I've create a blueprint covering the basic functions in libav that are
implemented as inline assembly:
https://blueprints.launchpad.net/gcc-linaro/+spec/investigate-libav-inline-…
These are a mix of multiplies, clipping, byte swap, and unaligned
access. We do OK on half of them but at least byte swap and 32x32 ->
top half of 64 aren't as good as they could be.
Let's discuss the investigation at the next performance meeting.
-- Michael
Hi,
Linaro bugs:
- #926472: Native gdb can't know multithreads of Android processes
Tried to reproduce but couldn't do a GDB build similar to
the bug reporter's. Found a workaround for him anyway.
- #942529: gdb failed to debug application (gdb reports about
internal error)
Reproduced the bug and found out it's fixed in GDB 7.4. It's a bug
in Ubuntu GDB not Linaro, so reassigned to the Ubuntu project.
Linaro GDB is already at version 7.4 so we're good.
- #954714: dropped patch reintroduces gcore relro backtrace problem
Reviewed the proposed patch. It's good.
GDB for Android:
- Started creating patches for build problems when trying to compile
gdbserver for Android. Will have patches to submit in the next
few days.
- Studied libthread_db.c to see what's special about it that it has
to be statically linked. Didn't find anything and indeed Android's
own libthread_db.so works fine. Didn't get an answer yet when
I asked about it on the Android mailing list.
--
[]'s
Thiago Jung Bauermann
Linaro Toolchain Working Group
==Progress===
* Did some work to improve code generation for addressing modes in VFP
registers.
* Looked at some more cases and detected that there are 2 areas of
improvement for
the future
* shrink-wrapping
* Too many unnecessary ldm / stm stacking for cases with small
structures being
passed .
* Committed configure.ac changes.
* Committed MALLOC_ABI_ALIGNMENT changes.
* Looked at some upstream testsuite failures. There are about 54 too
many test failures in vect.exp that
need to be looked at at some point.
=== Plans ===
* Push vfp wb patches upstream.
* Resurrect partial-partial PRE
Absences.
* 1 week holiday sometime before that - to be booked
* Linaro Connect Q2.12 - May 28 - June 1 - travel booked - hotel to be booked.
I've created blueprints for the libav missed optimisations and added
references to Ramana's page at:
https://wiki.linaro.org/RamanaRadhakrishnan/Sandbox/RRQ112ConnectLibavgcc46…
Those that are new and in progress I've assigned to the relevant
person. Ulrich, I couldn't find a end of loop counter blueprint so I
created a new one.
-- Michael
This is the first announcement on upcoming changes to the supported
Linaro GCC versions.
GCC 4.7 is expected out in the next two weeks. We plan to switch to
4.7 for the Linaro GCC 2012.04 release and, as part of that, will put
Linaro GCC 4.6 into maintenance and retire Linaro GCC 4.5. While in
maintenance we will continue to update, fix bugs, and do releases on
4.6. No further changes or releases will be made to 4.5. All
historical releases and branches will stay available.
For more informatio, please see the flyer at:
https://wiki.linaro.org/WorkingGroups/ToolChain/Flyer
especially the section on Lifecycle:
https://wiki.linaro.org/WorkingGroups/ToolChain/Flyer#Lifecycle
The formal change notes will be sent after the 2012.04 release.
-- Michael
Hi,
OpenEmbedded:
* rebased against current OE-core
* my patch that introduces an image fs alignment is now upstream
* noticed that the newly introduced bdwgc recipe (the Hans Boehm
Garbage collector which is needed by guile) uses a version of libatomic
that fails when building in Thumb mode
* has been fixed upstream already
* pulseaudio/libatomics-ops_1.2.bb has the same issue and just sets
ARM_INSTRUCTION_SET to "arm" (probably until a new version gets pulled in)
* we can use the same workaround till then
* found a bug in the recipe for the Linaro external binary toolchain
where some header files are placed into a wrong directory during the
install step
* which prevents python from being built
* now fixed
Please note that I'll be on vacation till 04/13/2012.
