Jun Liu
@Slimakanzer Can we verify if the following W/A works? > The simplest proposed fix is limiting applicability of the solvers that use Im2col/col2im machinery.
@DrizztDoUrden and @atamazov : we are having some issues with `Fp32 OpenCL Debug gfx1030` stage ``` [2022-07-25T20:02:37.066Z] MIOpen(OpenCL): Error [BuildProgram] Build log: /tmp/comgr-45f8cd/input/CompileSource:106:5: error: 'USE_LARGE_BUFFER_INDEX' is not defined, evaluates to...
@illsilin could you take a look at this issue?
Ping @DrizztDoUrden to also take a look from the host/API side.
@jerryyin ping for the above analysis from MIOpen team.
Let's maybe disable this perf stage by default for now? I see several pipelines stuck in this stage: http://micimaster.amd.com/blue/organizations/jenkins/MLLibs%2FMIOpen/detail/PR-2163/3/pipeline
@jerryyin any follow-ups on this issue?
@atamazov 4.5.2 in dockerfiles seem to have no problems on the current CI pipelines (while base OS ROCm are not updated): https://github.com/ROCmSoftwarePlatform/MIOpen/commits/jenkins-ci-rocm-4.5 http://micimaster.amd.com/blue/organizations/jenkins/MLLibs%2FMIOpen/detail/jenkins-ci-rocm-4.5/9/pipeline
@jbakhrai could you align with @okakarpa for the best time window for this task? Thanks!
Adding @atamazov , this looks like *adding* a functionality which will eventually replace/convert an existing one, I would suggest moving forward if no objections.