Initial OpenCL implementation of the VP8 decoder.

Change-Id: I74c334af09f13473ce07bbac74b0f9ea57573347
Note: very slow, but functional.  Encoder is untested, but should still work.

.gitignore

@@ -60,3 +60,9 @@
 /vpx_config.h
 /vpx_version.h
 TAGS
+vpxdec
+vpxenc
+.project
+.cproject
+*.csv
+*.oclpj

.mailmap

@@ -2,4 +2,3 @@ Adrian Grange <agrange@google.com>
 Johann Koenig <johannkoenig@google.com>
 Tero Rintaluoma <teror@google.com> <tero.rintaluoma@on2.com>
 Tom Finegan <tomfinegan@google.com>
-Ralph Giles <giles@xiph.org> <giles@entropywave.com>

AUTHORS

@@ -4,11 +4,8 @@
 Aaron Watry <awatry@gmail.com>
 Adrian Grange <agrange@google.com>
 Alex Converse <alex.converse@gmail.com>
-Alexis Ballier <aballier@gentoo.org>
-Alok Ahuja <waveletcoeff@gmail.com>
 Andoni Morales Alastruey <ylatuya@gmail.com>
 Andres Mejia <mcitadel@gmail.com>
-Aron Rosenberg <arosenberg@logitech.com>
 Attila Nagy <attilanagy@google.com>
 Fabio Pedretti <fabio.ped@libero.it>
 Frank Galligan <fgalligan@google.com>
@@ -25,29 +22,20 @@ Jeff Muizelaar <jmuizelaar@mozilla.com>
 Jim Bankoski <jimbankoski@google.com>
 Johann Koenig <johannkoenig@google.com>
 John Koleszar <jkoleszar@google.com>
-Joshua Bleecher Snyder <josh@treelinelabs.com>
 Justin Clift <justin@salasaga.org>
 Justin Lebar <justin.lebar@gmail.com>
-Lou Quillio <louquillio@google.com>
 Luca Barbato <lu_zero@gentoo.org>
 Makoto Kato <makoto.kt@gmail.com>
 Martin Ettl <ettl.martin78@googlemail.com>
 Michael Kohler <michaelkohler@live.com>
-Mike Hommey <mhommey@mozilla.com>
 Mikhal Shemer <mikhal@google.com>
 Pascal Massimino <pascal.massimino@gmail.com>
 Patrik Westin <patrik.westin@gmail.com>
 Paul Wilkins <paulwilkins@google.com>
 Pavol Rusnak <stick@gk2.sk>
 Philip Jägenstedt <philipj@opera.com>
-Rafael Ávila de Espíndola <rafael.espindola@gmail.com>
-Ralph Giles <giles@xiph.org>
-Ronald S. Bultje <rbultje@google.com>
 Scott LaVarnway <slavarnway@google.com>
-Stefan Holmer <holmer@google.com>
-Taekhyun Kim <takim@nvidia.com>
 Tero Rintaluoma <teror@google.com>
-Thijs Vermeir <thijsvermeir@gmail.com>
 Timothy B. Terriberry <tterribe@xiph.org>
 Tom Finegan <tomfinegan@google.com>
 Yaowu Xu <yaowu@google.com>

CHANGELOG

@@ -1,85 +1,3 @@
-2011-08-02 v0.9.7 "Cayuga"
-  Our third named release, focused on a faster, higher quality, encoder.
-
-  - Upgrading:
-    This release is backwards compatible with Aylesbury (v0.9.5) and
-    Bali (v0.9.6). Users of older releases should refer to the Upgrading
-    notes in this document for that release.
-
-  - Enhancements:
-          Stereo 3D format support for vpxenc
-          Runtime detection of available processor cores.
-          Allow specifying --end-usage by enum name
-          vpxdec: test for frame corruption
-          vpxenc: add quantizer histogram display
-          vpxenc: add rate histogram display
-          Set VPX_FRAME_IS_DROPPABLE
-          update configure for ios sdk 4.3
-          Avoid text relocations in ARM vp8 decoder
-          Generate a vpx.pc file for pkg-config.
-          New ways of passing encoded data between encoder and decoder.
-
-  - Speed:
-      This release includes across-the-board speed improvements to the
-      encoder. On x86, these measure at approximately 11.5% in Best mode,
-      21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
-      On ARM Cortex A9 with Neon extensions, real-time encoding of video
-      telephony content is 35% faster than Bali on single core and 48%
-      faster on multi-core. On the NVidia Tegra2 platform, real time
-      encoding is 40% faster than Bali.
-
-      Decoder speed was not a priority for this release, but improved
-      approximately 8.4% on x86.
-
-          Reduce motion vector search on alt-ref frame.
-          Encoder loopfilter running in its own thread
-          Reworked loopfilter to precalculate more parameters
-          SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
-          Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
-          Removed redundant checks
-          Reduced structure sizes
-          utilize preload in ARMv6 MC/LPF/Copy routines
-          ARM optimized quantization, dfct, variance, subtract
-          Increase chrow row alignment to 16 bytes.
-          disable trellis optimization for first pass
-          Write SSSE3 sub-pixel filter function
-          Improve SSE2 half-pixel filter funtions
-          Add vp8_sub_pixel_variance16x8_ssse3 function
-          Reduce unnecessary distortion computation
-          Use diamond search to replace full search
-          Preload reference area in sub-pixel motion search (real-time mode)
-
-  - Quality:
-      This release focused primarily on one-pass use cases, including
-      video conferencing. Low latency data rate control was significantly
-      improved, improving streamability over bandwidth constrained links.
-      Added support for error concealment, allowing frames to maintain
-      visual quality in the presence of substantial packet loss.
-
-          Add rc_max_intra_bitrate_pct control
-          Limit size of initial keyframe in one-pass.
-          Improve framerate adaptation
-          Improved 1-pass CBR rate control
-          Improved KF insertion after fades to still.
-          Improved key frame detection.
-          Improved activity masking (lower PSNR impact for same SSIM boost)
-          Improved interaction between GF and ARFs
-          Adding error-concealment to the decoder.
-          Adding support for independent partitions
-          Adjusted rate-distortion constants
-
-
-  - Bug Fixes:
-          Removed firstpass motion map
-          Fix parallel make install
-          Fix multithreaded encoding for 1 MB wide frame
-          Fixed iwalsh_neon build problems with RVDS4.1
-          Fix semaphore emulation, spin-wait intrinsics on Windows
-          Fix build with xcode4 and simplify GLOBAL.
-          Mark ARM asm objects as allowing a non-executable stack.
-          Fix vpxenc encoding incorrect webm file header on big endian
-
-
 2011-03-07 v0.9.6 "Bali"
   Our second named release, focused on a faster, higher quality, encoder.
 

build/make/Makefile

@@ -98,11 +98,11 @@ install::
 $(BUILD_PFX)%.c.d: %.c
 	$(if $(quiet),@echo "    [DEP] $@")
 	$(qexec)mkdir -p $(dir $@)
-	$(qexec)$(CC) $(INTERNAL_CFLAGS) $(CFLAGS) -M $< | $(fmt_deps) > $@
+	$(qexec)$(CC) $(CFLAGS) -M $< | $(fmt_deps) > $@
 
 $(BUILD_PFX)%.c.o: %.c
 	$(if $(quiet),@echo "    [CC] $@")
-	$(qexec)$(CC) $(INTERNAL_CFLAGS) $(CFLAGS) -c -o $@ $<
+	$(qexec)$(CC) $(CFLAGS) -c -o $@ $<
 
 $(BUILD_PFX)%.asm.d: %.asm
 	$(if $(quiet),@echo "    [DEP] $@")
@@ -124,12 +124,6 @@ $(BUILD_PFX)%.s.o: %.s
 	$(if $(quiet),@echo "    [AS] $@")
 	$(qexec)$(AS) $(ASFLAGS) -o $@ $<
 
-.PRECIOUS: %.c.S
-%.c.S: CFLAGS += -DINLINE_ASM
-$(BUILD_PFX)%.c.S: %.c
-	$(if $(quiet),@echo "    [GEN] $@")
-	$(qexec)$(CC) -S $(CFLAGS) -o $@ $<
-
 .PRECIOUS: %.asm.s
 $(BUILD_PFX)%.asm.s: %.asm
 	$(if $(quiet),@echo "    [ASM CONVERSION] $@")
@@ -194,7 +188,7 @@ define linker_template
 $(1): $(filter-out -%,$(2))
 $(1):
 	$(if $(quiet),@echo    "    [LD] $$@")
-	$(qexec)$$(LD) $$(strip $$(INTERNAL_LDFLAGS) $$(LDFLAGS) -o $$@ $(2) $(3) $$(extralibs))
+	$(qexec)$$(LD) $$(strip $$(LDFLAGS) -o $$@ $(2) $(3) $$(extralibs))
 endef
 # make-3.80 has a bug with expanding large input strings to the eval function,
 # which was triggered in some cases by the following component of
@@ -336,7 +330,6 @@ ifneq ($(call enabled,DIST-SRCS),)
     DIST-SRCS-$(CONFIG_MSVS)  += build/make/gen_msvs_proj.sh
     DIST-SRCS-$(CONFIG_MSVS)  += build/make/gen_msvs_sln.sh
     DIST-SRCS-$(CONFIG_MSVS)  += build/x86-msvs/yasm.rules
-    DIST-SRCS-$(CONFIG_MSVS)  += build/x86-msvs/obj_int_extract.bat
     DIST-SRCS-$(CONFIG_RVCT) += build/make/armlink_adapter.sh
     # Include obj_int_extract if we use offsets from asm_*_offsets
     DIST-SRCS-$(ARCH_ARM)$(ARCH_X86)$(ARCH_X86_64)    += build/make/obj_int_extract.c

build/make/ads2gas.pl

@@ -21,14 +21,8 @@ print "@ This file was created from a .asm file\n";
 print "@  using the ads2gas.pl script.\n";
 print "\t.equ DO1STROUNDING, 0\n";
 
-# Stack of procedure names.
-@proc_stack = ();
-
 while (<STDIN>)
 {
-    # Load and store alignment
-    s/@/,:/g;
-
     # Comment character
     s/;/@/g;
 
@@ -85,10 +79,7 @@ while (<STDIN>)
     s/CODE([0-9][0-9])/.code $1/;
 
     # No AREA required
-    # But ALIGNs in AREA must be obeyed
-    s/^\s*AREA.*ALIGN=([0-9])$/.text\n.p2align $1/;
-    # If no ALIGN, strip the AREA and align to 4 bytes
-    s/^\s*AREA.*$/.text\n.p2align 2/;
+    s/^\s*AREA.*$/.text/;
 
     # DCD to .word
     # This one is for incoming symbols
@@ -123,8 +114,8 @@ while (<STDIN>)
     # put the colon at the end of the line in the macro
     s/^([a-zA-Z_0-9\$]+)/$1:/ if !/EQU/;
 
-    # ALIGN directive
-    s/ALIGN/.balign/g;
+    # Strip ALIGN
+    s/\sALIGN/@ ALIGN/g;
 
     # Strip ARM
     s/\sARM/@ ARM/g;
@@ -136,23 +127,9 @@ while (<STDIN>)
     # Strip PRESERVE8
     s/\sPRESERVE8/@ PRESERVE8/g;
 
-    # Use PROC and ENDP to give the symbols a .size directive.
-    # This makes them show up properly in debugging tools like gdb and valgrind.
-    if (/\bPROC\b/)
-    {
-        my $proc;
-        /^_([\.0-9A-Z_a-z]\w+)\b/;
-        $proc = $1;
-        push(@proc_stack, $proc) if ($proc);
-        s/\bPROC\b/@ $&/;
-    }
-    if (/\bENDP\b/)
-    {
-        my $proc;
-        s/\bENDP\b/@ $&/;
-        $proc = pop(@proc_stack);
-        $_ = "\t.size $proc, .-$proc".$_ if ($proc);
-    }
+    # Strip PROC and ENDPROC
+    s/\sPROC/@/g;
+    s/\sENDP/@/g;
 
     # EQU directive
     s/(.*)EQU(.*)/.equ $1, $2/;
@@ -171,6 +148,3 @@ while (<STDIN>)
     next if /^\s*END\s*$/;
     print;
 }
-
-# Mark that this object doesn't need an executable stack.
-printf ("\t.section\t.note.GNU-stack,\"\",\%\%progbits\n");

build/make/ads2gas_apple.pl

@@ -41,9 +41,6 @@ sub trim($)
 
 while (<STDIN>)
 {
-    # Load and store alignment
-    s/@/,:/g;
-
     # Comment character
     s/;/@/g;
 
@@ -100,10 +97,7 @@ while (<STDIN>)
     s/CODE([0-9][0-9])/.code $1/;
 
     # No AREA required
-    # But ALIGNs in AREA must be obeyed
-    s/^\s*AREA.*ALIGN=([0-9])$/.text\n.p2align $1/;
-    # If no ALIGN, strip the AREA and align to 4 bytes
-    s/^\s*AREA.*$/.text\n.p2align 2/;
+    s/^\s*AREA.*$/.text/;
 
     # DCD to .word
     # This one is for incoming symbols
@@ -143,8 +137,8 @@ while (<STDIN>)
     # put the colon at the end of the line in the macro
     s/^([a-zA-Z_0-9\$]+)/$1:/ if !/EQU/;
 
-    # ALIGN directive
-    s/ALIGN/.balign/g;
+    # Strip ALIGN
+    s/\sALIGN/@ ALIGN/g;
 
     # Strip ARM
     s/\sARM/@ ARM/g;

build/make/configure.sh

@@ -412,14 +412,11 @@ EOF
 write_common_target_config_h() {
     cat > ${TMP_H} << EOF
 /* This file automatically generated by configure. Do not edit! */
-#ifndef VPX_CONFIG_H
-#define VPX_CONFIG_H
 #define RESTRICT    ${RESTRICT}
 EOF
     print_config_h ARCH   "${TMP_H}" ${ARCH_LIST}
     print_config_h HAVE   "${TMP_H}" ${HAVE_LIST}
     print_config_h CONFIG "${TMP_H}" ${CONFIG_LIST}
-    echo "#endif /* VPX_CONFIG_H */" >> ${TMP_H}
     mkdir -p `dirname "$1"`
     cmp "$1" ${TMP_H} >/dev/null 2>&1 || mv ${TMP_H} "$1"
 }
@@ -629,7 +626,7 @@ process_common_toolchain() {
     case ${toolchain} in
         sparc-solaris-*)
             add_extralibs -lposix4
-            disable fast_unaligned
+            add_cflags "-DMUST_BE_ALIGNED"
             ;;
         *-solaris-*)
             add_extralibs -lposix4
@@ -642,8 +639,8 @@ process_common_toolchain() {
     # on arm, isa versions are supersets
     enabled armv7a && soft_enable armv7 ### DEBUG
     enabled armv7 && soft_enable armv6
-    enabled armv7 || enabled armv6 && soft_enable armv5te
-    enabled armv7 || enabled armv6 && soft_enable fast_unaligned
+    enabled armv6 && soft_enable armv5te
+    enabled armv6 && soft_enable fast_unaligned
     enabled iwmmxt2 && soft_enable iwmmxt
     enabled iwmmxt && soft_enable armv5te
 
@@ -692,7 +689,7 @@ process_common_toolchain() {
             if enabled armv7
                 then
                     check_add_cflags --cpu=Cortex-A8 --fpu=softvfp+vfpv3
-                    check_add_asflags --cpu=Cortex-A8 --fpu=softvfp+vfpv3
+                    check_add_asflags --cpu=Cortex-A8 --fpu=none
                 else
                     check_add_cflags --cpu=${tgt_isa##armv}
                     check_add_asflags --cpu=${tgt_isa##armv}
@@ -754,24 +751,41 @@ process_common_toolchain() {
         linux*)
             enable linux
             if enabled rvct; then
-                # Check if we have CodeSourcery GCC in PATH. Needed for
-                # libraries
-                hash arm-none-linux-gnueabi-gcc 2>&- || \
-                  die "Couldn't find CodeSourcery GCC from PATH"
+                # Compiling with RVCT requires an alternate libc (glibc) when
+                # targetting linux.
+                disabled builtin_libc \
+                    || die "Must supply --libc when targetting *-linux-rvct"
 
-                # Use armcc as a linker to enable translation of
-                # some gcc specific options such as -lm and -lpthread.
-                LD="armcc --translate_gcc"
+                # Set up compiler
+                add_cflags --library_interface=aeabi_glibc
+                add_cflags --no_hide_all
+                add_cflags --dwarf2
 
-                # create configuration file (uses path to CodeSourcery GCC)
-                armcc --arm_linux_configure --arm_linux_config_file=arm_linux.cfg
+                # Set up linker
+                add_ldflags --sysv --no_startup --no_ref_cpp_init
+                add_ldflags --entry=_start
+                add_ldflags --keep '"*(.init)"' --keep '"*(.fini)"'
+                add_ldflags --keep '"*(.init_array)"' --keep '"*(.fini_array)"'
+                add_ldflags --dynamiclinker=/lib/ld-linux.so.3
+                add_extralibs libc.so.6 -lc_nonshared crt1.o crti.o crtn.o
 
-                add_cflags --arm_linux_paths --arm_linux_config_file=arm_linux.cfg
-                add_asflags --no_hide_all --apcs=/interwork
-                add_ldflags --arm_linux_paths --arm_linux_config_file=arm_linux.cfg
-                enabled pic && add_cflags --apcs=/fpic
-                enabled pic && add_asflags --apcs=/fpic
-                enabled shared && add_cflags --shared
+                # Add the paths for the alternate libc
+                for d in usr/include; do
+                    try_dir="${alt_libc}/${d}"
+                    [ -d "${try_dir}" ] && add_cflags -J"${try_dir}"
+                done
+                add_cflags -J"${RVCT31INC}"
+                for d in lib usr/lib; do
+                    try_dir="${alt_libc}/${d}"
+                    [ -d "${try_dir}" ] && add_ldflags -L"${try_dir}"
+                done
+
+
+                # glibc has some struct members named __align, which is a
+                # storage modifier in RVCT. If we need to use this modifier,
+                # we'll have to #undef it in our code. Note that this must
+                # happen AFTER all libc inclues.
+                add_cflags -D__align=x_align_x
             fi
         ;;
 
@@ -939,23 +953,47 @@ process_common_toolchain() {
     enabled gcov &&
         check_add_cflags -fprofile-arcs -ftest-coverage &&
         check_add_ldflags -fprofile-arcs -ftest-coverage
-
     if enabled optimizations; then
-        if enabled rvct; then
-            enabled small && check_add_cflags -Ospace || check_add_cflags -Otime
-        else
+        enabled rvct && check_add_cflags -Otime
         enabled small && check_add_cflags -O2 || check_add_cflags -O3
     fi
+    
+    if enabled opencl; then
+        disable multithread
+        echo "  disabling multithread"
+        soft_enable opencl #Provide output to make user comfortable
+        enable runtime_cpu_detect
+	
+        #Use dlopen() to load OpenCL when possible.
+        case ${toolchain} in
+            *darwin10*)
+                check_add_cflags -D__APPLE__
+                add_extralibs -framework OpenCL
+                ;;
+            *-win32-gcc)
+                if check_header dlfcn.h; then
+                    add_extralibs -ldl 
+                    enable dlopen
+                else
+                    #This shouldn't be a hard-coded path in the long term
+                    add_extralibs -L/cygdrive/c/Windows/System32 -lOpenCL
+                fi
+                ;;
+            *)
+                if check_header dlfcn.h; then
+                    add_extralibs -ldl 
+                    enable dlopen
+                else
+                    add_extralibs -lOpenCL
+                fi
+                ;;
+        esac
     fi
 
     # Position Independent Code (PIC) support, for building relocatable
     # shared objects
     enabled gcc && enabled pic && check_add_cflags -fPIC
 
-    # Work around longjmp interception on glibc >= 2.11, to improve binary
-    # compatibility. See http://code.google.com/p/webm/issues/detail?id=166
-    enabled linux && check_add_cflags -D_FORTIFY_SOURCE=0
-
     # Check for strip utility variant
     ${STRIP} -V 2>/dev/null | grep GNU >/dev/null && enable gnu_strip
 
@@ -974,9 +1012,6 @@ EOF
         esac
     fi
 
-    # for sysconf(3) and friends.
-    check_header unistd.h
-
     # glibc needs these
     if enabled linux; then
         add_cflags -D_LARGEFILE_SOURCE

build/make/gen_msvs_proj.sh

@@ -365,7 +365,7 @@ generate_vcproj() {
                             DebugInformationFormat="1" \
                             Detect64BitPortabilityProblems="true" \
 
-                        $uses_asm && tag Tool Name="YASM"  IncludePaths="$incs" Debug="true"
+                        $uses_asm && tag Tool Name="YASM"  IncludePaths="$incs" Debug="1"
                     ;;
                     *)
                         tag Tool \
@@ -379,7 +379,7 @@ generate_vcproj() {
                             DebugInformationFormat="1" \
                             Detect64BitPortabilityProblems="true" \
 
-                        $uses_asm && tag Tool Name="YASM"  IncludePaths="$incs" Debug="true"
+                        $uses_asm && tag Tool Name="YASM"  IncludePaths="$incs" Debug="1"
                     ;;
                 esac
             ;;
@@ -447,8 +447,6 @@ generate_vcproj() {
                     obj_int_extract)
                         tag Tool \
                             Name="VCCLCompilerTool" \
-                            Optimization="2" \
-                            FavorSizeorSpeed="1" \
                             AdditionalIncludeDirectories="$incs" \
                             PreprocessorDefinitions="WIN32;NDEBUG;_CONSOLE;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE" \
                             RuntimeLibrary="$release_runtime" \
@@ -464,8 +462,6 @@ generate_vcproj() {
 
                         tag Tool \
                             Name="VCCLCompilerTool" \
-                            Optimization="2" \
-                            FavorSizeorSpeed="1" \
                             AdditionalIncludeDirectories="$incs" \
                             PreprocessorDefinitions="WIN32;NDEBUG;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE;$defines" \
                             RuntimeLibrary="$release_runtime" \
@@ -480,8 +476,6 @@ generate_vcproj() {
                         tag Tool \
                             Name="VCCLCompilerTool" \
                             AdditionalIncludeDirectories="$incs" \
-                            Optimization="2" \
-                            FavorSizeorSpeed="1" \
                             PreprocessorDefinitions="WIN32;NDEBUG;_CRT_SECURE_NO_WARNINGS;_CRT_SECURE_NO_DEPRECATE;$defines" \
                             RuntimeLibrary="$release_runtime" \
                             UsePrecompiledHeader="0" \

configure

@@ -31,17 +31,16 @@ Advanced options:
   ${toggle_md5}                   support for output of checksum data
   ${toggle_static_msvcrt}         use static MSVCRT (VS builds only)
   ${toggle_vp8}                   VP8 codec support
-  ${toggle_internal_stats}        output of encoder internal stats for debug, if supported (encoders)
+  ${toggle_psnr}                  output of PSNR data, if supported (encoders)
   ${toggle_mem_tracker}           track memory usage
   ${toggle_postproc}              postprocessing
   ${toggle_multithread}           multithreaded encoding and decoding.
   ${toggle_spatial_resampling}    spatial sampling (scaling) support
   ${toggle_realtime_only}         enable this option while building for real-time encoding
-  ${toggle_error_concealment}     enable this option to get a decoder which is able to conceal losses
   ${toggle_runtime_cpu_detect}    runtime cpu detection
   ${toggle_shared}                shared library support
-  ${toggle_static}                static library support
   ${toggle_small}                 favor smaller size over speed
+  ${toggle_opencl}                support for OpenCL-assisted VP8 decoding (experimental)
   ${toggle_postproc_visualizer}   macro block / block level visualizers
 
 Codecs:
@@ -107,6 +106,7 @@ all_platforms="${all_platforms} x86-darwin8-gcc"
 all_platforms="${all_platforms} x86-darwin8-icc"
 all_platforms="${all_platforms} x86-darwin9-gcc"
 all_platforms="${all_platforms} x86-darwin9-icc"
+all_platforms="${all_platforms} x86-darwin10-gcc"
 all_platforms="${all_platforms} x86-linux-gcc"
 all_platforms="${all_platforms} x86-linux-icc"
 all_platforms="${all_platforms} x86-solaris-gcc"
@@ -154,7 +154,6 @@ enabled doxygen && php -v >/dev/null 2>&1 && enable install_docs
 enable install_bins
 enable install_libs
 
-enable static
 enable optimizations
 enable fast_unaligned #allow unaligned accesses, if supported by hw
 enable md5
@@ -214,7 +213,7 @@ HAVE_LIST="
     alt_tree_layout
     pthread_h
     sys_mman_h
-    unistd_h
+    dlopen
 "
 CONFIG_LIST="
     external_build
@@ -244,7 +243,7 @@ CONFIG_LIST="
     runtime_cpu_detect
     postproc
     multithread
-    internal_stats
+    psnr
     ${CODECS}
     ${CODEC_FAMILIES}
     encoders
@@ -252,10 +251,9 @@ CONFIG_LIST="
     static_msvcrt
     spatial_resampling
     realtime_only
-    error_concealment
     shared
-    static
     small
+    opencl
     postproc_visualizer
     os_support
 "
@@ -287,17 +285,16 @@ CMDLINE_SELECT="
     dc_recon
     postproc
     multithread
-    internal_stats
+    psnr
     ${CODECS}
     ${CODEC_FAMILIES}
     static_msvcrt
     mem_tracker
     spatial_resampling
     realtime_only
-    error_concealment
     shared
-    static
     small
+    opencl
     postproc_visualizer
 "
 
@@ -564,4 +561,6 @@ process "$@"
 cat <<EOF > ${BUILD_PFX}vpx_config.c
 static const char* const cfg = "$CONFIGURE_ARGS";
 const char *vpx_codec_build_config(void) {return cfg;}
+static const char* const libdir = "$libdir";
+const char *vpx_codec_lib_dir(void) {return libdir;}
 EOF

examples.mk

@@ -77,11 +77,6 @@ GEN_EXAMPLES-$(CONFIG_ENCODERS) += decode_with_drops.c
 endif
 decode_with_drops.GUID           = CE5C53C4-8DDA-438A-86ED-0DDD3CDB8D26
 decode_with_drops.DESCRIPTION    = Drops frames while decoding
-ifeq ($(CONFIG_DECODERS),yes)
-GEN_EXAMPLES-$(CONFIG_ERROR_CONCEALMENT) += decode_with_partial_drops.c
-endif
-decode_with_partial_drops.GUID           = 61C2D026-5754-46AC-916F-1343ECC5537E
-decode_with_partial_drops.DESCRIPTION    = Drops parts of frames while decoding
 GEN_EXAMPLES-$(CONFIG_ENCODERS) += error_resilient.c
 error_resilient.GUID             = DF5837B9-4145-4F92-A031-44E4F832E00C
 error_resilient.DESCRIPTION      = Error Resiliency Feature
@@ -127,8 +122,8 @@ else
     LIB_PATH := $(call enabled,LIB_PATH)
     INC_PATH := $(call enabled,INC_PATH)
 endif
-INTERNAL_CFLAGS = $(addprefix -I,$(INC_PATH))
-INTERNAL_LDFLAGS += $(addprefix -L,$(LIB_PATH))
+CFLAGS += $(addprefix -I,$(INC_PATH))
+LDFLAGS += $(addprefix -L,$(LIB_PATH))
 
 
 # Expand list of selected examples to build (as specified above)
@@ -167,10 +162,8 @@ BINS-$(NOT_MSVS)           += $(addprefix $(BUILD_PFX),$(ALL_EXAMPLES:.c=))
 
 # Instantiate linker template for all examples.
 CODEC_LIB=$(if $(CONFIG_DEBUG_LIBS),vpx_g,vpx)
-CODEC_LIB_SUF=$(if $(CONFIG_SHARED),.so,.a)
 $(foreach bin,$(BINS-yes),\
-    $(if $(BUILD_OBJS),$(eval $(bin):\
-        $(LIB_PATH)/lib$(CODEC_LIB)$(CODEC_LIB_SUF)))\
+    $(if $(BUILD_OBJS),$(eval $(bin): $(LIB_PATH)/lib$(CODEC_LIB).a))\
     $(if $(BUILD_OBJS),$(eval $(call linker_template,$(bin),\
         $(call objs,$($(notdir $(bin)).SRCS)) \
         -l$(CODEC_LIB) $(addprefix -l,$(CODEC_EXTRA_LIBS))\
@@ -221,8 +214,7 @@ $(1): $($(1:.vcproj=).SRCS)
             --ver=$$(CONFIG_VS_VERSION)\
             --proj-guid=$$($$(@:.vcproj=).GUID)\
             $$(if $$(CONFIG_STATIC_MSVCRT),--static-crt) \
-            --out=$$@ $$(INTERNAL_CFLAGS) $$(CFLAGS) \
-            $$(INTERNAL_LDFLAGS) $$(LDFLAGS) -l$$(CODEC_LIB) -lwinmm $$^
+            --out=$$@ $$(CFLAGS) $$(LDFLAGS) -l$$(CODEC_LIB) -lwinmm $$^
 endef
 PROJECTS-$(CONFIG_MSVS) += $(ALL_EXAMPLES:.c=.vcproj)
 INSTALL-BINS-$(CONFIG_MSVS) += $(foreach p,$(VS_PLATFORMS),\

examples/decode_with_partial_drops.txt

@@ -1,238 +0,0 @@
-@TEMPLATE decoder_tmpl.c
-Decode With Partial Drops Example
-=========================
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ INTRODUCTION
-This is an example utility which drops a series of frames (or parts of frames),
-as specified on the command line. This is useful for observing the error
-recovery features of the codec.
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ INTRODUCTION
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ EXTRA_INCLUDES
-#include <time.h>
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ EXTRA_INCLUDES
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ HELPERS
-struct parsed_header
-{
-    char key_frame;
-    int version;
-    char show_frame;
-    int first_part_size;
-};
-
-int next_packet(struct parsed_header* hdr, int pos, int length, int mtu)
-{
-    int size = 0;
-    int remaining = length - pos;
-    /* Uncompressed part is 3 bytes for P frames and 10 bytes for I frames */
-    int uncomp_part_size = (hdr->key_frame ? 10 : 3);
-    /* number of bytes yet to send from header and the first partition */
-    int remainFirst = uncomp_part_size + hdr->first_part_size - pos;
-    if (remainFirst > 0)
-    {
-        if (remainFirst <= mtu)
-        {
-            size = remainFirst;
-        }
-        else
-        {
-            size = mtu;
-        }
-
-        return size;
-    }
-
-    /* second partition; just slot it up according to MTU */
-    if (remaining <= mtu)
-    {
-        size = remaining;
-        return size;
-    }
-    return mtu;
-}
-
-void throw_packets(unsigned char* frame, int* size, int loss_rate,
-                   int* thrown, int* kept)
-{
-    unsigned char loss_frame[256*1024];
-    int pkg_size = 1;
-    int pos = 0;
-    int loss_pos = 0;
-    struct parsed_header hdr;
-    unsigned int tmp;
-    int mtu = 1500;
-
-    if (*size < 3)
-    {
-        return;
-    }
-    putc('|', stdout);
-    /* parse uncompressed 3 bytes */
-    tmp = (frame[2] << 16) | (frame[1] << 8) | frame[0];
-    hdr.key_frame = !(tmp & 0x1); /* inverse logic */
-    hdr.version = (tmp >> 1) & 0x7;
-    hdr.show_frame = (tmp >> 4) & 0x1;
-    hdr.first_part_size = (tmp >> 5) & 0x7FFFF;
-
-    /* don't drop key frames */
-    if (hdr.key_frame)
-    {
-        int i;
-        *kept = *size/mtu + ((*size % mtu > 0) ? 1 : 0); /* approximate */
-        for (i=0; i < *kept; i++)
-            putc('.', stdout);
-        return;
-    }
-
-    while ((pkg_size = next_packet(&hdr, pos, *size, mtu)) > 0)
-    {
-        int loss_event = ((rand() + 1.0)/(RAND_MAX + 1.0) < loss_rate/100.0);
-        if (*thrown == 0 && !loss_event)
-        {
-            memcpy(loss_frame + loss_pos, frame + pos, pkg_size);
-            loss_pos += pkg_size;
-            (*kept)++;
-            putc('.', stdout);
-        }
-        else
-        {
-            (*thrown)++;
-            putc('X', stdout);
-        }
-        pos += pkg_size;
-    }
-    memcpy(frame, loss_frame, loss_pos);
-    memset(frame + loss_pos, 0, *size - loss_pos);
-    *size = loss_pos;
-}
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ HELPERS
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ DEC_INIT
-/* Initialize codec */
-flags = VPX_CODEC_USE_ERROR_CONCEALMENT;
-res = vpx_codec_dec_init(&codec, interface, &dec_cfg, flags);
-if(res)
-    die_codec(&codec, "Failed to initialize decoder");
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ DEC_INIT
-
-Usage
------
-This example adds a single argument to the `simple_decoder` example,
-which specifies the range or pattern of frames to drop. The parameter is
-parsed as follows:
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ USAGE
-if(argc < 4 || argc > 6)
-    die("Usage: %s <infile> <outfile> [-t <num threads>] <N-M|N/M|L,S>\n",
-        argv[0]);
-{
-    char *nptr;
-    int arg_num = 3;
-    if (argc == 6 && strncmp(argv[arg_num++], "-t", 2) == 0)
-        dec_cfg.threads = strtol(argv[arg_num++], NULL, 0);
-    n = strtol(argv[arg_num], &nptr, 0);
-    mode = (*nptr == '\0' || *nptr == ',') ? 2 : (*nptr == '-') ? 1 : 0;
-
-    m = strtol(nptr+1, NULL, 0);
-    if((!n && !m) || (*nptr != '-' && *nptr != '/' &&
-        *nptr != '\0' && *nptr != ','))
-        die("Couldn't parse pattern %s\n", argv[3]);
-}
-seed = (m > 0) ? m : (unsigned int)time(NULL);
-srand(seed);thrown_frame = 0;
-printf("Seed: %u\n", seed);
-printf("Threads: %d\n", dec_cfg.threads);
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ USAGE
-
-
-Dropping A Range Of Frames
---------------------------
-To drop a range of frames, specify the starting frame and the ending
-frame to drop, separated by a dash. The following command will drop
-frames 5 through 10 (base 1).
-
-  $ ./decode_with_partial_drops in.ivf out.i420 5-10
-
-
-Dropping A Pattern Of Frames
-----------------------------
-To drop a pattern of frames, specify the number of frames to drop and
-the number of frames after which to repeat the pattern, separated by
-a forward-slash. The following command will drop 3 of 7 frames.
-Specifically, it will decode 4 frames, then drop 3 frames, and then
-repeat.
-
-  $ ./decode_with_partial_drops in.ivf out.i420 3/7
-
-Dropping Random Parts Of Frames
--------------------------------
-A third argument tuple is available to split the frame into 1500 bytes pieces
-and randomly drop pieces rather than frames. The frame will be split at
-partition boundaries where possible. The following example will seed the RNG
-with the seed 123 and drop approximately 5% of the pieces. Pieces which
-are depending on an already dropped piece will also be dropped.
-
-  $ ./decode_with_partial_drops in.ivf out.i420 5,123
-
-
-Extra Variables
----------------
-This example maintains the pattern passed on the command line in the
-`n`, `m`, and `is_range` variables:
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ EXTRA_VARS
-int              n, m, mode;
-unsigned int     seed;
-int              thrown=0, kept=0;
-int              thrown_frame=0, kept_frame=0;
-vpx_codec_dec_cfg_t  dec_cfg = {0};
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ EXTRA_VARS
-
-
-Making The Drop Decision
-------------------------
-The example decides whether to drop the frame based on the current
-frame number, immediately before decoding the frame.
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ PRE_DECODE
-/* Decide whether to throw parts of the frame or the whole frame
-   depending on the drop mode */
-thrown_frame = 0;
-kept_frame = 0;
-switch (mode)
-{
-case 0:
-    if (m - (frame_cnt-1)%m <= n)
-    {
-        frame_sz = 0;
-    }
-    break;
-case 1:
-    if (frame_cnt >= n && frame_cnt <= m)
-    {
-        frame_sz = 0;
-    }
-    break;
-case 2:
-    throw_packets(frame, &frame_sz, n, &thrown_frame, &kept_frame);
-    break;
-default: break;
-}
-if (mode < 2)
-{
-    if (frame_sz == 0)
-    {
-        putc('X', stdout);
-        thrown_frame++;
-    }
-    else
-    {
-        putc('.', stdout);
-        kept_frame++;
-    }
-}
-thrown += thrown_frame;
-kept += kept_frame;
-fflush(stdout);
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ PRE_DECODE

examples/decoder_tmpl.c

@@ -42,8 +42,6 @@ static void die(const char *fmt, ...) {
 
 @DIE_CODEC
 
-@HELPERS
-
 int main(int argc, char **argv) {
     FILE            *infile, *outfile;
     vpx_codec_ctx_t  codec;

examples/encoder_tmpl.c

@@ -111,6 +111,8 @@ int main(int argc, char **argv) {
     vpx_codec_ctx_t      codec;
     vpx_codec_enc_cfg_t  cfg;
     int                  frame_cnt = 0;
+    unsigned char        file_hdr[IVF_FILE_HDR_SZ];
+    unsigned char        frame_hdr[IVF_FRAME_HDR_SZ];
     vpx_image_t          raw;
     vpx_codec_err_t      res;
     long                 width;

examples/postproc.txt

@@ -21,7 +21,7 @@ res = vpx_codec_dec_init(&codec, interface, NULL,
 if(res == VPX_CODEC_INCAPABLE) {
     printf("NOTICE: Postproc not supported by %s\n",
            vpx_codec_iface_name(interface));
-    res = vpx_codec_dec_init(&codec, interface, NULL, flags);
+    res = vpx_codec_dec_init(&codec, interface, NULL, 0);
 }
 if(res)
     die_codec(&codec, "Failed to initialize decoder");

libmkv/EbmlIDs.h

@@ -120,7 +120,7 @@ enum mkv
     //video
     Video = 0xE0,
     FlagInterlaced = 0x9A,
-    StereoMode = 0x53B8,
+//  StereoMode = 0x53B8,
     PixelWidth = 0xB0,
     PixelHeight = 0xBA,
     PixelCropBottom = 0x54AA,

libmkv/EbmlWriter.c

@@ -11,7 +11,6 @@
 #include <stdlib.h>
 #include <wchar.h>
 #include <string.h>
-#include <limits.h>
 #if defined(_MSC_VER)
 #define LITERALU64(n) n
 #else
@@ -34,7 +33,7 @@ void Ebml_WriteLen(EbmlGlobal *glob, long long val)
 
     val |= (LITERALU64(0x000000000000080) << ((size - 1) * 7));
 
-    Ebml_Serialize(glob, (void *) &val, sizeof(val), size);
+    Ebml_Serialize(glob, (void *) &val, size);
 }
 
 void Ebml_WriteString(EbmlGlobal *glob, const char *str)
@@ -61,26 +60,21 @@ void Ebml_WriteUTF8(EbmlGlobal *glob, const wchar_t *wstr)
 
 void Ebml_WriteID(EbmlGlobal *glob, unsigned long class_id)
 {
-    int len;
-
     if (class_id >= 0x01000000)
-        len = 4;
+        Ebml_Serialize(glob, (void *)&class_id, 4);
     else if (class_id >= 0x00010000)
-        len = 3;
+        Ebml_Serialize(glob, (void *)&class_id, 3);
     else if (class_id >= 0x00000100)
-        len = 2;
+        Ebml_Serialize(glob, (void *)&class_id, 2);
     else
-        len = 1;
-
-    Ebml_Serialize(glob, (void *)&class_id, sizeof(class_id), len);
+        Ebml_Serialize(glob, (void *)&class_id, 1);
 }
-
 void Ebml_SerializeUnsigned64(EbmlGlobal *glob, unsigned long class_id, uint64_t ui)
 {
     unsigned char sizeSerialized = 8 | 0x80;
     Ebml_WriteID(glob, class_id);
-    Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
-    Ebml_Serialize(glob, &ui, sizeof(ui), 8);
+    Ebml_Serialize(glob, &sizeSerialized, 1);
+    Ebml_Serialize(glob, &ui, 8);
 }
 
 void Ebml_SerializeUnsigned(EbmlGlobal *glob, unsigned long class_id, unsigned long ui)
@@ -103,8 +97,8 @@ void Ebml_SerializeUnsigned(EbmlGlobal *glob, unsigned long class_id, unsigned l
     }
 
     sizeSerialized = 0x80 | size;
-    Ebml_Serialize(glob, &sizeSerialized, sizeof(sizeSerialized), 1);
-    Ebml_Serialize(glob, &ui, sizeof(ui), size);
+    Ebml_Serialize(glob, &sizeSerialized, 1);
+    Ebml_Serialize(glob, &ui, size);
 }
 //TODO: perhaps this is a poor name for this id serializer helper function
 void Ebml_SerializeBinary(EbmlGlobal *glob, unsigned long class_id, unsigned long bin)
@@ -125,14 +119,14 @@ void Ebml_SerializeFloat(EbmlGlobal *glob, unsigned long class_id, double d)
     unsigned char len = 0x88;
 
     Ebml_WriteID(glob, class_id);
-    Ebml_Serialize(glob, &len, sizeof(len), 1);
-    Ebml_Serialize(glob,  &d, sizeof(d), 8);
+    Ebml_Serialize(glob, &len, 1);
+    Ebml_Serialize(glob,  &d, 8);
 }
 
 void Ebml_WriteSigned16(EbmlGlobal *glob, short val)
 {
     signed long out = ((val & 0x003FFFFF) | 0x00200000) << 8;
-    Ebml_Serialize(glob, &out, sizeof(out), 3);
+    Ebml_Serialize(glob, &out, 3);
 }
 
 void Ebml_SerializeString(EbmlGlobal *glob, unsigned long class_id, const char *s)
@@ -149,6 +143,7 @@ void Ebml_SerializeUTF8(EbmlGlobal *glob, unsigned long class_id, wchar_t *s)
 
 void Ebml_SerializeData(EbmlGlobal *glob, unsigned long class_id, unsigned char *data, unsigned long data_length)
 {
+    unsigned char size = 4;
     Ebml_WriteID(glob, class_id);
     Ebml_WriteLen(glob, data_length);
     Ebml_Write(glob,  data, data_length);

libmkv/EbmlWriter.h

@@ -15,7 +15,7 @@
 #include "vpx/vpx_integer.h"
 
 typedef struct EbmlGlobal EbmlGlobal;
-void  Ebml_Serialize(EbmlGlobal *glob, const void *, int, unsigned long);
+void  Ebml_Serialize(EbmlGlobal *glob, const void *, unsigned long);
 void  Ebml_Write(EbmlGlobal *glob, const void *, unsigned long);
 /////
 

libmkv/WebMElement.c

@@ -35,11 +35,11 @@ void writeSimpleBlock(EbmlGlobal *glob, unsigned char trackNumber, short timeCod
     Ebml_WriteID(glob, SimpleBlock);
     unsigned long blockLength = 4 + dataLength;
     blockLength |= 0x10000000; //TODO check length < 0x0FFFFFFFF
-    Ebml_Serialize(glob, &blockLength, sizeof(blockLength), 4);
+    Ebml_Serialize(glob, &blockLength, 4);
     trackNumber |= 0x80;  //TODO check track nubmer < 128
     Ebml_Write(glob, &trackNumber, 1);
     //Ebml_WriteSigned16(glob, timeCode,2); //this is 3 bytes
-    Ebml_Serialize(glob, &timeCode, sizeof(timeCode), 2);
+    Ebml_Serialize(glob, &timeCode, 2);
     unsigned char flags = 0x00 | (isKeyframe ? 0x80 : 0x00) | (lacingFlag << 1) | discardable;
     Ebml_Write(glob, &flags, 1);
     Ebml_Write(glob, data, dataLength);

libs.mk

@@ -35,7 +35,6 @@ ifeq ($(CONFIG_VP8_ENCODER),yes)
   CODEC_SRCS-yes += $(addprefix $(VP8_PREFIX),$(call enabled,VP8_CX_SRCS))
   CODEC_EXPORTS-yes += $(addprefix $(VP8_PREFIX),$(VP8_CX_EXPORTS))
   CODEC_SRCS-yes += $(VP8_PREFIX)vp8cx.mk vpx/vp8.h vpx/vp8cx.h vpx/vp8e.h
-  CODEC_SRCS-$(ARCH_ARM) += $(VP8_PREFIX)vp8cx_arm.mk
   INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8e.h include/vpx/vp8cx.h
   INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP8_PREFIX)/%
   CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8cx.h
@@ -48,7 +47,6 @@ ifeq ($(CONFIG_VP8_DECODER),yes)
   CODEC_SRCS-yes += $(addprefix $(VP8_PREFIX),$(call enabled,VP8_DX_SRCS))
   CODEC_EXPORTS-yes += $(addprefix $(VP8_PREFIX),$(VP8_DX_EXPORTS))
   CODEC_SRCS-yes += $(VP8_PREFIX)vp8dx.mk vpx/vp8.h vpx/vp8dx.h
-  CODEC_SRCS-$(ARCH_ARM) += $(VP8_PREFIX)vp8dx_arm.mk
   INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8dx.h
   INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP8_PREFIX)/%
   CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8dx.h
@@ -91,7 +89,6 @@ $(eval $(if $(filter universal%,$(TOOLCHAIN)),LIPO_LIBVPX,BUILD_LIBVPX):=yes)
 
 CODEC_SRCS-$(BUILD_LIBVPX) += build/make/version.sh
 CODEC_SRCS-$(BUILD_LIBVPX) += vpx/vpx_integer.h
-CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/asm_offsets.h
 CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/vpx_timer.h
 CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/mem.h
 CODEC_SRCS-$(BUILD_LIBVPX) += $(BUILD_PFX)vpx_config.c
@@ -103,7 +100,7 @@ CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/x86_abi_support.asm
 CODEC_SRCS-$(BUILD_LIBVPX) += vpx_ports/x86_cpuid.c
 endif
 CODEC_SRCS-$(ARCH_ARM) += vpx_ports/arm_cpudetect.c
-CODEC_SRCS-$(ARCH_ARM) += vpx_ports/arm.h
+CODEC_SRCS-$(ARCH_ARM) += $(BUILD_PFX)vpx_config.asm
 CODEC_EXPORTS-$(BUILD_LIBVPX) += vpx/exports_com
 CODEC_EXPORTS-$(CONFIG_ENCODERS) += vpx/exports_enc
 CODEC_EXPORTS-$(CONFIG_DECODERS) += vpx/exports_dec
@@ -124,8 +121,20 @@ INSTALL-LIBS-$(CONFIG_SHARED) += $(foreach p,$(VS_PLATFORMS),$(LIBSUBDIR)/$(p)/v
 INSTALL-LIBS-$(CONFIG_SHARED) += $(foreach p,$(VS_PLATFORMS),$(LIBSUBDIR)/$(p)/vpx.exp)
 endif
 else
-INSTALL-LIBS-$(CONFIG_STATIC) += $(LIBSUBDIR)/libvpx.a
+INSTALL-LIBS-yes += $(LIBSUBDIR)/libvpx.a
 INSTALL-LIBS-$(CONFIG_DEBUG_LIBS) += $(LIBSUBDIR)/libvpx_g.a
+
+#Install the OpenCL kernels if CL enabled.
+ifeq ($(CONFIG_OPENCL),yes)
+INSTALL-LIBS-yes += $(LIBSUBDIR)/vp8/common/opencl/filter_cl.cl
+INSTALL-LIBS-yes += $(LIBSUBDIR)/vp8/common/opencl/idctllm_cl.cl
+INSTALL-LIBS-yes += $(LIBSUBDIR)/vp8/common/opencl/loopfilter.cl
+#only install decoder CL files if VP8 decoder enabled
+ifeq ($(CONFIG_VP8_DECODER),yes)
+INSTALL-LIBS-yes += $(LIBSUBDIR)/vp8/decoder/opencl/dequantize_cl.cl
+endif
+endif #CONFIG_OPENCL=yes
+
 endif
 
 CODEC_SRCS=$(call enabled,CODEC_SRCS)
@@ -180,15 +189,14 @@ endif
 else
 LIBVPX_OBJS=$(call objs,$(CODEC_SRCS))
 OBJS-$(BUILD_LIBVPX) += $(LIBVPX_OBJS)
-LIBS-$(if $(BUILD_LIBVPX),$(CONFIG_STATIC)) += $(BUILD_PFX)libvpx.a $(BUILD_PFX)libvpx_g.a
+LIBS-$(BUILD_LIBVPX) += $(BUILD_PFX)libvpx.a $(BUILD_PFX)libvpx_g.a
 $(BUILD_PFX)libvpx_g.a: $(LIBVPX_OBJS)
 
 BUILD_LIBVPX_SO         := $(if $(BUILD_LIBVPX),$(CONFIG_SHARED))
 LIBVPX_SO               := libvpx.so.$(VERSION_MAJOR).$(VERSION_MINOR).$(VERSION_PATCH)
-LIBS-$(BUILD_LIBVPX_SO) += $(BUILD_PFX)$(LIBVPX_SO)\
-                           $(notdir $(LIBVPX_SO_SYMLINKS))
+LIBS-$(BUILD_LIBVPX_SO) += $(BUILD_PFX)$(LIBVPX_SO)
 $(BUILD_PFX)$(LIBVPX_SO): $(LIBVPX_OBJS) libvpx.ver
-$(BUILD_PFX)$(LIBVPX_SO): extralibs += -lm
+$(BUILD_PFX)$(LIBVPX_SO): extralibs += -lm -pthread
 $(BUILD_PFX)$(LIBVPX_SO): SONAME = libvpx.so.$(VERSION_MAJOR)
 $(BUILD_PFX)$(LIBVPX_SO): SO_VERSION_SCRIPT = libvpx.ver
 LIBVPX_SO_SYMLINKS      := $(addprefix $(LIBSUBDIR)/, \
@@ -202,18 +210,9 @@ libvpx.ver: $(call enabled,CODEC_EXPORTS)
 	$(qexec)echo "local: *; };" >> $@
 CLEAN-OBJS += libvpx.ver
 
-define libvpx_symlink_template
-$(1): $(2)
-	@echo "    [LN]      $$@"
-	$(qexec)ln -sf $(LIBVPX_SO) $$@
-endef
-
-$(eval $(call libvpx_symlink_template,\
-    $(addprefix $(BUILD_PFX),$(notdir $(LIBVPX_SO_SYMLINKS))),\
-    $(BUILD_PFX)$(LIBVPX_SO)))
-$(eval $(call libvpx_symlink_template,\
-    $(addprefix $(DIST_DIR)/,$(LIBVPX_SO_SYMLINKS)),\
-    $(DIST_DIR)/$(LIBSUBDIR)/$(LIBVPX_SO)))
+$(addprefix $(DIST_DIR)/,$(LIBVPX_SO_SYMLINKS)):
+	@echo "    [LN]      $@"
+	$(qexec)ln -sf $(LIBVPX_SO) $@
 
 INSTALL-LIBS-$(CONFIG_SHARED) += $(LIBVPX_SO_SYMLINKS)
 INSTALL-LIBS-$(CONFIG_SHARED) += $(LIBSUBDIR)/$(LIBVPX_SO)
@@ -270,38 +269,29 @@ $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)vpx_config.asm
 #
 # Calculate platform- and compiler-specific offsets for hand coded assembly
 #
-
-ifeq ($(filter icc gcc,$(TGT_CC)), $(TGT_CC))
-    $(BUILD_PFX)asm_com_offsets.asm: $(BUILD_PFX)$(VP8_PREFIX)common/asm_com_offsets.c.S
-	grep EQU $< | tr -d '$$\#' $(ADS2GAS) > $@
-    $(BUILD_PFX)$(VP8_PREFIX)common/asm_com_offsets.c.S: $(VP8_PREFIX)common/asm_com_offsets.c
-    CLEAN-OBJS += $(BUILD_PFX)asm_com_offsets.asm $(BUILD_PFX)$(VP8_PREFIX)common/asm_com_offsets.c.S
-
-    $(BUILD_PFX)asm_enc_offsets.asm: $(BUILD_PFX)$(VP8_PREFIX)encoder/asm_enc_offsets.c.S
-	grep EQU $< | tr -d '$$\#' $(ADS2GAS) > $@
-    $(BUILD_PFX)$(VP8_PREFIX)encoder/asm_enc_offsets.c.S: $(VP8_PREFIX)encoder/asm_enc_offsets.c
-    CLEAN-OBJS += $(BUILD_PFX)asm_enc_offsets.asm $(BUILD_PFX)$(VP8_PREFIX)encoder/asm_enc_offsets.c.S
-
-    $(BUILD_PFX)asm_dec_offsets.asm: $(BUILD_PFX)$(VP8_PREFIX)decoder/asm_dec_offsets.c.S
-	grep EQU $< | tr -d '$$\#' $(ADS2GAS) > $@
-    $(BUILD_PFX)$(VP8_PREFIX)decoder/asm_dec_offsets.c.S: $(VP8_PREFIX)decoder/asm_dec_offsets.c
-    CLEAN-OBJS += $(BUILD_PFX)asm_dec_offsets.asm $(BUILD_PFX)$(VP8_PREFIX)decoder/asm_dec_offsets.c.S
-else
-  ifeq ($(filter rvct,$(TGT_CC)), $(TGT_CC))
+ifeq ($(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
+  endif
 
+  ifeq ($(ARCH_ARM)$(ARCH_X86)$(ARCH_X86_64), yes)
+    ifeq ($(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
+  endif
 
+  ifeq ($(ARCH_ARM), yes)
+    ifeq ($(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) > $@
@@ -310,6 +300,7 @@ else
       $(filter %$(ASM).o,$(OBJS-yes)): $(BUILD_PFX)asm_dec_offsets.asm
     endif
   endif
+endif
 
 $(shell $(SRC_PATH_BARE)/build/make/version.sh "$(SRC_PATH_BARE)" $(BUILD_PFX)vpx_version.h)
 CLEAN-OBJS += $(BUILD_PFX)vpx_version.h

vp8/common/alloccommon.c

@@ -27,9 +27,6 @@ static void update_mode_info_border(MODE_INFO *mi, int rows, int cols)
 
     for (i = 0; i < rows; i++)
     {
-        /* TODO(holmer): Bug? This updates the last element of each row
-         * rather than the border element!
-         */
         vpx_memset(&mi[i*cols-1], 0, sizeof(MODE_INFO));
     }
 }
@@ -46,11 +43,9 @@ void vp8_de_alloc_frame_buffers(VP8_COMMON *oci)
 
     vpx_free(oci->above_context);
     vpx_free(oci->mip);
-    vpx_free(oci->prev_mip);
 
     oci->above_context = 0;
     oci->mip = 0;
-    oci->prev_mip = 0;
 
 }
 
@@ -70,8 +65,8 @@ int vp8_alloc_frame_buffers(VP8_COMMON *oci, int width, int height)
 
     for (i = 0; i < NUM_YV12_BUFFERS; i++)
     {
-        oci->fb_idx_ref_cnt[i] = 0;
-        oci->yv12_fb[i].flags = 0;
+      oci->fb_idx_ref_cnt[0] = 0;
+
       if (vp8_yv12_alloc_frame_buffer(&oci->yv12_fb[i],  width, height, VP8BORDERINPIXELS) < 0)
         {
             vp8_de_alloc_frame_buffers(oci);
@@ -115,21 +110,6 @@ int vp8_alloc_frame_buffers(VP8_COMMON *oci, int width, int height)
 
     oci->mi = oci->mip + oci->mode_info_stride + 1;
 
-    /* allocate memory for last frame MODE_INFO array */
-#if CONFIG_ERROR_CONCEALMENT
-    oci->prev_mip = vpx_calloc((oci->mb_cols + 1) * (oci->mb_rows + 1), sizeof(MODE_INFO));
-
-    if (!oci->prev_mip)
-    {
-        vp8_de_alloc_frame_buffers(oci);
-        return 1;
-    }
-
-    oci->prev_mi = oci->prev_mip + oci->mode_info_stride + 1;
-#else
-    oci->prev_mip = NULL;
-    oci->prev_mi = NULL;
-#endif
 
     oci->above_context = vpx_calloc(sizeof(ENTROPY_CONTEXT_PLANES) * oci->mb_cols, 1);
 
@@ -140,9 +120,6 @@ int vp8_alloc_frame_buffers(VP8_COMMON *oci, int width, int height)
     }
 
     update_mode_info_border(oci->mi, oci->mb_rows, oci->mb_cols);
-#if CONFIG_ERROR_CONCEALMENT
-    update_mode_info_border(oci->prev_mi, oci->mb_rows, oci->mb_cols);
-#endif
 
     return 0;
 }
@@ -152,33 +129,33 @@ void vp8_setup_version(VP8_COMMON *cm)
     {
     case 0:
         cm->no_lpf = 0;
-        cm->filter_type = NORMAL_LOOPFILTER;
-        cm->use_bilinear_mc_filter = 0;
+        cm->simpler_lpf = 0;
+        cm->mcomp_filter_type = SIXTAP;
         cm->full_pixel = 0;
         break;
     case 1:
         cm->no_lpf = 0;
-        cm->filter_type = SIMPLE_LOOPFILTER;
-        cm->use_bilinear_mc_filter = 1;
+        cm->simpler_lpf = 1;
+        cm->mcomp_filter_type = BILINEAR;
         cm->full_pixel = 0;
         break;
     case 2:
         cm->no_lpf = 1;
-        cm->filter_type = NORMAL_LOOPFILTER;
-        cm->use_bilinear_mc_filter = 1;
+        cm->simpler_lpf = 0;
+        cm->mcomp_filter_type = BILINEAR;
         cm->full_pixel = 0;
         break;
     case 3:
         cm->no_lpf = 1;
-        cm->filter_type = SIMPLE_LOOPFILTER;
-        cm->use_bilinear_mc_filter = 1;
+        cm->simpler_lpf = 1;
+        cm->mcomp_filter_type = BILINEAR;
         cm->full_pixel = 1;
         break;
     default:
         /*4,5,6,7 are reserved for future use*/
         cm->no_lpf = 0;
-        cm->filter_type = NORMAL_LOOPFILTER;
-        cm->use_bilinear_mc_filter = 0;
+        cm->simpler_lpf = 0;
+        cm->mcomp_filter_type = SIXTAP;
         cm->full_pixel = 0;
         break;
     }
@@ -192,8 +169,8 @@ void vp8_create_common(VP8_COMMON *oci)
 
     oci->mb_no_coeff_skip = 1;
     oci->no_lpf = 0;
-    oci->filter_type = NORMAL_LOOPFILTER;
-    oci->use_bilinear_mc_filter = 0;
+    oci->simpler_lpf = 0;
+    oci->mcomp_filter_type = SIXTAP;
     oci->full_pixel = 0;
     oci->multi_token_partition = ONE_PARTITION;
     oci->clr_type = REG_YUV;

vp8/common/arm/arm_systemdependent.c

@@ -24,17 +24,14 @@ void vp8_arch_arm_common_init(VP8_COMMON *ctx)
 #if CONFIG_RUNTIME_CPU_DETECT
     VP8_COMMON_RTCD *rtcd = &ctx->rtcd;
     int flags = arm_cpu_caps();
+    int has_edsp = flags & HAS_EDSP;
+    int has_media = flags & HAS_MEDIA;
+    int has_neon = flags & HAS_NEON;
     rtcd->flags = flags;
 
     /* Override default functions with fastest ones for this CPU. */
-#if HAVE_ARMV5TE
-    if (flags & HAS_EDSP)
-    {
-    }
-#endif
-
 #if HAVE_ARMV6
-    if (flags & HAS_MEDIA)
+    if (has_media)
     {
         rtcd->subpix.sixtap16x16   = vp8_sixtap_predict16x16_armv6;
         rtcd->subpix.sixtap8x8     = vp8_sixtap_predict8x8_armv6;
@@ -54,11 +51,9 @@ void vp8_arch_arm_common_init(VP8_COMMON *ctx)
         rtcd->loopfilter.normal_b_v  = vp8_loop_filter_bv_armv6;
         rtcd->loopfilter.normal_mb_h = vp8_loop_filter_mbh_armv6;
         rtcd->loopfilter.normal_b_h  = vp8_loop_filter_bh_armv6;
-        rtcd->loopfilter.simple_mb_v =
-                vp8_loop_filter_simple_vertical_edge_armv6;
+        rtcd->loopfilter.simple_mb_v = vp8_loop_filter_mbvs_armv6;
         rtcd->loopfilter.simple_b_v  = vp8_loop_filter_bvs_armv6;
-        rtcd->loopfilter.simple_mb_h =
-                vp8_loop_filter_simple_horizontal_edge_armv6;
+        rtcd->loopfilter.simple_mb_h = vp8_loop_filter_mbhs_armv6;
         rtcd->loopfilter.simple_b_h  = vp8_loop_filter_bhs_armv6;
 
         rtcd->recon.copy16x16   = vp8_copy_mem16x16_v6;
@@ -71,7 +66,7 @@ void vp8_arch_arm_common_init(VP8_COMMON *ctx)
 #endif
 
 #if HAVE_ARMV7
-    if (flags & HAS_NEON)
+    if (has_neon)
     {
         rtcd->subpix.sixtap16x16   = vp8_sixtap_predict16x16_neon;
         rtcd->subpix.sixtap8x8     = vp8_sixtap_predict8x8_neon;

vp8/common/arm/armv6/bilinearfilter_v6.asm

@@ -30,12 +30,12 @@
     ldr     r4, [sp, #36]                   ; width
 
     mov     r12, r3                         ; outer-loop counter
-
-    add     r7, r2, r4                      ; preload next row
-    pld     [r0, r7]
-
     sub     r2, r2, r4                      ; src increment for height loop
 
+    ;;IF ARCHITECTURE=6
+    pld     [r0]
+    ;;ENDIF
+
     ldr     r5, [r11]                       ; load up filter coefficients
 
     mov     r3, r3, lsl #1                  ; height*2
@@ -96,8 +96,9 @@
     add     r0, r0, r2                      ; move to next input row
     subs    r12, r12, #1
 
-    add     r9, r2, r4, lsl #1              ; adding back block width
-    pld     [r0, r9]                        ; preload next row
+    ;;IF ARCHITECTURE=6
+    pld     [r0]
+    ;;ENDIF
 
     add     r11, r11, #2                    ; move over to next column
     mov     r1, r11

vp8/common/arm/armv6/copymem16x16_v6.asm

@@ -22,7 +22,9 @@
     ;push   {r4-r7}
 
     ;preload
-    pld     [r0, #31]                ; preload for next 16x16 block
+    pld     [r0]
+    pld     [r0, r1]
+    pld     [r0, r1, lsl #1]
 
     ands    r4, r0, #15
     beq     copy_mem16x16_fast
@@ -88,8 +90,6 @@ copy_mem16x16_1_loop
     ldrneb  r6, [r0, #2]
     ldrneb  r7, [r0, #3]
 
-    pld     [r0, #31]               ; preload for next 16x16 block
-
     bne     copy_mem16x16_1_loop
 
     ldmia       sp!, {r4 - r7}
@@ -121,8 +121,6 @@ copy_mem16x16_4_loop
     ldrne   r6, [r0, #8]
     ldrne   r7, [r0, #12]
 
-    pld     [r0, #31]               ; preload for next 16x16 block
-
     bne     copy_mem16x16_4_loop
 
     ldmia       sp!, {r4 - r7}
@@ -150,7 +148,6 @@ copy_mem16x16_8_loop
 
     add     r2, r2, r3
 
-    pld     [r0, #31]               ; preload for next 16x16 block
     bne     copy_mem16x16_8_loop
 
     ldmia       sp!, {r4 - r7}
@@ -174,7 +171,6 @@ copy_mem16x16_fast_loop
     ;stm        r2, {r4-r7}
     add     r2, r2, r3
 
-    pld     [r0, #31]               ; preload for next 16x16 block
     bne     copy_mem16x16_fast_loop
 
     ldmia       sp!, {r4 - r7}

vp8/common/arm/armv6/filter_v6.asm

@@ -10,8 +10,6 @@
 
 
     EXPORT  |vp8_filter_block2d_first_pass_armv6|
-    EXPORT  |vp8_filter_block2d_first_pass_16x16_armv6|
-    EXPORT  |vp8_filter_block2d_first_pass_8x8_armv6|
     EXPORT  |vp8_filter_block2d_second_pass_armv6|
     EXPORT  |vp8_filter4_block2d_second_pass_armv6|
     EXPORT  |vp8_filter_block2d_first_pass_only_armv6|
@@ -42,6 +40,11 @@
     add     r12, r3, #16                    ; square off the output
     sub     sp, sp, #4
 
+    ;;IF ARCHITECTURE=6
+    ;pld        [r0, #-2]
+    ;;pld       [r0, #30]
+    ;;ENDIF
+
     ldr     r4, [r11]                       ; load up packed filter coefficients
     ldr     r5, [r11, #4]
     ldr     r6, [r11, #8]
@@ -98,10 +101,15 @@
 
     bne     width_loop_1st_6
 
+    ;;add       r9, r2, #30                 ; attempt to load 2 adjacent cache lines
+    ;;IF ARCHITECTURE=6
+    ;pld        [r0, r2]
+    ;;pld       [r0, r9]
+    ;;ENDIF
+
     ldr     r1, [sp]                        ; load and update dst address
     subs    r7, r7, #0x10000
     add     r0, r0, r2                      ; move to next input line
-
     add     r1, r1, #2                      ; move over to next column
     str     r1, [sp]
 
