[PATCH] D28626: RFC: Generalize inverted gc dependencies

[Robert Lytton via llvm-commits]

Hi Rafael,
> Robert, is the MC produced objects used or is another assembler used?
We don't currently use MC to produce objects.
(Some work was started and we plan to in the future, but it is unclear when)
Robert

________________________________________
From: Rafael Avila de Espindola [rafael.espindola at gmail.com]
Sent: 23 January 2017 19:38
To: reviews+D28626+public+0a7e58a8570fa72f at reviews.llvm.org; rnk at google.com; eugenis at google.com; peter at pcc.me.uk; ruiu at google.com
Cc: chisophugis at gmail.com; llvm-commits at lists.llvm.org; Robert Lytton
Subject: Re: [PATCH] D28626: RFC: Generalize inverted gc dependencies

Evgeniy Stepanov via Phabricator <reviews at reviews.llvm.org> writes:

> eugenis added inline comments.
>
>
> ================
> Comment at: llvm/include/llvm/Support/ELF.h:752
> +  // FIXME: Tentative value.
> +  SHF_ASSOCIATED = 0x1000U,
> +
> ----------------
> this is conflicting with XCORE_SHF_DP_SECTION a few lines below, which is printed in assembly as "d" flag instead of "o"


And XCORE_SHF_CP_SECTION is conflicting with SHF_COMPRESSED :-(

The story seems to be:
* Ever since xcore was added in r58838 it has sections named like cp.foo
  and dp.foo.
* The cp/dp become flags in r79077.
* When MC came we started writing them to disk.

This is plainly a bug. These flags have to be moved to SHF_MASKPROC and
until they do they cannot be expected to work with standard flags:
Thees sections cannot be compressed and cannot have inverted gc edges.

Robert, is the MC produced objects used or is another assembler used?

Cheers,
Rafael