inline future<void>
dispatch_fill_async(BufferIterator first,
size_t count,
const T &value,
command_queue &queue,
typename boost::enable_if<
is_valid_fill_buffer_iterator<BufferIterator>
>::type* = 0)
{
typedef typename std::iterator_traits<BufferIterator>::value_type value_type;
// check if the device supports OpenCL 1.2 (required for enqueue_fill_buffer)
if(!queue.check_device_version(1, 2)){
return fill_async_with_copy(first, count, value, queue);
}
value_type pattern = static_cast<value_type>(value);
size_t offset = static_cast<size_t>(first.get_index());
event event_ =
queue.enqueue_fill_buffer(first.get_buffer(),
&pattern,
sizeof(value_type),
offset * sizeof(value_type),
count * sizeof(value_type));
return future<void>(event_);
}
inline void
dispatch_fill(BufferIterator first,
size_t count,
const T &value,
command_queue &queue,
typename boost::enable_if<
is_valid_fill_buffer_iterator<BufferIterator>
>::type* = 0)
{
typedef typename std::iterator_traits<BufferIterator>::value_type value_type;
if(count == 0){
// nothing to do
return;
}
// check if the device supports OpenCL 1.2 (required for enqueue_fill_buffer)
if(!queue.check_device_version(1, 2)){
return fill_with_copy(first, count, value, queue);
}
value_type pattern = static_cast<value_type>(value);
size_t offset = static_cast<size_t>(first.get_index());
if(count == 1){
// use clEnqueueWriteBuffer() directly when writing a single value
// to the device buffer. this is potentially more efficient and also
// works around a bug in the intel opencl driver.
queue.enqueue_write_buffer(
first.get_buffer(),
offset * sizeof(value_type),
sizeof(value_type),
&pattern
);
}
else {
queue.enqueue_fill_buffer(
first.get_buffer(),
&pattern,
sizeof(value_type),
offset * sizeof(value_type),
count * sizeof(value_type)
);
}
}