void add_arguments_to_kernel_rec(evnt_vec& events, args_vec& arguments,
T0& arg0, Ts&... args) {
cl_int err{0};
size_t buffer_size = sizeof(typename T0::value_type) * arg0.size();
auto buffer = clCreateBuffer(m_context.get(),
CL_MEM_READ_ONLY,
buffer_size,
nullptr,
&err);
if (err != CL_SUCCESS) {
CPPA_LOGMF(CPPA_ERROR, "clCreateBuffer: " << get_opencl_error(err));
return;
}
cl_event event;
err = clEnqueueWriteBuffer(m_queue.get(), buffer, CL_FALSE, 0,
buffer_size, arg0.data(),
0, nullptr, &event);
if (err != CL_SUCCESS) {
CPPA_LOGMF(CPPA_ERROR, "clEnqueueWriteBuffer: "
<< get_opencl_error(err));
return;
}
events.push_back(std::move(event));
mem_ptr tmp;
tmp.adopt(std::move(buffer));
arguments.push_back(tmp);
add_arguments_to_kernel_rec(events, arguments, args...);
}
void add_arguments_to_kernel_rec(evnt_vec&, args_vec& arguments) {
cl_int err{0};
// rotate left (output buffer to the end)
rotate(begin(arguments), begin(arguments) + 1, end(arguments));
for(size_t i = 0; i < arguments.size(); ++i) {
err = clSetKernelArg(m_kernel.get(), i, sizeof(cl_mem),
static_cast<void*>(&arguments[i]));
CPPA_LOG_ERROR_IF(err != CL_SUCCESS,
"clSetKernelArg: " << get_opencl_error(err));
}
clFlush(m_queue.get());
}
void enqueue_read_buffers(cl_event& kernel_done, detail::int_list<I, Is...>) {
using container_type =
typename std::tuple_element<I, std::tuple<Ts...>>::type;
using value_type = typename container_type::value_type;
cl_event event;
auto size = result_sizes_[I];
auto buffer_size = sizeof(value_type) * result_sizes_[I];
std::get<I>(result_buffers_).resize(size);
auto err = clEnqueueReadBuffer(queue_.get(), output_buffers_[I].get(),
CL_FALSE, 0, buffer_size,
std::get<I>(result_buffers_).data(),
1, &kernel_done, &event);
if (err != CL_SUCCESS) {
this->deref(); // failed to enqueue command
throw std::runtime_error("clEnqueueReadBuffer: " +
get_opencl_error(err));
}
mem_out_events_.push_back(std::move(event));
enqueue_read_buffers(kernel_done, detail::int_list<Is...>{});
}
