[PATCH] D127440: llvm-reduce: Handle reducing FP values to nan

[Arthur Eubanks via Phabricator via llvm-commits]

aeubanks added a comment.

currently we prefer 0 over 1 over undef because having undef in reproducers makes the IR makes it more likely that the code has undefined behavior, causing future changes to potentially perturb optimized IR
I think we should prefer 0 over 1 over nan over undef


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D127440/new/

https://reviews.llvm.org/D127440