From 6c6c40edb37d672c65e17c0ad7db336b4909b7f8 Mon Sep 17 00:00:00 2001 From: nihui Date: Fri, 27 Oct 2023 19:07:39 +0800 Subject: [PATCH] fix deconvolution x86 unaligned bias load (#5112) --- src/layer/x86/deconvolution_x86.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/layer/x86/deconvolution_x86.cpp b/src/layer/x86/deconvolution_x86.cpp index 60e463507ca..fa959dcafe1 100644 --- a/src/layer/x86/deconvolution_x86.cpp +++ b/src/layer/x86/deconvolution_x86.cpp @@ -300,7 +300,7 @@ int Deconvolution_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Opti } else { - outm.fill(_mm512_load_ps((const float*)bias_data + p * 16)); + outm.fill(_mm512_loadu_ps((const float*)bias_data + p * 16)); } for (int u = 0; u < kernel_h; u++) @@ -344,7 +344,7 @@ int Deconvolution_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Opti } else { - outm.fill(_mm256_load_ps((const float*)bias_data + p * 8)); + outm.fill(_mm256_loadu_ps((const float*)bias_data + p * 8)); } for (int u = 0; u < kernel_h; u++) @@ -388,7 +388,7 @@ int Deconvolution_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Opti } else { - outm.fill(_mm_load_ps((const float*)bias_data + p * 4)); + outm.fill(_mm_loadu_ps((const float*)bias_data + p * 4)); } for (int u = 0; u < kernel_h; u++)