* Linaro GCC
Discovered that part of my conditional execution problems were caused by
some over-zealous peephole optimizations adding CC register clobbers to
arithmetic operations in order to take advantage of the 16-bit encoding.
This is a problem because they can't be converted to conditional
execution so freely because overwriting the CC register prevents any
further conditional execution (and forces a 32-bit encoding
unnecessarily). I wrote a patch to help ifcvt unpick all this and DTRT,
and posted it upstream. It turns out that Benrd (I think) already
reimplemented the peephole optimization in the reorg pass, so all that
is really needed is to delete the now obsolete peepholes. Richard E has
approved the revised patch. Unfortunately I've now discovered that that
exposed a bug in the machine description, so that'll have to be fixed also.
Tried to update the gcc-linaro/4.7 branch, but found the build broken.
In fact, Matthias had already reported it to Bugzilla. I did a binary
search, identified the problem commit, posted the info to bugzilla, and
pushed a merge request for the last working commit for testing. This
tested ok, so I've merged it to the Linaro 4.7 trunk.
Brought the Linaro GCC patch tracker up to date to aid identification of
patches we will need to forward port to 4.7. The remaining 'red' patches
are either not committed upstream, or I wasn't sure.
Attempted to benchmark the 64-bit shift improvements, but the target
board (ursa1) has developed an I/O error.
* Other
Joe Seymour and I went to HP Bristol to attend one of their science
lectures. Professor Jim Al-Khalili gave a talk about the recent
"faster-than-light" neutrinos experiments, and explained why it can't
possibly be true.
http://www.jimal-khalili.com/blog/faster-than-the-speed-of-light.html
Summary:
* crosstool-ng patches for linaro 2012.02 binary toolchain release
Details:
1. Revise the gdb lsb build patches based on gdb-7.4.
2. Clean up old patches and only keep the latest version, since we
only keep the latest version in config for linaro.
3. Window installer patch.
* Scripts to use installjammer to create the win32 install package.
* Apply linaro logo to the install package.
* Sync up the install directory and Start-menu layout with
embedded toolchain.
* Test the package on win7.
Plans:
* Finalize 2012.02 binary toolchain release.
Best regards!
-Zhenqiang
Hi,
libunwind
* discussions about local unwinding on Android
* problem: dl_iterate_phdr isn't available
* idea: parse proc/<pid>/maps but to find the ELF file for a given IP
and get to the ARM specific unwind info from there
* assisted an Android game developer who is using clang to compile
the native parts of the app, the ndk to link it and wondered why
the debuggerd doesn't catch any crashes
OpenEmbedded
* mainly worked on the linaro meta layer
* got the external binary toolchain recipe working
* added the Linaro 3.1 Kernel to the layer
* updated documentation:
https://wiki.linaro.org/KenWerner/Sandbox/OpenEmbedded-Core
Regards,
Ken
RAG:
Red:
Amber:
Green:
Current Milestones:
|| || Planned || Estimate || Actual ||
||cp15-rework || 2012-01-06 || 2012-02-20 || ||
(for blueprint definitions: https://wiki.linaro.org/PeterMaydell/QemuKVM)
Historical Milestones:
||a15-usermode-support || 2011-11-10 || 2011-11-10 || 2011-10-27 ||
||upstream-omap3-cleanup || 2011-11-10 || 2011-12-15 || 2011-12-12 ||
||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 ||
== other ==
* post-Connect miscellany:
* got boot-wrapper into an official location in linaro git repo
* Exynos4 patchset finally made it through a review and is upstream
* vexpress-a15 model also now upstream
* put together arm-devs pullreq and got it committed
* LP:932856 : fix broken omap3 bootrom emulation of raw boot mode
* LP:931940 : fix failure to raise interrupts in PL031 realtime
clock model
* reviewed some qemu gdbstub patches that should fix a problem
with semihosting-via-gdb
* rebased qemu-linaro: another nasty one after upstream QOM
patches landing
* spent some time finishing up cleanup of the omap_i2c patches
in qemu-linaro (a job started and left unfinished a few months back)
== GCC ==
* Verified status of patches in Linaro 4.6 that might not yet
be in mainline / Linaro 4.7. Ported two of Ira's vectorizer
patches and created merge request.
* Successfully tested and benchmarked 4.7 version of Richard's
sched-pressure patch, with similar results as seen on 4.6.
Rebased branch due to conflicts with recent mainline merge;
will commit after retesting.
* Investigated LP #923397 (Broken alignment attribute). Now
fixed in FSF mainline and 4.6 branch, will get merged into
Linaro GCC 4.6 automatically for next release.
