[llvm] [InstCombine][NFC] Replace buildAssumeFromKnowledge with CreateAlignmentAssumption (PR #196254)

[Nikolas Klauser via llvm-commits]

================
@@ -3740,15 +3728,8 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
           /// offset and alignment.
           /// TODO: we can generate a GEP instead of merging the alignment with
           /// the offset.
-          RetainedKnowledge RK{Attribute::Alignment,
-                               MinAlign(Offset, AlignMask + 1), A};
-          if (auto *Replacement =
-                  buildAssumeFromKnowledge(RK, Next, &AC, &DT)) {
-
-            Replacement->insertAfter(II->getIterator());
-            AC.registerAssumption(Replacement);
-          }
-          return RemoveConditionFromAssume(II);
+          Builder.CreateAlignmentAssumption(getDataLayout(), A, AlignMask + 1);
----------------
philnik777 wrote:

I'm not sure how to test this. It looks like the `add` and `icmp` get folded first, which means that we never try to compare to zero with an offset smaller than the alignment compare. Any thoughts on what to do?

This is what I tried:
```llvm
define void @align_with_offset(ptr %ptr) {
; CHECK-LABEL: @align_with_offset(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[INT:%.*]] = ptrtoint ptr [[PTR:%.*]] to i64
; CHECK-NEXT:    [[TMP0:%.*]] = and i64 [[INT]], 7
; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[TMP0]], 5
; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP]])
; CHECK-NEXT:    ret void
;
entry:
  %int = ptrtoint ptr %ptr to i64
  %add = add i64 %int, 3
  %and = and i64 %add, 7
  %cmp = icmp eq i64 0, %and
  call void @llvm.assume(i1 %cmp)
  ret void
}
```

https://github.com/llvm/llvm-project/pull/196254