================
@@ -630,14 +630,24 @@ std::string Attribute::getAsString(bool InAttrGrp) const {
// Print access kind for "other" as the default access kind. This way it
// will apply to any new location kinds that get split out of "other".
ModRefInfo OtherMR = ME.getModRef(IRMemLocation::Other);
+ ModRefInfo InaccessibleMemMR =
ME.getModRef(IRMemLocation::InaccessibleMem);
if (OtherMR != ModRefInfo::NoModRef || ME.getModRef() == OtherMR) {
First = false;
OS << getModRefStr(OtherMR);
}
for (auto Loc : MemoryEffects::locations()) {
ModRefInfo MR = ME.getModRef(Loc);
- if (MR == OtherMR)
+ // Skip target memory location
+ if (MR == OtherMR && !ME.isTargetMemLoc(Loc))
+ continue;
+
+ // Prints Target Memory Location if ModRefInfo from InaccessibleMem
+ // differs from Target_MemX
+ // or ModRefInfo from Target Memory Location is NoModRef as OtherMR
+ // Otherwise skip Target_Mem print
+ if ((ME.isTargetMemLoc(Loc) && MR == InaccessibleMemMR) ||
+ (MR == OtherMR && OtherMR == ModRefInfo::NoModRef))
----------------
paulwalker-arm wrote:
I've been trying to figure out why this is more complicated than I expected,
especially because I suspect it doesn't quite matches the new rule used for
parsing.
I think it comes from fixing output difference caused by tests that use the old
style memory attributes prior to `memory()`. The real cause for those
differences is their parsing not implementing the new rule relating to the
relationship between InaccessibleMem and TargetMem. However, I've done a
deeper investigation and I now think we're just looking for trouble[1] when
trying to minimise the test changes. It seems I also misunderstood the extent
of those changes, which looks to be only ~10 files?
Assuming I've not misrepresented something, let's keep things simple and just
remove the special print/parse code and update the tests accordingly.
[1] Part of the problem here is bitcode. For those cases target memory
effectively defaults to `none`, so when tying to be clever we've actually just
caused a discrepancy between how ll and bc files interpret the same data. We
will need to change bitcode but that cannot happen until we start actively
using the new target memory locations.'
https://github.com/llvm/llvm-project/pull/148650
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
