[PATCH] R600/SI: Allow commuting some 3 op instructions
Matt Arsenault
arsenm2 at gmail.com
Fri Oct 31 11:50:17 PDT 2014
> On Oct 21, 2014, at 9:26 AM, Matt Arsenault <arsenm2 at gmail.com> wrote:
>
>>
>> On Oct 21, 2014, at 6:38 AM, Tom Stellard <tom at stellard.net> wrote:
>>
>> On Sat, Oct 18, 2014 at 01:57:11AM +0000, Matt Arsenault wrote:
>>> e.g. v_mad_f32 a, b, c -> v_mad_f32 b, a, c
>>>
>>> This simplifies matching v_madmk_f32.
>>>
>>> This looks somewhat surprising, but it appears to be
>>> OK to do this. We can commute src0 and src1 in all
>>> of these instructions, and that's all that appears
>>> to matter.
>>>
>>
>> Are there any advantages to being able to commute VOP3 instructions?
>>
>> -Tom
>
> Mostly just canonicalization to simplify other checks. For example, I have another patch that forms v_madmk_f32 which relies on this to avoid checking for an immediate in the 2nd operand
ping
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141031/10da3662/attachment.html>
More information about the llvm-commits
mailing list