@@ -112,192 +120,6 @@
 
     ENDP
 
-; --------------------------
-; 16x16 version
-; -----------------------------
-|vp8_filter_block2d_first_pass_16x16_armv6| PROC
-    stmdb   sp!, {r4 - r11, lr}
-
-    ldr     r11, [sp, #40]                  ; vp8_filter address
-    ldr     r7, [sp, #36]                   ; output height
-
-    add     r4, r2, #18                     ; preload next low
-    pld     [r0, r4]
-
-    sub     r2, r2, r3                      ; inside loop increments input array,
-                                            ; so the height loop only needs to add
-                                            ; r2 - width to the input pointer
-
-    mov     r3, r3, lsl #1                  ; multiply width by 2 because using shorts
-    add     r12, r3, #16                    ; square off the output
-    sub     sp, sp, #4
-
-    ldr     r4, [r11]                       ; load up packed filter coefficients
-    ldr     r5, [r11, #4]
-    ldr     r6, [r11, #8]
-
-    str     r1, [sp]                        ; push destination to stack
-    mov     r7, r7, lsl #16                 ; height is top part of counter
-
-; six tap filter
-|height_loop_1st_16_6|
-    ldrb    r8, [r0, #-2]                   ; load source data
-    ldrb    r9, [r0, #-1]
-    ldrb    r10, [r0], #2
-    orr     r7, r7, r3, lsr #2              ; construct loop counter
-
-|width_loop_1st_16_6|
-    ldrb    r11, [r0, #-1]
-
-    pkhbt   lr, r8, r9, lsl #16             ; r9 | r8
-    pkhbt   r8, r9, r10, lsl #16            ; r10 | r9
-
-    ldrb    r9, [r0]
-
-    smuad   lr, lr, r4                      ; apply the filter
-    pkhbt   r10, r10, r11, lsl #16          ; r11 | r10
-    smuad   r8, r8, r4
-    pkhbt   r11, r11, r9, lsl #16           ; r9 | r11
-
-    smlad   lr, r10, r5, lr
-    ldrb    r10, [r0, #1]
-    smlad   r8, r11, r5, r8
-    ldrb    r11, [r0, #2]
-
-    sub     r7, r7, #1
-
-    pkhbt   r9, r9, r10, lsl #16            ; r10 | r9
-    pkhbt   r10, r10, r11, lsl #16          ; r11 | r10
-
-    smlad   lr, r9, r6, lr
-    smlad   r11, r10, r6, r8
-
-    ands    r10, r7, #0xff                  ; test loop counter
-
-    add     lr, lr, #0x40                   ; round_shift_and_clamp
-    ldrneb  r8, [r0, #-2]                   ; load data for next loop
-    usat    lr, #8, lr, asr #7
-    add     r11, r11, #0x40
-    ldrneb  r9, [r0, #-1]
-    usat    r11, #8, r11, asr #7
-
-    strh    lr, [r1], r12                   ; result is transposed and stored, which
-                                            ; will make second pass filtering easier.
-    ldrneb  r10, [r0], #2
-    strh    r11, [r1], r12
-
-    bne     width_loop_1st_16_6
-
-    ldr     r1, [sp]                        ; load and update dst address
-    subs    r7, r7, #0x10000
-    add     r0, r0, r2                      ; move to next input line
-
-    add     r11, r2, #34                    ; adding back block width(=16)
-    pld     [r0, r11]                       ; preload next low
-
-    add     r1, r1, #2                      ; move over to next column
-    str     r1, [sp]
-
-    bne     height_loop_1st_16_6
-
-    add     sp, sp, #4
-    ldmia   sp!, {r4 - r11, pc}
-
-    ENDP
-
-; --------------------------
-; 8x8 version
-; -----------------------------
-|vp8_filter_block2d_first_pass_8x8_armv6| PROC
-    stmdb   sp!, {r4 - r11, lr}
-
-    ldr     r11, [sp, #40]                  ; vp8_filter address
-    ldr     r7, [sp, #36]                   ; output height
-
-    add     r4, r2, #10                     ; preload next low
-    pld     [r0, r4]
-
-    sub     r2, r2, r3                      ; inside loop increments input array,
-                                            ; so the height loop only needs to add
-                                            ; r2 - width to the input pointer
-
-    mov     r3, r3, lsl #1                  ; multiply width by 2 because using shorts
-    add     r12, r3, #16                    ; square off the output
-    sub     sp, sp, #4
-
-    ldr     r4, [r11]                       ; load up packed filter coefficients
-    ldr     r5, [r11, #4]
-    ldr     r6, [r11, #8]
-
-    str     r1, [sp]                        ; push destination to stack
-    mov     r7, r7, lsl #16                 ; height is top part of counter
-
-; six tap filter
-|height_loop_1st_8_6|
-    ldrb    r8, [r0, #-2]                   ; load source data
-    ldrb    r9, [r0, #-1]
-    ldrb    r10, [r0], #2
-    orr     r7, r7, r3, lsr #2              ; construct loop counter
-
-|width_loop_1st_8_6|
-    ldrb    r11, [r0, #-1]
-
-    pkhbt   lr, r8, r9, lsl #16             ; r9 | r8
-    pkhbt   r8, r9, r10, lsl #16            ; r10 | r9
-
-    ldrb    r9, [r0]
-
-    smuad   lr, lr, r4                      ; apply the filter
-    pkhbt   r10, r10, r11, lsl #16          ; r11 | r10
-    smuad   r8, r8, r4
-    pkhbt   r11, r11, r9, lsl #16           ; r9 | r11
-
-    smlad   lr, r10, r5, lr
-    ldrb    r10, [r0, #1]
-    smlad   r8, r11, r5, r8
-    ldrb    r11, [r0, #2]
-
-    sub     r7, r7, #1
-
-    pkhbt   r9, r9, r10, lsl #16            ; r10 | r9
-    pkhbt   r10, r10, r11, lsl #16          ; r11 | r10
-
-    smlad   lr, r9, r6, lr
-    smlad   r11, r10, r6, r8
-
-    ands    r10, r7, #0xff                  ; test loop counter
-
-    add     lr, lr, #0x40                   ; round_shift_and_clamp
-    ldrneb  r8, [r0, #-2]                   ; load data for next loop
-    usat    lr, #8, lr, asr #7
-    add     r11, r11, #0x40
-    ldrneb  r9, [r0, #-1]
-    usat    r11, #8, r11, asr #7
-
-    strh    lr, [r1], r12                   ; result is transposed and stored, which
-                                            ; will make second pass filtering easier.
-    ldrneb  r10, [r0], #2
-    strh    r11, [r1], r12
-
-    bne     width_loop_1st_8_6
-
-    ldr     r1, [sp]                        ; load and update dst address
-    subs    r7, r7, #0x10000
-    add     r0, r0, r2                      ; move to next input line
-
-    add     r11, r2, #18                    ; adding back block width(=8)
-    pld     [r0, r11]                       ; preload next low
-
-    add     r1, r1, #2                      ; move over to next column
-    str     r1, [sp]
-
-    bne     height_loop_1st_8_6
-
-    add     sp, sp, #4
-    ldmia   sp!, {r4 - r11, pc}
-
-    ENDP
-
 ;---------------------------------
 ; r0    short         *src_ptr,
 ; r1    unsigned char *output_ptr,
@@ -440,10 +262,6 @@
 |vp8_filter_block2d_first_pass_only_armv6| PROC
     stmdb   sp!, {r4 - r11, lr}
 
-    add     r7, r2, r3                      ; preload next low
-    add     r7, r7, #2
-    pld     [r0, r7]
-
     ldr     r4, [sp, #36]                   ; output pitch
     ldr     r11, [sp, #40]                  ; HFilter address
     sub     sp, sp, #8
@@ -512,15 +330,16 @@
 
     bne     width_loop_1st_only_6
 
+    ;;add       r9, r2, #30                 ; attempt to load 2 adjacent cache lines
+    ;;IF ARCHITECTURE=6
+    ;pld        [r0, r2]
+    ;;pld       [r0, r9]
+    ;;ENDIF
+
     ldr     lr, [sp]                        ; load back output pitch
     ldr     r12, [sp, #4]                   ; load back output pitch
     subs    r7, r7, #1
     add     r0, r0, r12                     ; updata src for next loop
-
-    add     r11, r12, r3                    ; preload next low
-    add     r11, r11, #2
-    pld     [r0, r11]
-
     add     r1, r1, lr                      ; update dst for next loop
 
     bne     height_loop_1st_only_6

vp8/common/arm/armv6/loopfilter_v6.asm

@@ -53,11 +53,14 @@ count       RN  r5
 
 ;r0     unsigned char *src_ptr,
 ;r1     int src_pixel_step,
-;r2     const char *blimit,
+;r2     const char *flimit,
 ;r3     const char *limit,
 ;stack  const char *thresh,
 ;stack  int  count
 
+;Note: All 16 elements in flimit are equal. So, in the code, only one load is needed
+;for flimit. Same way applies to limit and thresh.
+
 ;-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
 |vp8_loop_filter_horizontal_edge_armv6| PROC
 ;-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
@@ -69,18 +72,14 @@ count       RN  r5
     sub         sp, sp, #16                 ; create temp buffer
 
     ldr         r9, [src], pstep            ; p3
-    ldrb        r4, [r2]                    ; blimit
+    ldr         r4, [r2], #4                ; flimit
     ldr         r10, [src], pstep           ; p2
-    ldrb        r2, [r3]                    ; limit
+    ldr         r2, [r3], #4                ; limit
     ldr         r11, [src], pstep           ; p1
-    orr         r4, r4, r4, lsl #8
-    ldrb        r3, [r6]                    ; thresh
-    orr         r2, r2, r2, lsl #8
+    uadd8       r4, r4, r4                  ; flimit * 2
+    ldr         r3, [r6], #4                ; thresh
     mov         count, count, lsl #1        ; 4-in-parallel
-    orr         r4, r4, r4, lsl #16
-    orr         r3, r3, r3, lsl #8
-    orr         r2, r2, r2, lsl #16
-    orr         r3, r3, r3, lsl #16
+    uadd8       r4, r4, r2                  ; flimit * 2 + limit
 
 |Hnext8|
     ; vp8_filter_mask() function
@@ -254,6 +253,12 @@ count       RN  r5
 
     subs        count, count, #1
 
+    ;pld            [src]
+    ;pld            [src, pstep]
+    ;pld            [src, pstep, lsl #1]
+    ;pld            [src, pstep, lsl #2]
+    ;pld            [src, pstep, lsl #3]
+
     ldrne       r9, [src], pstep            ; p3
     ldrne       r10, [src], pstep           ; p2
     ldrne       r11, [src], pstep           ; p1
@@ -276,18 +281,14 @@ count       RN  r5
     sub         sp, sp, #16                 ; create temp buffer
 
     ldr         r9, [src], pstep            ; p3
-    ldrb        r4, [r2]                    ; blimit
+    ldr         r4, [r2], #4                ; flimit
     ldr         r10, [src], pstep           ; p2
-    ldrb        r2, [r3]                    ; limit
+    ldr         r2, [r3], #4                ; limit
     ldr         r11, [src], pstep           ; p1
-    orr         r4, r4, r4, lsl #8
-    ldrb        r3, [r6]                    ; thresh
-    orr         r2, r2, r2, lsl #8
+    uadd8       r4, r4, r4                  ; flimit * 2
+    ldr         r3, [r6], #4                ; thresh
     mov         count, count, lsl #1        ; 4-in-parallel
-    orr         r4, r4, r4, lsl #16
-    orr         r3, r3, r3, lsl #8
-    orr         r2, r2, r2, lsl #16
-    orr         r3, r3, r3, lsl #16
+    uadd8       r4, r4, r2                  ; flimit * 2 + limit
 
 |MBHnext8|
 
@@ -589,19 +590,15 @@ count       RN  r5
     sub         sp, sp, #16                 ; create temp buffer
 
     ldr         r6, [src], pstep            ; load source data
-    ldrb        r4, [r2]                    ; blimit
+    ldr         r4, [r2], #4                ; flimit
     ldr         r7, [src], pstep
-    ldrb        r2, [r3]                    ; limit
+    ldr         r2, [r3], #4                ; limit
     ldr         r8, [src], pstep
-    orr         r4, r4, r4, lsl #8
-    ldrb        r3, [r12]                   ; thresh
-    orr         r2, r2, r2, lsl #8
+    uadd8       r4, r4, r4                  ; flimit * 2
+    ldr         r3, [r12], #4               ; thresh
     ldr         lr, [src], pstep
     mov         count, count, lsl #1        ; 4-in-parallel
-    orr         r4, r4, r4, lsl #16
-    orr         r3, r3, r3, lsl #8
-    orr         r2, r2, r2, lsl #16
-    orr         r3, r3, r3, lsl #16
+    uadd8       r4, r4, r2                  ; flimit * 2 + limit
 
 |Vnext8|
 
@@ -860,26 +857,18 @@ count       RN  r5
     sub         src, src, #4                ; move src pointer down by 4
     ldr         count, [sp, #40]            ; count for 8-in-parallel
     ldr         r12, [sp, #36]              ; load thresh address
-    pld         [src, #23]                  ; preload for next block
     sub         sp, sp, #16                 ; create temp buffer
 
     ldr         r6, [src], pstep            ; load source data
-    ldrb        r4, [r2]                    ; blimit
-    pld         [src, #23]
+    ldr         r4, [r2], #4                ; flimit
     ldr         r7, [src], pstep
-    ldrb        r2, [r3]                    ; limit
-    pld         [src, #23]
+    ldr         r2, [r3], #4                ; limit
     ldr         r8, [src], pstep
-    orr         r4, r4, r4, lsl #8
-    ldrb        r3, [r12]                   ; thresh
-    orr         r2, r2, r2, lsl #8
-    pld         [src, #23]
+    uadd8       r4, r4, r4                  ; flimit * 2
+    ldr         r3, [r12], #4               ; thresh
     ldr         lr, [src], pstep
     mov         count, count, lsl #1        ; 4-in-parallel
-    orr         r4, r4, r4, lsl #16
-    orr         r3, r3, r3, lsl #8
-    orr         r2, r2, r2, lsl #16
-    orr         r3, r3, r3, lsl #16
+    uadd8       r4, r4, r2                  ; flimit * 2 + limit
 
 |MBVnext8|
     ; vp8_filter_mask() function
@@ -919,7 +908,6 @@ count       RN  r5
     str         lr, [sp, #8]
     ldr         lr, [src], pstep
 
-
     TRANSPOSE_MATRIX r6, r7, r8, lr, r9, r10, r11, r12
 
     ldr         lr, [sp, #8]                ; load back (f)limit accumulator
@@ -968,7 +956,6 @@ count       RN  r5
     beq         mbvskip_filter               ; skip filtering
 
 
-
     ;vp8_hevmask() function
     ;calculate high edge variance
 
@@ -1136,7 +1123,6 @@ count       RN  r5
     smlabb      r8, r6, lr, r7
     smlatb      r6, r6, lr, r7
     smlabb      r9, r10, lr, r7
-
     smlatb      r10, r10, lr, r7
     ssat        r8, #8, r8, asr #7
     ssat        r6, #8, r6, asr #7
@@ -1256,13 +1242,9 @@ count       RN  r5
     sub         src, src, #4
     subs        count, count, #1
 
-    pld         [src, #23]                  ; preload for next block
     ldrne       r6, [src], pstep            ; load source data
-    pld         [src, #23]
     ldrne       r7, [src], pstep
-    pld         [src, #23]
     ldrne       r8, [src], pstep
-    pld         [src, #23]
     ldrne       lr, [src], pstep
 
     bne         MBVnext8

vp8/common/arm/armv6/simpleloopfilter_v6.asm

@@ -45,28 +45,35 @@
     MEND
 
 
-
 src         RN  r0
 pstep       RN  r1
 
 ;r0     unsigned char *src_ptr,
 ;r1     int src_pixel_step,
-;r2     const char *blimit
+;r2     const char *flimit,
+;r3     const char *limit,
+;stack  const char *thresh,
+;stack  int  count
+
+; All 16 elements in flimit are equal. So, in the code, only one load is needed
+; for flimit. Same applies to limit. thresh is not used in simple looopfilter
 
 ;-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
 |vp8_loop_filter_simple_horizontal_edge_armv6| PROC
 ;-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
     stmdb       sp!, {r4 - r11, lr}
 
-    ldrb        r12, [r2]                   ; blimit
+    ldr         r12, [r3]                   ; limit
     ldr         r3, [src, -pstep, lsl #1]   ; p1
     ldr         r4, [src, -pstep]           ; p0
     ldr         r5, [src]                   ; q0
     ldr         r6, [src, pstep]            ; q1
-    orr         r12, r12, r12, lsl #8       ; blimit
+    ldr         r7, [r2]                    ; flimit
     ldr         r2, c0x80808080
-    orr         r12, r12, r12, lsl #16      ; blimit
-    mov         r9, #4                      ; double the count. we're doing 4 at a time
+    ldr         r9, [sp, #40]               ; count for 8-in-parallel
+    uadd8       r7, r7, r7                  ; flimit * 2
+    mov         r9, r9, lsl #1              ; double the count. we're doing 4 at a time
+    uadd8       r12, r7, r12                ; flimit * 2 + limit
     mov         lr, #0                      ; need 0 in a couple places
 
 |simple_hnext8|
@@ -141,32 +148,30 @@ pstep       RN  r1
 ;-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
     stmdb       sp!, {r4 - r11, lr}
 
-    ldrb        r12, [r2]                   ; r12: blimit
+    ldr         r12, [r2]                   ; r12: flimit
     ldr         r2, c0x80808080
-    orr         r12, r12, r12, lsl #8
+    ldr         r7, [r3]                    ; limit
 
     ; load soure data to r7, r8, r9, r10
     ldrh        r3, [src, #-2]
-    pld         [src, #23]                  ; preload for next block
     ldrh        r4, [src], pstep
-    orr         r12, r12, r12, lsl #16
+    uadd8       r12, r12, r12               ; flimit * 2
 
     ldrh        r5, [src, #-2]
-    pld         [src, #23]
     ldrh        r6, [src], pstep
+    uadd8       r12, r12, r7                ; flimit * 2 + limit
 
     pkhbt       r7, r3, r4, lsl #16
 
     ldrh        r3, [src, #-2]
-    pld         [src, #23]
     ldrh        r4, [src], pstep
+    ldr         r11, [sp, #40]              ; count (r11) for 8-in-parallel
 
     pkhbt       r8, r5, r6, lsl #16
 
     ldrh        r5, [src, #-2]
-    pld         [src, #23]
     ldrh        r6, [src], pstep
-    mov         r11, #4                     ; double the count. we're doing 4 at a time
+    mov         r11, r11, lsl #1            ; 4-in-parallel
 
 |simple_vnext8|
     ; vp8_simple_filter_mask() function
@@ -254,23 +259,19 @@ pstep       RN  r1
 
     ; load soure data to r7, r8, r9, r10
     ldrneh      r3, [src, #-2]
-    pld         [src, #23]                  ; preload for next block
     ldrneh      r4, [src], pstep
 
     ldrneh      r5, [src, #-2]
-    pld         [src, #23]
     ldrneh      r6, [src], pstep
 
     pkhbt       r7, r3, r4, lsl #16
 
     ldrneh      r3, [src, #-2]
-    pld         [src, #23]
     ldrneh      r4, [src], pstep
 
     pkhbt       r8, r5, r6, lsl #16
 
     ldrneh      r5, [src, #-2]
-    pld         [src, #23]
     ldrneh      r6, [src], pstep
 
     bne         simple_vnext8

vp8/common/arm/armv6/sixtappredict8x4_v6.asm

@@ -32,12 +32,9 @@
     beq         skip_firstpass_filter
 
 ;first-pass filter
-    adr         r12, filter8_coeff
+    ldr         r12, _filter8_coeff_
     sub         r0, r0, r1, lsl #1
 
-    add         r3, r1, #10                 ; preload next low
-    pld         [r0, r3]
-
     add         r2, r12, r2, lsl #4         ;calculate filter location
     add         r0, r0, #3                  ;adjust src only for loading convinience
 
@@ -113,9 +110,6 @@
 
     add         r0, r0, r1                  ; move to next input line
 
-    add         r11, r1, #18                ; preload next low. adding back block width(=8), which is subtracted earlier
-    pld         [r0, r11]
-
     bne         first_pass_hloop_v6
 
 ;second pass filter
@@ -127,7 +121,7 @@ secondpass_filter
     cmp         r3, #0
     beq         skip_secondpass_filter
 
-    adr         r12, filter8_coeff
+    ldr         r12, _filter8_coeff_
     add         lr, r12, r3, lsl #4         ;calculate filter location
 
     mov         r2, #0x00080000
@@ -251,6 +245,8 @@ skip_secondpass_hloop
 ;-----------------
 ;One word each is reserved. Label filter_coeff can be used to access the data.
 ;Data address: filter_coeff, filter_coeff+4, filter_coeff+8 ...
+_filter8_coeff_
+    DCD     filter8_coeff
 filter8_coeff
     DCD     0x00000000,     0x00000080,     0x00000000,     0x00000000
     DCD     0xfffa0000,     0x000c007b,     0x0000ffff,     0x00000000

vp8/common/arm/filter_arm.c

@@ -25,28 +25,6 @@ extern void vp8_filter_block2d_first_pass_armv6
     const short *vp8_filter
 );
 
-// 8x8
-extern void vp8_filter_block2d_first_pass_8x8_armv6
-(
-    unsigned char *src_ptr,
-    short         *output_ptr,
-    unsigned int src_pixels_per_line,
-    unsigned int output_width,
-    unsigned int output_height,
-    const short *vp8_filter
-);
-
-// 16x16
-extern void vp8_filter_block2d_first_pass_16x16_armv6
-(
-    unsigned char *src_ptr,
-    short         *output_ptr,
-    unsigned int src_pixels_per_line,
-    unsigned int output_width,
-    unsigned int output_height,
-    const short *vp8_filter
-);
-
 extern void vp8_filter_block2d_second_pass_armv6
 (
     short         *src_ptr,
@@ -165,12 +143,12 @@ void vp8_sixtap_predict8x8_armv6
     {
         if (yoffset & 0x1)
         {
-            vp8_filter_block2d_first_pass_8x8_armv6(src_ptr - src_pixels_per_line, FData + 1, src_pixels_per_line, 8, 11, HFilter);
+            vp8_filter_block2d_first_pass_armv6(src_ptr - src_pixels_per_line, FData + 1, src_pixels_per_line, 8, 11, HFilter);
             vp8_filter4_block2d_second_pass_armv6(FData + 2, dst_ptr, dst_pitch, 8, VFilter);
         }
         else
         {
-            vp8_filter_block2d_first_pass_8x8_armv6(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 8, 13, HFilter);
+            vp8_filter_block2d_first_pass_armv6(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 8, 13, HFilter);
             vp8_filter_block2d_second_pass_armv6(FData + 2, dst_ptr, dst_pitch, 8, VFilter);
         }
     }
@@ -207,12 +185,12 @@ void vp8_sixtap_predict16x16_armv6
     {
         if (yoffset & 0x1)
         {
-            vp8_filter_block2d_first_pass_16x16_armv6(src_ptr - src_pixels_per_line, FData + 1, src_pixels_per_line, 16, 19, HFilter);
+            vp8_filter_block2d_first_pass_armv6(src_ptr - src_pixels_per_line, FData + 1, src_pixels_per_line, 16, 19, HFilter);
             vp8_filter4_block2d_second_pass_armv6(FData + 2, dst_ptr, dst_pitch, 16, VFilter);
         }
         else
         {
-            vp8_filter_block2d_first_pass_16x16_armv6(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 16, 21, HFilter);
+            vp8_filter_block2d_first_pass_armv6(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 16, 21, HFilter);
             vp8_filter_block2d_second_pass_armv6(FData + 2, dst_ptr, dst_pitch, 16, VFilter);
         }
     }

vp8/common/arm/loopfilter_arm.c

@@ -9,107 +9,135 @@
  */
 
 
-#include "vpx_config.h"
+#include "vpx_ports/config.h"
+#include <math.h>
 #include "vp8/common/loopfilter.h"
 #include "vp8/common/onyxc_int.h"
 
-#if HAVE_ARMV6
 extern prototype_loopfilter(vp8_loop_filter_horizontal_edge_armv6);
 extern prototype_loopfilter(vp8_loop_filter_vertical_edge_armv6);
 extern prototype_loopfilter(vp8_mbloop_filter_horizontal_edge_armv6);
 extern prototype_loopfilter(vp8_mbloop_filter_vertical_edge_armv6);
-#endif
+extern prototype_loopfilter(vp8_loop_filter_simple_horizontal_edge_armv6);
+extern prototype_loopfilter(vp8_loop_filter_simple_vertical_edge_armv6);
 
-#if HAVE_ARMV7
-typedef void loopfilter_y_neon(unsigned char *src, int pitch,
-        unsigned char blimit, unsigned char limit, unsigned char thresh);
-typedef void loopfilter_uv_neon(unsigned char *u, int pitch,
-        unsigned char blimit, unsigned char limit, unsigned char thresh,
-        unsigned char *v);
+extern prototype_loopfilter(vp8_loop_filter_horizontal_edge_y_neon);
+extern prototype_loopfilter(vp8_loop_filter_vertical_edge_y_neon);
+extern prototype_loopfilter(vp8_mbloop_filter_horizontal_edge_y_neon);
+extern prototype_loopfilter(vp8_mbloop_filter_vertical_edge_y_neon);
+extern prototype_loopfilter(vp8_loop_filter_simple_horizontal_edge_neon);
+extern prototype_loopfilter(vp8_loop_filter_simple_vertical_edge_neon);
 
-extern loopfilter_y_neon vp8_loop_filter_horizontal_edge_y_neon;
-extern loopfilter_y_neon vp8_loop_filter_vertical_edge_y_neon;
-extern loopfilter_y_neon vp8_mbloop_filter_horizontal_edge_y_neon;
-extern loopfilter_y_neon vp8_mbloop_filter_vertical_edge_y_neon;
+extern loop_filter_uvfunction vp8_loop_filter_horizontal_edge_uv_neon;
+extern loop_filter_uvfunction vp8_loop_filter_vertical_edge_uv_neon;
+extern loop_filter_uvfunction vp8_mbloop_filter_horizontal_edge_uv_neon;
+extern loop_filter_uvfunction vp8_mbloop_filter_vertical_edge_uv_neon;
 
-extern loopfilter_uv_neon vp8_loop_filter_horizontal_edge_uv_neon;
-extern loopfilter_uv_neon vp8_loop_filter_vertical_edge_uv_neon;
-extern loopfilter_uv_neon vp8_mbloop_filter_horizontal_edge_uv_neon;
-extern loopfilter_uv_neon vp8_mbloop_filter_vertical_edge_uv_neon;
-#endif
 
 #if HAVE_ARMV6
 /*ARMV6 loopfilter functions*/
 /* Horizontal MB filtering */
 void vp8_loop_filter_mbh_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                               int y_stride, int uv_stride, loop_filter_info *lfi)
+                               int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_mbloop_filter_horizontal_edge_armv6(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2);
+    (void) simpler_lpf;
+    vp8_mbloop_filter_horizontal_edge_armv6(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_mbloop_filter_horizontal_edge_armv6(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+        vp8_mbloop_filter_horizontal_edge_armv6(u_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1);
 
     if (v_ptr)
-        vp8_mbloop_filter_horizontal_edge_armv6(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+        vp8_mbloop_filter_horizontal_edge_armv6(v_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1);
+}
+
+void vp8_loop_filter_mbhs_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                                int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 }
 
 /* Vertical MB Filtering */
 void vp8_loop_filter_mbv_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                               int y_stride, int uv_stride, loop_filter_info *lfi)
+                               int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_mbloop_filter_vertical_edge_armv6(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2);
+    (void) simpler_lpf;
+    vp8_mbloop_filter_vertical_edge_armv6(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_mbloop_filter_vertical_edge_armv6(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+        vp8_mbloop_filter_vertical_edge_armv6(u_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1);
 
     if (v_ptr)
-        vp8_mbloop_filter_vertical_edge_armv6(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+        vp8_mbloop_filter_vertical_edge_armv6(v_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1);
+}
+
+void vp8_loop_filter_mbvs_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                                int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_vertical_edge_armv6(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 }
 
 /* Horizontal B Filtering */
 void vp8_loop_filter_bh_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                              int y_stride, int uv_stride, loop_filter_info *lfi)
+                              int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_loop_filter_horizontal_edge_armv6(y_ptr + 4 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
-    vp8_loop_filter_horizontal_edge_armv6(y_ptr + 8 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
-    vp8_loop_filter_horizontal_edge_armv6(y_ptr + 12 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
+    (void) simpler_lpf;
+    vp8_loop_filter_horizontal_edge_armv6(y_ptr + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_horizontal_edge_armv6(y_ptr + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_horizontal_edge_armv6(y_ptr + 12 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_loop_filter_horizontal_edge_armv6(u_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
+        vp8_loop_filter_horizontal_edge_armv6(u_ptr + 4 * uv_stride, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1);
 
     if (v_ptr)
-        vp8_loop_filter_horizontal_edge_armv6(v_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
+        vp8_loop_filter_horizontal_edge_armv6(v_ptr + 4 * uv_stride, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1);
 }
 
-void vp8_loop_filter_bhs_armv6(unsigned char *y_ptr, int y_stride,
-                               const unsigned char *blimit)
+void vp8_loop_filter_bhs_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                               int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 4 * y_stride, y_stride, blimit);
-    vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 8 * y_stride, y_stride, blimit);
-    vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 12 * y_stride, y_stride, blimit);
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_horizontal_edge_armv6(y_ptr + 12 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 }
 
 /* Vertical B Filtering */
 void vp8_loop_filter_bv_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                              int y_stride, int uv_stride, loop_filter_info *lfi)
+                              int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_loop_filter_vertical_edge_armv6(y_ptr + 4, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
-    vp8_loop_filter_vertical_edge_armv6(y_ptr + 8, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
-    vp8_loop_filter_vertical_edge_armv6(y_ptr + 12, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
+    (void) simpler_lpf;
+    vp8_loop_filter_vertical_edge_armv6(y_ptr + 4, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_vertical_edge_armv6(y_ptr + 8, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_vertical_edge_armv6(y_ptr + 12, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_loop_filter_vertical_edge_armv6(u_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
+        vp8_loop_filter_vertical_edge_armv6(u_ptr + 4, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1);
 
     if (v_ptr)
-        vp8_loop_filter_vertical_edge_armv6(v_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
+        vp8_loop_filter_vertical_edge_armv6(v_ptr + 4, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1);
 }
 
-void vp8_loop_filter_bvs_armv6(unsigned char *y_ptr, int y_stride,
-                               const unsigned char *blimit)
+void vp8_loop_filter_bvs_armv6(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                               int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 4, y_stride, blimit);
-    vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 8, y_stride, blimit);
-    vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 12, y_stride, blimit);
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 4, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 8, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_vertical_edge_armv6(y_ptr + 12, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 }
 #endif
 
@@ -117,60 +145,93 @@ void vp8_loop_filter_bvs_armv6(unsigned char *y_ptr, int y_stride,
 /* NEON loopfilter functions */
 /* Horizontal MB filtering */
 void vp8_loop_filter_mbh_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                              int y_stride, int uv_stride, loop_filter_info *lfi)
+                              int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    unsigned char mblim = *lfi->mblim;
-    unsigned char lim = *lfi->lim;
-    unsigned char hev_thr = *lfi->hev_thr;
-    vp8_mbloop_filter_horizontal_edge_y_neon(y_ptr, y_stride, mblim, lim, hev_thr);
+    (void) simpler_lpf;
+    vp8_mbloop_filter_horizontal_edge_y_neon(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_mbloop_filter_horizontal_edge_uv_neon(u_ptr, uv_stride, mblim, lim, hev_thr, v_ptr);
+        vp8_mbloop_filter_horizontal_edge_uv_neon(u_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, v_ptr);
+}
+
+void vp8_loop_filter_mbhs_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                               int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 }
 
 /* Vertical MB Filtering */
 void vp8_loop_filter_mbv_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                              int y_stride, int uv_stride, loop_filter_info *lfi)
+                              int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    unsigned char mblim = *lfi->mblim;
-    unsigned char lim = *lfi->lim;
-    unsigned char hev_thr = *lfi->hev_thr;
-
-    vp8_mbloop_filter_vertical_edge_y_neon(y_ptr, y_stride, mblim, lim, hev_thr);
+    (void) simpler_lpf;
+    vp8_mbloop_filter_vertical_edge_y_neon(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_mbloop_filter_vertical_edge_uv_neon(u_ptr, uv_stride, mblim, lim, hev_thr, v_ptr);
+        vp8_mbloop_filter_vertical_edge_uv_neon(u_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, v_ptr);
+}
+
+void vp8_loop_filter_mbvs_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                               int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 }
 
 /* Horizontal B Filtering */
 void vp8_loop_filter_bh_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                             int y_stride, int uv_stride, loop_filter_info *lfi)
+                             int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    unsigned char blim = *lfi->blim;
-    unsigned char lim = *lfi->lim;
-    unsigned char hev_thr = *lfi->hev_thr;
-
-    vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 4 * y_stride, y_stride, blim, lim, hev_thr);
-    vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 8 * y_stride, y_stride, blim, lim, hev_thr);
-    vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 12 * y_stride, y_stride, blim, lim, hev_thr);
+    (void) simpler_lpf;
+    vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_horizontal_edge_y_neon(y_ptr + 12 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_loop_filter_horizontal_edge_uv_neon(u_ptr + 4 * uv_stride, uv_stride, blim, lim, hev_thr, v_ptr + 4 * uv_stride);
+        vp8_loop_filter_horizontal_edge_uv_neon(u_ptr + 4 * uv_stride, uv_stride, lfi->flim, lfi->lim, lfi->thr, v_ptr + 4 * uv_stride);
+}
+
+void vp8_loop_filter_bhs_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                              int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_horizontal_edge_neon(y_ptr + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_horizontal_edge_neon(y_ptr + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_horizontal_edge_neon(y_ptr + 12 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 }
 
 /* Vertical B Filtering */
 void vp8_loop_filter_bv_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                             int y_stride, int uv_stride, loop_filter_info *lfi)
+                             int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    unsigned char blim = *lfi->blim;
-    unsigned char lim = *lfi->lim;
-    unsigned char hev_thr = *lfi->hev_thr;
-
-    vp8_loop_filter_vertical_edge_y_neon(y_ptr + 4, y_stride, blim, lim, hev_thr);
-    vp8_loop_filter_vertical_edge_y_neon(y_ptr + 8, y_stride, blim, lim, hev_thr);
-    vp8_loop_filter_vertical_edge_y_neon(y_ptr + 12, y_stride, blim, lim, hev_thr);
+    (void) simpler_lpf;
+    vp8_loop_filter_vertical_edge_y_neon(y_ptr + 4, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_vertical_edge_y_neon(y_ptr + 8, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_vertical_edge_y_neon(y_ptr + 12, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_loop_filter_vertical_edge_uv_neon(u_ptr + 4, uv_stride, blim, lim, hev_thr, v_ptr + 4);
+        vp8_loop_filter_vertical_edge_uv_neon(u_ptr + 4, uv_stride, lfi->flim, lfi->lim, lfi->thr, v_ptr + 4);
+}
+
+void vp8_loop_filter_bvs_neon(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                              int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_vertical_edge_neon(y_ptr + 4, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_vertical_edge_neon(y_ptr + 8, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_vertical_edge_neon(y_ptr + 12, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 }
 #endif

vp8/common/arm/loopfilter_arm.h

@@ -12,17 +12,15 @@
 #ifndef LOOPFILTER_ARM_H
 #define LOOPFILTER_ARM_H
 
-#include "vpx_config.h"
-
 #if HAVE_ARMV6
 extern prototype_loopfilter_block(vp8_loop_filter_mbv_armv6);
 extern prototype_loopfilter_block(vp8_loop_filter_bv_armv6);
 extern prototype_loopfilter_block(vp8_loop_filter_mbh_armv6);
 extern prototype_loopfilter_block(vp8_loop_filter_bh_armv6);
-extern prototype_simple_loopfilter(vp8_loop_filter_bvs_armv6);
-extern prototype_simple_loopfilter(vp8_loop_filter_bhs_armv6);
-extern prototype_simple_loopfilter(vp8_loop_filter_simple_horizontal_edge_armv6);
-extern prototype_simple_loopfilter(vp8_loop_filter_simple_vertical_edge_armv6);
+extern prototype_loopfilter_block(vp8_loop_filter_mbvs_armv6);
+extern prototype_loopfilter_block(vp8_loop_filter_bvs_armv6);
+extern prototype_loopfilter_block(vp8_loop_filter_mbhs_armv6);
+extern prototype_loopfilter_block(vp8_loop_filter_bhs_armv6);
 
 #if !CONFIG_RUNTIME_CPU_DETECT
 #undef  vp8_lf_normal_mb_v
@@ -38,29 +36,28 @@ extern prototype_simple_loopfilter(vp8_loop_filter_simple_vertical_edge_armv6);
 #define vp8_lf_normal_b_h vp8_loop_filter_bh_armv6
 
 #undef  vp8_lf_simple_mb_v
-#define vp8_lf_simple_mb_v vp8_loop_filter_simple_vertical_edge_armv6
+#define vp8_lf_simple_mb_v vp8_loop_filter_mbvs_armv6
 
 #undef  vp8_lf_simple_b_v
 #define vp8_lf_simple_b_v vp8_loop_filter_bvs_armv6
 
 #undef  vp8_lf_simple_mb_h
-#define vp8_lf_simple_mb_h vp8_loop_filter_simple_horizontal_edge_armv6
+#define vp8_lf_simple_mb_h vp8_loop_filter_mbhs_armv6
 
 #undef  vp8_lf_simple_b_h
 #define vp8_lf_simple_b_h vp8_loop_filter_bhs_armv6
-#endif /* !CONFIG_RUNTIME_CPU_DETECT */
-
-#endif /* HAVE_ARMV6 */
+#endif
+#endif
 
 #if HAVE_ARMV7
 extern prototype_loopfilter_block(vp8_loop_filter_mbv_neon);
 extern prototype_loopfilter_block(vp8_loop_filter_bv_neon);
 extern prototype_loopfilter_block(vp8_loop_filter_mbh_neon);
 extern prototype_loopfilter_block(vp8_loop_filter_bh_neon);
-extern prototype_simple_loopfilter(vp8_loop_filter_mbvs_neon);
-extern prototype_simple_loopfilter(vp8_loop_filter_bvs_neon);
-extern prototype_simple_loopfilter(vp8_loop_filter_mbhs_neon);
-extern prototype_simple_loopfilter(vp8_loop_filter_bhs_neon);
+extern prototype_loopfilter_block(vp8_loop_filter_mbvs_neon);
+extern prototype_loopfilter_block(vp8_loop_filter_bvs_neon);
+extern prototype_loopfilter_block(vp8_loop_filter_mbhs_neon);
+extern prototype_loopfilter_block(vp8_loop_filter_bhs_neon);
 
 #if !CONFIG_RUNTIME_CPU_DETECT
 #undef  vp8_lf_normal_mb_v
@@ -86,8 +83,7 @@ extern prototype_simple_loopfilter(vp8_loop_filter_bhs_neon);
 
 #undef  vp8_lf_simple_b_h
 #define vp8_lf_simple_b_h vp8_loop_filter_bhs_neon
-#endif /* !CONFIG_RUNTIME_CPU_DETECT */
+#endif
+#endif
 
-#endif /* HAVE_ARMV7 */
-
-#endif /* LOOPFILTER_ARM_H */
+#endif

vp8/common/arm/neon/bilinearpredict16x16_neon.asm

@@ -25,7 +25,7 @@
 |vp8_bilinear_predict16x16_neon| PROC
     push            {r4-r5, lr}
 
-    adr             r12, bifilter16_coeff
+    ldr             r12, _bifilter16_coeff_
     ldr             r4, [sp, #12]           ;load parameters from stack
     ldr             r5, [sp, #16]           ;load parameters from stack
 
@@ -351,6 +351,8 @@ filt_blk2d_spo16x16_loop_neon
 
 ;-----------------
 
+_bifilter16_coeff_
+    DCD     bifilter16_coeff
 bifilter16_coeff
     DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
 

vp8/common/arm/neon/bilinearpredict4x4_neon.asm

@@ -25,7 +25,7 @@
 |vp8_bilinear_predict4x4_neon| PROC
     push            {r4, lr}
 
-    adr             r12, bifilter4_coeff
+    ldr             r12, _bifilter4_coeff_
     ldr             r4, [sp, #8]            ;load parameters from stack
     ldr             lr, [sp, #12]           ;load parameters from stack
 
@@ -124,6 +124,8 @@ skip_secondpass_filter
 
 ;-----------------
 
+_bifilter4_coeff_
+    DCD     bifilter4_coeff
 bifilter4_coeff
     DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
 

vp8/common/arm/neon/bilinearpredict8x4_neon.asm

@@ -25,7 +25,7 @@
 |vp8_bilinear_predict8x4_neon| PROC
     push            {r4, lr}
 
-    adr             r12, bifilter8x4_coeff
+    ldr             r12, _bifilter8x4_coeff_
     ldr             r4, [sp, #8]            ;load parameters from stack
     ldr             lr, [sp, #12]           ;load parameters from stack
 
@@ -129,6 +129,8 @@ skip_secondpass_filter
 
 ;-----------------
 
+_bifilter8x4_coeff_
+    DCD     bifilter8x4_coeff
 bifilter8x4_coeff
     DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
 

vp8/common/arm/neon/bilinearpredict8x8_neon.asm

@@ -25,7 +25,7 @@
 |vp8_bilinear_predict8x8_neon| PROC
     push            {r4, lr}
 
-    adr             r12, bifilter8_coeff
+    ldr             r12, _bifilter8_coeff_
     ldr             r4, [sp, #8]            ;load parameters from stack
     ldr             lr, [sp, #12]           ;load parameters from stack
 
@@ -177,6 +177,8 @@ skip_secondpass_filter
 
 ;-----------------
 
+_bifilter8_coeff_
+    DCD     bifilter8_coeff
 bifilter8_coeff
     DCD     128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
 

vp8/common/arm/neon/iwalsh_neon.asm

@@ -20,16 +20,19 @@
 |vp8_short_inv_walsh4x4_neon| PROC
 
     ; read in all four lines of values: d0->d3
-    vld1.i16 {q0-q1}, [r0@128]
+    vldm.64 r0, {q0, q1}
 
     ; first for loop
-    vadd.s16 d4, d0, d3 ;a = [0] + [12]
-    vadd.s16 d6, d1, d2 ;b = [4] + [8]
-    vsub.s16 d5, d0, d3 ;d = [0] - [12]
-    vsub.s16 d7, d1, d2 ;c = [4] - [8]
 
-    vadd.s16 q0, q2, q3 ; a+b d+c
-    vsub.s16 q1, q2, q3 ; a-b d-c
+    vadd.s16 d4, d0, d3 ;a = [0] + [12]
+    vadd.s16 d5, d1, d2 ;b = [4] + [8]
+    vsub.s16 d6, d1, d2 ;c = [4] - [8]
+    vsub.s16 d7, d0, d3 ;d = [0] - [12]
+
+    vadd.s16 d0, d4, d5 ;a + b
+    vadd.s16 d1, d6, d7 ;c + d
+    vsub.s16 d2, d4, d5 ;a - b
+    vsub.s16 d3, d7, d6 ;d - c
 
     vtrn.32 d0, d2 ;d0:  0  1  8  9
                    ;d2:  2  3 10 11
@@ -44,22 +47,29 @@
     ; second for loop
 
     vadd.s16 d4, d0, d3 ;a = [0] + [3]
-    vadd.s16 d6, d1, d2 ;b = [1] + [2]
-    vsub.s16 d5, d0, d3 ;d = [0] - [3]
-    vsub.s16 d7, d1, d2 ;c = [1] - [2]
+    vadd.s16 d5, d1, d2 ;b = [1] + [2]
+    vsub.s16 d6, d1, d2 ;c = [1] - [2]
+    vsub.s16 d7, d0, d3 ;d = [0] - [3]
 
-    vmov.i16 q8, #3
+    vadd.s16 d0, d4, d5 ;e = a + b
+    vadd.s16 d1, d6, d7 ;f = c + d
+    vsub.s16 d2, d4, d5 ;g = a - b
+    vsub.s16 d3, d7, d6 ;h = d - c
 
-    vadd.s16 q0, q2, q3 ; a+b d+c
-    vsub.s16 q1, q2, q3 ; a-b d-c
-
-    vadd.i16 q0, q0, q8 ;e/f += 3
-    vadd.i16 q1, q1, q8 ;g/h += 3
+    vmov.i16 q2, #3
+    vadd.i16 q0, q0, q2 ;e/f += 3
+    vadd.i16 q1, q1, q2 ;g/h += 3
 
     vshr.s16 q0, q0, #3 ;e/f >> 3
     vshr.s16 q1, q1, #3 ;g/h >> 3
 
-    vst4.i16 {d0,d1,d2,d3}, [r1@128]
+    vtrn.32 d0, d2
+    vtrn.32 d1, d3
+    vtrn.16 d0, d1
+    vtrn.16 d2, d3
+
+    vstmia.16 r1!, {q0}
+    vstmia.16 r1!, {q1}
 
     bx lr
     ENDP    ; |vp8_short_inv_walsh4x4_neon|
@@ -67,13 +77,19 @@
 
 ;short vp8_short_inv_walsh4x4_1_neon(short *input, short *output)
 |vp8_short_inv_walsh4x4_1_neon| PROC
-    ldrsh r2, [r0]          ; load input[0]
-    add r3, r2, #3          ; add 3
-    add r2, r1, #16         ; base for last 8 output
-    asr r0, r3, #3          ; right shift 3
-    vdup.16 q0, r0          ; load and duplicate
-    vst1.16 {q0}, [r1@128]  ; write back 8
-    vst1.16 {q0}, [r2@128]  ; write back last 8
+    ; load a full line into a neon register
+    vld1.16  {q0}, [r0]
+    ; extract first element and replicate
+    vdup.16 q1, d0[0]
+    ; add 3 to all values
+    vmov.i16 q2, #3
+    vadd.i16 q3, q1, q2
+    ; right shift
+    vshr.s16 q3, q3, #3
+    ; write it back
+    vstmia.16 r1!, {q3}
+    vstmia.16 r1!, {q3}
+
     bx lr
     ENDP    ; |vp8_short_inv_walsh4x4_1_neon|
 

vp8/common/arm/neon/loopfilter_neon.asm

@@ -14,97 +14,109 @@
     EXPORT  |vp8_loop_filter_vertical_edge_y_neon|
     EXPORT  |vp8_loop_filter_vertical_edge_uv_neon|
     ARM
+    REQUIRE8
+    PRESERVE8
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
 
+; flimit, limit, and thresh should be positive numbers.
+; All 16 elements in these variables are equal.
+
+; void vp8_loop_filter_horizontal_edge_y_neon(unsigned char *src, int pitch,
+;                                             const signed char *flimit,
+;                                             const signed char *limit,
+;                                             const signed char *thresh,
+;                                             int count)
 ; r0    unsigned char *src
 ; r1    int pitch
-; r2    unsigned char blimit
-; r3    unsigned char limit
-; sp    unsigned char thresh,
+; r2    const signed char *flimit
+; r3    const signed char *limit
+; sp    const signed char *thresh,
+; sp+4  int count (unused)
 |vp8_loop_filter_horizontal_edge_y_neon| PROC
-    push        {lr}
-    vdup.u8     q0, r2                     ; duplicate blimit
-    vdup.u8     q1, r3                     ; duplicate limit
+    stmdb       sp!, {lr}
+    vld1.s8     {d0[], d1[]}, [r2]          ; flimit
+    vld1.s8     {d2[], d3[]}, [r3]          ; limit
     sub         r2, r0, r1, lsl #2          ; move src pointer down by 4 lines
-    ldr         r3, [sp, #4]               ; load thresh
-    add         r12, r2, r1
-    add         r1, r1, r1
+    ldr         r12, [sp, #4]               ; load thresh pointer
 
-    vdup.u8     q2, r3                     ; duplicate thresh
-
-    vld1.u8     {q3}, [r2@128], r1              ; p3
-    vld1.u8     {q4}, [r12@128], r1             ; p2
-    vld1.u8     {q5}, [r2@128], r1              ; p1
-    vld1.u8     {q6}, [r12@128], r1             ; p0
-    vld1.u8     {q7}, [r2@128], r1              ; q0
-    vld1.u8     {q8}, [r12@128], r1             ; q1
-    vld1.u8     {q9}, [r2@128]                  ; q2
-    vld1.u8     {q10}, [r12@128]                ; q3
-
-    sub         r2, r2, r1, lsl #1
-    sub         r12, r12, r1, lsl #1
+    vld1.u8     {q3}, [r2], r1              ; p3
+    vld1.u8     {q4}, [r2], r1              ; p2
+    vld1.u8     {q5}, [r2], r1              ; p1
+    vld1.u8     {q6}, [r2], r1              ; p0
+    vld1.u8     {q7}, [r2], r1              ; q0
+    vld1.u8     {q8}, [r2], r1              ; q1
+    vld1.u8     {q9}, [r2], r1              ; q2
+    vld1.u8     {q10}, [r2]                 ; q3
+    vld1.s8     {d4[], d5[]}, [r12]         ; thresh
+    sub         r0, r0, r1, lsl #1
 
     bl          vp8_loop_filter_neon
 
-    vst1.u8     {q5}, [r2@128], r1              ; store op1
-    vst1.u8     {q6}, [r12@128], r1             ; store op0
-    vst1.u8     {q7}, [r2@128], r1              ; store oq0
-    vst1.u8     {q8}, [r12@128], r1             ; store oq1
+    vst1.u8     {q5}, [r0], r1              ; store op1
+    vst1.u8     {q6}, [r0], r1              ; store op0
+    vst1.u8     {q7}, [r0], r1              ; store oq0
+    vst1.u8     {q8}, [r0], r1              ; store oq1
 
-    pop         {pc}
+    ldmia       sp!, {pc}
     ENDP        ; |vp8_loop_filter_horizontal_edge_y_neon|
 
-
+; void vp8_loop_filter_horizontal_edge_uv_neon(unsigned char *u, int pitch
+;                                              const signed char *flimit,
+;                                              const signed char *limit,
+;                                              const signed char *thresh,
+;                                              unsigned char *v)
 ; r0    unsigned char *u,
 ; r1    int pitch,
-; r2    unsigned char blimit
-; r3    unsigned char limit
-; sp    unsigned char thresh,
+; r2    const signed char *flimit,
+; r3    const signed char *limit,
+; sp    const signed char *thresh,
 ; sp+4  unsigned char *v
 |vp8_loop_filter_horizontal_edge_uv_neon| PROC
-    push        {lr}
-    vdup.u8     q0, r2                      ; duplicate blimit
-    vdup.u8     q1, r3                      ; duplicate limit
-    ldr         r12, [sp, #4]               ; load thresh
+    stmdb       sp!, {lr}
+    vld1.s8     {d0[], d1[]}, [r2]          ; flimit
+    vld1.s8     {d2[], d3[]}, [r3]          ; limit
     ldr         r2, [sp, #8]                ; load v ptr
-    vdup.u8     q2, r12                     ; duplicate thresh
 
     sub         r3, r0, r1, lsl #2          ; move u pointer down by 4 lines
-    sub         r12, r2, r1, lsl #2         ; move v pointer down by 4 lines
+    vld1.u8     {d6}, [r3], r1              ; p3
+    vld1.u8     {d8}, [r3], r1              ; p2
+    vld1.u8     {d10}, [r3], r1             ; p1
+    vld1.u8     {d12}, [r3], r1             ; p0
+    vld1.u8     {d14}, [r3], r1             ; q0
+    vld1.u8     {d16}, [r3], r1             ; q1
+    vld1.u8     {d18}, [r3], r1             ; q2
+    vld1.u8     {d20}, [r3]                 ; q3
 
-    vld1.u8     {d6}, [r3@64], r1              ; p3
-    vld1.u8     {d7}, [r12@64], r1             ; p3
-    vld1.u8     {d8}, [r3@64], r1              ; p2
-    vld1.u8     {d9}, [r12@64], r1             ; p2
-    vld1.u8     {d10}, [r3@64], r1             ; p1
-    vld1.u8     {d11}, [r12@64], r1            ; p1
-    vld1.u8     {d12}, [r3@64], r1             ; p0
-    vld1.u8     {d13}, [r12@64], r1            ; p0
-    vld1.u8     {d14}, [r3@64], r1             ; q0
-    vld1.u8     {d15}, [r12@64], r1            ; q0
-    vld1.u8     {d16}, [r3@64], r1             ; q1
-    vld1.u8     {d17}, [r12@64], r1            ; q1
-    vld1.u8     {d18}, [r3@64], r1             ; q2
-    vld1.u8     {d19}, [r12@64], r1            ; q2
-    vld1.u8     {d20}, [r3@64]                 ; q3
-    vld1.u8     {d21}, [r12@64]                ; q3
+    ldr         r3, [sp, #4]                ; load thresh pointer
+
+    sub         r12, r2, r1, lsl #2         ; move v pointer down by 4 lines
+    vld1.u8     {d7}, [r12], r1             ; p3
+    vld1.u8     {d9}, [r12], r1             ; p2
+    vld1.u8     {d11}, [r12], r1            ; p1
+    vld1.u8     {d13}, [r12], r1            ; p0
+    vld1.u8     {d15}, [r12], r1            ; q0
+    vld1.u8     {d17}, [r12], r1            ; q1
+    vld1.u8     {d19}, [r12], r1            ; q2
+    vld1.u8     {d21}, [r12]                ; q3
+
+    vld1.s8     {d4[], d5[]}, [r3]          ; thresh
 
     bl          vp8_loop_filter_neon
 
     sub         r0, r0, r1, lsl #1
     sub         r2, r2, r1, lsl #1
 
-    vst1.u8     {d10}, [r0@64], r1             ; store u op1
-    vst1.u8     {d11}, [r2@64], r1             ; store v op1
-    vst1.u8     {d12}, [r0@64], r1             ; store u op0
-    vst1.u8     {d13}, [r2@64], r1             ; store v op0
-    vst1.u8     {d14}, [r0@64], r1             ; store u oq0
-    vst1.u8     {d15}, [r2@64], r1             ; store v oq0
-    vst1.u8     {d16}, [r0@64]                 ; store u oq1
-    vst1.u8     {d17}, [r2@64]                 ; store v oq1
+    vst1.u8     {d10}, [r0], r1             ; store u op1
+    vst1.u8     {d11}, [r2], r1             ; store v op1
+    vst1.u8     {d12}, [r0], r1             ; store u op0
+    vst1.u8     {d13}, [r2], r1             ; store v op0
+    vst1.u8     {d14}, [r0], r1             ; store u oq0
+    vst1.u8     {d15}, [r2], r1             ; store v oq0
+    vst1.u8     {d16}, [r0]                 ; store u oq1
+    vst1.u8     {d17}, [r2]                 ; store v oq1
 
-    pop         {pc}
+    ldmia       sp!, {pc}
     ENDP        ; |vp8_loop_filter_horizontal_edge_uv_neon|
 
 ; void vp8_loop_filter_vertical_edge_y_neon(unsigned char *src, int pitch,
@@ -112,38 +124,39 @@
 ;                                           const signed char *limit,
 ;                                           const signed char *thresh,
 ;                                           int count)
-; r0    unsigned char *src
-; r1    int pitch
-; r2    unsigned char blimit
-; r3    unsigned char limit
-; sp    unsigned char thresh,
-
+; r0    unsigned char *src,
+; r1    int pitch,
+; r2    const signed char *flimit,
+; r3    const signed char *limit,
+; sp    const signed char *thresh,
+; sp+4  int count (unused)
 |vp8_loop_filter_vertical_edge_y_neon| PROC
-    push        {lr}
-    vdup.u8     q0, r2                     ; duplicate blimit
-    vdup.u8     q1, r3                     ; duplicate limit
+    stmdb       sp!, {lr}
+    vld1.s8     {d0[], d1[]}, [r2]          ; flimit
+    vld1.s8     {d2[], d3[]}, [r3]          ; limit
     sub         r2, r0, #4                  ; src ptr down by 4 columns
-    add         r1, r1, r1
-    ldr         r3, [sp, #4]               ; load thresh
-    add         r12, r2, r1, asr #1
+    sub         r0, r0, #2                  ; dst ptr
+    ldr         r12, [sp, #4]               ; load thresh pointer
 
-    vld1.u8     {d6}, [r2], r1
-    vld1.u8     {d8}, [r12], r1
+    vld1.u8     {d6}, [r2], r1              ; load first 8-line src data
+    vld1.u8     {d8}, [r2], r1
     vld1.u8     {d10}, [r2], r1
-    vld1.u8     {d12}, [r12], r1
+    vld1.u8     {d12}, [r2], r1
     vld1.u8     {d14}, [r2], r1
-    vld1.u8     {d16}, [r12], r1
+    vld1.u8     {d16}, [r2], r1
     vld1.u8     {d18}, [r2], r1
-    vld1.u8     {d20}, [r12], r1
+    vld1.u8     {d20}, [r2], r1
+
+    vld1.s8     {d4[], d5[]}, [r12]         ; thresh
 
     vld1.u8     {d7}, [r2], r1              ; load second 8-line src data
-    vld1.u8     {d9}, [r12], r1
+    vld1.u8     {d9}, [r2], r1
     vld1.u8     {d11}, [r2], r1
-    vld1.u8     {d13}, [r12], r1
+    vld1.u8     {d13}, [r2], r1
     vld1.u8     {d15}, [r2], r1
-    vld1.u8     {d17}, [r12], r1
-    vld1.u8     {d19}, [r2]
-    vld1.u8     {d21}, [r12]
+    vld1.u8     {d17}, [r2], r1
+    vld1.u8     {d19}, [r2], r1
+    vld1.u8     {d21}, [r2]
 
     ;transpose to 8x16 matrix
     vtrn.32     q3, q7
@@ -151,8 +164,6 @@
     vtrn.32     q5, q9
     vtrn.32     q6, q10
 
-    vdup.u8     q2, r3                     ; duplicate thresh
-
     vtrn.16     q3, q5
     vtrn.16     q4, q6
     vtrn.16     q7, q9
@@ -167,34 +178,28 @@
 
     vswp        d12, d11
     vswp        d16, d13
-
-    sub         r0, r0, #2                 ; dst ptr
-
     vswp        d14, d12
     vswp        d16, d15
 
-    add         r12, r0, r1, asr #1
-
     ;store op1, op0, oq0, oq1
     vst4.8      {d10[0], d11[0], d12[0], d13[0]}, [r0], r1
-    vst4.8      {d10[1], d11[1], d12[1], d13[1]}, [r12], r1
+    vst4.8      {d10[1], d11[1], d12[1], d13[1]}, [r0], r1
     vst4.8      {d10[2], d11[2], d12[2], d13[2]}, [r0], r1
-    vst4.8      {d10[3], d11[3], d12[3], d13[3]}, [r12], r1
+    vst4.8      {d10[3], d11[3], d12[3], d13[3]}, [r0], r1
     vst4.8      {d10[4], d11[4], d12[4], d13[4]}, [r0], r1
-    vst4.8      {d10[5], d11[5], d12[5], d13[5]}, [r12], r1
+    vst4.8      {d10[5], d11[5], d12[5], d13[5]}, [r0], r1
     vst4.8      {d10[6], d11[6], d12[6], d13[6]}, [r0], r1
-    vst4.8      {d10[7], d11[7], d12[7], d13[7]}, [r12], r1
-
+    vst4.8      {d10[7], d11[7], d12[7], d13[7]}, [r0], r1
     vst4.8      {d14[0], d15[0], d16[0], d17[0]}, [r0], r1
-    vst4.8      {d14[1], d15[1], d16[1], d17[1]}, [r12], r1
+    vst4.8      {d14[1], d15[1], d16[1], d17[1]}, [r0], r1
     vst4.8      {d14[2], d15[2], d16[2], d17[2]}, [r0], r1
-    vst4.8      {d14[3], d15[3], d16[3], d17[3]}, [r12], r1
+    vst4.8      {d14[3], d15[3], d16[3], d17[3]}, [r0], r1
     vst4.8      {d14[4], d15[4], d16[4], d17[4]}, [r0], r1
-    vst4.8      {d14[5], d15[5], d16[5], d17[5]}, [r12], r1
-    vst4.8      {d14[6], d15[6], d16[6], d17[6]}, [r0]
-    vst4.8      {d14[7], d15[7], d16[7], d17[7]}, [r12]
+    vst4.8      {d14[5], d15[5], d16[5], d17[5]}, [r0], r1
+    vst4.8      {d14[6], d15[6], d16[6], d17[6]}, [r0], r1
+    vst4.8      {d14[7], d15[7], d16[7], d17[7]}, [r0]
 
-    pop         {pc}
+    ldmia       sp!, {pc}
     ENDP        ; |vp8_loop_filter_vertical_edge_y_neon|
 
 ; void vp8_loop_filter_vertical_edge_uv_neon(unsigned char *u, int pitch
@@ -204,36 +209,38 @@
 ;                                            unsigned char *v)
 ; r0    unsigned char *u,
 ; r1    int pitch,
-; r2    unsigned char blimit
-; r3    unsigned char limit
-; sp    unsigned char thresh,
+; r2    const signed char *flimit,
+; r3    const signed char *limit,
+; sp    const signed char *thresh,
 ; sp+4  unsigned char *v
 |vp8_loop_filter_vertical_edge_uv_neon| PROC
-    push        {lr}
-    vdup.u8     q0, r2                      ; duplicate blimit
+    stmdb       sp!, {lr}
     sub         r12, r0, #4                  ; move u pointer down by 4 columns
+    vld1.s8     {d0[], d1[]}, [r2]          ; flimit
+    vld1.s8     {d2[], d3[]}, [r3]          ; limit
+
     ldr         r2, [sp, #8]                ; load v ptr
-    vdup.u8     q1, r3                      ; duplicate limit
-    sub         r3, r2, #4                  ; move v pointer down by 4 columns
 
     vld1.u8     {d6}, [r12], r1              ;load u data
-    vld1.u8     {d7}, [r3], r1              ;load v data
     vld1.u8     {d8}, [r12], r1
-    vld1.u8     {d9}, [r3], r1
     vld1.u8     {d10}, [r12], r1
-    vld1.u8     {d11}, [r3], r1
     vld1.u8     {d12}, [r12], r1
-    vld1.u8     {d13}, [r3], r1
     vld1.u8     {d14}, [r12], r1
-    vld1.u8     {d15}, [r3], r1
     vld1.u8     {d16}, [r12], r1
-    vld1.u8     {d17}, [r3], r1
     vld1.u8     {d18}, [r12], r1
-    vld1.u8     {d19}, [r3], r1
     vld1.u8     {d20}, [r12]
+
+    sub         r3, r2, #4                  ; move v pointer down by 4 columns
+    vld1.u8     {d7}, [r3], r1              ;load v data
+    vld1.u8     {d9}, [r3], r1
+    vld1.u8     {d11}, [r3], r1
+    vld1.u8     {d13}, [r3], r1
+    vld1.u8     {d15}, [r3], r1
+    vld1.u8     {d17}, [r3], r1
+    vld1.u8     {d19}, [r3], r1
     vld1.u8     {d21}, [r3]
 
-    ldr        r12, [sp, #4]               ; load thresh
+    ldr         r12, [sp, #4]               ; load thresh pointer
 
     ;transpose to 8x16 matrix
     vtrn.32     q3, q7
@@ -241,8 +248,6 @@
     vtrn.32     q5, q9
     vtrn.32     q6, q10
 
-    vdup.u8     q2, r12                     ; duplicate thresh
-
     vtrn.16     q3, q5
     vtrn.16     q4, q6
     vtrn.16     q7, q9
@@ -253,16 +258,18 @@
     vtrn.8      q7, q8
     vtrn.8      q9, q10
 
+    vld1.s8     {d4[], d5[]}, [r12]         ; thresh
+
     bl          vp8_loop_filter_neon
 
+    sub         r0, r0, #2
+    sub         r2, r2, #2
+
     vswp        d12, d11
     vswp        d16, d13
     vswp        d14, d12
     vswp        d16, d15
 
-    sub         r0, r0, #2
-    sub         r2, r2, #2
-
     ;store op1, op0, oq0, oq1
     vst4.8      {d10[0], d11[0], d12[0], d13[0]}, [r0], r1
     vst4.8      {d14[0], d15[0], d16[0], d17[0]}, [r2], r1
@@ -281,7 +288,7 @@
     vst4.8      {d10[7], d11[7], d12[7], d13[7]}, [r0]
     vst4.8      {d14[7], d15[7], d16[7], d17[7]}, [r2]
 
-    pop         {pc}
+    ldmia       sp!, {pc}
     ENDP        ; |vp8_loop_filter_vertical_edge_uv_neon|
 
 ; void vp8_loop_filter_neon();
@@ -301,6 +308,7 @@
 ; q9    q2
 ; q10   q3
 |vp8_loop_filter_neon| PROC
+    ldr         r12, _lf_coeff_
 
     ; vp8_filter_mask
     vabd.u8     q11, q3, q4                 ; abs(p3 - p2)
@@ -309,44 +317,42 @@
     vabd.u8     q14, q8, q7                 ; abs(q1 - q0)
     vabd.u8     q3, q9, q8                  ; abs(q2 - q1)
     vabd.u8     q4, q10, q9                 ; abs(q3 - q2)
+    vabd.u8     q9, q6, q7                  ; abs(p0 - q0)
 
     vmax.u8     q11, q11, q12
     vmax.u8     q12, q13, q14
     vmax.u8     q3, q3, q4
     vmax.u8     q15, q11, q12
 
-    vabd.u8     q9, q6, q7                  ; abs(p0 - q0)
-
     ; vp8_hevmask
     vcgt.u8     q13, q13, q2                ; (abs(p1 - p0) > thresh)*-1
     vcgt.u8     q14, q14, q2                ; (abs(q1 - q0) > thresh)*-1
     vmax.u8     q15, q15, q3
 
-    vmov.u8     q10, #0x80                   ; 0x80
+    vadd.u8     q0, q0, q0                  ; flimit * 2
+    vadd.u8     q0, q0, q1                  ; flimit * 2 + limit
+    vcge.u8     q15, q1, q15
 
     vabd.u8     q2, q5, q8                  ; a = abs(p1 - q1)
     vqadd.u8    q9, q9, q9                  ; b = abs(p0 - q0) * 2
+    vshr.u8     q2, q2, #1                  ; a = a / 2
+    vqadd.u8    q9, q9, q2                  ; a = b + a
+    vcge.u8     q9, q0, q9                  ; (a > flimit * 2 + limit) * -1
 
-    vcge.u8     q15, q1, q15
+    vld1.u8     {q0}, [r12]!
 
     ; vp8_filter() function
     ; convert to signed
-    veor        q7, q7, q10                 ; qs0
-    vshr.u8     q2, q2, #1                  ; a = a / 2
-    veor        q6, q6, q10                 ; ps0
+    veor        q7, q7, q0                  ; qs0
+    veor        q6, q6, q0                  ; ps0
+    veor        q5, q5, q0                  ; ps1
+    veor        q8, q8, q0                  ; qs1
 
-    veor        q5, q5, q10                 ; ps1
-    vqadd.u8    q9, q9, q2                  ; a = b + a
-
-    veor        q8, q8, q10                 ; qs1
-
-    vmov.u8     q10, #3                     ; #3
+    vld1.u8     {q10}, [r12]!
 
     vsubl.s8    q2, d14, d12                ; ( qs0 - ps0)
     vsubl.s8    q11, d15, d13
 
-    vcge.u8     q9, q0, q9                  ; (a > flimit * 2 + limit) * -1
-
     vmovl.u8    q4, d20
 
     vqsub.s8    q1, q5, q8                  ; vp8_filter = clamp(ps1-qs1)
@@ -361,7 +367,7 @@
     vaddw.s8    q2, q2, d2
     vaddw.s8    q11, q11, d3
 
-    vmov.u8     q9, #4                      ; #4
+    vld1.u8     {q9}, [r12]!
 
     ; vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
     vqmovn.s16  d2, q2
@@ -373,20 +379,19 @@
     vshr.s8     q2, q2, #3                  ; Filter2 >>= 3
     vshr.s8     q1, q1, #3                  ; Filter1 >>= 3
 
-
     vqadd.s8    q11, q6, q2                 ; u = clamp(ps0 + Filter2)
     vqsub.s8    q10, q7, q1                 ; u = clamp(qs0 - Filter1)
 
     ; outer tap adjustments: ++vp8_filter >> 1
     vrshr.s8    q1, q1, #1
     vbic        q1, q1, q14                 ; vp8_filter &= ~hev
-    vmov.u8     q0, #0x80                   ; 0x80
+
     vqadd.s8    q13, q5, q1                 ; u = clamp(ps1 + vp8_filter)
     vqsub.s8    q12, q8, q1                 ; u = clamp(qs1 - vp8_filter)
 
+    veor        q5, q13, q0                 ; *op1 = u^0x80
     veor        q6, q11, q0                 ; *op0 = u^0x80
     veor        q7, q10, q0                 ; *oq0 = u^0x80
-    veor        q5, q13, q0                 ; *op1 = u^0x80
     veor        q8, q12, q0                 ; *oq1 = u^0x80
 
     bx          lr
@@ -394,4 +399,12 @@
 
 ;-----------------
 
+_lf_coeff_
+    DCD     lf_coeff
+lf_coeff
+    DCD     0x80808080, 0x80808080, 0x80808080, 0x80808080
+    DCD     0x03030303, 0x03030303, 0x03030303, 0x03030303
+    DCD     0x04040404, 0x04040404, 0x04040404, 0x04040404
+    DCD     0x01010101, 0x01010101, 0x01010101, 0x01010101
+
     END

vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.asm

@@ -9,109 +9,107 @@
 ;
 
 
-    ;EXPORT  |vp8_loop_filter_simple_horizontal_edge_neon|
-    EXPORT  |vp8_loop_filter_bhs_neon|
-    EXPORT  |vp8_loop_filter_mbhs_neon|
+    EXPORT  |vp8_loop_filter_simple_horizontal_edge_neon|
     ARM
+    REQUIRE8
     PRESERVE8
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
-
-; r0    unsigned char *s, PRESERVE
-; r1    int p, PRESERVE
-; q1    limit, PRESERVE
+;Note: flimit, limit, and thresh shpuld be positive numbers. All 16 elements in flimit
+;are equal. So, in the code, only one load is needed
+;for flimit. Same way applies to limit and thresh.
+; r0    unsigned char *s,
+; r1    int p, //pitch
+; r2    const signed char *flimit,
+; r3    const signed char *limit,
+; stack(r4) const signed char *thresh,
+; //stack(r5)   int count --unused
 
 |vp8_loop_filter_simple_horizontal_edge_neon| PROC
+    sub         r0, r0, r1, lsl #1          ; move src pointer down by 2 lines
 
-    sub         r3, r0, r1, lsl #1          ; move src pointer down by 2 lines
-
-    vld1.u8     {q7}, [r0@128], r1          ; q0
-    vld1.u8     {q5}, [r3@128], r1          ; p0
-    vld1.u8     {q8}, [r0@128]              ; q1
-    vld1.u8     {q6}, [r3@128]              ; p1
+    ldr         r12, _lfhy_coeff_
+    vld1.u8     {q5}, [r0], r1              ; p1
+    vld1.s8     {d2[], d3[]}, [r2]          ; flimit
+    vld1.s8     {d26[], d27[]}, [r3]        ; limit -> q13
+    vld1.u8     {q6}, [r0], r1              ; p0
+    vld1.u8     {q0}, [r12]!                ; 0x80
+    vld1.u8     {q7}, [r0], r1              ; q0
+    vld1.u8     {q10}, [r12]!               ; 0x03
+    vld1.u8     {q8}, [r0]                  ; q1
 
+    ;vp8_filter_mask() function
     vabd.u8     q15, q6, q7                 ; abs(p0 - q0)
     vabd.u8     q14, q5, q8                 ; abs(p1 - q1)
-
     vqadd.u8    q15, q15, q15               ; abs(p0 - q0) * 2
     vshr.u8     q14, q14, #1                ; abs(p1 - q1) / 2
-    vmov.u8     q0, #0x80                   ; 0x80
-    vmov.s16    q13, #3
     vqadd.u8    q15, q15, q14               ; abs(p0 - q0) * 2 + abs(p1 - q1) / 2
 
+    ;vp8_filter() function
     veor        q7, q7, q0                  ; qs0: q0 offset to convert to a signed value
     veor        q6, q6, q0                  ; ps0: p0 offset to convert to a signed value
     veor        q5, q5, q0                  ; ps1: p1 offset to convert to a signed value
     veor        q8, q8, q0                  ; qs1: q1 offset to convert to a signed value
 
-    vcge.u8     q15, q1, q15                ; (abs(p0 - q0)*2 + abs(p1-q1)/2 > limit)*-1
+    vadd.u8     q1, q1, q1                  ; flimit * 2
+    vadd.u8     q1, q1, q13                 ; flimit * 2 + limit
+    vcge.u8     q15, q1, q15                ; (abs(p0 - q0)*2 + abs(p1-q1)/2 > flimit*2 + limit)*-1
 
+;;;;;;;;;;
+    ;vqsub.s8   q2, q7, q6                  ; ( qs0 - ps0)
     vsubl.s8    q2, d14, d12                ; ( qs0 - ps0)
     vsubl.s8    q3, d15, d13
 
     vqsub.s8    q4, q5, q8                  ; q4: vp8_filter = vp8_signed_char_clamp(ps1-qs1)
 
-    vmul.s16    q2, q2, q13                 ;  3 * ( qs0 - ps0)
-    vmul.s16    q3, q3, q13
+    ;vmul.i8    q2, q2, q10                 ;  3 * ( qs0 - ps0)
+    vadd.s16    q11, q2, q2                 ;  3 * ( qs0 - ps0)
+    vadd.s16    q12, q3, q3
 
-    vmov.u8     q10, #0x03                  ; 0x03
-    vmov.u8     q9, #0x04                   ; 0x04
+    vld1.u8     {q9}, [r12]!                ; 0x04
+
+    vadd.s16    q2, q2, q11
+    vadd.s16    q3, q3, q12
 
     vaddw.s8    q2, q2, d8                  ; vp8_filter + 3 * ( qs0 - ps0)
     vaddw.s8    q3, q3, d9
 
+    ;vqadd.s8   q4, q4, q2                  ; vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0))
     vqmovn.s16  d8, q2                      ; vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0))
     vqmovn.s16  d9, q3
+;;;;;;;;;;;;;
 
-    vand        q14, q4, q15                ; vp8_filter &= mask
+    vand        q4, q4, q15                 ; vp8_filter &= mask
 
-    vqadd.s8    q2, q14, q10                ; Filter2 = vp8_signed_char_clamp(vp8_filter+3)
-    vqadd.s8    q3, q14, q9                 ; Filter1 = vp8_signed_char_clamp(vp8_filter+4)
+    vqadd.s8    q2, q4, q10                 ; Filter2 = vp8_signed_char_clamp(vp8_filter+3)
+    vqadd.s8    q4, q4, q9                  ; Filter1 = vp8_signed_char_clamp(vp8_filter+4)
     vshr.s8     q2, q2, #3                  ; Filter2 >>= 3
-    vshr.s8     q4, q3, #3                  ; Filter1 >>= 3
+    vshr.s8     q4, q4, #3                  ; Filter1 >>= 3
 
-    sub         r0, r0, r1
+    sub         r0, r0, r1, lsl #1
 
     ;calculate output
     vqadd.s8    q11, q6, q2                 ; u = vp8_signed_char_clamp(ps0 + Filter2)
     vqsub.s8    q10, q7, q4                 ; u = vp8_signed_char_clamp(qs0 - Filter1)
 
+    add         r3, r0, r1
+
     veor        q6, q11, q0                 ; *op0 = u^0x80
     veor        q7, q10, q0                 ; *oq0 = u^0x80
 
-    vst1.u8     {q6}, [r3@128]              ; store op0
-    vst1.u8     {q7}, [r0@128]              ; store oq0
+    vst1.u8     {q6}, [r0]                  ; store op0
+    vst1.u8     {q7}, [r3]                  ; store oq0
 
     bx          lr
     ENDP        ; |vp8_loop_filter_simple_horizontal_edge_neon|
 
-; r0    unsigned char *y
-; r1    int ystride
-; r2    const unsigned char *blimit
+;-----------------
 
-|vp8_loop_filter_bhs_neon| PROC
-    push        {r4, lr}
-    ldrb        r3, [r2]                    ; load blim from mem
-    vdup.s8     q1, r3                      ; duplicate blim
-
-    add         r0, r0, r1, lsl #2          ; src = y_ptr + 4 * y_stride
-    bl          vp8_loop_filter_simple_horizontal_edge_neon
-    ; vp8_loop_filter_simple_horizontal_edge_neon preserves r0, r1 and q1
-    add         r0, r0, r1, lsl #2          ; src = y_ptr + 8* y_stride
-    bl          vp8_loop_filter_simple_horizontal_edge_neon
-    add         r0, r0, r1, lsl #2          ; src = y_ptr + 12 * y_stride
-    pop         {r4, lr}
-    b           vp8_loop_filter_simple_horizontal_edge_neon
-    ENDP        ;|vp8_loop_filter_bhs_neon|
-
-; r0    unsigned char *y
-; r1    int ystride
-; r2    const unsigned char *blimit
-
-|vp8_loop_filter_mbhs_neon| PROC
-    ldrb        r3, [r2]                   ; load blim from mem
-    vdup.s8     q1, r3                     ; duplicate mblim
-    b           vp8_loop_filter_simple_horizontal_edge_neon
-    ENDP        ;|vp8_loop_filter_bhs_neon|
+_lfhy_coeff_
+    DCD     lfhy_coeff
+lfhy_coeff
+    DCD     0x80808080, 0x80808080, 0x80808080, 0x80808080
+    DCD     0x03030303, 0x03030303, 0x03030303, 0x03030303
+    DCD     0x04040404, 0x04040404, 0x04040404, 0x04040404
 
     END

vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm

@@ -9,54 +9,60 @@
 ;
 
 
-    ;EXPORT  |vp8_loop_filter_simple_vertical_edge_neon|
-    EXPORT |vp8_loop_filter_bvs_neon|
-    EXPORT |vp8_loop_filter_mbvs_neon|
+    EXPORT  |vp8_loop_filter_simple_vertical_edge_neon|
     ARM
+    REQUIRE8
     PRESERVE8
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
-
-; r0    unsigned char *s, PRESERVE
-; r1    int p, PRESERVE
-; q1    limit, PRESERVE
+;Note: flimit, limit, and thresh should be positive numbers. All 16 elements in flimit
+;are equal. So, in the code, only one load is needed
+;for flimit. Same way applies to limit and thresh.
+; r0    unsigned char *s,
+; r1    int p, //pitch
+; r2    const signed char *flimit,
+; r3    const signed char *limit,
+; stack(r4) const signed char *thresh,
+; //stack(r5)   int count --unused
 
 |vp8_loop_filter_simple_vertical_edge_neon| PROC
     sub         r0, r0, #2                  ; move src pointer down by 2 columns
-    add         r12, r1, r1
-    add         r3, r0, r1
 
-    vld4.8      {d6[0], d7[0], d8[0], d9[0]}, [r0], r12
-    vld4.8      {d6[1], d7[1], d8[1], d9[1]}, [r3], r12
-    vld4.8      {d6[2], d7[2], d8[2], d9[2]}, [r0], r12
-    vld4.8      {d6[3], d7[3], d8[3], d9[3]}, [r3], r12
-    vld4.8      {d6[4], d7[4], d8[4], d9[4]}, [r0], r12
-    vld4.8      {d6[5], d7[5], d8[5], d9[5]}, [r3], r12
-    vld4.8      {d6[6], d7[6], d8[6], d9[6]}, [r0], r12
-    vld4.8      {d6[7], d7[7], d8[7], d9[7]}, [r3], r12
+    vld4.8      {d6[0], d7[0], d8[0], d9[0]}, [r0], r1
+    vld1.s8     {d2[], d3[]}, [r2]          ; flimit
+    vld1.s8     {d26[], d27[]}, [r3]        ; limit -> q13
+    vld4.8      {d6[1], d7[1], d8[1], d9[1]}, [r0], r1
+    ldr         r12, _vlfy_coeff_
+    vld4.8      {d6[2], d7[2], d8[2], d9[2]}, [r0], r1
+    vld4.8      {d6[3], d7[3], d8[3], d9[3]}, [r0], r1
+    vld4.8      {d6[4], d7[4], d8[4], d9[4]}, [r0], r1
+    vld4.8      {d6[5], d7[5], d8[5], d9[5]}, [r0], r1
+    vld4.8      {d6[6], d7[6], d8[6], d9[6]}, [r0], r1
+    vld4.8      {d6[7], d7[7], d8[7], d9[7]}, [r0], r1
 
-    vld4.8      {d10[0], d11[0], d12[0], d13[0]}, [r0], r12
-    vld4.8      {d10[1], d11[1], d12[1], d13[1]}, [r3], r12
-    vld4.8      {d10[2], d11[2], d12[2], d13[2]}, [r0], r12
-    vld4.8      {d10[3], d11[3], d12[3], d13[3]}, [r3], r12
-    vld4.8      {d10[4], d11[4], d12[4], d13[4]}, [r0], r12
-    vld4.8      {d10[5], d11[5], d12[5], d13[5]}, [r3], r12
-    vld4.8      {d10[6], d11[6], d12[6], d13[6]}, [r0], r12
-    vld4.8      {d10[7], d11[7], d12[7], d13[7]}, [r3]
+    vld4.8      {d10[0], d11[0], d12[0], d13[0]}, [r0], r1
+    vld1.u8     {q0}, [r12]!                ; 0x80
+    vld4.8      {d10[1], d11[1], d12[1], d13[1]}, [r0], r1
+    vld1.u8     {q11}, [r12]!               ; 0x03
+    vld4.8      {d10[2], d11[2], d12[2], d13[2]}, [r0], r1
+    vld1.u8     {q12}, [r12]!               ; 0x04
+    vld4.8      {d10[3], d11[3], d12[3], d13[3]}, [r0], r1
+    vld4.8      {d10[4], d11[4], d12[4], d13[4]}, [r0], r1
+    vld4.8      {d10[5], d11[5], d12[5], d13[5]}, [r0], r1
+    vld4.8      {d10[6], d11[6], d12[6], d13[6]}, [r0], r1
+    vld4.8      {d10[7], d11[7], d12[7], d13[7]}, [r0], r1
 
     vswp        d7, d10
     vswp        d12, d9
+    ;vswp       q4, q5                      ; p1:q3, p0:q5, q0:q4, q1:q6
 
     ;vp8_filter_mask() function
     ;vp8_hevmask() function
     sub         r0, r0, r1, lsl #4
     vabd.u8     q15, q5, q4                 ; abs(p0 - q0)
     vabd.u8     q14, q3, q6                 ; abs(p1 - q1)
-
     vqadd.u8    q15, q15, q15               ; abs(p0 - q0) * 2
     vshr.u8     q14, q14, #1                ; abs(p1 - q1) / 2
-    vmov.u8     q0, #0x80                   ; 0x80
-    vmov.s16    q11, #3
     vqadd.u8    q15, q15, q14               ; abs(p0 - q0) * 2 + abs(p1 - q1) / 2
 
     veor        q4, q4, q0                  ; qs0: q0 offset to convert to a signed value
@@ -64,91 +70,87 @@
     veor        q3, q3, q0                  ; ps1: p1 offset to convert to a signed value
     veor        q6, q6, q0                  ; qs1: q1 offset to convert to a signed value
 
+    vadd.u8     q1, q1, q1                  ; flimit * 2
+    vadd.u8     q1, q1, q13                 ; flimit * 2 + limit
     vcge.u8     q15, q1, q15                ; abs(p0 - q0)*2 + abs(p1-q1)/2 > flimit*2 + limit)*-1
 
+    ;vp8_filter() function
+;;;;;;;;;;
+    ;vqsub.s8   q2, q5, q4                  ; ( qs0 - ps0)
     vsubl.s8    q2, d8, d10                 ; ( qs0 - ps0)
     vsubl.s8    q13, d9, d11
 
-    vqsub.s8    q14, q3, q6                  ; vp8_filter = vp8_signed_char_clamp(ps1-qs1)
+    vqsub.s8    q1, q3, q6                  ; vp8_filter = vp8_signed_char_clamp(ps1-qs1)
 
-    vmul.s16    q2, q2, q11                 ;  3 * ( qs0 - ps0)
-    vmul.s16    q13, q13, q11
+    ;vmul.i8    q2, q2, q11                 ; vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0))
+    vadd.s16    q10, q2, q2                 ;  3 * ( qs0 - ps0)
+    vadd.s16    q14, q13, q13
+    vadd.s16    q2, q2, q10
+    vadd.s16    q13, q13, q14
 
-    vmov.u8     q11, #0x03                  ; 0x03
-    vmov.u8     q12, #0x04                  ; 0x04
+    ;vqadd.s8   q1, q1, q2
+    vaddw.s8    q2, q2, d2                  ; vp8_filter + 3 * ( qs0 - ps0)
+    vaddw.s8    q13, q13, d3
 
-    vaddw.s8    q2, q2, d28                  ; vp8_filter + 3 * ( qs0 - ps0)
-    vaddw.s8    q13, q13, d29
-
-    vqmovn.s16  d28, q2                      ; vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0))
-    vqmovn.s16  d29, q13
+    vqmovn.s16  d2, q2                      ; vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0))
+    vqmovn.s16  d3, q13
 
     add         r0, r0, #1
-    add         r3, r0, r1
+    add         r2, r0, r1
+;;;;;;;;;;;
 
-    vand        q14, q14, q15                 ; vp8_filter &= mask
+    vand        q1, q1, q15                 ; vp8_filter &= mask
 
-    vqadd.s8    q2, q14, q11                 ; Filter2 = vp8_signed_char_clamp(vp8_filter+3)
-    vqadd.s8    q3, q14, q12                 ; Filter1 = vp8_signed_char_clamp(vp8_filter+4)
+    vqadd.s8    q2, q1, q11                 ; Filter2 = vp8_signed_char_clamp(vp8_filter+3)
+    vqadd.s8    q1, q1, q12                 ; Filter1 = vp8_signed_char_clamp(vp8_filter+4)
     vshr.s8     q2, q2, #3                  ; Filter2 >>= 3
-    vshr.s8     q14, q3, #3                  ; Filter1 >>= 3
+    vshr.s8     q1, q1, #3                  ; Filter1 >>= 3
 
     ;calculate output
+    vqsub.s8    q10, q4, q1                 ; u = vp8_signed_char_clamp(qs0 - Filter1)
     vqadd.s8    q11, q5, q2                 ; u = vp8_signed_char_clamp(ps0 + Filter2)
-    vqsub.s8    q10, q4, q14                 ; u = vp8_signed_char_clamp(qs0 - Filter1)
 
-    veor        q6, q11, q0                 ; *op0 = u^0x80
     veor        q7, q10, q0                 ; *oq0 = u^0x80
-    add         r12, r1, r1
+    veor        q6, q11, q0                 ; *op0 = u^0x80
+
+    add         r3, r2, r1
     vswp        d13, d14
+    add         r12, r3, r1
 
     ;store op1, op0, oq0, oq1
-    vst2.8      {d12[0], d13[0]}, [r0], r12
-    vst2.8      {d12[1], d13[1]}, [r3], r12
-    vst2.8      {d12[2], d13[2]}, [r0], r12
-    vst2.8      {d12[3], d13[3]}, [r3], r12
-    vst2.8      {d12[4], d13[4]}, [r0], r12
-    vst2.8      {d12[5], d13[5]}, [r3], r12
-    vst2.8      {d12[6], d13[6]}, [r0], r12
-    vst2.8      {d12[7], d13[7]}, [r3], r12
-    vst2.8      {d14[0], d15[0]}, [r0], r12
-    vst2.8      {d14[1], d15[1]}, [r3], r12
-    vst2.8      {d14[2], d15[2]}, [r0], r12
-    vst2.8      {d14[3], d15[3]}, [r3], r12
-    vst2.8      {d14[4], d15[4]}, [r0], r12
-    vst2.8      {d14[5], d15[5]}, [r3], r12
-    vst2.8      {d14[6], d15[6]}, [r0], r12
-    vst2.8      {d14[7], d15[7]}, [r3]
+    vst2.8      {d12[0], d13[0]}, [r0]
+    vst2.8      {d12[1], d13[1]}, [r2]
+    vst2.8      {d12[2], d13[2]}, [r3]
+    vst2.8      {d12[3], d13[3]}, [r12], r1
+    add         r0, r12, r1
+    vst2.8      {d12[4], d13[4]}, [r12]
+    vst2.8      {d12[5], d13[5]}, [r0], r1
+    add         r2, r0, r1
+    vst2.8      {d12[6], d13[6]}, [r0]
+    vst2.8      {d12[7], d13[7]}, [r2], r1
+    add         r3, r2, r1
+    vst2.8      {d14[0], d15[0]}, [r2]
+    vst2.8      {d14[1], d15[1]}, [r3], r1
+    add         r12, r3, r1
+    vst2.8      {d14[2], d15[2]}, [r3]
+    vst2.8      {d14[3], d15[3]}, [r12], r1
+    add         r0, r12, r1
+    vst2.8      {d14[4], d15[4]}, [r12]
+    vst2.8      {d14[5], d15[5]}, [r0], r1
+    add         r2, r0, r1
+    vst2.8      {d14[6], d15[6]}, [r0]
+    vst2.8      {d14[7], d15[7]}, [r2]
 
     bx          lr
     ENDP        ; |vp8_loop_filter_simple_vertical_edge_neon|
 
-; r0    unsigned char *y
-; r1    int ystride
-; r2    const unsigned char *blimit
+;-----------------
 
-|vp8_loop_filter_bvs_neon| PROC
-    push        {r4, lr}
-    ldrb        r3, [r2]                   ; load blim from mem
-    mov         r4, r0
-    add         r0, r0, #4
-    vdup.s8     q1, r3                     ; duplicate blim
-    bl          vp8_loop_filter_simple_vertical_edge_neon
-    ; vp8_loop_filter_simple_vertical_edge_neon preserves  r1 and q1
-    add         r0, r4, #8
-    bl          vp8_loop_filter_simple_vertical_edge_neon
-    add         r0, r4, #12
-    pop         {r4, lr}
-    b           vp8_loop_filter_simple_vertical_edge_neon
-    ENDP        ;|vp8_loop_filter_bvs_neon|
+_vlfy_coeff_
+    DCD     vlfy_coeff
+vlfy_coeff
+    DCD     0x80808080, 0x80808080, 0x80808080, 0x80808080
+    DCD     0x03030303, 0x03030303, 0x03030303, 0x03030303
+    DCD     0x04040404, 0x04040404, 0x04040404, 0x04040404
 
-; r0    unsigned char *y
-; r1    int ystride
-; r2    const unsigned char *blimit
-
-|vp8_loop_filter_mbvs_neon| PROC
-    ldrb        r3, [r2]                   ; load mblim from mem
-    vdup.s8     q1, r3                     ; duplicate mblim
-    b           vp8_loop_filter_simple_vertical_edge_neon
-    ENDP        ;|vp8_loop_filter_bvs_neon|
     END

vp8/common/arm/neon/mbloopfilter_neon.asm

@@ -14,143 +14,155 @@
     EXPORT  |vp8_mbloop_filter_vertical_edge_y_neon|
     EXPORT  |vp8_mbloop_filter_vertical_edge_uv_neon|
     ARM
+    REQUIRE8
+    PRESERVE8
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
 
+; flimit, limit, and thresh should be positive numbers.
+; All 16 elements in these variables are equal.
+
 ; void vp8_mbloop_filter_horizontal_edge_y_neon(unsigned char *src, int pitch,
-;                                               const unsigned char *blimit,
-;                                               const unsigned char *limit,
-;                                               const unsigned char *thresh)
+;                                               const signed char *flimit,
+;                                               const signed char *limit,
+;                                               const signed char *thresh,
+;                                               int count)
 ; r0    unsigned char *src,
 ; r1    int pitch,
-; r2    unsigned char blimit
-; r3    unsigned char limit
-; sp    unsigned char thresh,
+; r2    const signed char *flimit,
+; r3    const signed char *limit,
+; sp    const signed char *thresh,
+; sp+4  int count (unused)
 |vp8_mbloop_filter_horizontal_edge_y_neon| PROC
-    push        {lr}
-    add         r1, r1, r1                  ; double stride
-    ldr         r12, [sp, #4]               ; load thresh
-    sub         r0, r0, r1, lsl #1          ; move src pointer down by 4 lines
-    vdup.u8     q2, r12                     ; thresh
-    add         r12, r0, r1,  lsr #1        ; move src pointer up by 1 line
+    stmdb       sp!, {lr}
+    sub         r0, r0, r1, lsl #2          ; move src pointer down by 4 lines
+    ldr         r12, [sp, #4]               ; load thresh pointer
 
-    vld1.u8     {q3}, [r0@128], r1              ; p3
-    vld1.u8     {q4}, [r12@128], r1             ; p2
-    vld1.u8     {q5}, [r0@128], r1              ; p1
-    vld1.u8     {q6}, [r12@128], r1             ; p0
-    vld1.u8     {q7}, [r0@128], r1              ; q0
-    vld1.u8     {q8}, [r12@128], r1             ; q1
-    vld1.u8     {q9}, [r0@128], r1              ; q2
-    vld1.u8     {q10}, [r12@128], r1            ; q3
-
-    bl          vp8_mbloop_filter_neon
-
-    sub         r12, r12, r1, lsl #2
-    add         r0, r12, r1, lsr #1
-
-    vst1.u8     {q4}, [r12@128],r1         ; store op2
-    vst1.u8     {q5}, [r0@128],r1          ; store op1
-    vst1.u8     {q6}, [r12@128], r1        ; store op0
-    vst1.u8     {q7}, [r0@128],r1          ; store oq0
-    vst1.u8     {q8}, [r12@128]            ; store oq1
-    vst1.u8     {q9}, [r0@128]             ; store oq2
-
-    pop         {pc}
-    ENDP        ; |vp8_mbloop_filter_horizontal_edge_y_neon|
-
-; void vp8_mbloop_filter_horizontal_edge_uv_neon(unsigned char *u, int pitch,
-;                                                const unsigned char *blimit,
-;                                                const unsigned char *limit,
-;                                                const unsigned char *thresh,
-;                                                unsigned char *v)
-; r0    unsigned char *u,
-; r1    int pitch,
-; r2    unsigned char blimit
-; r3    unsigned char limit
-; sp    unsigned char thresh,
-; sp+4  unsigned char *v
-
-|vp8_mbloop_filter_horizontal_edge_uv_neon| PROC
-    push        {lr}
-    ldr         r12, [sp, #4]                 ; load thresh
-    sub         r0, r0, r1, lsl #2            ; move u pointer down by 4 lines
-    vdup.u8     q2, r12                       ; thresh
-    ldr         r12, [sp, #8]                 ; load v ptr
-    sub         r12, r12, r1, lsl #2          ; move v pointer down by 4 lines
-
-    vld1.u8     {d6}, [r0@64], r1              ; p3
-    vld1.u8     {d7}, [r12@64], r1              ; p3
-    vld1.u8     {d8}, [r0@64], r1              ; p2
-    vld1.u8     {d9}, [r12@64], r1              ; p2
-    vld1.u8     {d10}, [r0@64], r1             ; p1
-    vld1.u8     {d11}, [r12@64], r1             ; p1
-    vld1.u8     {d12}, [r0@64], r1             ; p0
-    vld1.u8     {d13}, [r12@64], r1             ; p0
-    vld1.u8     {d14}, [r0@64], r1             ; q0
-    vld1.u8     {d15}, [r12@64], r1             ; q0
-    vld1.u8     {d16}, [r0@64], r1             ; q1
-    vld1.u8     {d17}, [r12@64], r1             ; q1
-    vld1.u8     {d18}, [r0@64], r1             ; q2
-    vld1.u8     {d19}, [r12@64], r1             ; q2
-    vld1.u8     {d20}, [r0@64], r1             ; q3
-    vld1.u8     {d21}, [r12@64], r1             ; q3
+    vld1.u8     {q3}, [r0], r1              ; p3
+    vld1.s8     {d2[], d3[]}, [r3]          ; limit
+    vld1.u8     {q4}, [r0], r1              ; p2
+    vld1.s8     {d4[], d5[]}, [r12]         ; thresh
+    vld1.u8     {q5}, [r0], r1              ; p1
+    vld1.u8     {q6}, [r0], r1              ; p0
+    vld1.u8     {q7}, [r0], r1              ; q0
+    vld1.u8     {q8}, [r0], r1              ; q1
+    vld1.u8     {q9}, [r0], r1              ; q2
+    vld1.u8     {q10}, [r0], r1             ; q3
 
     bl          vp8_mbloop_filter_neon
 
     sub         r0, r0, r1, lsl #3
-    sub         r12, r12, r1, lsl #3
+    add         r0, r0, r1
+    add         r2, r0, r1
+    add         r3, r2, r1
+
+    vst1.u8     {q4}, [r0]                  ; store op2
+    vst1.u8     {q5}, [r2]                  ; store op1
+    vst1.u8     {q6}, [r3], r1              ; store op0
+    add         r12, r3, r1
+    vst1.u8     {q7}, [r3]                  ; store oq0
+    vst1.u8     {q8}, [r12], r1             ; store oq1
+    vst1.u8     {q9}, [r12]             ; store oq2
+
+    ldmia       sp!, {pc}
+    ENDP        ; |vp8_mbloop_filter_horizontal_edge_y_neon|
+
+; void vp8_mbloop_filter_horizontal_edge_uv_neon(unsigned char *u, int pitch,
+;                                                const signed char *flimit,
+;                                                const signed char *limit,
+;                                                const signed char *thresh,
+;                                                unsigned char *v)
+; r0    unsigned char *u,
+; r1    int pitch,
+; r2    const signed char *flimit,
+; r3    const signed char *limit,
+; sp    const signed char *thresh,
+; sp+4  unsigned char *v
+|vp8_mbloop_filter_horizontal_edge_uv_neon| PROC
+    stmdb       sp!, {lr}
+    sub         r0, r0, r1, lsl #2          ; move u pointer down by 4 lines
+    vld1.s8     {d2[], d3[]}, [r3]          ; limit
+    ldr         r3, [sp, #8]                ; load v ptr
+    ldr         r12, [sp, #4]               ; load thresh pointer
+    sub         r3, r3, r1, lsl #2          ; move v pointer down by 4 lines
+
+    vld1.u8     {d6}, [r0], r1              ; p3
+    vld1.u8     {d7}, [r3], r1              ; p3
+    vld1.u8     {d8}, [r0], r1              ; p2
+    vld1.u8     {d9}, [r3], r1              ; p2
+    vld1.u8     {d10}, [r0], r1             ; p1
+    vld1.u8     {d11}, [r3], r1             ; p1
+    vld1.u8     {d12}, [r0], r1             ; p0
+    vld1.u8     {d13}, [r3], r1             ; p0
+    vld1.u8     {d14}, [r0], r1             ; q0
+    vld1.u8     {d15}, [r3], r1             ; q0
+    vld1.u8     {d16}, [r0], r1             ; q1
+    vld1.u8     {d17}, [r3], r1             ; q1
+    vld1.u8     {d18}, [r0], r1             ; q2
+    vld1.u8     {d19}, [r3], r1             ; q2
+    vld1.u8     {d20}, [r0], r1             ; q3
+    vld1.u8     {d21}, [r3], r1             ; q3
+
+    vld1.s8     {d4[], d5[]}, [r12]         ; thresh
+
+    bl          vp8_mbloop_filter_neon
+
+    sub         r0, r0, r1, lsl #3
+    sub         r3, r3, r1, lsl #3
 
     add         r0, r0, r1
-    add         r12, r12, r1
+    add         r3, r3, r1
 
-    vst1.u8     {d8}, [r0@64], r1              ; store u op2
-    vst1.u8     {d9}, [r12@64], r1              ; store v op2
-    vst1.u8     {d10}, [r0@64], r1             ; store u op1
-    vst1.u8     {d11}, [r12@64], r1             ; store v op1
-    vst1.u8     {d12}, [r0@64], r1             ; store u op0
-    vst1.u8     {d13}, [r12@64], r1             ; store v op0
-    vst1.u8     {d14}, [r0@64], r1             ; store u oq0
-    vst1.u8     {d15}, [r12@64], r1             ; store v oq0
-    vst1.u8     {d16}, [r0@64], r1             ; store u oq1
-    vst1.u8     {d17}, [r12@64], r1             ; store v oq1
-    vst1.u8     {d18}, [r0@64], r1             ; store u oq2
-    vst1.u8     {d19}, [r12@64], r1             ; store v oq2
+    vst1.u8     {d8}, [r0], r1              ; store u op2
+    vst1.u8     {d9}, [r3], r1              ; store v op2
+    vst1.u8     {d10}, [r0], r1             ; store u op1
+    vst1.u8     {d11}, [r3], r1             ; store v op1
+    vst1.u8     {d12}, [r0], r1             ; store u op0
+    vst1.u8     {d13}, [r3], r1             ; store v op0
+    vst1.u8     {d14}, [r0], r1             ; store u oq0
+    vst1.u8     {d15}, [r3], r1             ; store v oq0
+    vst1.u8     {d16}, [r0], r1             ; store u oq1
+    vst1.u8     {d17}, [r3], r1             ; store v oq1
+    vst1.u8     {d18}, [r0], r1             ; store u oq2
+    vst1.u8     {d19}, [r3], r1             ; store v oq2
 
-    pop         {pc}
+    ldmia       sp!, {pc}
     ENDP        ; |vp8_mbloop_filter_horizontal_edge_uv_neon|
 
 ; void vp8_mbloop_filter_vertical_edge_y_neon(unsigned char *src, int pitch,
-;                                             const unsigned char *blimit,
-;                                             const unsigned char *limit,
-;                                             const unsigned char *thresh)
+;                                             const signed char *flimit,
+;                                             const signed char *limit,
+;                                             const signed char *thresh,
+;                                             int count)
 ; r0    unsigned char *src,
 ; r1    int pitch,
-; r2    unsigned char blimit
-; r3    unsigned char limit
-; sp    unsigned char thresh,
+; r2    const signed char *flimit,
+; r3    const signed char *limit,
+; sp    const signed char *thresh,
+; sp+4  int count (unused)
 |vp8_mbloop_filter_vertical_edge_y_neon| PROC
-    push        {lr}
-    ldr         r12, [sp, #4]               ; load thresh
+    stmdb       sp!, {lr}
     sub         r0, r0, #4                  ; move src pointer down by 4 columns
-    vdup.s8     q2, r12                     ; thresh
-    add         r12, r0, r1, lsl #3         ; move src pointer down by 8 lines
 
     vld1.u8     {d6}, [r0], r1              ; load first 8-line src data
-    vld1.u8     {d7}, [r12], r1             ; load second 8-line src data
+    ldr         r12, [sp, #4]               ; load thresh pointer
     vld1.u8     {d8}, [r0], r1
-    vld1.u8     {d9}, [r12], r1
+    sub         sp, sp, #32
     vld1.u8     {d10}, [r0], r1
-    vld1.u8     {d11}, [r12], r1
     vld1.u8     {d12}, [r0], r1
-    vld1.u8     {d13}, [r12], r1
     vld1.u8     {d14}, [r0], r1
-    vld1.u8     {d15}, [r12], r1
     vld1.u8     {d16}, [r0], r1
-    vld1.u8     {d17}, [r12], r1
     vld1.u8     {d18}, [r0], r1
-    vld1.u8     {d19}, [r12], r1
     vld1.u8     {d20}, [r0], r1
-    vld1.u8     {d21}, [r12], r1
+
+    vld1.u8     {d7}, [r0], r1              ; load second 8-line src data
+    vld1.u8     {d9}, [r0], r1
+    vld1.u8     {d11}, [r0], r1
+    vld1.u8     {d13}, [r0], r1
+    vld1.u8     {d15}, [r0], r1
+    vld1.u8     {d17}, [r0], r1
+    vld1.u8     {d19}, [r0], r1
+    vld1.u8     {d21}, [r0], r1
 
     ;transpose to 8x16 matrix
     vtrn.32     q3, q7
@@ -168,17 +180,29 @@
     vtrn.8      q7, q8
     vtrn.8      q9, q10
 
-    sub         r0, r0, r1, lsl #3
+    vld1.s8     {d4[], d5[]}, [r12]         ; thresh
+    vld1.s8     {d2[], d3[]}, [r3]          ; limit
+    mov         r12, sp
+    vst1.u8     {q3}, [r12]!
+    vst1.u8     {q10}, [r12]!
 
     bl          vp8_mbloop_filter_neon
 
-    sub         r12, r12, r1, lsl #3
+    sub         r0, r0, r1, lsl #4
+
+    add         r2, r0, r1
+
+    add         r3, r2, r1
+
+    vld1.u8     {q3}, [sp]!
+    vld1.u8     {q10}, [sp]!
 
     ;transpose to 16x8 matrix
     vtrn.32     q3, q7
     vtrn.32     q4, q8
     vtrn.32     q5, q9
     vtrn.32     q6, q10
+    add         r12, r3, r1
 
     vtrn.16     q3, q5
     vtrn.16     q4, q6
@@ -191,30 +215,36 @@
     vtrn.8      q9, q10
 
     ;store op2, op1, op0, oq0, oq1, oq2
-    vst1.8      {d6}, [r0], r1
-    vst1.8      {d7}, [r12], r1
-    vst1.8      {d8}, [r0], r1
-    vst1.8      {d9}, [r12], r1
-    vst1.8      {d10}, [r0], r1
-    vst1.8      {d11}, [r12], r1
-    vst1.8      {d12}, [r0], r1
-    vst1.8      {d13}, [r12], r1
-    vst1.8      {d14}, [r0], r1
-    vst1.8      {d15}, [r12], r1
+    vst1.8      {d6}, [r0]
+    vst1.8      {d8}, [r2]
+    vst1.8      {d10}, [r3]
+    vst1.8      {d12}, [r12], r1
+    add         r0, r12, r1
+    vst1.8      {d14}, [r12]
     vst1.8      {d16}, [r0], r1
-    vst1.8      {d17}, [r12], r1
-    vst1.8      {d18}, [r0], r1
-    vst1.8      {d19}, [r12], r1
-    vst1.8      {d20}, [r0]
-    vst1.8      {d21}, [r12]
+    add         r2, r0, r1
+    vst1.8      {d18}, [r0]
+    vst1.8      {d20}, [r2], r1
+    add         r3, r2, r1
+    vst1.8      {d7}, [r2]
+    vst1.8      {d9}, [r3], r1
+    add         r12, r3, r1
+    vst1.8      {d11}, [r3]
+    vst1.8      {d13}, [r12], r1
+    add         r0, r12, r1
+    vst1.8      {d15}, [r12]
+    vst1.8      {d17}, [r0], r1
+    add         r2, r0, r1
+    vst1.8      {d19}, [r0]
+    vst1.8      {d21}, [r2]
 
-    pop         {pc}
+    ldmia       sp!, {pc}
     ENDP        ; |vp8_mbloop_filter_vertical_edge_y_neon|
 
 ; void vp8_mbloop_filter_vertical_edge_uv_neon(unsigned char *u, int pitch,
-;                                              const unsigned char *blimit,
-;                                              const unsigned char *limit,
-;                                              const unsigned char *thresh,
+;                                              const signed char *flimit,
+;                                              const signed char *limit,
+;                                              const signed char *thresh,
 ;                                              unsigned char *v)
 ; r0    unsigned char *u,
 ; r1    int pitch,
@@ -223,29 +253,30 @@
 ; sp    const signed char *thresh,
 ; sp+4  unsigned char *v
 |vp8_mbloop_filter_vertical_edge_uv_neon| PROC
-    push        {lr}
-    ldr         r12, [sp, #4]               ; load thresh
-    sub         r0, r0, #4                  ; move u pointer down by 4 columns
-    vdup.u8     q2, r12                     ; thresh
-    ldr         r12, [sp, #8]               ; load v ptr
-    sub         r12, r12, #4                ; move v pointer down by 4 columns
+    stmdb       sp!, {lr}
+    sub         r0, r0, #4                  ; move src pointer down by 4 columns
+    vld1.s8     {d2[], d3[]}, [r3]          ; limit
+    ldr         r3, [sp, #8]                ; load v ptr
+    ldr         r12, [sp, #4]               ; load thresh pointer
+
+    sub         r3, r3, #4                  ; move v pointer down by 4 columns
 
     vld1.u8     {d6}, [r0], r1              ;load u data
-    vld1.u8     {d7}, [r12], r1             ;load v data
+    vld1.u8     {d7}, [r3], r1              ;load v data
     vld1.u8     {d8}, [r0], r1
-    vld1.u8     {d9}, [r12], r1
+    vld1.u8     {d9}, [r3], r1
     vld1.u8     {d10}, [r0], r1
-    vld1.u8     {d11}, [r12], r1
+    vld1.u8     {d11}, [r3], r1
     vld1.u8     {d12}, [r0], r1
-    vld1.u8     {d13}, [r12], r1
+    vld1.u8     {d13}, [r3], r1
     vld1.u8     {d14}, [r0], r1
-    vld1.u8     {d15}, [r12], r1
+    vld1.u8     {d15}, [r3], r1
     vld1.u8     {d16}, [r0], r1
-    vld1.u8     {d17}, [r12], r1
+    vld1.u8     {d17}, [r3], r1
     vld1.u8     {d18}, [r0], r1
-    vld1.u8     {d19}, [r12], r1
+    vld1.u8     {d19}, [r3], r1
     vld1.u8     {d20}, [r0], r1
-    vld1.u8     {d21}, [r12], r1
+    vld1.u8     {d21}, [r3], r1
 
     ;transpose to 8x16 matrix
     vtrn.32     q3, q7
@@ -263,11 +294,19 @@
     vtrn.8      q7, q8
     vtrn.8      q9, q10
 
-    sub         r0, r0, r1, lsl #3
+    sub         sp, sp, #32
+    vld1.s8     {d4[], d5[]}, [r12]         ; thresh
+    mov         r12, sp
+    vst1.u8     {q3}, [r12]!
+    vst1.u8     {q10}, [r12]!
 
     bl          vp8_mbloop_filter_neon
 
-    sub         r12, r12, r1, lsl #3
+    sub         r0, r0, r1, lsl #3
+    sub         r3, r3, r1, lsl #3
+
+    vld1.u8     {q3}, [sp]!
+    vld1.u8     {q10}, [sp]!
 
     ;transpose to 16x8 matrix
     vtrn.32     q3, q7
@@ -287,23 +326,23 @@
 
     ;store op2, op1, op0, oq0, oq1, oq2
     vst1.8      {d6}, [r0], r1
-    vst1.8      {d7}, [r12], r1
+    vst1.8      {d7}, [r3], r1
     vst1.8      {d8}, [r0], r1
-    vst1.8      {d9}, [r12], r1
+    vst1.8      {d9}, [r3], r1
     vst1.8      {d10}, [r0], r1
-    vst1.8      {d11}, [r12], r1
+    vst1.8      {d11}, [r3], r1
     vst1.8      {d12}, [r0], r1
-    vst1.8      {d13}, [r12], r1
+    vst1.8      {d13}, [r3], r1
     vst1.8      {d14}, [r0], r1
-    vst1.8      {d15}, [r12], r1
+    vst1.8      {d15}, [r3], r1
     vst1.8      {d16}, [r0], r1
-    vst1.8      {d17}, [r12], r1
+    vst1.8      {d17}, [r3], r1
     vst1.8      {d18}, [r0], r1
-    vst1.8      {d19}, [r12], r1
-    vst1.8      {d20}, [r0]
-    vst1.8      {d21}, [r12]
+    vst1.8      {d19}, [r3], r1
+    vst1.8      {d20}, [r0], r1
+    vst1.8      {d21}, [r3], r1
 
-    pop         {pc}
+    ldmia       sp!, {pc}
     ENDP        ; |vp8_mbloop_filter_vertical_edge_uv_neon|
 
 ; void vp8_mbloop_filter_neon()
@@ -311,33 +350,41 @@
 ; functions do the necessary load, transpose (if necessary), preserve (if
 ; necessary) and store.
 
-; r0,r1 PRESERVE
-; r2    mblimit
-; r3    limit
+; TODO:
+; The vertical filter writes p3/q3 back out because two 4 element writes are
+; much simpler than ordering and writing two 3 element sets (or three 2 elements
+; sets, or whichever other combinations are possible).
+; If we can preserve q3 and q10, the vertical filter will be able to avoid
+; storing those values on the stack and reading them back after the filter.
 
+; r0,r1 PRESERVE
+; r2    flimit
+; r3    PRESERVE
+; q1    limit
 ; q2    thresh
-; q3    p3 PRESERVE
+; q3    p3
 ; q4    p2
 ; q5    p1
 ; q6    p0
 ; q7    q0
 ; q8    q1
 ; q9    q2
-; q10   q3 PRESERVE
+; q10   q3
 
 |vp8_mbloop_filter_neon| PROC
+    ldr         r12, _mblf_coeff_
 
     ; vp8_filter_mask
     vabd.u8     q11, q3, q4                 ; abs(p3 - p2)
     vabd.u8     q12, q4, q5                 ; abs(p2 - p1)
     vabd.u8     q13, q5, q6                 ; abs(p1 - p0)
     vabd.u8     q14, q8, q7                 ; abs(q1 - q0)
-    vabd.u8     q1, q9, q8                  ; abs(q2 - q1)
+    vabd.u8     q3, q9, q8                  ; abs(q2 - q1)
     vabd.u8     q0, q10, q9                 ; abs(q3 - q2)
 
     vmax.u8     q11, q11, q12
     vmax.u8     q12, q13, q14
-    vmax.u8     q1, q1, q0
+    vmax.u8     q3, q3, q0
     vmax.u8     q15, q11, q12
 
     vabd.u8     q12, q6, q7                 ; abs(p0 - q0)
@@ -345,53 +392,51 @@
     ; vp8_hevmask
     vcgt.u8     q13, q13, q2                ; (abs(p1 - p0) > thresh) * -1
     vcgt.u8     q14, q14, q2                ; (abs(q1 - q0) > thresh) * -1
-    vmax.u8     q15, q15, q1
+    vmax.u8     q15, q15, q3
 
-    vdup.u8     q1, r3                      ; limit
-    vdup.u8     q2, r2                      ; mblimit
+    vld1.s8     {d4[], d5[]}, [r2]          ; flimit
 
-    vmov.u8     q0, #0x80                   ; 0x80
+    vld1.u8     {q0}, [r12]!
 
+    vadd.u8     q2, q2, q2                  ; flimit * 2
+    vadd.u8     q2, q2, q1                  ; flimit * 2 +  limit
     vcge.u8     q15, q1, q15
 
     vabd.u8     q1, q5, q8                  ; a = abs(p1 - q1)
     vqadd.u8    q12, q12, q12               ; b = abs(p0 - q0) * 2
-    vmov.u16    q11, #3                     ; #3
+    vshr.u8     q1, q1, #1                  ; a = a / 2
+    vqadd.u8    q12, q12, q1                ; a = b + a
+    vcge.u8     q12, q2, q12                ; (a > flimit * 2 + limit) * -1
 
     ; vp8_filter
     ; convert to signed
     veor        q7, q7, q0                  ; qs0
-    vshr.u8     q1, q1, #1                  ; a = a / 2
     veor        q6, q6, q0                  ; ps0
     veor        q5, q5, q0                  ; ps1
-
-    vqadd.u8    q12, q12, q1                ; a = b + a
-
     veor        q8, q8, q0                  ; qs1
     veor        q4, q4, q0                  ; ps2
     veor        q9, q9, q0                  ; qs2
 
     vorr        q14, q13, q14               ; vp8_hevmask
 
-    vcge.u8     q12, q2, q12                ; (a > flimit * 2 + limit) * -1
-
     vsubl.s8    q2, d14, d12                ; qs0 - ps0
     vsubl.s8    q13, d15, d13
 
     vqsub.s8    q1, q5, q8                  ; vp8_filter = clamp(ps1-qs1)
 
-    vmul.i16    q2, q2, q11                 ; 3 * ( qs0 - ps0)
-
+    vadd.s16    q10, q2, q2                 ; 3 * (qs0 - ps0)
+    vadd.s16    q11, q13, q13
     vand        q15, q15, q12               ; vp8_filter_mask
 
-    vmul.i16    q13, q13, q11
+    vadd.s16    q2, q2, q10
+    vadd.s16    q13, q13, q11
 
-    vmov.u8     q12, #3                     ; #3
+    vld1.u8     {q12}, [r12]!               ; #3
 
     vaddw.s8    q2, q2, d2                  ; vp8_filter + 3 * ( qs0 - ps0)
     vaddw.s8    q13, q13, d3
 
-    vmov.u8     q11, #4                     ; #4
+    vld1.u8     {q11}, [r12]!               ; #4
 
     ; vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
     vqmovn.s16  d2, q2
@@ -399,23 +444,27 @@
 
     vand        q1, q1, q15                 ; vp8_filter &= mask
 
-    vmov.u16    q15, #63                    ; #63
-
+    vld1.u8     {q15}, [r12]!               ; #63
+    ;
     vand        q13, q1, q14                ; Filter2 &= hev
 
+    vld1.u8     {d7}, [r12]!                ; #9
+
     vqadd.s8    q2, q13, q11                ; Filter1 = clamp(Filter2+4)
     vqadd.s8    q13, q13, q12               ; Filter2 = clamp(Filter2+3)
 
-    vmov        q0, q15
+    vld1.u8     {d6}, [r12]!                ; #18
 
     vshr.s8     q2, q2, #3                  ; Filter1 >>= 3
     vshr.s8     q13, q13, #3                ; Filter2 >>= 3
 
-    vmov        q11, q15
+    vmov        q10, q15
     vmov        q12, q15
 
     vqsub.s8    q7, q7, q2                  ; qs0 = clamp(qs0 - Filter1)
 
+    vld1.u8     {d5}, [r12]!                ; #27
+
     vqadd.s8    q6, q6, q13                 ; ps0 = clamp(ps0 + Filter2)
 
     vbic        q1, q1, q14                 ; vp8_filter &= ~hev
@@ -423,47 +472,49 @@
     ; roughly 1/7th difference across boundary
     ; roughly 2/7th difference across boundary
     ; roughly 3/7th difference across boundary
-
-    vmov.u8     d5, #9                      ; #9
-    vmov.u8     d4, #18                     ; #18
-
+    vmov        q11, q15
     vmov        q13, q15
     vmov        q14, q15
 
-    vmlal.s8    q0, d2, d5                  ; 63 + Filter2 * 9
-    vmlal.s8    q11, d3, d5
-    vmov.u8     d5, #27                     ; #27
-    vmlal.s8    q12, d2, d4                 ; 63 + Filter2 * 18
-    vmlal.s8    q13, d3, d4
-    vmlal.s8    q14, d2, d5                 ; 63 + Filter2 * 27
+    vmlal.s8    q10, d2, d7                 ; Filter2 * 9
+    vmlal.s8    q11, d3, d7
+    vmlal.s8    q12, d2, d6                 ; Filter2 * 18
+    vmlal.s8    q13, d3, d6
+    vmlal.s8    q14, d2, d5                 ; Filter2 * 27
     vmlal.s8    q15, d3, d5
-
-    vqshrn.s16  d0, q0, #7                  ; u = clamp((63 + Filter2 * 9)>>7)
-    vqshrn.s16  d1, q11, #7
+    vqshrn.s16  d20, q10, #7                ; u = clamp((63 + Filter2 * 9)>>7)
+    vqshrn.s16  d21, q11, #7
     vqshrn.s16  d24, q12, #7                ; u = clamp((63 + Filter2 * 18)>>7)
     vqshrn.s16  d25, q13, #7
     vqshrn.s16  d28, q14, #7                ; u = clamp((63 + Filter2 * 27)>>7)
     vqshrn.s16  d29, q15, #7
 
-    vmov.u8     q1, #0x80                   ; 0x80
-
-    vqsub.s8    q11, q9, q0                 ; s = clamp(qs2 - u)
-    vqadd.s8    q0, q4, q0                  ; s = clamp(ps2 + u)
+    vqsub.s8    q11, q9, q10                ; s = clamp(qs2 - u)
+    vqadd.s8    q10, q4, q10                ; s = clamp(ps2 + u)
     vqsub.s8    q13, q8, q12                ; s = clamp(qs1 - u)
     vqadd.s8    q12, q5, q12                ; s = clamp(ps1 + u)
     vqsub.s8    q15, q7, q14                ; s = clamp(qs0 - u)
     vqadd.s8    q14, q6, q14                ; s = clamp(ps0 + u)
-
-    veor        q9, q11, q1                 ; *oq2 = s^0x80
-    veor        q4, q0, q1                  ; *op2 = s^0x80
-    veor        q8, q13, q1                 ; *oq1 = s^0x80
-    veor        q5, q12, q1                 ; *op2 = s^0x80
-    veor        q7, q15, q1                 ; *oq0 = s^0x80
-    veor        q6, q14, q1                 ; *op0 = s^0x80
+    veor        q9, q11, q0                 ; *oq2 = s^0x80
+    veor        q4, q10, q0                 ; *op2 = s^0x80
+    veor        q8, q13, q0                 ; *oq1 = s^0x80
+    veor        q5, q12, q0                 ; *op2 = s^0x80
+    veor        q7, q15, q0                 ; *oq0 = s^0x80
+    veor        q6, q14, q0                 ; *op0 = s^0x80
 
     bx          lr
     ENDP        ; |vp8_mbloop_filter_neon|
 
 ;-----------------
 
+_mblf_coeff_
+    DCD     mblf_coeff
+mblf_coeff
+    DCD     0x80808080, 0x80808080, 0x80808080, 0x80808080
+    DCD     0x03030303, 0x03030303, 0x03030303, 0x03030303
+    DCD     0x04040404, 0x04040404, 0x04040404, 0x04040404
+    DCD     0x003f003f, 0x003f003f, 0x003f003f, 0x003f003f
+    DCD     0x09090909, 0x09090909, 0x12121212, 0x12121212
+    DCD     0x1b1b1b1b, 0x1b1b1b1b
+
     END

vp8/common/arm/neon/shortidct4x4llm_neon.asm

@@ -31,7 +31,7 @@
 ;result of the multiplication that is needed in IDCT.
 
 |vp8_short_idct4x4llm_neon| PROC
-    adr             r12, idct_coeff
+    ldr             r12, _idct_coeff_
     vld1.16         {q1, q2}, [r0]
     vld1.16         {d0}, [r12]
 
@@ -114,6 +114,8 @@
 
 ;-----------------
 
+_idct_coeff_
+    DCD     idct_coeff
 idct_coeff
     DCD     0x4e7b4e7b, 0x8a8c8a8c
 

vp8/common/arm/neon/sixtappredict16x16_neon.asm

@@ -15,17 +15,6 @@
     PRESERVE8
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
-
-filter16_coeff
-    DCD     0,  0,  128,    0,   0,  0,   0,  0
-    DCD     0, -6,  123,   12,  -1,  0,   0,  0
-    DCD     2, -11, 108,   36,  -8,  1,   0,  0
-    DCD     0, -9,   93,   50,  -6,  0,   0,  0
-    DCD     3, -16,  77,   77, -16,  3,   0,  0
-    DCD     0, -6,   50,   93,  -9,  0,   0,  0
-    DCD     1, -8,   36,  108, -11,  2,   0,  0
-    DCD     0, -1,   12,  123,  -6,   0,  0,  0
-
 ; r0    unsigned char  *src_ptr,
 ; r1    int  src_pixels_per_line,
 ; r2    int  xoffset,
@@ -44,7 +33,7 @@ filter16_coeff
 |vp8_sixtap_predict16x16_neon| PROC
     push            {r4-r5, lr}
 
-    adr             r12, filter16_coeff
+    ldr             r12, _filter16_coeff_
     ldr             r4, [sp, #12]           ;load parameters from stack
     ldr             r5, [sp, #16]           ;load parameters from stack
 
@@ -487,4 +476,17 @@ secondpass_only_inner_loop_neon
     ENDP
 
 ;-----------------
+
+_filter16_coeff_
+    DCD     filter16_coeff
+filter16_coeff
+    DCD     0,  0,  128,    0,   0,  0,   0,  0
+    DCD     0, -6,  123,   12,  -1,  0,   0,  0
+    DCD     2, -11, 108,   36,  -8,  1,   0,  0
+    DCD     0, -9,   93,   50,  -6,  0,   0,  0
+    DCD     3, -16,  77,   77, -16,  3,   0,  0
+    DCD     0, -6,   50,   93,  -9,  0,   0,  0
+    DCD     1, -8,   36,  108, -11,  2,   0,  0
+    DCD     0, -1,   12,  123,  -6,   0,  0,  0
+
     END

vp8/common/arm/neon/sixtappredict4x4_neon.asm

@@ -15,17 +15,6 @@
     PRESERVE8
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
-
-filter4_coeff
-    DCD     0,  0,  128,    0,   0,  0,   0,  0
-    DCD     0, -6,  123,   12,  -1,  0,   0,  0
-    DCD     2, -11, 108,   36,  -8,  1,   0,  0
-    DCD     0, -9,   93,   50,  -6,  0,   0,  0
-    DCD     3, -16,  77,   77, -16,  3,   0,  0
-    DCD     0, -6,   50,   93,  -9,  0,   0,  0
-    DCD     1, -8,   36,  108, -11,  2,   0,  0
-    DCD     0, -1,   12,  123,  -6,   0,  0,  0
-
 ; r0    unsigned char  *src_ptr,
 ; r1    int  src_pixels_per_line,
 ; r2    int  xoffset,
@@ -36,7 +25,7 @@ filter4_coeff
 |vp8_sixtap_predict_neon| PROC
     push            {r4, lr}
 
-    adr             r12, filter4_coeff
+    ldr             r12, _filter4_coeff_
     ldr             r4, [sp, #8]            ;load parameters from stack
     ldr             lr, [sp, #12]           ;load parameters from stack
 
@@ -419,4 +408,16 @@ secondpass_filter4x4_only
 
 ;-----------------
 
+_filter4_coeff_
+    DCD     filter4_coeff
+filter4_coeff
+    DCD     0,  0,  128,    0,   0,  0,   0,  0
+    DCD     0, -6,  123,   12,  -1,  0,   0,  0
+    DCD     2, -11, 108,   36,  -8,  1,   0,  0
+    DCD     0, -9,   93,   50,  -6,  0,   0,  0
+    DCD     3, -16,  77,   77, -16,  3,   0,  0
+    DCD     0, -6,   50,   93,  -9,  0,   0,  0
+    DCD     1, -8,   36,  108, -11,  2,   0,  0
+    DCD     0, -1,   12,  123,  -6,   0,  0,  0
+
     END

vp8/common/arm/neon/sixtappredict8x4_neon.asm

@@ -15,17 +15,6 @@
     PRESERVE8
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
-
-filter8_coeff
-    DCD     0,  0,  128,    0,   0,  0,   0,  0
-    DCD     0, -6,  123,   12,  -1,  0,   0,  0
-    DCD     2, -11, 108,   36,  -8,  1,   0,  0
-    DCD     0, -9,   93,   50,  -6,  0,   0,  0
-    DCD     3, -16,  77,   77, -16,  3,   0,  0
-    DCD     0, -6,   50,   93,  -9,  0,   0,  0
-    DCD     1, -8,   36,  108, -11,  2,   0,  0
-    DCD     0, -1,   12,  123,  -6,   0,  0,  0
-
 ; r0    unsigned char  *src_ptr,
 ; r1    int  src_pixels_per_line,
 ; r2    int  xoffset,
@@ -36,7 +25,7 @@ filter8_coeff
 |vp8_sixtap_predict8x4_neon| PROC
     push            {r4-r5, lr}
 
-    adr             r12, filter8_coeff
+    ldr             r12, _filter8_coeff_
     ldr             r4, [sp, #12]           ;load parameters from stack
     ldr             r5, [sp, #16]           ;load parameters from stack
 
@@ -470,4 +459,16 @@ secondpass_filter8x4_only
 
 ;-----------------
 
+_filter8_coeff_
+    DCD     filter8_coeff
+filter8_coeff
+    DCD     0,  0,  128,    0,   0,  0,   0,  0
+    DCD     0, -6,  123,   12,  -1,  0,   0,  0
+    DCD     2, -11, 108,   36,  -8,  1,   0,  0
+    DCD     0, -9,   93,   50,  -6,  0,   0,  0
+    DCD     3, -16,  77,   77, -16,  3,   0,  0
+    DCD     0, -6,   50,   93,  -9,  0,   0,  0
+    DCD     1, -8,   36,  108, -11,  2,   0,  0
+    DCD     0, -1,   12,  123,  -6,   0,  0,  0
+
     END

vp8/common/arm/neon/sixtappredict8x8_neon.asm

@@ -15,17 +15,6 @@
     PRESERVE8
 
     AREA ||.text||, CODE, READONLY, ALIGN=2
-
-filter8_coeff
-    DCD     0,  0,  128,    0,   0,  0,   0,  0
-    DCD     0, -6,  123,   12,  -1,  0,   0,  0
-    DCD     2, -11, 108,   36,  -8,  1,   0,  0
-    DCD     0, -9,   93,   50,  -6,  0,   0,  0
-    DCD     3, -16,  77,   77, -16,  3,   0,  0
-    DCD     0, -6,   50,   93,  -9,  0,   0,  0
-    DCD     1, -8,   36,  108, -11,  2,   0,  0
-    DCD     0, -1,   12,  123,  -6,   0,  0,  0
-
 ; r0    unsigned char  *src_ptr,
 ; r1    int  src_pixels_per_line,
 ; r2    int  xoffset,
@@ -36,7 +25,7 @@ filter8_coeff
 |vp8_sixtap_predict8x8_neon| PROC
     push            {r4-r5, lr}
 
-    adr             r12, filter8_coeff
+    ldr             r12, _filter8_coeff_
 
     ldr             r4, [sp, #12]           ;load parameters from stack
     ldr             r5, [sp, #16]           ;load parameters from stack
@@ -521,4 +510,16 @@ filt_blk2d_spo8x8_loop_neon
 
 ;-----------------
 
+_filter8_coeff_
+    DCD     filter8_coeff
+filter8_coeff
+    DCD     0,  0,  128,    0,   0,  0,   0,  0
+    DCD     0, -6,  123,   12,  -1,  0,   0,  0
+    DCD     2, -11, 108,   36,  -8,  1,   0,  0
+    DCD     0, -9,   93,   50,  -6,  0,   0,  0
+    DCD     3, -16,  77,   77, -16,  3,   0,  0
+    DCD     0, -6,   50,   93,  -9,  0,   0,  0
+    DCD     1, -8,   36,  108, -11,  2,   0,  0
+    DCD     0, -1,   12,  123,  -6,   0,  0,  0
+
     END

vp8/common/asm_com_offsets.c

@@ -9,14 +9,27 @@
  */
 
 
-#include "vpx_config.h"
-#include "vpx/vpx_codec.h"
-#include "vpx_ports/asm_offsets.h"
+#include "vpx_ports/config.h"
+#include <stddef.h>
+
 #include "vpx_scale/yv12config.h"
 
-BEGIN
+#define ct_assert(name,cond) \
+    static void assert_##name(void) UNUSED;\
+    static void assert_##name(void) {switch(0){case 0:case !!(cond):;}}
 
-/* vpx_scale */
+#define DEFINE(sym, val) int sym = val;
+
+/*
+#define BLANK() asm volatile("\n->" : : )
+*/
+
+/*
+ * int main(void)
+ * {
+ */
+
+//vpx_scale
 DEFINE(yv12_buffer_config_y_width,              offsetof(YV12_BUFFER_CONFIG, y_width));
 DEFINE(yv12_buffer_config_y_height,             offsetof(YV12_BUFFER_CONFIG, y_height));
 DEFINE(yv12_buffer_config_y_stride,             offsetof(YV12_BUFFER_CONFIG, y_stride));
@@ -27,14 +40,10 @@ DEFINE(yv12_buffer_config_y_buffer,             offsetof(YV12_BUFFER_CONFIG, y_b
 DEFINE(yv12_buffer_config_u_buffer,             offsetof(YV12_BUFFER_CONFIG, u_buffer));
 DEFINE(yv12_buffer_config_v_buffer,             offsetof(YV12_BUFFER_CONFIG, v_buffer));
 DEFINE(yv12_buffer_config_border,               offsetof(YV12_BUFFER_CONFIG, border));
-DEFINE(VP8BORDERINPIXELS_VAL,                   VP8BORDERINPIXELS);
 
-END
-
-/* add asserts for any offset that is not supported by assembly code */
-/* add asserts for any size that is not supported by assembly code */
-
-#if HAVE_ARMV7
-/* vp8_yv12_extend_frame_borders_neon makes several assumptions based on this */
-ct_assert(VP8BORDERINPIXELS_VAL, VP8BORDERINPIXELS == 32)
-#endif
+//add asserts for any offset that is not supported by assembly code
+//add asserts for any size that is not supported by assembly code
+/*
+ * return 0;
+ * }
+ */

vp8/common/blockd.h

@@ -14,12 +14,17 @@
 
 void vpx_log(const char *format, ...);
 
-#include "vpx_ports/config.h"
-#include "vpx_scale/yv12config.h"
+#include "../../vpx_ports/config.h"
+#include "../../vpx_scale/yv12config.h"
 #include "mv.h"
 #include "treecoder.h"
 #include "subpixel.h"
-#include "vpx_ports/mem.h"
+#include "../../vpx_ports/mem.h"
+
+#include "../../vpx_config.h"
+#if CONFIG_OPENCL
+#include "opencl/vp8_opencl.h"
+#endif
 
 #define TRUE    1
 #define FALSE   0
@@ -73,19 +78,19 @@ typedef enum
 
 typedef enum
 {
-    DC_PRED,            /* average of above and left pixels */
-    V_PRED,             /* vertical prediction */
-    H_PRED,             /* horizontal prediction */
-    TM_PRED,            /* Truemotion prediction */
-    B_PRED,             /* block based prediction, each block has its own prediction mode */
+    DC_PRED = 0,            /* average of above and left pixels */
+    V_PRED = 1,             /* vertical prediction */
+    H_PRED = 2,             /* horizontal prediction */
+    TM_PRED = 3,            /* Truemotion prediction */
+    B_PRED = 4,             /* block based prediction, each block has its own prediction mode */
 
-    NEARESTMV,
-    NEARMV,
-    ZEROMV,
-    NEWMV,
-    SPLITMV,
+    NEARESTMV = 5,
+    NEARMV = 6,
+    ZEROMV = 7,
+    NEWMV = 8,
+    SPLITMV = 9,
 
-    MB_MODE_COUNT
+    MB_MODE_COUNT = 10
 } MB_PREDICTION_MODE;
 
 /* Macroblock level features */
@@ -137,11 +142,16 @@ typedef enum
    modes for the Y blocks to the left and above us; for interframes, there
    is a single probability table. */
 
-union b_mode_info
+typedef struct
 {
-    B_PREDICTION_MODE as_mode;
-    int_mv mv;
-};
+    B_PREDICTION_MODE mode;
+    union
+    {
+        int as_int;
+        MV  as_mv;
+    } mv;
+} B_MODE_INFO;
+
 
 typedef enum
 {
@@ -156,43 +166,79 @@ typedef struct
 {
     MB_PREDICTION_MODE mode, uv_mode;
     MV_REFERENCE_FRAME ref_frame;
-    int_mv mv;
+    union
+    {
+        int as_int;
+        MV  as_mv;
+    } mv;
 
     unsigned char partitioning;
     unsigned char mb_skip_coeff;                                /* does this mb has coefficients at all, 1=no coefficients, 0=need decode tokens */
+    unsigned char dc_diff;
     unsigned char need_to_clamp_mvs;
+
     unsigned char segment_id;                  /* Which set of segmentation parameters should be used for this MB */
+
+    unsigned char force_no_skip; /* encoder only */
 } MB_MODE_INFO;
 
+
 typedef struct
 {
     MB_MODE_INFO mbmi;
-    union b_mode_info bmi[16];
+    B_MODE_INFO bmi[16];
 } MODE_INFO;
 
+
 typedef struct
 {
-    short *qcoeff;
-    short *dqcoeff;
-    unsigned char  *predictor;
-    short *diff;
+    short *qcoeff_base;
+    int qcoeff_offset;
+
+    short *dqcoeff_base;
+    int dqcoeff_offset;
+
+    unsigned char *predictor_base;
+    int predictor_offset;
+
+    short *diff_base;
+    int diff_offset;
+
     short *dequant;
 
+#if CONFIG_OPENCL
+    cl_command_queue cl_commands; //pointer to macroblock CL command queue
+
+    cl_mem cl_diff_mem;
+    cl_mem cl_predictor_mem;
+    cl_mem cl_qcoeff_mem;
+    cl_mem cl_dqcoeff_mem;
+    cl_mem cl_eobs_mem;
+
+    cl_mem cl_dequant_mem; //Block-specific, not shared
+
+    cl_bool sixtap_filter; //Subpixel Prediction type (true=sixtap, false=bilinear)
+
+#endif
+
     /* 16 Y blocks, 4 U blocks, 4 V blocks each with 16 entries */
-    unsigned char **base_pre;
+    unsigned char **base_pre; //previous frame, same Macroblock, base pointer
     int pre;
     int pre_stride;
 
-    unsigned char **base_dst;
+    unsigned char **base_dst; //destination base pointer
     int dst;
     int dst_stride;
 
-    int eob;
+    int eob; //only used in encoder? Decoder uses MBD.eobs
+
+    char *eobs_base; //beginning of MB.eobs
+
+    B_MODE_INFO bmi;
 
-    union b_mode_info bmi;
 } BLOCKD;
 
-typedef struct MacroBlockD
+typedef struct
 {
     DECLARE_ALIGNED(16, short, diff[400]);      /* from idct diff */
     DECLARE_ALIGNED(16, unsigned char,  predictor[384]);
@@ -200,11 +246,22 @@ typedef struct MacroBlockD
     DECLARE_ALIGNED(16, short, dqcoeff[400]);
     DECLARE_ALIGNED(16, char,  eobs[25]);
 
+#if CONFIG_OPENCL
+    cl_command_queue cl_commands; //Each macroblock gets its own command queue.
+    cl_mem cl_diff_mem;
+    cl_mem cl_predictor_mem;
+    cl_mem cl_qcoeff_mem;
+    cl_mem cl_dqcoeff_mem;
+    cl_mem cl_eobs_mem;
+
+    cl_bool sixtap_filter;
+#endif
+
     /* 16 Y blocks, 4 U, 4 V, 1 DC 2nd order block, each with 16 entries. */
     BLOCKD block[25];
 
     YV12_BUFFER_CONFIG pre; /* Filtered copy of previous frame reconstruction */
-    YV12_BUFFER_CONFIG dst;
+    YV12_BUFFER_CONFIG dst; /* Destination buffer for current frame */
 
     MODE_INFO *mode_info_context;
     int mode_info_stride;
@@ -252,11 +309,9 @@ typedef struct MacroBlockD
     int mb_to_top_edge;
     int mb_to_bottom_edge;
 
-    int ref_frame_cost[MAX_REF_FRAMES];
-
-
     unsigned int frames_since_golden;
     unsigned int frames_till_alt_ref_frame;
+
     vp8_subpix_fn_t  subpixel_predict;
     vp8_subpix_fn_t  subpixel_predict8x4;
     vp8_subpix_fn_t  subpixel_predict8x8;
@@ -266,14 +321,6 @@ typedef struct MacroBlockD
 
     int corrupted;
 
-#if ARCH_X86 || ARCH_X86_64
-    /* This is an intermediate buffer currently used in sub-pixel motion search
-     * to keep a copy of the reference area. This buffer can be used for other
-     * purpose.
-     */
-    DECLARE_ALIGNED(32, unsigned char, y_buf[22*32]);
-#endif
-
 #if CONFIG_RUNTIME_CPU_DETECT
     struct VP8_COMMON_RTCD  *rtcd;
 #endif
@@ -283,20 +330,4 @@ typedef struct MacroBlockD
 extern void vp8_build_block_doffsets(MACROBLOCKD *x);
 extern void vp8_setup_block_dptrs(MACROBLOCKD *x);
 
-static void update_blockd_bmi(MACROBLOCKD *xd)
-{
-    int i;
-    int is_4x4;
-    is_4x4 = (xd->mode_info_context->mbmi.mode == SPLITMV) ||
-              (xd->mode_info_context->mbmi.mode == B_PRED);
-
-    if (is_4x4)
-    {
-        for (i = 0; i < 16; i++)
-        {
-            xd->block[i].bmi = xd->mode_info_context->bmi[i];
-        }
-    }
-}
-
 #endif  /* __INC_BLOCKD_H */

vp8/common/coefupdateprobs.h

@@ -12,7 +12,7 @@
 /* Update probabilities for the nodes in the token entropy tree.
    Generated file included by entropy.c */
 
