[cfe-commits] [PATCH] fix failling regression tests due to calling convention markers on recent ARM ubuntu platforms

[David Tweed]

Hi, ping:

is there anyone who thinks this patch is actually the wrong thing to do?

Cheers,
Dave
-----Original Message-----
From: cfe-commits-bounces at cs.uiuc.edu [mailto:cfe-commits-bounces at cs.uiuc.edu] On Behalf Of David Tweed
Sent: 19 October 2012 08:52
To: 'llvm cfe'; '陳韋任 (Wei-Ren Chen)'; 'Galina Kistanova'
Subject: [cfe-commits] [PATCH] fix failling regression tests due to calling convention markers on recent ARM ubuntu platforms

Hi, I think that I've finally got a "correct" solution for the regression
tests failing due to the change to gnueabihf in the latest ubuntu which
implies aapcs-vfp should be the "defaultCC" unless additional command line
switches are set. (The code in target info somewhat ad hoc dealing with an
ad hoc reality so it's... challenging to follow but I believe the patch is
handling things correctly.) It looks like the
default-from-ambient-environment code hasn't gone beyond the aapcs calling
convention. This patch makes clang recognise that the defaultCC is aapcs_vfp
on gnueabihf platforms when no other arguments are specified; I believe it
doesn't alter behaviour if either explicit arguments are given or we're on a
gnueabihf platform. This has been tested on x86_64 and on a gnueabihf ubuntu
Linux pandaboard and it fixes the tests that were failing due to explicit
calling-convention markers being emitted and confusing FileCheck.

(Note that in the patch CodeGenOpts.FloatABI can be, eg, "" so that the test
doesn't contain redundant elements.)

Please review and, equally usefully, test. If there are no problems I'll
then commit it.

Cheers,
Dave
-------------- next part --------------
A non-text attachment was scrubbed...
Name: defaultSetting.diff
Type: application/octet-stream
Size: 1204 bytes
Desc: defaultSetting.diff
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20121022/070f4ad4/attachment.obj>