bogner wrote:
> The backend part looks good, but I am confused about the tests. I get that a
> lot of the CBufferAccess tests have been removed because big part of the the
> cbuffer access logic is now in DXILResourceAccesPass, but the Clang codegen
> tests in this change all generate typed GEPs such as:
>
> `getelementptr (<{ float, target("dx.Padding", 12) }>, ptr addrspace(2) @c1,
> i32 7, i32 0)`,
>
> And the few remaining tests in the backend are only testing untyped GEPs:
>
> `getelementptr inbounds nuw (i8, ptr addrspace(2) @a1, i32 16)`.
>
> Shouldn't we be testing the typed variants too? Or is something always
> translating typed GEPs into untyped before CBufferAccess runs?
The typed GEPs are all translated to untyped in InstCombine (See #68882), so we
never see them in the backend. I've added back the array-typedgep test that
shows that they're handled correctly regardless, as I do think it makes sense
to have that test.
https://github.com/llvm/llvm-project/pull/167404
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
