[llvm] [Float2Int] Fix miscompile with floats that can be converted to large values (PR #85996)

[Craig Topper via llvm-commits]

================
@@ -409,7 +409,8 @@ Value *Float2IntPass::convert(Instruction *I, Type *ToTy) {
     } else if (Instruction *VI = dyn_cast<Instruction>(V)) {
       NewOperands.push_back(convert(VI, ToTy));
     } else if (ConstantFP *CF = dyn_cast<ConstantFP>(V)) {
-      APSInt Val(ToTy->getPrimitiveSizeInBits(), /*isUnsigned=*/false);
+      APSInt Val(ToTy->getPrimitiveSizeInBits(),
+                 !CF->getValueAPF().isNegative());
----------------
topperc wrote:

The parameter to the APSInt constructor should tell how the integer is going to be interpreted by the users. I'm not sure it should be based on the data. We had a bug like this on RISC-V recently. https://github.com/llvm/llvm-project/pull/84070

https://github.com/llvm/llvm-project/pull/85996