diff --git a/include/bx/simd256_ref.inl b/include/bx/simd256_ref.inl
index 15e2754..c781a30 100644
--- a/include/bx/simd256_ref.inl
+++ b/include/bx/simd256_ref.inl
@@ -15,25 +15,25 @@ namespace bx
 	{
 		const simd128_t* ptr = reinterpret_cast<const simd128_t*>(_ptr);
 		simd256_ref_t result;
-		result.simd128[0] = simd_ld<simd128_t>(&ptr[0]);
-		result.simd128[1] = simd_ld<simd128_t>(&ptr[1]);
+		result.simd128_0 = simd_ld<simd128_t>(&ptr[0]);
+		result.simd128_1 = simd_ld<simd128_t>(&ptr[1]);
 		return result;
 	}
 
 	template<>
-	BX_SIMD_FORCE_INLINE void simd_st(void* _ptr, simd256_ref_t _a)
+	BX_SIMD_FORCE_INLINE void simd_st(void* _ptr, simd256_ref_t& _a)
 	{
 		simd128_t* result = reinterpret_cast<simd128_t*>(_ptr);
-		simd_st<simd128_t>(&result[0], _a.simd128[0]);
-		simd_st<simd128_t>(&result[1], _a.simd128[1]);
+		simd_st<simd128_t>(&result[0], _a.simd128_0);
+		simd_st<simd128_t>(&result[1], _a.simd128_1);
 	}
 
 	template<>
 	BX_SIMD_FORCE_INLINE simd256_ref_t simd_ld(float _x, float _y, float _z, float _w, float _A, float _B, float _C, float _D)
 	{
 		simd256_ref_t result;
-		result.simd128[0] = simd_ld<simd128_t>(_x, _y, _z, _w);
-		result.simd128[1] = simd_ld<simd128_t>(_A, _B, _C, _D);
+		result.simd128_0 = simd_ld<simd128_t>(_x, _y, _z, _w);
+		result.simd128_1 = simd_ld<simd128_t>(_A, _B, _C, _D);
 		return result;
 	}
 
@@ -41,8 +41,8 @@ namespace bx
 	BX_SIMD_FORCE_INLINE simd256_ref_t simd_ild(uint32_t _x, uint32_t _y, uint32_t _z, uint32_t _w, uint32_t _A, uint32_t _B, uint32_t _C, uint32_t _D)
 	{
 		simd256_ref_t result;
-		result.simd128[0] = simd_ild<simd128_t>(_x, _y, _z, _w);
-		result.simd128[1] = simd_ild<simd128_t>(_A, _B, _C, _D);
+		result.simd128_0 = simd_ild<simd128_t>(_x, _y, _z, _w);
+		result.simd128_1 = simd_ild<simd128_t>(_A, _B, _C, _D);
 		return result;
 	}
 
diff --git a/include/bx/simd_t.h b/include/bx/simd_t.h
index c10a337..14e3789 100644
--- a/include/bx/simd_t.h
+++ b/include/bx/simd_t.h
@@ -400,9 +400,10 @@ namespace bx
 	typedef simd128_ref_t simd128_t;
 #endif //
 
-	union simd256_ref_t
+	BX_ALIGN_DECL(32, struct) simd256_ref_t
 	{
-		simd128_t simd128[2];
+		simd128_t simd128_0;
+		simd128_t simd128_1;
 	};
 
 #if !BX_SIMD_AVX