Couple of fixes to allow SIMD on Emscripten

 - Switch from `-msse` to `-msse4.2`. In theory, those should be identical, but Emscripten currently has a bug where only latter form is supported: emscripten-core/emscripten#12746. Since we're using only SSE 4.2, an explicit form seems to be preferable anyway.
 - Add `-msimd128` to flags to actually enable WebAssembly SIMD when performing SIMD detection. It's currently required in addition to `-msse*` / `-mfpu=neon` flags which only perform translation of corresponding intrinsics to Wasm SIMD ones. See a discussion at emscripten-core/emscripten#12714 for automating this and making easier in the future.
 - Remove compilation branch that prevented definitions of `WP2_USE_SSE` and `WP2_USE_NEON` on Emscripten even when SIMD support was detected at compile-time.
 - Add an implementation of `WP2GetCPUInfo` for Emscripten which uses static `WP2_USE_*` flags to determine if a corresponding SIMD instruction is supported. This is because Wasm doesn't have proper feature detection (yet) and requires making separate build for SIMD version anyway.

CMakeLists.txt

@@ -33,7 +33,12 @@ if(POLICY CMP0048)
 endif(POLICY CMP0048)
 
 # Options for coder / decoder executables.
-option(WP2_ENABLE_SIMD "Enable any SIMD optimization." ON)
+if(NOT EMSCRIPTEN)
+  # Disable SIMD on Emscripten by default, as it's a new unstable Wasm feature.
+  # Users can still explicitly opt-in to make a SIMD-enabled build.
+  set(WP2_ENABLE_SIMD_DEFAULT ON)
+endif()
+option(WP2_ENABLE_SIMD "Enable any SIMD optimization." ${WP2_ENABLE_SIMD_DEFAULT})
 set(WP2_ENABLE_CONTEXT_SWITCH
     AUTO
     CACHE STRING "Use context switch optimization for incremental decoding.")
@@ -96,7 +101,6 @@ endif()
 
 # disable some configurations for WP2_JS
 if(WP2_BUILD_WP2_JS)
-  set(WP2_ENABLE_SIMD OFF)
   set(WP2_ENABLE_TESTS OFF)
 endif()
 

cmake/cpu.cmake

@@ -56,7 +56,7 @@ if(MSVC)
   set(SIMD_ENABLE_FLAGS "/arch:AVX;")
   set(SIMD_DISABLE_FLAGS)
 else()
-  set(SIMD_ENABLE_FLAGS "-msse4;-mfpu=neon")
+  set(SIMD_ENABLE_FLAGS "-msse4.2;-mfpu=neon")
   set(SIMD_DISABLE_FLAGS "-mno-sse4;")
 endif()
 
@@ -83,6 +83,9 @@ foreach(I_SIMD RANGE ${WP2_SIMD_DEFINES_RANGE})
   wp2_check_compiler_flag(${WP2_SIMD_DEFINE} ${WP2_ENABLE_SIMD})
   if(NOT WP2_HAVE_${WP2_SIMD_DEFINE})
     list(GET SIMD_ENABLE_FLAGS ${I_SIMD} SIMD_COMPILE_FLAG)
+    if(EMSCRIPTEN)
+      set(SIMD_COMPILE_FLAG "-msimd128 ${SIMD_COMPILE_FLAG}")
+    endif()
     set(CMAKE_REQUIRED_FLAGS ${SIMD_COMPILE_FLAG})
     wp2_check_compiler_flag(${WP2_SIMD_DEFINE} ${WP2_ENABLE_SIMD})
   else()

src/dsp/cpu.cc

@@ -191,6 +191,36 @@ static bool AndroidCPUInfo(WP2CPUFeature feature) {
   return false;
 }
 WP2CPUInfo WP2GetCPUInfo = AndroidCPUInfo;
+#elif defined(EMSCRIPTEN) // also needs to be before generic NEON test
+// Use compile flags as an indicator of SIMD support instead of a runtime check.
+static bool wasmCPUInfo(WP2CPUFeature feature) {
+  switch (feature) {
+    case kSSE2:
+    case kSSE3:
+    case kSlowSSSE3:
+    case kSSE4_1:
+    case kSSE4_2:
+    case kSSE:
+#ifdef WP2_USE_SSE
+      return true;
+#endif
+      return false;
+    case kAVX:
+    case kAVX2:
+#ifdef WP2_USE_AVX2:
+      return true;
+#endif
+      return false;
+    case kNEON:
+#ifdef WP2_USE_NEON
+      return true;
+#endif
+      return false;
+    default:
+      return false;
+  }
+}
+WP2CPUInfo WP2GetCPUInfo = wasmCPUInfo;
 #elif defined(WP2_USE_NEON)
 // define a dummy function to enable turning off NEON at runtime by setting
 // WP2DecGetCPUInfo = NULL

src/dsp/dsp.h

@@ -56,9 +56,6 @@ static_assert(_MSC_VER >= 1900, "Visual Studio needs to be at version >= 2015");
 # define __has_builtin(x) 0
 #endif
 
-// for now, none of the optimizations below are available in emscripten
-#if !defined(EMSCRIPTEN)
-
 #if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
 #define WP2_MSC_SSE   // Visual C++ SSE4.2 targets
 #endif
@@ -113,8 +110,6 @@ static_assert(_MSC_VER >= 1900, "Visual Studio needs to be at version >= 2015");
 #define WP2_USE_MSA
 #endif
 
-#endif  /* EMSCRIPTEN */
-
 // This macro prevents thread_sanitizer from reporting known concurrent writes.
 #define WP2_TSAN_IGNORE_FUNCTION
 #if defined(__has_feature)