README.wasm: add instructions for enabling mulhi

Change-Id: I1e9dd737f06ad76f73824352291a6e129ca5ded1

Android.mk

@@ -15,7 +15,10 @@ endif
 # https://bugs.chromium.org/p/webp/issues/detail?id=343
 ifeq ($(findstring clang,$(NDK_TOOLCHAIN_VERSION)),clang)
   ifeq ($(TARGET_ARCH),mips)
-    WEBP_CFLAGS += -no-integrated-as
+    clang_version := $(shell $(TARGET_CC) --version)
+    ifneq ($(findstring clang version 3,$(clang_version)),)
+      WEBP_CFLAGS += -no-integrated-as
+    endif
   endif
 endif
 
@@ -25,6 +28,7 @@ ifneq ($(findstring armeabi-v7a, $(TARGET_ARCH_ABI)),)
   # specifically.
   NEON := c.neon
   USE_CPUFEATURES := yes
+  WEBP_CFLAGS += -DHAVE_CPU_FEATURES_H
 else
   NEON := c
 endif

CMakeLists.txt

@@ -4,6 +4,7 @@ project(libwebp C)
 
 # Options for coder / decoder executables.
 option(WEBP_ENABLE_SIMD "Enable any SIMD optimization." ON)
+option(WEBP_ENABLE_WASM "Enable WebAssembly optimizations." OFF)
 option(WEBP_BUILD_CWEBP "Build the cwebp command line tool." OFF)
 option(WEBP_BUILD_DWEBP "Build the dwebp command line tool." OFF)
 option(WEBP_BUILD_GIF2WEBP "Build the gif2webp conversion tool." OFF)
@@ -13,7 +14,7 @@ option(WEBP_BUILD_WEBP_JS "Emscripten build of webp.js." OFF)
 option(WEBP_EXPERIMENTAL_FEATURES "Build with experimental features." OFF)
 option(WEBP_ENABLE_SWAP_16BIT_CSP "Enable byte swap for 16 bit colorspaces." OFF)
 
-if(WEBP_BUILD_WEBP_JS)
+if(WEBP_BUILD_WEBP_JS OR WEBP_ENABLE_WASM)
   set(WEBP_ENABLE_SIMD OFF)
 endif()
 
@@ -37,6 +38,9 @@ string(REGEX MATCH "[0-9.]+" WEBP_VERSION ${SOURCE_FILE})
 if(WEBP_ENABLE_SWAP_16BIT_CSP)
   add_definitions(-DWEBP_SWAP_16BIT_CSP)
 endif()
+if(WEBP_ENABLE_WASM)
+  add_definitions(-DWEBP_USE_WASM)
+endif()
 
 ################################################################################
 # Android only.
@@ -50,6 +54,7 @@ if(ANDROID)
   set(WEBP_DEP_INCLUDE_DIRS ${WEBP_DEP_INCLUDE_DIRS}
     ${ANDROID_NDK}/sources/android/cpufeatures
   )
+  add_definitions(-DHAVE_CPU_FEATURES_H)
 endif()
 
 ################################################################################
@@ -162,9 +167,11 @@ math(EXPR WEBP_SIMD_FILES_TO_INCLUDE_RANGE
 foreach(I_FILE RANGE ${WEBP_SIMD_FILES_TO_INCLUDE_RANGE})
   list(GET WEBP_SIMD_FILES_TO_INCLUDE ${I_FILE} FILE)
   list(GET WEBP_SIMD_FLAGS_TO_INCLUDE ${I_FILE} SIMD_COMPILE_FLAG)
-  set_source_files_properties(${FILE} PROPERTIES
+  if(NOT ${SIMD_COMPILE_FLAG} STREQUAL "NOTFOUND")
-    COMPILE_FLAGS ${SIMD_COMPILE_FLAG}
+    set_source_files_properties(${FILE} PROPERTIES
-  )
+      COMPILE_FLAGS ${SIMD_COMPILE_FLAG}
+    )
+  endif()
 endforeach()
 
 # Build the executables if asked for.
@@ -228,12 +235,12 @@ endif()
 
 if(WEBP_BUILD_GIF2WEBP)
   # gif2webp
-  include_directories(${WEBP_DEP_IMG_INCLUDE_DIRS})
+  include_directories(${WEBP_DEP_GIF_INCLUDE_DIRS})
   parse_Makefile_am(${CMAKE_CURRENT_SOURCE_DIR}/examples "GIF2WEBP_SRCS"
     "gif2webp")
   add_executable(gif2webp ${GIF2WEBP_SRCS})
   target_link_libraries(gif2webp exampleutil imageioutil webp webpmux
-    ${WEBP_DEP_IMG_LIBRARIES})
+    ${WEBP_DEP_GIF_LIBRARIES})
   install(TARGETS gif2webp RUNTIME DESTINATION bin)
 endif()
 

