[llvm] [LVI] Support using block values when handling conditions (PR #75311)

[Nikita Popov via llvm-commits]

================
@@ -1068,15 +1094,26 @@ static bool matchICmpOperand(APInt &Offset, Value *LHS, Value *Val,
 }
 
 /// Get value range for a "(Val + Offset) Pred RHS" condition.
-static ValueLatticeElement getValueFromSimpleICmpCondition(
-    CmpInst::Predicate Pred, Value *RHS, const APInt &Offset) {
+std::optional<ValueLatticeElement>
+LazyValueInfoImpl::getValueFromSimpleICmpCondition(CmpInst::Predicate Pred,
+                                                   Value *RHS,
+                                                   const APInt &Offset,
+                                                   Instruction *CxtI,
+                                                   bool UseBlockValue) {
   ConstantRange RHSRange(RHS->getType()->getIntegerBitWidth(),
                          /*isFullSet=*/true);
-  if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS))
+  if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) {
     RHSRange = ConstantRange(CI->getValue());
-  else if (Instruction *I = dyn_cast<Instruction>(RHS))
+  } else if (UseBlockValue) {
----------------
nikic wrote:

This is the core change, the rest is just threading the new nullopt return value through everything.

https://github.com/llvm/llvm-project/pull/75311