[llvm-bugs] [Bug 51836] New: [zorg] stage-1 SVE builds use unsupported flags with system compiler

via llvm-bugs llvm-bugs at lists.llvm.org
Mon Sep 13 08:02:34 PDT 2021


https://bugs.llvm.org/show_bug.cgi?id=51836

            Bug ID: 51836
           Summary: [zorg] stage-1 SVE builds use unsupported flags with
                    system compiler
           Product: clang
           Version: unspecified
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: -New Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: peter.waller at arm.com
                CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org,
                    neeilans at live.com, richard-llvm at metafoo.co.uk

The builders clang-aarch64-sve-vla-2stage and clang-aarch64-sve-vls-2stage both
use flags not understood by the stage1 compiler provided by the host which at
the moment is clang-12.

https://github.com/llvm/llvm-zorg/commit/6fc25a74c84e064b375e320b918b62fed3a2bde0

At stage 1:

* The VLA bot silently ignores the -mllvm -scalable-vectorization=preferred
flag passed to it.

* The VLS bot compiles with -msve-vector-bits=...

It seems to me it would be preferable to have a three-stage build, where the
first stage uses relatively boring flags. If that's not possible, at least
there needs to be a way to specify different flags between stage1 and stage2
which it seems isn't supported by zorg.

If at some point these builders are using clang-13, and it turns out there are
latent SVE codegen bugs, then this builder could fail in surprising ways when
stage1 executes and it won't be possible to fix it by submitting a patch to the
main branch. We're effectively in that situation in clang-12, although happily
I think the flags are being mostly ignored. The linked patch did change the
behaviour of the stage-1 build however, where previously it was testing the vls
codegen (because it understood -mllvm -aarch64-sve-vector-bits-min) but now it
does not.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210913/8a8dfd66/attachment.html>


More information about the llvm-bugs mailing list