We are really excited about this new version and were hoping to try it on the Summit supercomputer at the OLCF. Unfortunately Summit is IBM and Power9 architecture and we do not have the MKL library from Intel. We are trying to build the code with alternate libraries (pgi, netlib), but run into the problem that the MKL vderfc function is not available.
Is there anyway to work around this? We are tempted to alter the code to use a non-vectorized erfc but since we are not familiar with it, we are hesitant. We are hoping to make use of the acceleration on the GPU so it is possible that performance loss with not having this function vectorized would be reduced if more more is done on the GPU.
HI!
We are really excited about this new version and were hoping to try it on the Summit supercomputer at the OLCF. Unfortunately Summit is IBM and Power9 architecture and we do not have the MKL library from Intel. We are trying to build the code with alternate libraries (pgi, netlib), but run into the problem that the MKL vderfc function is not available.
Is there anyway to work around this? We are tempted to alter the code to use a non-vectorized erfc but since we are not familiar with it, we are hesitant. We are hoping to make use of the acceleration on the GPU so it is possible that performance loss with not having this function vectorized would be reduced if more more is done on the GPU.
Thanks so much for any help!