_ptr1 = img12_mul_sum + temp;
-
- for (x = 0; x < width; x++)
- {
- img1_sum_ptr1[x] = img1_sum_ptr2[x] - img1_sum_ptr1[x];
- img2_sum_ptr1[x] = img2_sum_ptr2[x] - img2_sum_ptr1[x];
- img1_sq_sum_ptr1[x] = img1_sq_sum_ptr2[x] - img1_sq_sum_ptr1[x];
- img2_sq_sum_ptr1[x] = img2_sq_sum_ptr2[x] - img2_sq_sum_ptr1[x];
- img12_mul_sum_ptr1[x] = img12_mul_sum_ptr2[x] - img12_mul_sum_ptr1[x];
- }
-
- //here we calculate the sum over the 8x8 block of pixels
- //this is done by sliding a window across the column sums for the last 8 lines
- //each time adding the new column sum, and subtracting the one which fell out of the window
- img1_block = 0;
- img2_block = 0;
- img1_sq_block = 0;
- img2_sq_block = 0;
- img12_mul_block = 0;
-
- //prologue, and calculation of simularity measure from the first 8 column sums
- for (x = 0; x < 8; x++)
- {
- img1_block += img1_sum_ptr1[x];
- img2_block += img2_sum_ptr1[x];
- img1_sq_block += img1_sq_sum_ptr1[x];
- img2_sq_block += img2_sq_sum_ptr1[x];
- img12_mul_block += img12_mul_sum_ptr1[x];
- }
-
- if (lumimask)
- {
- y2 = y - 7;
- x2 = 0;
-
- if (luminance)
- {
- mean = (img2_block + img1_block) / 128.0f;
-
- if (!(y2 % 2 || x2 % 2))
- *(img12_sum_block + y2 / 2 * width_uv + x2 / 2) = img2_block + img1_block;
- }
- else
- {
- mean = *(img12_sum_block + y2 * width_uv + x2);
- mean += *(img12_sum_block + y2 * width_uv + x2 + 4);
- mean += *(img12_sum_block + (y2 + 4) * width_uv + x2);
- mean += *(img12_sum_block + (y2 + 4) * width_uv + x2 + 4);
-
- mean /= 512.0f;
- }
-
- weight = mean < 40 ? 0.0f :
- (mean < 50 ? (mean - 40.0f) / 10.0f : 1.0f);
- plane_summed_weights += weight;
-
- plane_quality += weight * vp8_similarity(img1_block, img2_block, img1_sq_block, img2_sq_block, img12_mul_block);
- }
- else
- plane_quality += vp8_similarity(img1_block, img2_block, img1_sq_block, img2_sq_block, img12_mul_block);
-
- //and for the rest
- for (x = 8; x < width; x++)
- {
- img1_block = img1_block + img1_sum_ptr1[x] - img1_sum_ptr1[x - 8];
- img2_block = img2_block + img2_sum_ptr1[x] - img2_sum_ptr1[x - 8];
- img1_sq_block = img1_sq_block + img1_sq_sum_ptr1[x] - img1_sq_sum_ptr1[x - 8];
- img2_sq_block = img2_sq_block + img2_sq_sum_ptr1[x] - img2_sq_sum_ptr1[x - 8];
- img12_mul_block = img12_mul_block + img12_mul_sum_ptr1[x] - img12_mul_sum_ptr1[x - 8];
-
- if (lumimask)
- {
- y2 = y - 7;
- x2 = x - 7;
-
- if (luminance)
- {
- mean = (img2_block + img1_block) / 128.0f;
-
- if (!(y2 % 2 || x2 % 2))
- *(img12_sum_block + y2 / 2 * width_uv + x2 / 2) = img2_block + img1_block;
- }
- else
- {
- mean = *(img12_sum_block + y2 * width_uv + x2);
- mean += *(img12_sum_block + y2 * width_uv + x2 + 4);
- mean += *(img12_sum_block + (y2 + 4) * width_uv + x2);
- mean += *(img12_sum_block + (y2 + 4) * width_uv + x2 + 4);
-
- mean /= 512.0f;
- }
-
- weight = mean < 40 ? 0.0f :
- (mean < 50 ? (mean - 40.0f) / 10.0f : 1.0f);
- plane_summed_weights += weight;
-
