I have an embedded motherboard from Fujitsu with a GX-424CC apu on it. As I understand it this is a puma cpu core with a mullins sea island gpu. I am running Ubuntu 16.04.04 and have installed amdgpu-pro 17.50 with opencl set to legacy.
I am trying to run a custom opencl application on the hardware and am seeing a number of kernel errors and opecl hangs after running for a long time (can happen after 5minutes, but some times up to 48 hours). The kernel errors I see are:
[ 99.348774] amdgpu 0000:00:01.0: GPU fault detected: 146 0x08492014
[ 99.355041] amdgpu 0000:00:01.0: VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x00103042
[ 99.362509] amdgpu 0000:00:01.0: VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x09020014
[ 99.369980] VM fault (0x14, vmid 4) at page 1060930, write from 'TC0' (0x54433000) (32)
[ 7612.741693] amdgpu 0000:00:01.0: IH ring buffer overflow (0x00000010, 0x00000000, 0x00000020)
So my first question is amdgpu-pro even supported on this hardware? I believe it should be fine as it is a sea island part? Should I be running another driver instead (catalyst?)
Secondly any suggestions on how to narrow down this bug. From other posts it seems like this may be a bug in my opencl code or could be a driver issue. How do I tell?
Looking at other posts on the forum with similar kernel errors people have suggested going back to old drivers. Is that likely to help?
If it is a driver bug how should this be reported to AMD?
Thanks for any help