README.wasm

@@ -0,0 +1,91 @@
+Description:
+============
+
+This file describes the compilation of libwebp using portable intrinsics /
+WebAssembly (wasm) to native targets using clang and CMake.
+
+Prerequisites:
+==============
+
+- cmake 2.8+
+
+- clang 3.9+ for portable intrinsics support; as wasm progresses a tip of tree
+  build may be necessary.
+
+Building:
+=========
+
+ - configure the project with CMake using:
+
+ $ mkdir -p build && \
+   cd build && \
+   cmake -DWEBP_BUILD_DWEBP=1 -DCMAKE_C_COMPILER=clang -DWEBP_ENABLE_WASM=1 ../
+
+ - compile dwebp using 'make'.
+
+ - Note this currently generates native executables only and is incompatible
+   with -DWEBP_BUILD_WEBP_JS.
+
+Build options:
+==============
+
+- platform specific multiply high (mulhi) implementation, disabled by default.
+  arm: -DCMAKE_C_FLAGS='-DENABLE_NEON_BUILTIN_MULHI_INT16X8 ...'
+  x86: -DCMAKE_C_FLAGS='-DENABLE_X86_BUILTIN_MULHI_INT16X8 ...'
+
+Cross compilation:
+==================
+
+ - arm toolchains can be obtained from:
+   http://www.linaro.org/downloads/
+
+ - the android ndk can be obtained from:
+   https://developer.android.com/ndk/downloads/index.html
+
+armv7:
+------
+
+Android:
+ $ ./android-ndk-r15b/build/tools/make_standalone_toolchain.py \
+   --arch arm --api 24 --stl gnustl --install-dir /opt/android-arm-24
+ $ mkdir -p build && cd build
+ $ cmake ../libwebp \
+   -DWEBP_BUILD_DWEBP=1 \
+   -DCMAKE_C_COMPILER=/opt/android-arm-24/bin/clang \
+   -DCMAKE_PREFIX_PATH=/opt/android-arm-24/sysroot/usr/lib \
+   -DCMAKE_C_FLAGS=-fPIE \
+   -DCMAKE_EXE_LINKER_FLAGS=-Wl,-pie \
+   -DCMAKE_BUILD_TYPE=Release \
+   -DWEBP_ENABLE_WASM=1
+
+Linux:
+ $ gcc_arm=/opt/gcc-arm; target=arm-linux-gnueabihf
+ $ mkdir -p build && cd build
+ $ cmake ../libwebp -DWEBP_BUILD_DWEBP=1 -DWEBP_ENABLE_WASM=1 \
+   -DCMAKE_C_COMPILER=clang \
+   -DCMAKE_C_FLAGS="--target=$target --gcc-toolchain=$gcc_arm --sysroot=$gcc_arm/$target/libc -march=armv7-a -mfpu=neon" \
+   -DCMAKE_PREFIX_PATH=$gcc_arm/$target/libc/usr
+
+aarch64 / arm64:
+----------------
+
+Android:
+ $ ./android-ndk-r15b/build/tools/make_standalone_toolchain.py \
+   --arch arm64 --api 24 --stl gnustl --install-dir /opt/android-arm64-24
+ $ mkdir -p build && cd build
+ $ cmake ../libwebp \
+   -DWEBP_BUILD_DWEBP=1 \
+   -DCMAKE_C_COMPILER=/opt/android-arm64-24/bin/clang \
+   -DCMAKE_PREFIX_PATH=/opt/android-arm64-24/sysroot/usr/lib \
+   -DCMAKE_C_FLAGS=-fPIE \
+   -DCMAKE_EXE_LINKER_FLAGS=-Wl,-pie \
+   -DCMAKE_BUILD_TYPE=Release \
+   -DWEBP_ENABLE_WASM=1
+
+Linux:
+ $ gcc_arm=/opt/gcc-aarch64; target=aarch64-linux-gnu
+ $ mkdir -p build && cd build
+ $ cmake ../libwebp -DWEBP_BUILD_DWEBP=1 -DWEBP_ENABLE_WASM=1 \
+   -DCMAKE_C_COMPILER=clang \
+   -DCMAKE_C_FLAGS="--target=$target --gcc-toolchain=$gcc_arm --sysroot=$gcc_arm/$target/libc" \
+   -DCMAKE_PREFIX_PATH=$gcc_arm/$target/libc/usr

README.webp_js

@@ -31,6 +31,11 @@ using Emscripten and CMake.
  - that's it! Upon completion, you should have the webp.js and
    webp.js.mem files generated.
 