- plane_quality += weight * vp8_similarity(img1_block, img2_block, img1_sq_block, img2_sq_block, img12_mul_block);
- }
- else
- plane_quality += vp8_similarity(img1_block, img2_block, img1_sq_block, img2_sq_block, img12_mul_block);
- }
- }
- }
-
- if (plane_summed_weights == 0)
- return 1.0f;
- else
- return plane_quality / plane_summed_weights;
-}
-
-double vp8_calc_ssim
-(
- YV12_BUFFER_CONFIG *source,
- YV12_BUFFER_CONFIG *dest,
- int lumamask,
- double *weight
-)
-{
- double a, b, c;
- double frame_weight;
- double ssimv;
-
- width_y = source->y_width;
- height_y = source->y_height;
- height_uv = source->uv_height;
- width_uv = source->uv_width;
- stride_uv = dest->uv_stride;
- stride = dest->y_stride;
-
- lumimask = lumamask;
-
- luminance = 1;
- a = vp8_ssim(source->y_buffer, dest->y_buffer,
- source->y_stride, dest->y_stride, source->y_width, source->y_height);
- luminance = 0;
-
- frame_weight = plane_summed_weights / ((width_y - 7) * (height_y - 7));
-
- if (frame_weight == 0)
- a = b = c = 1.0f;
- else
- {
- b = vp8_ssim(source->u_buffer, dest->u_buffer,
- source->uv_stride, dest->uv_stride, source->uv_width, source->uv_height);
-
- c = vp8_ssim(source->v_buffer, dest->v_buffer,
- source->uv_stride, dest->uv_stride, source->uv_width, source->uv_height);
- }
-
- ssimv = a * .8 + .1 * (b + c);
-
- *weight = frame_weight;
-
- return ssimv;
-}
-
+#if CONFIG_RUNTIME_CPU_DETECT
+#define IF_RTCD(x) (x)
+#else
+#define IF_RTCD(x) NULL
+#endif
// Google version of SSIM
// SSIM
#define KERNEL 3
@@ -520,3 +235,174 @@
*ssim_v /= uvsize;
return ssim_all;
}
+
+
+void ssim_parms_c
+(
+ unsigned char *s,
+ int sp,
+ unsigned char *r,
+ int rp,
+ unsigned long *sum_s,
+ unsigned long *sum_r,
+ unsigned long *sum_sq_s,
+ unsigned long *sum_sq_r,
+ unsigned long *sum_sxr
+)
+{
+ int i,j;
+ for(i=0;i<16;i++,s+=sp,r+=rp)
+ {
+ for(j=0;j<16;j++)
+ {
+ *sum_s += s[j];
+ *sum_r += r[j];
+ *sum_sq_s += s[j] * s[j];
+ *sum_sq_r += r[j] * r[j];
+ *sum_sxr += s[j] * r[j];
+ }
+ }
+}
+void ssim_parms_8x8_c
+(
+ unsigned char *s,
+ int sp,
+ unsigned char *r,
+ int rp,
+ unsigned long *sum_s,
+ unsigned long *sum_r,
+ unsigned long *sum_sq_s,
+ unsigned long *sum_sq_r,
+ unsigned long *sum_sxr
+)
+{
+ int i,j;
+ for(i=0;i<8;i++,s+=sp,r+=rp)
+ {
+ for(j=0;j<8;j++)
+ {
+ *sum_s += s[j];
+ *sum_r += r[j];
+ *sum_sq_s += s[j] * s[j];
+ *sum_sq_r += r[j] * r[j];
+ *sum_sxr += s[j] * r[j];
+ }
+ }
+}
+
+const static long long c1 = 426148; // (256^2*(.01*255)^2
+const static long long c2 = 3835331; //(256^2*(.03*255)^2
+
+static double similarity
+(
+ unsigned long sum_s,
+ unsigned long sum_r,
+ unsigned long sum_sq_s,
+ unsigned long sum_sq_r,
+ unsigned long sum_sxr,
+ int count
+)
+{
+ long long ssim_n = (2*sum_s*sum_r+ c1)*(2*count*sum_sxr-2*sum_s*sum_r+c2);
+
+ long long ssim_d = (sum_s*sum_s +sum_r*sum_r+c1)*
+ (count*sum_sq_s-sum_s*sum_s + count*sum_sq_r-sum_r*sum_r +c2) ;
+
+ return ssim_n * 1.0 / ssim_d;
+}
+
+static double ssim_16x16(unsigned char *s,int sp, unsigned char *r,int rp,
