to _range or not to _range . . .

to _range or not to _range . . .

I'm completely new to this arbb and have been looking at the tutorials and the online docs.
I had a basic question, in tutorial0 there are these lines

bind(vec_out, out, vec_length);
call(sum_kernel)(vec_a, vec_b, vec_out);
output_result(out);

and later

dense out_only;
call(sum_kernel)(vec_a, vec_b, out_only);
const arbb::const_range output = out_only.read_only_range();
float result[vec_length];
for (unsigned int i = 0; i < vec_length; ++i) {
result[i] = output[i];
}
output_result(result);

So my question is, in the second case we have to use a const_range and call read_only_range()
but in the first case we don't? Wahy is that? The documentation says

"// You must call a read range function before reading the values"

Is this always required ? Is it just good luck that the first case works ?

--
jason

3 posts / novo 0
Último post
Para obter mais informações sobre otimizações de compiladores, consulte Aviso sobre otimizações.

Hi Jason,

Thanks for your post. It is a good catch!

In ArBB, there are two ways to create and manage dense objects. First, if you've got a C/C++ array then you can bind a dense object to it. The bound dense object can be passed to ArBB functions. This is the case in tutorial0. Second, you can create a new dense entirely in the ArBB memory space. This is the case in your second example.

No matter how a dense object was created, if it was modified by a function then a read_only_range must be obtained before you can safely access the content of the dense object. A read_only_range is like a synchronization point telling the ArBB runtime to complete all pending data copying for this container and bring it up to the latest state.

The example in tutorial0 is missing a read_only_range. It just happens to work because the current beta version of ArBB performs frequent synchronizations for bound dense objects. But this is not a behavior that you should rely on.

Zhang

Hi,

Thanks for the clarification, just trying to get it sorted out in my own mind how things should work.

Jason

Faça login para deixar um comentário.