[llvm-dev] [ScalarEvolution][SCEV] no-wrap flags dependent on order of getSCEV() calls
Sanjoy Das via llvm-dev
llvm-dev at lists.llvm.org
Sun Aug 13 22:36:38 PDT 2017
Hi Geoff,
On Wed, Aug 9, 2017 at 8:58 AM, Geoff Berry <gberry at codeaurora.org> wrote:
> On 8/8/2017 8:38 PM, Sanjoy Das wrote:
>>
>> Hi,
>>
>> On Tue, Aug 8, 2017 at 12:58 PM, Friedman, Eli <efriedma at codeaurora.org>
>> wrote:
>>>
>>> Oh, I see... yes, we do stupid things involving mutating NoWrap flags
>>> after
>>> a SCEV is created. (grep for setNoWrapFlags in ScalarEvolution.cpp.)
>>
>>
>> That's really a compile time hack -- we defer some expensive tricks to
>> prove nsw/nuw on an add recurrences to when we've been asked to
>> sign/zero extend said add recurrence.
>
>
> If that is the case, preserving SCEV analysis where we didn't before should
> only
> result in later passes seeing more nsw/nuw flags, which should theoretically
> be
> beneficial?
There are downsides to preserving SCEV for "too long" though -- if you
simplified a loop in a way that SCEV can compute its trip count when
it couldn't before (or it could compute a trip count before, but now
it can compute a more precise trip count) then invalidating the cache
can be an improvement. However, in this case judicious use of
forgetLoop() (as opposed to re-constructing the whole of SCEV) will
also do the job.
-- Sanjoy
>> I would be okay if you wanted to expose those tricks via a helper in
>> SCEV that could be called independently of creating zero or sign
>> extensions. That way SCEV clients could trade-off compile time for> more
>> precise information in a granular way.
>
>
> I'm not concerned with getting better SCEV results at the moment, I just
> want to
> make sure that changing these passes to preserve SCEV analysis is behaving
> as
> expected (which it sounds like it is).
>
>>
>> -- Sanjoy
>>
>>>
>>> -Eli
>>>
>>> --
>>> Employee of Qualcomm Innovation Center, Inc.
>>> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a
>>> Linux
>>> Foundation Collaborative Project
>>>
>
> --
> Geoff Berry
> Employee of Qualcomm Datacenter Technologies, Inc.
> Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm
> Technologies, Inc. Qualcomm Technologies, Inc. is a member of the Code
> Aurora Forum, a Linux Foundation Collaborative Project.
More information about the llvm-dev
mailing list