cbuild2 benchmarking - TCWG-360 [4/10]
* SPEC 2006 cross-built binaries now running. Easy when you know how.
libm exercising - CARD-1693 [4/10]
* Understood improved libm usage on 1 benchmark.
* Tried 3 more. 1 shows significantly more time in libm on AArch64 over AArch32.
Meetings/mail/etc - [2/10]
=Plan=
cbuild2 benchmarking
* Make sure reporting works
* Get working in either LAVA or 'new TCWG infrastructure'
* Test repeatability
* Get SPEC 2000 working
* Possibly have another go at building tools on AArch64
libm exerising
* Keep working through the list of benchmarks
* Hopefully reach the end and do some analysis
== Issues ==
* none.
== Progress ==
* GCC 4.9 and 4.8 2014.11 (1/10)
- Reviewed FSF branch Merges.
* Lab move (3/10)
- Cherry-picked cbuildv2 master patches in schroot-test
- Created a new jenkins job to tst the "schroot lab move" branch
- The job is able to build all the targets but validation still not works
* Misc: (6/10)
- Caught up with mails and irc logs
- Prepared a patch that add __LINARO_[RELEASE|SPIN]__ macros.
- Reviewed libatomic_ops upstream patch
- Validate libunwind upstream patch
- Various meetings.
== Plan ==
* Back on backports
* Tuesday off (11/11)
= Progress ==
* TCWG-544 - Investigate core mark performance with both at -O3 with
LTO + PGO (6/10)
Different IPA inline decisions happening between Linaro and trunk.
No major differences in other IPA passes. Trunk Inlines crcu32 and not
inlines crcu16. Inlining crcu16 decreases instruction count. Honza
pointed out a patch which inlines fucntions based on profile feedback
and ignores max inlines parameters. It could be reason why trunk
inlines crcu32 and then prevents inlining crcu16. Need to explore on
this.
* Bug fix - Linaro 849. Reproduced and tested it with trunk revision
which fixes it. It looks to be same issue as upstream pr62308 (1/10)
* Misc [3/10]
Emails, linaro and AMD status meetings.
1-1 with inline mangers (Mev, Ryan).
1-1 with christophe.
== Plan ==
* Debug and understand the inline differences trunk vs linaro
compiler for core mark at -O3 with LTO + PGO.
* Expecting one Aarch64 Enablement task.
* Leave on 11/11/2014.
== This week ==
* GCC Modularization Project (9/10)
- Flattening header files
- gimple-streamer.h, tree-streamer.h, lto-streamer.h
- Additional updates based on review:
- Created new header files to contain prototypes from .c files
- Boostrapped changes
- New changes being reviewed
- tree-core.h, cfgloop.h, df.h
- Removed unnecessary includes
- Created new header files to contain exports from .c files
* Misc. (1/10)
- Conference calls
== Next week ==
- Continue flattening of header files on GCC modularization project
- Vacation November 12-14th
== Progress ==
* Zero/sign extension elimination with widening types (TCWG-546 - 9/10)
- benchmarked Spec2k and the improvements are very small
- Coremark fared worse. Looked into the cases and relaxed some of the
constraints.
- Subsequent passes are also not optimizing some of the expected cases
- Re-factored and posted an RFC patch for comments at
https://gcc.gnu.org/ml/gcc-patches/2014-11/msg00756.html
* Improve block memory operations by GCC (TCWG-142 - 1/10)
- Looked at the test-cases and gcc dumps
== Plan ==
* Continue with improve block memory operations by GCC.
* Continue with Zero/sign extension pass based on feedback.
== Progress ==
* GCC 4.8 and 4.9 releases (2/10)
- committed both branch merges after review+checked the validation results
- 4.8 needs another branch merge, to get the last errata-related backport
* GCC trunk/4.9 cross-validation (1/10)
- committed a couple of patches to clean the testsuite
- observed a couple of regressions, no time to investigate/report
* AArch64 sanitizer
- got agreement in principle for a new patch, which would test the
kernel headers version.
* Neon intrinsics tests update (1/10)
- the 2 PR I created last week are probably duplicate.
