ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add workaround for failing overhead estimation test on (Intel) GPUs that only support single precision

Browse Source
This commit is contained in:
Axel Kohlmeyer
2023-07-18 13:24:11 -04:00
parent 34fcf1361f
commit 021eeae860
1 changed files with 8 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
lib/gpu/lal_device.cpp
View File

@ -745,7 +745,14 @@ void DeviceT::estimate_gpu_overhead(const int kernel_calls,
gpu_overhead=0.0; gpu_overhead=0.0;
gpu_driver_overhead=0.0; gpu_driver_overhead=0.0;
for (int z=0; z<11; z++) { // TODO: XXX
// The following estimation currently fails on Intel GPUs
// that do not support double precision with OpenCL error code -5.
// Until we have a better solution, we just skip this test in this case.
int zloops = 11;
if (!gpu->double_precision()) zloops = 0;
for (int z=0; z < zloops; z++) {
gpu->sync(); gpu->sync();
gpu_barrier(); gpu_barrier();
over_timer.start(); over_timer.start();