-const vp8_prob vp8_coef_update_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES] =
+const vp8_prob vp8_coef_update_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens-1] =
 {
     {
         {

vp8/common/debugmodes.c

@@ -97,7 +97,7 @@ void vp8_print_modes_and_motion_vectors(MODE_INFO *mi, int rows, int cols, int f
                 bindex = (b_row & 3) * 4 + (b_col & 3);
 
                 if (mi[mb_index].mbmi.mode == B_PRED)
-                    fprintf(mvs, "%2d ", mi[mb_index].bmi[bindex].as_mode);
+                    fprintf(mvs, "%2d ", mi[mb_index].bmi[bindex].mode);
                 else
                     fprintf(mvs, "xx ");
 

vp8/common/defaultcoefcounts.c

@@ -1,225 +0,0 @@
-/*
- *  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 "defaultcoefcounts.h"
-
-/* Generated file, included by entropy.c */
-
-const unsigned int vp8_default_coef_counts[BLOCK_TYPES]
-                                          [COEF_BANDS]
-                                          [PREV_COEF_CONTEXTS]
-                                          [MAX_ENTROPY_TOKENS] =
-{
-
-    {
-        /* Block Type ( 0 ) */
-        {
-            /* Coeff Band ( 0 ) */
-            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-        },
-        {
-            /* Coeff Band ( 1 ) */
-            {30190, 26544, 225,  24,   4,   0,   0,   0,   0,   0,   0, 4171593,},
-            {26846, 25157, 1241, 130,  26,   6,   1,   0,   0,   0,   0, 149987,},
-            {10484, 9538, 1006, 160,  36,  18,   0,   0,   0,   0,   0, 15104,},
-        },
-        {
-            /* Coeff Band ( 2 ) */
-            {25842, 40456, 1126,  83,  11,   2,   0,   0,   0,   0,   0,   0,},
-            {9338, 8010, 512,  73,   7,   3,   2,   0,   0,   0,   0, 43294,},
-            {1047, 751, 149,  31,  13,   6,   1,   0,   0,   0,   0, 879,},
-        },
-        {
-            /* Coeff Band ( 3 ) */
-            {26136, 9826, 252,  13,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {8134, 5574, 191,  14,   2,   0,   0,   0,   0,   0,   0, 35302,},
-            { 605, 677, 116,   9,   1,   0,   0,   0,   0,   0,   0, 611,},
-        },
-        {
-            /* Coeff Band ( 4 ) */
-            {10263, 15463, 283,  17,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {2773, 2191, 128,   9,   2,   2,   0,   0,   0,   0,   0, 10073,},
-            { 134, 125,  32,   4,   0,   2,   0,   0,   0,   0,   0,  50,},
-        },
-        {
-            /* Coeff Band ( 5 ) */
-            {10483, 2663,  23,   1,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {2137, 1251,  27,   1,   1,   0,   0,   0,   0,   0,   0, 14362,},
-            { 116, 156,  14,   2,   1,   0,   0,   0,   0,   0,   0, 190,},
-        },
-        {
-            /* Coeff Band ( 6 ) */
-            {40977, 27614, 412,  28,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {6113, 5213, 261,  22,   3,   0,   0,   0,   0,   0,   0, 26164,},
-            { 382, 312,  50,  14,   2,   0,   0,   0,   0,   0,   0, 345,},
-        },
-        {
-            /* Coeff Band ( 7 ) */
-            {   0,  26,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {   0,  13,   0,   0,   0,   0,   0,   0,   0,   0,   0, 319,},
-            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   8,},
-        },
-    },
-    {
-        /* Block Type ( 1 ) */
-        {
-            /* Coeff Band ( 0 ) */
-            {3268, 19382, 1043, 250,  93,  82,  49,  26,  17,   8,  25, 82289,},
-            {8758, 32110, 5436, 1832, 827, 668, 420, 153,  24,   0,   3, 52914,},
-            {9337, 23725, 8487, 3954, 2107, 1836, 1069, 399,  59,   0,   0, 18620,},
-        },
-        {
-            /* Coeff Band ( 1 ) */
-            {12419, 8420, 452,  62,   9,   1,   0,   0,   0,   0,   0,   0,},
-            {11715, 8705, 693,  92,  15,   7,   2,   0,   0,   0,   0, 53988,},
-            {7603, 8585, 2306, 778, 270, 145,  39,   5,   0,   0,   0, 9136,},
-        },
-        {
-            /* Coeff Band ( 2 ) */
-            {15938, 14335, 1207, 184,  55,  13,   4,   1,   0,   0,   0,   0,},
-            {7415, 6829, 1138, 244,  71,  26,   7,   0,   0,   0,   0, 9980,},
-            {1580, 1824, 655, 241,  89,  46,  10,   2,   0,   0,   0, 429,},
-        },
-        {
-            /* Coeff Band ( 3 ) */
-            {19453, 5260, 201,  19,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {9173, 3758, 213,  22,   1,   1,   0,   0,   0,   0,   0, 9820,},
-            {1689, 1277, 276,  51,  17,   4,   0,   0,   0,   0,   0, 679,},
-        },
-        {
-            /* Coeff Band ( 4 ) */
-            {12076, 10667, 620,  85,  19,   9,   5,   0,   0,   0,   0,   0,},
-            {4665, 3625, 423,  55,  19,   9,   0,   0,   0,   0,   0, 5127,},
-            { 415, 440, 143,  34,  20,   7,   2,   0,   0,   0,   0, 101,},
-        },
-        {
-            /* Coeff Band ( 5 ) */
-            {12183, 4846, 115,  11,   1,   0,   0,   0,   0,   0,   0,   0,},
-            {4226, 3149, 177,  21,   2,   0,   0,   0,   0,   0,   0, 7157,},
-            { 375, 621, 189,  51,  11,   4,   1,   0,   0,   0,   0, 198,},
-        },
-        {
-            /* Coeff Band ( 6 ) */
-            {61658, 37743, 1203,  94,  10,   3,   0,   0,   0,   0,   0,   0,},
-            {15514, 11563, 903, 111,  14,   5,   0,   0,   0,   0,   0, 25195,},
-            { 929, 1077, 291,  78,  14,   7,   1,   0,   0,   0,   0, 507,},
-        },
-        {
-            /* Coeff Band ( 7 ) */
-            {   0, 990,  15,   3,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {   0, 412,  13,   0,   0,   0,   0,   0,   0,   0,   0, 1641,},
-            {   0,  18,   7,   1,   0,   0,   0,   0,   0,   0,   0,  30,},
-        },
-    },
-    {
-        /* Block Type ( 2 ) */
-        {
-            /* Coeff Band ( 0 ) */
-            { 953, 24519, 628, 120,  28,  12,   4,   0,   0,   0,   0, 2248798,},
-            {1525, 25654, 2647, 617, 239, 143,  42,   5,   0,   0,   0, 66837,},
-            {1180, 11011, 3001, 1237, 532, 448, 239,  54,   5,   0,   0, 7122,},
-        },
-        {
-            /* Coeff Band ( 1 ) */
-            {1356, 2220,  67,  10,   4,   1,   0,   0,   0,   0,   0,   0,},
-            {1450, 2544, 102,  18,   4,   3,   0,   0,   0,   0,   0, 57063,},
-            {1182, 2110, 470, 130,  41,  21,   0,   0,   0,   0,   0, 6047,},
-        },
-        {
-            /* Coeff Band ( 2 ) */
-            { 370, 3378, 200,  30,   5,   4,   1,   0,   0,   0,   0,   0,},
-            { 293, 1006, 131,  29,  11,   0,   0,   0,   0,   0,   0, 5404,},
-            { 114, 387,  98,  23,   4,   8,   1,   0,   0,   0,   0, 236,},
-        },
-        {
-            /* Coeff Band ( 3 ) */
-            { 579, 194,   4,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            { 395, 213,   5,   1,   0,   0,   0,   0,   0,   0,   0, 4157,},
-            { 119, 122,   4,   0,   0,   0,   0,   0,   0,   0,   0, 300,},
-        },
-        {
-            /* Coeff Band ( 4 ) */
-            {  38, 557,  19,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {  21, 114,  12,   1,   0,   0,   0,   0,   0,   0,   0, 427,},
-            {   0,   5,   0,   0,   0,   0,   0,   0,   0,   0,   0,   7,},
-        },
-        {
-            /* Coeff Band ( 5 ) */
-            {  52,   7,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {  18,   6,   0,   0,   0,   0,   0,   0,   0,   0,   0, 652,},
-            {   1,   1,   0,   0,   0,   0,   0,   0,   0,   0,   0,  30,},
-        },
-        {
-            /* Coeff Band ( 6 ) */
-            { 640, 569,  10,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {  25,  77,   2,   0,   0,   0,   0,   0,   0,   0,   0, 517,},
-            {   4,   7,   0,   0,   0,   0,   0,   0,   0,   0,   0,   3,},
-        },
-        {
-            /* Coeff Band ( 7 ) */
-            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-        },
-    },
-    {
-        /* Block Type ( 3 ) */
-        {
-            /* Coeff Band ( 0 ) */
-            {2506, 20161, 2707, 767, 261, 178, 107,  30,  14,   3,   0, 100694,},
-            {8806, 36478, 8817, 3268, 1280, 850, 401, 114,  42,   0,   0, 58572,},
-            {11003, 27214, 11798, 5716, 2482, 2072, 1048, 175,  32,   0,   0, 19284,},
-        },
-        {
-            /* Coeff Band ( 1 ) */
-            {9738, 11313, 959, 205,  70,  18,  11,   1,   0,   0,   0,   0,},
-            {12628, 15085, 1507, 273,  52,  19,   9,   0,   0,   0,   0, 54280,},
-            {10701, 15846, 5561, 1926, 813, 570, 249,  36,   0,   0,   0, 6460,},
-        },
-        {
-            /* Coeff Band ( 2 ) */
-            {6781, 22539, 2784, 634, 182, 123,  20,   4,   0,   0,   0,   0,},
-            {6263, 11544, 2649, 790, 259, 168,  27,   5,   0,   0,   0, 20539,},
-            {3109, 4075, 2031, 896, 457, 386, 158,  29,   0,   0,   0, 1138,},
-        },
-        {
-            /* Coeff Band ( 3 ) */
-            {11515, 4079, 465,  73,   5,  14,   2,   0,   0,   0,   0,   0,},
-            {9361, 5834, 650,  96,  24,   8,   4,   0,   0,   0,   0, 22181,},
-            {4343, 3974, 1360, 415, 132,  96,  14,   1,   0,   0,   0, 1267,},
-        },
-        {
-            /* Coeff Band ( 4 ) */
-            {4787, 9297, 823, 168,  44,  12,   4,   0,   0,   0,   0,   0,},
-            {3619, 4472, 719, 198,  60,  31,   3,   0,   0,   0,   0, 8401,},
-            {1157, 1175, 483, 182,  88,  31,   8,   0,   0,   0,   0, 268,},
-        },
-        {
-            /* Coeff Band ( 5 ) */
-            {8299, 1226,  32,   5,   1,   0,   0,   0,   0,   0,   0,   0,},
-            {3502, 1568,  57,   4,   1,   1,   0,   0,   0,   0,   0, 9811,},
-            {1055, 1070, 166,  29,   6,   1,   0,   0,   0,   0,   0, 527,},
-        },
-        {
-            /* Coeff Band ( 6 ) */
-            {27414, 27927, 1989, 347,  69,  26,   0,   0,   0,   0,   0,   0,},
-            {5876, 10074, 1574, 341,  91,  24,   4,   0,   0,   0,   0, 21954,},
-            {1571, 2171, 778, 324, 124,  65,  16,   0,   0,   0,   0, 979,},
-        },
-        {
-            /* Coeff Band ( 7 ) */
-            {   0,  29,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
-            {   0,  23,   0,   0,   0,   0,   0,   0,   0,   0,   0, 459,},
-            {   0,   1,   0,   0,   0,   0,   0,   0,   0,   0,   0,  13,},
-        },
-    },
-};

vp8/common/defaultcoefcounts.h

@@ -8,14 +8,214 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#ifndef __DEFAULTCOEFCOUNTS_H
-#define __DEFAULTCOEFCOUNTS_H
 
-#include "entropy.h"
+/* Generated file, included by entropy.c */
 
-extern const unsigned int vp8_default_coef_counts[BLOCK_TYPES]
-                                                 [COEF_BANDS]
-                                                 [PREV_COEF_CONTEXTS]
-                                                 [MAX_ENTROPY_TOKENS];
+static const unsigned int default_coef_counts [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens] =
+{
 
-#endif //__DEFAULTCOEFCOUNTS_H
+    {
+        /* Block Type ( 0 ) */
+        {
+            /* Coeff Band ( 0 ) */
+            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+        },
+        {
+            /* Coeff Band ( 1 ) */
+            {30190, 26544, 225,  24,   4,   0,   0,   0,   0,   0,   0, 4171593,},
+            {26846, 25157, 1241, 130,  26,   6,   1,   0,   0,   0,   0, 149987,},
+            {10484, 9538, 1006, 160,  36,  18,   0,   0,   0,   0,   0, 15104,},
+        },
+        {
+            /* Coeff Band ( 2 ) */
+            {25842, 40456, 1126,  83,  11,   2,   0,   0,   0,   0,   0,   0,},
+            {9338, 8010, 512,  73,   7,   3,   2,   0,   0,   0,   0, 43294,},
+            {1047, 751, 149,  31,  13,   6,   1,   0,   0,   0,   0, 879,},
+        },
+        {
+            /* Coeff Band ( 3 ) */
+            {26136, 9826, 252,  13,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {8134, 5574, 191,  14,   2,   0,   0,   0,   0,   0,   0, 35302,},
+            { 605, 677, 116,   9,   1,   0,   0,   0,   0,   0,   0, 611,},
+        },
+        {
+            /* Coeff Band ( 4 ) */
+            {10263, 15463, 283,  17,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {2773, 2191, 128,   9,   2,   2,   0,   0,   0,   0,   0, 10073,},
+            { 134, 125,  32,   4,   0,   2,   0,   0,   0,   0,   0,  50,},
+        },
+        {
+            /* Coeff Band ( 5 ) */
+            {10483, 2663,  23,   1,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {2137, 1251,  27,   1,   1,   0,   0,   0,   0,   0,   0, 14362,},
+            { 116, 156,  14,   2,   1,   0,   0,   0,   0,   0,   0, 190,},
+        },
+        {
+            /* Coeff Band ( 6 ) */
+            {40977, 27614, 412,  28,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {6113, 5213, 261,  22,   3,   0,   0,   0,   0,   0,   0, 26164,},
+            { 382, 312,  50,  14,   2,   0,   0,   0,   0,   0,   0, 345,},
+        },
+        {
+            /* Coeff Band ( 7 ) */
+            {   0,  26,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {   0,  13,   0,   0,   0,   0,   0,   0,   0,   0,   0, 319,},
+            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   8,},
+        },
+    },
+    {
+        /* Block Type ( 1 ) */
+        {
+            /* Coeff Band ( 0 ) */
+            {3268, 19382, 1043, 250,  93,  82,  49,  26,  17,   8,  25, 82289,},
+            {8758, 32110, 5436, 1832, 827, 668, 420, 153,  24,   0,   3, 52914,},
+            {9337, 23725, 8487, 3954, 2107, 1836, 1069, 399,  59,   0,   0, 18620,},
+        },
+        {
+            /* Coeff Band ( 1 ) */
+            {12419, 8420, 452,  62,   9,   1,   0,   0,   0,   0,   0,   0,},
+            {11715, 8705, 693,  92,  15,   7,   2,   0,   0,   0,   0, 53988,},
+            {7603, 8585, 2306, 778, 270, 145,  39,   5,   0,   0,   0, 9136,},
+        },
+        {
+            /* Coeff Band ( 2 ) */
+            {15938, 14335, 1207, 184,  55,  13,   4,   1,   0,   0,   0,   0,},
+            {7415, 6829, 1138, 244,  71,  26,   7,   0,   0,   0,   0, 9980,},
+            {1580, 1824, 655, 241,  89,  46,  10,   2,   0,   0,   0, 429,},
+        },
+        {
+            /* Coeff Band ( 3 ) */
+            {19453, 5260, 201,  19,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {9173, 3758, 213,  22,   1,   1,   0,   0,   0,   0,   0, 9820,},
+            {1689, 1277, 276,  51,  17,   4,   0,   0,   0,   0,   0, 679,},
+        },
+        {
+            /* Coeff Band ( 4 ) */
+            {12076, 10667, 620,  85,  19,   9,   5,   0,   0,   0,   0,   0,},
+            {4665, 3625, 423,  55,  19,   9,   0,   0,   0,   0,   0, 5127,},
+            { 415, 440, 143,  34,  20,   7,   2,   0,   0,   0,   0, 101,},
+        },
+        {
+            /* Coeff Band ( 5 ) */
+            {12183, 4846, 115,  11,   1,   0,   0,   0,   0,   0,   0,   0,},
+            {4226, 3149, 177,  21,   2,   0,   0,   0,   0,   0,   0, 7157,},
+            { 375, 621, 189,  51,  11,   4,   1,   0,   0,   0,   0, 198,},
+        },
+        {
+            /* Coeff Band ( 6 ) */
+            {61658, 37743, 1203,  94,  10,   3,   0,   0,   0,   0,   0,   0,},
+            {15514, 11563, 903, 111,  14,   5,   0,   0,   0,   0,   0, 25195,},
+            { 929, 1077, 291,  78,  14,   7,   1,   0,   0,   0,   0, 507,},
+        },
+        {
+            /* Coeff Band ( 7 ) */
+            {   0, 990,  15,   3,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {   0, 412,  13,   0,   0,   0,   0,   0,   0,   0,   0, 1641,},
+            {   0,  18,   7,   1,   0,   0,   0,   0,   0,   0,   0,  30,},
+        },
+    },
+    {
+        /* Block Type ( 2 ) */
+        {
+            /* Coeff Band ( 0 ) */
+            { 953, 24519, 628, 120,  28,  12,   4,   0,   0,   0,   0, 2248798,},
+            {1525, 25654, 2647, 617, 239, 143,  42,   5,   0,   0,   0, 66837,},
+            {1180, 11011, 3001, 1237, 532, 448, 239,  54,   5,   0,   0, 7122,},
+        },
+        {
+            /* Coeff Band ( 1 ) */
+            {1356, 2220,  67,  10,   4,   1,   0,   0,   0,   0,   0,   0,},
+            {1450, 2544, 102,  18,   4,   3,   0,   0,   0,   0,   0, 57063,},
+            {1182, 2110, 470, 130,  41,  21,   0,   0,   0,   0,   0, 6047,},
+        },
+        {
+            /* Coeff Band ( 2 ) */
+            { 370, 3378, 200,  30,   5,   4,   1,   0,   0,   0,   0,   0,},
+            { 293, 1006, 131,  29,  11,   0,   0,   0,   0,   0,   0, 5404,},
+            { 114, 387,  98,  23,   4,   8,   1,   0,   0,   0,   0, 236,},
+        },
+        {
+            /* Coeff Band ( 3 ) */
+            { 579, 194,   4,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            { 395, 213,   5,   1,   0,   0,   0,   0,   0,   0,   0, 4157,},
+            { 119, 122,   4,   0,   0,   0,   0,   0,   0,   0,   0, 300,},
+        },
+        {
+            /* Coeff Band ( 4 ) */
+            {  38, 557,  19,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {  21, 114,  12,   1,   0,   0,   0,   0,   0,   0,   0, 427,},
+            {   0,   5,   0,   0,   0,   0,   0,   0,   0,   0,   0,   7,},
+        },
+        {
+            /* Coeff Band ( 5 ) */
+            {  52,   7,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {  18,   6,   0,   0,   0,   0,   0,   0,   0,   0,   0, 652,},
+            {   1,   1,   0,   0,   0,   0,   0,   0,   0,   0,   0,  30,},
+        },
+        {
+            /* Coeff Band ( 6 ) */
+            { 640, 569,  10,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {  25,  77,   2,   0,   0,   0,   0,   0,   0,   0,   0, 517,},
+            {   4,   7,   0,   0,   0,   0,   0,   0,   0,   0,   0,   3,},
+        },
+        {
+            /* Coeff Band ( 7 ) */
+            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+        },
+    },
+    {
+        /* Block Type ( 3 ) */
+        {
+            /* Coeff Band ( 0 ) */
+            {2506, 20161, 2707, 767, 261, 178, 107,  30,  14,   3,   0, 100694,},
+            {8806, 36478, 8817, 3268, 1280, 850, 401, 114,  42,   0,   0, 58572,},
+            {11003, 27214, 11798, 5716, 2482, 2072, 1048, 175,  32,   0,   0, 19284,},
+        },
+        {
+            /* Coeff Band ( 1 ) */
+            {9738, 11313, 959, 205,  70,  18,  11,   1,   0,   0,   0,   0,},
+            {12628, 15085, 1507, 273,  52,  19,   9,   0,   0,   0,   0, 54280,},
+            {10701, 15846, 5561, 1926, 813, 570, 249,  36,   0,   0,   0, 6460,},
+        },
+        {
+            /* Coeff Band ( 2 ) */
+            {6781, 22539, 2784, 634, 182, 123,  20,   4,   0,   0,   0,   0,},
+            {6263, 11544, 2649, 790, 259, 168,  27,   5,   0,   0,   0, 20539,},
+            {3109, 4075, 2031, 896, 457, 386, 158,  29,   0,   0,   0, 1138,},
+        },
+        {
+            /* Coeff Band ( 3 ) */
+            {11515, 4079, 465,  73,   5,  14,   2,   0,   0,   0,   0,   0,},
+            {9361, 5834, 650,  96,  24,   8,   4,   0,   0,   0,   0, 22181,},
+            {4343, 3974, 1360, 415, 132,  96,  14,   1,   0,   0,   0, 1267,},
+        },
+        {
+            /* Coeff Band ( 4 ) */
+            {4787, 9297, 823, 168,  44,  12,   4,   0,   0,   0,   0,   0,},
+            {3619, 4472, 719, 198,  60,  31,   3,   0,   0,   0,   0, 8401,},
+            {1157, 1175, 483, 182,  88,  31,   8,   0,   0,   0,   0, 268,},
+        },
+        {
+            /* Coeff Band ( 5 ) */
+            {8299, 1226,  32,   5,   1,   0,   0,   0,   0,   0,   0,   0,},
+            {3502, 1568,  57,   4,   1,   1,   0,   0,   0,   0,   0, 9811,},
+            {1055, 1070, 166,  29,   6,   1,   0,   0,   0,   0,   0, 527,},
+        },
+        {
+            /* Coeff Band ( 6 ) */
+            {27414, 27927, 1989, 347,  69,  26,   0,   0,   0,   0,   0,   0,},
+            {5876, 10074, 1574, 341,  91,  24,   4,   0,   0,   0,   0, 21954,},
+            {1571, 2171, 778, 324, 124,  65,  16,   0,   0,   0,   0, 979,},
+        },
+        {
+            /* Coeff Band ( 7 ) */
+            {   0,  29,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,},
+            {   0,  23,   0,   0,   0,   0,   0,   0,   0,   0,   0, 459,},
+            {   0,   1,   0,   0,   0,   0,   0,   0,   0,   0,   0,  13,},
+        },
+    },
+};

vp8/common/entropy.c

@@ -26,32 +26,8 @@ typedef vp8_prob Prob;
 
 #include "coefupdateprobs.h"
 
-DECLARE_ALIGNED(16, const unsigned char, vp8_norm[256]) =
-{
-    0, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4,
-    3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
-    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
-    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
-    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
-};
-
-DECLARE_ALIGNED(16, cuchar, vp8_coef_bands[16]) =
-{ 0, 1, 2, 3, 6, 4, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7};
-
-DECLARE_ALIGNED(16, cuchar, vp8_prev_token_class[MAX_ENTROPY_TOKENS]) =
-{ 0, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0};
-
+DECLARE_ALIGNED(16, cuchar, vp8_coef_bands[16]) = { 0, 1, 2, 3, 6, 4, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7};
+DECLARE_ALIGNED(16, cuchar, vp8_prev_token_class[MAX_ENTROPY_TOKENS]) = { 0, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0};
 DECLARE_ALIGNED(16, const int, vp8_default_zig_zag1d[16]) =
 {
     0,  1,  4,  8,
@@ -89,7 +65,7 @@ const vp8_tree_index vp8_coef_tree[ 22] =     /* corresponding _CONTEXT_NODEs */
     -DCT_VAL_CATEGORY5, -DCT_VAL_CATEGORY6   /* 10 = CAT_FIVE */
 };
 
-struct vp8_token_struct vp8_coef_encodings[MAX_ENTROPY_TOKENS];
+struct vp8_token_struct vp8_coef_encodings[vp8_coef_tokens];
 
 /* Trees for extra bits.  Probabilities are constant and
    do not depend on previously encoded bits */
@@ -169,12 +145,10 @@ void vp8_default_coef_probs(VP8_COMMON *pc)
 
             do
             {
-                unsigned int branch_ct [ENTROPY_NODES] [2];
+                unsigned int branch_ct [vp8_coef_tokens-1] [2];
                 vp8_tree_probs_from_distribution(
-                    MAX_ENTROPY_TOKENS, vp8_coef_encodings, vp8_coef_tree,
-                    pc->fc.coef_probs[h][i][k],
-                    branch_ct,
-                    vp8_default_coef_counts[h][i][k],
+                    vp8_coef_tokens, vp8_coef_encodings, vp8_coef_tree,
+                    pc->fc.coef_probs [h][i][k], branch_ct, default_coef_counts [h][i][k],
                     256, 1);
 
             }

vp8/common/entropy.h

@@ -30,12 +30,13 @@
 #define DCT_VAL_CATEGORY6       10      /* 67+       Extra Bits 11+1 */
 #define DCT_EOB_TOKEN           11      /* EOB       Extra Bits 0+0 */
 
-#define MAX_ENTROPY_TOKENS 12
+#define vp8_coef_tokens 12
+#define MAX_ENTROPY_TOKENS vp8_coef_tokens
 #define ENTROPY_NODES 11
 
 extern const vp8_tree_index vp8_coef_tree[];
 
-extern struct vp8_token_struct vp8_coef_encodings[MAX_ENTROPY_TOKENS];
+extern struct vp8_token_struct vp8_coef_encodings[vp8_coef_tokens];
 
 typedef struct
 {
@@ -84,9 +85,9 @@ extern DECLARE_ALIGNED(16, const unsigned char, vp8_coef_bands[16]);
 /*# define DC_TOKEN_CONTEXTS        3*/ /* 00, 0!0, !0!0 */
 #   define PREV_COEF_CONTEXTS       3
 
-extern DECLARE_ALIGNED(16, const unsigned char, vp8_prev_token_class[MAX_ENTROPY_TOKENS]);
+extern DECLARE_ALIGNED(16, const unsigned char, vp8_prev_token_class[vp8_coef_tokens]);
 
-extern const vp8_prob vp8_coef_update_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES];
+extern const vp8_prob vp8_coef_update_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens-1];
 
 
 struct VP8Common;

vp8/common/entropymode.c

@@ -33,11 +33,11 @@ typedef enum
     SUBMVREF_LEFT_ABOVE_ZED
 } sumvfref_t;
 
-int vp8_mv_cont(const int_mv *l, const int_mv *a)
+int vp8_mv_cont(const MV *l, const MV *a)
 {
-    int lez = (l->as_int == 0);
-    int aez = (a->as_int == 0);
-    int lea = (l->as_int == a->as_int);
+    int lez = (l->row == 0 && l->col == 0);
+    int aez = (a->row == 0 && a->col == 0);
+    int lea = (l->row == a->row && l->col == a->col);
 
     if (lea && lez)
         return SUBMVREF_LEFT_ABOVE_ZED;

vp8/common/entropymode.h

@@ -25,7 +25,7 @@ extern const int vp8_mbsplit_count [VP8_NUMMBSPLITS];    /* # of subsets */
 
 extern const vp8_prob vp8_mbsplit_probs [VP8_NUMMBSPLITS-1];
 
-extern int vp8_mv_cont(const int_mv *l, const int_mv *a);
+extern int vp8_mv_cont(const MV *l, const MV *a);
 #define SUBMVREF_COUNT 5
 extern const vp8_prob vp8_sub_mv_ref_prob2 [SUBMVREF_COUNT][VP8_SUBMVREFS-1];
 

vp8/common/extend.c

@@ -85,10 +85,10 @@ void vp8_copy_and_extend_frame(YV12_BUFFER_CONFIG *src,
                           src->y_height, src->y_width,
                           et, el, eb, er);
 
-    et = dst->border >> 1;
-    el = dst->border >> 1;
-    eb = (dst->border >> 1) + dst->uv_height - src->uv_height;
-    er = (dst->border >> 1) + dst->uv_width - src->uv_width;
+    et = (et + 1) >> 1;
+    el = (el + 1) >> 1;
+    eb = (eb + 1) >> 1;
+    er = (er + 1) >> 1;
 
     copy_and_extend_plane(src->u_buffer, src->uv_stride,
                           dst->u_buffer, dst->uv_stride,

vp8/common/filter.c

@@ -10,6 +10,29 @@
 
 
 #include <stdlib.h>
+#include <stdio.h>
+
+#define REGISTER_FILTER 1
+#define CLAMP(x,min,max) if (x < min) x = min; else if ( x > max ) x = max;
+
+#if REGISTER_FILTER
+#define FILTER0 filter0
+#define FILTER1 filter1
+#define FILTER2 filter2
+#define FILTER3 filter3
+#define FILTER4 filter4
+#define FILTER5 filter5
+#else
+#define FILTER0 vp8_filter[0]
+#define FILTER1 vp8_filter[1]
+#define FILTER2 vp8_filter[2]
+#define FILTER3 vp8_filter[3]
+#define FILTER4 vp8_filter[4]
+#define FILTER5 vp8_filter[5]
+#endif
+
+#define SRC_INCREMENT src_increment
+
 #include "filter.h"
 #include "vpx_ports/mem.h"
 
@@ -27,7 +50,6 @@ DECLARE_ALIGNED(16, const short, vp8_bilinear_filters[8][2]) =
 
 DECLARE_ALIGNED(16, const short, vp8_sub_pel_filters[8][6]) =
 {
-
     { 0,  0,  128,    0,   0,  0 },         /* note that 1/8 pel positions are just as per alpha -0.5 bicubic */
     { 0, -6,  123,   12,  -1,  0 },
     { 2, -11, 108,   36,  -8,  1 },         /* New 1/4 pel 6 tap filter */
@@ -49,35 +71,45 @@ static void filter_block2d_first_pass
     const short *vp8_filter
 )
 {
+
     unsigned int i, j;
     int Temp;
 
+#if REGISTER_FILTER
+    short filter0 = vp8_filter[0];
+    short filter1 = vp8_filter[1];
+    short filter2 = vp8_filter[2];
+    short filter3 = vp8_filter[3];
+    short filter4 = vp8_filter[4];
+    short filter5 = vp8_filter[5];
+#endif
+
+    int ps2 = 2*(int)pixel_step;
+    int ps3 = 3*(int)pixel_step;
+
+    unsigned int src_increment = src_pixels_per_line - output_width;
     for (i = 0; i < output_height; i++)
     {
         for (j = 0; j < output_width; j++)
         {
-            Temp = ((int)src_ptr[-2 * (int)pixel_step] * vp8_filter[0]) +
-                   ((int)src_ptr[-1 * (int)pixel_step] * vp8_filter[1]) +
-                   ((int)src_ptr[0]                 * vp8_filter[2]) +
-                   ((int)src_ptr[pixel_step]         * vp8_filter[3]) +
-                   ((int)src_ptr[2*pixel_step]       * vp8_filter[4]) +
-                   ((int)src_ptr[3*pixel_step]       * vp8_filter[5]) +
+            Temp = ((int)src_ptr[-1*ps2]         * FILTER0);
+            Temp += ((int)src_ptr[-1*(int)pixel_step] * FILTER1) +
+               ((int)src_ptr[0]                * FILTER2) +
+               ((int)src_ptr[pixel_step]       * FILTER3) +
+               ((int)src_ptr[ps2]              * FILTER4) +
+               ((int)src_ptr[ps3]              * FILTER5) +
                (VP8_FILTER_WEIGHT >> 1);      /* Rounding */
 
             /* Normalize back to 0-255 */
             Temp = Temp >> VP8_FILTER_SHIFT;
-
-            if (Temp < 0)
-                Temp = 0;
-            else if (Temp > 255)
-                Temp = 255;
+            CLAMP(Temp, 0, 255);
 
             output_ptr[j] = Temp;
             src_ptr++;
         }
 
         /* Next row... */
-        src_ptr    += src_pixels_per_line - output_width;
+        src_ptr    += SRC_INCREMENT;
         output_ptr += output_width;
     }
 }
@@ -97,33 +129,42 @@ static void filter_block2d_second_pass
 	unsigned int i, j;
 	int  Temp;
 
+#if REGISTER_FILTER
+    short filter0 = vp8_filter[0];
+    short filter1 = vp8_filter[1];
+    short filter2 = vp8_filter[2];
+    short filter3 = vp8_filter[3];
+    short filter4 = vp8_filter[4];
+    short filter5 = vp8_filter[5];
+#endif
+
+    int ps2 = ((int)pixel_step) << 1;
+    int ps3 = ps2 + (int)pixel_step;
+    unsigned int src_increment = src_pixels_per_line - output_width;
+
     for (i = 0; i < output_height; i++)
     {
         for (j = 0; j < output_width; j++)
         {
             /* Apply filter */
-            Temp = ((int)src_ptr[-2 * (int)pixel_step] * vp8_filter[0]) +
-                   ((int)src_ptr[-1 * (int)pixel_step] * vp8_filter[1]) +
-                   ((int)src_ptr[0]                 * vp8_filter[2]) +
-                   ((int)src_ptr[pixel_step]         * vp8_filter[3]) +
-                   ((int)src_ptr[2*pixel_step]       * vp8_filter[4]) +
-                   ((int)src_ptr[3*pixel_step]       * vp8_filter[5]) +
+            Temp = ((int)src_ptr[-1*ps2] * FILTER0) +
+                   ((int)src_ptr[-1*(int)pixel_step] * FILTER1) +
+                   ((int)src_ptr[0]                  * FILTER2) +
+                   ((int)src_ptr[pixel_step]         * FILTER3) +
+                   ((int)src_ptr[ps2]       * FILTER4) +
+                   ((int)src_ptr[ps3]       * FILTER5) +
                    (VP8_FILTER_WEIGHT >> 1);   /* Rounding */
 
             /* Normalize back to 0-255 */
             Temp = Temp >> VP8_FILTER_SHIFT;
-
-            if (Temp < 0)
-                Temp = 0;
-            else if (Temp > 255)
-                Temp = 255;
+            CLAMP(Temp, 0, 255);
 
             output_ptr[j] = (unsigned char)Temp;
             src_ptr++;
         }
 
         /* Start next row */
-        src_ptr    += src_pixels_per_line - output_width;
+        src_ptr    += src_increment;
         output_ptr += output_pitch;
     }
 }
@@ -167,6 +208,7 @@ void vp8_sixtap_predict_c
 
     filter_block2d(src_ptr, dst_ptr, src_pixels_per_line, dst_pitch, HFilter, VFilter);
 }
+
 void vp8_sixtap_predict8x8_c
 (
     unsigned char  *src_ptr,

vp8/common/findnearmv.c

@@ -25,9 +25,9 @@ void vp8_find_near_mvs
 (
     MACROBLOCKD *xd,
     const MODE_INFO *here,
-    int_mv *nearest,
-    int_mv *nearby,
-    int_mv *best_mv,
+    MV *nearest,
+    MV *nearby,
+    MV *best_mv,
     int cnt[4],
     int refframe,
     int *ref_frame_sign_bias
@@ -131,14 +131,13 @@ void vp8_find_near_mvs
         near_mvs[CNT_INTRA] = near_mvs[CNT_NEAREST];
 
     /* Set up return values */
-    best_mv->as_int = near_mvs[0].as_int;
-    nearest->as_int = near_mvs[CNT_NEAREST].as_int;
-    nearby->as_int = near_mvs[CNT_NEAR].as_int;
+    *best_mv = near_mvs[0].as_mv;
+    *nearest = near_mvs[CNT_NEAREST].as_mv;
+    *nearby = near_mvs[CNT_NEAR].as_mv;
 
-    //TODO: move clamp outside findnearmv
-    vp8_clamp_mv2(nearest, xd);
-    vp8_clamp_mv2(nearby, xd);
-    vp8_clamp_mv2(best_mv, xd);
+    vp8_clamp_mv(nearest, xd);
+    vp8_clamp_mv(nearby, xd);
+    vp8_clamp_mv(best_mv, xd); /*TODO: move this up before the copy*/
 }
 
 vp8_prob *vp8_mv_ref_probs(
@@ -153,3 +152,26 @@ vp8_prob *vp8_mv_ref_probs(
     return p;
 }
 
+const B_MODE_INFO *vp8_left_bmi(const MODE_INFO *cur_mb, int b)
+{
+    if (!(b & 3))
+    {
+        /* On L edge, get from MB to left of us */
+        --cur_mb;
+        b += 4;
+    }
+
+    return cur_mb->bmi + b - 1;
+}
+
+const B_MODE_INFO *vp8_above_bmi(const MODE_INFO *cur_mb, int b, int mi_stride)
+{
+    if (!(b >> 2))
+    {
+        /* On top edge, get from MB above us */
+        cur_mb -= mi_stride;
+        b += 16;
+    }
+
+    return cur_mb->bmi + b - 4;
+}

vp8/common/findnearmv.h

@@ -17,6 +17,11 @@
 #include "modecont.h"
 #include "treecoder.h"
 
+typedef union
+{
+    unsigned int as_int;
+    MV           as_mv;
+} int_mv;        /* facilitates rapid equality tests */
 
 static void mv_bias(int refmb_ref_frame_sign_bias, int refframe, int_mv *mvp, const int *ref_frame_sign_bias)
 {
@@ -34,48 +39,24 @@ static void mv_bias(int refmb_ref_frame_sign_bias, int refframe, int_mv *mvp, co
 
 #define LEFT_TOP_MARGIN (16 << 3)
 #define RIGHT_BOTTOM_MARGIN (16 << 3)
-static void vp8_clamp_mv2(int_mv *mv, const MACROBLOCKD *xd)
+static void vp8_clamp_mv(MV *mv, const MACROBLOCKD *xd)
 {
-    if (mv->as_mv.col < (xd->mb_to_left_edge - LEFT_TOP_MARGIN))
-        mv->as_mv.col = xd->mb_to_left_edge - LEFT_TOP_MARGIN;
-    else if (mv->as_mv.col > xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN)
-        mv->as_mv.col = xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN;
+    if (mv->col < (xd->mb_to_left_edge - LEFT_TOP_MARGIN))
+        mv->col = xd->mb_to_left_edge - LEFT_TOP_MARGIN;
+    else if (mv->col > xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN)
+        mv->col = xd->mb_to_right_edge + RIGHT_BOTTOM_MARGIN;
 
-    if (mv->as_mv.row < (xd->mb_to_top_edge - LEFT_TOP_MARGIN))
-        mv->as_mv.row = xd->mb_to_top_edge - LEFT_TOP_MARGIN;
-    else if (mv->as_mv.row > xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN)
-        mv->as_mv.row = xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN;
-}
-
-static void vp8_clamp_mv(int_mv *mv, int mb_to_left_edge, int mb_to_right_edge,
-                         int mb_to_top_edge, int mb_to_bottom_edge)
-{
-    mv->as_mv.col = (mv->as_mv.col < mb_to_left_edge) ?
-        mb_to_left_edge : mv->as_mv.col;
-    mv->as_mv.col = (mv->as_mv.col > mb_to_right_edge) ?
-        mb_to_right_edge : mv->as_mv.col;
-    mv->as_mv.row = (mv->as_mv.row < mb_to_top_edge) ?
-        mb_to_top_edge : mv->as_mv.row;
-    mv->as_mv.row = (mv->as_mv.row > mb_to_bottom_edge) ?
-        mb_to_bottom_edge : mv->as_mv.row;
-}
-static unsigned int vp8_check_mv_bounds(int_mv *mv, int mb_to_left_edge,
-                                int mb_to_right_edge, int mb_to_top_edge,
-                                int mb_to_bottom_edge)
-{
-    unsigned int need_to_clamp;
-    need_to_clamp = (mv->as_mv.col < mb_to_left_edge) ? 1 : 0;
-    need_to_clamp |= (mv->as_mv.col > mb_to_right_edge) ? 1 : 0;
-    need_to_clamp |= (mv->as_mv.row < mb_to_top_edge) ? 1 : 0;
-    need_to_clamp |= (mv->as_mv.row > mb_to_bottom_edge) ? 1 : 0;
-    return need_to_clamp;
+    if (mv->row < (xd->mb_to_top_edge - LEFT_TOP_MARGIN))
+        mv->row = xd->mb_to_top_edge - LEFT_TOP_MARGIN;
+    else if (mv->row > xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN)
+        mv->row = xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN;
 }
 
 void vp8_find_near_mvs
 (
     MACROBLOCKD *xd,
     const MODE_INFO *here,
-    int_mv *nearest, int_mv *nearby, int_mv *best,
+    MV *nearest, MV *nearby, MV *best,
     int near_mv_ref_cts[4],
     int refframe,
     int *ref_frame_sign_bias
@@ -85,89 +66,10 @@ vp8_prob *vp8_mv_ref_probs(
     vp8_prob p[VP8_MVREFS-1], const int near_mv_ref_ct[4]
 );
 
+const B_MODE_INFO *vp8_left_bmi(const MODE_INFO *cur_mb, int b);
+
+const B_MODE_INFO *vp8_above_bmi(const MODE_INFO *cur_mb, int b, int mi_stride);
+
 extern const unsigned char vp8_mbsplit_offset[4][16];
 
-
-static int left_block_mv(const MODE_INFO *cur_mb, int b)
-{
-    if (!(b & 3))
-    {
-        /* On L edge, get from MB to left of us */
-        --cur_mb;
-
-        if(cur_mb->mbmi.mode != SPLITMV)
-            return cur_mb->mbmi.mv.as_int;
-        b += 4;
-    }
-
-    return (cur_mb->bmi + b - 1)->mv.as_int;
-}
-
-static int above_block_mv(const MODE_INFO *cur_mb, int b, int mi_stride)
-{
-    if (!(b >> 2))
-    {
-        /* On top edge, get from MB above us */
-        cur_mb -= mi_stride;
-
-        if(cur_mb->mbmi.mode != SPLITMV)
-            return cur_mb->mbmi.mv.as_int;
-        b += 16;
-    }
-
-    return (cur_mb->bmi + b - 4)->mv.as_int;
-}
-static B_PREDICTION_MODE left_block_mode(const MODE_INFO *cur_mb, int b)
-{
-    if (!(b & 3))
-    {
-        /* On L edge, get from MB to left of us */
-        --cur_mb;
-        switch (cur_mb->mbmi.mode)
-        {
-            case B_PRED:
-              return (cur_mb->bmi + b + 3)->as_mode;
-            case DC_PRED:
-                return B_DC_PRED;
-            case V_PRED:
-                return B_VE_PRED;
-            case H_PRED:
-                return B_HE_PRED;
-            case TM_PRED:
-                return B_TM_PRED;
-            default:
-                return B_DC_PRED;
-        }
-    }
-
-    return (cur_mb->bmi + b - 1)->as_mode;
-}
-
-static B_PREDICTION_MODE above_block_mode(const MODE_INFO *cur_mb, int b, int mi_stride)
-{
-    if (!(b >> 2))
-    {
-        /* On top edge, get from MB above us */
-        cur_mb -= mi_stride;
-
-        switch (cur_mb->mbmi.mode)
-        {
-            case B_PRED:
-              return (cur_mb->bmi + b + 12)->as_mode;
-            case DC_PRED:
-                return B_DC_PRED;
-            case V_PRED:
-                return B_VE_PRED;
-            case H_PRED:
-                return B_HE_PRED;
-            case TM_PRED:
-                return B_TM_PRED;
-            default:
-                return B_DC_PRED;
-        }
-    }
-
-    return (cur_mb->bmi + b - 4)->as_mode;
-}
-
 #endif

vp8/common/generic/systemdependent.c

@@ -17,53 +17,9 @@
 #include "vp8/common/idct.h"
 #include "vp8/common/onyxc_int.h"
 
-#if CONFIG_MULTITHREAD
-#if HAVE_UNISTD_H
-#include <unistd.h>
-#elif defined(_WIN32)
-#include <windows.h>
-typedef void (WINAPI *PGNSI)(LPSYSTEM_INFO);
-#endif
-#endif
-
 extern void vp8_arch_x86_common_init(VP8_COMMON *ctx);
 extern void vp8_arch_arm_common_init(VP8_COMMON *ctx);
-
-#if CONFIG_MULTITHREAD
-static int get_cpu_count()
-{
-    int core_count = 16;
-
-#if HAVE_UNISTD_H
-#if defined(_SC_NPROCESSORS_ONLN)
-    core_count = sysconf(_SC_NPROCESSORS_ONLN);
-#elif defined(_SC_NPROC_ONLN)
-    core_count = sysconf(_SC_NPROC_ONLN);
-#endif
-#elif defined(_WIN32)
-    {
-        PGNSI pGNSI;
-        SYSTEM_INFO sysinfo;
-
-        /* Call GetNativeSystemInfo if supported or
-         * GetSystemInfo otherwise. */
-
-        pGNSI = (PGNSI) GetProcAddress(
-                GetModuleHandle(TEXT("kernel32.dll")), "GetNativeSystemInfo");
-        if (pGNSI != NULL)
-            pGNSI(&sysinfo);
-        else
-            GetSystemInfo(&sysinfo);
-
-        core_count = sysinfo.dwNumberOfProcessors;
-    }
-#else
-    /* other platforms */
-#endif
-
-    return core_count > 0 ? core_count : 1;
-}
-#endif
+extern void vp8_arch_opencl_common_init(VP8_COMMON *ctx);
 
 void vp8_machine_specific_config(VP8_COMMON *ctx)
 {
@@ -88,12 +44,6 @@ void vp8_machine_specific_config(VP8_COMMON *ctx)
         vp8_build_intra_predictors_mby;
     rtcd->recon.build_intra_predictors_mby_s =
         vp8_build_intra_predictors_mby_s;
-    rtcd->recon.build_intra_predictors_mbuv =
-        vp8_build_intra_predictors_mbuv;
-    rtcd->recon.build_intra_predictors_mbuv_s =
-        vp8_build_intra_predictors_mbuv_s;
-    rtcd->recon.intra4x4_predict =
-        vp8_intra4x4_predict;
 
     rtcd->subpix.sixtap16x16   = vp8_sixtap_predict16x16_c;
     rtcd->subpix.sixtap8x8     = vp8_sixtap_predict8x8_c;
@@ -108,12 +58,12 @@ void vp8_machine_specific_config(VP8_COMMON *ctx)
     rtcd->loopfilter.normal_b_v  = vp8_loop_filter_bv_c;
     rtcd->loopfilter.normal_mb_h = vp8_loop_filter_mbh_c;
     rtcd->loopfilter.normal_b_h  = vp8_loop_filter_bh_c;
-    rtcd->loopfilter.simple_mb_v = vp8_loop_filter_simple_vertical_edge_c;
+    rtcd->loopfilter.simple_mb_v = vp8_loop_filter_mbvs_c;
     rtcd->loopfilter.simple_b_v  = vp8_loop_filter_bvs_c;
-    rtcd->loopfilter.simple_mb_h = vp8_loop_filter_simple_horizontal_edge_c;
+    rtcd->loopfilter.simple_mb_h = vp8_loop_filter_mbhs_c;
     rtcd->loopfilter.simple_b_h  = vp8_loop_filter_bhs_c;
 
-#if CONFIG_POSTPROC || (CONFIG_VP8_ENCODER && CONFIG_INTERNAL_STATS)
+#if CONFIG_POSTPROC || (CONFIG_VP8_ENCODER && CONFIG_PSNR)
     rtcd->postproc.down             = vp8_mbpost_proc_down_c;
     rtcd->postproc.across           = vp8_mbpost_proc_across_ip_c;
     rtcd->postproc.downacross       = vp8_post_proc_down_and_across_c;
@@ -133,7 +83,8 @@ void vp8_machine_specific_config(VP8_COMMON *ctx)
     vp8_arch_arm_common_init(ctx);
 #endif
 
-#if CONFIG_MULTITHREAD
-    ctx->processor_core_count = get_cpu_count();
-#endif /* CONFIG_MULTITHREAD */
+#if CONFIG_OPENCL && (ENABLE_CL_IDCT_DEQUANT || ENABLE_CL_SUBPIXEL || ENABLE_CL_LOOPFILTER)
+    vp8_arch_opencl_common_init(ctx);
+#endif
+
 }

vp8/common/idct.h

@@ -31,6 +31,10 @@
 #include "arm/idct_arm.h"
 #endif
 
+#if CONFIG_OPENCL
+#include "opencl/idct_cl.h"
+#endif
+
 #ifndef vp8_idct_idct1
 #define vp8_idct_idct1 vp8_short_idct4x4llm_1_c
 #endif

vp8/common/loopfilter.c

@@ -9,149 +9,164 @@
  */
 
 
-#include "vpx_config.h"
+#include "vpx_ports/config.h"
 #include "loopfilter.h"
 #include "onyxc_int.h"
-#include "vpx_mem/vpx_mem.h"
+
+#if CONFIG_OPENCL
+#include "opencl/loopfilter_cl.h"
+#endif
 
 typedef unsigned char uc;
 
+
 prototype_loopfilter(vp8_loop_filter_horizontal_edge_c);
 prototype_loopfilter(vp8_loop_filter_vertical_edge_c);
 prototype_loopfilter(vp8_mbloop_filter_horizontal_edge_c);
 prototype_loopfilter(vp8_mbloop_filter_vertical_edge_c);
-
-prototype_simple_loopfilter(vp8_loop_filter_simple_horizontal_edge_c);
-prototype_simple_loopfilter(vp8_loop_filter_simple_vertical_edge_c);
+prototype_loopfilter(vp8_loop_filter_simple_horizontal_edge_c);
+prototype_loopfilter(vp8_loop_filter_simple_vertical_edge_c);
 
 /* Horizontal MB filtering */
-void vp8_loop_filter_mbh_c(unsigned char *y_ptr, unsigned char *u_ptr,
-                           unsigned char *v_ptr, int y_stride, int uv_stride,
-                           loop_filter_info *lfi)
+void vp8_loop_filter_mbh_c(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                           int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_mbloop_filter_horizontal_edge_c(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2);
+    (void) simpler_lpf;
+    vp8_mbloop_filter_horizontal_edge_c(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_mbloop_filter_horizontal_edge_c(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+        vp8_mbloop_filter_horizontal_edge_c(u_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1);
 
     if (v_ptr)
-        vp8_mbloop_filter_horizontal_edge_c(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+        vp8_mbloop_filter_horizontal_edge_c(v_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1);
+}
+
+void vp8_loop_filter_mbhs_c(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                            int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_horizontal_edge_c(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 }
 
 /* Vertical MB Filtering */
-void vp8_loop_filter_mbv_c(unsigned char *y_ptr, unsigned char *u_ptr,
-                           unsigned char *v_ptr, int y_stride, int uv_stride,
-                           loop_filter_info *lfi)
+void vp8_loop_filter_mbv_c(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                           int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_mbloop_filter_vertical_edge_c(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2);
+    (void) simpler_lpf;
+    vp8_mbloop_filter_vertical_edge_c(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_mbloop_filter_vertical_edge_c(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+        vp8_mbloop_filter_vertical_edge_c(u_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1);
 
     if (v_ptr)
-        vp8_mbloop_filter_vertical_edge_c(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
+        vp8_mbloop_filter_vertical_edge_c(v_ptr, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1);
+}
+
+void vp8_loop_filter_mbvs_c(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                            int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_vertical_edge_c(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2);
 }
 
 /* Horizontal B Filtering */
-void vp8_loop_filter_bh_c(unsigned char *y_ptr, unsigned char *u_ptr,
-                          unsigned char *v_ptr, int y_stride, int uv_stride,
-                          loop_filter_info *lfi)
+void vp8_loop_filter_bh_c(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                          int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_loop_filter_horizontal_edge_c(y_ptr + 4 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
-    vp8_loop_filter_horizontal_edge_c(y_ptr + 8 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
-    vp8_loop_filter_horizontal_edge_c(y_ptr + 12 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
+    (void) simpler_lpf;
+    vp8_loop_filter_horizontal_edge_c(y_ptr + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_horizontal_edge_c(y_ptr + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_horizontal_edge_c(y_ptr + 12 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_loop_filter_horizontal_edge_c(u_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
+        vp8_loop_filter_horizontal_edge_c(u_ptr + 4 * uv_stride, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1);
 
     if (v_ptr)
-        vp8_loop_filter_horizontal_edge_c(v_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
+        vp8_loop_filter_horizontal_edge_c(v_ptr + 4 * uv_stride, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1);
 }
 
-void vp8_loop_filter_bhs_c(unsigned char *y_ptr, int y_stride,
-                           const unsigned char *blimit)
+void vp8_loop_filter_bhs_c(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                           int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 4 * y_stride, y_stride, blimit);
-    vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 8 * y_stride, y_stride, blimit);
-    vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 12 * y_stride, y_stride, blimit);
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_horizontal_edge_c(y_ptr + 12 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 }
 
 /* Vertical B Filtering */
-void vp8_loop_filter_bv_c(unsigned char *y_ptr, unsigned char *u_ptr,
-                          unsigned char *v_ptr, int y_stride, int uv_stride,
-                          loop_filter_info *lfi)
+void vp8_loop_filter_bv_c(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                          int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_loop_filter_vertical_edge_c(y_ptr + 4, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
-    vp8_loop_filter_vertical_edge_c(y_ptr + 8, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
-    vp8_loop_filter_vertical_edge_c(y_ptr + 12, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
+    (void) simpler_lpf;
+    vp8_loop_filter_vertical_edge_c(y_ptr + 4, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_vertical_edge_c(y_ptr + 8, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_vertical_edge_c(y_ptr + 12, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 
     if (u_ptr)
-        vp8_loop_filter_vertical_edge_c(u_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
+        vp8_loop_filter_vertical_edge_c(u_ptr + 4, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1);
 
     if (v_ptr)
-        vp8_loop_filter_vertical_edge_c(v_ptr + 4, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
+        vp8_loop_filter_vertical_edge_c(v_ptr + 4, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1);
 }
 
-void vp8_loop_filter_bvs_c(unsigned char *y_ptr, int y_stride,
-                           const unsigned char *blimit)
+void vp8_loop_filter_bvs_c(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
+                           int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
-    vp8_loop_filter_simple_vertical_edge_c(y_ptr + 4, y_stride, blimit);
-    vp8_loop_filter_simple_vertical_edge_c(y_ptr + 8, y_stride, blimit);
-    vp8_loop_filter_simple_vertical_edge_c(y_ptr + 12, y_stride, blimit);
+    (void) u_ptr;
+    (void) v_ptr;
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_vertical_edge_c(y_ptr + 4, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_vertical_edge_c(y_ptr + 8, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
+    vp8_loop_filter_simple_vertical_edge_c(y_ptr + 12, y_stride, lfi->flim, lfi->lim, lfi->thr, 2);
 }
 
-static void lf_init_lut(loop_filter_info_n *lfi)
+void vp8_init_loop_filter(VP8_COMMON *cm)
 {
-    int filt_lvl;
+    loop_filter_info *lfi = cm->lf_info;
+    LOOPFILTERTYPE lft = cm->filter_type;
+    int sharpness_lvl = cm->sharpness_level;
+    int frame_type = cm->frame_type;
+    int i, j;
 
-    for (filt_lvl = 0; filt_lvl <= MAX_LOOP_FILTER; filt_lvl++)
-    {
-        if (filt_lvl >= 40)
-        {
-            lfi->hev_thr_lut[KEY_FRAME][filt_lvl] = 2;
-            lfi->hev_thr_lut[INTER_FRAME][filt_lvl] = 3;
-        }
-        else if (filt_lvl >= 20)
-        {
-            lfi->hev_thr_lut[KEY_FRAME][filt_lvl] = 1;
-            lfi->hev_thr_lut[INTER_FRAME][filt_lvl] = 2;
-        }
-        else if (filt_lvl >= 15)
-        {
-            lfi->hev_thr_lut[KEY_FRAME][filt_lvl] = 1;
-            lfi->hev_thr_lut[INTER_FRAME][filt_lvl] = 1;
-        }
-        else
-        {
-            lfi->hev_thr_lut[KEY_FRAME][filt_lvl] = 0;
-            lfi->hev_thr_lut[INTER_FRAME][filt_lvl] = 0;
-        }
-    }
+    int block_inside_limit = 0;
+    int HEVThresh;
 
-    lfi->mode_lf_lut[DC_PRED] = 1;
-    lfi->mode_lf_lut[V_PRED] = 1;
-    lfi->mode_lf_lut[H_PRED] = 1;
-    lfi->mode_lf_lut[TM_PRED] = 1;
-    lfi->mode_lf_lut[B_PRED]  = 0;
-
-    lfi->mode_lf_lut[ZEROMV]  = 1;
-    lfi->mode_lf_lut[NEARESTMV] = 2;
-    lfi->mode_lf_lut[NEARMV] = 2;
-    lfi->mode_lf_lut[NEWMV] = 2;
-    lfi->mode_lf_lut[SPLITMV] = 3;
-
-}
-
-void vp8_loop_filter_update_sharpness(loop_filter_info_n *lfi,
-                                      int sharpness_lvl)
-{
-    int i;
-
-    /* For each possible value for the loop filter fill out limits */
+    /* For each possible value for the loop filter fill out a "loop_filter_info" entry. */
     for (i = 0; i <= MAX_LOOP_FILTER; i++)
     {
         int filt_lvl = i;
-        int block_inside_limit = 0;
+
+        if (frame_type == KEY_FRAME)
+        {
+            if (filt_lvl >= 40)
+                HEVThresh = 2;
+            else if (filt_lvl >= 15)
+                HEVThresh = 1;
+            else
+                HEVThresh = 0;
+        }
+        else
+        {
+            if (filt_lvl >= 40)
+                HEVThresh = 3;
+            else if (filt_lvl >= 20)
+                HEVThresh = 2;
+            else if (filt_lvl >= 15)
+                HEVThresh = 1;
+            else
+                HEVThresh = 0;
+        }
 
         /* Set loop filter paramaeters that control sharpness. */
         block_inside_limit = filt_lvl >> (sharpness_lvl > 0);
@@ -166,143 +181,177 @@ void vp8_loop_filter_update_sharpness(loop_filter_info_n *lfi,
         if (block_inside_limit < 1)
             block_inside_limit = 1;
 
-        vpx_memset(lfi->lim[i], block_inside_limit, SIMD_WIDTH);
-        vpx_memset(lfi->blim[i], (2 * filt_lvl + block_inside_limit),
-                SIMD_WIDTH);
-        vpx_memset(lfi->mblim[i], (2 * (filt_lvl + 2) + block_inside_limit),
-                SIMD_WIDTH);
+        for (j = 0; j < 16; j++)
+        {
+            lfi[i].lim[j] = block_inside_limit;
+            lfi[i].mbflim[j] = filt_lvl + 2;
+            lfi[i].flim[j] = filt_lvl;
+            lfi[i].thr[j] = HEVThresh;
+        }
+
+    }
+
+    /* Set up the function pointers depending on the type of loop filtering selected */
+    if (lft == NORMAL_LOOPFILTER)
+    {
+        cm->lf_mbv = LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_v);
+        cm->lf_bv  = LF_INVOKE(&cm->rtcd.loopfilter, normal_b_v);
+        cm->lf_mbh = LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_h);
+        cm->lf_bh  = LF_INVOKE(&cm->rtcd.loopfilter, normal_b_h);
+    }
+    else
+    {
+        cm->lf_mbv = LF_INVOKE(&cm->rtcd.loopfilter, simple_mb_v);
+        cm->lf_bv  = LF_INVOKE(&cm->rtcd.loopfilter, simple_b_v);
+        cm->lf_mbh = LF_INVOKE(&cm->rtcd.loopfilter, simple_mb_h);
+        cm->lf_bh  = LF_INVOKE(&cm->rtcd.loopfilter, simple_b_h);
     }
 }
 
-void vp8_loop_filter_init(VP8_COMMON *cm)
-{
-    loop_filter_info_n *lfi = &cm->lf_info;
-    int i;
-
-    /* init limits for given sharpness*/
-    vp8_loop_filter_update_sharpness(lfi, cm->sharpness_level);
-    cm->last_sharpness_level = cm->sharpness_level;
-
-    /* init LUT for lvl  and hev thr picking */
-    lf_init_lut(lfi);
-
-    /* init hev threshold const vectors */
-    for(i = 0; i < 4 ; i++)
-    {
-        vpx_memset(lfi->hev_thr[i], i, SIMD_WIDTH);
-    }
-}
-
-void vp8_loop_filter_frame_init(VP8_COMMON *cm,
-                                MACROBLOCKD *mbd,
-                                int default_filt_lvl)
-{
-    int seg,  /* segment number */
-        ref,  /* index in ref_lf_deltas */
-        mode; /* index in mode_lf_deltas */
-
-    loop_filter_info_n *lfi = &cm->lf_info;
-
-    /* update limits if sharpness has changed */
-    if(cm->last_sharpness_level != cm->sharpness_level)
-    {
-        vp8_loop_filter_update_sharpness(lfi, cm->sharpness_level);
-        cm->last_sharpness_level = cm->sharpness_level;
-    }
-
-    for(seg = 0; seg < MAX_MB_SEGMENTS; seg++)
-    {
-        int lvl_seg = default_filt_lvl;
-        int lvl_ref, lvl_mode;
-
-        /* Note the baseline filter values for each segment */
-        if (mbd->segmentation_enabled)
-        {
-            /* Abs value */
-            if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA)
-            {
-                lvl_seg = mbd->segment_feature_data[MB_LVL_ALT_LF][seg];
-            }
-            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;
-            }
-        }
-
-        if (!mbd->mode_ref_lf_delta_enabled)
-        {
-            /* we could get rid of this if we assume that deltas are set to
-             * zero when not in use; encoder always uses deltas
+/* Put vp8_init_loop_filter() in vp8dx_create_decompressor(). Only call vp8_frame_init_loop_filter() while decoding
+ * each frame. Check last_frame_type to skip the function most of times.
  */
-            vpx_memset(lfi->lvl[seg][0], lvl_seg, 4 * 4 );
-            continue;
+void vp8_frame_init_loop_filter(loop_filter_info *lfi, int frame_type)
+{
+    int HEVThresh;
+    int i, j;
+
+    /* For each possible value for the loop filter fill out a "loop_filter_info" entry. */
+    for (i = 0; i <= MAX_LOOP_FILTER; i++)
+    {
+        int filt_lvl = i;
+
+        if (frame_type == KEY_FRAME)
+        {
+            if (filt_lvl >= 40)
+                HEVThresh = 2;
+            else if (filt_lvl >= 15)
+                HEVThresh = 1;
+            else
+                HEVThresh = 0;
+        }
+        else
+        {
+            if (filt_lvl >= 40)
+                HEVThresh = 3;
+            else if (filt_lvl >= 20)
+                HEVThresh = 2;
+            else if (filt_lvl >= 15)
+                HEVThresh = 1;
+            else
+                HEVThresh = 0;
         }
 
-        lvl_ref = lvl_seg;
+        for (j = 0; j < 16; j++)
+        {
+            /*lfi[i].lim[j] = block_inside_limit;
+            lfi[i].mbflim[j] = filt_lvl+2;*/
+            /*lfi[i].flim[j] = filt_lvl;*/
+            lfi[i].thr[j] = HEVThresh;
+        }
+    }
+}
 
-        /* INTRA_FRAME */
-        ref = INTRA_FRAME;
 
+int vp8_adjust_mb_lf_value(MACROBLOCKD *mbd, int filter_level)
+{
+    MB_MODE_INFO *mbmi = &mbd->mode_info_context->mbmi;
+
+    if (mbd->mode_ref_lf_delta_enabled)
+    {
         /* Apply delta for reference frame */
-        lvl_ref += mbd->ref_lf_deltas[ref];
+        filter_level += mbd->ref_lf_deltas[mbmi->ref_frame];
 
-        /* Apply delta for Intra modes */
-        mode = 0; /* B_PRED */
+        /* Apply delta for mode */
+        if (mbmi->ref_frame == INTRA_FRAME)
+        {
             /* Only the split mode BPRED has a further special case */
-        lvl_mode = lvl_ref +  mbd->mode_lf_deltas[mode];
-        lvl_mode = (lvl_mode > 0) ? (lvl_mode > 63 ? 63 : lvl_mode) : 0; /* clamp */
-
-        lfi->lvl[seg][ref][mode] = lvl_mode;
-
-        mode = 1; /* all the rest of Intra modes */
-        lvl_mode = (lvl_ref > 0) ? (lvl_ref > 63 ? 63 : lvl_ref)  : 0; /* clamp */
-        lfi->lvl[seg][ref][mode] = lvl_mode;
-
-        /* LAST, GOLDEN, ALT */
-        for(ref = 1; ref < MAX_REF_FRAMES; ref++)
+            if (mbmi->mode == B_PRED)
+                filter_level +=  mbd->mode_lf_deltas[0];
+        }
+        else
         {
-            int lvl_ref = lvl_seg;
+            /* Zero motion mode */
+            if (mbmi->mode == ZEROMV)
+                filter_level +=  mbd->mode_lf_deltas[1];
 
-            /* Apply delta for reference frame */
-            lvl_ref += mbd->ref_lf_deltas[ref];
+            /* Split MB motion mode */
+            else if (mbmi->mode == SPLITMV)
+                filter_level +=  mbd->mode_lf_deltas[3];
 
-            /* Apply delta for Inter modes */
-            for (mode = 1; mode < 4; mode++)
-            {
-                lvl_mode = lvl_ref + mbd->mode_lf_deltas[mode];
-                lvl_mode = (lvl_mode > 0) ? (lvl_mode > 63 ? 63 : lvl_mode) : 0; /* clamp */
+            /* All other inter motion modes (Nearest, Near, New) */
+            else
+                filter_level +=  mbd->mode_lf_deltas[2];
+        }
 
-                lfi->lvl[seg][ref][mode] = lvl_mode;
-            }
-        }
+        /* Range check */
+        if (filter_level > MAX_LOOP_FILTER)
+            filter_level = MAX_LOOP_FILTER;
+        else if (filter_level < 0)
+            filter_level = 0;
     }
+    return filter_level;
 }
 
+
 void vp8_loop_filter_frame
 (
     VP8_COMMON *cm,
-    MACROBLOCKD *mbd
+    MACROBLOCKD *mbd,
+    int default_filt_lvl
 )
 {
     YV12_BUFFER_CONFIG *post = cm->frame_to_show;
-    loop_filter_info_n *lfi_n = &cm->lf_info;
-    loop_filter_info lfi;
-
+    loop_filter_info *lfi = cm->lf_info;
     FRAME_TYPE frame_type = cm->frame_type;
 
     int mb_row;
     int mb_col;
 
-    int filter_level;
 
+    int baseline_filter_level[MAX_MB_SEGMENTS];
+    int filter_level;
+    int alt_flt_enabled = mbd->segmentation_enabled;
+
+    int i;
     unsigned char *y_ptr, *u_ptr, *v_ptr;
 
-    /* Point at base of Mb MODE_INFO list */
-    const MODE_INFO *mode_info_context = cm->mi;
+#if CONFIG_OPENCL && ENABLE_CL_LOOPFILTER
+    if ( cl_initialized == CL_SUCCESS ){
+        vp8_loop_filter_frame_cl(cm,mbd,default_filt_lvl);
+        return;
+    }
+#endif
+
+    mbd->mode_info_context = cm->mi;          /* Point at base of Mb MODE_INFO list */
+
+    /* Note the baseline filter values for each segment */
+    if (alt_flt_enabled)
+    {
+        for (i = 0; i < MAX_MB_SEGMENTS; i++)
+        {
+            /* Abs value */
+            if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA)
+                baseline_filter_level[i] = mbd->segment_feature_data[MB_LVL_ALT_LF][i];
+            /* Delta Value */
+            else
+            {
+                baseline_filter_level[i] = default_filt_lvl + mbd->segment_feature_data[MB_LVL_ALT_LF][i];
+                baseline_filter_level[i] = (baseline_filter_level[i] >= 0) ? ((baseline_filter_level[i] <= MAX_LOOP_FILTER) ? baseline_filter_level[i] : MAX_LOOP_FILTER) : 0;  /* Clamp to valid range */
+            }
+        }
+    }
+    else
+    {
+        for (i = 0; i < MAX_MB_SEGMENTS; i++)
+            baseline_filter_level[i] = default_filt_lvl;
+    }
 
     /* Initialize the loop filter for this frame. */
-    vp8_loop_filter_frame_init(cm, mbd, cm->filter_level);
+    if ((cm->last_filter_type != cm->filter_type) || (cm->last_sharpness_level != cm->sharpness_level))
+        vp8_init_loop_filter(cm);
+    else if (frame_type != cm->last_frame_type)
+        vp8_frame_init_loop_filter(lfi, frame_type);
 
     /* Set up the buffer pointers */
     y_ptr = post->y_buffer;
@@ -314,108 +363,102 @@ void vp8_loop_filter_frame
     {
         for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
         {
-            int skip_lf = (mode_info_context->mbmi.mode != B_PRED &&
-                            mode_info_context->mbmi.mode != SPLITMV &&
-                            mode_info_context->mbmi.mb_skip_coeff);
+            int Segment = (alt_flt_enabled) ? mbd->mode_info_context->mbmi.segment_id : 0;
 