Regards,
Ken
Hi Ken. I've made a meta-linaro project on Launchpad and created the
near term blueprints. See:
https://blueprints.launchpad.net/meta-linaro
You're almost done with the initial layer work. Next is scripting up
the build and integrating with LAVA followed by turning on the build
for other architectures.
I've dropped the build in the cloud investigation. Let's use
tcserver01 as it works and is fast enough. Keep an eye on the
validation lab cloud the Validation guys are talking about - we might
shift there later.
I've sent the draft card to Loïc who will massage the text and get it
on the TSC agenda. The current blueprints focus on toolchain
validation and come to a hard stop past there. Once we have TSC
approval we can go further and talk with the Platforms guys about who
does what.
-- Michael
The Validation lab is shifting to a new location this week so I've
started a graceful shutdown of the tcpanda boards. Merge requests
will continue to queue and run on x86 but will back up until the lab
comes online at the end of this week. If the backlog gets too big
then I'll re-enable an ursa board or two.
Please work as normal but expect the builds results to take longer to
come through.
-- Michael
Hi there. Over the next three months both GCC 4.7 and Ubuntu 12.04
'Precise' are coming out. We'll switch over to these pretty quickly
which will affect our internal testing and anyone using the binary
toolchain.
The changeover plan including dates, details of what's happening, and
backwards compatibility is at:
https://wiki.linaro.org/MichaelHope/Sandbox/BinariesMigration
Comments and concerns are welcome,
-- Michael
* Linaro GCC
Spun release tarballs for Linaro GCC 4.5 and 4.6. Launched the build and
test runs. When those completed, briefly checked the results, and
launched the package build tests and benchmark runs.
Continued trying to figure out how my NEON 64-bit immediates patch had
caused a bootstrap failure. The problem turned out to be an assembler
bug ([1]). I've adjusted my patch to work around the problem. This is
the only real way to do it given that, even if I fixed the assembler
right away, there'll be broken assemblers knocking around for some time
to come. The compiler now bootstraps correctly in a manual build. I've
submitted it for testing on Michael's systems, so hopefully this patch
will be ready to post back upstream soonish.
Continued trying to figure out the neon shifts bootstrap failure. I ran
a cross-compiler test suite run, but didn't witness any obvious
problems. Launched a manual bootstrap to reproduce the problem (once the
board because available after testing the immediates patch), so I should
have something to work with next week.
Attempted to merge lp:gcc/4.7 to lp:gcc-linaro/4.7, but got a bzr error.
I asked for help on #bzr, and 'jelmer' is looking into it. Apparently
the history didn't import in quite the same way as lp:gcc/trunk, or
something.
Helped Dmitry Antipov with a relocation (possible) bug.
* Other
Two days leave (Thursday and Friday).
Booked flights and hotels to Hong Kong for Linaro Connect Q2.12
[1] http://sourceware.org/bugzilla/show_bug.cgi?id=13843
Summary:
* Linaro binary toolchain 2012.03 release.
* Make multilib and multiarch work together.
* Failed case analysis for -Os.
Details:
1. Handover Asa's linaro day-to-day jobs (check auto build health,
proposals health and ticket triage)
2. Linaro binary toolchain 2012.03 release.
* Update config to 2012.03
* Fix lp:939008 and patch for lp:939143
3. Multilib/multiarch investigation for linaro toolchain.
* Update the logic in set_multilib_dir to set multiarch_dir when it
matches the format (.:.:$(MULTIARCH_DIRNAME)). Now multilib and
multiarch can work together. A reference build to support
marm/march=armv5t is at http://people.linaro.org/~zhenqiangchen/
* Try to build -mfoat-abi=hard/-fpu=neon. But libgcc build fails
with message "ld: error ... uses VFP register arguments, ..."
4. Investigate new failed cases in -Os regression test.
* For gcc.dg/pr30951.c, both option -O0 and -O2 can optimize (x ==
x + 10) to (0). But -Os can not.
