[llvm] [SDAG] Fix type checks in `ShrinkDemandedOp` to avoid creating invalid truncates (PR #92730)

[Yingwei Zheng via llvm-commits]

================
@@ -599,7 +599,11 @@ bool TargetLowering::ShrinkDemandedOp(SDValue Op, unsigned BitWidth,
   for (unsigned SmallVTBits = llvm::bit_ceil(DemandedSize);
        SmallVTBits < BitWidth; SmallVTBits = NextPowerOf2(SmallVTBits)) {
     EVT SmallVT = EVT::getIntegerVT(*DAG.getContext(), SmallVTBits);
-    if (TLI.isTruncateFree(VT, SmallVT) && TLI.isZExtFree(SmallVT, VT)) {
+    // Types of LHS and RHS may differ before legalization (e.g., shl), so we
+    // need to check both.
+    if (TLI.isTruncateFree(Op.getOperand(0).getValueType(), SmallVT) &&
+        TLI.isTruncateFree(Op.getOperand(1).getValueType(), SmallVT) &&
----------------
dtcxzyw wrote:

> Don't understand the diff from the description. isTruncateFree can't guard against invalid transforms

Oops, it just so happens that all implementations of `XXXTargetLowering::isTruncateFree` guarantee this.
I will fix this.


https://github.com/llvm/llvm-project/pull/92730