-            const int mode_index = lfi_n->mode_lf_lut[mode_info_context->mbmi.mode];
-            const int seg = mode_info_context->mbmi.segment_id;
-            const int ref_frame = mode_info_context->mbmi.ref_frame;
+            filter_level = baseline_filter_level[Segment];
 
-            filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
+            /* Distance of Mb to the various image edges.
+             * These specified to 8th pel as they are always compared to values that are in 1/8th pel units
+             * Apply any context driven MB level adjustment
+             */
+            filter_level = vp8_adjust_mb_lf_value(mbd, filter_level);
 
             if (filter_level)
             {
-                if (cm->filter_type == NORMAL_LOOPFILTER)
-                {
-                    const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level];
-                    lfi.mblim = lfi_n->mblim[filter_level];
-                    lfi.blim = lfi_n->blim[filter_level];
-                    lfi.lim = lfi_n->lim[filter_level];
-                    lfi.hev_thr = lfi_n->hev_thr[hev_index];
-
                 if (mb_col > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_v)
-                        (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi);
+                    cm->lf_mbv(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
 
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_b_v)
-                        (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi);
+                if (mbd->mode_info_context->mbmi.dc_diff > 0)
+                    cm->lf_bv(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
 
                 /* don't apply across umv border */
                 if (mb_row > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_h)
-                        (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi);
+                    cm->lf_mbh(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
 
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_b_h)
-                        (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi);
-                }
-                else
-                {
-                    if (mb_col > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_mb_v)
-                        (y_ptr, post->y_stride, lfi_n->mblim[filter_level]);
-
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_b_v)
-                        (y_ptr, post->y_stride, lfi_n->blim[filter_level]);
-
-                    /* don't apply across umv border */
-                    if (mb_row > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_mb_h)
-                        (y_ptr, post->y_stride, lfi_n->mblim[filter_level]);
-
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_b_h)
-                        (y_ptr, post->y_stride, lfi_n->blim[filter_level]);
-                }
+                if (mbd->mode_info_context->mbmi.dc_diff > 0)
+                    cm->lf_bh(y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
             }
 
             y_ptr += 16;
             u_ptr += 8;
             v_ptr += 8;
 
-            mode_info_context++;     /* step to next MB */
+            mbd->mode_info_context++;     /* step to next MB */
         }
 
         y_ptr += post->y_stride  * 16 - post->y_width;
         u_ptr += post->uv_stride *  8 - post->uv_width;
         v_ptr += post->uv_stride *  8 - post->uv_width;
 
-        mode_info_context++;         /* Skip border mb */
+        mbd->mode_info_context++;         /* Skip border mb */
     }
 }
 
+
+/* Encoder only... */
 void vp8_loop_filter_frame_yonly
 (
     VP8_COMMON *cm,
     MACROBLOCKD *mbd,
-    int default_filt_lvl
+    int default_filt_lvl,
+    int sharpness_lvl
 )
 {
     YV12_BUFFER_CONFIG *post = cm->frame_to_show;
 
+    int i;
     unsigned char *y_ptr;
     int mb_row;
     int mb_col;
 
-    loop_filter_info_n *lfi_n = &cm->lf_info;
-    loop_filter_info lfi;
-
+    loop_filter_info *lfi = cm->lf_info;
+    int baseline_filter_level[MAX_MB_SEGMENTS];
     int filter_level;
+    int alt_flt_enabled = mbd->segmentation_enabled;
     FRAME_TYPE frame_type = cm->frame_type;
 
-    /* Point at base of Mb MODE_INFO list */
-    const MODE_INFO *mode_info_context = cm->mi;
+    (void) sharpness_lvl;
 
-#if 0
-    if(default_filt_lvl == 0) /* no filter applied */
-        return;
-#endif
+    /*MODE_INFO * this_mb_mode_info = cm->mi;*/ /* Point at base of Mb MODE_INFO list */
+    mbd->mode_info_context = cm->mi;          /* Point at base of Mb MODE_INFO list */
+
+    /* Note the baseline filter values for each segment */
+    if (alt_flt_enabled)
+    {
+        for (i = 0; i < MAX_MB_SEGMENTS; i++)
+        {
+            /* Abs value */
+            if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA)
+                baseline_filter_level[i] = mbd->segment_feature_data[MB_LVL_ALT_LF][i];
+            /* Delta Value */
+            else
+            {
+                baseline_filter_level[i] = default_filt_lvl + mbd->segment_feature_data[MB_LVL_ALT_LF][i];
+                baseline_filter_level[i] = (baseline_filter_level[i] >= 0) ? ((baseline_filter_level[i] <= MAX_LOOP_FILTER) ? baseline_filter_level[i] : MAX_LOOP_FILTER) : 0;  /* Clamp to valid range */
+            }
+        }
+    }
+    else
+    {
+        for (i = 0; i < MAX_MB_SEGMENTS; i++)
+            baseline_filter_level[i] = default_filt_lvl;
+    }
 
     /* Initialize the loop filter for this frame. */
-    vp8_loop_filter_frame_init( cm, mbd, default_filt_lvl);
+    if ((cm->last_filter_type != cm->filter_type) || (cm->last_sharpness_level != cm->sharpness_level))
+        vp8_init_loop_filter(cm);
+    else if (frame_type != cm->last_frame_type)
+        vp8_frame_init_loop_filter(lfi, frame_type);
 
     /* Set up the buffer pointers */
     y_ptr = post->y_buffer;
@@ -425,106 +468,72 @@ void vp8_loop_filter_frame_yonly
     {
         for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
         {
-            int skip_lf = (mode_info_context->mbmi.mode != B_PRED &&
-                            mode_info_context->mbmi.mode != SPLITMV &&
-                            mode_info_context->mbmi.mb_skip_coeff);
+            int Segment = (alt_flt_enabled) ? mbd->mode_info_context->mbmi.segment_id : 0;
+            filter_level = baseline_filter_level[Segment];
 
-            const int mode_index = lfi_n->mode_lf_lut[mode_info_context->mbmi.mode];
-            const int seg = mode_info_context->mbmi.segment_id;
-            const int ref_frame = mode_info_context->mbmi.ref_frame;
-
-            filter_level = lfi_n->lvl[seg][ref_frame][mode_index];
+            /* Apply any context driven MB level adjustment */
+            filter_level = vp8_adjust_mb_lf_value(mbd, filter_level);
 
             if (filter_level)
             {
-                if (cm->filter_type == NORMAL_LOOPFILTER)
-                {
-                    const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level];
-                    lfi.mblim = lfi_n->mblim[filter_level];
-                    lfi.blim = lfi_n->blim[filter_level];
-                    lfi.lim = lfi_n->lim[filter_level];
-                    lfi.hev_thr = lfi_n->hev_thr[hev_index];
-
                 if (mb_col > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_v)
-                        (y_ptr, 0, 0, post->y_stride, 0, &lfi);
+                    cm->lf_mbv(y_ptr, 0, 0, post->y_stride, 0, &lfi[filter_level], 0);
 
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_b_v)
-                        (y_ptr, 0, 0, post->y_stride, 0, &lfi);
+                if (mbd->mode_info_context->mbmi.dc_diff > 0)
+                    cm->lf_bv(y_ptr, 0, 0, post->y_stride, 0, &lfi[filter_level], 0);
 
                 /* don't apply across umv border */
                 if (mb_row > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_h)
-                        (y_ptr, 0, 0, post->y_stride, 0, &lfi);
+                    cm->lf_mbh(y_ptr, 0, 0, post->y_stride, 0, &lfi[filter_level], 0);
 
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_b_h)
-                        (y_ptr, 0, 0, post->y_stride, 0, &lfi);
-                }
-                else
-                {
-                    if (mb_col > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_mb_v)
-                        (y_ptr, post->y_stride, lfi_n->mblim[filter_level]);
-
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_b_v)
-                        (y_ptr, post->y_stride, lfi_n->blim[filter_level]);
-
-                    /* don't apply across umv border */
-                    if (mb_row > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_mb_h)
-                        (y_ptr, post->y_stride, lfi_n->mblim[filter_level]);
-
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_b_h)
-                        (y_ptr, post->y_stride, lfi_n->blim[filter_level]);
-                }
+                if (mbd->mode_info_context->mbmi.dc_diff > 0)
+                    cm->lf_bh(y_ptr, 0, 0, post->y_stride, 0, &lfi[filter_level], 0);
             }
 
             y_ptr += 16;
-            mode_info_context ++;        /* step to next MB */
+            mbd->mode_info_context ++;        /* step to next MB */
 
         }
 
         y_ptr += post->y_stride  * 16 - post->y_width;
-        mode_info_context ++;            /* Skip border mb */
+        mbd->mode_info_context ++;            /* Skip border mb */
     }
 
 }
 
+/* Encoder only... */
 void vp8_loop_filter_partial_frame
 (
     VP8_COMMON *cm,
     MACROBLOCKD *mbd,
-    int default_filt_lvl
+    int default_filt_lvl,
+    int sharpness_lvl,
+    int Fraction
 )
 {
     YV12_BUFFER_CONFIG *post = cm->frame_to_show;
 
+    int i;
     unsigned char *y_ptr;
     int mb_row;
     int mb_col;
+    /*int mb_rows = post->y_height >> 4;*/
     int mb_cols = post->y_width  >> 4;
 
-    int linestocopy, i;
-
-    loop_filter_info_n *lfi_n = &cm->lf_info;
-    loop_filter_info lfi;
+    int linestocopy;
 
+    loop_filter_info *lfi = cm->lf_info;
+    int baseline_filter_level[MAX_MB_SEGMENTS];
     int filter_level;
     int alt_flt_enabled = mbd->segmentation_enabled;
     FRAME_TYPE frame_type = cm->frame_type;
 
-    const MODE_INFO *mode_info_context;
+    (void) sharpness_lvl;
 
-    int lvl_seg[MAX_MB_SEGMENTS];
+    /*MODE_INFO * this_mb_mode_info = cm->mi + (post->y_height>>5) * (mb_cols + 1);*/ /* Point at base of Mb MODE_INFO list */
+    mbd->mode_info_context = cm->mi + (post->y_height >> 5) * (mb_cols + 1);        /* Point at base of Mb MODE_INFO list */
 
-    mode_info_context = cm->mi + (post->y_height >> 5) * (mb_cols + 1);
-
-    /* 3 is a magic number. 4 is probably magic too */
-    linestocopy = (post->y_height >> (4 + 3));
+    linestocopy = (post->y_height >> (4 + Fraction));
 
     if (linestocopy < 1)
         linestocopy = 1;
@@ -532,27 +541,32 @@ void vp8_loop_filter_partial_frame
     linestocopy <<= 4;
 
     /* Note the baseline filter values for each segment */
-    /* See vp8_loop_filter_frame_init. Rather than call that for each change
-     * to default_filt_lvl, copy the relevant calculation here.
-     */
     if (alt_flt_enabled)
     {
         for (i = 0; i < MAX_MB_SEGMENTS; i++)
-        {    /* Abs value */
-            if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA)
         {
-                lvl_seg[i] = mbd->segment_feature_data[MB_LVL_ALT_LF][i];
-            }
+            /* Abs value */
+            if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA)
+                baseline_filter_level[i] = mbd->segment_feature_data[MB_LVL_ALT_LF][i];
             /* Delta Value */
             else
             {
-                lvl_seg[i] = default_filt_lvl
-                        + mbd->segment_feature_data[MB_LVL_ALT_LF][i];
-                lvl_seg[i] = (lvl_seg[i] > 0) ?
-                        ((lvl_seg[i] > 63) ? 63: lvl_seg[i]) : 0;
+                baseline_filter_level[i] = default_filt_lvl + mbd->segment_feature_data[MB_LVL_ALT_LF][i];
+                baseline_filter_level[i] = (baseline_filter_level[i] >= 0) ? ((baseline_filter_level[i] <= MAX_LOOP_FILTER) ? baseline_filter_level[i] : MAX_LOOP_FILTER) : 0;  /* Clamp to valid range */
             }
         }
     }
+    else
+    {
+        for (i = 0; i < MAX_MB_SEGMENTS; i++)
+            baseline_filter_level[i] = default_filt_lvl;
+    }
+
+    /* Initialize the loop filter for this frame. */
+    if ((cm->last_filter_type != cm->filter_type) || (cm->last_sharpness_level != cm->sharpness_level))
+        vp8_init_loop_filter(cm);
+    else if (frame_type != cm->last_frame_type)
+        vp8_frame_init_loop_filter(lfi, frame_type);
 
     /* Set up the buffer pointers */
     y_ptr = post->y_buffer + (post->y_height >> 5) * 16 * post->y_stride;
@@ -562,64 +576,28 @@ void vp8_loop_filter_partial_frame
     {
         for (mb_col = 0; mb_col < mb_cols; mb_col++)
         {
-            int skip_lf = (mode_info_context->mbmi.mode != B_PRED &&
-                           mode_info_context->mbmi.mode != SPLITMV &&
-                           mode_info_context->mbmi.mb_skip_coeff);
-
-            if (alt_flt_enabled)
-                filter_level = lvl_seg[mode_info_context->mbmi.segment_id];
-            else
-                filter_level = default_filt_lvl;
+            int Segment = (alt_flt_enabled) ? mbd->mode_info_context->mbmi.segment_id : 0;
+            filter_level = baseline_filter_level[Segment];
 
             if (filter_level)
             {
-                if (cm->filter_type == NORMAL_LOOPFILTER)
-                {
-                    const int hev_index = lfi_n->hev_thr_lut[frame_type][filter_level];
-                    lfi.mblim = lfi_n->mblim[filter_level];
-                    lfi.blim = lfi_n->blim[filter_level];
-                    lfi.lim = lfi_n->lim[filter_level];
-                    lfi.hev_thr = lfi_n->hev_thr[hev_index];
-
                 if (mb_col > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_v)
-                        (y_ptr, 0, 0, post->y_stride, 0, &lfi);
+                    cm->lf_mbv(y_ptr, 0, 0, post->y_stride, 0, &lfi[filter_level], 0);
 
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_b_v)
-                        (y_ptr, 0, 0, post->y_stride, 0, &lfi);
+                if (mbd->mode_info_context->mbmi.dc_diff > 0)
+                    cm->lf_bv(y_ptr, 0, 0, post->y_stride, 0, &lfi[filter_level], 0);
 
-                    LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_h)
-                        (y_ptr, 0, 0, post->y_stride, 0, &lfi);
+                cm->lf_mbh(y_ptr, 0, 0, post->y_stride, 0, &lfi[filter_level], 0);
 
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, normal_b_h)
-                        (y_ptr, 0, 0, post->y_stride, 0, &lfi);
-                }
-                else
-                {
-                    if (mb_col > 0)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_mb_v)
-                        (y_ptr, post->y_stride, lfi_n->mblim[filter_level]);
-
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_b_v)
-                        (y_ptr, post->y_stride, lfi_n->blim[filter_level]);
-
-                    LF_INVOKE(&cm->rtcd.loopfilter, simple_mb_h)
-                        (y_ptr, post->y_stride, lfi_n->mblim[filter_level]);
-
-                    if (!skip_lf)
-                        LF_INVOKE(&cm->rtcd.loopfilter, simple_b_h)
-                        (y_ptr, post->y_stride, lfi_n->blim[filter_level]);
-                }
+                if (mbd->mode_info_context->mbmi.dc_diff > 0)
+                    cm->lf_bh(y_ptr, 0, 0, post->y_stride, 0, &lfi[filter_level], 0);
             }
 
             y_ptr += 16;
-            mode_info_context += 1;      /* step to next MB */
+            mbd->mode_info_context += 1;      /* step to next MB */
         }
 
         y_ptr += post->y_stride  * 16 - post->y_width;
-        mode_info_context += 1;          /* Skip border mb */
+        mbd->mode_info_context += 1;          /* Skip border mb */
     }
 }

vp8/common/loopfilter.h

@@ -13,7 +13,6 @@
 #define loopfilter_h
 
 #include "vpx_ports/mem.h"
-#include "vpx_config.h"
 
 #define MAX_LOOP_FILTER 63
 
@@ -23,45 +22,26 @@ typedef enum
     SIMPLE_LOOPFILTER = 1
 } LOOPFILTERTYPE;
 
-#if ARCH_ARM
-#define SIMD_WIDTH 1
-#else
-#define SIMD_WIDTH 16
-#endif
-
-/* Need to align this structure so when it is declared and
+/* FRK
+ * Need to align this structure so when it is declared and
  * passed it can be loaded into vector registers.
  */
 typedef struct
 {
-    DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, mblim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]);
-    DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, blim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]);
-    DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, lim[MAX_LOOP_FILTER + 1][SIMD_WIDTH]);
-    DECLARE_ALIGNED(SIMD_WIDTH, unsigned char, hev_thr[4][SIMD_WIDTH]);
-    unsigned char lvl[4][4][4];
-    unsigned char hev_thr_lut[2][MAX_LOOP_FILTER + 1];
-    unsigned char mode_lf_lut[10];
-} loop_filter_info_n;
-
-typedef struct
-{
-    const unsigned char * mblim;
-    const unsigned char * blim;
-    const unsigned char * lim;
-    const unsigned char * hev_thr;
+    DECLARE_ALIGNED(16, signed char, lim[16]);
+    DECLARE_ALIGNED(16, signed char, flim[16]);
+    DECLARE_ALIGNED(16, signed char, thr[16]);
+    DECLARE_ALIGNED(16, signed char, mbflim[16]);
 } loop_filter_info;
 
 
 #define prototype_loopfilter(sym) \
-    void sym(unsigned char *src, int pitch, const unsigned char *blimit,\
-             const unsigned char *limit, const unsigned char *thresh, int count)
+    void sym(unsigned char *src, int pitch, const signed char *flimit,\
+             const signed char *limit, const signed char *thresh, int count)
 
 #define prototype_loopfilter_block(sym) \
     void sym(unsigned char *y, unsigned char *u, unsigned char *v,\
-             int ystride, int uv_stride, loop_filter_info *lfi)
-
-#define prototype_simple_loopfilter(sym) \
-    void sym(unsigned char *y, int ystride, const unsigned char *blimit)
+             int ystride, int uv_stride, loop_filter_info *lfi, int simpler)
 
 #if ARCH_X86 || ARCH_X86_64
 #include "x86/loopfilter_x86.h"
@@ -91,39 +71,38 @@ extern prototype_loopfilter_block(vp8_lf_normal_mb_h);
 #endif
 extern prototype_loopfilter_block(vp8_lf_normal_b_h);
 
+
 #ifndef vp8_lf_simple_mb_v
-#define vp8_lf_simple_mb_v vp8_loop_filter_simple_vertical_edge_c
+#define vp8_lf_simple_mb_v vp8_loop_filter_mbvs_c
 #endif
-extern prototype_simple_loopfilter(vp8_lf_simple_mb_v);
+extern prototype_loopfilter_block(vp8_lf_simple_mb_v);
 
 #ifndef vp8_lf_simple_b_v
 #define vp8_lf_simple_b_v vp8_loop_filter_bvs_c
 #endif
-extern prototype_simple_loopfilter(vp8_lf_simple_b_v);
+extern prototype_loopfilter_block(vp8_lf_simple_b_v);
 
 #ifndef vp8_lf_simple_mb_h
-#define vp8_lf_simple_mb_h vp8_loop_filter_simple_horizontal_edge_c
+#define vp8_lf_simple_mb_h vp8_loop_filter_mbhs_c
 #endif
-extern prototype_simple_loopfilter(vp8_lf_simple_mb_h);
+extern prototype_loopfilter_block(vp8_lf_simple_mb_h);
 
 #ifndef vp8_lf_simple_b_h
 #define vp8_lf_simple_b_h vp8_loop_filter_bhs_c
 #endif
-extern prototype_simple_loopfilter(vp8_lf_simple_b_h);
+extern prototype_loopfilter_block(vp8_lf_simple_b_h);
 
 typedef prototype_loopfilter_block((*vp8_lf_block_fn_t));
-typedef prototype_simple_loopfilter((*vp8_slf_block_fn_t));
-
 typedef struct
 {
     vp8_lf_block_fn_t  normal_mb_v;
     vp8_lf_block_fn_t  normal_b_v;
     vp8_lf_block_fn_t  normal_mb_h;
     vp8_lf_block_fn_t  normal_b_h;
-    vp8_slf_block_fn_t  simple_mb_v;
-    vp8_slf_block_fn_t  simple_b_v;
-    vp8_slf_block_fn_t  simple_mb_h;
-    vp8_slf_block_fn_t  simple_b_h;
+    vp8_lf_block_fn_t  simple_mb_v;
+    vp8_lf_block_fn_t  simple_b_v;
+    vp8_lf_block_fn_t  simple_mb_h;
+    vp8_lf_block_fn_t  simple_b_h;
 } vp8_loopfilter_rtcd_vtable_t;
 
 #if CONFIG_RUNTIME_CPU_DETECT
@@ -136,33 +115,10 @@ typedef void loop_filter_uvfunction
 (
     unsigned char *u,   /* source pointer */
     int p,              /* pitch */
-    const unsigned char *blimit,
-    const unsigned char *limit,
-    const unsigned char *thresh,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
     unsigned char *v
 );
 
-/* assorted loopfilter functions which get used elsewhere */
-struct VP8Common;
-struct MacroBlockD;
-
-void vp8_loop_filter_init(struct VP8Common *cm);
-
-void vp8_loop_filter_frame_init(struct VP8Common *cm,
-                                struct MacroBlockD *mbd,
-                                int default_filt_lvl);
-
-void vp8_loop_filter_frame(struct VP8Common *cm, struct MacroBlockD *mbd);
-
-void vp8_loop_filter_partial_frame(struct VP8Common *cm,
-                                   struct MacroBlockD *mbd,
-                                   int default_filt_lvl);
-
-void vp8_loop_filter_frame_yonly(struct VP8Common *cm,
-                                 struct MacroBlockD *mbd,
-                                 int default_filt_lvl);
-
-void vp8_loop_filter_update_sharpness(loop_filter_info_n *lfi,
-                                      int sharpness_lvl);
-
 #endif

vp8/common/loopfilter_filters.c

@@ -24,9 +24,8 @@ static __inline signed char vp8_signed_char_clamp(int t)
 
 
 /* should we apply any filter at all ( 11111111 yes, 00000000 no) */
-static __inline signed char vp8_filter_mask(uc limit, uc blimit,
-                                     uc p3, uc p2, uc p1, uc p0,
-                                     uc q0, uc q1, uc q2, uc q3)
+static __inline signed char vp8_filter_mask(signed char limit, signed char flimit,
+                                     uc p3, uc p2, uc p1, uc p0, uc q0, uc q1, uc q2, uc q3)
 {
     signed char mask = 0;
     mask |= (abs(p3 - p2) > limit) * -1;
@@ -35,13 +34,13 @@ static __inline signed char vp8_filter_mask(uc limit, uc blimit,
     mask |= (abs(q1 - q0) > limit) * -1;
     mask |= (abs(q2 - q1) > limit) * -1;
     mask |= (abs(q3 - q2) > limit) * -1;
-    mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > blimit) * -1;
+    mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > flimit * 2 + limit) * -1;
     mask = ~mask;
     return mask;
 }
 
 /* is there high variance internal edge ( 11111111 yes, 00000000 no) */
-static __inline signed char vp8_hevmask(uc thresh, uc p1, uc p0, uc q0, uc q1)
+static __inline signed char vp8_hevmask(signed char thresh, uc p1, uc p0, uc q0, uc q1)
 {
     signed char hev = 0;
     hev  |= (abs(p1 - p0) > thresh) * -1;
@@ -49,9 +48,7 @@ static __inline signed char vp8_hevmask(uc thresh, uc p1, uc p0, uc q0, uc q1)
     return hev;
 }
 
-static __inline void vp8_filter(signed char mask, uc hev, uc *op1,
-        uc *op0, uc *oq0, uc *oq1)
-
+static __inline void vp8_filter(signed char mask, signed char hev, uc *op1, uc *op0, uc *oq0, uc *oq1)
 {
     signed char ps0, qs0;
     signed char ps1, qs1;
@@ -96,13 +93,14 @@ static __inline void vp8_filter(signed char mask, uc hev, uc *op1,
     *op1 = u ^ 0x80;
 
 }
+
 void vp8_loop_filter_horizontal_edge_c
 (
     unsigned char *s,
     int p, /* pitch */
-    const unsigned char *blimit,
-    const unsigned char *limit,
-    const unsigned char *thresh,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
     int count
 )
 {
@@ -115,11 +113,11 @@ void vp8_loop_filter_horizontal_edge_c
      */
     do
     {
-        mask = vp8_filter_mask(limit[0], blimit[0],
+        mask = vp8_filter_mask(limit[i], flimit[i],
                                s[-4*p], s[-3*p], s[-2*p], s[-1*p],
                                s[0*p], s[1*p], s[2*p], s[3*p]);
 
-        hev = vp8_hevmask(thresh[0], s[-2*p], s[-1*p], s[0*p], s[1*p]);
+        hev = vp8_hevmask(thresh[i], s[-2*p], s[-1*p], s[0*p], s[1*p]);
 
         vp8_filter(mask, hev, s - 2 * p, s - 1 * p, s, s + 1 * p);
 
@@ -132,9 +130,9 @@ void vp8_loop_filter_vertical_edge_c
 (
     unsigned char *s,
     int p,
-    const unsigned char *blimit,
-    const unsigned char *limit,
-    const unsigned char *thresh,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
     int count
 )
 {
@@ -147,10 +145,10 @@ void vp8_loop_filter_vertical_edge_c
      */
     do
     {
-        mask = vp8_filter_mask(limit[0], blimit[0],
+        mask = vp8_filter_mask(limit[i], flimit[i],
                                s[-4], s[-3], s[-2], s[-1], s[0], s[1], s[2], s[3]);
 
-        hev = vp8_hevmask(thresh[0], s[-2], s[-1], s[0], s[1]);
+        hev = vp8_hevmask(thresh[i], s[-2], s[-1], s[0], s[1]);
 
         vp8_filter(mask, hev, s - 2, s - 1, s, s + 1);
 
@@ -159,7 +157,7 @@ void vp8_loop_filter_vertical_edge_c
     while (++i < count * 8);
 }
 
-static __inline void vp8_mbfilter(signed char mask, uc hev,
+static __inline void vp8_mbfilter(signed char mask, signed char hev,
                            uc *op2, uc *op1, uc *op0, uc *oq0, uc *oq1, uc *oq2)
 {
     signed char s, u;
@@ -218,9 +216,9 @@ void vp8_mbloop_filter_horizontal_edge_c
 (
     unsigned char *s,
     int p,
-    const unsigned char *blimit,
-    const unsigned char *limit,
-    const unsigned char *thresh,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
     int count
 )
 {
@@ -234,11 +232,11 @@ void vp8_mbloop_filter_horizontal_edge_c
     do
     {
 
-        mask = vp8_filter_mask(limit[0], blimit[0],
+        mask = vp8_filter_mask(limit[i], flimit[i],
                                s[-4*p], s[-3*p], s[-2*p], s[-1*p],
                                s[0*p], s[1*p], s[2*p], s[3*p]);
 
-        hev = vp8_hevmask(thresh[0], s[-2*p], s[-1*p], s[0*p], s[1*p]);
+        hev = vp8_hevmask(thresh[i], s[-2*p], s[-1*p], s[0*p], s[1*p]);
 
         vp8_mbfilter(mask, hev, s - 3 * p, s - 2 * p, s - 1 * p, s, s + 1 * p, s + 2 * p);
 
@@ -253,9 +251,9 @@ void vp8_mbloop_filter_vertical_edge_c
 (
     unsigned char *s,
     int p,
-    const unsigned char *blimit,
-    const unsigned char *limit,
-    const unsigned char *thresh,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
     int count
 )
 {
@@ -266,10 +264,10 @@ void vp8_mbloop_filter_vertical_edge_c
     do
     {
 
-        mask = vp8_filter_mask(limit[0], blimit[0],
+        mask = vp8_filter_mask(limit[i], flimit[i],
                                s[-4], s[-3], s[-2], s[-1], s[0], s[1], s[2], s[3]);
 
-        hev = vp8_hevmask(thresh[0], s[-2], s[-1], s[0], s[1]);
+        hev = vp8_hevmask(thresh[i], s[-2], s[-1], s[0], s[1]);
 
         vp8_mbfilter(mask, hev, s - 3, s - 2, s - 1, s, s + 1, s + 2);
 
@@ -280,13 +278,13 @@ void vp8_mbloop_filter_vertical_edge_c
 }
 
 /* should we apply any filter at all ( 11111111 yes, 00000000 no) */
-static __inline signed char vp8_simple_filter_mask(uc blimit, uc p1, uc p0, uc q0, uc q1)
+static __inline signed char vp8_simple_filter_mask(signed char limit, signed char flimit, uc p1, uc p0, uc q0, uc q1)
 {
 /* Why does this cause problems for win32?
  * error C2143: syntax error : missing ';' before 'type'
  *  (void) limit;
  */
-    signed char mask = (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  <= blimit) * -1;
+    signed char mask = (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  <= flimit * 2 + limit) * -1;
     return mask;
 }
 
@@ -319,37 +317,47 @@ void vp8_loop_filter_simple_horizontal_edge_c
 (
     unsigned char *s,
     int p,
-    const unsigned char *blimit
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count
 )
 {
     signed char mask = 0;
     int i = 0;
+    (void) thresh;
 
     do
     {
-        mask = vp8_simple_filter_mask(blimit[0], s[-2*p], s[-1*p], s[0*p], s[1*p]);
+        /*mask = vp8_simple_filter_mask( limit[i], flimit[i],s[-1*p],s[0*p]);*/
+        mask = vp8_simple_filter_mask(limit[i], flimit[i], s[-2*p], s[-1*p], s[0*p], s[1*p]);
         vp8_simple_filter(mask, s - 2 * p, s - 1 * p, s, s + 1 * p);
         ++s;
     }
-    while (++i < 16);
+    while (++i < count * 8);
 }
 
 void vp8_loop_filter_simple_vertical_edge_c
 (
     unsigned char *s,
     int p,
-    const unsigned char *blimit
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count
 )
 {
     signed char mask = 0;
     int i = 0;
+    (void) thresh;
 
     do
     {
-        mask = vp8_simple_filter_mask(blimit[0], s[-2], s[-1], s[0], s[1]);
+        /*mask = vp8_simple_filter_mask( limit[i], flimit[i],s[-1],s[0]);*/
+        mask = vp8_simple_filter_mask(limit[i], flimit[i], s[-2], s[-1], s[0], s[1]);
         vp8_simple_filter(mask, s - 2, s - 1, s, s + 1);
         s += p;
     }
-    while (++i < 16);
+    while (++i < count * 8);
 
 }

vp8/common/mbpitch.c

@@ -11,6 +11,12 @@
 
 #include "blockd.h"
 
+#include "stdio.h"
+#include "vpx_config.h"
+#if CONFIG_OPENCL
+#include "opencl/vp8_opencl.h"
+#endif
+
 typedef enum
 {
     PRED = 0,
@@ -20,7 +26,6 @@ typedef enum
 static void setup_block
 (
     BLOCKD *b,
-    int mv_stride,
     unsigned char **base,
     int Stride,
     int offset,
@@ -49,81 +54,176 @@ static void setup_macroblock(MACROBLOCKD *x, BLOCKSET bs)
     int block;
 
     unsigned char **y, **u, **v;
+    unsigned char **buf_base;
+    int y_off, u_off, v_off;
 
     if (bs == DEST)
     {
+        buf_base = &x->dst.buffer_alloc;
+        y_off = x->dst.y_buffer - x->dst.buffer_alloc;
+        u_off = x->dst.u_buffer - x->dst.buffer_alloc;
+        v_off = x->dst.v_buffer - x->dst.buffer_alloc;
         y = &x->dst.y_buffer;
         u = &x->dst.u_buffer;
         v = &x->dst.v_buffer;
+        y_off = 0;
+
+        //y = buf_base;
+        //y_off = x->dst.y_buffer - x->dst.buffer_alloc;
+        
+        u = buf_base;
+        v = buf_base;
+
+        u_off = x->dst.u_buffer - x->dst.buffer_alloc;
+        v_off = x->dst.v_buffer - x->dst.buffer_alloc;
+
     }
     else
     {
+        buf_base = &x->pre.buffer_alloc;
         y = &x->pre.y_buffer;
         u = &x->pre.u_buffer;
         v = &x->pre.v_buffer;
+        y_off = u_off = v_off = 0;
+
+        //y = buf_base;
+        //y_off = x->pre.y_buffer - x->pre.buffer_alloc;
+        //u = buf_base;
+        //u_off = x->pre.u_buffer - x->pre.buffer_alloc;
+        //v = buf_base;
+        //v_off = x->pre.v_buffer - x->pre.buffer_alloc;
     }
 
     for (block = 0; block < 16; block++) /* y blocks */
     {
-        setup_block(&x->block[block], x->dst.y_stride, y, x->dst.y_stride,
-                        (block >> 2) * 4 * x->dst.y_stride + (block & 3) * 4, bs);
+        setup_block(&x->block[block], y, x->dst.y_stride,
+                        y_off + ((block >> 2) * 4 * x->dst.y_stride + (block & 3) * 4), bs);
     }
 
     for (block = 16; block < 20; block++) /* U and V blocks */
     {
-        setup_block(&x->block[block], x->dst.uv_stride, u, x->dst.uv_stride,
-                        ((block - 16) >> 1) * 4 * x->dst.uv_stride + (block & 1) * 4, bs);
+        int block_off = ((block - 16) >> 1) * 4 * x->dst.uv_stride + (block & 1) * 4;
 
-        setup_block(&x->block[block+4], x->dst.uv_stride, v, x->dst.uv_stride,
-                        ((block - 16) >> 1) * 4 * x->dst.uv_stride + (block & 1) * 4, bs);
+        setup_block(&x->block[block], u, x->dst.uv_stride,
+                        u_off + block_off, bs);
+
+        setup_block(&x->block[block+4], v, x->dst.uv_stride,
+                        v_off + block_off, bs);
     }
 }
 
 void vp8_setup_block_dptrs(MACROBLOCKD *x)
 {
     int r, c;
+    unsigned int offset;
 
+#if CONFIG_OPENCL && !ONE_CQ_PER_MB
+    cl_command_queue y_cq, u_cq, v_cq;
+    int err;
+    if (cl_initialized == CL_SUCCESS){
+        //Create command queue for Y/U/V Planes
+        y_cq = clCreateCommandQueue(cl_data.context, cl_data.device_id, 0, &err);
+        if (!y_cq || err != CL_SUCCESS) {
+            printf("Error: Failed to create a command queue!\n");
+            cl_destroy(NULL, VP8_CL_TRIED_BUT_FAILED);
+        }
+        u_cq = clCreateCommandQueue(cl_data.context, cl_data.device_id, 0, &err);
+        if (!u_cq || err != CL_SUCCESS) {
+            printf("Error: Failed to create a command queue!\n");
+            cl_destroy(NULL, VP8_CL_TRIED_BUT_FAILED);
+        }
+        v_cq = clCreateCommandQueue(cl_data.context, cl_data.device_id, 0, &err);
+        if (!v_cq || err != CL_SUCCESS) {
+            printf("Error: Failed to create a command queue!\n");
+            cl_destroy(NULL, VP8_CL_TRIED_BUT_FAILED);
+        }
+    }
+#endif
+
+    /* 16 Y blocks */
     for (r = 0; r < 4; r++)
     {
         for (c = 0; c < 4; c++)
         {
-            x->block[r*4+c].diff      = &x->diff[r * 4 * 16 + c * 4];
-            x->block[r*4+c].predictor = x->predictor + r * 4 * 16 + c * 4;
+            offset = r * 4 * 16 + c * 4;
+            x->block[r*4+c].diff_offset      = offset;
+            x->block[r*4+c].predictor_offset = offset;
+#if CONFIG_OPENCL && !ONE_CQ_PER_MB
+            if (cl_initialized == CL_SUCCESS)
+                x->block[r*4+c].cl_commands = y_cq;
+#endif
         }
     }
 
+    /* 4 U Blocks */
     for (r = 0; r < 2; r++)
     {
         for (c = 0; c < 2; c++)
         {
-            x->block[16+r*2+c].diff      = &x->diff[256 + r * 4 * 8 + c * 4];
-            x->block[16+r*2+c].predictor = x->predictor + 256 + r * 4 * 8 + c * 4;
+            offset = 256 + r * 4 * 8 + c * 4;
+            x->block[16+r*2+c].diff_offset      = offset;
+            x->block[16+r*2+c].predictor_offset = offset;
 
+#if CONFIG_OPENCL && !ONE_CQ_PER_MB
+            if (cl_initialized == CL_SUCCESS)
+                x->block[16+r*2+c].cl_commands = u_cq;
+#endif
         }
     }
 
+    /* 4 V Blocks */
     for (r = 0; r < 2; r++)
     {
         for (c = 0; c < 2; c++)
         {
-            x->block[20+r*2+c].diff      = &x->diff[320+ r * 4 * 8 + c * 4];
-            x->block[20+r*2+c].predictor = x->predictor + 320 + r * 4 * 8 + c * 4;
+            offset = 320+ r * 4 * 8 + c * 4;
+            x->block[20+r*2+c].diff_offset      = offset;
+            x->block[20+r*2+c].predictor_offset = offset;
 
+#if CONFIG_OPENCL && !ONE_CQ_PER_MB
+            if (cl_initialized == CL_SUCCESS)
+                x->block[20+r*2+c].cl_commands = v_cq;
+#endif
         }
     }
 
-    x->block[24].diff = &x->diff[384];
+    x->block[24].diff_offset = 384;
 
     for (r = 0; r < 25; r++)
     {
-        x->block[r].qcoeff  = x->qcoeff  + r * 16;
-        x->block[r].dqcoeff = x->dqcoeff + r * 16;
+    	x->block[r].qcoeff_base = x->qcoeff;
+    	x->block[r].qcoeff_offset = r * 16;
+        x->block[r].dqcoeff_base = x->dqcoeff;
+        x->block[r].dqcoeff_offset = r * 16;
+        
+        x->block[r].predictor_base = x->predictor;
+        x->block[r].diff_base = x->diff;
+        x->block[r].eobs_base = x->eobs;
+
+#if CONFIG_OPENCL
+        if (cl_initialized == CL_SUCCESS){
+            /* Copy command queue reference from macroblock */
+#if ONE_CQ_PER_MB
+            x->block[r].cl_commands = x->cl_commands;
+#endif
+
+            /* Set up CL memory buffers as appropriate */
+            x->block[r].cl_diff_mem = x->cl_diff_mem;
+            x->block[r].cl_dqcoeff_mem = x->cl_dqcoeff_mem;
+            x->block[r].cl_eobs_mem = x->cl_eobs_mem;
+            x->block[r].cl_predictor_mem = x->cl_predictor_mem;
+            x->block[r].cl_qcoeff_mem = x->cl_qcoeff_mem;
         }
+
+        //Copy filter type to block.
+        x->block[r].sixtap_filter = x->sixtap_filter;
+#endif
+    }
+
 }
 
 void vp8_build_block_doffsets(MACROBLOCKD *x)
 {
-
     /* handle the destination pitch features */
     setup_macroblock(x, DEST);
     setup_macroblock(x, PRED);

vp8/common/mv.h

@@ -11,7 +11,6 @@
 
 #ifndef __INC_MV_H
 #define __INC_MV_H
-#include "vpx/vpx_integer.h"
 
 typedef struct
 {
@@ -19,10 +18,4 @@ typedef struct
     short col;
 } MV;
 
-typedef union
-{
-    uint32_t  as_int;
-    MV        as_mv;
-} int_mv;        /* facilitates faster equality tests and copies */
-
 #endif

vp8/common/onyx.h

@@ -109,7 +109,6 @@ extern "C"
         int noise_sensitivity;   // parameter used for applying pre processing blur: recommendation 0
         int Sharpness;          // parameter used for sharpening output: recommendation 0:
         int cpu_used;
-        unsigned int rc_max_intra_bitrate_pct;
 
         // mode ->
         //(0)=Realtime/Live Encoding. This mode is optimized for realtim encoding (for example, capturing
@@ -140,9 +139,8 @@ extern "C"
 
         int end_usage; // vbr or cbr
 
-        // buffer targeting aggressiveness
+        // shoot to keep buffer full at all times by undershooting a bit 95 recommended
         int under_shoot_pct;
-        int over_shoot_pct;
 
         // buffering parameters
         int starting_buffer_level;  // in seconds
@@ -184,11 +182,8 @@ extern "C"
         int token_partitions; // how many token partitions to create for multi core decoding
         int encode_breakout;  // early breakout encode threshold : for video conf recommend 800
 
-        unsigned int error_resilient_mode; // Bitfield defining the error
-                                   // resiliency features to enable. Can provide
-                                   // decodable frames after losses in previous
-                                   // frames and decodable partitions after
-                                   // losses in the same frame.
+        int error_resilient_mode;  // if running over udp networks provides decodable frames after a
+        // dropped packet
 
         int arnr_max_frames;
         int arnr_strength ;
@@ -211,8 +206,8 @@ extern "C"
 
 // receive a frames worth of data caller can assume that a copy of this frame is made
 // and not just a copy of the pointer..
-    int vp8_receive_raw_frame(VP8_PTR comp, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, int64_t time_stamp, int64_t end_time_stamp);
-    int vp8_get_compressed_data(VP8_PTR comp, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, int64_t *time_stamp, int64_t *time_end, int flush);
+    int vp8_receive_raw_frame(VP8_PTR comp, unsigned int frame_flags, YV12_BUFFER_CONFIG *sd, INT64 time_stamp, INT64 end_time_stamp);
+    int vp8_get_compressed_data(VP8_PTR comp, unsigned int *frame_flags, unsigned long *size, unsigned char *dest, INT64 *time_stamp, INT64 *time_end, int flush);
     int vp8_get_preview_raw_frame(VP8_PTR comp, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t *flags);
 
     int vp8_use_as_reference(VP8_PTR comp, int ref_frame_flags);

vp8/common/onyxc_int.h

@@ -19,9 +19,7 @@
 #include "entropy.h"
 #include "idct.h"
 #include "recon.h"
-#if CONFIG_POSTPROC
 #include "postproc.h"
-#endif
 
 /*#ifdef PACKET_TESTING*/
 #include "header.h"
@@ -37,15 +35,13 @@ void vp8_initialize_common(void);
 
 #define NUM_YV12_BUFFERS 4
 
-#define MAX_PARTITIONS 9
-
 typedef struct frame_contexts
 {
     vp8_prob bmode_prob [VP8_BINTRAMODES-1];
     vp8_prob ymode_prob [VP8_YMODES-1];   /* interframe intra mode probs */
     vp8_prob uv_mode_prob [VP8_UV_MODES-1];
     vp8_prob sub_mv_ref_prob [VP8_SUBMVREFS-1];
-    vp8_prob coef_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [ENTROPY_NODES];
+    vp8_prob coef_probs [BLOCK_TYPES] [COEF_BANDS] [PREV_COEF_CONTEXTS] [vp8_coef_tokens-1];
     MV_CONTEXT mvc[2];
     MV_CONTEXT pre_mvc[2];  /* not to caculate the mvcost for the frame if mvc doesn't change. */
 } FRAME_CONTEXT;
@@ -77,9 +73,7 @@ typedef struct VP8_COMMON_RTCD
     vp8_recon_rtcd_vtable_t       recon;
     vp8_subpix_rtcd_vtable_t      subpix;
     vp8_loopfilter_rtcd_vtable_t  loopfilter;
-#if CONFIG_POSTPROC
     vp8_postproc_rtcd_vtable_t    postproc;
-#endif
     int                           flags;
 #else
     int unused;
@@ -87,7 +81,6 @@ typedef struct VP8_COMMON_RTCD
 } VP8_COMMON_RTCD;
 
 typedef struct VP8Common
-
 {
     struct vpx_internal_error_info  error;
 
@@ -112,8 +105,7 @@ typedef struct VP8Common
     YV12_BUFFER_CONFIG post_proc_buffer;
     YV12_BUFFER_CONFIG temp_scale_frame;
 
-
-    FRAME_TYPE last_frame_type;  /* Save last frame's frame type for motion search. */
+    FRAME_TYPE last_frame_type;  /* Save last frame's frame type for loopfilter init checking and motion search. */
     FRAME_TYPE frame_type;
 
     int show_frame;
@@ -127,7 +119,7 @@ typedef struct VP8Common
     /* profile settings */
     int mb_no_coeff_skip;
     int no_lpf;
-    int use_bilinear_mc_filter;
+    int simpler_lpf;
     int full_pixel;
 
     int base_qindex;
@@ -147,15 +139,16 @@ typedef struct VP8Common
 
     MODE_INFO *mip; /* Base of allocated array */
     MODE_INFO *mi;  /* Corresponds to upper left visible macroblock */
-    MODE_INFO *prev_mip; /* MODE_INFO array 'mip' from last decoded frame */
-    MODE_INFO *prev_mi;  /* 'mi' from last frame (points into prev_mip) */
 
 
     INTERPOLATIONFILTERTYPE mcomp_filter_type;
+    LOOPFILTERTYPE last_filter_type;
     LOOPFILTERTYPE filter_type;
-
-    loop_filter_info_n lf_info;
-
+    loop_filter_info lf_info[MAX_LOOP_FILTER+1];
+    prototype_loopfilter_block((*lf_mbv));
+    prototype_loopfilter_block((*lf_mbh));
+    prototype_loopfilter_block((*lf_bv));
+    prototype_loopfilter_block((*lf_bh));
     int filter_level;
     int last_sharpness_level;
     int sharpness_level;
@@ -202,12 +195,13 @@ typedef struct VP8Common
 #if CONFIG_RUNTIME_CPU_DETECT
     VP8_COMMON_RTCD rtcd;
 #endif
-#if CONFIG_MULTITHREAD
-    int processor_core_count;
-#endif
-#if CONFIG_POSTPROC
     struct postproc_state  postproc_state;
-#endif
 } VP8_COMMON;
 
+
+int vp8_adjust_mb_lf_value(MACROBLOCKD *mbd, int filter_level);
+void vp8_init_loop_filter(VP8_COMMON *cm);
+void vp8_frame_init_loop_filter(loop_filter_info *lfi, int frame_type);
+extern void vp8_loop_filter_frame(VP8_COMMON *cm,    MACROBLOCKD *mbd,  int filt_val);
+
 #endif

vp8/common/onyxd.h

@@ -18,12 +18,10 @@
 extern "C"
 {
 #endif
-#include "vpx/vpx_codec.h"
 #include "type_aliases.h"
 #include "vpx_scale/yv12config.h"
 #include "ppflags.h"
 #include "vpx_ports/mem.h"
-#include "vpx/vpx_codec.h"
 
     typedef void   *VP8D_PTR;
     typedef struct
@@ -33,8 +31,6 @@ extern "C"
         int     Version;
         int     postprocess;
         int     max_threads;
-        int     error_concealment;
-        int     input_partition;
     } VP8D_CONFIG;
     typedef enum
     {
@@ -54,11 +50,11 @@ extern "C"
 
     int vp8dx_get_setting(VP8D_PTR comp, VP8D_SETTING oxst);
 
-    int vp8dx_receive_compressed_data(VP8D_PTR comp, unsigned long size, const unsigned char *dest, int64_t time_stamp);
-    int vp8dx_get_raw_frame(VP8D_PTR comp, YV12_BUFFER_CONFIG *sd, int64_t *time_stamp, int64_t *time_end_stamp, vp8_ppflags_t *flags);
+    int vp8dx_receive_compressed_data(VP8D_PTR comp, unsigned long size, const unsigned char *dest, INT64 time_stamp);
+    int vp8dx_get_raw_frame(VP8D_PTR comp, YV12_BUFFER_CONFIG *sd, INT64 *time_stamp, INT64 *time_end_stamp, vp8_ppflags_t *flags);
 
-    vpx_codec_err_t vp8dx_get_reference(VP8D_PTR comp, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_CONFIG *sd);
-    vpx_codec_err_t vp8dx_set_reference(VP8D_PTR comp, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_CONFIG *sd);
+    int vp8dx_get_reference(VP8D_PTR comp, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_CONFIG *sd);
+    int vp8dx_set_reference(VP8D_PTR comp, VP8_REFFRAME ref_frame_flag, YV12_BUFFER_CONFIG *sd);
 
     VP8D_PTR vp8dx_create_decompressor(VP8D_CONFIG *oxcf);
 

vp8/common/opencl/blockd_cl.c

@@ -0,0 +1,233 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "../../decoder/onyxd_int.h"
+#include "../../../vpx_ports/config.h"
+#include "../../common/idct.h"
+#include "blockd_cl.h"
+#include "../../decoder/opencl/dequantize_cl.h"
+
+
+int vp8_cl_mb_prep(MACROBLOCKD *x, int flags){
+        int err;
+
+    if (cl_initialized != CL_SUCCESS){
+        return cl_initialized;
+    }
+
+    //Copy all blockd.cl_*_mem objects
+    if (flags & DIFF)
+        VP8_CL_SET_BUF(x->cl_commands, x->cl_diff_mem, sizeof(cl_short)*400, x->diff,
+            ,err
+        );
+
+    if (flags & PREDICTOR)
+        VP8_CL_SET_BUF(x->cl_commands, x->cl_predictor_mem, sizeof(cl_uchar)*384, x->predictor,
+            ,err
+        );
+
+    if (flags & QCOEFF)
+        VP8_CL_SET_BUF(x->cl_commands, x->cl_qcoeff_mem, sizeof(cl_short)*400, x->qcoeff,
+            ,err
+        );
+
+    if (flags & DQCOEFF)
+        VP8_CL_SET_BUF(x->cl_commands, x->cl_dqcoeff_mem, sizeof(cl_short)*400, x->dqcoeff,
+            ,err
+        );
+
+    if (flags & EOBS)
+        VP8_CL_SET_BUF(x->cl_commands, x->cl_eobs_mem, sizeof(cl_char)*25, x->eobs,
+            ,err
+        );
+
+    if (flags & PRE_BUF){
+        VP8_CL_SET_BUF(x->cl_commands, x->pre.buffer_mem, x->pre.buffer_size, x->pre.buffer_alloc,
+            ,err
+        );
+    }
+
+    if (flags & DST_BUF){
+        VP8_CL_SET_BUF(x->cl_commands, x->dst.buffer_mem, x->dst.buffer_size, x->dst.buffer_alloc,
+            ,err
+        );
+    }
+
+
+    return CL_SUCCESS;
+}
+
+int vp8_cl_mb_finish(MACROBLOCKD *x, int flags){
+    int err;
+
+    if (cl_initialized != CL_SUCCESS){
+        return cl_initialized;
+    }
+
+    if (flags & DIFF){
+        err = clEnqueueReadBuffer(x->cl_commands, x->cl_diff_mem, CL_FALSE, 0, sizeof(cl_short)*400, x->diff, 0, NULL, NULL);
+        VP8_CL_CHECK_SUCCESS( x->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+        );
+    }
+
+    if (flags & PREDICTOR){
+    err = clEnqueueReadBuffer(x->cl_commands, x->cl_predictor_mem, CL_FALSE, 0, sizeof(cl_uchar)*384, x->predictor, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( x->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+    );
+    }
+
+    if (flags & QCOEFF){
+    err = clEnqueueReadBuffer(x->cl_commands, x->cl_qcoeff_mem, CL_FALSE, 0, sizeof(cl_short)*400, x->qcoeff, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( x->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+    );
+    }
+
+    if (flags & DQCOEFF){
+    err = clEnqueueReadBuffer(x->cl_commands, x->cl_dqcoeff_mem, CL_FALSE, 0, sizeof(cl_short)*400, x->dqcoeff, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( x->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+    );
+    }
+
+    if (flags & EOBS){
+        err = clEnqueueReadBuffer(x->cl_commands, x->cl_eobs_mem, CL_FALSE, 0, sizeof(cl_char)*25, x->eobs, 0, NULL, NULL);
+        VP8_CL_CHECK_SUCCESS( x->cl_commands, err != CL_SUCCESS,
+          "Error: Failed to read from GPU!\n",
+            , err
+        );
+    }
+
+    if (flags & PRE_BUF){
+        err = clEnqueueReadBuffer(x->cl_commands, x->pre.buffer_mem, CL_FALSE, 
+                0, x->pre.buffer_size, x->pre.buffer_alloc, 0, NULL, NULL);
+        VP8_CL_CHECK_SUCCESS( x->cl_commands, err != CL_SUCCESS,
+          "Error: Failed to read from GPU!\n",
+            , err
+        );
+    }
+
+    if (flags & DST_BUF){
+        err = clEnqueueReadBuffer(x->cl_commands, x->dst.buffer_mem, CL_FALSE,
+                0, x->dst.buffer_size, x->dst.buffer_alloc, 0, NULL, NULL);
+        VP8_CL_CHECK_SUCCESS( x->cl_commands, err != CL_SUCCESS,
+          "Error: Failed to read from GPU!\n",
+            , err
+        );
+    }
+
+
+    return CL_SUCCESS;
+}
+
+int vp8_cl_block_prep(BLOCKD *b, int flags){
+    int err;
+
+    if (cl_initialized != CL_SUCCESS){
+        return cl_initialized;
+    }
+
+    //Copy all blockd.cl_*_mem objects
+    if (flags & DIFF)
+        VP8_CL_SET_BUF(b->cl_commands, b->cl_diff_mem, sizeof(cl_short)*400, b->diff_base,
+            ,err
+        );
+
+    if (flags & PREDICTOR)
+        VP8_CL_SET_BUF(b->cl_commands, b->cl_predictor_mem, sizeof(cl_uchar)*384, b->predictor_base,
+            ,err
+        );
+
+    if (flags & QCOEFF)
+        VP8_CL_SET_BUF(b->cl_commands, b->cl_qcoeff_mem, sizeof(cl_short)*400, b->qcoeff_base,
+            ,err
+        );
+
+    if (flags & DQCOEFF)
+        VP8_CL_SET_BUF(b->cl_commands, b->cl_dqcoeff_mem, sizeof(cl_short)*400, b->dqcoeff_base,
+            ,err
+        );
+
+    if (flags & EOBS)
+        VP8_CL_SET_BUF(b->cl_commands, b->cl_eobs_mem, sizeof(cl_char)*25, b->eobs_base,
+            ,err
+        );
+
+    if (flags & DEQUANT)
+        VP8_CL_SET_BUF(b->cl_commands, b->cl_dequant_mem, sizeof(cl_short)*16 ,b->dequant,
+            ,err
+        );
+
+    return CL_SUCCESS;
+}
+
+int vp8_cl_block_finish(BLOCKD *b, int flags){
+    int err;
+
+    if (cl_initialized != CL_SUCCESS){
+        return cl_initialized;
+    }
+
+    if (flags & DIFF){
+        err = clEnqueueReadBuffer(b->cl_commands, b->cl_diff_mem, CL_FALSE, 0, sizeof(cl_short)*400, b->diff_base, 0, NULL, NULL);
+        VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+        );
+    }
+
+    if (flags & PREDICTOR){
+    err = clEnqueueReadBuffer(b->cl_commands, b->cl_predictor_mem, CL_FALSE, 0, sizeof(cl_uchar)*384, b->predictor_base, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+    );
+    }
+
+    if (flags & QCOEFF){
+    err = clEnqueueReadBuffer(b->cl_commands, b->cl_qcoeff_mem, CL_FALSE, 0, sizeof(cl_short)*400, b->qcoeff_base, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+    );
+    }
+
+    if (flags & DQCOEFF){
+    err = clEnqueueReadBuffer(b->cl_commands, b->cl_dqcoeff_mem, CL_FALSE, 0, sizeof(cl_short)*400, b->dqcoeff_base, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+    );
+    }
+
+    if (flags & EOBS){
+    err = clEnqueueReadBuffer(b->cl_commands, b->cl_eobs_mem, CL_FALSE, 0, sizeof(cl_char)*25, b->eobs_base, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+    );
+    }
+
+    if (flags & DEQUANT){
+    err = clEnqueueReadBuffer(b->cl_commands, b->cl_dequant_mem, CL_FALSE, 0, sizeof(cl_short)*16 ,b->dequant, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read from GPU!\n",
+            , err
+    );
+    }
+
+    return CL_SUCCESS;
+}

vp8/common/opencl/blockd_cl.h

@@ -0,0 +1,64 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef BLOCKD_OPENCL_H
+#define BLOCKD_OPENCL_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#include "vp8_opencl.h"
+#include "../blockd.h"
+
+#define DIFF 0x0001
+#define PREDICTOR 0x0002
+#define QCOEFF 0x0004
+#define DQCOEFF 0x0008
+#define EOBS 0x0010
+#define DEQUANT 0x0020
+#define PRE_BUF 0x0040
+#define DST_BUF 0x0080
+    
+#define BLOCK_COPY_ALL 0xffff
+
+/*
+#define BLOCK_MEM_SIZE 6
+enum {
+    DIFF_MEM = 0,
+    PRED_MEM = 1,
+    QCOEFF_MEM = 2,
+    DQCOEFF_MEM = 3,
+    EOBS_MEM = 4,
+    DEQUANT_MEM = 5
+} BLOCK_MEM_TYPES;
+
+
+struct cl_block_mem{
+    cl_mem gpu_mem;
+    size_t size;
+    void *host_mem;
+};
+
+typedef struct cl_block_mem block_mem;
+*/
+    
+extern int vp8_cl_block_finish(BLOCKD *b, int flags);
+extern int vp8_cl_block_prep(BLOCKD *b, int flags);
+
+extern int vp8_cl_mb_prep(MACROBLOCKD *x, int flags);
+extern int vp8_cl_mb_finish(MACROBLOCKD *x, int flags);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif

vp8/common/opencl/dynamic_cl.c

@@ -0,0 +1,106 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vp8_opencl.h"
+
+#include <stdio.h>
+
+CL_FUNCTIONS cl;
+void *dll = NULL;
+int cl_loaded = VP8_CL_NOT_INITIALIZED;
+
+int close_cl(){
+    int ret = dlclose(dll);
+
+    if (ret != 0)
+        fprintf(stderr, "Error closing OpenCL library: %s", dlerror());
+
+    return ret;
+}
+
+int load_cl(char *lib_name){
+
+    //printf("Loading OpenCL library\n");
+    dll = dlopen(lib_name, RTLD_NOW|RTLD_LOCAL);
+    if (dll != NULL){
+        //printf("Found CL library\n");
+    } else {
+        //printf("Didn't find CL library\n");
+        return VP8_CL_TRIED_BUT_FAILED;
+    }
+
+    CL_LOAD_FN("clGetPlatformIDs", cl.getPlatformIDs);
+    CL_LOAD_FN("clGetPlatformInfo", cl.getPlatformInfo);
+    CL_LOAD_FN("clGetDeviceIDs", cl.getDeviceIDs);
+    CL_LOAD_FN("clGetDeviceInfo", cl.getDeviceInfo);
+    CL_LOAD_FN("clCreateContext", cl.createContext);
+//    CL_LOAD_FN("clCreateContextFromType", cl.createContextFromType);
+//    CL_LOAD_FN("clRetainContext", cl.retainContext);
+    CL_LOAD_FN("clReleaseContext", cl.releaseContext);
+//    CL_LOAD_FN("clGetContextInfo", cl.getContextInfo);
+    CL_LOAD_FN("clCreateCommandQueue", cl.createCommandQueue);
+//    CL_LOAD_FN("clRetainCommandQueue", cl.retainCommandQueue);
+    CL_LOAD_FN("clReleaseCommandQueue", cl.releaseCommandQueue);
+//    CL_LOAD_FN("clGetCommandQueueInfo", cl.getCommandQueue);
+    CL_LOAD_FN("clCreateBuffer", cl.createBuffer);
+//    CL_LOAD_FN("clCreateImage2D", cl.createImage2D);
+//    CL_LOAD_FN("clCreateImage3D", cl.createImage3D);
+//    CL_LOAD_FN("clRetainMemObject", cl.retainMemObject);
+    CL_LOAD_FN("clReleaseMemObject", cl.releaseMemObject);
+//    CL_LOAD_FN("clGetSupportedImageFormats", cl.getSupportedImageFormats);
+//    CL_LOAD_FN("clGetMemObjectInfo", cl.getMemObjectInfo);
+//    CL_LOAD_FN("clGetImageInfo", cl.getImageInfo);
+//    CL_LOAD_FN("clCreateSampler", cl.createSampler);
+//    CL_LOAD_FN("clRetainSampler", cl.retainSampler);
+//    CL_LOAD_FN("clReleaseSampler", cl.releaseSampler);
+//    CL_LOAD_FN("clGetSamplerInfo", cl.getSamplerInfo);
+    CL_LOAD_FN("clCreateProgramWithSource", cl.createProgramWithSource);
+//    CL_LOAD_FN("clCreateProgramWithBinary", cl.createProgramWithBinary);
+//    CL_LOAD_FN("clRetainProgram", cl.retainProgram);
+    CL_LOAD_FN("clReleaseProgram", cl.releaseProgram);
+    CL_LOAD_FN("clBuildProgram", cl.buildProgram);
+//    CL_LOAD_FN("clUnloadCompiler", cl.unloadCompiler);
+    CL_LOAD_FN("clGetProgramInfo", cl.getProgramInfo);
+    CL_LOAD_FN("clGetProgramBuildInfo", cl.getProgramBuildInfo);
+    CL_LOAD_FN("clCreateKernel", cl.createKernel);
+//    CL_LOAD_FN("clCreateKernelsInProgram", cl.createKernelsInProgram);
+//    CL_LOAD_FN("clRetainKernel", cl.retainKernel);
+    CL_LOAD_FN("clReleaseKernel", cl.releaseKernel);
+    CL_LOAD_FN("clSetKernelArg", cl.setKernelArg);
+//    CL_LOAD_FN("clGetKernelInfo", cl.getKernelInfo);
+    CL_LOAD_FN("clGetKernelWorkGroupInfo", cl.getKernelWorkGroupInfo);
+//    CL_LOAD_FN("clWaitForEvents", cl.waitForEvents);
+//    CL_LOAD_FN("clGetEventInfo", cl.getEventInfo);
+//    CL_LOAD_FN("clRetainEvent", cl.retainEvent);
+//    CL_LOAD_FN("clReleaseEvent", cl.releaseEvent);
+//    CL_LOAD_FN("clGetEventProfilingInfo", cl.getEventProfilingInfo);
+    CL_LOAD_FN("clFlush", cl.flush);
+    CL_LOAD_FN("clFinish", cl.finish);
+    CL_LOAD_FN("clEnqueueReadBuffer", cl.enqueueReadBuffer);
+    CL_LOAD_FN("clEnqueueWriteBuffer", cl.enqueueWriteBuffer);
+    CL_LOAD_FN("clEnqueueCopyBuffer", cl.enqueueCopyBuffer);
+//    CL_LOAD_FN("clEnqueueReadImage", cl.enqueueReadImage);
+//    CL_LOAD_FN("clEnqueueWriteImage", cl.enqueueWriteImage);
+//    CL_LOAD_FN("clEnqueueCopyImage", cl.enqueueCopyImage);
+//    CL_LOAD_FN("clEnqueueCopyImageToBuffer", cl.enqueueCopyImageToBuffer);
+//    CL_LOAD_FN("clEnqueueCopyBufferToImage", cl.enqueueCopyBufferToImage);
+//    CL_LOAD_FN("clEnqueueMapBuffer", cl.enqueueMapBuffer);
+//    CL_LOAD_FN("clEnqueueMapImage", cl.enqueueMapImage);
+//    CL_LOAD_FN("clEnqueueUnmapMemObject", cl.enqueueUnmapMemObject);
+    CL_LOAD_FN("clEnqueueNDRangeKernel", cl.enqueueNDRAngeKernel);
+//    CL_LOAD_FN("clEnqueueTask", cl.enqueueTask);
+//    CL_LOAD_FN("clEnqueueNativeKernel", cl.enqueueNativeKernel);
+//    CL_LOAD_FN("clEnqueueMarker", cl.enqueueMarker);
+//    CL_LOAD_FN("clEnqueueWaitForEvents", cl.enqueueWaitForEvents);
+    CL_LOAD_FN("clEnqueueBarrier", cl.enqueueBarrier);
+//    CL_LOAD_FN("clGetExtensionFunctionAddress", cl.getExtensionFunctionAddress);
+
+    return CL_SUCCESS;
+}

