Hi, I have read some issues of batched MM.
I think this may be featured in latest version 3.5.1, because this page (http://arrayfire.org/docs/page_gfor.htm) illustrates an usage of gfor-MM.
But, I can't apply a batched-MM as below.
gfor(af::seq i, n_sample)
{
af::array i_left = arr_left(af::span, af::span, i);
af::array i_right = arr_right(af::span, af::span, i);
arr(af::span, af::span, i) = af::matmul(i_left, i_right);
}
What should I do?
On the other hand, ArrayFire could not work with OpenMP as below.
#pragma omp parallel for
for(auto i=0; i<n_sample, ++i)
{
af::array i_left = arr_left(af::span, af::span, i);
af::array i_right = arr_right(af::span, af::span, i);
arr(af::span, af::span, i) = af::matmul(i_left, i_right);
}
It came up with an exception in af::gen_assign.
Could this be fixed?
Thanks, a lot.
Hi, I have read some issues of batched MM.
I think this may be featured in latest version 3.5.1, because this page (http://arrayfire.org/docs/page_gfor.htm) illustrates an usage of gfor-MM.
But, I can't apply a batched-MM as below.
What should I do?
On the other hand, ArrayFire could not work with OpenMP as below.
It came up with an exception in
af::gen_assign.Could this be fixed?
Thanks, a lot.