================
@@ -502,3 +497,27 @@ void NVPTXPassConfig::addMachineSSAOptimization() {
addPass(&PeepholeOptimizerLegacyID);
printAndVerify("After codegen peephole optimization pass");
}
+
+bool NVPTXTargetMachine::isCompatibleDataLayout(
+ const DataLayout &Candidate) const {
+ // XXX: Should we enforce that the Candidate DataLayout has the same address
+ // space for allocas?
+ if (DL == Candidate)
+ return true;
+
+ auto DLStr = DL.getStringRepresentation();
+ if (!StringRef(DLStr).contains("A"))
+ DLStr = DLStr.empty() ? "A" + std::to_string(ADDRESS_SPACE_LOCAL)
+ : DLStr + "-A" + std::to_string(ADDRESS_SPACE_LOCAL);
+ auto NewDL = DataLayout(DLStr);
+
+ return NewDL == Candidate;
----------------
Artem-B wrote:
So, the idea here is to effectively assume that if `Ax` is not explicitly
specified in the data layout, we assume that the layout is compatible with
`A5`. This contradicts LLVM spec which states that `A` `Defaults to the default
address space of 0.`
I think we may be working around the problem here, instead of addressing the
root cause.
https://github.com/llvm/llvm-project/pull/154814
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
