================ @@ -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"); ---------------- jcranmer-intel wrote:
The way I see it, nnan and ninf collectively define four modes for math: full types, no-nan, finite-math (nnan + ninf), and the extremely questionable (to me at least) ninf-without-nnan. I don't think `FiniteMathOnly` as a concept independent of the nnan/ninf flags makes much sense. There is perhaps room for disagreement as to whether `__FINITE_MATH_ONLY__` should be `nnan || ninf` or `nnan && ninf`; I lean towards && myself. Looking at sourcegraph, there seems to be essentially three uses of these macros: two to indicate that they don't care about min(NaN) results, and one to avoid using fpclassify. https://github.com/llvm/llvm-project/pull/97342 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits