[llvm-dev] (RFC) Adjusting default loop fully unroll threshold

Fri Feb 10 15:21:09 PST 2017

Thanks every for the comments.

Do we have a decision here?

Dehao

On Tue, Feb 7, 2017 at 10:24 PM, Hal Finkel <hfinkel at anl.gov> wrote:

>
> On 02/07/2017 05:29 PM, Sanjay Patel via llvm-dev wrote:
>
> Sorry if I missed it, but what machine/CPU are you using to collect the
> perf numbers?
>
> I am concerned that what may be a win on a CPU that keeps a couple of
> hundred instructions in-flight and has many MB of caches will not hold for
> a small core.
>
>
> In my experience, unrolling tends to help weaker cores even more than
> stronger ones because it allows the instruction scheduler more
> opportunities to hide latency. Obviously, instruction-cache pressure is an
> important consideration, but the code size changes here seems small.
>
>
> Is the proposed change universal? Is there a way to undo it?
>
>
> All of the unrolling thresholds should be target-adjustable using the
> TTI::getUnrollingPreferences hook.
>
>  -Hal
>
>
>
> On Tue, Feb 7, 2017 at 3:26 PM, Dehao Chen via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
>> Ping... with the updated code size impact data, any more comments? Any
>> more data that would be interesting to collect?
>>
>> Thanks,
>> Dehao
>>
>> On Thu, Feb 2, 2017 at 2:07 PM, Dehao Chen <dehao at google.com> wrote:
>>
>>> Here is the code size impact for clang, chrome and 24 google internal
>>> benchmarks (name omited, 14 15 16 are encoding/decoding benchmarks similar
>>> as h264). There are 2 columns, for threshold 300 and 450 respectively.
>>>
>>> I also tested the llvm test suite. Changing the threshold to 300/450
>>> does not affect code gen for any binary in the test suite.
>>>
>>>
>>>
>>> 300 450
>>> clang 0.30% 0.63%
>>> chrome 0.00% 0.00%
>>> 1 0.27% 0.67%
>>> 2 0.44% 0.93%
>>> 3 0.44% 0.93%
>>> 4 0.26% 0.53%
>>> 5 0.74% 2.21%
>>> 6 0.74% 2.21%
>>> 7 0.74% 2.21%
>>> 8 0.46% 1.05%
>>> 9 0.35% 0.86%
>>> 10 0.35% 0.86%
>>> 11 0.40% 0.83%
>>> 12 0.32% 0.65%
>>> 13 0.31% 0.64%
>>> 14 4.52% 8.23%
>>> 15 9.90% 19.38%
>>> 16 9.90% 19.38%
>>> 17 0.68% 1.97%
>>> 18 0.21% 0.48%
>>> 19 0.99% 3.44%
>>> 20 0.19% 0.46%
>>> 21 0.57% 1.62%
>>> 22 0.37% 1.05%
>>> 23 0.78% 1.30%
>>> 24 0.51% 1.54%
>>>
>>> On Wed, Feb 1, 2017 at 6:08 PM, Mikhail Zolotukhin via llvm-dev <
>>> llvm-dev at lists.llvm.org> wrote:
>>>
>>>> On Feb 1, 2017, at 4:57 PM, Xinliang David Li via llvm-dev <
>>>> llvm-dev at lists.llvm.org> wrote:
>>>>
>>>> clang, chrome, and some internal large apps are good candidates for
>>>> size metrics.
>>>>
>>>> I'd also add the standard LLVM testsuite just because it's the suite
>>>> everyone in the community can use.
>>>>
>>>> Michael
>>>>
>>>>
>>>> David
>>>>
>>>> On Wed, Feb 1, 2017 at 4:47 PM, Chandler Carruth via llvm-dev <
>>>> llvm-dev at lists.llvm.org> wrote:
>>>>
>>>>> I had suggested having size metrics from somewhat larger applications
>>>>> such as Chrome, Webkit, or Firefox; clang itself; and maybe some of our
>>>>> internal binaries with rough size brackets?
>>>>>
>>>>> On Wed, Feb 1, 2017 at 4:33 PM Dehao Chen <dehao at google.com> wrote:
>>>>>
>>>>>> With the new data points, any comments on whether this can justify
>>>>>> setting fully inline threshold to 300 (or any other number) in O2? I can
>>>>>> collect more data points if it's helpful.
>>>>>>
>>>>>> Thanks,
>>>>>> Dehao
>>>>>>
>>>>>> On Tue, Jan 31, 2017 at 3:20 PM, Dehao Chen <dehao at google.com> wrote:
>>>>>>
>>>>>> Recollected the data from trunk head with stddev data and more
>>>>>> threshold data points attached:
>>>>>>
>>>>>> Performance:
>>>>>>
>>>>>> stddev/mean 300 450 600 750