+ const vp8_variance_rtcd_vtable_t *rtcd)
+{
+ unsigned long sum_s=0,sum_r=0,sum_sq_s=0,sum_sq_r=0,sum_sxr=0;
+ rtcd->ssimpf(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
+ return similarity(sum_s, sum_r, sum_sq_s, sum_sq_r, sum_sxr, 256);
+}
+static double ssim_8x8(unsigned char *s,int sp, unsigned char *r,int rp,
+ const vp8_variance_rtcd_vtable_t *rtcd)
+{
+ unsigned long sum_s=0,sum_r=0,sum_sq_s=0,sum_sq_r=0,sum_sxr=0;
+ rtcd->ssimpf_8x8(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
+ return similarity(sum_s, sum_r, sum_sq_s, sum_sq_r, sum_sxr, 64);
+}
+
+// TODO: (jbb) tried to scale this function such that we may be able to use it
+// for distortion metric in mode selection code ( provided we do a reconstruction)
+long dssim(unsigned char *s,int sp, unsigned char *r,int rp,
+ const vp8_variance_rtcd_vtable_t *rtcd)
+{
+ unsigned long sum_s=0,sum_r=0,sum_sq_s=0,sum_sq_r=0,sum_sxr=0;
+ double ssim3;
+ long long ssim_n;
+ long long ssim_d;
+
+ rtcd->ssimpf(s, sp, r, rp, &sum_s, &sum_r, &sum_sq_s, &sum_sq_r, &sum_sxr);
+ ssim_n = (2*sum_s*sum_r+ c1)*(2*256*sum_sxr-2*sum_s*sum_r+c2);
+
+ ssim_d = (sum_s*sum_s +sum_r*sum_r+c1)*
+ (256*sum_sq_s-sum_s*sum_s + 256*sum_sq_r-sum_r*sum_r +c2) ;
+
+ ssim3 = 256 * (ssim_d-ssim_n) / ssim_d;
+ return (long)( 256*ssim3 * ssim3 );
+}
+// TODO: (jbb) this 8x8 window might be too big + we may want to pick pixels
+// such that the window regions overlap block boundaries to penalize blocking
+// artifacts.
+
+double vp8_ssim2
+(
+ unsigned char *img1,
+ unsigned char *img2,
+ int stride_img1,
+ int stride_img2,
+ int width,
+ int height,
+ const vp8_variance_rtcd_vtable_t *rtcd
+)
+{
+ int i,j;
+
+ double ssim_total=0;
+
+ // we can sample points as frequently as we like start with 1 per 8x8
+ for(i=0; i < height; i+=8, img1 += stride_img1*8, img2 += stride_img2*8)
+ {
+ for(j=0; j < width; j+=8 )
+ {
+ ssim_total += ssim_8x8(img1, stride_img1, img2, stride_img2, rtcd);
+ }
+ }
+ ssim_total /= (width/8 * height /8);
+ return ssim_total;
+
+}
+double vp8_calc_ssim
+(
+ YV12_BUFFER_CONFIG *source,
+ YV12_BUFFER_CONFIG *dest,
+ int lumamask,
+ double *weight,
+ const vp8_variance_rtcd_vtable_t *rtcd
+)
+{
+ double a, b, c;
+ double ssimv;
+
+ a = vp8_ssim2(source->y_buffer, dest->y_buffer,
+ source->y_stride, dest->y_stride, source->y_width,
+ source->y_height, rtcd);
+
+ b = vp8_ssim2(source->u_buffer, dest->u_buffer,
+ source->uv_stride, dest->uv_stride, source->uv_width,
+ source->uv_height, rtcd);
+
+ c = vp8_ssim2(source->v_buffer, dest->v_buffer,
+ source->uv_stride, dest->uv_stride, source->uv_width,
+ source->uv_height, rtcd);
+
+ ssimv = a * .8 + .1 * (b + c);
+
+ *weight = 1;
+
+ return ssimv;
+}
diff --git a/vp8/encoder/variance.h b/vp8/encoder/variance.h
index 5befd3b..bf17ea8 100644
--- a/vp8/encoder/variance.h
+++ b/vp8/encoder/variance.h
@@ -85,6 +85,19 @@
unsigned int *sse \
);
+#define prototype_ssimpf(sym) \
+ void (sym) \
