eago

cudaMalloc can no longer guarantee to return 64kB aligned address

3

GDRDRV needs 64kB aligned addresses. ``` gdrdrv_pin_buffer() { ... page_virt_start = params.addr & GPU_PAGE_MASK; page_virt_end = params.addr + params.size - 1; rounded_size = page_virt_end - page_virt_start + 1; mr->offset =...

On Ubuntu 20.04 with MOFED 5.7 and python3.8 I get this error: ``` ~/trex-core-2.96/scripts$ sudo ./t-rex-64 --cfg /etc/trex_cfg.yaml -i -c 12 Error while determining OS type: module 'platform' has no...

I tested on the Wilkes cluster (Tesla K20 GPUs) the CoMD-CUDA implementation using up to 16 nodes, and I got some errors: 8 processes, crash. All the processes on the...

In my asynchronous implementation ( https://github.com/e-ago/CoMD-CUDA-Async ) using runs with parameters like "-e -x 80 -y 80 -z 80" with up to 16 process I noticed that after the first...