Updated meshoptimizer.

3rdparty/meshoptimizer/src/vertexcodec.cpp

@@ -42,10 +42,8 @@
 #endif
 
 // When targeting Wasm SIMD we can't use runtime cpuid checks so we unconditionally enable SIMD
-// Note that we need unimplemented-simd128 subset for a few functions that are implemented de-facto
 #if defined(__wasm_simd128__)
 #define SIMD_WASM
-#define SIMD_TARGET __attribute__((target("unimplemented-simd128")))
 #endif
 
 #ifndef SIMD_TARGET
@@ -100,15 +98,6 @@
 #define wasmx_unpackhi_v64x2(a, b) wasm_v64x2_shuffle(a, b, 1, 3)
 #endif
 
-#if defined(SIMD_WASM)
-// v128_t wasm_v8x16_swizzle(v128_t a, v128_t b)
-SIMD_TARGET
-static __inline__ v128_t wasm_v8x16_swizzle(v128_t a, v128_t b)
-{
-	return (v128_t)__builtin_wasm_swizzle_v8x16((__i8x16)a, (__i8x16)b);
-}
-#endif
-
 namespace meshopt
 {
 
@@ -769,6 +758,7 @@ static void wasmMoveMask(v128_t mask, unsigned char& mask0, unsigned char& mask1
 	uint64_t mask_1a = wasm_i64x2_extract_lane(mask_0, 0) & 0x0804020108040201ull;
 	uint64_t mask_1b = wasm_i64x2_extract_lane(mask_0, 1) & 0x8040201080402010ull;
 
+	// TODO: This can use v8x16_bitmask in the future
 	uint64_t mask_2 = mask_1a | mask_1b;
 	uint64_t mask_4 = mask_2 | (mask_2 >> 16);
 	uint64_t mask_8 = mask_4 | (mask_4 >> 8);

3rdparty/meshoptimizer/src/vertexfilter.cpp

@@ -735,7 +735,8 @@ static void decodeFilterQuatSimd(short* data, size_t count)
 		v128_t res_1 = wasmx_unpackhi_v16x8(wyr, xzr);
 
 		// compute component index shifted left by 4 (and moved into i32x4 slot)
-		v128_t cm = wasm_i32x4_shl(cf, 4);
+		// TODO: volatile here works around LLVM mis-optimizing code; https://github.com/emscripten-core/emscripten/issues/11449
+		volatile v128_t cm = wasm_i32x4_shl(cf, 4);
 
 		// rotate and store
 		uint64_t* out = reinterpret_cast<uint64_t*>(&data[i * 4]);