On Thu, Oct 15, 2015 at 12:16 PM, Uros Bizjak <[email protected]> wrote:
> On Thu, Oct 15, 2015 at 6:59 PM, H.J. Lu <[email protected]> wrote:
>> On Thu, Oct 15, 2015 at 9:57 AM, Uros Bizjak <[email protected]> wrote:
>>> On Thu, Oct 15, 2015 at 1:45 PM, Uros Bizjak <[email protected]> wrote:
>>>> On Wed, Oct 14, 2015 at 5:08 PM, Yulia Koval <[email protected]> wrote:
>>>>> Hi,
>>>>>
>>>>> This patch fixes the issue:
>>>>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67963
>>>>>
>>>>> gcc/config/i386/i386.c (ix86_option_override_internal) Disable
>>>>> 80387 mask if lakemont target is set.
>>>>>
>>>>> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
>>>>> index 4c25c9e..db722aa 100644
>>>>> --- a/gcc/config/i386/i386.c
>>>>> +++ b/gcc/config/i386/i386.c
>>>>> @@ -4943,6 +4943,12 @@ ix86_option_override_internal (bool main_args_p,
>>>>> break;
>>>>> }
>>>>>
>>>>> + if (!strcmp (opts->x_ix86_arch_string, "lakemont"))
>>>>> + {
>>>>> + opts->x_target_flags &= ~MASK_80387;
>>>>> + opts_set->x_target_flags |= MASK_80387;
>>>>> + }
>>>>> +
>>>>> if (TARGET_X32 && (opts->x_ix86_isa_flags & OPTION_MASK_ISA_MPX))
>>>>> error ("Intel MPX does not support x32");
>>>>>
>>>>> Ok for trunk?
>>>>
>>>> The problem is in TARGET_SUBTARGET{32,64}_DEFAULT, that will override
>>>> set target flags, unless relevant bit of opts_set->x_target_flags is
>>>> set. However, we can't just set x_target_flag, because
>>>> __attribute__((target("arch=...."))) won't work. Unfortunately, my
>>>> proposed patch in the PR violates this last requirement.
>>>>
>>>> This can probably be solved by adding local "x_target_flags_mask" and
>>>> use it after TARGET_SUBTARGET processing.
>>>
>>> Attached is a final patch I plan to commit to mainline soon.
>>>
>>> 2015-10-15 Uros Bizjak <[email protected]>
>>>
>>> PR target/67963
>>> * config/i386/i386.c (ix86_option_override_internal): Add PTA_NO_80387.
>>> Add PTA_NO_80387 to "lakemont". Disable MASK_80387 target flag
>>> after target flags were initialized to target defaults.
>>>
>>> testsuite/ChangeLog:
>>>
>>> 2015-10-15 Uros Bizjak <[email protected]>
>>>
>>> PR target/67963
>>> * gcc.target/i386/pr67963-1.c: New test.
>>> * gcc.target/i386/pr67963-2.c: Ditto.
>>>
>>> Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.
>>>
>>> Uros.
>>
>> Do we support -O2 -march=lakemont with
>>
>> __attribute__((target("arch=silvermont")))
>
> Hm, no.
>
Do we issue an error or silently ignore
__attribute__((target("arch=silvermont")))?
If we don't support it, should we support
-O2 -march=silvermont
__attribute__((target("arch=lakemont")))
--
H.J.