vp8/common/opencl/dynamic_cl.h

@@ -0,0 +1,253 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef DYNAMIC_CL_H
+#define	DYNAMIC_CL_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#ifdef __APPLE__
+#include <OpenCL/cl.h>
+#else
+#include <CL/cl.h>
+#endif
+    
+#include <dlfcn.h>
+
+int load_cl(char *lib_name);
+int close_cl();
+
+extern int cl_loaded;
+
+typedef cl_int(*fn_clGetPlatformIDs_t)(cl_uint, cl_platform_id *, cl_uint *);
+typedef cl_int(*fn_clGetPlatformInfo_t)(cl_platform_id, cl_platform_info, size_t, void *, size_t *);
+typedef cl_int(*fn_clGetDeviceIDs_t)(cl_platform_id, cl_device_type, cl_uint, cl_device_id *, cl_uint *);
+typedef cl_int(*fn_clGetDeviceInfo_t)(cl_device_id, cl_device_info, size_t, void *, size_t *);
+typedef cl_context(*fn_clCreateContext_t)(const cl_context_properties *, cl_uint, const cl_device_id *, void (*pfn_notify)(const char *, const void *, size_t, void *), void *, cl_int *);
+typedef cl_context(*fn_clCreateContextFromType_t)(const cl_context_properties *, cl_device_type, void (*pfn_notify)(const char *, const void *, size_t, void *), void *, cl_int *);
+typedef cl_int(*fn_clRetainContext_t)(cl_context);
+typedef cl_int(*fn_clReleaseContext_t)(cl_context);
+typedef cl_int(*fn_clGetContextInfo_t)(cl_context, cl_context_info, size_t, void *, size_t *);
+typedef cl_command_queue(*fn_clCreateCommandQueue_t)(cl_context, cl_device_id, cl_command_queue_properties, cl_int *);
+typedef cl_int(*fn_clRetainCommandQueue_t)(cl_command_queue);
+typedef cl_int(*fn_clReleaseCommandQueue_t)(cl_command_queue);
+typedef cl_int(*fn_clGetCommandQueueInfo_t)(cl_command_queue, cl_command_queue_info, size_t, void *, size_t *);
+typedef cl_mem(*fn_clCreateBuffer_t)(cl_context, cl_mem_flags, size_t, void *, cl_int *);
+typedef cl_mem(*fn_clCreateImage2D_t)(cl_context, cl_mem_flags, const cl_image_format *, size_t, size_t, size_t, void *, cl_int *);
+typedef cl_mem(*fn_clCreateImage3D_t)(cl_context, cl_mem_flags, const cl_image_format *, size_t, size_t, size_t, size_t, size_t, void *, cl_int *);
+typedef cl_int(*fn_clRetainMemObject_t)(cl_mem);
+typedef cl_int(*fn_clReleaseMemObject_t)(cl_mem);
+typedef cl_int(*fn_clGetSupportedImageFormats_t)(cl_context, cl_mem_flags, cl_mem_object_type, cl_uint, cl_image_format *, cl_uint *);
+typedef cl_int(*fn_clGetMemObjectInfo_t)(cl_mem, cl_mem_info, size_t, void *, size_t *);
+typedef cl_int(*fn_clGetImageInfo_t)(cl_mem, cl_image_info, size_t, void *, size_t *);
+typedef cl_sampler(*fn_clCreateSampler_t)(cl_context, cl_bool, cl_addressing_mode, cl_filter_mode, cl_int *);
+typedef cl_int(*fn_clRetainSampler_t)(cl_sampler);
+typedef cl_int(*fn_clReleaseSampler_t)(cl_sampler);
+typedef cl_int(*fn_clGetSamplerInfo_t)(cl_sampler, cl_sampler_info, size_t, void *, size_t *);
+typedef cl_program(*fn_clCreateProgramWithSource_t)(cl_context, cl_uint, const char **, const size_t *, cl_int *);
+typedef cl_program(*fn_clCreateProgramWithBinary_t)(cl_context, cl_uint, const cl_device_id *, const size_t *, const unsigned char **, cl_int *, cl_int *);
+typedef cl_int(*fn_clRetainProgram_t)(cl_program);
+typedef cl_int(*fn_clReleaseProgram_t)(cl_program);
+typedef cl_int(*fn_clBuildProgram_t)(cl_program, cl_uint, const cl_device_id *, const char *,  void (*pfn_notify)(cl_program,void*), void *);
+typedef cl_int(*fn_clUnloadCompiler_t)(void);
+typedef cl_int(*fn_clGetProgramInfo_t)(cl_program, cl_program_info, size_t, void *, size_t *);
+typedef cl_int(*fn_clGetProgramBuildInfo_t)(cl_program, cl_device_id, cl_program_build_info, size_t, void *, size_t *);
+typedef cl_kernel(*fn_clCreateKernel_t)(cl_program, const char *, cl_int *);
+typedef cl_int(*fn_clCreateKernelsInProgram_t)(cl_program, cl_uint, cl_kernel *, cl_uint *);
+typedef cl_int(*fn_clRetainKernel_t)(cl_kernel);
+typedef cl_int(*fn_clReleaseKernel_t)(cl_kernel);
+typedef cl_int(*fn_clSetKernelArg_t)(cl_kernel, cl_uint, size_t, const void *);
+typedef cl_int(*fn_clGetKernelInfo_t)(cl_kernel, cl_kernel_info, size_t, void *, size_t *);
+typedef cl_int(*fn_clGetKernelWorkGroupInfo_t)(cl_kernel, cl_device_id, cl_kernel_work_group_info, size_t, void *, size_t *);
+typedef cl_int(*fn_clWaitForEvents_t)(cl_uint, const cl_event *);
+typedef cl_int(*fn_clGetEventInfo_t)(cl_event, cl_event_info, size_t, void *, size_t *);
+typedef cl_int(*fn_clRetainEvent_t)(cl_event);
+typedef cl_int(*fn_clReleaseEvent_t)(cl_event);
+typedef cl_int(*fn_clGetEventProfilingInfo_t)(cl_event, cl_profiling_info, size_t, void *, size_t *);
+typedef cl_int(*fn_clFlush_t)(cl_command_queue);
+typedef cl_int(*fn_clFinish_t)(cl_command_queue);
+typedef cl_int(*fn_clEnqueueReadBuffer_t)(cl_command_queue, cl_mem, cl_bool, size_t, size_t, void *, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueWriteBuffer_t)(cl_command_queue,  cl_mem,  cl_bool,  size_t,  size_t,  const void *,  cl_uint,  const cl_event *,  cl_event *);
+typedef cl_int(*fn_clEnqueueCopyBuffer_t)(cl_command_queue,  cl_mem, cl_mem, size_t, size_t, size_t, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueReadImage_t)(cl_command_queue, cl_mem, cl_bool, const size_t *, const size_t *, size_t, size_t, void *, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueWriteImage_t)(cl_command_queue, cl_mem, cl_bool, const size_t *, const size_t *, size_t, size_t, const void *, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueCopyImage_t)(cl_command_queue, cl_mem, cl_mem, const size_t *, const size_t *, const size_t *, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueCopyImageToBuffer_t)(cl_command_queue, cl_mem, cl_mem, const size_t *, const size_t *, size_t, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueCopyBufferToImage_t)(cl_command_queue, cl_mem, cl_mem, size_t, const size_t *, const size_t *, cl_uint, const cl_event *, cl_event *);
+typedef void*(*fn_clEnqueueMapBuffer_t)(cl_command_queue, cl_mem, cl_bool, cl_map_flags, size_t, size_t, cl_uint, const cl_event *, cl_event *, cl_int *);
+typedef void*(*fn_clEnqueueMapImage_t)(cl_command_queue, cl_mem, cl_bool, cl_map_flags, const size_t *, const size_t *, size_t *, size_t *, cl_uint, const cl_event *, cl_event *, cl_int *);
+typedef cl_int(*fn_clEnqueueUnmapMemObject_t)(cl_command_queue, cl_mem, void *, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueNDRangeKernel_t)(cl_command_queue, cl_kernel, cl_uint, const size_t *, const size_t *, const size_t *, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueTask_t)(cl_command_queue, cl_kernel, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueNativeKernel_t)(cl_command_queue,					 void (*user_func)(void *), void *, size_t, cl_uint, const cl_mem *, const void **, cl_uint, const cl_event *, cl_event *);
+typedef cl_int(*fn_clEnqueueMarker_t)(cl_command_queue, cl_event *);
+typedef cl_int(*fn_clEnqueueWaitForEvents_t)(cl_command_queue, cl_uint, const cl_event *);
+typedef cl_int(*fn_clEnqueueBarrier_t)(cl_command_queue);
+typedef void*(*fn_clGetExtensionFunctionAddress_t)(const char *);
+
+typedef struct CL_FUNCTIONS {
+    fn_clGetPlatformIDs_t getPlatformIDs;
+    fn_clGetPlatformInfo_t getPlatformInfo;
+    fn_clGetDeviceIDs_t getDeviceIDs;
+    fn_clGetDeviceInfo_t getDeviceInfo;
+    fn_clCreateContext_t createContext;
+    fn_clCreateContextFromType_t createContextFromType;
+    fn_clRetainContext_t retainContext;
+    fn_clReleaseContext_t releaseContext;
+    fn_clGetContextInfo_t getContextInfo;
+    fn_clCreateCommandQueue_t createCommandQueue;
+    fn_clRetainCommandQueue_t retainCommandQueue;
+    fn_clReleaseCommandQueue_t releaseCommandQueue;
+    fn_clGetCommandQueueInfo_t getCommandQueue;
+    fn_clCreateBuffer_t createBuffer;
+    fn_clCreateImage2D_t createImage2D;
+    fn_clCreateImage3D_t createImage3D;
+    fn_clRetainMemObject_t retainMemObject;
+    fn_clReleaseMemObject_t releaseMemObject;
+    fn_clGetSupportedImageFormats_t getSupportedImageFormats;
+    fn_clGetMemObjectInfo_t getMemObjectInfo;
+    fn_clGetImageInfo_t getImageInfo;
+    fn_clCreateSampler_t createSampler;
+    fn_clRetainSampler_t retainSampler;
+    fn_clReleaseSampler_t releaseSampler;
+    fn_clGetSamplerInfo_t getSamplerInfo;
+    fn_clCreateProgramWithSource_t createProgramWithSource;
+    fn_clCreateProgramWithBinary_t createProgramWithBinary;
+    fn_clRetainProgram_t retainProgram;
+    fn_clReleaseProgram_t releaseProgram;
+    fn_clBuildProgram_t buildProgram;
+    fn_clUnloadCompiler_t unloadCompiler;
+    fn_clGetProgramInfo_t getProgramInfo;
+    fn_clGetProgramBuildInfo_t getProgramBuildInfo;
+    fn_clCreateKernel_t createKernel;
+    fn_clCreateKernelsInProgram_t createKernelsInProgram;
+    fn_clRetainKernel_t retainKernel;
+    fn_clReleaseKernel_t releaseKernel;
+    fn_clSetKernelArg_t setKernelArg;
+    fn_clGetKernelInfo_t getKernelInfo;
+    fn_clGetKernelWorkGroupInfo_t getKernelWorkGroupInfo;
+    fn_clWaitForEvents_t waitForEvents;
+    fn_clGetEventInfo_t getEventInfo;
+    fn_clRetainEvent_t retainEvent;
+    fn_clReleaseEvent_t releaseEvent;
+    fn_clGetEventProfilingInfo_t getEventProfilingInfo;
+    fn_clFlush_t flush;
+    fn_clFinish_t finish;
+    fn_clEnqueueReadBuffer_t enqueueReadBuffer;
+    fn_clEnqueueWriteBuffer_t enqueueWriteBuffer;
+    fn_clEnqueueCopyBuffer_t enqueueCopyBuffer;
+    fn_clEnqueueReadImage_t enqueueReadImage;
+    fn_clEnqueueWriteImage_t enqueueWriteImage;
+    fn_clEnqueueCopyImage_t enqueueCopyImage;
+    fn_clEnqueueCopyImageToBuffer_t enqueueCopyImageToBuffer;
+    fn_clEnqueueCopyBufferToImage_t enqueueCopyBufferToImage;
+    fn_clEnqueueMapBuffer_t enqueueMapBuffer;
+    fn_clEnqueueMapImage_t enqueueMapImage;
+    fn_clEnqueueUnmapMemObject_t enqueueUnmapMemObject;
+    fn_clEnqueueNDRangeKernel_t enqueueNDRAngeKernel;
+    fn_clEnqueueTask_t enqueueTask;
+    fn_clEnqueueNativeKernel_t enqueueNativeKernel;
+    fn_clEnqueueMarker_t enqueueMarker;
+    fn_clEnqueueWaitForEvents_t enqueueWaitForEvents;
+    fn_clEnqueueBarrier_t enqueueBarrier;
+    fn_clGetExtensionFunctionAddress_t getExtensionFunctionAddress;
+} CL_FUNCTIONS;
+
+extern CL_FUNCTIONS cl;
+
+#define clGetPlatformIDs cl.getPlatformIDs
+#define clGetPlatformInfo cl.getPlatformInfo
+#define clGetDeviceIDs cl.getDeviceIDs
+#define clGetDeviceInfo cl.getDeviceInfo
+#define clCreateContext cl.createContext
+#define clCreateContextFromType cl.createContextFromType
+#define clRetainContext cl.retainContext
+#define clReleaseContext cl.releaseContext
+#define clGetContextInfo cl.getContextInfo
+#define clCreateCommandQueue cl.createCommandQueue
+#define clRetainCommandQueue cl.retainCommandQueue
+#define clReleaseCommandQueue cl.releaseCommandQueue
+#define clGetCommandQueueInfo cl.getCommandQueue
+#define clCreateBuffer cl.createBuffer
+#define clCreateSubBuffer cl.createSubBuffer
+#define clCreateImage2D cl.createImage2D
+#define clCreateImage3D cl.createImage3D
+#define clRetainMemObject cl.retainMemObject
+#define clReleaseMemObject cl.releaseMemObject
+#define clGetSupportedImageFormats cl.getSupportedImageFormats
+#define clGetMemObjectInfo cl.getMemObjectInfo
+#define clGetImageInfo cl.getImageInfo
+#define clSetMemObjectDestructorCallback cl.setMemObjectDestructorCallback
+#define clCreateSampler cl.createSampler
+#define clRetainSampler cl.retainSampler
+#define clReleaseSampler cl.releaseSampler
+#define clGetSamplerInfo cl.getSamplerInfo
+#define clCreateProgramWithSource cl.createProgramWithSource
+#define clCreateProgramWithBinary cl.createProgramWithBinary
+#define clRetainProgram cl.retainProgram
+#define clReleaseProgram cl.releaseProgram
+#define clBuildProgram cl.buildProgram
+#define clUnloadCompiler cl.unloadCompiler
+#define clGetProgramInfo cl.getProgramInfo
+#define clGetProgramBuildInfo cl.getProgramBuildInfo
+#define clCreateKernel cl.createKernel
+#define clCreateKernelsInProgram cl.createKernelsInProgram
+#define clRetainKernel cl.retainKernel
+#define clReleaseKernel cl.releaseKernel
+#define clSetKernelArg cl.setKernelArg
+#define clGetKernelInfo cl.getKernelInfo
+#define clGetKernelWorkGroupInfo cl.getKernelWorkGroupInfo
+#define clWaitForEvents cl.waitForEvents
+#define clGetEventInfo cl.getEventInfo
+#define clCreateUserEvent cl.createUserEvent
+#define clRetainEvent cl.retainEvent
+#define clReleaseEvent cl.releaseEvent
+#define clSetUserEventStatus cl.setUserEventStatus
+#define clSetEventCallback cl.setEventCallback
+#define clGetEventProfilingInfo cl.getEventProfilingInfo
+#define clFlush cl.flush
+#define clFinish cl.finish
+#define clEnqueueReadBuffer cl.enqueueReadBuffer
+#define clEnqueueReadBufferRect cl.enqueueReadBufferRect
+#define clEnqueueWriteBuffer cl.enqueueWriteBuffer
+#define clEnqueueWriteBufferRect cl.enqueueWriteBufferRect
+#define clEnqueueCopyBuffer cl.enqueueCopyBuffer
+#define clEnqueueCopyBufferRect cl.enqueueCopyBufferRect
+#define clEnqueueReadImage cl.enqueueReadImage
+#define clEnqueueWriteImage cl.enqueueWriteImage
+#define clEnqueueCopyImage cl.enqueueCopyImage
+#define clEnqueueCopyImageToBuffer cl.enqueueCopyImageToBuffer
+#define clEnqueueCopyBufferToImage cl.enqueueCopyBufferToImage
+#define clEnqueueMapBuffer cl.enqueueMapBuffer
+#define clEnqueueMapImage cl.enqueueMapImage
+#define clEnqueueUnmapMemObject cl.enqueueUnmapMemObject
+#define clEnqueueNDRangeKernel cl.enqueueNDRAngeKernel
+#define clEnqueueTask cl.enqueueTask
+#define clEnqueueNativeKernel cl.enqueueNativeKernel
+#define clEnqueueMarker cl.enqueueMarker
+#define clEnqueueWaitForEvents cl.enqueueWaitForEvents
+#define clEnqueueBarrier cl.enqueueBarrier
+#define clGetExtensionFunctionAddress cl.getExtensionFunctionAddress
+
+#define CL_LOAD_FN(name, ref) \
+    ref = dlsym(dll,name); \
+    if (ref == NULL){ \
+        dlclose(dll); \
+        return CL_INVALID_PLATFORM; \
+    }
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* DYNAMIC_CL_H */

vp8/common/opencl/filter_cl.c

@@ -0,0 +1,824 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdlib.h>
+
+//ACW: Remove me after debugging.
+#include <stdio.h>
+#include <string.h>
+
+#include "vp8_opencl.h"
+#include "filter_cl.h"
+#include "../blockd.h"
+
+#define SIXTAP_FILTER_LEN 6
+
+const char *filterCompileOptions = "-Ivp8/common/opencl -DVP8_FILTER_WEIGHT=128 -DVP8_FILTER_SHIFT=7 -DFILTER_OFFSET";
+const char *filter_cl_file_name = "vp8/common/opencl/filter_cl.cl";
+
+#define STATIC_MEM 1
+#if STATIC_MEM
+static cl_mem int_mem = NULL;
+#endif
+
+void cl_destroy_filter(){
+
+    if (cl_data.filter_program)
+        clReleaseProgram(cl_data.filter_program);
+
+    //VP8_CL_RELEASE_KERNEL(cl_data.vp8_block_variation_kernel);
+#if !TWO_PASS_SIXTAP
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_sixtap_predict_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_sixtap_predict8x8_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_sixtap_predict8x4_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_sixtap_predict16x16_kernel);
+#else
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_filter_block2d_first_pass_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_filter_block2d_second_pass_kernel);
+#endif
+    //VP8_CL_RELEASE_KERNEL(cl_data.vp8_bilinear_predict4x4_kernel);
+    //VP8_CL_RELEASE_KERNEL(cl_data.vp8_bilinear_predict8x4_kernel);
+    //VP8_CL_RELEASE_KERNEL(cl_data.vp8_bilinear_predict8x8_kernel);
+    //VP8_CL_RELEASE_KERNEL(cl_data.vp8_bilinear_predict16x16_kernel);
+
+#if MEM_COPY_KERNEL
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_memcpy_kernel);
+#endif
+
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_filter_block2d_bil_first_pass_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_filter_block2d_bil_second_pass_kernel);
+
+#if STATIC_MEM
+    if (int_mem != NULL)
+        clReleaseMemObject(int_mem);
+    int_mem = NULL;
+#endif
+
+    cl_data.filter_program = NULL;
+}
+
+int cl_init_filter() {
+    int err;
+
+
+    // Create the filter compute program from the file-defined source code
+    if ( cl_load_program(&cl_data.filter_program, filter_cl_file_name,
+            filterCompileOptions) != CL_SUCCESS )
+        return VP8_CL_TRIED_BUT_FAILED;
+
+    // Create the compute kernel in the program we wish to run
+#if TWO_PASS_SIXTAP
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_filter_block2d_first_pass_kernel,"vp8_filter_block2d_first_pass_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_filter_block2d_second_pass_kernel,"vp8_filter_block2d_second_pass_kernel");
+    VP8_CL_CALC_LOCAL_SIZE(vp8_filter_block2d_first_pass_kernel,vp8_filter_block2d_first_pass_kernel_size);
+    VP8_CL_CALC_LOCAL_SIZE(vp8_filter_block2d_second_pass_kernel,vp8_filter_block2d_second_pass_kernel_size);
+#else
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_sixtap_predict_kernel,"vp8_sixtap_predict_kernel");
+    VP8_CL_CALC_LOCAL_SIZE(vp8_sixtap_predict_kernel,vp8_sixtap_predict_kernel_size);
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_sixtap_predict8x8_kernel,"vp8_sixtap_predict8x8_kernel");
+    VP8_CL_CALC_LOCAL_SIZE(vp8_sixtap_predict8x8_kernel,vp8_sixtap_predict8x8_kernel_size);
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_sixtap_predict8x4_kernel,"vp8_sixtap_predict8x4_kernel");
+    VP8_CL_CALC_LOCAL_SIZE(vp8_sixtap_predict8x4_kernel,vp8_sixtap_predict8x4_kernel_size);
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_sixtap_predict16x16_kernel,"vp8_sixtap_predict16x16_kernel");
+    VP8_CL_CALC_LOCAL_SIZE(vp8_sixtap_predict16x16_kernel,vp8_sixtap_predict16x16_kernel_size);
+#endif
+    
+    //VP8_CL_CALC_LOCAL_SIZE(vp8_filter_block2d_bil_first_pass_kernel,vp8_filter_block2d_bil_first_pass_kernel_size);
+    //VP8_CL_CALC_LOCAL_SIZE(vp8_filter_block2d_bil_second_pass_kernel,vp8_filter_block2d_bil_second_pass_kernel_size);
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_filter_block2d_bil_first_pass_kernel,"vp8_filter_block2d_bil_first_pass_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_filter_block2d_bil_second_pass_kernel,"vp8_filter_block2d_bil_second_pass_kernel");
+
+
+    //VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_bilinear_predict4x4_kernel,"vp8_bilinear_predict4x4_kernel");
+    //VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_bilinear_predict8x4_kernel,"vp8_bilinear_predict8x4_kernel");
+    //VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_bilinear_predict8x8_kernel,"vp8_bilinear_predict8x8_kernel");
+    //VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_bilinear_predict16x16_kernel,"vp8_bilinear_predict16x16_kernel");
+
+#if MEM_COPY_KERNEL
+    VP8_CL_CREATE_KERNEL(cl_data,filter_program,vp8_memcpy_kernel,"vp8_memcpy_kernel");
+    VP8_CL_CALC_LOCAL_SIZE(vp8_memcpy_kernel,vp8_memcpy_kernel_size);
+#endif
+
+#if STATIC_MEM
+    VP8_CL_CREATE_BUF(NULL, int_mem, NULL, sizeof(cl_int)*21*16, NULL, ,err);
+#endif
+
+    return CL_SUCCESS;
+}
+
+void vp8_filter_block2d_first_pass_cl(
+    cl_command_queue cq,
+    cl_mem src_mem,
+    int src_offset,
+    cl_mem int_mem,
+    unsigned int src_pixels_per_line,
+    unsigned int int_height,
+    unsigned int int_width,
+    int xoffset
+){
+    int err;
+    size_t global = int_width*int_height;
+    size_t local = cl_data.vp8_filter_block2d_first_pass_kernel_size;
+    if (local > global)
+        local = global;
+
+    err =  clSetKernelArg(cl_data.vp8_filter_block2d_first_pass_kernel, 0, sizeof (cl_mem), &src_mem);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_first_pass_kernel, 1, sizeof (int), &src_offset);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_first_pass_kernel, 2, sizeof (cl_mem), &int_mem);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_first_pass_kernel, 3, sizeof (cl_uint), &src_pixels_per_line);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_first_pass_kernel, 4, sizeof (cl_uint), &int_height);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_first_pass_kernel, 5, sizeof (cl_int), &int_width);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_first_pass_kernel, 6, sizeof (int), &xoffset);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        ,
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel( cq, cl_data.vp8_filter_block2d_first_pass_kernel, 1, NULL, &global, &local , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);,
+    );
+}
+
+void vp8_filter_block2d_second_pass_cl(
+    cl_command_queue cq,
+    cl_mem int_mem,
+    int int_offset,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch,
+    unsigned int output_height,
+    unsigned int output_width,
+    int yoffset
+){
+    int err;
+    size_t global = output_width*output_height;
+    size_t local = cl_data.vp8_filter_block2d_second_pass_kernel_size;
+    if (local > global){
+        //printf("Local is now %ld\n",global);
+        local = global;
+    }
+
+    /* Set kernel arguments */
+    err =  clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 0, sizeof (cl_mem), &int_mem);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 1, sizeof (int), &int_offset);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 2, sizeof (cl_mem), &dst_mem);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 3, sizeof (int), &dst_offset);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 4, sizeof (int), &dst_pitch);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 5, sizeof (int), &output_width);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 6, sizeof (int), &output_width);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 7, sizeof (int), &output_height);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 8, sizeof (int), &output_width);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_second_pass_kernel, 9, sizeof (int), &yoffset);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        ,
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel( cq, cl_data.vp8_filter_block2d_second_pass_kernel, 1, NULL, &global, &local , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);,
+    );
+}
+
+void vp8_sixtap_single_pass(
+    cl_command_queue cq,
+    cl_kernel kernel,
+    size_t local,
+    size_t global,
+    cl_mem src_mem,
+    cl_mem dst_mem,
+    unsigned char *src_base,
+    int src_offset,
+    size_t src_len,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    int dst_offset,
+    int dst_pitch,
+    size_t dst_len
+){
+    int err;
+
+#if !STATIC_MEM
+    cl_mem int_mem;
+#endif
+
+    int free_src = 0, free_dst = 0;
+
+    if (local > global){
+        local = global;
+    }
+
+    /* Make space for kernel input/output data.
+     * Initialize the buffer as well if needed.
+     */
+    if (src_mem == NULL){
+        VP8_CL_CREATE_BUF( cq, src_mem,, sizeof (unsigned char) * src_len, src_base-2,,);
+        src_offset = 2;
+        free_src = 1;
+    } else {
+        src_offset -= 2*src_pixels_per_line;
+    }
+
+    if (dst_mem == NULL){
+        VP8_CL_CREATE_BUF( cq, dst_mem,, sizeof (unsigned char) * dst_len + dst_offset, dst_base,, );
+        free_dst = 1;
+    }
+
+#if !STATIC_MEM
+    CL_CREATE_BUF( cq, int_mem,, sizeof(cl_int)*FData_height*FData_width, NULL,, );
+#endif
+
+    err =  clSetKernelArg(kernel, 0, sizeof (cl_mem), &src_mem);
+    err |= clSetKernelArg(kernel, 1, sizeof (int), &src_offset);
+    err |= clSetKernelArg(kernel, 2, sizeof (cl_int), &src_pixels_per_line);
+    err |= clSetKernelArg(kernel, 3, sizeof (cl_int), &xoffset);
+    err |= clSetKernelArg(kernel, 4, sizeof (cl_int), &yoffset);
+    err |= clSetKernelArg(kernel, 5, sizeof (cl_mem), &dst_mem);
+    err |= clSetKernelArg(kernel, 6, sizeof (cl_int), &dst_offset);
+    err |= clSetKernelArg(kernel, 7, sizeof (int), &dst_pitch);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        ,
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel( cq, kernel, 1, NULL, &global, &local , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);,
+    );
+
+    if (free_src == 1)
+        clReleaseMemObject(src_mem);
+
+    if (free_dst == 1){
+        /* Read back the result data from the device */
+        err = clEnqueueReadBuffer(cq, dst_mem, CL_FALSE, 0, sizeof (unsigned char) * dst_len + dst_offset, dst_base, 0, NULL, NULL);
+        VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+            "Error: Failed to read output array!\n",
+            ,
+        );
+        clReleaseMemObject(dst_mem);
+    }
+}
+
+void vp8_sixtap_run_cl(
+    cl_command_queue cq,
+    cl_mem src_mem,
+    cl_mem dst_mem,
+    unsigned char *src_base,
+    int src_offset,
+    size_t src_len,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    int dst_offset,
+    int dst_pitch,
+    size_t dst_len,
+    unsigned int FData_height,
+    unsigned int FData_width,
+    unsigned int output_height,
+    unsigned int output_width,
+    int int_offset
+)
+{
+    int err;
+
+#if !STATIC_MEM
+    cl_mem int_mem;
+#endif
+
+    int free_src = 0, free_dst = 0;
+
+    /* Make space for kernel input/output data.
+     * Initialize the buffer as well if needed.
+     */
+    if (src_mem == NULL){
+        VP8_CL_CREATE_BUF( cq, src_mem,, sizeof (unsigned char) * src_len, src_base-2,,);
+        src_offset = 2;
+        free_src = 1;
+    } else {
+        src_offset -= 2*src_pixels_per_line;
+    }
+
+    if (dst_mem == NULL){
+        VP8_CL_CREATE_BUF( cq, dst_mem,, sizeof (unsigned char) * dst_len + dst_offset, dst_base,, );
+        free_dst = 1;
+    }
+
+#if !STATIC_MEM
+    CL_CREATE_BUF( cq, int_mem,, sizeof(cl_int)*FData_height*FData_width, NULL,, );
+#endif
+
+    vp8_filter_block2d_first_pass_cl(
+        cq, src_mem, src_offset, int_mem, src_pixels_per_line,
+        FData_height, FData_width, xoffset
+    );
+
+    vp8_filter_block2d_second_pass_cl(cq,int_mem,int_offset,dst_mem,dst_offset,dst_pitch,
+            output_height,output_width,yoffset);
+
+    if (free_src == 1)
+        clReleaseMemObject(src_mem);
+
+    if (free_dst == 1){
+        /* Read back the result data from the device */
+        err = clEnqueueReadBuffer(cq, dst_mem, CL_FALSE, 0, sizeof (unsigned char) * dst_len + dst_offset, dst_base, 0, NULL, NULL);
+        VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+            "Error: Failed to read output array!\n",
+            ,
+        );
+        clReleaseMemObject(dst_mem);
+    }
+
+#if !STATIC_MEM
+    clReleaseMemObject(int_mem);
+#endif
+}
+
+void vp8_sixtap_predict4x4_cl
+(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch
+) {
+
+    int output_width=4, output_height=4, FData_height=9, FData_width=4;
+
+    //Size of output to transfer
+    int dst_len = DST_LEN(dst_pitch,output_height,output_width);
+    int src_len = SIXTAP_SRC_LEN(FData_width,FData_height,src_pixels_per_line);
+
+#if TWO_PASS_SIXTAP
+    int int_offset = 8;
+    unsigned char *src_ptr = src_base + src_offset;
+
+    vp8_sixtap_run_cl(cq, src_mem, dst_mem,
+            (src_ptr-2*src_pixels_per_line),src_offset, src_len,
+            src_pixels_per_line, xoffset,yoffset,dst_base,dst_offset,
+            dst_pitch,dst_len,FData_height,FData_width,output_height,
+            output_width,int_offset
+    );
+#else
+    vp8_sixtap_single_pass(
+            cq,
+            cl_data.vp8_sixtap_predict_kernel,
+            cl_data.vp8_sixtap_predict_kernel_size,
+            FData_height*FData_width,
+            src_mem,
+            dst_mem,
+            src_base,
+            src_offset,
+            src_len,
+            src_pixels_per_line,
+            xoffset,
+            yoffset,
+            dst_base,
+            dst_offset,
+            dst_pitch,
+            dst_len
+    );
+#endif
+
+
+    return;
+}
+
+void vp8_sixtap_predict8x8_cl
+(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch
+) {
+    int output_width=8, output_height=8, FData_height=13, FData_width=8;
+
+    //Size of output to transfer
+    int dst_len = DST_LEN(dst_pitch,output_height,output_width);
+    int src_len = SIXTAP_SRC_LEN(FData_width,FData_height,src_pixels_per_line);
+
+#if TWO_PASS_SIXTAP
+    int int_offset = 16;
+    unsigned char *src_ptr = src_base + src_offset;
+
+    vp8_sixtap_run_cl(cq, src_mem, dst_mem,
+            (src_ptr-2*src_pixels_per_line),src_offset, src_len,
+            src_pixels_per_line, xoffset,yoffset,dst_base,dst_offset,
+            dst_pitch,dst_len,FData_height,FData_width,output_height,
+            output_width,int_offset
+    );
+#else
+    vp8_sixtap_single_pass(
+            cq,
+            cl_data.vp8_sixtap_predict8x8_kernel,
+            cl_data.vp8_sixtap_predict8x8_kernel_size,
+            FData_height*FData_width,
+            src_mem,
+            dst_mem,
+            src_base,
+            src_offset,
+            src_len,
+            src_pixels_per_line,
+            xoffset,
+            yoffset,
+            dst_base,
+            dst_offset,
+            dst_pitch,
+            dst_len
+    );
+#endif
+
+    return;
+}
+
+void vp8_sixtap_predict8x4_cl
+(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch
+) {
+
+    int output_width=8, output_height=4, FData_height=9, FData_width=8;
+
+    //Size of output to transfer
+    int dst_len = DST_LEN(dst_pitch,output_height,output_width);
+    int src_len = SIXTAP_SRC_LEN(FData_width,FData_height,src_pixels_per_line);
+
+#if TWO_PASS_SIXTAP
+    int int_offset = 16;
+    unsigned char *src_ptr = src_base + src_offset;
+    
+    vp8_sixtap_run_cl(cq, src_mem, dst_mem,
+            (src_ptr-2*src_pixels_per_line),src_offset, src_len,
+            src_pixels_per_line, xoffset,yoffset,dst_base,dst_offset,
+            dst_pitch,dst_len,FData_height,FData_width,output_height,
+            output_width,int_offset
+    );
+#else
+    vp8_sixtap_single_pass(
+            cq,
+            cl_data.vp8_sixtap_predict8x4_kernel,
+            cl_data.vp8_sixtap_predict8x4_kernel_size,
+            FData_height*FData_width,
+            src_mem,
+            dst_mem,
+            src_base,
+            src_offset,
+            src_len,
+            src_pixels_per_line,
+            xoffset,
+            yoffset,
+            dst_base,
+            dst_offset,
+            dst_pitch,
+            dst_len
+    );
+#endif
+
+    return;
+}
+
+void vp8_sixtap_predict16x16_cl
+(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch
+) {
+
+    int output_width=16, output_height=16, FData_height=21, FData_width=16;
+
+    //Size of output to transfer
+    int dst_len = DST_LEN(dst_pitch,output_height,output_width);
+    int src_len = SIXTAP_SRC_LEN(FData_width,FData_height,src_pixels_per_line);
+
+#if TWO_PASS_SIXTAP
+    int int_offset = 32;
+    unsigned char *src_ptr = src_base + src_offset;
+
+    vp8_sixtap_run_cl(cq, src_mem, dst_mem,
+            (src_ptr-2*src_pixels_per_line),src_offset, src_len,
+            src_pixels_per_line, xoffset,yoffset,dst_base,dst_offset,
+            dst_pitch,dst_len,FData_height,FData_width,output_height,
+            output_width,int_offset
+    );
+#else
+    vp8_sixtap_single_pass(
+            cq,
+            cl_data.vp8_sixtap_predict16x16_kernel,
+            cl_data.vp8_sixtap_predict16x16_kernel_size,
+            FData_height*FData_width,
+            src_mem,
+            dst_mem,
+            src_base,
+            src_offset,
+            src_len,
+            src_pixels_per_line,
+            xoffset,
+            yoffset,
+            dst_base,
+            dst_offset,
+            dst_pitch,
+            dst_len
+    );
+#endif
+
+    return;
+
+}
+
+
+
+void vp8_filter_block2d_bil_first_pass_cl(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    cl_mem int_mem,
+    int src_pixels_per_line,
+    int height,
+    int width,
+    int xoffset
+)
+{
+    int err;
+    size_t global = width*height;
+    int free_src = 0;
+
+    if (src_mem == NULL){
+        int src_len = BIL_SRC_LEN(width,height,src_pixels_per_line);
+
+        /*Make space for kernel input/output data. Initialize the buffer as well if needed. */
+        VP8_CL_CREATE_BUF(cq, src_mem, CL_MEM_READ_ONLY|CL_MEM_COPY_HOST_PTR,
+            sizeof (unsigned char) * src_len, src_base+src_offset,,
+        );
+        src_offset = 0; //Set to zero as long as src_mem starts at base+offset
+        free_src = 1;
+    }
+
+    err =  clSetKernelArg(cl_data.vp8_filter_block2d_bil_first_pass_kernel, 0, sizeof (cl_mem), &src_mem);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_first_pass_kernel, 1, sizeof (int), &src_offset);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_first_pass_kernel, 2, sizeof (cl_mem), &int_mem);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_first_pass_kernel, 3, sizeof (int), &src_pixels_per_line);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_first_pass_kernel, 4, sizeof (int), &height);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_first_pass_kernel, 5, sizeof (int), &width);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_first_pass_kernel, 6, sizeof (int), &xoffset);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        ,
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel( cq, cl_data.vp8_filter_block2d_bil_first_pass_kernel, 1, NULL, &global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);,
+    );
+
+    if (free_src == 1)
+        clReleaseMemObject(src_mem);
+}
+
+
+void vp8_filter_block2d_bil_second_pass_cl(
+    cl_command_queue cq,
+    cl_mem int_mem,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch,
+    int height,
+    int width,
+    int yoffset
+)
+{
+    int err;
+    size_t global = width*height;
+
+    //Size of output data
+    int dst_len = DST_LEN(dst_pitch,height,width);
+
+    int free_dst = 0;
+    if (dst_mem == NULL){
+        VP8_CL_CREATE_BUF(cq, dst_mem, CL_MEM_WRITE_ONLY|CL_MEM_COPY_HOST_PTR,
+            sizeof (unsigned char) * dst_len + dst_offset, dst_base,,
+        );
+        free_dst = 1;
+    }
+
+    err =  clSetKernelArg(cl_data.vp8_filter_block2d_bil_second_pass_kernel, 0, sizeof (cl_mem), &int_mem);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_second_pass_kernel, 1, sizeof (cl_mem), &dst_mem);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_second_pass_kernel, 2, sizeof (int), &dst_offset);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_second_pass_kernel, 3, sizeof (int), &dst_pitch);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_second_pass_kernel, 4, sizeof (int), &height);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_second_pass_kernel, 5, sizeof (int), &width);
+    err |= clSetKernelArg(cl_data.vp8_filter_block2d_bil_second_pass_kernel, 6, sizeof (int), &yoffset);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        ,
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel( cq, cl_data.vp8_filter_block2d_bil_second_pass_kernel, 1, NULL, &global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);,
+    );
+
+    if (free_dst == 1){
+        /* Read back the result data from the device */
+        err = clEnqueueReadBuffer(cq, dst_mem, CL_FALSE, 0, sizeof (unsigned char) * dst_len + dst_offset, dst_base, 0, NULL, NULL);
+        VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+            "Error: Failed to read output array!\n",
+            ,
+        );
+        clReleaseMemObject(dst_mem);
+    }
+
+}
+
+void vp8_bilinear_predict4x4_cl
+(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch
+) {
+
+    const int height = 4, width = 4;
+
+#if !STATIC_MEM
+    int err;
+    cl_mem int_mem = NULL;
+    VP8_CL_CREATE_BUF(NULL, int_mem, NULL, sizeof(cl_int)*21*16, NULL, ,);
+#endif
+    
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_bil_first_pass_cl(cq, src_base, src_mem, src_offset, int_mem, src_pixels_per_line, height + 1, width, xoffset);
+
+    /* then 1-D vertically... */
+    vp8_filter_block2d_bil_second_pass_cl(cq, int_mem, dst_base, dst_mem, dst_offset, dst_pitch, height, width, yoffset);
+
+#if !STATIC_MEM
+    clReleaseMemObject(int_mem);
+#endif
+
+}
+
+void vp8_bilinear_predict8x8_cl
+(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch
+) {
+
+    const int height = 8, width = 8;
+
+#if !STATIC_MEM
+    int err;
+    cl_mem int_mem = NULL;
+    VP8_CL_CREATE_BUF(NULL, int_mem, NULL, sizeof(cl_int)*21*16, NULL, ,);
+#endif
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_bil_first_pass_cl(cq, src_base, src_mem, src_offset, int_mem, src_pixels_per_line, height + 1, width, xoffset);
+
+    /* then 1-D vertically... */
+    vp8_filter_block2d_bil_second_pass_cl(cq, int_mem, dst_base, dst_mem, dst_offset, dst_pitch, height, width, yoffset);
+
+#if !STATIC_MEM
+    clReleaseMemObject(int_mem);
+#endif
+    
+}
+
+void vp8_bilinear_predict8x4_cl
+(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch
+) {
+
+    const int height = 4, width = 8;
+
+#if !STATIC_MEM
+    int err;
+    cl_mem int_mem = NULL;
+    VP8_CL_CREATE_BUF(NULL, int_mem, NULL, sizeof(cl_int)*21*16, NULL, ,);
+#endif
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_bil_first_pass_cl(cq, src_base, src_mem, src_offset, int_mem, src_pixels_per_line, height + 1, width, xoffset);
+
+    /* then 1-D vertically... */
+    vp8_filter_block2d_bil_second_pass_cl(cq, int_mem, dst_base, dst_mem, dst_offset, dst_pitch, height, width, yoffset);
+
+#if !STATIC_MEM
+    clReleaseMemObject(int_mem);
+#endif
+
+}
+
+void vp8_bilinear_predict16x16_cl
+(
+    cl_command_queue cq,
+    unsigned char *src_base,
+    cl_mem src_mem,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    unsigned char *dst_base,
+    cl_mem dst_mem,
+    int dst_offset,
+    int dst_pitch
+) {
+
+    const int height = 16, width = 16;
+
+#if !STATIC_MEM
+    int err;
+    cl_mem int_mem = NULL;
+    VP8_CL_CREATE_BUF(NULL, int_mem, NULL, sizeof(cl_int)*21*16, NULL, ,);
+#endif
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_bil_first_pass_cl(cq, src_base, src_mem, src_offset, int_mem, src_pixels_per_line, height + 1, width, xoffset);
+
+    /* then 1-D vertically... */
+    vp8_filter_block2d_bil_second_pass_cl(cq, int_mem, dst_base, dst_mem, dst_offset, dst_pitch, height, width, yoffset);
+
+#if !STATIC_MEM
+    clReleaseMemObject(int_mem);
+#endif
+
+}

vp8/common/opencl/filter_cl.cl

@@ -0,0 +1,562 @@
+#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
+#pragma OPENCL EXTENSION cl_amd_printf : enable
+
+__constant int bilinear_filters[8][2] = {
+    { 128, 0},
+    { 112, 16},
+    { 96, 32},
+    { 80, 48},
+    { 64, 64},
+    { 48, 80},
+    { 32, 96},
+    { 16, 112}
+};
+
+__constant short sub_pel_filters[8][8] = {
+    //These were originally 8x6, but are padded for vector ops
+    { 0, 0, 128, 0, 0, 0, 0, 0}, /* note that 1/8 pel positions are just as per alpha -0.5 bicubic */
+    { 0, -6, 123, 12, -1, 0, 0, 0},
+    { 2, -11, 108, 36, -8, 1, 0, 0}, /* New 1/4 pel 6 tap filter */
+    { 0, -9, 93, 50, -6, 0, 0, 0},
+    { 3, -16, 77, 77, -16, 3, 0, 0}, /* New 1/2 pel 6 tap filter */
+    { 0, -6, 50, 93, -9, 0, 0, 0},
+    { 1, -8, 36, 108, -11, 2, 0, 0}, /* New 1/4 pel 6 tap filter */
+    { 0, -1, 12, 123, -6, 0, 0, 0},
+};
+
+
+kernel void vp8_filter_block2d_first_pass_kernel(
+    __global unsigned char *src_base,
+    int src_offset,
+    __global int *output_ptr,
+    unsigned int src_pixels_per_line,
+    unsigned int output_height,
+    unsigned int output_width,
+    int filter_offset
+){
+    uint tid = get_global_id(0);
+
+    global unsigned char *src_ptr = &src_base[src_offset];
+    //Note that src_offset will be reset later, which is why we use it now
+
+    int Temp;
+
+    __constant short *vp8_filter = sub_pel_filters[filter_offset];
+
+    if (tid < (output_width*output_height)){
+        src_offset = tid + (tid/output_width * (src_pixels_per_line - output_width));
+
+        Temp = (int)(src_ptr[src_offset - 2] * vp8_filter[0]) +
+           (int)(src_ptr[src_offset - 1] * vp8_filter[1]) +
+           (int)(src_ptr[src_offset]     * vp8_filter[2]) +
+           (int)(src_ptr[src_offset + 1] * vp8_filter[3]) +
+           (int)(src_ptr[src_offset + 2] * vp8_filter[4]) +
+           (int)(src_ptr[src_offset + 3] * vp8_filter[5]) +
+           (VP8_FILTER_WEIGHT >> 1);      /* Rounding */
+
+        /* Normalize back to 0-255 */
+        Temp = Temp >> VP8_FILTER_SHIFT;
+
+        if (Temp < 0)
+            Temp = 0;
+        else if ( Temp > 255 )
+            Temp = 255;
+
+        output_ptr[tid] = Temp;
+    }
+
+}
+
+kernel void vp8_filter_block2d_second_pass_kernel
+(
+    __global int *src_base,
+    int src_offset,
+    __global unsigned char *output_base,
+    int output_offset,
+    int output_pitch,
+    unsigned int src_pixels_per_line,
+    unsigned int pixel_step,
+    unsigned int output_height,
+    unsigned int output_width,
+    int filter_offset
+) {
+
+    uint i = get_global_id(0);
+
+    global int *src_ptr = &src_base[src_offset];
+    global unsigned char *output_ptr = &output_base[output_offset];
+
+    int out_offset; //Not same as output_offset...
+    int Temp;
+    int PS2 = 2*(int)pixel_step;
+    int PS3 = 3*(int)pixel_step;
+
+    unsigned int src_increment = src_pixels_per_line - output_width;
+
+    __constant short *vp8_filter = sub_pel_filters[filter_offset];
+
+    if (i < (output_width * output_height)){
+        out_offset = i/output_width;
+        src_offset = out_offset;
+
+        src_offset = i + (src_offset * src_increment);
+        out_offset = i%output_width + (out_offset * output_pitch);
+
+        /* Apply filter */
+        Temp = ((int)src_ptr[src_offset - PS2] * vp8_filter[0]) +
+           ((int)src_ptr[src_offset -(int)pixel_step] * vp8_filter[1]) +
+           ((int)src_ptr[src_offset]                  * vp8_filter[2]) +
+           ((int)src_ptr[src_offset + pixel_step]     * vp8_filter[3]) +
+           ((int)src_ptr[src_offset + PS2]       * vp8_filter[4]) +
+           ((int)src_ptr[src_offset + PS3]       * vp8_filter[5]) +
+           (VP8_FILTER_WEIGHT >> 1);   /* Rounding */
+
+        /* Normalize back to 0-255 */
+        Temp = Temp >> VP8_FILTER_SHIFT;
+        if (Temp < 0)
+            Temp = 0;
+        else if (Temp > 255)
+            Temp = 255;
+
+        output_ptr[out_offset] = (unsigned char)Temp;
+    }
+}
+
+
+kernel void vp8_filter_block2d_bil_first_pass_kernel(
+    __global unsigned char *src_base,
+    int src_offset,
+    __global int *output_ptr,
+    unsigned int src_pixels_per_line,
+    unsigned int output_height,
+    unsigned int output_width,
+    int filter_offset
+)
+{
+    uint tid = get_global_id(0);
+
+    if (tid < output_width * output_height){
+        global unsigned char *src_ptr = &src_base[src_offset];
+
+        unsigned int i, j;
+        __constant int *vp8_filter = bilinear_filters[filter_offset];
+
+        unsigned int out_row,out_offset;
+        int src_increment = src_pixels_per_line - output_width;
+
+        i = tid / output_width;
+        j = tid % output_width;
+
+        src_offset = i*(output_width+src_increment) + j;
+        out_row = output_width * i;
+
+        out_offset = out_row + j;
+
+        /* Apply bilinear filter */
+        output_ptr[out_offset] = (((int)src_ptr[src_offset]   * vp8_filter[0]) +
+                 ((int)src_ptr[src_offset+1] * vp8_filter[1]) +
+                 (VP8_FILTER_WEIGHT / 2)) >> VP8_FILTER_SHIFT;
+    }
+}
+
+kernel void vp8_filter_block2d_bil_second_pass_kernel
+(
+    __global int *src_ptr,
+    __global unsigned char *output_base,
+    int output_offset,
+    int output_pitch,
+    unsigned int output_height,
+    unsigned int output_width,
+    int filter_offset
+)
+{
+
+    uint tid = get_global_id(0);
+
+    if (tid < output_width * output_height){
+        global unsigned char *output_ptr = &output_base[output_offset];
+
+        unsigned int i, j;
+        int Temp;
+        __constant int *vp8_filter = bilinear_filters[filter_offset];
+
+        int out_offset;
+        int src_offset;
+
+        i = tid / output_width;
+        j = tid % output_width;
+
+        src_offset = i*(output_width) + j;
+        out_offset = i*output_pitch + j;
+
+        /* Apply filter */
+        Temp = ((int)src_ptr[src_offset]         * vp8_filter[0]) +
+               ((int)src_ptr[src_offset+output_width] * vp8_filter[1]) +
+               (VP8_FILTER_WEIGHT / 2);
+
+        output_ptr[out_offset++] = (unsigned int)(Temp >> VP8_FILTER_SHIFT);
+    }
+}
+
+
+
+
+//Called from reconinter_cl.c
+kernel void vp8_memcpy_kernel(
+    global unsigned char *src_base,
+    int src_offset,
+    int src_stride,
+    global unsigned char *dst_base,
+    int dst_offset,
+    int dst_stride,
+    int num_bytes,
+    int num_iter
+){
+
+    int i,r;
+    global unsigned char *src = &src_base[src_offset];
+    global unsigned char *dst = &dst_base[dst_offset];
+    src_offset = dst_offset = 0;
+
+    r = get_global_id(1);
+    if (r < get_global_size(1)){
+        i = get_global_id(0);
+        if (i < get_global_size(0)){
+            src_offset = r*src_stride + i;
+            dst_offset = r*dst_stride + i;
+            dst[dst_offset] = src[src_offset];
+        }
+    }
+}
+
+//Not used currently.
+void vp8_memset_short(
+    global short *mem,
+    int offset,
+    short newval,
+    unsigned int size
+)
+{
+    int tid = get_global_id(0);
+
+    if (tid < (size/2)){
+        mem[offset+tid/2] = newval;
+    }
+}
+
+
+
+__kernel void vp8_bilinear_predict4x4_kernel
+(
+        __global unsigned char *src_base,
+        int src_offset,
+        int src_pixels_per_line,
+        int xoffset,
+        int yoffset,
+        __global unsigned char *dst_base,
+        int dst_offset,
+        int dst_pitch,
+        __global int *int_mem
+)
+{
+    int Height = 4, Width = 4;
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_bil_first_pass_kernel(src_base, src_offset, int_mem, src_pixels_per_line, Height + 1, Width, xoffset);
+
+    /* then 1-D vertically... */
+    vp8_filter_block2d_bil_second_pass_kernel(int_mem, dst_base, dst_offset, dst_pitch, Height, Width, yoffset);
+}
+
+__kernel void vp8_bilinear_predict8x8_kernel
+(
+    __global unsigned char *src_base,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    __global unsigned char *dst_base,
+    int dst_offset,
+    int dst_pitch,
+    __global int *int_mem
+)
+{
+    int Height = 8, Width = 8;
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_bil_first_pass_kernel(src_base, src_offset, int_mem, src_pixels_per_line, Height + 1, Width, xoffset);
+
+    /* then 1-D vertically... */
+    vp8_filter_block2d_bil_second_pass_kernel(int_mem, dst_base, dst_offset, dst_pitch, Height, Width, yoffset);
+
+}
+
+__kernel void vp8_bilinear_predict8x4_kernel
+(
+    __global unsigned char *src_base,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    __global unsigned char *dst_base,
+    int dst_offset,
+    int dst_pitch,
+    __global int *int_mem
+)
+{
+    int Height = 4, Width = 8;
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_bil_first_pass_kernel(src_base, src_offset, int_mem, src_pixels_per_line, Height + 1, Width, xoffset);
+
+    /* then 1-D vertically... */
+    vp8_filter_block2d_bil_second_pass_kernel(int_mem, dst_base, dst_offset, dst_pitch, Height, Width, yoffset);
+}
+
+__kernel void vp8_bilinear_predict16x16_kernel
+(
+    __global unsigned char *src_base,
+    int src_offset,
+    int src_pixels_per_line,
+    int xoffset,
+    int yoffset,
+    __global unsigned char *dst_base,
+    int dst_offset,
+    int dst_pitch,
+    __global int *int_mem
+)
+{
+
+    int Height = 16, Width = 16;
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_bil_first_pass_kernel(src_base, src_offset, int_mem, src_pixels_per_line, Height + 1, Width, xoffset);
+
+    /* then 1-D vertically... */
+    vp8_filter_block2d_bil_second_pass_kernel(int_mem, dst_base, dst_offset, dst_pitch, Height, Width, yoffset);
+
+}
+
+void vp8_filter_block2d_first_pass(
+    global unsigned char *src_base,
+    int src_offset,
+    local int *output_ptr,
+    unsigned int src_pixels_per_line,
+    unsigned int pixel_step,
+    unsigned int output_height,
+    unsigned int output_width,
+    int filter_offset
+){
+    uint tid = get_global_id(0);
+    uint i = tid;
+
+    int nthreads = get_global_size(0);
+    int ngroups = nthreads / get_local_size(0);
+
+    global unsigned char *src_ptr = &src_base[src_offset];
+    //Note that src_offset will be reset later, which is why we capture it now
+
+    int Temp;
+
+    __constant short *vp8_filter = sub_pel_filters[filter_offset];
+
+    if (tid < (output_width*output_height)){
+        short filter0 = vp8_filter[0];
+        short filter1 = vp8_filter[1];
+        short filter2 = vp8_filter[2];
+        short filter3 = vp8_filter[3];
+        short filter4 = vp8_filter[4];
+        short filter5 = vp8_filter[5];
+
+        if (ngroups > 1){
+            //This is generally only true on Apple CPU-CL, which gives a group
+            //size of 1, regardless of the CPU core count.
+            for (i=0; i < output_width*output_height; i++){
+                src_offset = i + (i/output_width * (src_pixels_per_line - output_width));
+
+                Temp = (int)(src_ptr[src_offset - 2] * filter0) +
+                       (int)(src_ptr[src_offset - 1] * filter1) +
+                       (int)(src_ptr[src_offset]     * filter2) +
+                       (int)(src_ptr[src_offset + 1] * filter3) +
+                       (int)(src_ptr[src_offset + 2] * filter4) +
+                       (int)(src_ptr[src_offset + 3] * filter5) +
+                       (VP8_FILTER_WEIGHT >> 1);      /* Rounding */
+
+                /* Normalize back to 0-255 */
+                Temp >>= VP8_FILTER_SHIFT;
+
+                if (Temp < 0)
+                    Temp = 0;
+                else if ( Temp > 255 )
+                    Temp = 255;
+
+                output_ptr[i] = Temp;
+            }
+        } else {
+            src_offset = i + (i/output_width * (src_pixels_per_line - output_width));
+
+            Temp = (int)(src_ptr[src_offset - 2] * filter0) +
+                   (int)(src_ptr[src_offset - 1] * filter1) +
+                   (int)(src_ptr[src_offset]     * filter2) +
+                   (int)(src_ptr[src_offset + 1] * filter3) +
+                   (int)(src_ptr[src_offset + 2] * filter4) +
+                   (int)(src_ptr[src_offset + 3] * filter5) +
+                   (VP8_FILTER_WEIGHT >> 1);      /* Rounding */
+
+            /* Normalize back to 0-255 */
+            Temp >>= VP8_FILTER_SHIFT;
+
+            if (Temp < 0)
+                Temp = 0;
+            else if ( Temp > 255 )
+                Temp = 255;
+
+            output_ptr[i] = Temp;
+        }
+    }
+
+    //Add a fence so that no 2nd pass stuff starts before 1st pass writes are done.
+    barrier(CLK_LOCAL_MEM_FENCE);
+}
+
+void vp8_filter_block2d_second_pass
+(
+    local int *src_ptr,
+    global unsigned char *output_base,
+    int output_offset,
+    int output_pitch,
+    unsigned int src_pixels_per_line,
+    unsigned int pixel_step,
+    unsigned int output_height,
+    unsigned int output_width,
+    int filter_offset
+) {
+
+    global unsigned char *output_ptr = &output_base[output_offset];
+
+    int out_offset; //Not same as output_offset...
+    int src_offset;
+    int Temp;
+    int PS2 = 2*(int)pixel_step;
+    int PS3 = 3*(int)pixel_step;
+
+    unsigned int src_increment = src_pixels_per_line - output_width;
+
+    uint i = get_global_id(0);
+
+    __constant short *vp8_filter = sub_pel_filters[filter_offset];
+
+    if (i < (output_width * output_height)){
+        out_offset = i/output_width;
+        src_offset = out_offset;
+
+        src_offset = i + (src_offset * src_increment);
+        out_offset = i%output_width + (out_offset * output_pitch);
+
+        /* Apply filter */
+        Temp = ((int)src_ptr[src_offset - PS2] * vp8_filter[0]) +
+           ((int)src_ptr[src_offset -(int)pixel_step] * vp8_filter[1]) +
+           ((int)src_ptr[src_offset]                  * vp8_filter[2]) +
+           ((int)src_ptr[src_offset + pixel_step]     * vp8_filter[3]) +
+           ((int)src_ptr[src_offset + PS2]            * vp8_filter[4]) +
+           ((int)src_ptr[src_offset + PS3]       * vp8_filter[5]) +
+           (VP8_FILTER_WEIGHT >> 1);   /* Rounding */
+
+        /* Normalize back to 0-255 */
+        Temp = Temp >> VP8_FILTER_SHIFT;
+        if (Temp < 0)
+            Temp = 0;
+        else if (Temp > 255)
+            Temp = 255;
+
+        output_ptr[out_offset] = (unsigned char)Temp;
+    }
+}
+
+__kernel void vp8_sixtap_predict_kernel
+(
+    __global unsigned char  *src_ptr,
+    int src_offset,
+    int  src_pixels_per_line,
+    int  xoffset,
+    int  yoffset,
+    __global unsigned char *dst_ptr,
+    int dst_offset,
+    int  dst_pitch
+)
+{
+
+    local int FData[9*4];
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_first_pass(src_ptr, src_offset, FData, src_pixels_per_line, 1, 9, 4, xoffset);
+
+    /* then filter vertically... */
+    vp8_filter_block2d_second_pass(&FData[8], dst_ptr, dst_offset, dst_pitch, 4, 4, 4, 4, yoffset);
+}
+
+__kernel void vp8_sixtap_predict8x8_kernel
+(
+    __global unsigned char  *src_ptr,
+    int src_offset,
+    int  src_pixels_per_line,
+    int  xoffset,
+    int  yoffset,
+    __global unsigned char *dst_ptr,
+    int dst_offset,
+    int  dst_pitch
+)
+{
+    local int FData[13*16];   /* Temp data bufffer used in filtering */
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_first_pass(src_ptr, src_offset, FData, src_pixels_per_line, 1, 13, 8, xoffset);
+
+    /* then filter vertically... */
+    vp8_filter_block2d_second_pass(&FData[16], dst_ptr, dst_offset, dst_pitch, 8, 8, 8, 8, yoffset);
+
+}
+
+__kernel void vp8_sixtap_predict8x4_kernel
+(
+    __global unsigned char  *src_ptr,
+    int src_offset,
+    int  src_pixels_per_line,
+    int  xoffset,
+    int  yoffset,
+    __global unsigned char *dst_ptr,
+    int dst_offset,
+    int  dst_pitch
+)
+{
+    local int FData[13*16];   /* Temp data buffer used in filtering */
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_first_pass(src_ptr, src_offset, FData, src_pixels_per_line, 1, 9, 8, xoffset);
+
+    /* then filter verticaly... */
+    vp8_filter_block2d_second_pass(&FData[16], dst_ptr, dst_offset, dst_pitch, 8, 8, 4, 8, yoffset);
+}
+
+__kernel void vp8_sixtap_predict16x16_kernel
+(
+    __global unsigned char  *src_ptr,
+    int src_offset,
+    int  src_pixels_per_line,
+    int  xoffset,
+    int  yoffset,
+    __global unsigned char *dst_ptr,
+    int dst_offset,
+    int  dst_pitch
+)
+{
+    local int FData[21*24];   /* Temp data buffer used in filtering */
+
+    /* First filter 1-D horizontally... */
+    vp8_filter_block2d_first_pass(src_ptr, src_offset, FData, src_pixels_per_line, 1, 21, 16, xoffset);
+
+    /* then filter verticaly... */
+    vp8_filter_block2d_second_pass(&FData[32], dst_ptr, dst_offset, dst_pitch, 16, 16, 16, 16, yoffset);
+
+    return;
+}

vp8/common/opencl/filter_cl.h

@@ -0,0 +1,74 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef FILTER_CL_H_
+#define FILTER_CL_H_
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#include "vp8_opencl.h"
+
+#define VP8_FILTER_WEIGHT 128
+#define VP8_FILTER_SHIFT  7
+
+#define REGISTER_FILTER 1
+#define CLAMP(x,min,max) if (x < min) x = min; else if ( x > max ) x = max;
+#define PRE_CALC_PIXEL_STEPS 1
+#define PRE_CALC_SRC_INCREMENT 1
+
+#if PRE_CALC_PIXEL_STEPS
+#define PS2 two_pixel_steps
+#define PS3 three_pixel_steps
+#else
+#define PS2 2*(int)pixel_step
+#define PS3 3*(int)pixel_step
+#endif
+
+#if REGISTER_FILTER
+#define FILTER0 filter0
+#define FILTER1 filter1
+#define FILTER2 filter2
+#define FILTER3 filter3
+#define FILTER4 filter4
+#define FILTER5 filter5
+#else
+#define FILTER0 vp8_filter[0]
+#define FILTER1 vp8_filter[1]
+#define FILTER2 vp8_filter[2]
+#define FILTER3 vp8_filter[3]
+#define FILTER4 vp8_filter[4]
+#define FILTER5 vp8_filter[5]
+#endif
+
+#if PRE_CALC_SRC_INCREMENT
+#define SRC_INCREMENT src_increment
+#else
+#define SRC_INCREMENT (src_pixels_per_line - output_width)
+#endif
+
+#define FILTER_OFFSET //Filter data stored as CL constant memory
+#define FILTER_REF sub_pel_filters[filter_offset]
+
+extern const char *filterCompileOptions;
+extern const char *filter_cl_file_name;
+
+//Copy the -2*pixel_step (and ps*3) bytes because the filter algorithm
+//accesses negative indexes
+#define SIXTAP_SRC_LEN(out_width,out_height,src_px) ((out_width)*(out_height) + (((out_width)*(out_height)-1)/(out_width))*(src_px - out_width) + 5)
+#define BIL_SRC_LEN(out_width,out_height,src_px) ((out_height) * src_px + out_width)
+#define DST_LEN(dst_pitch,dst_height,dst_width) (dst_pitch * (dst_height) + (dst_width))
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* FILTER_CL_H_ */

vp8/common/opencl/idct_cl.h

@@ -0,0 +1,45 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef IDCT_OPENCL_H
+#define IDCT_OPENCL_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#include "vp8_opencl.h"
+#include "vp8/common/blockd.h"
+
+#define prototype_second_order_cl(sym) \
+    void sym(BLOCKD *b)
+
+#define prototype_idct_cl(sym) \
+    void sym(BLOCKD *b, int pitch)
+
+#define prototype_idct_scalar_add_cl(sym) \
+    void sym(BLOCKD *b, cl_int use_diff, int diff_offset, int qcoeff_offset, \
+             int pred_offset, unsigned char *output, cl_mem out_mem, int out_offset, size_t out_size, \
+             int pitch, int stride)\
+
+
+extern prototype_idct_cl(vp8_short_idct4x4llm_1_cl);
+extern prototype_idct_cl(vp8_short_idct4x4llm_cl);
+extern prototype_idct_scalar_add_cl(vp8_dc_only_idct_add_cl);
+
+extern prototype_second_order_cl(vp8_short_inv_walsh4x4_1_cl);
+extern prototype_second_order_cl(vp8_short_inv_walsh4x4_cl);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif

vp8/common/opencl/idctllm_cl.c

@@ -0,0 +1,325 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdlib.h>
+
+//ACW: Remove me after debugging.
+#include <stdio.h>
+#include <string.h>
+
+#include "idct_cl.h"
+#include "idctllm_cl.h"
+#include "blockd_cl.h"
+
+void cl_destroy_idct(){
+
+    if (cl_data.idct_program)
+        clReleaseProgram(cl_data.idct_program);
+
+    cl_data.idct_program = NULL;
+    
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_short_inv_walsh4x4_1_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_short_inv_walsh4x4_1st_pass_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_short_inv_walsh4x4_2nd_pass_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_dc_only_idct_add_kernel);
+    //VP8_CL_RELEASE_KERNEL(cl_data.vp8_short_idct4x4llm_1_kernel);
+    //VP8_CL_RELEASE_KERNEL(cl_data.vp8_short_idct4x4llm_kernel);
+
+}
+
+int cl_init_idct() {
+    int err;
+
+    // Create the filter compute program from the file-defined source code
+    if (cl_load_program(&cl_data.idct_program, idctllm_cl_file_name,
+            idctCompileOptions) != CL_SUCCESS)
+        return VP8_CL_TRIED_BUT_FAILED;
+
+    // Create the compute kernel in the program we wish to run
+    VP8_CL_CREATE_KERNEL(cl_data,idct_program,vp8_short_inv_walsh4x4_1_kernel,"vp8_short_inv_walsh4x4_1_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,idct_program,vp8_short_inv_walsh4x4_1st_pass_kernel,"vp8_short_inv_walsh4x4_1st_pass_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,idct_program,vp8_short_inv_walsh4x4_2nd_pass_kernel,"vp8_short_inv_walsh4x4_2nd_pass_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,idct_program,vp8_dc_only_idct_add_kernel,"vp8_dc_only_idct_add_kernel");
+
+    ////idct4x4llm kernels are only useful for the encoder
+    //VP8_CL_CREATE_KERNEL(cl_data,idct_program,vp8_short_idct4x4llm_1_kernel,"vp8_short_idct4x4llm_1_kernel");
+    //VP8_CL_CREATE_KERNEL(cl_data,idct_program,vp8_short_idct4x4llm_kernel,"vp8_short_idct4x4llm_kernel");
+
+    return CL_SUCCESS;
+}
+
+#define max(x,y) (x > y ? x: y)
+//#define NO_CL
+
+/* Only useful for encoder... Untested... */
+void vp8_short_idct4x4llm_cl(BLOCKD *b, int pitch)
+{
+    int err;
+
+    short *input = b->dqcoeff_base + b->dqcoeff_offset;
+    short *output = &b->diff_base[b->diff_offset];
+
+    cl_mem src_mem, dst_mem;
+
+    //1 instance for now. This should be split into 2-pass * 4 thread.
+    size_t global = 1;
+
+    if (cl_initialized != CL_SUCCESS){
+        vp8_short_idct4x4llm_c(input,output,pitch);
+        return;
+    }
+
+    VP8_CL_CREATE_BUF(b->cl_commands, src_mem,,
+            sizeof(short)*16, input,
+            vp8_short_idct4x4llm_c(input,output,pitch),
+    );
+
+    VP8_CL_CREATE_BUF(b->cl_commands, dst_mem,,
+            sizeof(short)*(4+(pitch/2)*3), output,
+            vp8_short_idct4x4llm_c(input,output,pitch),
+    );
+
+    //Set arguments and run kernel
+    err = 0;
+    err = clSetKernelArg(cl_data.vp8_short_idct4x4llm_kernel, 0, sizeof (cl_mem), &src_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_idct4x4llm_kernel, 1, sizeof (cl_mem), &dst_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_idct4x4llm_kernel, 2, sizeof (int), &pitch);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        vp8_short_idct4x4llm_c(input,output,pitch),
+    );
+    
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel(b->cl_commands, cl_data.vp8_short_idct4x4llm_kernel, 1, NULL, &global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);
+        vp8_short_idct4x4llm_c(input,output,pitch),
+    );
+
+    /* Read back the result data from the device */
+    err = clEnqueueReadBuffer(b->cl_commands, dst_mem, CL_FALSE, 0, sizeof(short)*(4+pitch/2*3), output, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS(b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read output array!\n",
+        vp8_short_idct4x4llm_c(input,output,pitch),
+    );
+
+    clReleaseMemObject(src_mem);
+    clReleaseMemObject(dst_mem);
+
+    return;
+}
+
+/* Only useful for encoder... Untested... */
+void vp8_short_idct4x4llm_1_cl(BLOCKD *b, int pitch)
+{
+    int err;
+    size_t global = 4;
+
+    short *input = b->dqcoeff_base + b->dqcoeff_offset;
+    short *output = &b->diff_base[b->diff_offset];
+
+    cl_mem src_mem, dst_mem;
+
+    if (cl_initialized != CL_SUCCESS){
+        vp8_short_idct4x4llm_1_c(input,output,pitch);
+        return;
+    }
+
+    printf("vp8_short_idct4x4llm_1_cl\n");
+
+    VP8_CL_CREATE_BUF(b->cl_commands, src_mem,,
+            sizeof(short), input,
+            vp8_short_idct4x4llm_1_c(input,output,pitch),
+    );
+
+    VP8_CL_CREATE_BUF(b->cl_commands, dst_mem,,
+            sizeof(short)*(4+(pitch/2)*3), output,
+            vp8_short_idct4x4llm_1_c(input,output,pitch),
+    );
+
+    //Set arguments and run kernel
+    err = 0;
+    err = clSetKernelArg(cl_data.vp8_short_idct4x4llm_1_kernel, 0, sizeof (cl_mem), &src_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_idct4x4llm_1_kernel, 1, sizeof (cl_mem), &dst_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_idct4x4llm_1_kernel, 2, sizeof (int), &pitch);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        vp8_short_idct4x4llm_1_c(input,output,pitch),
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel(b->cl_commands, cl_data.vp8_short_idct4x4llm_1_kernel, 1, NULL, &global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);
+        vp8_short_idct4x4llm_1_c(input,output,pitch),
+    );
+
+    /* Read back the result data from the device */
+    err = clEnqueueReadBuffer(b->cl_commands, dst_mem, CL_FALSE, 0, sizeof(short)*(4+pitch/2*3), output, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS(b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read output array!\n",
+        vp8_short_idct4x4llm_1_c(input,output,pitch),
+    );
+
+    clReleaseMemObject(src_mem);
+    clReleaseMemObject(dst_mem);
+
+    return;
+
+}
+
+void vp8_dc_only_idct_add_cl(BLOCKD *b, cl_int use_diff, int diff_offset, 
+        int qcoeff_offset, int pred_offset,
+        unsigned char *dst_base, cl_mem dst_mem, int dst_offset, size_t dest_size,
+        int pitch, int stride
+)
+{
+    
+    int err;
+    size_t global = 16;
+
+    int free_mem = 0;
+    //cl_mem dest_mem = NULL;
+
+    if (dst_mem == NULL){
+        VP8_CL_CREATE_BUF(b->cl_commands, dst_mem,,
+                dest_size, dst_base,,
+        );
+        free_mem = 1;
+    }
+
+    //Set arguments and run kernel
+    err =  clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 0, sizeof (cl_mem), &b->cl_predictor_mem);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 1, sizeof (int), &pred_offset);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 2, sizeof (cl_mem), &dst_mem);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 3, sizeof (int), &dst_offset);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 4, sizeof (int), &pitch);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 5, sizeof (int), &stride);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 6, sizeof (cl_int), &use_diff);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 7, sizeof (cl_mem), &b->cl_diff_mem);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 8, sizeof (int), &diff_offset);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 9, sizeof (cl_mem), &b->cl_qcoeff_mem);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 10, sizeof (int), &qcoeff_offset);
+    err |= clSetKernelArg(cl_data.vp8_dc_only_idct_add_kernel, 11, sizeof (cl_mem), &b->cl_dequant_mem);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",,
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel(b->cl_commands, cl_data.vp8_dc_only_idct_add_kernel, 1, NULL, &global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);,
+    );
+
+
+    if (free_mem == 1){
+    /* Read back the result data from the device */
+        err = clEnqueueReadBuffer(b->cl_commands, dst_mem, CL_FALSE, 0,
+                dest_size, dst_base, 0, NULL, NULL);
+
+        VP8_CL_CHECK_SUCCESS(b->cl_commands, err != CL_SUCCESS,
+            "Error: Failed to read output array!\n",,
+        );
+
+        clReleaseMemObject(dst_mem);
+    }
+
+    return;
+}
+
+void vp8_short_inv_walsh4x4_cl(BLOCKD *b)
+{
+    int err;
+    size_t global = 4;
+
+    if (cl_initialized != CL_SUCCESS){
+        vp8_short_inv_walsh4x4_c(b->dqcoeff_base+b->dqcoeff_offset,&b->diff_base[b->diff_offset]);
+        return;
+    }
+
+    //Set arguments and run kernel
+    err = 0;
+    err = clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_1st_pass_kernel, 0, sizeof (cl_mem), &b->cl_dqcoeff_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_1st_pass_kernel, 1, sizeof(int), &b->dqcoeff_offset);
+    err |= clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_1st_pass_kernel, 2, sizeof (cl_mem), &b->cl_diff_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_1st_pass_kernel, 3, sizeof(int), &b->diff_offset);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        vp8_short_inv_walsh4x4_c(b->dqcoeff_base+b->dqcoeff_offset, &b->diff_base[b->diff_offset]),
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel(b->cl_commands, cl_data.vp8_short_inv_walsh4x4_1st_pass_kernel, 1, NULL, &global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);
+        vp8_short_inv_walsh4x4_c(b->dqcoeff_base+b->dqcoeff_offset, &b->diff_base[b->diff_offset]),
+    );
+
+    //Second pass
+    //Set arguments and run kernel
+    err = 0;
+    err = clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_2nd_pass_kernel, 0, sizeof (cl_mem), &b->cl_diff_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_2nd_pass_kernel, 1, sizeof(int), &b->diff_offset);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        vp8_short_inv_walsh4x4_c(b->dqcoeff_base+b->dqcoeff_offset, &b->diff_base[b->diff_offset]),
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel(b->cl_commands, cl_data.vp8_short_inv_walsh4x4_2nd_pass_kernel, 1, NULL, &global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);
+        vp8_short_inv_walsh4x4_c(b->dqcoeff_base+b->dqcoeff_offset, &b->diff_base[b->diff_offset]),
+    );
+
+    return;
+}
+
+void vp8_short_inv_walsh4x4_1_cl(BLOCKD *b)
+{
+    
+    int err;
+    size_t global = 4;
+
+    if (cl_initialized != CL_SUCCESS){
+        vp8_short_inv_walsh4x4_1_c(b->dqcoeff_base + b->dqcoeff_offset,
+            &b->diff_base[b->diff_offset]);
+        return;
+    }
+
+    //Set arguments and run kernel
+    err = 0;
+    err = clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_1_kernel, 0, sizeof (cl_mem), &b->cl_dqcoeff_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_1_kernel, 1, sizeof (int), &b->dqcoeff_offset);
+    err |= clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_1_kernel, 2, sizeof (cl_mem), &b->cl_diff_mem);
+    err |= clSetKernelArg(cl_data.vp8_short_inv_walsh4x4_1_kernel, 3, sizeof (int), &b->diff_offset);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        vp8_short_inv_walsh4x4_1_c(b->dqcoeff_base + b->dqcoeff_offset,
+            &b->diff_base[b->diff_offset]),
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel(b->cl_commands, cl_data.vp8_short_inv_walsh4x4_1_kernel, 1, NULL, &global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( b->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);
+        vp8_short_inv_walsh4x4_1_c(b->dqcoeff_base + b->dqcoeff_offset,
+                &b->diff_base[b->diff_offset]),
+    );
+
+    return;
+}

