[Mlir-commits] [mlir] [mlir][spirv] Fix SSA Handling in SPIRV -> MLIR Translation (PR #123371)

[Jakub Kuderski]

kuhar wrote:

> I have been wondering what is the proper way to write tests for these changes though -- it appears to me that an MLIR based test can never actually represent the valid SPVASM -> invalid MLIR scenario that our changes seek to resolve.
> @kuhar perhaps you have an opinion on that subject or know how to point us in the right direction?

I think that the llvm spirv backend uses spirv tools conditionally -- we could do something similar in a very limited set of tests. See https://github.com/llvm/llvm-project/blob/main/llvm/test/Other/spirv-sim/simple.spv .
In general, we want to avoid external dependencies unless strictly necessary.

https://github.com/llvm/llvm-project/pull/123371