[PATCH] [AArch64] Handle vec4 sitofp and uitofp for half

Ahmed Bougacha ahmed.bougacha at gmail.com
Tue Apr 21 15:14:29 PDT 2015

On top of James' request, a test for smaller vector sizes might also
be useful, see https://llvm.org/bugs/show_bug.cgi?id=21277.  Feel free
to take it and mark as fixed, the patch should be enough.

Also, I realize this might require a different fix (if any), but what
about i8->f16, or i16->f16 conversions?  I don't know whether we widen
v4i8 or not, but if we do, to avoid scalarizing, you can probably test
v8i8 -> v8f16.


On Tue, Apr 21, 2015 at 12:49 PM, James Molloy <james at jamesmolloy.co.uk> wrote:
> Hi Pirama,
> This looks OK to me, but could you please add a test for v8f16 and v16f16 to
> check oversize vectors are handled sanely? I'm trying to keep the backend
> oversize-safe.
> Cheers,
> James
> On Tue, 21 Apr 2015 at 20:44, Pirama Arumuga Nainar <pirama at google.com>
> wrote:
>> IIUC, this patch shouldn't affect v4i32 to v4f32 or v4f64 operations, but
>> couldn't find any tests for those operations. I confirmed that locally, and
>> can add tests with this patch if requested.
>> In a test of v4f16 operations (similar to the comprehensive test in
>> http://reviews.llvm.org/D8648), only sitofp and uitofp cause problems, and
>> will be fixed with this patch.
>> http://reviews.llvm.org/D9166
>>   http://reviews.llvm.org/settings/panel/emailpreferences/
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

More information about the llvm-commits mailing list