[BasicAA] Fix aliasGEP/DecomposeGEPExpression for scalable type.

[huihuiz]

Summary

Don't attempt to analyze the decomposed GEP for scalable type.
GEP index scale is not compile-time constant for scalable type.
Be conservative, return MayAlias.

Explicitly call TypeSize::getFixedSize() to assert on places where
scalable type doesn't make sense.

Add unit tests to check functionality of -basicaa for scalable type.

This patch is needed for D76944.

[huihuiz]

Current upstream is producing wrong AliasResult for scalable type.

take test.ll , run: opt -basicaa -aa-eval -print-all-alias-modref-info -disable-output test.ll.

Then you will see " MustAlias: i32* %gep1, i32* %gep2 "
gep1,gep2 should be MayAlias, not MustAlias.

define void @gep_bitcast_1(<vscale x 4 x i32>* %p) {
  %gep1 = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* %p, i64 1, i64 0
  %p2 = bitcast <vscale x 4 x i32>* %p to i32*
  %gep2 = getelementptr i32, i32* %p2, i64 4
  ret void
}

[efriedma]

llvm/lib/Analysis/BasicAliasAnalysis.cpp

1326

I think this still doesn't completely block off all the relevant cases.

DecomposeGEPExpression is recursive, so there are can be multiple levels of GEPs. The GEP over the scalable vector may not be GEP1; it could be GEP1->getPointerOperand(), or something like that.

[huihuiz]

Address review comments.

llvm/lib/Analysis/BasicAliasAnalysis.cpp

1326

Good catch, Thanks Eli!

Moving the checks to DecomposeGEPExpression, so we have knowledge of scalable type given multiple levels of GEPs.

Patch updated accordingly, and with more tests in recursive geps.

[efriedma]

LGTM