From 69c5e9c07941212ac77368effd1c60db3140d4a3 Mon Sep 17 00:00:00 2001 From: Nikita Kostovsky Date: Sat, 8 Nov 2025 18:24:01 +0100 Subject: use vld1q_u32/vst1q_u32 for memcpy --- src/mem_utils.h | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+) create mode 100644 src/mem_utils.h (limited to 'src/mem_utils.h') diff --git a/src/mem_utils.h b/src/mem_utils.h new file mode 100644 index 0000000..8601f78 --- /dev/null +++ b/src/mem_utils.h @@ -0,0 +1,25 @@ +#pragma once + +#include + +#include + +template +void memcpy_1by1(std::byte *dst, const std::byte *src) +{ + for (std::size_t i{0}; i < S; ++i) { + dst[i] = src[i]; + } +} + +using ARRAY_TYPE = uint32_t; +template +void memcpy_neon(ARRAY_TYPE *dst, const ARRAY_TYPE *src) +{ + uint32x4_t tmp; + + for (std::size_t i{0}; i < (S / 4); i += 4) { + tmp = vld1q_u32(src + i); + vst1q_u32(&dst[i], tmp); + } +} -- cgit v1.2.3-70-g09d2