AnsweredAssumed Answered

Hang in clFinish on gfx902 (Vega M GH)

Question asked by timchist on Mar 12, 2020
Latest reply on Apr 9, 2020 by timchist

Steps to reproduce:

1. Prepare a binary for gfx902 using the CL_CONTEXT_OFFLINE_DEVICES_AMD approach (offline compilation)

2. Find the AMD platform by name ('Advanced Micro Devices, Inc')

3. Create an OpenCL context with CL_DEVICE_TYPE_GPU

4. Load and build the program (clCreateProgramWithBinary, clBuildProgram)

5. Create 4 queues

6. Enqueue several kernel calls in one of the queues

7. Call clFinish on the queues

 

Expected outcome:

clFinish returns

 

Observed outcome:

clFinish hangs, call stack:

 

     ntdll.dll!NtWaitForSingleObject()    Unknown
     KERNELBASE.dll!WaitForSingleObjectEx()    Unknown
     amdocl64.dll!00007ff806264e49()    Unknown
     amdocl64.dll!00007ff80625df88()    Unknown
     amdocl64.dll!00007ff80625fd5a()    Unknown
     amdocl64.dll!00007ff80624a51c()    Unknown

 

Unfortunately I do not have direct access to the device to prepare a mimimalistic test case, but can provide an EXE for further investigation on request.

Outcomes