提交 345737d0 编写于 作者: C chengduoZH

add sync

上级 187e23a7
......@@ -63,6 +63,7 @@ __device__ T reduceSum(T val, int tid, int len) {
val += platform::CudaShuffleDownSync(mask, val, offset);
if (tid < warpSize) shm[tid] = 0;
__syncthreads();
if (tid % warpSize == 0) {
shm[tid / warpSize] = val;
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册