[clang] [llvm] [NVPTX] Auto-Upgrade llvm.nvvm.atomic.load.{inc,dec}.32 (PR #134111)

[via llvm-commits]

================
@@ -2314,6 +2317,12 @@ static Value *upgradeNVVMIntrinsicCall(StringRef Name, CallBase *CI,
     Value *Val = CI->getArgOperand(1);
     Rep = Builder.CreateAtomicRMW(AtomicRMWInst::FAdd, Ptr, Val, MaybeAlign(),
                                   AtomicOrdering::SequentiallyConsistent);
+  } else if (Name.consume_front("atomic.load.") && Name.consume_back(".32")) {
+    Value *Ptr = CI->getArgOperand(0);
+    Value *Val = CI->getArgOperand(1);
+    auto Op = Name == "inc" ? AtomicRMWInst::UIncWrap : AtomicRMWInst::UDecWrap;
+    Rep = Builder.CreateAtomicRMW(Op, Ptr, Val, MaybeAlign(),
+                                  AtomicOrdering::SequentiallyConsistent);
----------------
gonzalobg wrote:

For historical reasons, frontends generate `seq_cst` and "system scope", and we (LLVM and NVVM) lower that to `relaxed+gpu scope`. We should definetly prioritize fixing these bugs by lowering these properly (there are also a few other related bugs that we need to fix).

However, these fixes will break all the frontends, so we need to give some thought to how to approach this to minimize churn.

Therefore, i don't know if this PR is the right place to make this fix (this has been broken for a long time).

https://github.com/llvm/llvm-project/pull/134111