OpenACC in an AMD-APU

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Good afternoon,
I wonder if anyone has managed to compile with OpenACC on an AMD-APU.

After a series of tricks, the pgaccelinfo returns two devices:
OpenCL Platform: AMD Accelerated Parallel Processing
OpenCL Vendor: Advanced Micro Devices, Inc.
Device Number: 0

and

Device Number: 1
Device Name: AMD A10-6800K APU with Radeon ™ HD Graphics
Available: Yes
Compiler Available: Yes


But when I try to run the binary executable , it doesn’t seem to find the device: The output obtained is
“Accelerator Fatal Error: No AMD GPU version of this construct available for the current device”

Which compilation flags are necessary? I used -ta=acc-radeon:specter without much success.

I’ve added in the source code acc_init(mydevice) with both mydevice=acc_device_not_host and mydevice=acc_device_radeon but it cashes with the same error: “Accelerator Fatal Error: No AMD GPU version of this construct available for the current device”

Any ideas?

Many thanks

Carlos

2

Hi Carlos,

Unfortunately, we don’t support these older “Trinity” based APUs. Currently only “Spectre”, “Cape Verde” or the “Tahiti” architecture.

See: PGI Compilers with OpenACC | PGI

  • Mat