[PATCH] D81981: [PGO] Supplement PGO profile with Sample profile

[Rong Xu via llvm-commits]

BTW, I used google internal benchmarks that are highly threaded.  I don't
remember the exact numbers. But should be more than 10% runtime increase.

On Thu, Jun 18, 2020 at 11:17 AM Rong Xu <xur at google.com> wrote:

> Ok, got it . That's probably a single thread program. Instrumentation
> overhead is low.
>
> On Thu, Jun 18, 2020 at 11:13 AM Wei Mi <wmi at google.com> wrote:
>
>>
>>
>> On Thu, Jun 18, 2020 at 10:49 AM Rong Xu <xur at google.com> wrote:
>>
>>>
>>>
>>> On Wed, Jun 17, 2020 at 3:07 PM Wei Mi via Phabricator <
>>> reviews at reviews.llvm.org> wrote:
>>>
>>>> wmi added a comment.
>>>>
>>>> In D81981#2099070 <https://reviews.llvm.org/D81981#2099070>, @davidxl
>>>> wrote:
>>>>
>>>> > I think it is good to have an entry counter always, so that the
>>>> profile dump is more readable. Do you have data showing the instrumentation
>>>> overhead and profile size impact (clang and some large app)?
>>>>
>>>>
>>>> Yes, I tried clang. The instrumentation runtime overhead increases by
>>>> about 0.8%. The raw profile size increases by 1.8%. The zipped profile size
>>>> increases by 0.15%.
>>>> Right now in the patch, inserting entry counter is guarded by a flag
>>>> with default value being false.
>>>>
>>>> How did you test clang. This is for one file or for the whole clang?
>>> To me, 0.8% is close to nothing.
>>> My previous patch shows much more than this for a highly threaded
>>> program.
>>>
>>
>> This is to use clang to compile a single large file as a test.
>>
>>
>>
>>>
>>>> Repository:
>>>>   rL LLVM
>>>>
>>>> CHANGES SINCE LAST ACTION
>>>>   https://reviews.llvm.org/D81981/new/
>>>>
>>>> https://reviews.llvm.org/D81981
>>>>
>>>>
>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200618/b0eaed89/attachment.html>