+ ( \
+ unsigned char *s, \
+ int sp, \
+ unsigned char *r, \
+ int rp, \
+ unsigned long *sum_s, \
+ unsigned long *sum_r, \
+ unsigned long *sum_sq_s, \
+ unsigned long *sum_sq_r, \
+ unsigned long *sum_sxr \
+ );
#define prototype_getmbss(sym) unsigned int (sym)(const short *)
@@ -306,6 +319,15 @@
#endif
extern prototype_sad(vp8_variance_get4x4sse_cs);
+#ifndef vp8_ssimpf
+#define vp8_ssimpf ssim_parms_c
+#endif
+extern prototype_ssimpf(vp8_ssimpf)
+
+#ifndef vp8_ssimpf_8x8
+#define vp8_ssimpf_8x8 ssim_parms_8x8_c
+#endif
+extern prototype_ssimpf(vp8_ssimpf_8x8)
typedef prototype_sad(*vp8_sad_fn_t);
typedef prototype_sad_multi_same_address(*vp8_sad_multi_fn_t);
@@ -315,6 +337,10 @@
typedef prototype_variance2(*vp8_variance2_fn_t);
typedef prototype_subpixvariance(*vp8_subpixvariance_fn_t);
typedef prototype_getmbss(*vp8_getmbss_fn_t);
+
+typedef prototype_ssimpf(*vp8_ssimpf_fn_t)
+
+
typedef struct
{
vp8_sad_fn_t sad4x4;
@@ -365,6 +391,11 @@
vp8_sad_multi_d_fn_t sad8x8x4d;
vp8_sad_multi_d_fn_t sad4x4x4d;
+#if CONFIG_PSNR
+ vp8_ssimpf_fn_t ssimpf_8x8;
+ vp8_ssimpf_fn_t ssimpf;
+#endif
+
} vp8_variance_rtcd_vtable_t;
typedef struct
@@ -378,6 +409,7 @@
vp8_sad_multi_fn_t sdx3f;
vp8_sad_multi1_fn_t sdx8f;
vp8_sad_multi_d_fn_t sdx4df;
+
} vp8_variance_fn_ptr_t;
#if CONFIG_RUNTIME_CPU_DETECT
diff --git a/vp8/encoder/x86/ssim_opt.asm b/vp8/encoder/x86/ssim_opt.asm
new file mode 100644
index 0000000..c267cdb
--- /dev/null
+++ b/vp8/encoder/x86/ssim_opt.asm
@@ -0,0 +1,215 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+; tabulate_ssim - sums sum_s,sum_r,sum_sq_s,sum_sq_r, sum_sxr
+%macro TABULATE_SSIM 0
+ paddusw xmm15, xmm3 ; sum_s
+ paddusw xmm14, xmm4 ; sum_r
+ movdqa xmm1, xmm3
+ pmaddwd xmm1, xmm1
+ paddq xmm13, xmm1 ; sum_sq_s
+ movdqa xmm2, xmm4
+ pmaddwd xmm2, xmm2
+ paddq xmm12, xmm2 ; sum_sq_r
+ pmaddwd xmm3, xmm4
+ paddq xmm11, xmm3 ; sum_sxr
+%endmacro
+
+; Sum across the register %1 starting with q words
+%macro SUM_ACROSS_Q 1
+ movdqa xmm2,%1
+ punpckldq %1,xmm0
+ punpckhdq xmm2,xmm0
+ paddq %1,xmm2
+ movdqa xmm2,%1
+ punpcklqdq %1,xmm0
+ punpckhqdq xmm2,xmm0
+ paddq %1,xmm2
+%endmacro
+
+; Sum across the register %1 starting with q words
+%macro SUM_ACROSS_W 1
+ movdqa xmm1, %1
+ punpcklwd %1,xmm0
+ punpckhwd xmm1,xmm0
+ paddd %1, xmm1
+ SUM_ACROSS_Q %1
+%endmacro
+;void ssim_parms_sse3(
+; unsigned char *s,
+; int sp,
+; unsigned char *r,
+; int rp
+; unsigned long *sum_s,
+; unsigned long *sum_r,
+; unsigned long *sum_sq_s,
+; unsigned long *sum_sq_r,
+; unsigned long *sum_sxr);
+;
+; TODO: Use parm passing through structure, probably don't need the pxors
+; ( calling app will initialize to 0 ) could easily fit everything in sse2
+; without too much hastle, and can probably do better estimates with psadw
+; or pavgb At this point this is just meant to be first pass for calculating
+; all the parms needed for 16x16 ssim so we can play with dssim as distortion
+; in mode selection code.
