summaryrefslogtreecommitdiff
path: root/test/routines/levelx/xim2col.hpp
diff options
context:
space:
mode:
Diffstat (limited to 'test/routines/levelx/xim2col.hpp')
-rw-r--r--test/routines/levelx/xim2col.hpp34
1 files changed, 23 insertions, 11 deletions
diff --git a/test/routines/levelx/xim2col.hpp b/test/routines/levelx/xim2col.hpp
index 4124190f..ebffe85e 100644
--- a/test/routines/levelx/xim2col.hpp
+++ b/test/routines/levelx/xim2col.hpp
@@ -84,17 +84,29 @@ public:
// Describes how to run the CLBlast routine
static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
- auto queue_plain = queue();
- auto event = cl_event{};
- auto status = Im2col<T>(args.channels, args.height, args.width,
- args.kernel_h, args.kernel_w,
- args.pad_h, args.pad_w,
- args.stride_h, args.stride_w,
- args.dilation_h, args.dilation_w,
- buffers.a_mat(), args.a_offset,
- buffers.b_mat(), args.b_offset,
- &queue_plain, &event);
- if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+ #ifdef OPENCL_API
+ auto queue_plain = queue();
+ auto event = cl_event{};
+ auto status = Im2col<T>(args.channels, args.height, args.width,
+ args.kernel_h, args.kernel_w,
+ args.pad_h, args.pad_w,
+ args.stride_h, args.stride_w,
+ args.dilation_h, args.dilation_w,
+ buffers.a_mat(), args.a_offset,
+ buffers.b_mat(), args.b_offset,
+ &queue_plain, &event);
+ if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
+ #elif CUDA_API
+ auto status = Im2col<T>(args.channels, args.height, args.width,
+ args.kernel_h, args.kernel_w,
+ args.pad_h, args.pad_w,
+ args.stride_h, args.stride_w,
+ args.dilation_h, args.dilation_w,
+ buffers.a_mat(), args.a_offset,
+ buffers.b_mat(), args.b_offset,
+ queue.GetContext()(), queue.GetDevice()());
+ cuStreamSynchronize(queue());
+ #endif
return status;
}