* For gcc.dg/atomic-lockfree.c and gcc.dg/atomic-noinline.c, the
optimization is correct. The test cases can only be test with -O0.
Plans:
* Code size benchmark tuning.
Best regards!
-Zhenqiang
Hi!
* V8 - SunSpider
The pieces for building in cbuild and parsing with the
linaro-toolchain-benchmark scripts are in place. Ran the SunSpider
benchmark across a few toolchains with the o2-neon and o3-neon variants.
I have documented my work on this page, but not analyzed the results yet.
https://wiki.linaro.org/AsaSandahl/Sandbox/JavaScriptBenchmark
What worries me is that there is too much variation in some of the test
results. Probably caused by the garbage collection kicking in at
unpredictable times. I will try to monitor the gc and investigate if it can
be controlled somehow.
I would like to do a gcc-4.4 run as well. Android's original toolchain is
based on gcc-4.4.3(?) and that is what I compared against when doing
measurements internally.
However, I have problem compiling C++ with gcc-4.4 - it is this one again:
home/asa-san/cbuild/slaves/ursa3/gcc-4.4.5/gcc-binary/bin/../libexec/gcc/armv7l-unknown-linux-gnueabi/4.4.5/cc1plus:
/home/asa-san/cbuild/slaves/ursa3/gcc-4.4.5/gcc-binary/lib/libstdc++.so.6:
version `GLIBCXX_3.4.14' not found (required by /usr/lib/libppl.so.7)
* Handed over daily tasks to Zhenqiang.
Regards
Åsa
Hi,
OpenEmbedded:
* verified that the release candidate of our 2012.03 toolchain
(both source and binary) is able to build the sato and qt4e images of
oe-core+meta-linaro - they are booting fine using QEMU
* out sick starting from Tue :/
Regards,
Ken
Current Milestones:
|| || Planned || Estimate || Actual ||
||cp15-rework || 2012-01-06 || 2012-03-30 || ||
Historical Milestones:
||initial-a15-system-model || 2012-01-27 || 2012-01-27 || 2012-01-17 ||
||qemu-kvm-getting-started || 2012-03-04?|| 2012-03-04?|| 2012-02-01 ||
== cp15-rework ==
* converted crn=1; still TODO: crn=0, some loose ends; then reassess
the design in the light of experience doing register conversion
* I've estimated another two weeks here but this might be out, because
in practice much of my time is sucked up by 'other' issues
== other ==
* tracked down cause of LP:947888 gpg crash bug: newer glibc use
/proc/self/maps to decide whether a printf format string using '%n'
is in writable memory, and qemu's maps emulation wasn't good enough
* fixed a thumb decoder bug where we were treating 'setend' like 'cps'
* investigated whether we had any conveniently testable cores which take
advantage of the ARM ARM latitude for UNDEFfing even on failed condition
code checks (answer, not really but the KVM code to handle this case
should be small enough not to worry about its not-yet-tested nature)
* qemu-linaro 2012.03 release (lots of bug fixes, plus exynos4 and
highbank models thanks to Samsung and Calxeda)
* code review: imx31 board patches
* rebased qemu-linaro on upstream and applied some new patches from
Christoffer for ARM KVM support
* LP:956799: added ppoll to QEMU arm-linux-user (a one liner...)
* boot-wrapper: moved initrd load address up so we can handle large
kernels (like Android!)
* sent pullreqs for target-arm, arm-devs patchqueues
== GCC ==
* Checked first part of fwprop-subreg patch into mainline.
* Checked Ira's vectorizer patches into mainline.
* Ongoing work on improving end-of-loop value computation.
Mit freundlichen Gruessen / Best Regards
Ulrich Weigand
--
Dr. Ulrich Weigand | Phone: +49-7031/16-3727
STSM, GNU compiler and toolchain for Linux on System z and Cell/B.E.