+global sym(vp8_ssim_parms_16x16_sse3)
+sym(vp8_ssim_parms_16x16_sse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 9
+ push rsi
+ push rdi
+ ; end prolog
+
+ mov rsi, arg(0) ;s
+ mov rcx, arg(1) ;sp
+ mov rdi, arg(2) ;r
+ mov rax, arg(3) ;rp
+
+ pxor xmm0, xmm0
+ pxor xmm15,xmm15 ;sum_s
+ pxor xmm14,xmm14 ;sum_r
+ pxor xmm13,xmm13 ;sum_sq_s
+ pxor xmm12,xmm12 ;sum_sq_r
+ pxor xmm11,xmm11 ;sum_sxr
+
+ mov rdx, 16 ;row counter
+NextRow:
+
+ ;grab source and reference pixels
+ movdqu xmm5, [rsi]
+ movdqu xmm6, [rdi]
+ movdqa xmm3, xmm5
+ movdqa xmm4, xmm6
+ punpckhbw xmm3, xmm0 ; high_s
+ punpckhbw xmm4, xmm0 ; high_r
+
+ TABULATE_SSIM
+
+ movdqa xmm3, xmm5
+ movdqa xmm4, xmm6
+ punpcklbw xmm3, xmm0 ; low_s
+ punpcklbw xmm4, xmm0 ; low_r
+
+ TABULATE_SSIM
+
+ add rsi, rcx ; next s row
+ add rdi, rax ; next r row
+
+ dec rdx ; counter
+ jnz NextRow
+
+ SUM_ACROSS_W xmm15
+ SUM_ACROSS_W xmm14
+ SUM_ACROSS_Q xmm13
+ SUM_ACROSS_Q xmm12
+ SUM_ACROSS_Q xmm11
+
+ mov rdi,arg(4)
+ movq [rdi], xmm15;
+ mov rdi,arg(5)
+ movq [rdi], xmm14;
+ mov rdi,arg(6)
+ movq [rdi], xmm13;
+ mov rdi,arg(7)
+ movq [rdi], xmm12;
+ mov rdi,arg(8)
+ movq [rdi], xmm11;
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void ssim_parms_sse3(
+; unsigned char *s,
+; int sp,
+; unsigned char *r,
+; int rp
+; unsigned long *sum_s,
+; unsigned long *sum_r,
+; unsigned long *sum_sq_s,
+; unsigned long *sum_sq_r,
+; unsigned long *sum_sxr);
+;
+; TODO: Use parm passing through structure, probably don't need the pxors
+; ( calling app will initialize to 0 ) could easily fit everything in sse2
+; without too much hastle, and can probably do better estimates with psadw
+; or pavgb At this point this is just meant to be first pass for calculating
+; all the parms needed for 16x16 ssim so we can play with dssim as distortion
+; in mode selection code.