+ - Note this generates both webp_js and webp_wasm without any SIMD enabled due
+   to bugs with this toolchain associated with the SSE2 code.
+   -DWEBP_ENABLE_WASM is currently meant to generate native (x86, arm)
+   executables (dwebp, cwebp) and is incompatible with -DWEBP_BUILD_WEBP_JS.
+
 The callable JavaScript function is WebPToSDL(), which decodes a raw WebP
 bitstream into a canvas. See webp_js/index.html for a simple usage sample.
 

build.gradle

@@ -84,6 +84,7 @@ model {
       // Check for NEON usage.
       if (getTargetPlatform() == "arm" || getTargetPlatform() == "arm64") {
         NEON = "c.neon"
+        cCompiler.define "HAVE_CPU_FEATURES_H"
       } else {
         NEON = "c"
       }

cmake/config.h.cmake

@@ -65,7 +65,7 @@ endif()
 # Find the standard image libraries.
 set(WEBP_DEP_IMG_LIBRARIES)
 set(WEBP_DEP_IMG_INCLUDE_DIRS)
-foreach(I_LIB PNG JPEG TIFF GIF)
+foreach(I_LIB PNG JPEG TIFF)
   find_package(${I_LIB})
   set(WEBP_HAVE_${I_LIB} ${${I_LIB}_FOUND})
   if(${I_LIB}_FOUND)
@@ -74,6 +74,16 @@ foreach(I_LIB PNG JPEG TIFF GIF)
   endif()
 endforeach()
 
+# GIF detection, gifdec isn't part of the imageio lib.
+set(WEBP_DEP_GIF_LIBRARIES)
+set(WEBP_DEP_GIF_INCLUDE_DIRS)
+find_package(GIF)
+set(WEBP_HAVE_GIF ${GIF_FOUND})
+if(GIF_FOUND)
+  list(APPEND WEBP_DEP_GIF_LIBRARIES ${GIF_LIBRARIES})
+  list(APPEND WEBP_DEP_GIF_INCLUDE_DIRS ${GIF_INCLUDE_DIR})
+endif()
+
 ## Check for specific headers.
 include(CheckIncludeFiles)
 check_include_files("stdlib.h;stdarg.h;string.h;float.h" STDC_HEADERS)

cmake/cpu.cmake

@@ -85,8 +85,11 @@ foreach(I_SIMD RANGE ${WEBP_SIMD_FLAGS_RANGE})
     foreach(FILE ${SIMD_FILES})
       list(APPEND WEBP_SIMD_FILES_NOT_TO_INCLUDE ${FILE})
     endforeach()
-    # Explicitly disable SIMD.
+    # Explicitly disable SIMD. Avoid this with WASM to avoid an ICE with clang:
-    if(SIMD_DISABLE_FLAGS)
+    # https://bugs.chromium.org/p/webp/issues/detail?id=350
+    # WASM overrides the native SIMD so building it in is harmless aside from
+    # binary size.
+    if(NOT WEBP_ENABLE_WASM AND SIMD_DISABLE_FLAGS)
       list(GET SIMD_DISABLE_FLAGS ${I_SIMD} SIMD_COMPILE_FLAG)
       include(CheckCCompilerFlag)
       if(SIMD_COMPILE_FLAG)
@@ -116,3 +119,13 @@ foreach(I_SIMD RANGE ${WEBP_SIMD_FLAGS_RANGE})
     endif()
   endif()
 endforeach()
+
+## Add *_wasm.c files if enabled.
+if(WEBP_ENABLE_WASM)
+  file(GLOB SIMD_FILES "${CMAKE_CURRENT_LIST_DIR}/../"
+    "src/dsp/*_wasm.c"
+  )
+  foreach(FILE ${SIMD_FILES})
+    list(APPEND WEBP_SIMD_FILES_TO_INCLUDE ${FILE})
+  endforeach()
+endif()

configure.ac

@@ -243,9 +243,13 @@ AS_IF([test "x$enable_neon" != "xno"], [
           NEON_FLAGS=""],
           [AC_DEFINE(WEBP_HAVE_NEON_RTCD, [1],
                      [Set to 1 if runtime detection of NEON is enabled])])])
-        ;;
+
-    esac
+      case "$host_os" in
-    AC_SUBST([NEON_FLAGS])])
+        *android*) AC_CHECK_HEADERS([cpu-features.h]) ;;
+      esac
+      ;;
+  esac
+  AC_SUBST([NEON_FLAGS])])
 
 dnl === CLEAR_LIBVARS([var_pfx])
 dnl ===   Clears <var_pfx>_{INCLUDES,LIBS}.

