[llvm] [AMDGPU] prevent shrinking udiv/urem if operands exceed smax_bitwidth (PR #116733)

[via llvm-commits]

================
@@ -1193,6 +1193,17 @@ int AMDGPUCodeGenPrepareImpl::getDivNumBits(BinaryOperator &I, Value *Num,
                                             Value *Den, unsigned AtLeast,
                                             bool IsSigned) const {
   const DataLayout &DL = Mod->getDataLayout();
+  if (!IsSigned) {
+    KnownBits Known = computeKnownBits(Num, DL, 0, AC, &I);
+    // We know all bits are used for division for Operand > smax_bitwidth
----------------
choikwa wrote:

I had meant smax_bitwidth to be signed max of bit width which corresponds to that of Num/Den. Will reword it as such. The operand range that's handled incorrectly is [SignedMax, UnsignedMax].

https://github.com/llvm/llvm-project/pull/116733