[clang] [llvm] [BPF] Add load-acquire and store-release instructions under -mcpu=v4 (PR #108636)

Peilin Ye via llvm-commits llvm-commits at lists.llvm.org
Fri Oct 25 18:00:11 PDT 2024


================
@@ -703,6 +715,39 @@ SDValue BPFTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
   return DAG.getNode(BPFISD::SELECT_CC, DL, VTs, Ops);
 }
 
+SDValue BPFTargetLowering::LowerATOMIC_LOAD(SDValue Op,
+                                            SelectionDAG &DAG) const {
+  const char *Msg =
+      "sequentially consistent (seq_cst) atomic load is not supported";
+  SDNode *N = Op.getNode();
+  SDLoc DL(N);
+
+  if (cast<AtomicSDNode>(N)->getMergedOrdering() ==
+      AtomicOrdering::SequentiallyConsistent)
+    fail(DL, DAG, Msg);
+
+  return Op;
+}
+
+SDValue BPFTargetLowering::LowerATOMIC_STORE(SDValue Op,
+                                             SelectionDAG &DAG) const {
+  const char *Msg =
+      "sequentially consistent (seq_cst) atomic store is not supported";
+  EVT VT = Op.getOperand(1).getValueType();
+  SDNode *N = Op.getNode();
+  SDLoc DL(N);
+
+  // Promote operand #1 (value to store) if necessary.
+  if (!isTypeLegal(VT))
+    return SDValue();
----------------
peilin-ye wrote:

```c
// clang --target=bpf -mcpu=v4 -O2 -g -c -o demo.bpf.o demo.bpf.c
void bar(char *ptr, char val) {
    __atomic_store_n(ptr, val, __ATOMIC_RELEASE);
}
```
Compiling the above program would trigger that if clause:
```
...
llvm::SelectionDAG::LegalizeTypes()
  llvm::DAGTypeLegalizer::run()
    llvm::DAGTypeLegalizer::PromoteIntegerOperand()
      llvm::DAGTypeLegalizer::CustomLowerNode()
        llvm::TargetLowering::LowerOperationWrapper()
          llvm::BPFTargetLowering::LowerATOMIC_STORE()
```
BPF doesn't have native support for `i8` (see `BPFISelLowering.cpp`) :
```cpp
  // Set up the register classes.
  addRegisterClass(MVT::i64, &BPF::GPRRegClass);
  if (STI.getHasAlu32())
    addRegisterClass(MVT::i32, &BPF::GPR32RegClass);
```

https://github.com/llvm/llvm-project/pull/108636


More information about the llvm-commits mailing list