diff --git a/src/util/arch/arm/simd_utils.h b/src/util/arch/arm/simd_utils.h
index f7b92e70..dcf3fe58 100644
--- a/src/util/arch/arm/simd_utils.h
+++ b/src/util/arch/arm/simd_utils.h
@@ -202,6 +202,18 @@ static really_inline u64a extract64from128(const m128 in, unsigned imm) {
 #endif
 }
 
+static really_inline m128 low64from128(const m128 in) {
+    return vcombine_u64(vget_low_u64(in), vdup_n_u64(0));
+}
+
+static really_inline m128 high64from128(const m128 in) {
+    return vcombine_u64(vget_high_u64(in), vdup_n_u64(0));
+}
+
+static really_inline m128 add128(m128 a, m128 b) {
+    return (m128) vaddq_u64((uint64x2_t)a, (uint64x2_t)b);
+}
+
 static really_inline m128 and128(m128 a, m128 b) {
     return (m128) vandq_s8((int8x16_t)a, (int8x16_t)b);
 }
@@ -381,13 +393,13 @@ m128 sub_u8_m128(m128 a, m128 b) {
 
 static really_inline
 m128 set4x32(u32 x3, u32 x2, u32 x1, u32 x0) {
-    uint32_t __attribute__((aligned(16))) data[4] = { x0, x1, x2, x3 };
+    uint32_t ALIGN_ATTR(16) data[4] = { x0, x1, x2, x3 };
     return (m128) vld1q_u32((uint32_t *) data);
 }
 
 static really_inline
 m128 set2x64(u64a hi, u64a lo) {
-    uint64_t __attribute__((aligned(16))) data[2] = { lo, hi };
+    uint64_t ALIGN_ATTR(16) data[2] = { lo, hi };
     return (m128) vld1q_u64((uint64_t *) data);
 }
 
diff --git a/src/util/arch/common/simd_utils.h b/src/util/arch/common/simd_utils.h
index 0c67ee94..b20becdc 100644
--- a/src/util/arch/common/simd_utils.h
+++ b/src/util/arch/common/simd_utils.h
@@ -46,7 +46,7 @@
 
 #ifdef DEBUG
 static inline void print_m128_16x8(char *label, m128 vector) {
-    uint8_t __attribute__((aligned(16))) data[16];
+    uint8_t ALIGN_ATTR(16) data[16];
     store128(data, vector);
     DEBUG_PRINTF("%s: ", label);
     for(int i=0; i < 16; i++)
@@ -55,7 +55,7 @@ static inline void print_m128_16x8(char *label, m128 vector) {
 }
 
 static inline void print_m128_8x16(char *label, m128 vector) {
-    uint16_t __attribute__((aligned(16))) data[8];
+    uint16_t ALIGN_ATTR(16) data[8];
     store128(data, vector);
     DEBUG_PRINTF("%s: ", label);
     for(int i=0; i < 8; i++)
@@ -64,7 +64,7 @@ static inline void print_m128_8x16(char *label, m128 vector) {
 }
 
 static inline void print_m128_4x32(char *label, m128 vector) {
-    uint32_t __attribute__((aligned(16))) data[4];
+    uint32_t ALIGN_ATTR(16) data[4];
     store128(data, vector);
     DEBUG_PRINTF("%s: ", label);
     for(int i=0; i < 4; i++)
@@ -73,7 +73,7 @@ static inline void print_m128_4x32(char *label, m128 vector) {
 }
 
 static inline void print_m128_2x64(char *label, m128 vector) {
-    uint64_t __attribute__((aligned(16))) data[2];
+    uint64_t ALIGN_ATTR(16) data[2];
     store128(data, vector);
     DEBUG_PRINTF("%s: ", label);
     for(int i=0; i < 2; i++)
@@ -146,6 +146,13 @@ static really_inline m256 ones256(void) {
     return rv;
 }
 
+static really_inline m256 add256(m256 a, m256 b) {
+    m256 rv;
+    rv.lo = add128(a.lo, b.lo);
+    rv.hi = add128(a.hi, b.hi);
+    return rv;
+}
+
 static really_inline m256 and256(m256 a, m256 b) {
     m256 rv;
     rv.lo = and128(a.lo, b.lo);
@@ -585,6 +592,13 @@ m512 set1_4x128(m128 a) {
     return rv;
 }
 
+static really_inline
+m512 add512(m512 a, m512 b) {
+    m512 rv;
+    rv.lo = add256(a.lo, b.lo);
+    rv.hi = add256(a.hi, b.hi);
+    return rv;
+}
 
 static really_inline
 m512 and512(m512 a, m512 b) {