From a284a7da8d1facbf984a22302665a2a50295a687 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Wed, 08 Aug 2018 16:08:58 +0000
Subject: [PATCH] Try to use avx_hs() - slow and requires alignment 4096 bits < (l.size*l.size*l.c) May be faster only from 8192 bits and more.
---
src/im2col_kernels.cu | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/src/im2col_kernels.cu b/src/im2col_kernels.cu
index d42d600..8a15e50 100644
--- a/src/im2col_kernels.cu
+++ b/src/im2col_kernels.cu
@@ -54,7 +54,7 @@
int width_col = (width + 2 * pad - ksize) / stride + 1;
int num_kernels = channels * height_col * width_col;
im2col_gpu_kernel<<<(num_kernels+BLOCK-1)/BLOCK,
- BLOCK>>>(
+ BLOCK, 0, get_cuda_stream()>>>(
num_kernels, im, height, width, ksize, pad,
stride, height_col,
width_col, data_col);
--
Gitblit v1.10.0