[DAG] Apply Disjoint flag.

[fengfeng09]

or disjoint (or disjoint (x, c0), c1)
-->
or disjont x, or (c0, c1)
Alive2: https://alive2.llvm.org/ce/z/3wPth5

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-llvm-selectiondag

Author: fengfeng (fengfeng09)

Changes

or disjoint (or disjoint (x, c0), c1)
-->
or disjont x, or (c0, c1)
Alive2: https://alive2.llvm.org/ce/z/3wPth5

---

Full diff: https://github.com/llvm/llvm-project/pull/118045.diff

1 Files Affected:

• (modified) llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (+4-1)

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 521829675ae7c3..f56930f08f68ef 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -1220,8 +1220,11 @@ SDValue DAGCombiner::reassociateOpsCommutative(unsigned Opc, const SDLoc &DL,
 
     if (DAG.isConstantIntBuildVectorOrConstantInt(N1)) {
       // Reassociate: (op (op x, c1), c2) -> (op x, (op c1, c2))
-      if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, DL, VT, {N01, N1}))
+      if (SDValue OpNode = DAG.FoldConstantArithmetic(Opc, DL, VT, {N01, N1})) {
+        NewFlags.setDisjoint(Flags.hasDisjoint() &&
+                             N0->getFlags().hasDisjoint());
         return DAG.getNode(Opc, DL, VT, N00, OpNode, NewFlags);
+      }
       return SDValue();
     }
     if (TLI.isReassocProfitable(DAG, N0, N1)) {

[dtcxzyw]

Can you add some tests to demonstrate that this patch improves the codegen?

[fengfeng09]

Can you add some tests to demonstrate that this patch improves the codegen?

Ok.

[fengfeng09]

Can you add some tests to demonstrate that this patch improves the codegen?

Updated. A test case is appended.

[dtcxzyw]

LGTM.

[fengfeng09]

@dtcxzyw Thanks, could you help me merge this PR? I have no write access to this repo.

[nikic]

Why does this need -start-before=aarch64-isel? (Or -verify-machineinstrs for that matter)

[fengfeng09]

Cleared. That is not really needed.