[PATCH] New SystemZ backend: LLVM code changes

[Rafael Espíndola]

> Hmm, I was just trying to mimic the AArch64 tests, like Eric said. :-)
> Do you mean that there's something wrong with the way those work,
> or with the way I adapted them?

In general, we like to test systems in isolation. I am sorry I never
got time to look at AArch64, but any architecture that has at least a
basic asm parser should test codegen and MC individually.

Tests using -flietype=obj should only exist to test problems specific
to direct code emission, which, given the MCStreamer design, shouldn't
be very many.

Historically we had many of them because some targets got the ability
to output objects directly before they got the ability to parse ASM,
but that is an historical issue, not a design goal.

> In case it helps, the CodeGen/SystemZ/frame-*.ll tests that I mentioned
> are attached (all of them, including some that aren't CFI-related).
> I think it makes sense to test the .cfi_* .s output alongside the
> prologue and epilogue tests.
>
> Richard
>

Cheers,
Rafael