[PATCH 1/3, v2] rs6000: Add base support and types for defining MMA built-ins.
Peter Bergner
bergner@linux.ibm.com
Fri Jun 19 19:33:21 GMT 2020
On 6/19/20 1:12 PM, Segher Boessenkool wrote:
> On Fri, Jun 19, 2020 at 11:47:35AM -0500, Peter Bergner wrote:
>>> Why are OImode and XImode handled here?
>>>
>>>> static bool
>>>> rs6000_modes_tieable_p (machine_mode mode1, machine_mode mode2)
>>>> {
>>
>> Do you mean why *aren't* they handled in rs6000_modes_tieable_p?
>
> No, this is a comment about the stuff above my comment, so
>
>> + /* MMA accumulator modes need FPR registers divisible by 4. We need to allow
>> + XImode to have the same registers as PXImode, even though we do not enable
>> + the move pattern for XImode. */
>> + if (mode == PXImode || mode == XImode)
>> + return (TARGET_MMA && FP_REGNO_P (regno)
>> + && (regno & 3) == 0);
>
> and the one with
>
>> + if (mode == POImode || mode == OImode)
>
> before it.
Ah, ok. Yeah, I think that was an oversight and we shouldn't need those.
I'll remove them.
>>> Same for the CCFP one here.
>>
>> Mike added those. I guess I thought they were needed. Mike?
>> If they're not needed for MMA, I'll remove them from this patch
>> and they be submitted in a separate patch if they are needed for
>> something else.
>
> You can keep them, it's compiler debug only, but the changelog should
> mention it (it looks like an accident now, which maybe it was ;-) )
Ok, I'll add a changelog entry for them then...unless Mike comes back
before my testing is done and says we don't need them at all.
Peter
More information about the Gcc-patches
mailing list