+global sym(vp8_ssim_parms_8x8_sse3)
+sym(vp8_ssim_parms_8x8_sse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 9
+ push rsi
+ push rdi
+ ; end prolog
+
+ mov rsi, arg(0) ;s
+ mov rcx, arg(1) ;sp
+ mov rdi, arg(2) ;r
+ mov rax, arg(3) ;rp
+
+ pxor xmm0, xmm0
+ pxor xmm15,xmm15 ;sum_s
+ pxor xmm14,xmm14 ;sum_r
+ pxor xmm13,xmm13 ;sum_sq_s
+ pxor xmm12,xmm12 ;sum_sq_r
+ pxor xmm11,xmm11 ;sum_sxr
+
+ mov rdx, 8 ;row counter
+NextRow2:
+
+ ;grab source and reference pixels
+ movq xmm5, [rsi]
+ movq xmm6, [rdi]
+
+ movdqa xmm3, xmm5
+ movdqa xmm4, xmm6
+ punpcklbw xmm3, xmm0 ; low_s
+ punpcklbw xmm4, xmm0 ; low_r
+
+ TABULATE_SSIM
+
+ add rsi, rcx ; next s row
+ add rdi, rax ; next r row
+
+ dec rdx ; counter
+ jnz NextRow2
+
+ SUM_ACROSS_W xmm15
+ SUM_ACROSS_W xmm14
+ SUM_ACROSS_Q xmm13
+ SUM_ACROSS_Q xmm12
+ SUM_ACROSS_Q xmm11
+
+ mov rdi,arg(4)
+ movq [rdi], xmm15;
+ mov rdi,arg(5)
+ movq [rdi], xmm14;
+ mov rdi,arg(6)
+ movq [rdi], xmm13;
+ mov rdi,arg(7)
+ movq [rdi], xmm12;
+ mov rdi,arg(8)
+ movq [rdi], xmm11;
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c
index 3158ac1..5ab3641 100644
--- a/vp8/encoder/x86/x86_csystemdependent.c
+++ b/vp8/encoder/x86/x86_csystemdependent.c
@@ -176,6 +176,25 @@
d->dqcoeff
);
}
+#if CONFIG_PSNR
+#if ARCH_X86_64
+typedef void ssimpf
+(
+ unsigned char *s,
+ int sp,
+ unsigned char *r,
+ int rp,
+ unsigned long *sum_s,
+ unsigned long *sum_r,
+ unsigned long *sum_sq_s,
+ unsigned long *sum_sq_r,
+ unsigned long *sum_sxr
+);
+
+extern ssimpf vp8_ssim_parms_16x16_sse3;
+extern ssimpf vp8_ssim_parms_8x8_sse3;
+#endif
+#endif
#endif
@@ -280,6 +299,8 @@
cpi->rtcd.variance.get16x16prederror = vp8_get16x16pred_error_sse2;
cpi->rtcd.variance.get8x8var = vp8_get8x8var_sse2;
cpi->rtcd.variance.get16x16var = vp8_get16x16var_sse2;
+
+
/* cpi->rtcd.variance.get4x4sse_cs not implemented for wmt */;
cpi->rtcd.fdct.short4x4 = vp8_short_fdct4x4_sse2;
@@ -339,9 +360,18 @@
cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_ssse3;
+#if CONFIG_PSNR
+#if ARCH_X86_64
+ cpi->rtcd.variance.ssimpf_8x8 = vp8_ssim_parms_8x8_sse3;
+ cpi->rtcd.variance.ssimpf = vp8_ssim_parms_16x16_sse3;
+#endif
+#endif
+
}
#endif
+
+
#if HAVE_SSE4_1
if (SSE4_1Enabled)
{
diff --git a/vp8/vp8cx.mk b/vp8/vp8cx.mk
index c0ae250..670c022 100644
--- a/vp8/vp8cx.mk
+++ b/vp8/vp8cx.mk
@@ -116,6 +116,7 @@
VP8_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/sad_sse4.asm
VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/quantize_mmx.asm
VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/encodeopt.asm
+VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/ssim_opt.asm
ifeq ($(CONFIG_REALTIME_ONLY),yes)
VP8_CX_SRCS_REMOVE-$(HAVE_SSE2) += encoder/x86/temporal_filter_apply_sse2.asm
