[llvm] [GlobalISel] Catching inconsistencies in load memory, result, and range metadata type (PR #121247)

[Matt Arsenault via llvm-commits]

================
@@ -1274,6 +1274,24 @@ void MachineVerifier::verifyPreISelGenericInstruction(const MachineInstr *MI) {
         if (TypeSize::isKnownGT(MMO.getSize().getValue(),
                                 ValTy.getSizeInBytes()))
           report("load memory size cannot exceed result size", MI);
+
+        if (MMO.getRanges()) {
+          auto operandBits = [](const MDOperand &o) -> unsigned {
+            ConstantInt *i = mdconst::dyn_extract<ConstantInt>(o);
+            if (!i->isNegative())
+              return i->getValue().getActiveBits();
+            APInt reversed(i->getValue());
----------------
arsenm wrote:

>  there is a match between 2*32 == 64

This is the problem, it's not a match. The total bitwidth is not what matters, the scalar type does. It would be OK if this was a `<2 x s64>` result and the range was using i64. tttt

i.e. the vectorness doesn't matter, nor the total bit width. The scalar type interpretation on both halves is what matters 

https://github.com/llvm/llvm-project/pull/121247