IBM Deutschland Research & Development GmbH
Vorsitzender des Aufsichtsrats: Martin Jetter | Geschäftsführung: Dirk
Wittkopp
Sitz der Gesellschaft: Böblingen | Registergericht: Amtsgericht
Stuttgart, HRB 243294
The final version of the 'Building at -O3' writeup is at:
https://wiki.linaro.org/Internal/ToolChain/BuildingAtO3
This updates the SPEC 2000 results to show that there is a net win
which is held back by a bad (but understood) regression in one
benchmark.
Thank you all for your work on this,
-- Michael
The Linaro Toolchain Working Group is pleased to announce the release of
Linaro QEMU 2012.03.
Linaro QEMU 2012.03 is the latest monthly release of qemu-linaro. Based
off upstream (trunk) QEMU, it includes a number of ARM-focused bug fixes
and enhancements.
Highlights in this month's release:
- we now default to enabling 'reserve memory for guest' on 64 bit hosts
in linux-user-mode. This significantly reduces the chances of QEMU
being unable to satisfy a guest process mmap() request.
- Fix for a bug that was causing spurious failures of the glibc check
for "%n in a format string must be in a read-only area of memory"
when running in linux-user-mode.
- QEMU's built-in boot loader now supports passing a device tree blob
to the kernel: if you boot with -kernel mykernel (and optionally
-initrd myinitrd) you can now also use the new command line option
-dtb my.dtb to pass a device tree.
- This version includes an initial implementation of a model of the
Samsung Exynos4210 SoC, used by board models 'nuri' and 'smdkc210'
(thanks to Evgeny Voevodin, Maksim Kozlov, Igor Mitsyanko and
Dmitry Solodkiy from Samsung, who submitted this work to upstream
QEMU).
- This version includes an initial implementation of a model of the
Calxeda Highbank SoC, used by board model 'highbank' (thanks to Rob
Herring and Mark Langsdorf of Calxeda, who submitted this work to
upstream QEMU).
- various other minor bug fixes (detailed in the Changelog.LINARO).
Known issues:
- Graphics do not work for OMAP3 based models (beagle, overo)
with 11.10 Linaro images.
The source tarball is available at:
https://launchpad.net/qemu-linaro/+milestone/2012.03
More information on Linaro QEMU is available at:
https://launchpad.net/qemu-linaro
The GCC release tested up just fine. The branch is now open for commits.
The next release is Thursday the 12th of April. Note that this is the
week after Easter.
-- Michael
The Linaro Toolchain Working Group is pleased to announce the 2012.03
release of both Linaro GCC 4.6 and Linaro GCC 4.5.
Linaro GCC 4.6 2012.03 is the thirteenth release in the 4.6 series. Based
off the latest GCC 4.6.3 release, it contains a new scheduler pressure pass,
implements new instructions, and contains a number of bug fixes.
Interesting changes include:
* Updates to 4.6.3.
* Better performance by accounting for register pressure when
scheduling instructions.
* Support for the ARMv6 USAT/SSAT saturation instructions.
* Support for the VFP VCVT fixed to floating point conversion instruction.
Fixes:
* LP: #922474 Bug in __sync_lock_release with 64 bit primitives
* LP: #923397 Alignment attribute has no effect under certain conditions
* LP: #926855 [ARMhf] gcc produces assembler it can't compile
* LP: #936863 ICE in constprop.2 (ARM NEON related?)
* LP: #942307 'asm' operand requires impossible reload
* LP: #952565 Not compliant with the ABI for multi-register NEON intrinsics
Linaro GCC 4.5 2012.03 is the nineteenth release in the 4.5 series. Based
off the latest GCC 4.5.3+svn184976 release, this is a maintenance only
update.
Interesting changes include:
* Updates to 4.5.3+svn184976.
The source tarballs are available from:
https://launchpad.net/gcc-linaro/+milestone/4.6-2012.03https://launchpad.net/gcc-linaro/+milestone/4.5-2012.03
Downloads are available from the Linaro GCC page on Launchpad:
https://launchpad.net/gcc-linaro
More information on the features and issues are available from the
release page:
https://launchpad.net/gcc-linaro/4.6/4.6-2012.03
Mailing list: http://lists.linaro.org/mailman/listinfo/linaro-toolchain
Bugs: https://bugs.launchpad.net/gcc-linaro/
Questions? https://ask.linaro.org/
Interested in commercial support? Inquire at support(a)linaro.org
-- Michael
EEMBC have announced AndEBench, a mixed native/Java benchmark for
Android. It's available on the Market.
