Merged latest reduction improvements

This commit is contained in:
Benoit Steiner
2016-05-26 12:19:33 -07:00
3 changed files with 73 additions and 0 deletions

View File

@@ -255,8 +255,13 @@ void test_cuda_reductions(int size1, int size2, int redux) {
Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream);
<<<<<<< local
int num_elem = size1*size2;
int result_size = (redux == 1 ? size1 : size2);
=======
int size = 40;
int num_elem = size*size;
>>>>>>> other
float* d_float1 = (float*)gpu_device.allocate(num_elem * sizeof(float));
float* d_float2 = (float*)gpu_device.allocate(num_elem * sizeof(float));