Monday Patch Queue Review update (2021-11-01)

Monday Patch Queue Review update (2021-11-01)

[Carlos O'Donell]

Most recent meeting status is always here:
https://sourceware.org/glibc/wiki/PatchworkReviewMeetings#Update

Meeting: 2021-11-01 @ 0900h EST5EDT

Video/Audio: https://bluejeans.com/9093064454

IRC: #glibc on OFTC.

Review new patches and restart review at the top.

 * State NEW, Delegate NOBODY, 317 patches.
 * Series 4330: Enable EVEX strcmp
  * Carlos noted that with Noah and HJ working together this is enough review
    along with the microbenchmark to commit this code.
 * Series 4329: Fix lld build for armhf
  * HJ: Binary with ARM_EXIDX inbetween to relocation sections looks odd.
  * We fixed a similar issue with fa19d5c48a6b36518ce68412e3bdde6bfa8aa4a6.
  * Rewrite as cleanup and provide timing cycle information.
 * Series 4318: [v1,1/5] string: Make tests birdirectional test-memcpy.c
  * HJ to review.
 * Series 4294: benchtests: Add atan2f, expm1f, log1pf and tanf to bench-math (4 patches)
 * Series 4292: benchtests: Add float version of math functions to bench-math (12 patches)
  * Convert double input to float input.
 * Series 4208: regex: Unnest nested functions in regcomp.c
  * Carlos to look at it.
 * 46656: [PiNg,9] localedata: add new locale ab_GE
 * 46551: powerpc: Use the correct flag for 32-bit known libraries
  * Needs a new commit message.
 * 46481: manual: remove an obsolete requirement on aligned_alloc() usage
  * Status? Suggest version the interface, return error without alignment.
 * Enable hard-coded path in tests.
  * 46294: elf: Also try DT_RUNPATH for internal dlopen [BZ #28455]
  * 46295: Add and use link-test-modules-rpath-link [BZ #28455]
  * Carlos to review.
 * Series 2732: malloc: Improve Huge Page support
 * Maxim Kuvyrkov attended. Hello Maxim!
  * Linaro looking at benchmarking toolchains.
  * Patrick: libmicro has many mirobenchmarks.
  * Stream benchmarks were showing regression for the L3 memcpy issue we fixed before.
  * It is very important that you must find the patch that makes the change.
  * We are doing post-commit testing.
  * Pre-commit testing is the point where it is easiest to prevent a patch going.

-- 
Cheers,
Carlos.

Re: Monday Patch Queue Review update (2021-11-01)

[Noah Goldstein]

On Mon, Nov 1, 2021 at 9:26 PM Carlos O'Donell via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> Most recent meeting status is always here:
> https://sourceware.org/glibc/wiki/PatchworkReviewMeetings#Update
>
> Meeting: 2021-11-01 @ 0900h EST5EDT
>
> Video/Audio: https://bluejeans.com/9093064454
>
> IRC: #glibc on OFTC.
>
> Review new patches and restart review at the top.
>
>  * State NEW, Delegate NOBODY, 317 patches.
>  * Series 4330: Enable EVEX strcmp
>   * Carlos noted that with Noah and HJ working together this is enough review
>     along with the microbenchmark to commit this code.
>  * Series 4329: Fix lld build for armhf
>   * HJ: Binary with ARM_EXIDX inbetween to relocation sections looks odd.
>   * We fixed a similar issue with fa19d5c48a6b36518ce68412e3bdde6bfa8aa4a6.
>   * Rewrite as cleanup and provide timing cycle information.
>  * Series 4318: [v1,1/5] string: Make tests birdirectional test-memcpy.c
>   * HJ to review.
>  * Series 4294: benchtests: Add atan2f, expm1f, log1pf and tanf to bench-math (4 patches)
>  * Series 4292: benchtests: Add float version of math functions to bench-math (12 patches)
>   * Convert double input to float input.
>  * Series 4208: regex: Unnest nested functions in regcomp.c
>   * Carlos to look at it.
>  * 46656: [PiNg,9] localedata: add new locale ab_GE
>  * 46551: powerpc: Use the correct flag for 32-bit known libraries
>   * Needs a new commit message.
>  * 46481: manual: remove an obsolete requirement on aligned_alloc() usage
>   * Status? Suggest version the interface, return error without alignment.
>  * Enable hard-coded path in tests.
>   * 46294: elf: Also try DT_RUNPATH for internal dlopen [BZ #28455]
>   * 46295: Add and use link-test-modules-rpath-link [BZ #28455]
>   * Carlos to review.
>  * Series 2732: malloc: Improve Huge Page support
>  * Maxim Kuvyrkov attended. Hello Maxim!
>   * Linaro looking at benchmarking toolchains.
>   * Patrick: libmicro has many mirobenchmarks.
>   * Stream benchmarks were showing regression for the L3 memcpy issue we fixed before.

