[llvm-dev] Regarding basic block layout/code placement optimizations of profile guided optimization (PGO)

[Michael Spencer via llvm-dev]

On Thu, Aug 9, 2018 at 2:07 PM Xiaowan Dong <xdong at cs.rochester.edu> wrote:

> Dear Davide,
>
> We noticed that C^3 will only be available for the upcoming versions of
> LLVM. I was wondering what it is like in LLVM 6.0.1?
>
> Thank you so much!
>
> Best regards,
> Xiaowan
>

LLVM currently does basic block layout per function based on profile data.
See lib/CodeGen/MachineBlockPlacement.cpp .

- Michael Spencer


>
> On Tue, Aug 7, 2018 at 5:14 PM, Davide Italiano <davide at freebsd.org>
> wrote:
>
>> On Tue, Aug 7, 2018 at 11:53 AM Xiaowan Dong via llvm-dev
>> <llvm-dev at lists.llvm.org> wrote:
>> >
>> > Hi,
>> >
>> > I would like to learn the details regarding what exactly PGO does for
>> basic block layout/code placement optimizations in llvm. Could you please
>> point me to some descriptions? Is it close to this paper (Karl Pettis and
>> Robert C. Hansen. 1990. Profile guided code positioning.  PLDI'90)
>> https://urldefense.proofpoint.com/v2/url?u=http-3A__perso.ensta-2Dparistech.fr_-7Ebmonsuez_Cours_B6-2D4_Articles_papers15.pdf-3F&d=DwIFaQ&c=kbmfwr1Yojg42sGEpaQh5ofMHBeTl9EI2eaqQZhHbOU&r=9lVsUv0m4rDQCgSupIkpgbizkVWsSr2-oDgbaUgY_C0&m=YPozo_Hjjg-5VCjvXCWDF1LsHc3pKWNWySkv3nXqSFc&s=Hbba3m9Gul1G8YndUAW1N9qQpS3nHs9DA7c0CJVNebg&e=
>> Whether it is purely intra-procedural or include inter-procedural
>> optimizations?
>> >
>> > Thank you so much!
>> >
>>
>> lld implements C^3 which is basically a variation of Pettis-Hansen.
>> The CGO 17 paper describes the approach with a great level of detail
>>
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__research.fb.com_wp-2Dcontent_uploads_2017_01_cgo2017-2Dhfsort-2Dfinal1.pdf-3F&d=DwIFaQ&c=kbmfwr1Yojg42sGEpaQh5ofMHBeTl9EI2eaqQZhHbOU&r=9lVsUv0m4rDQCgSupIkpgbizkVWsSr2-oDgbaUgY_C0&m=YPozo_Hjjg-5VCjvXCWDF1LsHc3pKWNWySkv3nXqSFc&s=pIdMmcvpHh75XIH29J98lMchoSRMHF4PjYjsw5fWV2c&e=
>> (the algorithm per-se is fairly straightforward).
>> Pointers to the code at
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__reviews.llvm.org_D36351&d=DwIFaQ&c=kbmfwr1Yojg42sGEpaQh5ofMHBeTl9EI2eaqQZhHbOU&r=9lVsUv0m4rDQCgSupIkpgbizkVWsSr2-oDgbaUgY_C0&m=YPozo_Hjjg-5VCjvXCWDF1LsHc3pKWNWySkv3nXqSFc&s=zyCFsKdWzhdEH2AAV0y8itEL3f3MCq_97koJeZEBsGY&e=
>> You can also find the details in Michael's original RFC
>>
>> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.llvm.org_pipermail_llvm-2Ddev_2017-2DJune_114178.html&d=DwIFaQ&c=kbmfwr1Yojg42sGEpaQh5ofMHBeTl9EI2eaqQZhHbOU&r=9lVsUv0m4rDQCgSupIkpgbizkVWsSr2-oDgbaUgY_C0&m=YPozo_Hjjg-5VCjvXCWDF1LsHc3pKWNWySkv3nXqSFc&s=DGrA_unYXo8wrHhtd-cK2GiyQOfsYqvIz016-wcV4iU&e=
>>
>> There's also a pass in the backend (codegen) to perform profile-driven
>> Machine BB placement (CodeGen/MachineBlockPlacement.cpp).
>>
>> --
>> Davide
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20180809/dfd0b560/attachment.html>