vp8/common/opencl/idctllm_cl.cl

@@ -0,0 +1,309 @@
+#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
+#pragma OPENCL EXTENSION cl_amd_printf : enable
+
+__constant int cospi8sqrt2minus1 = 20091;
+__constant int sinpi8sqrt2      = 35468;
+__constant int rounding = 0;
+
+
+kernel void vp8_short_idct4x4llm_1st_pass_kernel(global short*,global short *,int);
+kernel void vp8_short_idct4x4llm_2nd_pass_kernel(global short*,int);
+
+
+__kernel void vp8_short_idct4x4llm_kernel(
+    __global short *input,
+    __global short *output,
+    int pitch
+){
+    vp8_short_idct4x4llm_1st_pass_kernel(input,output,pitch);
+    vp8_short_idct4x4llm_2nd_pass_kernel(output,pitch);
+}
+
+__kernel void vp8_short_idct4x4llm_1st_pass_kernel(
+    __global short *ip,
+    __global short *op,
+    int pitch
+)
+{
+    int i;
+    int a1, b1, c1, d1;
+
+    int temp1, temp2;
+    int shortpitch = pitch >> 1;
+
+    for (i = 0; i < 4; i++)
+    {
+        a1 = ip[0] + ip[8];
+        b1 = ip[0] - ip[8];
+
+        temp1 = (ip[4] * sinpi8sqrt2 + rounding) >> 16;
+        temp2 = ip[12] + ((ip[12] * cospi8sqrt2minus1 + rounding) >> 16);
+        c1 = temp1 - temp2;
+
+        temp1 = ip[4] + ((ip[4] * cospi8sqrt2minus1 + rounding) >> 16);
+        temp2 = (ip[12] * sinpi8sqrt2 + rounding) >> 16;
+        d1 = temp1 + temp2;
+
+        op[shortpitch*0] = a1 + d1;
+        op[shortpitch*3] = a1 - d1;
+
+        op[shortpitch*1] = b1 + c1;
+        op[shortpitch*2] = b1 - c1;
+
+        ip++;
+        op++;
+    }
+
+    return;
+}
+
+__kernel void vp8_short_idct4x4llm_2nd_pass_kernel(
+    __global short *output,
+    int pitch
+)
+{
+    int i;
+    int a1, b1, c1, d1;
+
+    int temp1, temp2;
+    int shortpitch = pitch >> 1;
+    __global short *ip = output;
+    __global short *op = output;
+
+    for (i = 0; i < 4; i++)
+    {
+        a1 = ip[0] + ip[2];
+        b1 = ip[0] - ip[2];
+
+        temp1 = (ip[1] * sinpi8sqrt2 + rounding) >> 16;
+        temp2 = ip[3] + ((ip[3] * cospi8sqrt2minus1 + rounding) >> 16);
+        c1 = temp1 - temp2;
+
+        temp1 = ip[1] + ((ip[1] * cospi8sqrt2minus1 + rounding) >> 16);
+        temp2 = (ip[3] * sinpi8sqrt2 + rounding) >> 16;
+        d1 = temp1 + temp2;
+
+        op[0] = (a1 + d1 + 4) >> 3;
+        op[3] = (a1 - d1 + 4) >> 3;
+
+        op[1] = (b1 + c1 + 4) >> 3;
+        op[2] = (b1 - c1 + 4) >> 3;
+
+        ip += shortpitch;
+        op += shortpitch;
+    }
+
+    return;
+}
+
+__kernel void vp8_short_idct4x4llm_1_kernel(
+    __global short *input,
+    __global short *output,
+    int pitch
+)
+{
+    int a1;
+    int out_offset;
+    int shortpitch = pitch >> 1;
+
+    //short4 a;
+    a1 = ((input[0] + 4) >> 3);
+    //a = a1;
+
+    int tid = get_global_id(0);
+    if (tid < 4){
+        out_offset = shortpitch * tid;
+
+        //vstore4(a,0,&output[out_offset];
+        output[out_offset] = a1;
+        output[out_offset+1] = a1;
+        output[out_offset+2] = a1;
+        output[out_offset+3] = a1;
+    }
+}
+
+__kernel void vp8_dc_only_idct_add_kernel(
+    __global unsigned char *pred_base,
+    int pred_offset,
+    __global unsigned char *dst_base,
+    int dst_offset,
+    int pitch,
+    int stride,
+    int use_diff,
+    global short *diff_base,
+    int diff_offset,
+    global short *qcoeff_base,
+    int qcoeff_offset,
+    global short *dequant
+)
+{
+    int r, c;
+    //int pred_offset;
+    global unsigned char *pred_ptr = &pred_base[pred_offset];
+    global unsigned char *dst_ptr = &dst_base[dst_offset];
+
+    int tid = get_global_id(0);
+
+    int a1;
+
+    if (tid < 16){
+
+        if (use_diff == 1){
+            a1 = diff_base[diff_offset];
+        } else {
+            a1 = qcoeff_base[qcoeff_offset] * dequant[0];
+        }
+        a1 = (a1 + 4)>>3;
+
+        r = tid / 4;
+        c = tid % 4;
+
+        pred_offset = r * pitch;
+        dst_offset += r * stride;
+        int a = a1 + pred_ptr[pred_offset + c] ;
+
+        if (a < 0)
+            a = 0;
+        else if (a > 255)
+            a = 255;
+
+        dst_base[dst_offset + c] = (unsigned char) a ;
+    }
+}
+
+
+__kernel void vp8_short_inv_walsh4x4_1st_pass_kernel(
+    __global short *src_base,
+    int src_offset,
+    __global short *output_base,
+    int out_offset
+)
+{
+
+    __global short *input = src_base + src_offset;
+    __global short *output = output_base + src_offset;
+    int tid = get_global_id(0);
+
+#define VEC_WALSH 0
+#if VEC_WALSH
+    //4-short vectors to calculate things in
+    short4 a,b,c,d, a2v, b2v, c2v, d2v, a1t, b1t, c1t, d1t;
+    short16 out;
+
+    if (tid == 0){
+        //first pass loop in vector form
+        a = vload4(0,input) + vload4(3,input);
+        b = vload4(1,input) + vload4(2,input);
+        c = vload4(1,input) - vload4(2,input);
+        d = vload4(0,input) - vload4(3,input);
+        vstore4(a + b, 0, output);
+        vstore4(c + d, 1, output);
+        vstore4(a - b, 2, output);
+        vstore4(d - c, 3, output);
+
+        return;
+
+        //2nd pass
+        a = (short4)(output[0], output[4], output[8], output[12]);
+        b = (short4)(output[1], output[5], output[9], output[13]);
+        c = (short4)(output[1], output[5], output[9], output[13]);
+        d = (short4)(output[0], output[4], output[8], output[12]);
+        a1t = (short4)(output[3], output[7], output[11], output[15]);
+        b1t = (short4)(output[2], output[6], output[10], output[14]);
+        c1t = (short4)(output[2], output[6], output[10], output[14]);
+        d1t = (short4)(output[3], output[7], output[11], output[15]);
+
+        a = a + a1t + (short)3;
+        b = b + b1t;
+        c = c - c1t;
+        d = d - d1t + (short)3;
+
+        a2v = (a + b) >> (short)3;
+        b2v = (c + d) >> (short)3;
+        c2v = (a - b) >> (short)3;
+        d2v = (d - c) >> (short)3;
+
+        out.s048c = a2v;
+        out.s159d = b2v;
+        out.s26ae = c2v;
+        out.s37bf = d2v;
+        vstore16(out,0,output);
+    }
+#else
+
+    int i;
+    int a1, b1, c1, d1;
+    int a2, b2, c2, d2;
+    global short *ip = input;
+    global short *op = output;
+
+    int offset;
+
+    if (tid < 4){
+        offset = tid;
+        a1 = ip[offset] + ip[offset + 12];
+        b1 = ip[offset + 4] + ip[offset + 8];
+        c1 = ip[offset + 4] - ip[offset + 8];
+        d1 = ip[offset] - ip[offset + 12];
+
+        op[offset] = a1 + b1;
+        op[offset + 4] = c1 + d1;
+        op[offset + 8] = a1 - b1;
+        op[offset + 12] = d1 - c1;
+    }
+#endif
+}
+
+__kernel void vp8_short_inv_walsh4x4_2nd_pass_kernel(
+    __global short *output_base,
+    int out_offset
+)
+{
+    int i;
+    int a1, b1, c1, d1;
+    int a2, b2, c2, d2;
+
+    __global short *output = output_base + out_offset;
+    int tid = get_global_id(0);
+    int offset = 0;
+
+    if (tid < 4){
+        offset = 4*tid;
+        a1 = output[offset] + output[offset + 3];
+        b1 = output[offset + 1] + output[offset + 2];
+        c1 = output[offset + 1] - output[offset + 2];
+        d1 = output[offset + 0] - output[offset + 3];
+
+        a2 = a1 + b1;
+        b2 = c1 + d1;
+        c2 = a1 - b1;
+        d2 = d1 - c1;
+
+        output[offset + 0] = (a2 + 3) >> 3;
+        output[offset + 1] = (b2 + 3) >> 3;
+        output[offset + 2] = (c2 + 3) >> 3;
+        output[offset + 3] = (d2 + 3) >> 3;
+    }
+}
+
+__kernel void vp8_short_inv_walsh4x4_1_kernel(
+    __global short *src_data,
+    int src_offset,
+    __global short *dst_data,
+    int dst_offset
+){
+    int a1;
+    int tid = get_global_id(0);
+    //short16 a;
+    int i;
+    short4 a;
+    __global short *input = src_data + src_offset;
+    __global short *output = dst_data + dst_offset;
+
+    if (tid < 4)
+    {
+        a1 = ((input[0] + 3) >> 3);
+        a = (short)a1; //Set all elements of vector to a1
+        vstore4(a, tid, output);
+    }
+}

vp8/common/opencl/idctllm_cl.h

@@ -0,0 +1,26 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_config.h"
+#include "vp8_opencl.h"
+#include "vp8/common/blockd.h"
+
+#define CLAMP(x,min,max) if (x < min) x = min; else if ( x > max ) x = max;
+
+//External functions that are fallbacks if CL is unavailable
+extern void vp8_short_idct4x4llm_c(short *input, short *output, int pitch);
+extern void vp8_short_idct4x4llm_1_c(short *input, short *output, int pitch);
+extern void vp8_dc_only_idct_add_c(short input_dc, unsigned char *pred_ptr, unsigned char *dst_ptr, int pitch, int stride);
+extern void vp8_short_inv_walsh4x4_c(short *input, short *output);
+extern void vp8_short_inv_walsh4x4_1_c(short *input, short *output);
+
+const char *idctCompileOptions = "-Ivp8/common/opencl";
+const char *idctllm_cl_file_name = "vp8/common/opencl/idctllm_cl.cl";
+

vp8/common/opencl/loopfilter.cl

@@ -0,0 +1,427 @@
+#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
+#pragma OPENCL EXTENSION cl_amd_printf : enable
+
+typedef unsigned char uc;
+typedef signed char sc;
+
+__inline signed char vp8_filter_mask(sc, sc, uc, uc, uc, uc, uc, uc, uc, uc);
+__inline signed char vp8_simple_filter_mask(signed char, signed char, uc, uc, uc, uc);
+__inline signed char vp8_hevmask(signed char, uc, uc, uc, uc);
+__inline signed char vp8_signed_char_clamp(int);
+
+__inline void vp8_mbfilter(signed char mask,signed char hev,global uc *op2,
+    global uc *op1,global uc *op0,global uc *oq0,global uc *oq1,global uc *oq2);
+
+void vp8_simple_filter(signed char mask,global uc *base, int op1_off,int op0_off,int oq0_off,int oq1_off);
+
+
+typedef struct
+{
+    signed char lim[16];
+    signed char flim[16];
+    signed char thr[16];
+    signed char mbflim[16];
+    signed char mbthr[16];
+    signed char uvlim[16];
+    signed char uvflim[16];
+    signed char uvthr[16];
+    signed char uvmbflim[16];
+    signed char uvmbthr[16];
+} loop_filter_info;
+
+
+
+
+void vp8_filter(
+    signed char mask,
+    signed char hev,
+    global uc *base,
+    int op1_off,
+    int op0_off,
+    int oq0_off,
+    int oq1_off
+)
+{
+
+    global uc *op1 = &base[op1_off];
+    global uc *op0 = &base[op0_off];
+    global uc *oq0 = &base[oq0_off];
+    global uc *oq1 = &base[oq1_off];
+
+    signed char ps0, qs0;
+    signed char ps1, qs1;
+    signed char vp8_filter, Filter1, Filter2;
+    signed char u;
+
+    ps1 = (signed char) * op1 ^ 0x80;
+    ps0 = (signed char) * op0 ^ 0x80;
+    qs0 = (signed char) * oq0 ^ 0x80;
+    qs1 = (signed char) * oq1 ^ 0x80;
+
+    /* add outer taps if we have high edge variance */
+    vp8_filter = vp8_signed_char_clamp(ps1 - qs1);
+    vp8_filter &= hev;
+
+    /* inner taps */
+    vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * (qs0 - ps0));
+    vp8_filter &= mask;
+
+    /* save bottom 3 bits so that we round one side +4 and the other +3
+     * if it equals 4 we'll set to adjust by -1 to account for the fact
+     * we'd round 3 the other way
+     */
+    Filter1 = vp8_signed_char_clamp(vp8_filter + 4);
+    Filter2 = vp8_signed_char_clamp(vp8_filter + 3);
+    Filter1 >>= 3;
+    Filter2 >>= 3;
+    u = vp8_signed_char_clamp(qs0 - Filter1);
+    *oq0 = u ^ 0x80;
+    u = vp8_signed_char_clamp(ps0 + Filter2);
+    *op0 = u ^ 0x80;
+    vp8_filter = Filter1;
+
+    /* outer tap adjustments */
+    vp8_filter += 1;
+    vp8_filter >>= 1;
+    vp8_filter &= ~hev;
+
+    u = vp8_signed_char_clamp(qs1 - vp8_filter);
+    *oq1 = u ^ 0x80;
+    u = vp8_signed_char_clamp(ps1 + vp8_filter);
+    *op1 = u ^ 0x80;
+}
+
+
+kernel void vp8_loop_filter_horizontal_edge_kernel
+(
+    global unsigned char *s_base,
+    int s_off,
+    int p, /* pitch */
+    global signed char *flimit,
+    global signed char *limit,
+    global signed char *thresh,
+    int off_stride
+)
+{
+    int  hev = 0; /* high edge variance */
+    signed char mask = 0;
+    int i = get_global_id(0);
+
+    if (i < get_global_size(0)){
+        s_off += i;
+
+        mask = vp8_filter_mask(limit[i], flimit[i], s_base[s_off - 4*p],
+                s_base[s_off - 3*p], s_base[s_off - 2*p], s_base[s_off - p],
+                s_base[s_off], s_base[s_off + p], s_base[s_off + 2*p],
+                s_base[s_off + 3*p]);
+
+        hev = vp8_hevmask(thresh[i], s_base[s_off - 2*p], s_base[s_off - p],
+                s_base[s_off], s_base[s_off+p]);
+
+        vp8_filter(mask, hev, s_base, s_off - 2 * p, s_off - p, s_off,
+                s_off + p);
+    }
+}
+
+
+kernel void vp8_loop_filter_vertical_edge_kernel
+(
+    global unsigned char *s_base,
+    int s_off,
+    int p,
+    global signed char *flimit,
+    global signed char *limit,
+    global signed char *thresh,
+    int off_stride
+)
+{
+
+    int  hev = 0; /* high edge variance */
+    signed char mask = 0;
+    int i = get_global_id(0);
+
+    if ( i < get_global_size(0) ){
+        s_off += p * i;
+        mask = vp8_filter_mask(limit[i], flimit[i],
+                s_base[s_off-4], s_base[s_off-3], s_base[s_off-2],
+                s_base[s_off-1], s_base[s_off], s_base[s_off+1],
+                s_base[s_off+2], s_base[s_off+3]);
+
+        hev = vp8_hevmask(thresh[i], s_base[s_off-2], s_base[s_off-1],
+                s_base[s_off], s_base[s_off+1]);
+
+        vp8_filter(mask, hev, s_base, s_off - 2, s_off - 1, s_off, s_off + 1);
+
+    }
+}
+
+
+kernel void vp8_mbloop_filter_horizontal_edge_kernel
+(
+    global unsigned char *s_base,
+    int s_off,
+    int p,
+    global signed char *flimit,
+    global signed char *limit,
+    global signed char *thresh,
+    int off_stride
+)
+{
+
+    global uc *s = s_base+s_off;
+
+    signed char hev = 0; /* high edge variance */
+    signed char mask = 0;
+    int i = get_global_id(0);
+
+    if (i < get_global_size(0)){
+        s += i;
+
+        mask = vp8_filter_mask(limit[i], flimit[i],
+                               s[-4*p], s[-3*p], s[-2*p], s[-1*p],
+                               s[0*p], s[1*p], s[2*p], s[3*p]);
+
+        hev = vp8_hevmask(thresh[i], s[-2*p], s[-1*p], s[0*p], s[1*p]);
+
+        vp8_mbfilter(mask, hev, s - 3 * p, s - 2 * p, s - 1 * p, s, s + 1 * p, s + 2 * p);
+
+    }
+}
+
+
+kernel void vp8_mbloop_filter_vertical_edge_kernel
+(
+    global unsigned char *s_base,
+    int s_off,
+    int p,
+    global signed char *flimit,
+    global signed char *limit,
+    global signed char *thresh,
+    int off_stride
+)
+{
+
+    global uc *s = s_base + s_off;
+
+    signed char hev = 0; /* high edge variance */
+    signed char mask = 0;
+    int i = get_global_id(0);
+
+    if (i < get_global_size(0)){
+        s += p * i;
+
+        mask = vp8_filter_mask(limit[i], flimit[i],
+                               s[-4], s[-3], s[-2], s[-1], s[0], s[1], s[2], s[3]);
+
+        hev = vp8_hevmask(thresh[i], s[-2], s[-1], s[0], s[1]);
+
+        vp8_mbfilter(mask, hev, s - 3, s - 2, s - 1, s, s + 1, s + 2);
+
+    }
+}
+
+
+kernel void vp8_loop_filter_simple_horizontal_edge_kernel
+(
+    global unsigned char *s_base,
+    int s_off,
+    int p,
+    global const signed char *flimit,
+    global const signed char *limit,
+    global const signed char *thresh,
+    int off_stride
+)
+{
+
+    signed char mask = 0;
+    int i = get_global_id(0);
+    (void) thresh;
+
+    if (i < get_global_size(0))
+    {
+        s_off += i;
+        mask = vp8_simple_filter_mask(limit[i], flimit[i], s_base[s_off-2*p], s_base[s_off-p], s_base[s_off], s_base[s_off+p]);
+        vp8_simple_filter(mask, s_base, s_off - 2 * p, s_off - 1 * p, s_off, s_off + 1 * p);
+    }
+}
+
+
+kernel void vp8_loop_filter_simple_vertical_edge_kernel
+(
+    global unsigned char *s_base,
+    int s_off,
+    int p,
+    global signed char *flimit,
+    global signed char *limit,
+    global signed char *thresh,
+    int off_stride
+)
+{
+
+    signed char mask = 0;
+    int i = get_global_id(0);
+    (void) thresh;
+
+    if (i < get_global_size(0)){
+        s_off += p * i;
+        mask = vp8_simple_filter_mask(limit[i], flimit[i], s_base[s_off-2], s_base[s_off-1], s_base[s_off], s_base[s_off+1]);
+        vp8_simple_filter(mask, s_base, s_off - 2, s_off - 1, s_off, s_off + 1);
+    }
+
+}
+
+
+
+//Inline and non-kernel functions follow.
+
+__inline void vp8_mbfilter(
+    signed char mask,
+    signed char hev,
+    global uc *op2,
+    global uc *op1,
+    global uc *op0,
+    global uc *oq0,
+    global uc *oq1,
+    global uc *oq2
+)
+{
+    signed char s, u;
+    signed char vp8_filter, Filter1, Filter2;
+    signed char ps2 = (signed char) * op2 ^ 0x80;
+    signed char ps1 = (signed char) * op1 ^ 0x80;
+    signed char ps0 = (signed char) * op0 ^ 0x80;
+    signed char qs0 = (signed char) * oq0 ^ 0x80;
+    signed char qs1 = (signed char) * oq1 ^ 0x80;
+    signed char qs2 = (signed char) * oq2 ^ 0x80;
+
+    /* add outer taps if we have high edge variance */
+    vp8_filter = vp8_signed_char_clamp(ps1 - qs1);
+    vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * (qs0 - ps0));
+    vp8_filter &= mask;
+
+    Filter2 = vp8_filter;
+    Filter2 &= hev;
+
+    /* save bottom 3 bits so that we round one side +4 and the other +3 */
+    Filter1 = vp8_signed_char_clamp(Filter2 + 4);
+    Filter2 = vp8_signed_char_clamp(Filter2 + 3);
+    Filter1 >>= 3;
+    Filter2 >>= 3;
+    qs0 = vp8_signed_char_clamp(qs0 - Filter1);
+    ps0 = vp8_signed_char_clamp(ps0 + Filter2);
+
+
+    /* only apply wider filter if not high edge variance */
+    vp8_filter &= ~hev;
+    Filter2 = vp8_filter;
+
+    /* roughly 3/7th difference across boundary */
+    u = vp8_signed_char_clamp((63 + Filter2 * 27) >> 7);
+    s = vp8_signed_char_clamp(qs0 - u);
+    *oq0 = s ^ 0x80;
+    s = vp8_signed_char_clamp(ps0 + u);
+    *op0 = s ^ 0x80;
+
+    /* roughly 2/7th difference across boundary */
+    u = vp8_signed_char_clamp((63 + Filter2 * 18) >> 7);
+    s = vp8_signed_char_clamp(qs1 - u);
+    *oq1 = s ^ 0x80;
+    s = vp8_signed_char_clamp(ps1 + u);
+    *op1 = s ^ 0x80;
+
+    /* roughly 1/7th difference across boundary */
+    u = vp8_signed_char_clamp((63 + Filter2 * 9) >> 7);
+    s = vp8_signed_char_clamp(qs2 - u);
+    *oq2 = s ^ 0x80;
+    s = vp8_signed_char_clamp(ps2 + u);
+    *op2 = s ^ 0x80;
+}
+
+
+__inline signed char vp8_signed_char_clamp(int t)
+{
+    t = (t < -128 ? -128 : t);
+    t = (t > 127 ? 127 : t);
+    return (signed char) t;
+}
+
+
+/* is there high variance internal edge ( 11111111 yes, 00000000 no) */
+__inline signed char vp8_hevmask(signed char thresh, uc p1, uc p0, uc q0, uc q1)
+{
+    signed char hev = 0;
+    hev  |= (abs(p1 - p0) > thresh) * -1;
+    hev  |= (abs(q1 - q0) > thresh) * -1;
+    return hev;
+}
+
+
+/* should we apply any filter at all ( 11111111 yes, 00000000 no) */
+__inline signed char vp8_filter_mask(
+    signed char limit,
+    signed char flimit,
+     uc p3, uc p2, uc p1, uc p0, uc q0, uc q1, uc q2, uc q3)
+{
+    signed char mask = 0;
+    mask |= (abs(p3 - p2) > limit) * -1;
+    mask |= (abs(p2 - p1) > limit) * -1;
+    mask |= (abs(p1 - p0) > limit) * -1;
+    mask |= (abs(q1 - q0) > limit) * -1;
+    mask |= (abs(q2 - q1) > limit) * -1;
+    mask |= (abs(q3 - q2) > limit) * -1;
+    mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  > flimit * 2 + limit) * -1;
+    mask = ~mask;
+    return mask;
+}
+
+/* should we apply any filter at all ( 11111111 yes, 00000000 no) */
+__inline signed char vp8_simple_filter_mask(
+    signed char limit,
+    signed char flimit,
+    uc p1,
+    uc p0,
+    uc q0,
+    uc q1
+)
+{
+    signed char mask = (abs(p0 - q0) * 2 + abs(p1 - q1) / 2  <= flimit * 2 + limit) * -1;
+    return mask;
+}
+
+void vp8_simple_filter(
+    signed char mask,
+    global uc *base,
+    int op1_off,
+    int op0_off,
+    int oq0_off,
+    int oq1_off
+)
+{
+
+    global uc *op1 = base + op1_off;
+    global uc *op0 = base + op0_off;
+    global uc *oq0 = base + oq0_off;
+    global uc *oq1 = base + oq1_off;
+
+    signed char vp8_filter, Filter1, Filter2;
+    signed char p1 = (signed char) * op1 ^ 0x80;
+    signed char p0 = (signed char) * op0 ^ 0x80;
+    signed char q0 = (signed char) * oq0 ^ 0x80;
+    signed char q1 = (signed char) * oq1 ^ 0x80;
+    signed char u;
+
+    vp8_filter = vp8_signed_char_clamp(p1 - q1);
+    vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * (q0 - p0));
+    vp8_filter &= mask;
+
+    /* save bottom 3 bits so that we round one side +4 and the other +3 */
+    Filter1 = vp8_signed_char_clamp(vp8_filter + 4);
+    Filter1 >>= 3;
+    u = vp8_signed_char_clamp(q0 - Filter1);
+    *oq0  = u ^ 0x80;
+
+    Filter2 = vp8_signed_char_clamp(vp8_filter + 3);
+    Filter2 >>= 3;
+    u = vp8_signed_char_clamp(p0 + Filter2);
+    *op0 = u ^ 0x80;
+}

vp8/common/opencl/loopfilter_cl.c

@@ -0,0 +1,457 @@
+/*
+ *  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/config.h"
+#include "loopfilter_cl.h"
+#include "../onyxc_int.h"
+
+#include "vpx_config.h"
+#include "vp8_opencl.h"
+#include "blockd_cl.h"
+
+const char *loopFilterCompileOptions = "-Ivp8/common/opencl";
+const char *loop_filter_cl_file_name = "vp8/common/opencl/loopfilter.cl";
+
+typedef unsigned char uc;
+
+extern void vp8_loop_filter_frame
+(
+    VP8_COMMON *cm,
+    MACROBLOCKD *mbd,
+    int default_filt_lvl
+);
+
+prototype_loopfilter_cl(vp8_loop_filter_horizontal_edge_cl);
+prototype_loopfilter_cl(vp8_loop_filter_vertical_edge_cl);
+prototype_loopfilter_cl(vp8_mbloop_filter_horizontal_edge_cl);
+prototype_loopfilter_cl(vp8_mbloop_filter_vertical_edge_cl);
+prototype_loopfilter_cl(vp8_loop_filter_simple_horizontal_edge_cl);
+prototype_loopfilter_cl(vp8_loop_filter_simple_vertical_edge_cl);
+
+/* Horizontal MB filtering */
+void vp8_loop_filter_mbh_cl(
+    MACROBLOCKD *x,
+    cl_mem buf_base,
+    int y_off,
+    int u_off,
+    int v_off,
+    int y_stride,
+    int uv_stride,
+    loop_filter_info *lfi,
+    int simpler_lpf
+)
+{
+    (void) simpler_lpf;
+
+    vp8_mbloop_filter_horizontal_edge_cl(x, buf_base, y_off, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2, 1);
+    vp8_mbloop_filter_horizontal_edge_cl(x, buf_base, u_off, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1, 1);
+    vp8_mbloop_filter_horizontal_edge_cl(x, buf_base, v_off, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1, 1);
+}
+
+void vp8_loop_filter_mbhs_cl(MACROBLOCKD *x, cl_mem buf_base, int y_off, int u_off, int v_off,
+                            int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_horizontal_edge_cl(x, buf_base, y_off, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2, 1);
+}
+
+/* Vertical MB Filtering */
+void vp8_loop_filter_mbv_cl(MACROBLOCKD *x, cl_mem buf_base, int y_off, int u_off, int v_off,
+                           int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) simpler_lpf;
+
+    vp8_mbloop_filter_vertical_edge_cl(x, buf_base, y_off, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2, 1);
+    vp8_mbloop_filter_vertical_edge_cl(x, buf_base, u_off, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1, 1);
+    vp8_mbloop_filter_vertical_edge_cl(x, buf_base, v_off, uv_stride, lfi->mbflim, lfi->lim, lfi->thr, 1, 1);
+}
+
+void vp8_loop_filter_mbvs_cl(MACROBLOCKD *x, cl_mem buf_base, int y_off, int u_off, int v_off,
+                            int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) uv_stride;
+    (void) simpler_lpf;
+    vp8_loop_filter_simple_vertical_edge_cl(x, buf_base, y_off, y_stride, lfi->mbflim, lfi->lim, lfi->thr, 2, 1);
+}
+
+/* Horizontal B Filtering */
+void vp8_loop_filter_bh_cl(MACROBLOCKD *x, cl_mem buf_base, int y_off, int u_off, int v_off,
+                          int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) simpler_lpf;
+
+    vp8_loop_filter_horizontal_edge_cl(x, buf_base, y_off + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_horizontal_edge_cl(x, buf_base, y_off + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_horizontal_edge_cl(x, buf_base, y_off + 12 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_horizontal_edge_cl(x, buf_base, u_off + 4 * uv_stride, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1, 1);
+    vp8_loop_filter_horizontal_edge_cl(x, buf_base, v_off + 4 * uv_stride, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1, 1);
+
+}
+
+void vp8_loop_filter_bhs_cl(MACROBLOCKD *x, cl_mem buf_base, int y_off, int u_off, int v_off,
+                           int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) uv_stride;
+    (void) simpler_lpf;
+
+    vp8_loop_filter_simple_horizontal_edge_cl(x, buf_base, y_off + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_simple_horizontal_edge_cl(x, buf_base, y_off + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_simple_horizontal_edge_cl(x, buf_base, y_off + 12 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+}
+
+/* Vertical B Filtering */
+void vp8_loop_filter_bv_cl(MACROBLOCKD *x, cl_mem buf_base, int y_off, int u_off, int v_off,
+                          int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) simpler_lpf;
+
+    vp8_loop_filter_vertical_edge_cl(x, buf_base, y_off + 4, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_vertical_edge_cl(x, buf_base, y_off + 8, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_vertical_edge_cl(x, buf_base, y_off + 12, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+
+    vp8_loop_filter_vertical_edge_cl(x, buf_base, u_off + 4, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1, 1);
+    vp8_loop_filter_vertical_edge_cl(x, buf_base, v_off + 4, uv_stride, lfi->flim, lfi->lim, lfi->thr, 1, 1);
+}
+
+void vp8_loop_filter_bvs_cl(MACROBLOCKD *x, cl_mem buf_base, int y_off, int u_off, int v_off,
+                           int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
+{
+    (void) uv_stride;
+    (void) simpler_lpf;
+
+    vp8_loop_filter_simple_vertical_edge_cl(x, buf_base, y_off + 4, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_simple_vertical_edge_cl(x, buf_base, y_off + 8, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+    vp8_loop_filter_simple_vertical_edge_cl(x, buf_base, y_off + 12, y_stride, lfi->flim, lfi->lim, lfi->thr, 2, 1);
+}
+
+void vp8_init_loop_filter_cl(VP8_COMMON *cm)
+{
+    loop_filter_info *lfi = cm->lf_info;
+    int sharpness_lvl = cm->sharpness_level;
+    int frame_type = cm->frame_type;
+    int i, j;
+
+    int block_inside_limit = 0;
+    int HEVThresh;
+    const int yhedge_boost  = 2;
+
+    /* For each possible value for the loop filter fill out a "loop_filter_info" entry. */
+    for (i = 0; i <= MAX_LOOP_FILTER; i++)
+    {
+        int filt_lvl = i;
+
+        if (frame_type == KEY_FRAME)
+        {
+            if (filt_lvl >= 40)
+                HEVThresh = 2;
+            else if (filt_lvl >= 15)
+                HEVThresh = 1;
+            else
+                HEVThresh = 0;
+        }
+        else
+        {
+            if (filt_lvl >= 40)
+                HEVThresh = 3;
+            else if (filt_lvl >= 20)
+                HEVThresh = 2;
+            else if (filt_lvl >= 15)
+                HEVThresh = 1;
+            else
+                HEVThresh = 0;
+        }
+
+        /* Set loop filter paramaeters that control sharpness. */
+        block_inside_limit = filt_lvl >> (sharpness_lvl > 0);
+        block_inside_limit = block_inside_limit >> (sharpness_lvl > 4);
+
+        if (sharpness_lvl > 0)
+        {
+            if (block_inside_limit > (9 - sharpness_lvl))
+                block_inside_limit = (9 - sharpness_lvl);
+        }
+
+        if (block_inside_limit < 1)
+            block_inside_limit = 1;
+
+        for (j = 0; j < 16; j++)
+        {
+            lfi[i].lim[j] = block_inside_limit;
+            lfi[i].mbflim[j] = filt_lvl + yhedge_boost;
+            lfi[i].flim[j] = filt_lvl;
+            lfi[i].thr[j] = HEVThresh;
+        }
+    }
+}
+
+/* Put vp8_init_loop_filter() in vp8dx_create_decompressor(). Only call vp8_frame_init_loop_filter() while decoding
+ * each frame. Check last_frame_type to skip the function most of times.
+ */
+void vp8_frame_init_loop_filter_cl(loop_filter_info *lfi, int frame_type)
+{
+    int HEVThresh;
+    int i, j;
+
+    /* For each possible value for the loop filter fill out a "loop_filter_info" entry. */
+    for (i = 0; i <= MAX_LOOP_FILTER; i++)
+    {
+        int filt_lvl = i;
+
+        if (frame_type == KEY_FRAME)
+        {
+            if (filt_lvl >= 40)
+                HEVThresh = 2;
+            else if (filt_lvl >= 15)
+                HEVThresh = 1;
+            else
+                HEVThresh = 0;
+        }
+        else
+        {
+            if (filt_lvl >= 40)
+                HEVThresh = 3;
+            else if (filt_lvl >= 20)
+                HEVThresh = 2;
+            else if (filt_lvl >= 15)
+                HEVThresh = 1;
+            else
+                HEVThresh = 0;
+        }
+
+        for (j = 0; j < 16; j++)
+        {
+            lfi[i].thr[j] = HEVThresh;
+        }
+    }
+}
+
+
+//This might not need to be copied from loopfilter.c
+void vp8_adjust_mb_lf_value_cl(MACROBLOCKD *mbd, int *filter_level)
+{
+    MB_MODE_INFO *mbmi = &mbd->mode_info_context->mbmi;
+
+    if (mbd->mode_ref_lf_delta_enabled)
+    {
+        /* Apply delta for reference frame */
+        *filter_level += mbd->ref_lf_deltas[mbmi->ref_frame];
+
+        /* Apply delta for mode */
+        if (mbmi->ref_frame == INTRA_FRAME)
+        {
+            /* Only the split mode BPRED has a further special case */
+            if (mbmi->mode == B_PRED)
+                *filter_level +=  mbd->mode_lf_deltas[0];
+        }
+        else
+        {
+            /* Zero motion mode */
+            if (mbmi->mode == ZEROMV)
+                *filter_level +=  mbd->mode_lf_deltas[1];
+
+            /* Split MB motion mode */
+            else if (mbmi->mode == SPLITMV)
+                *filter_level +=  mbd->mode_lf_deltas[3];
+
+            /* All other inter motion modes (Nearest, Near, New) */
+            else
+                *filter_level +=  mbd->mode_lf_deltas[2];
+        }
+
+        /* Range check */
+        if (*filter_level > MAX_LOOP_FILTER)
+            *filter_level = MAX_LOOP_FILTER;
+        else if (*filter_level < 0)
+            *filter_level = 0;
+    }
+}
+
+
+//Start of externally callable functions.
+
+int cl_init_loop_filter() {
+    int err;
+
+    // Create the filter compute program from the file-defined source code
+    if ( cl_load_program(&cl_data.loop_filter_program, loop_filter_cl_file_name,
+            loopFilterCompileOptions) != CL_SUCCESS )
+        return VP8_CL_TRIED_BUT_FAILED;
+
+    // Create the compute kernels in the program we wish to run
+    VP8_CL_CREATE_KERNEL(cl_data,loop_filter_program,vp8_loop_filter_horizontal_edge_kernel,"vp8_loop_filter_horizontal_edge_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,loop_filter_program,vp8_loop_filter_vertical_edge_kernel,"vp8_loop_filter_vertical_edge_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,loop_filter_program,vp8_mbloop_filter_horizontal_edge_kernel,"vp8_mbloop_filter_horizontal_edge_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,loop_filter_program,vp8_mbloop_filter_vertical_edge_kernel,"vp8_mbloop_filter_vertical_edge_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,loop_filter_program,vp8_loop_filter_simple_horizontal_edge_kernel,"vp8_loop_filter_simple_horizontal_edge_kernel");
+    VP8_CL_CREATE_KERNEL(cl_data,loop_filter_program,vp8_loop_filter_simple_vertical_edge_kernel,"vp8_loop_filter_simple_vertical_edge_kernel");
+
+    return CL_SUCCESS;
+}
+
+void cl_destroy_loop_filter(){
+
+    if (cl_data.loop_filter_program)
+        clReleaseProgram(cl_data.loop_filter_program);
+
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_loop_filter_horizontal_edge_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_loop_filter_vertical_edge_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_mbloop_filter_horizontal_edge_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_mbloop_filter_vertical_edge_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_loop_filter_simple_horizontal_edge_kernel);
+    VP8_CL_RELEASE_KERNEL(cl_data.vp8_loop_filter_simple_vertical_edge_kernel);
+
+    cl_data.loop_filter_program = NULL;
+}
+
+
+void vp8_loop_filter_set_baselines_cl(MACROBLOCKD *mbd, int default_filt_lvl, int *baseline_filter_level){
+    int alt_flt_enabled = mbd->segmentation_enabled;
+    int i;
+
+    if (alt_flt_enabled)
+    {
+        for (i = 0; i < MAX_MB_SEGMENTS; i++)
+        {
+            /* Abs value */
+            if (mbd->mb_segement_abs_delta == SEGMENT_ABSDATA)
+                baseline_filter_level[i] = mbd->segment_feature_data[MB_LVL_ALT_LF][i];
+            /* Delta Value */
+            else
+            {
+                baseline_filter_level[i] = default_filt_lvl + mbd->segment_feature_data[MB_LVL_ALT_LF][i];
+                baseline_filter_level[i] = (baseline_filter_level[i] >= 0) ? ((baseline_filter_level[i] <= MAX_LOOP_FILTER) ? baseline_filter_level[i] : MAX_LOOP_FILTER) : 0;  /* Clamp to valid range */
+            }
+        }
+    }
+    else
+    {
+        for (i = 0; i < MAX_MB_SEGMENTS; i++)
+            baseline_filter_level[i] = default_filt_lvl;
+    }
+}
+
+void vp8_loop_filter_frame_cl
+(
+    VP8_COMMON *cm,
+    MACROBLOCKD *mbd,
+    int default_filt_lvl
+)
+{
+    YV12_BUFFER_CONFIG *post = cm->frame_to_show;
+    loop_filter_info *lfi = cm->lf_info;
+    FRAME_TYPE frame_type = cm->frame_type;
+    LOOPFILTERTYPE filter_type = cm->filter_type;
+
+    int mb_row;
+    int mb_col;
+
+    int baseline_filter_level[MAX_MB_SEGMENTS];
+    int filter_level;
+    int alt_flt_enabled = mbd->segmentation_enabled;
+
+    int err;
+    unsigned char *buf_base;
+    int y_off, u_off, v_off;
+    //unsigned char *y_ptr, *u_ptr, *v_ptr;
+
+    mbd->mode_info_context = cm->mi;          /* Point at base of Mb MODE_INFO list */
+
+    /* Note the baseline filter values for each segment */
+    vp8_loop_filter_set_baselines_cl(mbd, default_filt_lvl, baseline_filter_level);
+
+    /* Initialize the loop filter for this frame. */
+    if ((cm->last_filter_type != cm->filter_type) || (cm->last_sharpness_level != cm->sharpness_level))
+        vp8_init_loop_filter_cl(cm);
+    else if (frame_type != cm->last_frame_type)
+        vp8_frame_init_loop_filter_cl(lfi, frame_type);
+
+    /* Set up the buffer pointers */
+
+    buf_base = post->buffer_alloc;
+    y_off = post->y_buffer - buf_base;
+    u_off = post->u_buffer - buf_base;
+    v_off = post->v_buffer - buf_base;
+
+    VP8_CL_SET_BUF(mbd->cl_commands, post->buffer_mem, post->buffer_size, post->buffer_alloc,
+            vp8_loop_filter_frame(cm,mbd,default_filt_lvl),);
+
+    /* vp8_filter each macro block */
+    for (mb_row = 0; mb_row < cm->mb_rows; mb_row++)
+    {
+        for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
+        {
+            int Segment = (alt_flt_enabled) ? mbd->mode_info_context->mbmi.segment_id : 0;
+
+            filter_level = baseline_filter_level[Segment];
+
+            /* Distance of Mb to the various image edges.
+             * These specified to 8th pel as they are always compared to values 
+             * that are in 1/8th pel units. Apply any context driven MB level
+             * adjustment
+             */
+            filter_level = vp8_adjust_mb_lf_value(mbd, filter_level);
+
+            if (filter_level)
+            {
+                if (mb_col > 0){
+                    if (filter_type == NORMAL_LOOPFILTER)
+                        vp8_loop_filter_mbv_cl(mbd, post->buffer_mem, y_off, u_off, v_off, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
+                    else
+                        vp8_loop_filter_mbvs_cl(mbd, post->buffer_mem, y_off, u_off, v_off, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
+                }
+
+                if (mbd->mode_info_context->mbmi.dc_diff > 0){
+                    if (filter_type == NORMAL_LOOPFILTER)
+                        vp8_loop_filter_bv_cl(mbd, post->buffer_mem, y_off, u_off, v_off, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
+                    else
+                        vp8_loop_filter_bvs_cl(mbd, post->buffer_mem, y_off, u_off, v_off, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
+                }
+
+                /* don't apply across umv border */
+                if (mb_row > 0){
+                    if (filter_type == NORMAL_LOOPFILTER)
+                        vp8_loop_filter_mbh_cl(mbd, post->buffer_mem, y_off, u_off, v_off, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
+                    else
+                        vp8_loop_filter_mbhs_cl(mbd, post->buffer_mem, y_off, u_off, v_off, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
+                }
+
+                if (mbd->mode_info_context->mbmi.dc_diff > 0){
+                    if (filter_type == NORMAL_LOOPFILTER)
+                        vp8_loop_filter_bh_cl(mbd, post->buffer_mem, y_off, u_off, v_off, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
+                    else
+                        vp8_loop_filter_bhs_cl(mbd, post->buffer_mem, y_off, u_off, v_off, post->y_stride, post->uv_stride, &lfi[filter_level], cm->simpler_lpf);
+                }
+            }
+
+            y_off += 16;
+            u_off += 8;
+            v_off += 8;
+
+            mbd->mode_info_context++;     /* step to next MB */
+        }
+
+        y_off += post->y_stride  * 16 - post->y_width;
+        u_off += post->uv_stride *  8 - post->uv_width;
+        v_off += post->uv_stride *  8 - post->uv_width;
+
+        mbd->mode_info_context++;         /* Skip border mb */
+    }
+
+    //Retrieve buffer contents
+    err = clEnqueueReadBuffer(mbd->cl_commands, post->buffer_mem, CL_FALSE, 0, post->buffer_size, post->buffer_alloc, 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS(mbd->cl_commands, err != CL_SUCCESS,
+        "Error: Failed to read loop filter output!\n",
+        ,
+    );
+
+    VP8_CL_FINISH(mbd->cl_commands);
+}

vp8/common/opencl/loopfilter_cl.h

@@ -0,0 +1,48 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef loopfilter_cl_h
+#define loopfilter_cl_h
+
+#include "../../../vpx_ports/mem.h"
+
+#include "../onyxc_int.h"
+#include "blockd_cl.h"
+#include "../loopfilter.h"
+
+#define prototype_loopfilter_cl(sym) \
+    void sym(MACROBLOCKD*, cl_mem src_base, int src_offset,  \
+             int pitch, const signed char *flimit, \
+             const signed char *limit, const signed char *thresh, int count, int block_cnt)
+
+#define prototype_loopfilter_block_cl(sym) \
+    void sym(MACROBLOCKD*, unsigned char *y, unsigned char *u, unsigned char *v,\
+             int ystride, int uv_stride, loop_filter_info *lfi, int simpler)
+
+extern void vp8_loop_filter_frame_cl
+(
+    VP8_COMMON *cm,
+    MACROBLOCKD *mbd,
+    int default_filt_lvl
+);
+
+extern prototype_loopfilter_block_cl(vp8_lf_normal_mb_v_cl);
+extern prototype_loopfilter_block_cl(vp8_lf_normal_b_v_cl);
+extern prototype_loopfilter_block_cl(vp8_lf_normal_mb_h_cl);
+extern prototype_loopfilter_block_cl(vp8_lf_normal_b_h_cl);
+extern prototype_loopfilter_block_cl(vp8_lf_simple_mb_v_cl);
+extern prototype_loopfilter_block_cl(vp8_lf_simple_b_v_cl);
+extern prototype_loopfilter_block_cl(vp8_lf_simple_mb_h_cl);
+extern prototype_loopfilter_block_cl(vp8_lf_simple_b_h_cl);
+
+typedef prototype_loopfilter_block_cl((*vp8_lf_block_cl_fn_t));
+
+#endif

vp8/common/opencl/loopfilter_filters_cl.c

@@ -0,0 +1,187 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#include <stdlib.h>
+
+#include <stdio.h>
+
+#include "vpx_ports/config.h"
+#include "vp8_opencl.h"
+#include "blockd_cl.h"
+
+//#include "loopfilter_cl.h"
+//#include "../onyxc_int.h"
+
+typedef unsigned char uc;
+
+static void vp8_loop_filter_cl_run(
+    cl_command_queue cq,
+    cl_kernel kernel,
+    cl_mem buf_mem,
+    int s_off,
+    int p,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count,
+    int block_cnt
+){
+    size_t global[] = {count,block_cnt};
+    int err;
+
+    cl_mem flimit_mem;
+    cl_mem limit_mem;
+    cl_mem thresh_mem;
+
+    VP8_CL_CREATE_BUF(cq, flimit_mem, , sizeof(uc)*16, flimit,, );
+    VP8_CL_CREATE_BUF(cq, limit_mem, , sizeof(uc)*16, limit,, );
+    VP8_CL_CREATE_BUF(cq, thresh_mem, , sizeof(uc)*16, thresh,, );
+
+    err = 0;
+    err = clSetKernelArg(kernel, 0, sizeof (cl_mem), &buf_mem);
+    err |= clSetKernelArg(kernel, 1, sizeof (cl_int), &s_off);
+    err |= clSetKernelArg(kernel, 2, sizeof (cl_int), &p);
+    err |= clSetKernelArg(kernel, 3, sizeof (cl_mem), &flimit_mem);
+    err |= clSetKernelArg(kernel, 4, sizeof (cl_mem), &limit_mem);
+    err |= clSetKernelArg(kernel, 5, sizeof (cl_mem), &thresh_mem);
+    err |= clSetKernelArg(kernel, 6, sizeof (cl_int), &block_cnt);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",,
+    );
+
+    /* Execute the kernel */
+    err = clEnqueueNDRangeKernel(cq, kernel, 2, NULL, global, NULL , 0, NULL, NULL);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to execute kernel!\n",
+        printf("err = %d\n",err);,
+    );
+
+    clReleaseMemObject(flimit_mem);
+    clReleaseMemObject(limit_mem);
+    clReleaseMemObject(thresh_mem);
+
+    VP8_CL_FINISH(cq);
+}
+
+void vp8_loop_filter_horizontal_edge_cl
+(
+    MACROBLOCKD *x,
+    cl_mem s_base,
+    int s_off,
+    int p, /* pitch */
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count,
+    int block_cnt
+)
+{
+    vp8_loop_filter_cl_run(x->cl_commands,
+        cl_data.vp8_loop_filter_horizontal_edge_kernel, s_base, s_off,
+        p, flimit, limit, thresh, count*8, block_cnt
+    );
+}
+
+void vp8_loop_filter_vertical_edge_cl
+(
+    MACROBLOCKD *x,
+    cl_mem s_base,
+    int s_off,
+    int p,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count,
+    int block_cnt
+)
+{
+    vp8_loop_filter_cl_run(x->cl_commands,
+        cl_data.vp8_loop_filter_vertical_edge_kernel, s_base, s_off,
+        p, flimit, limit, thresh, count*8, block_cnt
+    );
+}
+
+void vp8_mbloop_filter_horizontal_edge_cl
+(
+    MACROBLOCKD *x,
+    cl_mem s_base,
+    int s_off,
+    int p,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count,
+    int block_cnt
+)
+{
+    vp8_loop_filter_cl_run(x->cl_commands,
+        cl_data.vp8_mbloop_filter_horizontal_edge_kernel, s_base, s_off,
+        p, flimit, limit, thresh, count*8, block_cnt
+    );
+}
+
+
+void vp8_mbloop_filter_vertical_edge_cl
+(
+    MACROBLOCKD *x,
+    cl_mem s_base,
+    int s_off,
+    int p,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count,
+    int block_cnt
+)
+{
+    vp8_loop_filter_cl_run(x->cl_commands,
+        cl_data.vp8_mbloop_filter_vertical_edge_kernel, s_base, s_off,
+        p, flimit, limit, thresh, count*8, block_cnt
+    );
+}
+
+void vp8_loop_filter_simple_horizontal_edge_cl
+(
+    MACROBLOCKD *x,
+    cl_mem s_base,
+    int s_off,
+    int p,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count,
+    int block_cnt
+)
+{
+    vp8_loop_filter_cl_run(x->cl_commands,
+        cl_data.vp8_loop_filter_simple_horizontal_edge_kernel, s_base, s_off,
+        p, flimit, limit, thresh, count*8, block_cnt
+    );
+}
+
+void vp8_loop_filter_simple_vertical_edge_cl
+(
+    MACROBLOCKD *x,
+    cl_mem s_base,
+    int s_off,
+    int p,
+    const signed char *flimit,
+    const signed char *limit,
+    const signed char *thresh,
+    int count,
+    int block_cnt
+)
+{
+    vp8_loop_filter_cl_run(x->cl_commands,
+        cl_data.vp8_loop_filter_simple_vertical_edge_kernel, s_base, s_off,
+        p, flimit, limit, thresh, count*8, block_cnt
+    );
+}

vp8/common/opencl/opencl_systemdependent.c

@@ -0,0 +1,41 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_ports/config.h"
+#include "../subpixel.h"
+#include "subpixel_cl.h"
+#include "../onyxc_int.h"
+#include "vp8_opencl.h"
+
+#if HAVE_DLOPEN
+#include "dynamic_cl.h"
+#endif
+
+void vp8_arch_opencl_common_init(VP8_COMMON *ctx)
+{
+
+#if HAVE_DLOPEN
+
+#if WIN32 //Windows .dll has no lib prefix and no extension
+    	cl_loaded = load_cl("OpenCL");
+#else   //But *nix needs full name
+    	cl_loaded = load_cl("libOpenCL.so");
+#endif
+
+        if (cl_loaded == CL_SUCCESS)
+            cl_initialized = cl_common_init();
+        else
+            cl_initialized = VP8_CL_TRIED_BUT_FAILED;
+
+#else //!HAVE_DLOPEN (e.g. Apple)
+        cl_initialized = cl_common_init();
+#endif
+
+}

vp8/common/opencl/reconinter_cl.c

@@ -0,0 +1,641 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+//for the decoder, all subpixel prediction is done in this file.
+//
+//Need to determine some sort of mechanism for easily determining SIXTAP/BILINEAR
+//and what arguments to feed into the kernels. These kernels SHOULD be 2-pass,
+//and ideally there'd be a data structure that determined what static arguments
+//to pass in.
+//
+//Also, the only external functions being called here are the subpixel prediction
+//functions. Hopefully this means no worrying about when to copy data back/forth.
+
+#include "../../../vpx_ports/config.h"
+//#include "../recon.h"
+#include "../subpixel.h"
+//#include "../blockd.h"
+//#include "../reconinter.h"
+#if CONFIG_RUNTIME_CPU_DETECT
+//#include "../onyxc_int.h"
+#endif
+
+#include "vp8_opencl.h"
+#include "filter_cl.h"
+#include "reconinter_cl.h"
+#include "blockd_cl.h"
+
+#include <stdio.h>
+
+/* use this define on systems where unaligned int reads and writes are
+ * not allowed, i.e. ARM architectures
+ */
+/*#define MUST_BE_ALIGNED*/
+
+static const int bbb[4] = {0, 2, 8, 10};
+
+static void vp8_memcpy(
+    unsigned char *src_base,
+    int src_offset,
+    int src_stride,
+    unsigned char *dst_base,
+    int dst_offset,
+    int dst_stride,
+    int num_bytes,
+    int num_iter
+){
+
+    int i,r;
+    unsigned char *src = &src_base[src_offset];
+    unsigned char *dst = &dst_base[dst_offset];
+    src_offset = dst_offset = 0;
+
+    for (r = 0; r < num_iter; r++){
+        for (i = 0; i < num_bytes; i++){
+            src_offset = r*src_stride + i;
+            dst_offset = r*dst_stride + i;
+            dst[dst_offset] = src[src_offset];
+        }
+    }
+}
+
+static void vp8_copy_mem_cl(
+    cl_command_queue cq,
+    cl_mem src_mem,
+    int *src_offsets,
+    int src_stride,
+    cl_mem dst_mem,
+    int *dst_offsets,
+    int dst_stride,
+    int num_bytes,
+    int num_iter,
+    int num_blocks
+){
+
+    int err,block;
+
+#if MEM_COPY_KERNEL
+    size_t global[3] = {num_bytes, num_iter, num_blocks};
+
+    size_t local[3];
+    local[0] = global[0];
+    local[1] = global[1];
+    local[2] = global[2];
+
+    err  = clSetKernelArg(cl_data.vp8_memcpy_kernel, 0, sizeof (cl_mem), &src_mem);
+    err |= clSetKernelArg(cl_data.vp8_memcpy_kernel, 2, sizeof (int), &src_stride);
+    err |= clSetKernelArg(cl_data.vp8_memcpy_kernel, 3, sizeof (cl_mem), &dst_mem);
+    err |= clSetKernelArg(cl_data.vp8_memcpy_kernel, 5, sizeof (int), &dst_stride);
+    err |= clSetKernelArg(cl_data.vp8_memcpy_kernel, 6, sizeof (int), &num_bytes);
+    err |= clSetKernelArg(cl_data.vp8_memcpy_kernel, 7, sizeof (int), &num_iter);
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+        "Error: Failed to set kernel arguments!\n",
+        return,
+    );
+
+    for (block = 0; block < num_blocks; block++){
+
+        /* Set kernel arguments */
+        err = clSetKernelArg(cl_data.vp8_memcpy_kernel, 1, sizeof (int), &src_offsets[block]);
+        err |= clSetKernelArg(cl_data.vp8_memcpy_kernel, 4, sizeof (int), &dst_offsets[block]);
+        VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+            "Error: Failed to set kernel arguments!\n",
+            return,
+        );
+
+        /* Execute the kernel */
+        if (num_bytes * num_iter > cl_data.vp8_memcpy_kernel_size){
+            err = clEnqueueNDRangeKernel( cq, cl_data.vp8_memcpy_kernel, 2, NULL, global, NULL , 0, NULL, NULL);
+        } else {
+            err = clEnqueueNDRangeKernel( cq, cl_data.vp8_memcpy_kernel, 2, NULL, global, local , 0, NULL, NULL);
+        }
+
+        VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS,
+            "Error: Failed to execute kernel!\n",
+            return,
+        );
+    }
+#else
+    int iter;
+    for (block=0; block < num_blocks; block++){
+        for (iter = 0; iter < num_iter; iter++){
+            err = clEnqueueCopyBuffer(cq, src_mem, dst_mem,
+                    src_offsets[block]+iter*src_stride,
+                    dst_offsets[block]+iter*dst_stride,
+                    num_bytes, 0, NULL, NULL
+                  );
+            VP8_CL_CHECK_SUCCESS(cq, err != CL_SUCCESS, "Error copying between buffers\n",
+                    ,
+            );
+        }
+    }
+#endif
+}
+
+static void vp8_build_inter_predictors_b_cl(MACROBLOCKD *x, BLOCKD *d, int pitch)
+{
+    unsigned char *ptr_base = *(d->base_pre);
+    int ptr_offset = d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
+
+    vp8_subpix_cl_fn_t sppf;
+
+    int pre_dist = *d->base_pre - x->pre.buffer_alloc;
+    cl_mem pre_mem = x->pre.buffer_mem;
+    int pre_off = pre_dist+ptr_offset;
+
+    if (d->sixtap_filter == CL_TRUE)
+        sppf = vp8_sixtap_predict4x4_cl;
+    else
+        sppf = vp8_bilinear_predict4x4_cl;
+
+    //ptr_base a.k.a. d->base_pre is the start of the
+    //Macroblock's y_buffer, u_buffer, or v_buffer
+
+    if ( (d->bmi.mv.as_mv.row | d->bmi.mv.as_mv.col) & 7)
+    {
+        sppf(d->cl_commands, ptr_base, pre_mem, pre_off, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, d->predictor_base, d->cl_predictor_mem, d->predictor_offset, pitch);
+    }
+    else
+    {
+        vp8_copy_mem_cl(d->cl_commands, pre_mem, &pre_off, d->pre_stride,d->cl_predictor_mem, &d->predictor_offset,pitch,4,4,1);
+    }
+}
+
+
+static void vp8_build_inter_predictors4b_cl(MACROBLOCKD *x, BLOCKD *d, int pitch)
+{
+    unsigned char *ptr_base = *(d->base_pre);
+    int ptr_offset = d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
+
+    int pre_dist = *d->base_pre - x->pre.buffer_alloc;
+    cl_mem pre_mem = x->pre.buffer_mem;
+    int pre_off = pre_dist + ptr_offset;
+
+    //If there's motion in the bottom 8 subpixels, need to do subpixel prediction
+    if ( (d->bmi.mv.as_mv.row | d->bmi.mv.as_mv.col) & 7)
+    {
+            if (d->sixtap_filter == CL_TRUE)
+                vp8_sixtap_predict8x8_cl(d->cl_commands, ptr_base, pre_mem, pre_off, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, d->predictor_base, d->cl_predictor_mem, d->predictor_offset, pitch);
+            else
+                vp8_bilinear_predict8x8_cl(d->cl_commands, ptr_base, pre_mem, pre_off, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, d->predictor_base, d->cl_predictor_mem, d->predictor_offset, pitch);
+    }
+    //Otherwise copy memory directly from src to dest
+    else
+    {
+        vp8_copy_mem_cl(d->cl_commands, pre_mem, &pre_off, d->pre_stride, d->cl_predictor_mem, &d->predictor_offset, pitch, 8, 8, 1);
+    }
+
+
+}
+
+static void vp8_build_inter_predictors2b_cl(MACROBLOCKD *x, BLOCKD *d, int pitch)
+{
+    unsigned char *ptr_base = *(d->base_pre);
+
+    int ptr_offset = d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
+
+    int pre_dist = *d->base_pre - x->pre.buffer_alloc;
+    cl_mem pre_mem = x->pre.buffer_mem;
+    int pre_off = pre_dist+ptr_offset;
+
+    if ( (d->bmi.mv.as_mv.row | d->bmi.mv.as_mv.col) & 7)
+    {
+        if (d->sixtap_filter == CL_TRUE)
+            vp8_sixtap_predict8x4_cl(d->cl_commands,ptr_base,pre_mem,pre_off, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, d->predictor_base, d->cl_predictor_mem, d->predictor_offset, pitch);
+        else
+            vp8_bilinear_predict8x4_cl(d->cl_commands,ptr_base,pre_mem,pre_off, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, d->predictor_base, d->cl_predictor_mem, d->predictor_offset, pitch);
+    }
+    else
+    {
+        vp8_copy_mem_cl(d->cl_commands, pre_mem, &pre_off, d->pre_stride, d->cl_predictor_mem, &d->predictor_offset, pitch, 8, 4, 1);
+    }
+}
+
+
+void vp8_build_inter_predictors_mbuv_cl(MACROBLOCKD *x)
+{
+    int i;
+
+    vp8_cl_mb_prep(x, PREDICTOR|PRE_BUF);
+
+#if !ONE_CQ_PER_MB
+    VP8_CL_FINISH(x->cl_commands);
+#endif
+
+    if (x->mode_info_context->mbmi.ref_frame != INTRA_FRAME &&
+        x->mode_info_context->mbmi.mode != SPLITMV)
+    {
+
+        unsigned char *pred_base = x->predictor;
+        int upred_offset = 256;
+        int vpred_offset = 320;
+
+        int mv_row = x->block[16].bmi.mv.as_mv.row;
+        int mv_col = x->block[16].bmi.mv.as_mv.col;
+        int offset;
+
+        unsigned char *pre_base = x->pre.buffer_alloc;
+        cl_mem pre_mem = x->pre.buffer_mem;
+        int upre_off = x->pre.u_buffer - pre_base;
+        int vpre_off = x->pre.v_buffer - pre_base;
+        int pre_stride = x->block[16].pre_stride;
+
+        offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
+
+        if ((mv_row | mv_col) & 7)
+        {
+            if (cl_initialized == CL_SUCCESS && x->sixtap_filter == CL_TRUE){
+                vp8_sixtap_predict8x8_cl(x->block[16].cl_commands,pre_base, pre_mem, upre_off+offset, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, upred_offset, 8);
+                vp8_sixtap_predict8x8_cl(x->block[20].cl_commands,pre_base, pre_mem, vpre_off+offset, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, vpred_offset, 8);
+            }
+            else{
+                vp8_bilinear_predict8x8_cl(x->block[16].cl_commands,pre_base, pre_mem, upre_off+offset, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, upred_offset, 8);
+                vp8_bilinear_predict8x8_cl(x->block[20].cl_commands,pre_base, pre_mem, vpre_off+offset, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, vpred_offset, 8);
+            }
+        }
+        else
+        {
+            int pre_offsets[2] = {upre_off+offset, vpre_off+offset};
+            int pred_offsets[2] = {upred_offset,vpred_offset};
+            vp8_copy_mem_cl(x->block[16].cl_commands, pre_mem, pre_offsets, pre_stride, x->cl_predictor_mem, pred_offsets, 8, 8, 8, 2);
+        }
+    }
+    else
+    {
+        // Can probably batch these operations as well, but not tested in decoder
+        // (or at least the test videos I've been using.
+        for (i = 16; i < 24; i += 2)
+        {
+            BLOCKD *d0 = &x->block[i];
+            BLOCKD *d1 = &x->block[i+1];
+            if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
+                vp8_build_inter_predictors2b_cl(x, d0, 8);
+            else
+            {
+                vp8_build_inter_predictors_b_cl(x, d0, 8);
+                vp8_build_inter_predictors_b_cl(x, d1, 8);
+            }
+        }
+    }
+
+#if !ONE_CQ_PER_MB
+    VP8_CL_FINISH(x->block[0].cl_commands);
+    VP8_CL_FINISH(x->block[16].cl_commands);
+    VP8_CL_FINISH(x->block[20].cl_commands);
+#endif
+
+    vp8_cl_mb_finish(x, PREDICTOR);
+}
+
+void vp8_build_inter_predictors_mb_cl(MACROBLOCKD *x)
+{
+    //If CL is running in encoder, need to call following before proceeding.
+    //vp8_cl_mb_prep(x, PRE_BUF);
+
+#if !ONE_CQ_PER_MB
+    VP8_CL_FINISH(x->cl_commands);
+#endif
+
+    if (x->mode_info_context->mbmi.ref_frame != INTRA_FRAME &&
+        x->mode_info_context->mbmi.mode != SPLITMV)
+    {
+        int offset;
+        unsigned char *pred_base = x->predictor;
+        int upred_offset = 256;
+        int vpred_offset = 320;
+
+        int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
+        int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
+        int pre_stride = x->block[0].pre_stride;
+
+        unsigned char *pre_base = x->pre.buffer_alloc;
+        cl_mem pre_mem = x->pre.buffer_mem;
+        int ypre_off = x->pre.y_buffer - pre_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
+        int upre_off = x->pre.u_buffer - pre_base;
+        int vpre_off = x->pre.v_buffer - pre_base;
+
+        if ((mv_row | mv_col) & 7)
+        {
+            if (cl_initialized == CL_SUCCESS && x->sixtap_filter == CL_TRUE){
+                vp8_sixtap_predict16x16_cl(x->block[0].cl_commands, pre_base, pre_mem, ypre_off, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, 0, 16);
+            }
+            else
+                vp8_bilinear_predict16x16_cl(x->block[0].cl_commands, pre_base, pre_mem,  ypre_off, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, 0, 16);
+        }
+        else
+        {
+            //16x16 copy
+            int pred_off = 0;
+            vp8_copy_mem_cl(x->block[0].cl_commands, pre_mem, &ypre_off, pre_stride, x->cl_predictor_mem, &pred_off, 16, 16, 16, 1);
+        }
+
+
+        mv_row = x->block[16].bmi.mv.as_mv.row;
+        mv_col = x->block[16].bmi.mv.as_mv.col;
+        pre_stride >>= 1;
+        offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
+
+        if ((mv_row | mv_col) & 7)
+        {
+            if (x->sixtap_filter == CL_TRUE){
+                vp8_sixtap_predict8x8_cl(x->block[16].cl_commands, pre_base, pre_mem, upre_off+offset, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, upred_offset, 8);
+                vp8_sixtap_predict8x8_cl(x->block[20].cl_commands, pre_base, pre_mem, vpre_off+offset, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, vpred_offset, 8);
+            }
+            else {
+                vp8_bilinear_predict8x8_cl(x->block[16].cl_commands, pre_base, pre_mem, upre_off+offset, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, upred_offset, 8);
+                vp8_bilinear_predict8x8_cl(x->block[20].cl_commands, pre_base, pre_mem, vpre_off+offset, pre_stride, mv_col & 7, mv_row & 7, pred_base, x->cl_predictor_mem, vpred_offset, 8);
+            }
+        }
+        else
+        {
+            int pre_off = upre_off + offset;
+            vp8_copy_mem_cl(x->block[16].cl_commands, pre_mem, &pre_off, pre_stride, x->cl_predictor_mem, &upred_offset, 8, 8, 8, 1);
+            pre_off = vpre_off + offset;
+            vp8_copy_mem_cl(x->block[20].cl_commands, pre_mem, &pre_off, pre_stride, x->cl_predictor_mem, &vpred_offset, 8, 8, 8, 1);
+        }
+    }
+    else
+    {
+        int i;
+
+        if (x->mode_info_context->mbmi.partitioning < 3)
+        {
+            for (i = 0; i < 4; i++)
+            {
+                BLOCKD *d = &x->block[bbb[i]];
+                vp8_build_inter_predictors4b_cl(x, d, 16);
+            }
+        }
+        else
+        {
+            /* This loop can be done in any order... No dependencies.*/
+            /* Also, d0/d1 can be decoded simultaneously */
+            for (i = 0; i < 16; i += 2)
+            {
+                BLOCKD *d0 = &x->block[i];
+                BLOCKD *d1 = &x->block[i+1];
+
+                if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
+                    vp8_build_inter_predictors2b_cl(x, d0, 16);
+                else
+                {
+                    vp8_build_inter_predictors_b_cl(x, d0, 16);
+                    vp8_build_inter_predictors_b_cl(x, d1, 16);
+                }
+            }
+        }
+
+        /* Another case of re-orderable/batchable loop */
+        for (i = 16; i < 24; i += 2)
+        {
+            BLOCKD *d0 = &x->block[i];
+            BLOCKD *d1 = &x->block[i+1];
+
+            if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
+                vp8_build_inter_predictors2b_cl(x, d0, 8);
+            else
+            {
+                vp8_build_inter_predictors_b_cl(x, d0, 8);
+                vp8_build_inter_predictors_b_cl(x, d1, 8);
+            }
+        }
+    }
+
+#if !ONE_CQ_PER_MB
+    VP8_CL_FINISH(x->block[0].cl_commands);
+    VP8_CL_FINISH(x->block[16].cl_commands);
+    VP8_CL_FINISH(x->block[20].cl_commands);
+#endif
+
+    vp8_cl_mb_finish(x, PREDICTOR);
+}
+
+
+/* The following functions are written for skip_recon_mb() to call. Since there is no recon in this
+ * situation, we can write the result directly to dst buffer instead of writing it to predictor
+ * buffer and then copying it to dst buffer.
+ */
+static void vp8_build_inter_predictors_b_s_cl(MACROBLOCKD *x, BLOCKD *d, int dst_offset)
+{
+    unsigned char *ptr_base = *(d->base_pre);
+    int dst_stride = d->dst_stride;
+    int pre_stride = d->pre_stride;
+    int ptr_offset = d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
+    vp8_subpix_cl_fn_t sppf;
+
+    int pre_dist = *d->base_pre - x->pre.buffer_alloc;
+    cl_mem pre_mem = x->pre.buffer_mem;
+    cl_mem dst_mem = x->dst.buffer_mem;
+
+    if (d->sixtap_filter == CL_TRUE){
+        sppf = vp8_sixtap_predict4x4_cl;
+    } else
+        sppf = vp8_bilinear_predict4x4_cl;
+        
+    if ( (d->bmi.mv.as_mv.row | d->bmi.mv.as_mv.col) & 7)
+    {
+        sppf(d->cl_commands, ptr_base, pre_mem, pre_dist+ptr_offset, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, NULL, dst_mem, dst_offset, dst_stride);
+    }
+    else
+    {
+        int pre_off = pre_dist+ptr_offset;
+        vp8_copy_mem_cl(d->cl_commands, pre_mem,&pre_off,pre_stride, dst_mem, &dst_offset,dst_stride,4,4,1);
+    }
+}
+
+
+void vp8_build_inter_predictors_mb_s_cl(MACROBLOCKD *x)
+{
+    cl_mem dst_mem = NULL;
+    cl_mem pre_mem = x->pre.buffer_mem;
+
+    unsigned char *dst_base = x->dst.buffer_alloc;
+    int ydst_off = x->dst.y_buffer - dst_base;
+    int udst_off = x->dst.u_buffer - dst_base;
+    int vdst_off = x->dst.v_buffer - dst_base;
+
+    dst_mem = x->dst.buffer_mem;
+    vp8_cl_mb_prep(x, DST_BUF);
+
+#if !ONE_CQ_PER_MB
+    VP8_CL_FINISH(x->cl_commands);
+#endif
+
+    if (x->mode_info_context->mbmi.mode != SPLITMV)
+    {
+        int offset;
+        unsigned char *pre_base = x->pre.buffer_alloc;
+        int ypre_off = x->pre.y_buffer - pre_base;
+        int upre_off = x->pre.u_buffer - pre_base;
+        int vpre_off = x->pre.v_buffer - pre_base;
+
+        int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
+        int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
+        int pre_stride = x->dst.y_stride;
+
+        int ptr_offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
+
+        if ((mv_row | mv_col) & 7)
+        {
+            if (x->sixtap_filter == CL_TRUE){
+                vp8_sixtap_predict16x16_cl(x->block[0].cl_commands, pre_base, pre_mem, ypre_off+ptr_offset, pre_stride, mv_col & 7, mv_row & 7, dst_base, dst_mem, ydst_off, x->dst.y_stride);
+            }
+            else
+                vp8_bilinear_predict16x16_cl(x->block[0].cl_commands, pre_base, pre_mem, ypre_off+ptr_offset, pre_stride, mv_col & 7, mv_row & 7, dst_base, dst_mem, ydst_off, x->dst.y_stride);
+        }
+        else
+        {
+            int pre_off = ypre_off+ptr_offset;
+            vp8_copy_mem_cl(x->block[0].cl_commands, pre_mem, &pre_off, pre_stride, dst_mem, &ydst_off, x->dst.y_stride, 16, 16, 1);
+        }
+
+        mv_row = x->block[16].bmi.mv.as_mv.row;
+        mv_col = x->block[16].bmi.mv.as_mv.col;
+        pre_stride >>= 1;
+        offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
+
+        if ((mv_row | mv_col) & 7)
+        {
+            if (x->sixtap_filter == CL_TRUE){
+                vp8_sixtap_predict8x8_cl(x->block[16].cl_commands, pre_base, pre_mem, upre_off+offset, pre_stride, mv_col & 7, mv_row & 7, dst_base, dst_mem, udst_off, x->dst.uv_stride);
+                vp8_sixtap_predict8x8_cl(x->block[20].cl_commands, pre_base, pre_mem, vpre_off+offset, pre_stride, mv_col & 7, mv_row & 7, dst_base, dst_mem, vdst_off, x->dst.uv_stride);
+            } else {
+                vp8_bilinear_predict8x8_cl(x->block[16].cl_commands, pre_base, pre_mem, upre_off+offset, pre_stride, mv_col & 7, mv_row & 7, dst_base, dst_mem, udst_off, x->dst.uv_stride);
+                vp8_bilinear_predict8x8_cl(x->block[20].cl_commands, pre_base, pre_mem, vpre_off+offset, pre_stride, mv_col & 7, mv_row & 7, dst_base, dst_mem, vdst_off, x->dst.uv_stride);
+            }
+        }
+        else
+        {
+            int pre_offsets[2] = {upre_off+offset, vpre_off+offset};
+            int dst_offsets[2] = {udst_off,vdst_off};
+            vp8_copy_mem_cl(x->block[16].cl_commands, pre_mem, pre_offsets, pre_stride, dst_mem, dst_offsets, x->dst.uv_stride, 8, 8, 2);
+        }
+
+    }
+    else
+    {
+        /* note: this whole ELSE part is not executed at all. So, no way to test the correctness of my modification. Later,
+         * if sth is wrong, go back to what it is in build_inter_predictors_mb.
+         *
+         * ACW: Not sure who the above comment belongs to, but it is
+         *      accurate for the decoder. Verified by reverse trace of source
+         */
+        int i;
+
+        if (x->mode_info_context->mbmi.partitioning < 3)
+        {
+            for (i = 0; i < 4; i++)
+            {
+                BLOCKD *d = &x->block[bbb[i]];
+
+                {
+                    unsigned char *ptr_base = *(d->base_pre);
+                    int pre_off = ptr_base - x->pre.buffer_alloc;
+                    
+                    int ptr_offset = d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
+
+                    pre_off += ptr_offset;
+
+                    if ( (d->bmi.mv.as_mv.row | d->bmi.mv.as_mv.col) & 7)
+                    {
+                        if (x->sixtap_filter == CL_TRUE)
+                            vp8_sixtap_predict8x8_cl(d->cl_commands, ptr_base, pre_mem, pre_off, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst_base, dst_mem, ydst_off, x->dst.y_stride);
+                        else
+                            vp8_bilinear_predict8x8_cl(d->cl_commands, ptr_base, pre_mem, pre_off, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst_base, dst_mem, ydst_off, x->dst.y_stride);
+                    }
+                    else
+                    {
+                        vp8_copy_mem_cl(x->block[0].cl_commands, pre_mem, &pre_off, d->pre_stride, dst_mem, &ydst_off, x->dst.y_stride, 8, 8, 1);
+                    }
+                }
+            }
+        }
+        else
+        {
+            for (i = 0; i < 16; i += 2)
+            {
+                BLOCKD *d0 = &x->block[i];
+                BLOCKD *d1 = &x->block[i+1];
+
+                if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
+                {
+                    /*vp8_build_inter_predictors2b(x, d0, 16);*/
+                    unsigned char *ptr_base = *(d0->base_pre);
+
+                    int pre_off = ptr_base - x->pre.buffer_alloc;
+
+                    int ptr_offset = d0->pre + (d0->bmi.mv.as_mv.row >> 3) * d0->pre_stride + (d0->bmi.mv.as_mv.col >> 3);
+                    pre_off += ptr_offset;
+
+                    if ( (d0->bmi.mv.as_mv.row | d0->bmi.mv.as_mv.col) & 7)
+                    {
+                        if (d0->sixtap_filter == CL_TRUE)
+                            vp8_sixtap_predict8x4_cl(d0->cl_commands, ptr_base, pre_mem, pre_off, d0->pre_stride, d0->bmi.mv.as_mv.col & 7, d0->bmi.mv.as_mv.row & 7, dst_base, dst_mem, ydst_off, x->dst.y_stride);
+                        else
+                            vp8_bilinear_predict8x4_cl(d0->cl_commands, ptr_base, pre_mem,pre_off, d0->pre_stride, d0->bmi.mv.as_mv.col & 7, d0->bmi.mv.as_mv.row & 7, dst_base, dst_mem, ydst_off, x->dst.y_stride);
+                    }
+                    else
+                    {
+                        vp8_copy_mem_cl(x->block[0].cl_commands, pre_mem, &pre_off, d0->pre_stride, dst_mem, &ydst_off, x->dst.y_stride, 8, 4, 1);
+                    }
+                }
+                else
+                {
+                    vp8_build_inter_predictors_b_s_cl(x,d0, ydst_off);
+                    vp8_build_inter_predictors_b_s_cl(x,d1, ydst_off);
+                }
+            }
+        }
+
+        for (i = 16; i < 24; i += 2)
+        {
+            BLOCKD *d0 = &x->block[i];
+            BLOCKD *d1 = &x->block[i+1];
+
+            if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
+            {
+                /*vp8_build_inter_predictors2b(x, d0, 8);*/
+                unsigned char *ptr_base = *(d0->base_pre);
+                int ptr_offset = d0->pre + (d0->bmi.mv.as_mv.row >> 3) * d0->pre_stride + (d0->bmi.mv.as_mv.col >> 3);
+                int pre_off = ptr_base - x->pre.buffer_alloc + ptr_offset;
+
+                if ( (d0->bmi.mv.as_mv.row | d0->bmi.mv.as_mv.col) & 7)
+                {
+                    if (d0->sixtap_filter || CL_TRUE)
+                        vp8_sixtap_predict8x4_cl(d0->cl_commands, ptr_base, pre_mem, pre_off, d0->pre_stride,
+                            d0->bmi.mv.as_mv.col & 7, d0->bmi.mv.as_mv.row & 7,
+                            dst_base, dst_mem, ydst_off, x->dst.uv_stride);
+                    else
+                        vp8_bilinear_predict8x4_cl(d0->cl_commands, ptr_base, pre_mem, pre_off, d0->pre_stride,
+                            d0->bmi.mv.as_mv.col & 7, d0->bmi.mv.as_mv.row & 7,
+                            dst_base, dst_mem, ydst_off, x->dst.uv_stride);
+                }
+                else
+                {
+                    vp8_copy_mem_cl(x->block[0].cl_commands, pre_mem, &pre_off,
+                        d0->pre_stride, dst_mem, &ydst_off, x->dst.uv_stride, 8, 4, 1);
+                }
+            }
+            else
+            {
+                vp8_build_inter_predictors_b_s_cl(x,d0, ydst_off);
+                vp8_build_inter_predictors_b_s_cl(x,d1, ydst_off);
+            }
+        } //end for
+    }
+
+#if !ONE_CQ_PER_MB
+    VP8_CL_FINISH(x->block[0].cl_commands);
+    VP8_CL_FINISH(x->block[16].cl_commands);
+    VP8_CL_FINISH(x->block[20].cl_commands);
+#endif
+
+    vp8_cl_mb_finish(x, DST_BUF);
+}

