[llvm] [MachinePipeliner] Add an abstract layer to manipulate Data Dependenc… (PR #109918)

Ryotaro Kasuga via llvm-commits llvm-commits at lists.llvm.org
Wed Feb 12 22:14:42 PST 2025


================
@@ -114,10 +115,123 @@ class MachinePipeliner : public MachineFunctionPass {
   bool useWindowScheduler(bool Changed);
 };
 
+/// Represents a dependence between two instruction.
+class SwingSchedulerDDGEdge {
+  SUnit *Dst = nullptr;
+  SDep Pred;
+  unsigned Distance = 0;
+
+public:
+  /// Creates an edge corresponding to an edge represented by \p PredOrSucc and
+  /// \p Dep in the original DAG. This pair has no information about the
+  /// direction of the edge, so we need to pass an additional argument \p
+  /// IsSucc.
+  SwingSchedulerDDGEdge(SUnit *PredOrSucc, const SDep &Dep, bool IsSucc)
+      : Dst(PredOrSucc), Pred(Dep), Distance(0u) {
+    SUnit *Src = Dep.getSUnit();
+
+    if (IsSucc) {
+      std::swap(Src, Dst);
+      Pred.setSUnit(Src);
+    }
+
+    // An anti-dependence to PHI means loop-carried dependence.
+    if (Pred.getKind() == SDep::Anti && Src->getInstr()->isPHI()) {
+      Distance = 1;
+      std::swap(Src, Dst);
+      auto Reg = Pred.getReg();
+      Pred = SDep(Src, SDep::Kind::Data, Reg);
----------------
kasuga-fj wrote:

> "The anti dependence from some_instr to phi becomes the data dependence from phi to some_instr, as I intended."
> 
> By looking at this code:
> Src will be some_instr
> Dst will be the phi.
>
> So I believe you meant to tell me that data dep is from some_instr to phi?

Yes, I meant that, but now I think something is incorrect. In your previous comment, you said:

> Input:
> some_instruction -> phi; anti dep

Also

> PredOrSucc: phi
> IsSucc = true,
> Dep: some_instruction; anti dep

This seems to be wrong. IIUIC, `isSucc` should be `false` in this case. This argument doesn't mean the "actual semantics" of the dependency. It just indicates whether the `Dep` is contained in `PredOrSucc->Preds` or `PredOrSucc->Succs`.  Inserting some validation code like the following at the top of the ctor might help (I don't check if this code works fine).

```c++
const auto &Edges = (IsSucc ? PredOrSucc->Succs : PredOrSucc->Preds);
bool Valid = false;
for (const SDep &D : Edges)
  if (D == Dep)
    Valid = true;
assert(Valid && "isSucc may be incorrect");
```

https://github.com/llvm/llvm-project/pull/109918


More information about the llvm-commits mailing list