From 20ece4b12a8d0f0d545b36cb0ec6fc6a2d29dc81 Mon Sep 17 00:00:00 2001 From: Matt Corallo Date: Sat, 16 Dec 2023 19:06:50 +0000 Subject: [PATCH] with less shuf --- lightning/src/util/simd_f32.rs | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-) diff --git a/lightning/src/util/simd_f32.rs b/lightning/src/util/simd_f32.rs index 68db25af4..c7c04baac 100644 --- a/lightning/src/util/simd_f32.rs +++ b/lightning/src/util/simd_f32.rs @@ -146,14 +146,10 @@ mod x86_sse { let res_b = _mm256_unpackhi_epi32(lo, zeros); let suma = _mm256_add_epi64(res_a, res_b); - let res_a = _mm256_unpacklo_epi64(suma, zeros); - let res_b = _mm256_unpackhi_epi64(suma, zeros); - - let sumb = _mm256_add_epi64(res_a, res_b); let mut res_bytes = AlignedInts([0; 4]); - _mm256_store_si256(&mut res_bytes.0[0] as *mut u64 as *mut __m256i, sumb); - res_bytes.0[0] + res_bytes.0[2] + _mm256_store_si256(&mut res_bytes.0[0] as *mut u64 as *mut __m256i, suma); + res_bytes.0[0] + res_bytes.0[1] + res_bytes.0[2] + res_bytes.0[3] } } } -- 2.39.5