[PATCHES] R600/SI: New V_FRACT fix, intrinsic for S_FLBIT_I32, and more

[Marek Olšák]

New patches are attached. I only added -enable-unsafe-fp-math to the
tests, so that folding NEG modifiers can be tested. Please review.

Marek

On Thu, Mar 12, 2015 at 4:48 PM, Marek Olšák <maraeo at gmail.com> wrote:
> So can I consider that patches 2 & 3 have your "LGTM"?
>
> Marek
>
> On Tue, Mar 10, 2015 at 11:28 PM, Matt Arsenault
> <Matthew.Arsenault at amd.com> wrote:
>> On 03/10/2015 03:21 PM, Marek Olšák wrote:
>>>
>>> If we didn't have to deal with fsub, only one pattern would be needed.
>>> A possible solution is to expand fsub, so that it's translated into
>>> (v_add_f32 a, -b), and then convert it to v_sub_f32 in the shrinking
>>> pass if it's possible.
>>>
>>> The hardware internally expands (v_sub_f32 a, b) into (v_add_f32 a, -b)
>>> anyway.
>>>
>>> Marek
>>
>> When I removed the fsub pseudo, I was originally going to handle it with the
>> expanded pattern, but then I noticed the expansion for fsub did the same
>> thing anyway and was slightly easier. Another option might be to do that
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-R600-SI-Expand-fract-to-floor-then-only-select-V_FRA.patch
Type: text/x-patch
Size: 7897 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150313/a262aa77/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0002-R600-SI-Use-V_FRACT_F64-for-faster-64-bit-floor-on-S.patch
Type: text/x-patch
Size: 12930 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150313/a262aa77/attachment-0001.bin>