[llvm] [DAGCombine] Remove OneUse restriction when folding (shl (add x, c1), c2) and (shl (sext (add x, c1)), c2) (PR #101294)

[via llvm-commits]

================
@@ -59541,3 +59541,18 @@ Align X86TargetLowering::getPrefLoopAlignment(MachineLoop *ML) const {
     return Align(1ULL << ExperimentalPrefInnermostLoopAlignment);
   return TargetLowering::getPrefLoopAlignment();
 }
+
+bool X86TargetLowering::isDesirableToCommuteWithShift(
+    const SDNode *N, CombineLevel Level) const {
+  assert((N->getOpcode() == ISD::SHL || N->getOpcode() == ISD::SRA ||
+          N->getOpcode() == ISD::SRL) &&
+         "Expected shift op");
+
+  SDValue ShiftLHS = N->getOperand(0);
+  if ((ShiftLHS.getOpcode() == ISD::SIGN_EXTEND &&
+       !(ShiftLHS->hasOneUse() && ShiftLHS.getOperand(0)->hasOneUse())) ||
+      !ShiftLHS->hasOneUse())
+    return false;
----------------
goldsteinn wrote:

Here (and everywhere else) think it would be much cleaner to just do:

```
match ShiftLHS->hasOneUse() || match(ShiftLHS, m_SExtLike(m_OneUse(m_Value(X)))
```

NB: The `m_SExtLike` needs to be added to `SDPatternMatch.h`, its just `ZEXT` + `NNEG`.

https://github.com/llvm/llvm-project/pull/101294