// Describes how to run the CLBlast routine
static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
#ifdef OPENCL_API
auto queue_plain = queue();
auto event = cl_event{};
auto status = Col2im<T>(args.kernel_mode,
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.b_mat(), args.b_offset, // col
buffers.a_mat(), args.a_offset, // im
&queue_plain, &event);
if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
#elif CUDA_API
auto status = Col2im<T>(args.kernel_mode,
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.b_mat(), args.b_offset, // col
buffers.a_mat(), args.a_offset, // im
queue.GetContext()(), queue.GetDevice()());
cuStreamSynchronize(queue());
#endif
return status;
}
// 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 = Omatcopy<T>(args.layout, args.a_transpose,
args.m, args.n, args.alpha,
buffers.a_mat(), args.a_offset, args.a_ld,
buffers.b_mat(), args.b_offset, args.b_ld,
&queue_plain, &event);
if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
return status;
}
// 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 = Hemm(args.layout, args.side, args.triangle,
args.m, args.n, args.alpha,
buffers.a_mat(), args.a_offset, args.a_ld,
buffers.b_mat(), args.b_offset, args.b_ld, args.beta,
buffers.c_mat(), args.c_offset, args.c_ld,
&queue_plain, &event);
clWaitForEvents(1, &event);
return status;
}