Charlie Lin
Charlie Lin
Failing to compile, needs fix. Check CI runs for the errors.
I'm going to help this PR along and make some changes to pass the rest of the CI.
Right, I see the limit for the direct algorithm for convolution; but this is for pooling. Technically pooling can be implemented as a convolution, but the pooling page doesn't mention...
> Now my question is do I have to test each configuration for pooling, and check when it first fails, and put on those condition to fallback on cpu? I...
> Is this required if we were to pull in changes from MLIR ? We won't hit the verify "accuracy" bug for the quantized resnet50 model with the change from...
> > We won't hit the verify "accuracy" bug for the quantized resnet50 model with the change from MLIR. > > If we dont hit an accuracy error then we...
Closing: Should be able to pass verify accuracy with this compiler pass enable with the MLIR fix https://github.com/ROCm/rocMLIR/pull/1567
Am I missing something for this PR or is it supposed to be this short?
editted your comment @aarushijai to be readable on Github
Can we move the `find_splits{}` matcher in `simplify_algebra.cpp` up or have a copy of it run before `find_mul_add{}`? That should find the fusion opportunity over the `slice` instructions.