Any bugzilla for this? Don't see one.

>   * It is very important that you must find the patch that makes the change.
>   * We are doing post-commit testing.
>   * Pre-commit testing is the point where it is easiest to prevent a patch going.
>
> --
> Cheers,
> Carlos.
>

Re: Monday Patch Queue Review update (2021-11-01)

[Carlos O'Donell]

On 11/1/21 23:19, Noah Goldstein wrote:
> On Mon, Nov 1, 2021 at 9:26 PM Carlos O'Donell via Libc-alpha
> <libc-alpha@sourceware.org> wrote:
>>
>> Most recent meeting status is always here:
>> https://sourceware.org/glibc/wiki/PatchworkReviewMeetings#Update
>>
>> Meeting: 2021-11-01 @ 0900h EST5EDT
>>
>> Video/Audio: https://bluejeans.com/9093064454
>>
>> IRC: #glibc on OFTC.
>>
>> Review new patches and restart review at the top.
>>
>>  * State NEW, Delegate NOBODY, 317 patches.
>>  * Series 4330: Enable EVEX strcmp
>>   * Carlos noted that with Noah and HJ working together this is enough review
>>     along with the microbenchmark to commit this code.
>>  * Series 4329: Fix lld build for armhf
>>   * HJ: Binary with ARM_EXIDX inbetween to relocation sections looks odd.
>>   * We fixed a similar issue with fa19d5c48a6b36518ce68412e3bdde6bfa8aa4a6.
>>   * Rewrite as cleanup and provide timing cycle information.
>>  * Series 4318: [v1,1/5] string: Make tests birdirectional test-memcpy.c
>>   * HJ to review.
>>  * Series 4294: benchtests: Add atan2f, expm1f, log1pf and tanf to bench-math (4 patches)
>>  * Series 4292: benchtests: Add float version of math functions to bench-math (12 patches)
>>   * Convert double input to float input.
>>  * Series 4208: regex: Unnest nested functions in regcomp.c
>>   * Carlos to look at it.
>>  * 46656: [PiNg,9] localedata: add new locale ab_GE
>>  * 46551: powerpc: Use the correct flag for 32-bit known libraries
>>   * Needs a new commit message.
>>  * 46481: manual: remove an obsolete requirement on aligned_alloc() usage
>>   * Status? Suggest version the interface, return error without alignment.
>>  * Enable hard-coded path in tests.
>>   * 46294: elf: Also try DT_RUNPATH for internal dlopen [BZ #28455]
>>   * 46295: Add and use link-test-modules-rpath-link [BZ #28455]
>>   * Carlos to review.
>>  * Series 2732: malloc: Improve Huge Page support
>>  * Maxim Kuvyrkov attended. Hello Maxim!
>>   * Linaro looking at benchmarking toolchains.
>>   * Patrick: libmicro has many mirobenchmarks.
>>   * Stream benchmarks were showing regression for the L3 memcpy issue we fixed before.
> 
> Any bugzilla for this? Don't see one.

No, it was commit d3c57027470b78dba79c6d931e4e409b1fecfc80

Let me expand on this.

Maxim at the meeting asked if there were benchmarks they could add to their CI/CD benchmark program.

Maxim was looking for benchmarks that had previously caught performance problems.

Patrick McGehearty noted that the x86 memcpy issue related to L3 usage was caught downstream.

It was caught downstream by running stream and seeing a performance regression.

The fix is already in glibc.

Maxim was going to look at stream as a performance benchmark that has visibly observed a regression.

Does that answer your question?
 
