[PATCH] D84879: [X86] Separate CPU Feature lists in X86.td between architecture features and tuning features

[Craig Topper via Phabricator via llvm-commits]

craig.topper added inline comments.


================
Comment at: llvm/lib/Target/X86/X86.td:562
+  list<SubtargetFeature> NHMTuning = [FeatureMacroFusion,
+                                      FeatureInsertVZEROUPPER];
 
----------------
RKSimon wrote:
> I keep forgetting to ask - why do targets without AVX implement FeatureInsertVZEROUPPER?
It starts with an assumption that "-mavx" without a -march should insert vzeroupper. Since in that case the CPU will be pentium4 or x86-64 so those CPUs need the feature. Maybe also i486 or i686 if you're on the BSD that uses those as the default 32-bit CPU. After that it was just to apply consistent behavior if you wrote something like "-march=nehalem -mavx" for some reason.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D84879/new/

https://reviews.llvm.org/D84879