[llvm] [InstCombine] Handle scalable geps in EmitGEPOffset (PR #71565)

[Philip Reames via llvm-commits]

================
@@ -45,7 +45,8 @@ Value *llvm::emitGEPOffset(IRBuilderBase *Builder, const DataLayout &DL,
   for (User::op_iterator i = GEP->op_begin() + 1, e = GEP->op_end(); i != e;
        ++i, ++GTI) {
     Value *Op = *i;
-    uint64_t Size = DL.getTypeAllocSize(GTI.getIndexedType()) & PtrSizeMask;
+    TypeSize TSize = DL.getTypeAllocSize(GTI.getIndexedType());
+    uint64_t Size = TSize.getKnownMinValue() & PtrSizeMask;
----------------
preames wrote:

I'm not entirely clear on the role of PtrSizeMask, but it seems odd to be applying this to the known min value for a scalable vector and not the full dynamic type size.  Doesn't seem to match the fixed vector semantics.  

https://github.com/llvm/llvm-project/pull/71565