>>   * It is very important that you must find the patch that makes the change.
>>   * We are doing post-commit testing.
>>   * Pre-commit testing is the point where it is easiest to prevent a patch going.
>>
>> --
>> Cheers,
>> Carlos.
>>
> 


-- 
Cheers,
Carlos.

Re: Monday Patch Queue Review update (2021-11-01)

[Noah Goldstein]

On Tue, Nov 2, 2021 at 11:19 AM Carlos O'Donell <carlos@redhat.com> wrote:
>
> On 11/1/21 23:19, Noah Goldstein wrote:
> > On Mon, Nov 1, 2021 at 9:26 PM Carlos O'Donell via Libc-alpha
> > <libc-alpha@sourceware.org> wrote:
> >>
> >> Most recent meeting status is always here:
> >> https://sourceware.org/glibc/wiki/PatchworkReviewMeetings#Update
> >>
> >> Meeting: 2021-11-01 @ 0900h EST5EDT
> >>
> >> Video/Audio: https://bluejeans.com/9093064454
> >>
> >> IRC: #glibc on OFTC.
> >>
> >> Review new patches and restart review at the top.
> >>
> >>  * State NEW, Delegate NOBODY, 317 patches.
> >>  * Series 4330: Enable EVEX strcmp
> >>   * Carlos noted that with Noah and HJ working together this is enough review
> >>     along with the microbenchmark to commit this code.
> >>  * Series 4329: Fix lld build for armhf
> >>   * HJ: Binary with ARM_EXIDX inbetween to relocation sections looks odd.
> >>   * We fixed a similar issue with fa19d5c48a6b36518ce68412e3bdde6bfa8aa4a6.
> >>   * Rewrite as cleanup and provide timing cycle information.
> >>  * Series 4318: [v1,1/5] string: Make tests birdirectional test-memcpy.c
> >>   * HJ to review.
> >>  * Series 4294: benchtests: Add atan2f, expm1f, log1pf and tanf to bench-math (4 patches)
> >>  * Series 4292: benchtests: Add float version of math functions to bench-math (12 patches)
> >>   * Convert double input to float input.
> >>  * Series 4208: regex: Unnest nested functions in regcomp.c
> >>   * Carlos to look at it.
> >>  * 46656: [PiNg,9] localedata: add new locale ab_GE
> >>  * 46551: powerpc: Use the correct flag for 32-bit known libraries
> >>   * Needs a new commit message.
> >>  * 46481: manual: remove an obsolete requirement on aligned_alloc() usage
> >>   * Status? Suggest version the interface, return error without alignment.
> >>  * Enable hard-coded path in tests.
> >>   * 46294: elf: Also try DT_RUNPATH for internal dlopen [BZ #28455]
> >>   * 46295: Add and use link-test-modules-rpath-link [BZ #28455]
> >>   * Carlos to review.
> >>  * Series 2732: malloc: Improve Huge Page support
> >>  * Maxim Kuvyrkov attended. Hello Maxim!
> >>   * Linaro looking at benchmarking toolchains.
> >>   * Patrick: libmicro has many mirobenchmarks.
> >>   * Stream benchmarks were showing regression for the L3 memcpy issue we fixed before.
> >
> > Any bugzilla for this? Don't see one.
>
> No, it was commit d3c57027470b78dba79c6d931e4e409b1fecfc80
>
> Let me expand on this.
>
> Maxim at the meeting asked if there were benchmarks they could add to their CI/CD benchmark program.
>
> Maxim was looking for benchmarks that had previously caught performance problems.
>
> Patrick McGehearty noted that the x86 memcpy issue related to L3 usage was caught downstream.
>
> It was caught downstream by running stream and seeing a performance regression.
>
> The fix is already in glibc.
>
> Maxim was going to look at stream as a performance benchmark that has visibly observed a regression.
>
> Does that answer your question?

Got it.

Thought there was an active regression in head or 2.34 release


>
> >>   * It is very important that you must find the patch that makes the change.
> >>   * We are doing post-commit testing.
> >>   * Pre-commit testing is the point where it is easiest to prevent a patch going.
> >>
> >> --
> >> Cheers,
> >> Carlos.
> >>
> >
>
>
> --
> Cheers,
> Carlos.
>