[clang] [NFC] Add assertion to ensure FiniteMathOnly is in sync with HonorINFs and HonorNANs. (PR #97342)

Aaron Ballman via cfe-commits cfe-commits at lists.llvm.org
Tue Jul 2 11:40:56 PDT 2024


================
@@ -816,6 +816,11 @@ class FPOptions {
     setAllowFPReassociate(LO.AllowFPReassoc);
     setNoHonorNaNs(LO.NoHonorNaNs);
     setNoHonorInfs(LO.NoHonorInfs);
+    // Ensure that if FiniteMathOnly is enabled, NoHonorNaNs and NoHonorInfs are
+    // also enabled. This is because FiniteMathOnly mode assumes no NaNs or Infs
+    // are present in computations.
+    if (!LO.NoHonorInfs || !LO.NoHonorInfs)
+      assert(!LO.FiniteMathOnly && "FiniteMathOnly implies NoHonorInfs");
----------------
AaronBallman wrote:

That sounds like a logic bug to me -- if infs and nans are disabled, then it's definitely finite math only mode. If infs *OR* nans are disabled... I think it's also finite math only mode.

Ideally, I think we should get rid of the internal notion of `FiniteMathOnly` and replace it with the honor infs/nans flags explicitly so we don't get into these sort of situations to begin with. We'd still have to answer how to define `__FINITE_MATH_ONLY__`, but that seems doable.

CC @andykaylor @jcranmer-intel 

https://github.com/llvm/llvm-project/pull/97342


More information about the cfe-commits mailing list