I don't know much more. It seems to be CPU bound and the test names
sound a bit like CoreMark. The source is available for license and
might be worth the Android guys looking into.
The awesome thing is they used the 2012.01 Linaro Binary toolchain
release to build the native parts - see the compiler version string in
the log :)
-- Michael
Hi,
GDB for Android:
* Wrote first draft of the GDB for Android card.
* Found out that there actually is a libthread_db.so in /system/lib
and was able to compile (after a small hack) an FSF gdbserver 7.3
which uses the libthread_db.so from Android and correctly shows
all threads in the process. So the libthread_db.so from Android
actually works, still have to learn why they don't use it.
* Tried to compile a native GDB which uses libthread_db.so but no
luck so far. There are many differences in bionic's header files
which upset libiberty and gnulib which are not easy to work around.
--
[]'s
Thiago Jung Bauermann
Linaro Toolchain Working Group
We now do a SPEC ref run when benchmarking. I've updated the
linaro-toolchain-benchmarks scripts to handle the changes and spawned
some historical builds to do comparisons. All earlier builds have
been archived to prevent confusion.
The human readable summary is now also included in the results. See:
http://ex.seabright.co.nz/benchmarks/gcc-linaro-4.6-2012.02/logs/armv7l-nat…
for an example.
Half an hour to build, 26 hours to run. Not too bad.
-- Michael
Here's brief notes on running different benchmark variants across the
auto builders. Asa, could you pull these plus your notes into a wiki
page?
Spawn a job:
http://ex.seabright.co.nz/helpers/scheduler/spawn
Merge requests are automatically built. Otherwise, drop arbitrary
tarballs into cbuild@orion:~/snapshots and spawn <tarball name minus
extension>. For example, scp gcc-4.6.3.tar.gz
cbuild@orion:~/snapshots; spawn gcc-4.6.3 into a9-builder.
Jobs:
* gcc-version - build and test GCC
* benchmarks-gcc-version - run coremark, denbench, eembc against the
already built version
* benchmarks-spec2000-gcc-version - run spec2000
Queues:
* a9-builders: anything that can naive build a A9 compiler
* a9-ref: reference A9 boards
* a8-ref: reference A8 boards
Variables:
* BENCHMARKS = list, such as coremark spec2000 pybench - run these
benchmarks instead of the defaults
Variants:
* By default we build o3-neon
* See http://bazaar.launchpad.net/~linaro-toolchain-dev/cbuild/trunk/view/head:/l…
for all names
* Spawn a job with VARIANT_SRC = all and VARIANT_LIST = glob-pattern
Examples:
* VARIANT_LIST = o3-neon o3-vfpv3 (compare NEON with VFPv3D32)
* VARIANT_LIST = o3-neon-cortexa8 o3-neon-cortexa9 (compare
-mtune=cortexa8 vs -mtune=cortexa9(
* VARIANT_LIST = o3-neon-novect o3-neon (compare with/without the vectoriser)
-- Michael
I'd like to announce a change in how the Linaro Toolchain group notify
about our monthly releases. In the past we've sent one email per
product to the linaro-announce list. From this week forward a summary
of all products will be included in the main, end of month
announcement instead.
We'll continue to send a per product emails to the linaro-toolchain
list when the mid-month release is available. If you'd like to get
things two weeks early, please subscribe[1]. You can filter on the
word '[ANNOUNCE]' to filter out the development chatter. A RSS feed
is also available[2].
-- Michael
[1] http://lists.linaro.org/mailman/listinfo/linaro-toolchain
[2] http://feeds.launchpad.net/linaro-toolchain/announcements.atom
