[llvm] [SCCP] Handle llvm.experimental.get.vector.length calls (PR #169527)

[Luke Lau via llvm-commits]

================
@@ -2098,6 +2098,38 @@ void SCCPInstVisitor::handleCallResult(CallBase &CB) {
       return (void)mergeInValue(ValueState[II], II,
                                 ValueLatticeElement::getRange(Result));
     }
+    if (II->getIntrinsicID() == Intrinsic::experimental_get_vector_length) {
+      Value *CountArg = II->getArgOperand(0);
+      Value *VF = II->getArgOperand(1);
+      bool Scalable = cast<ConstantInt>(II->getArgOperand(2))->isOne();
+
+      // Computation happens in the larger type.
+      unsigned BitWidth = std::max(CountArg->getType()->getScalarSizeInBits(),
+                                   VF->getType()->getScalarSizeInBits());
+
+      ConstantRange Count = getValueState(CountArg)
+                                .asConstantRange(CountArg->getType(), false)
+                                .zextOrTrunc(BitWidth);
----------------
lukel97 wrote:

ConstantRange::zeroExtend unfortunately asserts that the new BitWidth needs to be larger, not just equal to in size. So I used zextOrTrunc which returns the input when OldBitWidth == NewBitWidth. 

Should we relax the assertion? It seems to be from the early days in 2004. Allowing equal bitwidths would make it more consistent with IRBuilder::CreateZExt

https://github.com/llvm/llvm-project/pull/169527