I use Fortran PGI 14.10 with OpenACC.
If I use the compiler-option “-cuda” the environment-variable “ACC_DEVICE_NUM” will not be evaluated from my program.
I use -Mcuda to create data only on the device (https://forums.developer.nvidia.com/t/how-to-create-data-on-device/134350/1)
I have a computer with 4 GPU’s and I’d like to be able to select the device.
I tried to call acc_set_device_num “by hand”. But the I get an errormessage
line 41: cudaLaunch returned status 33: invalid resource handle