Avoid scalarization in _mm_madd_epi16

When compiling with `-munimplemented-simd128`.

Author: kleisauke

system/include/compat/emmintrin.h

@@ -669,6 +669,9 @@ _mm_avg_epu16(__m128i __a, __m128i __b)
 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_madd_epi16(__m128i __a, __m128i __b)
 {
+#ifdef __wasm_unimplemented_simd128__
+  return (__m128i)__builtin_wasm_dot_s_i32x4_i16x8((v128_t)__a, (v128_t)__b);
+#else
   // TODO: optimize
   union {
     signed short x[8];
@@ -683,6 +686,7 @@ _mm_madd_epi16(__m128i __a, __m128i __b)
   for(int i = 0; i < 4; ++i)
     dst.x[i] = src.x[i*2] * src2.x[i*2] + src.x[i*2+1] * src2.x[i*2+1];
   return dst.m;
+#endif
 }
 
 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))