[llvm-branch-commits] [llvm] AMDGPU: Replace unused permlane inputs with poison instead of undef (PR #131288)

[Jay Foad via llvm-branch-commits]

jayfoad wrote:

> > Same kind of objection as #131287: as a general strategy, "replace unused inputs with poison"
> 
> Repeating from the other review, but this is not the case. Poison does not unconditionally fold through intrinsics. This is specific to an operand for an intrinsic. It just happens in the common case, normal arithmetic-like intrinsics propagate poison in any inputs

Then I think the poison-propagating rules for this intrinsic should be documented. They're not obvious and "it just does what the underlying instruction does" is no longer sufficient.

https://github.com/llvm/llvm-project/pull/131288