cory-master/BabelStream
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

proper declaration of tb_sum with HIP_DYNAMIC_SHARED macro

Browse Source
This commit is contained in:
Peter Steinbach 2017-02-28 10:07:48 +01:00
parent 350a151c3b
commit ceada6922f
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
HIPStream.cpp
View File

@ -188,9 +188,9 @@ template <class T>
__global__ void dot_kernel(hipLaunchParm lp, const T * a, const T * b, T * sum, unsigned int array_size) __global__ void dot_kernel(hipLaunchParm lp, const T * a, const T * b, T * sum, unsigned int array_size)
{ {
//extern __shared__ __align__(sizeof(T)) unsigned char smem[]; HIP_DYNAMIC_SHARED(T,tb_sum);
HIP_DYNAMIC_SHARED(unsigned char,smem); // HIP_DYNAMIC_SHARED(unsigned char,smem);
T *tb_sum = reinterpret_cast<T*>(smem); // T *tb_sum = reinterpret_cast<T*>(smem);
int i = hipBlockDim_x * hipBlockIdx_x + hipThreadIdx_x; int i = hipBlockDim_x * hipBlockIdx_x + hipThreadIdx_x;
const size_t local_i = hipThreadIdx_x; const size_t local_i = hipThreadIdx_x;