On 18 March 2016 at 17:51, Jeff Law <law@redhat.com> wrote:
> On 03/18/2016 06:25 AM, Bernd Schmidt wrote:
>>
>> This fixes an oversight in my previous patch here. I used biggest_mode
>> in the assumption that if the reg was used in the function, it would be
>> set to something other than VOIDmode, but that fails if we have a
>> multiword access - only the first hard reg gets its biggest_mode
>> assigned in that case.
>>
>> Bootstrapped and tested on x86_64-linux, ran (just) the new arm testcase
>> manually with arm-eabi. Ok?
>>
>> (The testcase seems to be from glibc. Do we keep the copyright notices
>> on the reduced form)?
>
> I don't recall specific guidance on including the copyright notice on a
> reduced/derived test.
>
> Given the actual copyright on the original code, ISTM the safest thing to do
> is keep the notice intact.
>
> A long long time ago I receive guidance from the FSF WRT what could be
> included in the testsuite -- unfortunately I didn't keep that message. I
> probably should have.
>
>>
>> Bernd
>>
>> 70278.diff
>>
>>
>>         PR rtl-optimization/70278
>>         * lra-constraints.c (split_reg): Handle the case where
>> biggest_mode is
>>         VOIDmode.
>>
>> testsuite/
>>         * gcc.dg/torture/pr70278.c: New test.
>>         * gcc.target/arm/pr70278.c: New test.
>

The ARM test isn't sufficiently protected against non-compliant configurations,
and fails if GCC is configured for arm*linux-gnueabihf for instance
(see http://people.linaro.org/~christophe.lyon/cross-validation/gcc/trunk/234342/report-build-info.html)

The attached small patch fixes that by requiring arm_arch_v4t_multilib
effective target.

I used arm_arch_v4t_multilib instead of arm_arch_v4t because, as I
reported a long time ago
the later does not complain in some unsupported configuration because
the sample effective
target test does not contain actual code. In particular it's not
sufficient to reject thumb-1 with
hard-float.

OK?

Thanks,

Christophe.

> OK.
> jeff
>