- ARM posted patches to fix problems in the same area, but don't fix these PRs
* cbuild2
- no progress
* Misc (6/10)
- calls, meetings, support
== Next ==
- Tuesday 11th public holiday
* AArch64 sanitizer
* Neon intrinsics update
* GCC linaro 4.8/4.9 releases
* cbuild2:
- analyze previous results
- look at backport-test and tcwgweb scripts+logs
== Progress ==
* Automation Framework (CARD-1378 4/8)
- Moving TCWG machines to the new office
- Setting up new rack, stacking boards, etc.
- Setting up Junos
* Buildbots (TCWG-76 2/8)
- Moving buildbots to the new office
- Checking libcxxabi failure, marking as XFAIL
- All bots green
* Background (2/8)
- Code review, meetings, discussions, etc.
* 1 day ill
== Plan ==
* Continue lab migration
Hi,
When updating:
>From git://git.linaro.org/toolchain/binutils-gdb
e0f5246..336649d master -> linaro/master
I now get the following build error:
gdb/binutils/readelf.c: In function ‘process_mips_specific’:
gdb/binutils/readelf.c:13522:3: error: format ‘%lu’ expects argument of type
‘long unsigned int’, but argument 2 has type ‘size_t’ [-Werror=format=]
printf (_("<symbol index %lu exceeds number of dynamic symbols>"), i);
^
Thanks,
Cov
--
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
Hi,
after the recent lkml thread on blacklisting some GCC versions (see
below) and the issue in identifying accurately our releases, I propose
to add some Linaro specific macros in our branches (i.e this patch
will not go upstream) to be able to check the Linaro version at
preprocessor time. It will not solve the kernel issue with 4.8.N but
hopefully help if such issues happen again the the futur.
http://thread.gmane.org/gmane.linux.ports.arm.omap/119412
What GCC has for the moment is 3 macros __GNUC__, __GNUC_MINOR__ and
__GNUC_PATCHLEVEL__ that are filled by parsing version number
contained in BASE-VER file, for instance on our 4.9 branch:
__GNUC__ = 4
__GNUC_MINOR__ = 9
__GNUC_PATCHLEVEL__ = 2
In our branches, the Linaro version number is in the LINARO-VERSION
file and has this format:
At release point : 4.9-2014.10
Head of our branch: 4.9-2014.10-1~dev
I want your (the team and users) point of view on the macros we need
to create from it. Here is the options I see:
A - Be fully Linaro consistent:
__LINARO_MAJOR__ = 4
__LINARO_MINOR__ = 9
__LINARO_YEAR__ = 2014
__LINARO_MONTH__ = 10
__LINARO_SPIN__ = 0 or N
__LINARO_STATE = 0 for release or 1 for dev
B - Only give information that are not in the __GNUC* macros:
__LINARO_YEAR__ = 2014
__LINARO_MONTH__ = 10
__LINARO_SPIN__ = 0 or N
__LINARO_STATE = 0 for release or 1 for dev
C - Be more concise:
__LINARO_VERSION__ = 201410
__LINARO_SPIN__ = 0 or N
__LINARO_STATE = 0 for release or 1 for dev
D - Even more:
__LINARO_VERSION__ = 201410N (with N the spin number)
__LINARO_STATE = 0 for release or 1 for dev
E - Hardcore conciseness:
__LINARO__ = 201410NM (N = SPIN M = state)
F - One of the previous ones without STATE information.
G - One of the previous ones without SPIN information.
Do you think it is something we need ?
Do we already have that kind of macros in some products (binutils,
gdb, glibc, ...) ?
What option do you prefer ?
My own feeling is that C+F is sufficient as STATE information is
useless for releases and I don't think dev builds checking have to be
used in another project. But SPIN information can be useful has we're
doing respin because an outstanding issue/improvement has to be
fixed/added to the current release, thus it is the kind of thing you
want to check if the version of the compiler you are using contains.
Thanks,
Yvan
Dear all concerned:
ARM has reported it's 53's bug:AArch64 multiply-accumulate instruction might produce incorrect result
and developed the patch descriped below. will the patch be backported to Linaro 4.9 this month's release.
https://gcc.gnu.org/ml/gcc-cvs/2014-10/msg00335.html
thanks
Peter