[Mlir-commits] [mlir] [mlir][gpu][NVPTX] Enable NVIDIA GPU JIT compilation path (PR #66220)

[Fabian Mora]

fabianmcg wrote:

> Would it be okay if we didn't do this for default behaviour?

The default behavior remains `fatbin`. However, I made `isa` (JIT) the default behavior for the sparse compiler.

Another option, is setting `MLIR_GPU_COMPILATION_TEST_FORMAT=isa` so tests run with JIT, but everywhere else keep `fatbin` as the default behavior.

I'm inclined to keep `fatbin` as default everywhere and make downstream users set `MLIR_GPU_COMPILATION_TEST_FORMAT=isa` in their builds.

https://github.com/llvm/llvm-project/pull/66220