[OpenWrt-Devel] Broken GCC 4.9 (and 5.2) on ARM - asking for review/test
john at szakmeister.net
Fri Aug 21 07:13:03 EDT 2015
On Thu, Aug 20, 2015 at 6:57 AM, Jonas Gorski <jogo at openwrt.org> wrote:
> On Thu, Aug 20, 2015 at 11:40 AM, Dirk Neukirchen <dirkneukirchen at web.de> wrote:
>> GCC 4.9 was marked BROKEN with r46089. Recently GCC 5.2 was added (marked as broken too)
>> commit msg: " toolchain: mark gcc 4.9 as broken, it miscompiles some code especially on ARM"
>> Looking through some GCC 5 serious regressions I found:
>> I cannot replicate the ARM issue because of missing details and I dont have ARM hw where it occurs (arch: malta seems fine)
>> - but there was an open ticket from a dev on github regarding the boot problem, which might be a symptom of code miscompile
>> According to his tests GCC4.9 and GCC5.2 now work with this 2 changes:
>> adding -fno-ipa-sra to Kernel/CPU CFLAGS
> Readin the bug ticket, it seems that
> should fix the issue at the root cause, but with the "disadvantage" of
> breaking some testcases relying on the (mis-?)behaviour, which sounds
> like the "blocker" for this change.
> So I wonder if adding that wouldn't be a better solution.
I don't think that's the only bug affecting ARM. At least for us,
against an ARM11, we found that there is some other optimization that
results in broken code generation (numbers where being converted to
unsigned, and the optimization failed to convert them back to signed
correctly). We reported it several months ago, but haven't heard
anything. I'm not sure what else it broke, but the kernel logging was
the most noticeable.
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
More information about the openwrt-devel