[FR] Expect to support PoolingBWD with zero workspace

[ghostplant]

MIOpenPoolingBackwards requires additional workspace while CUDNN doesn't, and this complicates the porting work from CUDNN to MIOpen since it is not easy sometimes to pair the Forwards and Backwards invocations among quantities of pooling operations.

[lsder]

Is this FR solved?

[atamazov]

@lsder I do not think so. IIRC existing pooling implementations in MIOpen require workspace.

/cc @JehandadKhan @dmikushin

[atamazov]

/cc @junliume

[ppanchad-amd]

@ghostplant Apologies for the lack of response. Please try latest ROCm 6.1.2 to see if your issue has been resolved? If resolved, please close the ticket. Thanks.