* Ongoing work on a patch to allow memory operands with vec_set
and vec_extract to avoid excessive vmov generation in the
PR 51819 test case.
* Started working on a patch to generate usat/ssat instructions
where appropriate.
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
Hi Andrew,
thanks for bringing the 4.6 patch tracker up to date. I've had a look over
everything that's marked red to see what we need to do to bring Linaro 4.7
up to date:
106869 Add -fsched-pressure-algorithm=model and enable it by default on
ARM.
My recent commit of Richard's patch. I've tested a forward-port to the
Linaro 4.7 branch already, but had to rebase it due to merge conflicts
with the recent upstream merge, so I'll commit after retesting:
lp:~uweigand/gcc-linaro/sched-pressure-v2-4.7
106846 Handle non-constants in mixed size condition
106844 Support pattern detection for basic blocks.
Two of Ira's patches that have not yet landed upstream due to stage 3.
I've created a forward-port branch and will merge into Linaro 4.7:
lp:~uweigand/gcc-linaro/vect-patches-4.7
106789 Fix rs6000 vector expansion.
106768 Merge fix for lp721513.
These are actually already upstream (modulo minor cosmetic changes)
and present in Linaro 4.7.
All in all, once the two branches mentioned above are merged,
Linaro 4.7 will have everything that's today in Linaro 4.6.
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
Vorsitzende des Aufsichtsrats: Martina Koederitz | Geschäftsführung: Dirk
Wittkopp
Sitz der Gesellschaft: Böblingen | Registergericht: Amtsgericht
Stuttgart, HRB 243294
Hi!
* Finished gcc-4.7 benchmarking:
https://wiki.linaro.org/Internal/ToolChain/Benchmarks/gcc-4.7_benchmarking
Results uploaded to
lp:~linaro-toolchain-wg/linaro-toolchain-benchmarks/private-runs
* Development benchmarks:
Continued the work started at connect.
Got stuck for a while with a problem with the launchpad staging area. I
could no longer log in. Seems the staging area is reset at weekends. The
user credentials has to be given again, but you first have to remove them
manually from the GNOME keyring (or whatever place they are stored).
* Bug triaging:
Triaged these bugs: https://launchpad.net/bugs/926855,
https://launchpad.net/bugs/925991, https://launchpad.net/bugs/924726.
Regards
Åsa
Hi there. I've cancelled the Monday and stand up calls for this week.
Andrew, could you run the standup call on Thursday please? The
moderator details are at:
https://wiki.linaro.org/Internal/ToolChain
-- Michael
* Linaro GCC
Continued work on 64-bit shifts. My patch for shifts in core registers
was rejected because I used conditional-execution before it's safe,
apparently, so I've spent some time looking at the alternatives.
First, conditional move instructions were unsatisfactory - the final
instruction sequence was longer and therefore less efficient.
Second, if_then_else instructions (basically conditional execution
expressed a different way) was to limited - there were no patterns
compatible with "or with shifted input" available in ARM mode, and no
patterns available at all in Thumb mode. I solved that easily enough, of
course, but the result was still unsatisfactory. In Thumb mode the
result was roughly equivalent (conditional-execution is constrained by
IT blocks), but in ARM mode it uses patterns that output two opposing
instructions at once, without allowing for the possibility of scheduling
or other optimization.
Third, I've tried putting in ordinary branches and relying on the
if-conversion pass to transform that to conditional execution. I had not
done this initially, partially because I didn't know how, but mostly
because the original implementation I was trying to replace resulted in
branches in the output code, and I was trying to improve on that.
Anyway, I've now figured out how to do that, and initial testing
suggests that the if-conversion really is working this time. Even if it
doesn't always, the other improvements to the shift sequence should
still give a boost.
My other patch, to do 64-bits shifts in NEON (and fall back to the above
code when register allocation sees fit) has not yet been reviewed.
* Linaro Connect
This week was Linaro Connect week. I could not attend the event in San
Francisco in person this time, so I took part in some of the morning
sessions remotely.
Linaro have switched to using Google Hangouts for the remote
participation, and it went really rather well. In the old system there
was an audio stream to listen to (on a 20 second delay or so, usually),
and an IRC room projected onto a large screen for remote attendees to
respond. The new Hangout system is effectively a simple in-browser
video-conference system that allowed us to not only see the room, but
also respond verbally in real time. The only disadvantage is the
10-person limit, but I never saw any room get near that all week.
I successfully attended several sessions and one regular meeting
remotely by these means. And no jet lag or eating so much!