vp8/common/opencl/reconinter_cl.h

@@ -0,0 +1,25 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef __INC_RECONINTER_CL_H
+#define __INC_RECONINTER_CL_H
+
+#include "blockd_cl.h"
+#include "subpixel_cl.h"
+#include "filter_cl.h"
+
+extern void vp8_build_inter_predictors_mb_cl(MACROBLOCKD *x);
+extern void vp8_build_inter_predictors_mbuv_cl(MACROBLOCKD *x);
+
+extern void vp8_build_inter_predictors_mb_s_cl(MACROBLOCKD *x);
+//extern void vp8_build_inter_predictors_b_cl(BLOCKD *d, int pitch);
+
+#endif

vp8/common/opencl/subpixel_cl.h

@@ -0,0 +1,46 @@
+/*
+ *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+#ifndef SUBPIXEL_CL_H
+#define SUBPIXEL_CL_H
+
+#include "../blockd.h"
+
+/* Note:
+ *
+ * This platform is commonly built for runtime CPU detection. If you modify
+ * any of the function mappings present in this file, be sure to also update
+ * them in the function pointer initialization code
+ */
+
+#define prototype_subpixel_predict_cl(sym) \
+    void sym(cl_command_queue cq, unsigned char *src_base, cl_mem src_mem, int src_offset, \
+            int src_pitch, int xofst, int yofst, \
+             unsigned char *dst_base, cl_mem dst_mem, int dst_offset, int dst_pitch)
+
+extern prototype_subpixel_predict_cl(vp8_sixtap_predict16x16_cl);
+extern prototype_subpixel_predict_cl(vp8_sixtap_predict8x8_cl);
+extern prototype_subpixel_predict_cl(vp8_sixtap_predict8x4_cl);
+extern prototype_subpixel_predict_cl(vp8_sixtap_predict4x4_cl);
+extern prototype_subpixel_predict_cl(vp8_bilinear_predict16x16_cl);
+extern prototype_subpixel_predict_cl(vp8_bilinear_predict8x8_cl);
+extern prototype_subpixel_predict_cl(vp8_bilinear_predict8x4_cl);
+extern prototype_subpixel_predict_cl(vp8_bilinear_predict4x4_cl);
+
+typedef prototype_subpixel_predict_cl((*vp8_subpix_cl_fn_t));
+
+//typedef enum
+//{
+//    SIXTAP = 0,
+//    BILINEAR = 1
+//} SUBPIX_TYPE;
+
+#endif

vp8/common/opencl/vp8_opencl.c

@@ -0,0 +1,342 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include "vp8_opencl.h"
+
+int cl_initialized = VP8_CL_NOT_INITIALIZED;
+VP8_COMMON_CL cl_data;
+
+//Initialization functions for various CL programs.
+extern int cl_init_filter();
+extern int cl_init_idct();
+extern int cl_init_loop_filter();
+
+//Common CL destructors
+extern void cl_destroy_loop_filter();
+extern void cl_destroy_filter();
+extern void cl_destroy_idct();
+
+//Destructors for encoder/decoder-specific bits
+extern void cl_decode_destroy();
+extern void cl_encode_destroy();
+
+/**
+ * 
+ * @param cq
+ * @param new_status
+ */
+void cl_destroy(cl_command_queue cq, int new_status) {
+
+    if (cl_initialized != CL_SUCCESS)
+        return;
+
+    //Wait on any pending operations to complete... frees up all of our pointers
+    if (cq != NULL)
+        clFinish(cq);
+
+#if ENABLE_CL_SUBPIXEL
+    //Release the objects that we've allocated on the GPU
+    cl_destroy_filter();
+#endif
+
+#if ENABLE_CL_IDCT_DEQUANT
+    cl_destroy_idct();
+
+#if CONFIG_VP8_DECODER
+    if (cl_data.cl_decode_initialized == CL_SUCCESS)
+        cl_decode_destroy();
+#endif
+
+#endif
+#if ENABLE_CL_LOOPFILTER
+    cl_destroy_loop_filter();
+#endif
+
+
+#if CONFIG_VP8_ENCODER
+    //placeholder for if/when encoder CL gets implemented
+#endif
+
+    if (cq){
+        clReleaseCommandQueue(cq);
+    }
+
+    if (cl_data.context){
+        clReleaseContext(cl_data.context);
+        cl_data.context = NULL;
+    }
+
+    cl_initialized = new_status;
+
+    return;
+}
+
+/**
+ * 
+ * @param dev
+ * @return
+ */
+cl_device_type device_type(cl_device_id dev){
+    cl_device_type type;
+    int err;
+
+    err = clGetDeviceInfo(dev, CL_DEVICE_TYPE, sizeof(type),&type,NULL);
+    if (err != CL_SUCCESS)
+        return CL_INVALID_DEVICE;
+    return type;
+}
+
+/**
+ * 
+ * @return
+ */
+int cl_common_init() {
+    int err,i,dev;
+    cl_platform_id platform_ids[MAX_NUM_PLATFORMS];
+    cl_uint num_found, num_devices;
+    cl_device_id devices[MAX_NUM_DEVICES];
+
+    //Don't allow multiple CL contexts..
+    if (cl_initialized != VP8_CL_NOT_INITIALIZED)
+        return cl_initialized;
+
+    // Connect to a compute device
+    err = clGetPlatformIDs(MAX_NUM_PLATFORMS, platform_ids, &num_found);
+
+    if (err != CL_SUCCESS) {
+        fprintf(stderr, "Couldn't query platform IDs\n");
+        return VP8_CL_TRIED_BUT_FAILED;
+    }
+
+    if (num_found == 0) {
+        fprintf(stderr, "No platforms found\n");
+        return VP8_CL_TRIED_BUT_FAILED;
+    }
+
+    //printf("Enumerating %d platform(s)\n", num_found);
+    //Enumerate the platforms found
+    for (i = 0; i < num_found; i++){
+    	char buf[2048];
+        size_t len;
+        
+    	err = clGetPlatformInfo( platform_ids[i], CL_PLATFORM_VENDOR, sizeof(buf), buf, &len);
+    	if (err != CL_SUCCESS){
+            fprintf(stderr, "Error retrieving platform vendor for platform %d",i);
+            continue;
+    	}
+    	//printf("Platform %d: %s\n",i,buf);
+
+        //If you need to force a platform (e.g. CPU-only testing), uncomment this
+        //if (strstr(buf,"NVIDIA"))
+        //    continue;
+
+    	//Try to find a valid compute device
+    	//Favor the GPU, but fall back to any other available device if necessary
+#ifdef __APPLE__
+    	printf("Apple system. Running CL as CPU-only for now...\n");
+        err = clGetDeviceIDs(platform_ids[i], CL_DEVICE_TYPE_CPU, MAX_NUM_DEVICES, devices, &num_devices);
+#else
+        err = clGetDeviceIDs(platform_ids[i], CL_DEVICE_TYPE_ALL, MAX_NUM_DEVICES, devices, &num_devices);
+#endif //__APPLE__
+        //printf("found %d devices\n", num_devices);
+        cl_data.device_id = NULL;
+        for( dev = 0; dev < num_devices; dev++ ){
+            char ext[2048];
+            //Get info for this device.
+            err = clGetDeviceInfo(devices[dev], CL_DEVICE_EXTENSIONS,
+                    sizeof(ext),ext,NULL);
+            VP8_CL_CHECK_SUCCESS(NULL,err != CL_SUCCESS,
+                    "Error retrieving device extension list",continue, 0);
+            //printf("Device %d supports: %s\n",dev,ext);
+            
+            //The kernels in VP8 require byte-addressable stores, which is an
+            //extension. It's required in OpenCL 1.1, but not all devices
+            //support it.
+            if (strstr(ext,"cl_khr_byte_addressable_store")){
+                //We found a valid device, so use it. But if we find a GPU
+                //(maybe this is one), prefer that.
+                cl_data.device_id = devices[dev];
+
+                if ( device_type(devices[dev]) == CL_DEVICE_TYPE_GPU ){
+                    //printf("Device %d is a GPU\n",dev);
+                    break;
+                }
+            }
+        }
+
+        //If we've found a usable GPU, stop looking.
+        if (cl_data.device_id != NULL && device_type(cl_data.device_id) == CL_DEVICE_TYPE_GPU )
+            break;
+
+    }
+
+    if (cl_data.device_id == NULL){
+    	printf("Error: Failed to find a valid OpenCL device. Using CPU paths\n");
+    	return VP8_CL_TRIED_BUT_FAILED;
+    }
+
+    // Create the compute context
+    cl_data.context = clCreateContext(0, 1, &cl_data.device_id, NULL, NULL, &err);
+    if (!cl_data.context) {
+        printf("Error: Failed to create a compute context!\n");
+        return VP8_CL_TRIED_BUT_FAILED;
+    }
+
+    //Initialize programs to null value
+    //Enables detection of if they've been initialized as well.
+    cl_data.filter_program = NULL;
+    cl_data.idct_program = NULL;
+    cl_data.loop_filter_program = NULL;
+
+#if ENABLE_CL_SUBPIXEL
+    err = cl_init_filter();
+    if (err != CL_SUCCESS)
+        return err;
+#endif
+
+#if ENABLE_CL_IDCT_DEQUANT
+    err = cl_init_idct();
+    if (err != CL_SUCCESS)
+        return err;
+#endif
+
+#if ENABLE_CL_LOOPFILTER
+
+    err = cl_init_loop_filter();
+    if (err != CL_SUCCESS)
+        return err;
+#endif
+
+    return CL_SUCCESS;
+}
+
+char *cl_read_file(const char* file_name) {
+    long pos;
+    char *bytes;
+    size_t amt_read;
+    FILE *f;
+
+    f = fopen(file_name, "rb");
+    
+    if (f == NULL) {
+        char *fullpath;
+        //printf("Couldn't find %s\n", file_name);
+
+        //Generate a file path for the CL sources using the library install dir
+        fullpath = malloc(strlen(vpx_codec_lib_dir()) + strlen(file_name) + 2);
+        if (fullpath == NULL) {
+           return NULL;
+        }
+        strcpy(fullpath, vpx_codec_lib_dir());
+        strcat(fullpath, "/"); //Will need to be changed for MSVS
+        strcat(fullpath, file_name);
+
+        //printf("Looking in %s\n", fullpath);
+
+        f = fopen(fullpath, "rb");
+        if (f == NULL) {
+            fprintf(stderr,"Couldn't find CL source at %s or %s\n", file_name, fullpath);
+            free(fullpath);
+            return NULL;
+        }
+
+        //printf("Found cl source at %s\n", fullpath);
+        free(fullpath);
+    } else {
+        //printf("Found cl source at %s\n", file_name);
+    }
+
+    fseek(f, 0, SEEK_END);
+    pos = ftell(f);
+    fseek(f, 0, SEEK_SET);
+    bytes = malloc(pos+1);
+
+    if (bytes == NULL) {
+        fclose(f);
+        return NULL;
+    }
+
+    amt_read = fread(bytes, pos, 1, f);
+    if (amt_read != 1) {
+        free(bytes);
+        fclose(f);
+        return NULL;
+    }
+
+    bytes[pos] = '\0'; //null terminate the source string
+    fclose(f);
+
+
+    return bytes;
+}
+
+void show_build_log(cl_program *prog_ref){
+    size_t len;
+    char *buffer;
+    int err = clGetProgramBuildInfo(*prog_ref, cl_data.device_id, CL_PROGRAM_BUILD_LOG, 0, NULL, &len);
+
+    if (err != CL_SUCCESS){
+        printf("Error: Could not get length of CL build log\n");
+    }
+
+    buffer = (char*) malloc(len);
+    if (buffer == NULL) {
+        printf("Error: Couldn't allocate compile output buffer memory\n");
+    }
+
+    err = clGetProgramBuildInfo(*prog_ref, cl_data.device_id, CL_PROGRAM_BUILD_LOG, len, buffer, NULL);
+    if (err != CL_SUCCESS) {
+        printf("Error: Could not get CL build log\n");
+
+    } else {
+        printf("Compile output: %s\n", buffer);
+    }
+    free(buffer);
+}
+
+int cl_load_program(cl_program *prog_ref, const char *file_name, const char *opts) {
+
+    int err;
+    char *kernel_src = cl_read_file(file_name);
+    
+    *prog_ref = NULL;
+    if (kernel_src != NULL) {
+        *prog_ref = clCreateProgramWithSource(cl_data.context, 1, (const char**)&kernel_src, NULL, &err);
+        free(kernel_src);
+    } else {
+        cl_destroy(NULL, VP8_CL_TRIED_BUT_FAILED);
+        printf("Couldn't find OpenCL source files. \nUsing software path.\n");
+        return VP8_CL_TRIED_BUT_FAILED;
+    }
+
+    if (*prog_ref == NULL) {
+        printf("Error: Couldn't create program\n");
+        return VP8_CL_TRIED_BUT_FAILED;
+    }
+
+    if (err != CL_SUCCESS) {
+        printf("Error creating program: %d\n", err);
+    }
+
+    /* Build the program executable */
+    err = clBuildProgram(*prog_ref, 0, NULL, opts, NULL, NULL);
+    if (err != CL_SUCCESS) {
+        printf("Error: Failed to build program executable for %s!\n", file_name);
+
+        show_build_log(prog_ref);
+
+        return VP8_CL_TRIED_BUT_FAILED;
+    }
+
+    return CL_SUCCESS;
+}

vp8/common/opencl/vp8_opencl.h

@@ -0,0 +1,192 @@
+/*
+ *  Copyright (c) 2011 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP8_OPENCL_H
+#define	VP8_OPENCL_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+#include "../../../vpx_config.h"
+
+#ifdef __APPLE__
+#include <OpenCL/cl.h>
+#else
+#include <CL/cl.h>
+#endif
+
+#if HAVE_DLOPEN
+#include "dynamic_cl.h"
+#endif
+
+#define ENABLE_CL_IDCT_DEQUANT 0
+#define ENABLE_CL_SUBPIXEL 1
+#define TWO_PASS_SIXTAP 0
+#define MEM_COPY_KERNEL 1
+#define ONE_CQ_PER_MB 1 //Value of 0 is racey... still experimental.
+#define ENABLE_CL_LOOPFILTER 0
+
+extern char *cl_read_file(const char* file_name);
+extern int cl_common_init();
+extern void cl_destroy(cl_command_queue cq, int new_status);
+extern int cl_load_program(cl_program *prog_ref, const char *file_name, const char *opts);
+
+#define MAX_NUM_PLATFORMS 4
+#define MAX_NUM_DEVICES 10
+
+#define VP8_CL_TRIED_BUT_FAILED 1
+#define VP8_CL_NOT_INITIALIZED -1
+extern int cl_initialized;
+
+extern const char *vpx_codec_lib_dir(void);
+
+#define VP8_CL_FINISH(cq) \
+    if (cl_initialized == CL_SUCCESS){ \
+        /* Wait for kernels to finish. */ \
+        clFinish(cq); \
+    }
+
+#define VP8_CL_BARRIER(cq) \
+    if (cl_initialized == CL_SUCCESS){ \
+        /* Insert a barrier into the command queue. */ \
+        clEnqueueBarrier(cq); \
+    }
+
+#define VP8_CL_CHECK_SUCCESS(cq,cond,msg,alt,retCode) \
+    if ( cond ){ \
+        fprintf(stderr, msg);  \
+        cl_destroy(cq, VP8_CL_TRIED_BUT_FAILED); \
+        alt; \
+        return retCode; \
+    }
+
+#define VP8_CL_CALC_LOCAL_SIZE(kernel, kernel_size) \
+    err = clGetKernelWorkGroupInfo( cl_data.kernel, \
+  	cl_data.device_id, \
+  	CL_KERNEL_WORK_GROUP_SIZE, \
+  	sizeof(size_t), \
+  	&cl_data.kernel_size, \
+  	NULL);\
+    VP8_CL_CHECK_SUCCESS(NULL, err != CL_SUCCESS, \
+        "Error: Failed to calculate local size of kernel!\n", \
+        ,\
+        VP8_CL_TRIED_BUT_FAILED \
+    ); \
+
+#define VP8_CL_CREATE_KERNEL(data,program,name,str_name) \
+    data.name = clCreateKernel(data.program, str_name , &err); \
+    VP8_CL_CHECK_SUCCESS(NULL, err != CL_SUCCESS || !data.name, \
+        "Error: Failed to create compute kernel "#str_name"!\n", \
+        ,\
+        VP8_CL_TRIED_BUT_FAILED \
+    );
+
+#define VP8_CL_READ_BUF(cq, bufRef, bufSize, dstPtr) \
+    err = clEnqueueReadBuffer(cq, bufRef, CL_FALSE, 0, bufSize , dstPtr, 0, NULL, NULL); \
+    VP8_CL_CHECK_SUCCESS( cq, err != CL_SUCCESS, \
+        "Error: Failed to read from GPU!\n",, err \
+    ); \
+
+#define VP8_CL_SET_BUF(cq, bufRef, bufSize, dataPtr, altPath, retCode) \
+    { \
+        err = clEnqueueWriteBuffer(cq, bufRef, CL_FALSE, 0, \
+            bufSize, dataPtr, 0, NULL, NULL); \
+        \
+        VP8_CL_CHECK_SUCCESS(cq, err != CL_SUCCESS, \
+            "Error: Failed to write to buffer!\n", \
+            altPath, retCode\
+        ); \
+    } \
+
+#define VP8_CL_CREATE_BUF(cq, bufRef, bufType, bufSize, dataPtr, altPath, retCode) \
+    bufRef = clCreateBuffer(cl_data.context, CL_MEM_READ_WRITE, bufSize, NULL, NULL); \
+    if (dataPtr != NULL && bufRef != NULL){ \
+        VP8_CL_SET_BUF(cq, bufRef, bufSize, dataPtr, altPath, retCode)\
+    } \
+    VP8_CL_CHECK_SUCCESS(cq, !bufRef, \
+        "Error: Failed to allocate buffer. Using CPU path!\n", \
+        altPath, retCode\
+    ); \
+
+#define VP8_CL_RELEASE_KERNEL(kernel) \
+    if (kernel) \
+        clReleaseKernel(kernel); \
+    kernel = NULL;
+
+typedef struct VP8_COMMON_CL {
+    cl_device_id device_id; // compute device id
+    cl_context context; // compute context
+    //cl_command_queue commands; // compute command queue
+
+    cl_program filter_program; // compute program for subpixel/bilinear filters
+    cl_kernel vp8_sixtap_predict_kernel;
+    size_t    vp8_sixtap_predict_kernel_size;
+    cl_kernel vp8_sixtap_predict8x4_kernel;
+    size_t    vp8_sixtap_predict8x4_kernel_size;
+    cl_kernel vp8_sixtap_predict8x8_kernel;
+    size_t    vp8_sixtap_predict8x8_kernel_size;
+    cl_kernel vp8_sixtap_predict16x16_kernel;
+    size_t    vp8_sixtap_predict16x16_kernel_size;
+
+    cl_kernel vp8_bilinear_predict4x4_kernel;
+    cl_kernel vp8_bilinear_predict8x4_kernel;
+    cl_kernel vp8_bilinear_predict8x8_kernel;
+    cl_kernel vp8_bilinear_predict16x16_kernel;
+
+    cl_kernel vp8_filter_block2d_first_pass_kernel;
+    size_t    vp8_filter_block2d_first_pass_kernel_size;
+    cl_kernel vp8_filter_block2d_second_pass_kernel;
+    size_t    vp8_filter_block2d_second_pass_kernel_size;
+
+    cl_kernel vp8_filter_block2d_bil_first_pass_kernel;
+    size_t    vp8_filter_block2d_bil_first_pass_kernel_size;
+    cl_kernel vp8_filter_block2d_bil_second_pass_kernel;
+    size_t    vp8_filter_block2d_bil_second_pass_kernel_size;
+
+    cl_kernel vp8_memcpy_kernel;
+    size_t    vp8_memcpy_kernel_size;
+    cl_kernel vp8_memset_short_kernel;
+
+    cl_program idct_program;
+    cl_kernel vp8_short_inv_walsh4x4_1_kernel;
+    cl_kernel vp8_short_inv_walsh4x4_1st_pass_kernel;
+    cl_kernel vp8_short_inv_walsh4x4_2nd_pass_kernel;
+    cl_kernel vp8_dc_only_idct_add_kernel;
+    //Note that the following 2 kernels are encoder-only. Not used in decoder.
+    cl_kernel vp8_short_idct4x4llm_1_kernel;
+    cl_kernel vp8_short_idct4x4llm_kernel;
+
+    cl_program loop_filter_program;
+    cl_kernel vp8_loop_filter_horizontal_edge_kernel;
+    cl_kernel vp8_loop_filter_vertical_edge_kernel;
+    cl_kernel vp8_mbloop_filter_horizontal_edge_kernel;
+    cl_kernel vp8_mbloop_filter_vertical_edge_kernel;
+    cl_kernel vp8_loop_filter_simple_horizontal_edge_kernel;
+    cl_kernel vp8_loop_filter_simple_vertical_edge_kernel;
+
+    cl_program dequant_program;
+    cl_kernel vp8_dequant_dc_idct_add_kernel;
+    cl_kernel vp8_dequant_idct_add_kernel;
+    cl_kernel vp8_dequantize_b_kernel;
+
+    cl_int cl_decode_initialized;
+    cl_int cl_encode_initialized;
+    
+} VP8_COMMON_CL;
+
+extern VP8_COMMON_CL cl_data;
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* VP8_OPENCL_H */
+

vp8/common/postproc.c

@@ -804,14 +804,11 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
             for (j = 0; j < mb_cols; j++)
             {
                 char zz[4];
-                int dc_diff = !(mi[mb_index].mbmi.mode != B_PRED &&
-                              mi[mb_index].mbmi.mode != SPLITMV &&
-                              mi[mb_index].mbmi.mb_skip_coeff);
 
                 if (oci->frame_type == KEY_FRAME)
                     sprintf(zz, "a");
                 else
-                    sprintf(zz, "%c", dc_diff + '0');
+                    sprintf(zz, "%c", mi[mb_index].mbmi.dc_diff + '0');
 
                 vp8_blit_text(zz, y_ptr, post->y_stride);
                 mb_index ++;
@@ -837,6 +834,7 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
         YV12_BUFFER_CONFIG *post = &oci->post_proc_buffer;
         int width  = post->y_width;
         int height = post->y_height;
+        int mb_cols = width  >> 4;
         unsigned char *y_buffer = oci->post_proc_buffer.y_buffer;
         int y_stride = oci->post_proc_buffer.y_stride;
         MODE_INFO *mi = oci->mi;
@@ -860,7 +858,7 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
                     {
                         case 0 :    /* mv_top_bottom */
                         {
-                            union b_mode_info *bmi = &mi->bmi[0];
+                            B_MODE_INFO *bmi = &mi->bmi[0];
                             MV *mv = &bmi->mv.as_mv;
 
                             x1 = x0 + 8 + (mv->col >> 3);
@@ -881,7 +879,7 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
                         }
                         case 1 :    /* mv_left_right */
                         {
-                            union b_mode_info *bmi = &mi->bmi[0];
+                            B_MODE_INFO *bmi = &mi->bmi[0];
                             MV *mv = &bmi->mv.as_mv;
 
                             x1 = x0 + 4 + (mv->col >> 3);
@@ -902,7 +900,7 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
                         }
                         case 2 :    /* mv_quarters   */
                         {
-                            union b_mode_info *bmi = &mi->bmi[0];
+                            B_MODE_INFO *bmi = &mi->bmi[0];
                             MV *mv = &bmi->mv.as_mv;
 
                             x1 = x0 + 4 + (mv->col >> 3);
@@ -938,7 +936,7 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
                         }
                         default :
                         {
-                            union b_mode_info *bmi = mi->bmi;
+                            B_MODE_INFO *bmi = mi->bmi;
                             int bx0, by0;
 
                             for (by0 = y0; by0 < (y0+16); by0 += 4)
@@ -1011,7 +1009,7 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
                 {
                     int by, bx;
                     unsigned char *yl, *ul, *vl;
-                    union b_mode_info *bmi = mi->bmi;
+                    B_MODE_INFO *bmi = mi->bmi;
 
                     yl = y_ptr + x;
                     ul = u_ptr + (x>>1);
@@ -1024,9 +1022,9 @@ int vp8_post_proc_frame(VP8_COMMON *oci, YV12_BUFFER_CONFIG *dest, vp8_ppflags_t
                             if ((ppflags->display_b_modes_flag & (1<<mi->mbmi.mode))
                                 || (ppflags->display_mb_modes_flag & B_PRED))
                             {
-                                Y = B_PREDICTION_MODE_colors[bmi->as_mode][0];
-                                U = B_PREDICTION_MODE_colors[bmi->as_mode][1];
-                                V = B_PREDICTION_MODE_colors[bmi->as_mode][2];
+                                Y = B_PREDICTION_MODE_colors[bmi->mode][0];
+                                U = B_PREDICTION_MODE_colors[bmi->mode][1];
+                                V = B_PREDICTION_MODE_colors[bmi->mode][2];
 
                                 POSTPROC_INVOKE(RTCD_VTABLE(oci), blend_b)
                                     (yl+bx, ul+(bx>>1), vl+(bx>>1), Y, U, V, 0xc000, y_stride);

vp8/common/ppc/loopfilter_altivec.c

@@ -53,8 +53,9 @@ loop_filter_function_s_ppc loop_filter_simple_vertical_edge_ppc;
 
 // Horizontal MB filtering
 void loop_filter_mbh_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                         int y_stride, int uv_stride, loop_filter_info *lfi)
+                         int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
+    (void)simpler_lpf;
     mbloop_filter_horizontal_edge_y_ppc(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr);
 
     if (u_ptr)
@@ -62,8 +63,9 @@ void loop_filter_mbh_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned ch
 }
 
 void loop_filter_mbhs_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                          int y_stride, int uv_stride, loop_filter_info *lfi)
+                          int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
+    (void)simpler_lpf;
     (void)u_ptr;
     (void)v_ptr;
     (void)uv_stride;
@@ -72,8 +74,9 @@ void loop_filter_mbhs_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned c
 
 // Vertical MB Filtering
 void loop_filter_mbv_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                         int y_stride, int uv_stride, loop_filter_info *lfi)
+                         int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
+    (void)simpler_lpf;
     mbloop_filter_vertical_edge_y_ppc(y_ptr, y_stride, lfi->mbflim, lfi->lim, lfi->thr);
 
     if (u_ptr)
@@ -81,8 +84,9 @@ void loop_filter_mbv_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned ch
 }
 
 void loop_filter_mbvs_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                          int y_stride, int uv_stride, loop_filter_info *lfi)
+                          int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
+    (void)simpler_lpf;
     (void)u_ptr;
     (void)v_ptr;
     (void)uv_stride;
@@ -91,8 +95,9 @@ void loop_filter_mbvs_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned c
 
 // Horizontal B Filtering
 void loop_filter_bh_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                        int y_stride, int uv_stride, loop_filter_info *lfi)
+                        int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
+    (void)simpler_lpf;
     // These should all be done at once with one call, instead of 3
     loop_filter_horizontal_edge_y_ppc(y_ptr + 4 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr);
     loop_filter_horizontal_edge_y_ppc(y_ptr + 8 * y_stride, y_stride, lfi->flim, lfi->lim, lfi->thr);
@@ -103,8 +108,9 @@ void loop_filter_bh_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned cha
 }
 
 void loop_filter_bhs_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                         int y_stride, int uv_stride, loop_filter_info *lfi)
+                         int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
+    (void)simpler_lpf;
     (void)u_ptr;
     (void)v_ptr;
     (void)uv_stride;
@@ -115,8 +121,9 @@ void loop_filter_bhs_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned ch
 
 // Vertical B Filtering
 void loop_filter_bv_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                        int y_stride, int uv_stride, loop_filter_info *lfi)
+                        int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
+    (void)simpler_lpf;
     loop_filter_vertical_edge_y_ppc(y_ptr, y_stride, lfi->flim, lfi->lim, lfi->thr);
 
     if (u_ptr)
@@ -124,8 +131,9 @@ void loop_filter_bv_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned cha
 }
 
 void loop_filter_bvs_ppc(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
-                         int y_stride, int uv_stride, loop_filter_info *lfi)
+                         int y_stride, int uv_stride, loop_filter_info *lfi, int simpler_lpf)
 {
+    (void)simpler_lpf;
     (void)u_ptr;
     (void)v_ptr;
     (void)uv_stride;

vp8/common/quant_common.c

@@ -66,6 +66,7 @@ int vp8_dc2quant(int QIndex, int Delta)
     return retval;
 
 }
+
 int vp8_dc_uv_quant(int QIndex, int Delta)
 {
     int retval;
@@ -116,6 +117,7 @@ int vp8_ac2quant(int QIndex, int Delta)
 
     return retval;
 }
+
 int vp8_ac_uv_quant(int QIndex, int Delta)
 {
     int retval;

vp8/common/recon.c

@@ -110,19 +110,19 @@ void vp8_recon_mby_c(const vp8_recon_rtcd_vtable_t *rtcd, MACROBLOCKD *x)
 {
 #if ARCH_ARM
     BLOCKD *b = &x->block[0];
-    RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
 
     /*b = &x->block[4];*/
     b += 4;
-    RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
 
     /*b = &x->block[8];*/
     b += 4;
-    RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
 
     /*b = &x->block[12];*/
     b += 4;
-    RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
 #else
     int i;
 
@@ -130,7 +130,7 @@ void vp8_recon_mby_c(const vp8_recon_rtcd_vtable_t *rtcd, MACROBLOCKD *x)
     {
         BLOCKD *b = &x->block[i];
 
-        RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+        RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     }
 #endif
 }
@@ -140,27 +140,27 @@ void vp8_recon_mb_c(const vp8_recon_rtcd_vtable_t *rtcd, MACROBLOCKD *x)
 #if ARCH_ARM
     BLOCKD *b = &x->block[0];
 
-    RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     b += 4;
-    RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     b += 4;
-    RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     b += 4;
-    RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     b += 4;
 
     /*b = &x->block[16];*/
 
-    RECON_INVOKE(rtcd, recon2)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon2)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     b++;
     b++;
-    RECON_INVOKE(rtcd, recon2)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon2)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     b++;
     b++;
-    RECON_INVOKE(rtcd, recon2)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon2)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     b++;
     b++;
-    RECON_INVOKE(rtcd, recon2)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+    RECON_INVOKE(rtcd, recon2)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
 #else
     int i;
 
@@ -168,14 +168,14 @@ void vp8_recon_mb_c(const vp8_recon_rtcd_vtable_t *rtcd, MACROBLOCKD *x)
     {
         BLOCKD *b = &x->block[i];
 
-        RECON_INVOKE(rtcd, recon4)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+        RECON_INVOKE(rtcd, recon4)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     }
 
     for (i = 16; i < 24; i += 2)
     {
         BLOCKD *b = &x->block[i];
 
-        RECON_INVOKE(rtcd, recon2)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+        RECON_INVOKE(rtcd, recon2)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     }
 #endif
 }

vp8/common/recon.h

@@ -26,9 +26,6 @@
 #define prototype_build_intra_predictors(sym) \
     void sym(MACROBLOCKD *x)
 
-#define prototype_intra4x4_predict(sym) \
-    void sym(BLOCKD *x, int b_mode, unsigned char *predictor)
-
 struct vp8_recon_rtcd_vtable;
 
 #if ARCH_X86 || ARCH_X86_64
@@ -91,30 +88,11 @@ extern prototype_build_intra_predictors\
 extern prototype_build_intra_predictors\
     (vp8_recon_build_intra_predictors_mby_s);
 
-#ifndef vp8_recon_build_intra_predictors_mbuv
-#define vp8_recon_build_intra_predictors_mbuv vp8_build_intra_predictors_mbuv
-#endif
-extern prototype_build_intra_predictors\
-    (vp8_recon_build_intra_predictors_mbuv);
-
-#ifndef vp8_recon_build_intra_predictors_mbuv_s
-#define vp8_recon_build_intra_predictors_mbuv_s vp8_build_intra_predictors_mbuv_s
-#endif
-extern prototype_build_intra_predictors\
-    (vp8_recon_build_intra_predictors_mbuv_s);
-
-#ifndef vp8_recon_intra4x4_predict
-#define vp8_recon_intra4x4_predict vp8_intra4x4_predict
-#endif
-extern prototype_intra4x4_predict\
-    (vp8_recon_intra4x4_predict);
-
 
 typedef prototype_copy_block((*vp8_copy_block_fn_t));
 typedef prototype_recon_block((*vp8_recon_fn_t));
 typedef prototype_recon_macroblock((*vp8_recon_mb_fn_t));
 typedef prototype_build_intra_predictors((*vp8_build_intra_pred_fn_t));
-typedef prototype_intra4x4_predict((*vp8_intra4x4_pred_fn_t));
 typedef struct vp8_recon_rtcd_vtable
 {
     vp8_copy_block_fn_t  copy16x16;
@@ -127,9 +105,6 @@ typedef struct vp8_recon_rtcd_vtable
     vp8_recon_mb_fn_t    recon_mby;
     vp8_build_intra_pred_fn_t  build_intra_predictors_mby_s;
     vp8_build_intra_pred_fn_t  build_intra_predictors_mby;
-    vp8_build_intra_pred_fn_t  build_intra_predictors_mbuv_s;
-    vp8_build_intra_pred_fn_t  build_intra_predictors_mbuv;
-    vp8_intra4x4_pred_fn_t intra4x4_predict;
 } vp8_recon_rtcd_vtable_t;
 
 #if CONFIG_RUNTIME_CPU_DETECT

vp8/common/reconinter.c

@@ -8,9 +8,7 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-
 #include "vpx_ports/config.h"
-#include "vpx/vpx_integer.h"
 #include "recon.h"
 #include "subpixel.h"
 #include "blockd.h"
@@ -19,10 +17,22 @@
 #include "onyxc_int.h"
 #endif
 
+#if CONFIG_OPENCL
+#include "opencl/vp8_opencl.h"
+#include "opencl/filter_cl.h"
+#include "opencl/reconinter_cl.h"
+#endif
+
+/* use this define on systems where unaligned int reads and writes are
+ * not allowed, i.e. ARM architectures
+ */
+/*#define MUST_BE_ALIGNED*/
+
+
 static const int bbb[4] = {0, 2, 8, 10};
 
 
-
+//Copy 16 x 16-bytes from src to dst.
 void vp8_copy_mem16x16_c(
     unsigned char *src,
     int src_stride,
@@ -32,9 +42,12 @@ void vp8_copy_mem16x16_c(
 
     int r;
 
+	//Set this up as a 2D kernel. Each loop iteration is X, each byte/int within
+	//is the Y address.
+
     for (r = 0; r < 16; r++)
     {
-#if !(CONFIG_FAST_UNALIGNED)
+#ifdef MUST_BE_ALIGNED
         dst[0] = src[0];
         dst[1] = src[1];
         dst[2] = src[2];
@@ -53,10 +66,10 @@ void vp8_copy_mem16x16_c(
         dst[15] = src[15];
 
 #else
-        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
-        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
-        ((uint32_t *)dst)[2] = ((uint32_t *)src)[2] ;
-        ((uint32_t *)dst)[3] = ((uint32_t *)src)[3] ;
+        ((int *)dst)[0] = ((int *)src)[0] ;
+        ((int *)dst)[1] = ((int *)src)[1] ;
+        ((int *)dst)[2] = ((int *)src)[2] ;
+        ((int *)dst)[3] = ((int *)src)[3] ;
 
 #endif
         src += src_stride;
@@ -66,6 +79,7 @@ void vp8_copy_mem16x16_c(
 
 }
 
+//Copy 8 x 8-bytes
 void vp8_copy_mem8x8_c(
     unsigned char *src,
     int src_stride,
@@ -76,7 +90,7 @@ void vp8_copy_mem8x8_c(
 
     for (r = 0; r < 8; r++)
     {
-#if !(CONFIG_FAST_UNALIGNED)
+#ifdef MUST_BE_ALIGNED
         dst[0] = src[0];
         dst[1] = src[1];
         dst[2] = src[2];
@@ -86,8 +100,8 @@ void vp8_copy_mem8x8_c(
         dst[6] = src[6];
         dst[7] = src[7];
 #else
-        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
-        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
+        ((int *)dst)[0] = ((int *)src)[0] ;
+        ((int *)dst)[1] = ((int *)src)[1] ;
 #endif
         src += src_stride;
         dst += dst_stride;
@@ -106,7 +120,7 @@ void vp8_copy_mem8x4_c(
 
     for (r = 0; r < 4; r++)
     {
-#if !(CONFIG_FAST_UNALIGNED)
+#ifdef MUST_BE_ALIGNED
         dst[0] = src[0];
         dst[1] = src[1];
         dst[2] = src[2];
@@ -116,8 +130,8 @@ void vp8_copy_mem8x4_c(
         dst[6] = src[6];
         dst[7] = src[7];
 #else
-        ((uint32_t *)dst)[0] = ((uint32_t *)src)[0] ;
-        ((uint32_t *)dst)[1] = ((uint32_t *)src)[1] ;
+        ((int *)dst)[0] = ((int *)src)[0] ;
+        ((int *)dst)[1] = ((int *)src)[1] ;
 #endif
         src += src_stride;
         dst += dst_stride;
@@ -131,34 +145,32 @@ void vp8_copy_mem8x4_c(
 void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, vp8_subpix_fn_t sppf)
 {
     int r;
-    unsigned char *ptr_base;
-    unsigned char *ptr;
-    unsigned char *pred_ptr = d->predictor;
 
-    ptr_base = *(d->base_pre);
+    //d->base_pre is the start of the previous frame's y_buffer, u_buffer, or v_buffer
+    unsigned char *ptr_base = *(d->base_pre);
+    int ptr_offset = d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
+
+    unsigned char *pred_ptr = d->predictor_base + d->predictor_offset;
 
     if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
     {
-        ptr = ptr_base + d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
-        sppf(ptr, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, pred_ptr, pitch);
+        sppf(ptr_base+ptr_offset, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, pred_ptr, pitch);
     }
     else
     {
-        ptr_base += d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
-        ptr = ptr_base;
 
         for (r = 0; r < 4; r++)
         {
-#if !(CONFIG_FAST_UNALIGNED)
-            pred_ptr[0]  = ptr[0];
-            pred_ptr[1]  = ptr[1];
-            pred_ptr[2]  = ptr[2];
-            pred_ptr[3]  = ptr[3];
+#ifdef MUST_BE_ALIGNED
+            pred_ptr[0]  = ptr_base[ptr_offset];
+            pred_ptr[1]  = ptr_base[ptr_offset+1];
+            pred_ptr[2]  = ptr_base[ptr_offset+2];
+            pred_ptr[3]  = ptr_base[ptr_offset+3];
 #else
-            *(uint32_t *)pred_ptr = *(uint32_t *)ptr ;
+            *(int *)pred_ptr = *(int *)(ptr_base+ptr_offset) ;
 #endif
             pred_ptr     += pitch;
-            ptr         += d->pre_stride;
+            ptr_offset   += d->pre_stride;
         }
     }
 }
@@ -167,7 +179,7 @@ static void build_inter_predictors4b(MACROBLOCKD *x, BLOCKD *d, int pitch)
 {
     unsigned char *ptr_base;
     unsigned char *ptr;
-    unsigned char *pred_ptr = d->predictor;
+    unsigned char *pred_ptr = d->predictor_base + d->predictor_offset;
 
     ptr_base = *(d->base_pre);
     ptr = ptr_base + d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
@@ -186,7 +198,7 @@ static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d, int pitch)
 {
     unsigned char *ptr_base;
     unsigned char *ptr;
-    unsigned char *pred_ptr = d->predictor;
+    unsigned char *pred_ptr = d->predictor_base + d->predictor_offset;
 
     ptr_base = *(d->base_pre);
     ptr = ptr_base + d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
@@ -201,13 +213,24 @@ static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d, int pitch)
     }
 }
 
-
-/*encoder only*/
+/* Encoder only */
 void vp8_build_inter_predictors_mbuv(MACROBLOCKD *x)
 {
     int i;
 
-    if (x->mode_info_context->mbmi.mode != SPLITMV)
+#if CONFIG_OPENCL
+    if ( 0 && cl_initialized == CL_SUCCESS ){
+        vp8_build_inter_predictors_mbuv_cl(x);
+        VP8_CL_FINISH(x->cl_commands);
+        VP8_CL_FINISH(x->block[0].cl_commands);
+        VP8_CL_FINISH(x->block[16].cl_commands);
+        VP8_CL_FINISH(x->block[20].cl_commands);
+        return;
+    }
+#endif
+
+    if (x->mode_info_context->mbmi.ref_frame != INTRA_FRAME &&
+        x->mode_info_context->mbmi.mode != SPLITMV)
     {
         unsigned char *uptr, *vptr;
         unsigned char *upred_ptr = &x->predictor[256];
@@ -252,7 +275,11 @@ void vp8_build_inter_predictors_mbuv(MACROBLOCKD *x)
 }
 
 /*encoder only*/
-void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x)
+void vp8_build_inter_predictors_mby(MACROBLOCKD *x)
+{
+
+    if (x->mode_info_context->mbmi.ref_frame != INTRA_FRAME &&
+        x->mode_info_context->mbmi.mode != SPLITMV)
     {
         unsigned char *ptr_base;
         unsigned char *ptr;
@@ -273,33 +300,67 @@ void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x)
             RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, pred_ptr, 16);
         }
     }
+    else
+    {
+        int i;
 
-void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x,
-                                        unsigned char *dst_y,
-                                        unsigned char *dst_u,
-                                        unsigned char *dst_v,
-                                        int dst_ystride,
-                                        int dst_uvstride)
+        if (x->mode_info_context->mbmi.partitioning < 3)
+        {
+            for (i = 0; i < 4; i++)
+            {
+                BLOCKD *d = &x->block[bbb[i]];
+                build_inter_predictors4b(x, d, 16);
+            }
+
+        }
+        else
+        {
+            for (i = 0; i < 16; i += 2)
+            {
+                BLOCKD *d0 = &x->block[i];
+                BLOCKD *d1 = &x->block[i+1];
+
+                if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
+                    build_inter_predictors2b(x, d0, 16);
+                else
+                {
+                    vp8_build_inter_predictors_b(d0, 16, x->subpixel_predict);
+                    vp8_build_inter_predictors_b(d1, 16, x->subpixel_predict);
+                }
+
+            }
+        }
+    }
+}
+
+void vp8_build_inter_predictors_mb(MACROBLOCKD *x)
+{
+
+    if (x->mode_info_context->mbmi.ref_frame != INTRA_FRAME &&
+        x->mode_info_context->mbmi.mode != SPLITMV)
     {
         int offset;
+        unsigned char *ptr_base;
         unsigned char *ptr;
         unsigned char *uptr, *vptr;
+        unsigned char *pred_ptr = x->predictor;
+        unsigned char *upred_ptr = &x->predictor[256];
+        unsigned char *vpred_ptr = &x->predictor[320];
 
         int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
         int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
-
-    unsigned char *ptr_base = x->pre.y_buffer;
         int pre_stride = x->block[0].pre_stride;
 
+        ptr_base = x->pre.y_buffer;
         ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
 
         if ((mv_row | mv_col) & 7)
         {
-        x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, dst_y, dst_ystride);
+            x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, pred_ptr, 16);
         }
         else
         {
-        RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, dst_y, dst_ystride);
+            RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, pred_ptr, 16);
         }
 
         mv_row = x->block[16].bmi.mv.as_mv.row;
@@ -311,18 +372,16 @@ void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x,
 
         if ((mv_row | mv_col) & 7)
         {
-        x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, dst_u, dst_uvstride);
-        x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, dst_v, dst_uvstride);
+                x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, upred_ptr, 8);
+                x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vpred_ptr, 8);
         }
         else
         {
-        RECON_INVOKE(&x->rtcd->recon, copy8x8)(uptr, pre_stride, dst_u, dst_uvstride);
-        RECON_INVOKE(&x->rtcd->recon, copy8x8)(vptr, pre_stride, dst_v, dst_uvstride);
+            RECON_INVOKE(&x->rtcd->recon, copy8x8)(uptr, pre_stride, upred_ptr, 8);
+            RECON_INVOKE(&x->rtcd->recon, copy8x8)(vptr, pre_stride, vpred_ptr, 8);
         }
-
     }
-
-void vp8_build_inter4x4_predictors_mb(MACROBLOCKD *x)
+    else
     {
         int i;
 
@@ -365,19 +424,9 @@ void vp8_build_inter4x4_predictors_mb(MACROBLOCKD *x)
                 vp8_build_inter_predictors_b(d0, 8, x->subpixel_predict);
                 vp8_build_inter_predictors_b(d1, 8, x->subpixel_predict);
             }
-    }
+
         }
 
-void vp8_build_inter_predictors_mb(MACROBLOCKD *x)
-{
-    if (x->mode_info_context->mbmi.mode != SPLITMV)
-    {
-        vp8_build_inter16x16_predictors_mb(x, x->predictor, &x->predictor[256],
-                                           &x->predictor[320], 16, 8);
-    }
-    else
-    {
-        vp8_build_inter4x4_predictors_mb(x);
     }
 }
 
@@ -461,5 +510,202 @@ void vp8_build_uvmvs(MACROBLOCKD *x, int fullpixel)
 }
 
 
+/* The following functions are written for skip_recon_mb() to call. Since there is no recon in this
+ * situation, we can write the result directly to dst buffer instead of writing it to predictor
+ * buffer and then copying it to dst buffer.
+ */
+static void vp8_build_inter_predictors_b_s(BLOCKD *d, unsigned char *dst_ptr, vp8_subpix_fn_t sppf)
+{
+    int r;
+    unsigned char *ptr_base;
+    unsigned char *ptr;
+    /*unsigned char *pred_ptr = d->predictor_base + d->predictor_offset;*/
+    int dst_stride = d->dst_stride;
+    int pre_stride = d->pre_stride;
+    int ptr_offset = d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
+
+    ptr_base = *(d->base_pre);
+    ptr = ptr_base + ptr_offset;
+
+    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
+    {
+        sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst_ptr, dst_stride);
+    }
+    else
+    {
+        for (r = 0; r < 4; r++)
+        {
+#ifdef MUST_BE_ALIGNED
+            dst_ptr[0]   = ptr[0];
+            dst_ptr[1]   = ptr[1];
+            dst_ptr[2]   = ptr[2];
+            dst_ptr[3]   = ptr[3];
+#else
+            *(int *)dst_ptr = *(int *)ptr ;
+#endif
+            dst_ptr      += dst_stride;
+            ptr         += pre_stride;
+        }
+    }
+}
 
 
+void vp8_build_inter_predictors_mb_s(MACROBLOCKD *x)
+{
+    unsigned char *dst_ptr = x->dst.y_buffer;
+
+#if CONFIG_OPENCL && ENABLE_CL_SUBPIXEL
+    if (cl_initialized == CL_SUCCESS){
+        vp8_build_inter_predictors_mb_s_cl(x);
+        return;
+    }
+#endif
+
+    if (x->mode_info_context->mbmi.mode != SPLITMV)
+    {
+        int offset;
+        unsigned char *ptr_base;
+        unsigned char *ptr;
+        unsigned char *uptr, *vptr;
+        /*unsigned char *pred_ptr = x->predictor;
+        unsigned char *upred_ptr = &x->predictor[256];
+        unsigned char *vpred_ptr = &x->predictor[320];*/
+        unsigned char *udst_ptr = x->dst.u_buffer;
+        unsigned char *vdst_ptr = x->dst.v_buffer;
+
+        int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
+        int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
+        int pre_stride = x->dst.y_stride; /*x->block[0].pre_stride;*/
+
+        ptr_base = x->pre.y_buffer;
+        ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
+
+        if ((mv_row | mv_col) & 7)
+        {
+                x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
+        }
+        else
+        {
+            RECON_INVOKE(&x->rtcd->recon, copy16x16)(ptr, pre_stride, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
+        }
+
+        mv_row = x->block[16].bmi.mv.as_mv.row;
+        mv_col = x->block[16].bmi.mv.as_mv.col;
+        pre_stride >>= 1;
+        offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
+        uptr = x->pre.u_buffer + offset;
+        vptr = x->pre.v_buffer + offset;
+
+        if ((mv_row | mv_col) & 7)
+        {
+                x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, udst_ptr, x->dst.uv_stride);
+                x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vdst_ptr, x->dst.uv_stride);
+        }
+        else
+        {
+            RECON_INVOKE(&x->rtcd->recon, copy8x8)(uptr, pre_stride, udst_ptr, x->dst.uv_stride);
+            RECON_INVOKE(&x->rtcd->recon, copy8x8)(vptr, pre_stride, vdst_ptr, x->dst.uv_stride);
+        }
+    }
+    else
+    {
+        /* note: this whole ELSE part is not executed at all. So, no way to test the correctness of my modification. Later,
+         * if sth is wrong, go back to what it is in build_inter_predictors_mb.
+         *
+         * ACW: note: Not sure who the above comment belongs to.
+         */
+        int i;
+
+        if (x->mode_info_context->mbmi.partitioning < 3)
+        {
+            for (i = 0; i < 4; i++)
+            {
+                BLOCKD *d = &x->block[bbb[i]];
+                /*build_inter_predictors4b(x, d, 16);*/
+
+                {
+                    unsigned char *ptr_base;
+                    unsigned char *ptr;
+
+                    ptr_base = *(d->base_pre);
+                    ptr = ptr_base + d->pre + (d->bmi.mv.as_mv.row >> 3) * d->pre_stride + (d->bmi.mv.as_mv.col >> 3);
+
+                    if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7)
+                    {
+                        x->subpixel_predict8x8(ptr, d->pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
+                    }
+                    else
+                    {
+                        RECON_INVOKE(&x->rtcd->recon, copy8x8)(ptr, d->pre_stride, dst_ptr, x->dst.y_stride); /*x->block[0].dst_stride);*/
+                    }
+                }
+            }
+        }
+		else
+        {
+            for (i = 0; i < 16; i += 2)
+            {
+                BLOCKD *d0 = &x->block[i];
+                BLOCKD *d1 = &x->block[i+1];
+
+                if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
+                {
+                    /*build_inter_predictors2b(x, d0, 16);*/
+                    unsigned char *ptr_base;
+                    unsigned char *ptr;
+
+                    ptr_base = *(d0->base_pre);
+                    ptr = ptr_base + d0->pre + (d0->bmi.mv.as_mv.row >> 3) * d0->pre_stride + (d0->bmi.mv.as_mv.col >> 3);
+
+                    if (d0->bmi.mv.as_mv.row & 7 || d0->bmi.mv.as_mv.col & 7)
+                    {
+                        x->subpixel_predict8x4(ptr, d0->pre_stride, d0->bmi.mv.as_mv.col & 7, d0->bmi.mv.as_mv.row & 7, dst_ptr, x->dst.y_stride);
+                    }
+                    else
+                    {
+                        RECON_INVOKE(&x->rtcd->recon, copy8x4)(ptr, d0->pre_stride, dst_ptr, x->dst.y_stride);
+                    }
+                }
+                else
+                {
+                    vp8_build_inter_predictors_b_s(d0, dst_ptr, x->subpixel_predict);
+                    vp8_build_inter_predictors_b_s(d1, dst_ptr, x->subpixel_predict);
+                }
+            }
+        }
+
+        for (i = 16; i < 24; i += 2)
+        {
+            BLOCKD *d0 = &x->block[i];
+            BLOCKD *d1 = &x->block[i+1];
+
+            if (d0->bmi.mv.as_int == d1->bmi.mv.as_int)
+            {
+                /*build_inter_predictors2b(x, d0, 8);*/
+                unsigned char *ptr_base;
+                unsigned char *ptr;
+
+                ptr_base = *(d0->base_pre);
+                ptr = ptr_base + d0->pre + (d0->bmi.mv.as_mv.row >> 3) * d0->pre_stride + (d0->bmi.mv.as_mv.col >> 3);
+
+                if (d0->bmi.mv.as_mv.row & 7 || d0->bmi.mv.as_mv.col & 7)
+                {
+                    x->subpixel_predict8x4(ptr, d0->pre_stride,
+                        d0->bmi.mv.as_mv.col & 7,
+                        d0->bmi.mv.as_mv.row & 7,
+                        dst_ptr, x->dst.uv_stride);
+                }
+                else
+                {
+                    RECON_INVOKE(&x->rtcd->recon, copy8x4)(ptr,
+                        d0->pre_stride, dst_ptr, x->dst.uv_stride);
+                }
+            }
+            else
+            {
+                vp8_build_inter_predictors_b_s(d0, dst_ptr, x->subpixel_predict);
+                vp8_build_inter_predictors_b_s(d1, dst_ptr, x->subpixel_predict);
+            }
+        }
+    }
+}

vp8/common/reconinter.h

@@ -13,15 +13,9 @@
 #define __INC_RECONINTER_H
 
 extern void vp8_build_inter_predictors_mb(MACROBLOCKD *x);
-extern void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x,
-                                               unsigned char *dst_y,
-                                               unsigned char *dst_u,
-                                               unsigned char *dst_v,
-                                               int dst_ystride,
-                                               int dst_uvstride);
+extern void vp8_build_inter_predictors_mb_s(MACROBLOCKD *x);
 
-
-extern void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x);
+extern void vp8_build_inter_predictors_mby(MACROBLOCKD *x);
 extern void vp8_build_uvmvs(MACROBLOCKD *x, int fullpixel);
 extern void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, vp8_subpix_fn_t sppf);
 extern void vp8_build_inter_predictors_mbuv(MACROBLOCKD *x);

vp8/common/reconintra.c

@@ -24,7 +24,7 @@ void vp8_recon_intra_mbuv(const vp8_recon_rtcd_vtable_t *rtcd, MACROBLOCKD *x)
     for (i = 16; i < 24; i += 2)
     {
         BLOCKD *b = &x->block[i];
-        RECON_INVOKE(rtcd, recon2)(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride);
+        RECON_INVOKE(rtcd, recon2)(b->predictor_base + b->predictor_offset, &b->diff_base[b->diff_offset], *(b->base_dst) + b->dst, b->dst_stride);
     }
 }
 

vp8/common/reconintra.h

@@ -14,4 +14,9 @@
 
 extern void init_intra_left_above_pixels(MACROBLOCKD *x);
 
+extern void vp8_build_intra_predictors_mbuv(MACROBLOCKD *x);
+extern void vp8_build_intra_predictors_mbuv_s(MACROBLOCKD *x);
+
+extern void vp8_predict_intra4x4(BLOCKD *x, int b_mode, unsigned char *Predictor);
+
 #endif

vp8/common/reconintra4x4.c

@@ -14,7 +14,7 @@
 #include "vpx_mem/vpx_mem.h"
 #include "reconintra.h"
 
-void vp8_intra4x4_predict(BLOCKD *x,
+void vp8_predict_intra4x4(BLOCKD *x,
                           int b_mode,
                           unsigned char *predictor)
 {
@@ -124,6 +124,18 @@ void vp8_intra4x4_predict(BLOCKD *x,
     case B_LD_PRED:
     {
         unsigned char *ptr = Above;
+
+#if 0
+        //More readable version of the unrolled loop
+        int stride = 16, r=0, c=0;
+        for (r=0; r < 4; r++){
+            for (c=0; c < 4; c++){
+                int off = r+c;
+                int off2 = off > 5 ? 5: off; //Clamp so [3,3] has max off2 of 7
+                predictor[r*stride+c] = (ptr[off] + ptr[off+1]*2 + ptr[off2+2] + 2)>>2;
+            }
+        }
+#else
         predictor[0 * 16 + 0] = (ptr[0] + ptr[1] * 2 + ptr[2] + 2) >> 2;
         predictor[0 * 16 + 1] =
             predictor[1 * 16 + 0] = (ptr[1] + ptr[2] * 2 + ptr[3] + 2) >> 2;
@@ -140,6 +152,7 @@ void vp8_intra4x4_predict(BLOCKD *x,
         predictor[2 * 16 + 3] =
             predictor[3 * 16 + 2] = (ptr[5] + ptr[6] * 2 + ptr[7] + 2) >> 2;
         predictor[3 * 16 + 3] = (ptr[6] + ptr[7] * 2 + ptr[7] + 2) >> 2;
+#endif
         
     }
     break;
@@ -311,5 +324,3 @@ void vp8_intra_prediction_down_copy(MACROBLOCKD *x)
     *dst_ptr1 = *src_ptr;
     *dst_ptr2 = *src_ptr;
 }
-
-