fix arm build

diff --git a/src/util/arch/arm/simd_utils.h b/src/util/arch/arm/simd_utils.h
index 858866d77b08a48c1c626af36610d5797db11b29..b097e66a883b4f4152de531b1d2e5c14401b99a9 100644 (file)
--- a/src/util/arch/arm/simd_utils.h
+++ b/src/util/arch/arm/simd_utils.h
@@ -182,7 +182,7 @@ static really_inline m128 set1_2x64(u64a c) {
 }
 
 static really_inline m128 insert32_m128(m128 in, u32 val, const int imm) {
-    return vsetq_lane_u32((uint32x4_t)in, val, imm);
+    return (m128) vsetq_lane_u32(val, (uint32x4_t)in, imm);
 }
 
 static really_inline u32 movd(const m128 in) {
@@ -199,12 +199,6 @@ m128 load_m128_from_u64a(const u64a *p) {
     return (m128) vsetq_lane_u64(*p, (uint64x2_t) zeroes128(), 0);
 }
 
-/* another form of movq */
-static really_inline
-m128 load_m128_from_u64a(const u64a *p) {
-    return (m128) vsetq_lane_u64(*p, (uint64x2_t) zeroes128(), 0);
-}
-
 static really_inline u32 extract32from128(const m128 in, unsigned imm) {
 #if defined(HAVE__BUILTIN_CONSTANT_P)
     if (__builtin_constant_p(imm)) {
@@ -449,4 +443,14 @@ m128 set2x64(u64a hi, u64a lo) {
     return (m128) vld1q_u64((uint64_t *) data);
 }
 
+static really_inline
+m128 widenlo128(m128 x) {
+    return (m128) vmovl_u32(vget_low_u32((uint32x4_t)x));
+}
+
+static really_inline
+m128 widenhi128(m128 x) {
+    return (m128) vmovl_u32(vget_high_u32((uint32x4_t)x));
+}
+
 #endif // ARCH_ARM_SIMD_UTILS_H