================
Comment at: lib/Basic/Targets.cpp:5679
@@ +5678,3 @@
+ bool isNan2008() const override {
+ return IsNan2008 || isNaN2008Default();
+ }
----------------
petarj wrote:
> vradosavljevic wrote:
> > dsanders wrote:
> > > This isn't quite right. It should just be IsNan2008 since IsNan2008 is
> > > initialized with the result of isNaN2008Default().
> > When you generating code for mips32r6/mips64r6 with option -mnan=legacy,
> > IsNan2008 will be set to false.
> > In function handleTargetFeatures there are lines that change that.
> > else if (*it == "-nan2008")
> > IsNan2008 = false;
> > Because of that, I use the result of isNaN2008Default().
> We should not allow that case. Clang should (at least) issue a warning if
> "-mnan=legacy" is passed for the revisions that do not support the NaN legacy
> encoding.
Agreed. I'd prefer an error* since the user is requesting an impossible
combination of options but it should at least warn instead of quietly ignoring
the user.
*Ideally with a suggestion for fixing it, e.g. "error: IEEE754(2008) compliant
NaN's are mandatory on Mips32r6. Did you mean -mnan=2008?"
http://reviews.llvm.org/D7882
EMAIL PREFERENCES
http://reviews.llvm.org/settings/panel/emailpreferences/
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
