andrew.w.kaylor added inline comments.
================
Comment at: clang/test/CodeGen/pragma-fenv_access.c:239
+// CHECK-LABEL: @func_20
+// STRICT: call float @llvm.experimental.constrained.fadd.f32(float {{.*}},
float {{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+// DEFAULT: fadd float
----------------
pengfei wrote:
> Should this be `ignore`?
This is a tricky case. By a strict reading of the C standard, this could be
ignore, because the standard says the compiler can assume the default floating
point environment when FENV_ACCESS is OFF and that if code compiled with
FENV_ACCESS OFF is executed with anything other than the default environment
the behavior is undefined. However, in this case strict exception semantics
have been enabled elsewhere in the compilation unit, so floating point
exceptions may be unmasked. The standard allows us to ignore exceptions, but
raising a spurious exception may be bad for users.
I'm unsure about this case. I lean towards leaving it as Zahira has it here
because I don't think the use of strict exception semantics will be common
enough to justify the less conservative behavior.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D147733/new/
https://reviews.llvm.org/D147733
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
