================
@@ -743,6 +800,37 @@ class MapInfoFinalizationPass
return mlir::WalkResult::advance();
});
+ func->walk([&](mlir::omp::MapInfoOp op) {
+ if (!op.getMembers().empty())
+ return;
+
+ if (!mlir::isa<fir::BoxCharType>(fir::unwrapRefType(op.getVarType())))
+ return;
+
+ // POSSIBLE_HACK_ALERT: If the boxchar has been implicitly mapped then
+ // it is likely that the underlying pointer to the data
----------------
agozillon wrote:
I think as long as it's the same address, the runtime wouldn't necessarily
invoke another data transferal, just bump the ref counter, so it wouldn't be
too costly in these cases I think! But it is nice to avoid generating extra
maps where feasible :-)
https://github.com/llvm/llvm-project/pull/141715
_______________________________________________
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