>>>>>> 403 0.37% 0.11% 0.11% 0.09% 0.79%
>>>>>> 433 0.14% 0.51% 0.25% -0.63% -0.29%
>>>>>> 445 0.08% 0.48% 0.89% 0.12% 0.83%
>>>>>> 447 0.16% 3.50% 2.69% 3.66% 3.59%
>>>>>> 453 0.11% 1.49% 0.45% -0.07% 0.78%
>>>>>> 464 0.17% 0.75% 1.80% 1.86% 1.54%
>>>>>> Code size:
>>>>>>
>>>>>> 300 450 600 750
>>>>>> 403 0.56% 2.41% 2.74% 3.75%
>>>>>> 433 0.96% 2.84% 4.19% 4.87%
>>>>>> 445 2.16% 3.62% 4.48% 5.88%
>>>>>> 447 2.96% 5.09% 6.74% 8.89%
>>>>>> 453 0.94% 1.67% 2.73% 2.96%
>>>>>> 464 8.02% 13.50% 20.51% 26.59%
>>>>>> Compile time is proportional in the experiments and more noisy, so I
>>>>>> did not include it.
>>>>>>
>>>>>> We have >2% speedup on some google internal benchmarks when switching
>>>>>> the threshold from 150 to 300.
>>>>>>
>>>>>> Dehao
>>>>>>
>>>>>> On Mon, Jan 30, 2017 at 5:06 PM, Chandler Carruth <
>>>>>> chandlerc at google.com> wrote:
>>>>>>
>>>>>> On Mon, Jan 30, 2017 at 4:59 PM Mehdi Amini <mehdi.amini at apple.com>
>>>>>> wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>> Another question is about PGO integration: is it already hooked
>>>>>> there? Should we have a more aggressive threshold in a hot function?
>>>>>> (Assuming we’re willing to spend some binary size there but not on the cold
>>>>>> path).
>>>>>>
>>>>>>
>>>>>> I would even wire the *unrolling* the other way: just suppress
>>>>>> unrolling in cold paths to save binary size. rolled loops seem like a
>>>>>> generally good thing in cold code unless they are having some larger impact
>>>>>> (IE, the loop itself is more expensive than the unrolled form).
>>>>>>
>>>>>>
>>>>>>
>>>>>> Agree that we could suppress unrolling in cold path to save code
>>>>>> size. But that's orthogonal with the propose here. This proposal focuses on
>>>>>> O2 performance: shall we have different (higher) fully unroll threshold
>>>>>> than dynamic/partial unroll.
>>>>>>
>>>>>>
>>>>>> I agree that this is (to some extent) orthogonal, and it makes sense
>>>>>> to me to differentiate the threshold for full unroll and the
>>>>>> dynamic/partial case.
>>>>>>
>>>>>>
>>>>>> There is one issue that makes these not orthogonal.
>>>>>>
>>>>>> If even *static* profile hints will reduce some of the code size
>>>>>> increase caused by higher unrolling thresholds for non-cold code, we should
>>>>>> factor that into the tradeoff in picking where the threshold goes.
>>>>>>
>>>>>> However, getting PGO into the full unroller is currently challenging
>>>>>> outside of the new pass manager. We already have some unfortunate hacks
>>>>>> around this in LoopUnswitch that are making the port of it to the new PM
>>>>>> more annoying.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>> _______________________________________________
>>>>> LLVM Developers mailing list
>>>>> llvm-dev at lists.llvm.org
>>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>>>
>>>>>
>>>> _______________________________________________
>>>> LLVM Developers mailing list
>>>> llvm-dev at lists.llvm.org
>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> LLVM Developers mailing list
>>>> llvm-dev at lists.llvm.org
>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>>
>>>>
>>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> llvm-dev at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>
>>
>
>
> _______________________________________________
> LLVM Developers mailing listllvm-dev at lists.llvm.orghttp://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
> --
> Hal Finkel
> Lead, Compiler Technology and Programming Languages
> Leadership Computing Facility
> Argonne National Laboratory
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170210/4fbe2b72/attachment-0001.html>