src/dsp/Makefile.am

@@ -3,6 +3,7 @@ noinst_LTLIBRARIES += libwebpdsp_sse2.la libwebpdspdecode_sse2.la
 noinst_LTLIBRARIES += libwebpdsp_sse41.la libwebpdspdecode_sse41.la
 noinst_LTLIBRARIES += libwebpdsp_neon.la libwebpdspdecode_neon.la
 noinst_LTLIBRARIES += libwebpdsp_msa.la libwebpdspdecode_msa.la
+noinst_LTLIBRARIES += libwebpdspdecode_wasm.la
 
 if BUILD_LIBWEBPDECODER
   noinst_LTLIBRARIES += libwebpdspdecode.la
@@ -96,6 +97,10 @@ libwebpdspdecode_msa_la_SOURCES += upsampling_msa.c
 libwebpdspdecode_msa_la_CPPFLAGS = $(libwebpdsp_msa_la_CPPFLAGS)
 libwebpdspdecode_msa_la_CFLAGS = $(libwebpdsp_msa_la_CFLAGS)
 
+# WASM is not fully integrated into configure; the addition here keeps source
+# extraction by cmake simple.
+libwebpdspdecode_wasm_la_SOURCES = dec_wasm.c
+
 libwebpdsp_sse2_la_SOURCES =
 libwebpdsp_sse2_la_SOURCES += argb_sse2.c
 libwebpdsp_sse2_la_SOURCES += cost_sse2.c

src/dsp/cpu.c

@@ -23,11 +23,13 @@
 #endif
 
 //------------------------------------------------------------------------------
-// SSE2 detection.
+// x86/x86-64 micro-arch detection.
 //
 
+// skip x86 specific code for WASM builds
+#if defined(WEBP_USE_WASM)
 // apple/darwin gcc-4.0.1 defines __PIC__, but not __pic__ with -fPIC.
-#if (defined(__pic__) || defined(__PIC__)) && defined(__i386__)
+#elif (defined(__pic__) || defined(__PIC__)) && defined(__i386__)
 static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
   __asm__ volatile (
     "mov %%ebx, %%edi\n"
@@ -63,8 +65,10 @@ static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
 #define GetCPUInfo __cpuid
 #endif
 
+// skip xgetbv definition for WASM builds
+#if defined(WEBP_USE_WASM)
 // NaCl has no support for xgetbv or the raw opcode.
-#if !defined(__native_client__) && (defined(__i386__) || defined(__x86_64__))
+#elif !defined(__native_client__) && (defined(__i386__) || defined(__x86_64__))
 static WEBP_INLINE uint64_t xgetbv(void) {
   const uint32_t ecx = 0;
   uint32_t eax, edx;
@@ -94,7 +98,19 @@ static WEBP_INLINE uint64_t xgetbv(void) {
 #define xgetbv() 0U  // no AVX for older x64 or unrecognized toolchains.
 #endif
 
-#if defined(__i386__) || defined(__x86_64__) || defined(WEBP_MSC_SSE2)
+//------------------------------------------------------------------------------
+// Platform specific VP8CPUInfo functions.
+//
+
+// WASM needs to precede platform specific architecture checks as the defines
+// will still be present when building this target.
+#if defined(WEBP_USE_WASM)
+static int wasmCPUInfo(CPUFeature feature) {
+  if (feature != kWASM) return 0;
+  return 1;
+}
+VP8CPUInfo VP8GetCPUInfo = wasmCPUInfo;
+#elif defined(__i386__) || defined(__x86_64__) || defined(WEBP_MSC_SSE2)
 
 // helper function for run-time detection of slow SSSE3 platforms
 static int CheckSlowModel(int info) {

src/dsp/dec.c

@@ -700,6 +700,7 @@ extern void VP8DspInitNEON(void);
 extern void VP8DspInitMIPS32(void);
 extern void VP8DspInitMIPSdspR2(void);
 extern void VP8DspInitMSA(void);
+extern void VP8DspInitWASM(void);
 
 static volatile VP8CPUInfo dec_last_cpuinfo_used =
     (VP8CPUInfo)&dec_last_cpuinfo_used;
@@ -789,6 +790,11 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8DspInit(void) {
     if (VP8GetCPUInfo(kMSA)) {
       VP8DspInitMSA();
     }
+#endif
+#if defined(WEBP_USE_WASM)
+    if (VP8GetCPUInfo(kWASM)) {
+      VP8DspInitWASM();
+    }
 #endif
   }
   dec_last_cpuinfo_used = VP8GetCPUInfo;

