From 57e878b4f9512cf9995ff6b5cd6e0d7dc1da9eaf Mon Sep 17 00:00:00 2001 From: AlexeyAB Date: Fri, 7 Sep 2018 15:25:37 +0300 Subject: [PATCH] another compile fix --- src/im2col_kernels.cu | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/src/im2col_kernels.cu b/src/im2col_kernels.cu index e545260d..f407e976 100644 --- a/src/im2col_kernels.cu +++ b/src/im2col_kernels.cu @@ -131,10 +131,11 @@ __global__ void float_to_bit_gpu_kernel(float *src, unsigned char *dst, size_t s //for (; index < size_aligned; index += blockDim.x*gridDim.x) { - src_val = src[index]; - //if(index < size) src_val = src[index]; - //else src_val = 0; - unsigned int bit_mask = __ballot_sync(0xffffffff, src_val > 0); + //src_val = src[index]; + if(index < size) src_val = src[index]; + else src_val = 0; + //unsigned int bit_mask = __ballot_sync(0xffffffff, src_val > 0); + unsigned int bit_mask = __ballot(src_val > 0); if (threadIdx.x % WARP_SIZE == 0) ((unsigned int*)dst)[index / 32] = bit_mask; } }