Hello,
I need to offload a ZGEMM call to the xeon phi. Unfortunately, the compiler is complaining that "variable 'A' used in this offload region is not bitwise copyable". However, A is just a MKL_Complex16 pointer. So far the MKL_Complex16 type never gave me any problems.
Since MKL_Complex16 is just a plain struct with two doubles, the following explanation does not really apply: https://software.intel.com/en-us/articles/effective-use-of-the-intel-compilers-offload-features
void foo(MKL_Complex16 *A, MKL_Complex16 *B, MKL_Complex16 *C, ... some more arguments...)
{
//... some initialization ...
alpha = MKL_Complex16 (sigma_1 / e, 0.0);
beta = MKL_Complex16 (0.0, 0.0);
const int matrix_elements = n*n;
#pragma offload target(mic:0) \
in(A: length(matrix_elements) free_if(1)) \
in(B: length(matrix_elements) free_if(1)) \
in(C:length(matrix_elements)) \
out(C:length(matrix_elements) free_if(1))
zgemm("N", "N", &n, &n, &n, &alpha, (const MKL_Complex16*)(A), &n,
(const MKL_Complex16*)B, &n, &beta, C, &n);
//... something else ...
}
What could be the problem here?
Thanks,
Paul