src/dsp/dsp.h

@@ -38,12 +38,22 @@ extern "C" {
 # define LOCAL_GCC_PREREQ(maj, min) 0
 #endif
 
+#if defined(__clang__)
+# define LOCAL_CLANG_VERSION ((__clang_major__ << 8) | __clang_minor__)
+# define LOCAL_CLANG_PREREQ(maj, min) \
+    (LOCAL_CLANG_VERSION >= (((maj) << 8) | (min)))
+#else
+# define LOCAL_CLANG_VERSION 0
+# define LOCAL_CLANG_PREREQ(maj, min) 0
+#endif
+
 #ifndef __has_builtin
 # define __has_builtin(x) 0
 #endif
 
-// for now, none of the optimizations below are available in emscripten
+// For now, none of the optimizations below are available in emscripten.
-#if !defined(EMSCRIPTEN)
+// WebAssembly overrides native optimizations.
+#if !(defined(EMSCRIPTEN) || defined(WEBP_USE_WASM))
 
 #if defined(_MSC_VER) && _MSC_VER > 1310 && \
     (defined(_M_X64) || defined(_M_IX86))
@@ -71,18 +81,20 @@ extern "C" {
 #define WEBP_USE_AVX2
 #endif
 
-#if defined(__ANDROID__) && defined(__ARM_ARCH_7A__)
-#define WEBP_ANDROID_NEON  // Android targets that might support NEON
-#endif
-
 // The intrinsics currently cause compiler errors with arm-nacl-gcc and the
 // inline assembly would need to be modified for use with Native Client.
-#if (defined(__ARM_NEON__) || defined(WEBP_ANDROID_NEON) || \
+#if (defined(__ARM_NEON__) || \
      defined(__aarch64__) || defined(WEBP_HAVE_NEON)) && \
     !defined(__native_client__)
 #define WEBP_USE_NEON
 #endif
 
+#if !defined(WEBP_USE_NEON) && defined(__ANDROID__) && \
+    defined(__ARM_ARCH_7A__) && defined(HAVE_CPU_FEATURES_H)
+#define WEBP_ANDROID_NEON  // Android targets that may have NEON
+#define WEBP_USE_NEON
+#endif
+
 #if defined(_MSC_VER) && _MSC_VER >= 1700 && defined(_M_ARM)
 #define WEBP_USE_NEON
 #define WEBP_USE_INTRINSICS
@@ -144,7 +156,8 @@ typedef enum {
   kNEON,
   kMIPS32,
   kMIPSdspR2,
-  kMSA
+  kMSA,
+  kWASM
 } CPUFeature;
 // returns true if the CPU supports the feature.
 typedef int (*VP8CPUInfo)(CPUFeature feature);

src/dsp/neon.h

@@ -17,8 +17,9 @@
 #include "./dsp.h"
 
 // Right now, some intrinsics functions seem slower, so we disable them
-// everywhere except aarch64 where the inline assembly is incompatible.
+// everywhere except newer clang/gcc or aarch64 where the inline assembly is
-#if defined(__aarch64__)
+// incompatible.
+#if LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,9) || defined(__aarch64__)
 #define WEBP_USE_INTRINSICS   // use intrinsics when possible
 #endif
 
@@ -43,7 +44,7 @@
 // if using intrinsics, this flag avoids some functions that make gcc-4.6.3
 // crash ("internal compiler error: in immed_double_const, at emit-rtl.").
 // (probably similar to gcc.gnu.org/bugzilla/show_bug.cgi?id=48183)
-#if !(LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
+#if !(LOCAL_CLANG_PREREQ(3,8) || LOCAL_GCC_PREREQ(4,8) || defined(__aarch64__))
 #define WORK_AROUND_GCC
 #endif
 

src/utils/utils.h

@@ -66,7 +66,7 @@ WEBP_EXTERN(void) WebPSafeFree(void* const ptr);
 // memcpy() is the safe way of moving potentially unaligned 32b memory.
 static WEBP_INLINE uint32_t WebPMemToUint32(const uint8_t* const ptr) {
   uint32_t A;
-  memcpy(&A, (const int*)ptr, sizeof(A));
+  memcpy(&A, ptr, sizeof(A));
   return A;
 }
 static WEBP_INLINE void WebPUint32ToMem(uint8_t* const ptr, uint32_t val) {