[Mlir-commits] [mlir] [mlir][VectorToGPU] Only convert an scf.yield that terminates an scf.for (PR #218226)

[Federico Bruzzone]

https://github.com/FedericoBruzzone commented:

Given that `getOpToConvert` (the only user of `supportsMMaMatrixType`) is also used in `convertVectorToNVVMCompatibleMMASync`:

https://github.com/llvm/llvm-project/blob/90a6975dc8384f894668049477698c2fb32e80dc/mlir/lib/Conversion/VectorToGPU/VectorToGPU.cpp#L1301-L1303

I was wondering if this check remains valid even when `useNvGpu` is `true`. Just curious :'D

https://github.com/llvm/llvm-project/pull/218226