== Progress ==
* Trying to profile ImageMagick on A9
- perf is broken on Exynos 4, used Panda
- Linaro image doesn't have all PPAs
- Clang code 10% worse than GCC on some operations
- Code doesn't look that bad, but Clang on x86 is faster
- Will look into that next week
Background
- Meetings at ARM, Linaro
- Omair in Cambridge
- Patch review, fixing bots, discussions, etc
- Moved from Ubuntu to Mint 15, using Gnome Shell
== Plan ==
* Finally look into the profiler output
* Increase the profiling to other image operations
* Profile Phoronix results?
Hi,
In building some code (CRIU) that includes linux/if_packet.h for
x86_64-linux-gnu, arm-linux-gnueabihf, and aarch64-linux-gnu, some
discrepancies popped up relating to whether the TPACKET_V3 structures were
defined. It looks like the headers shipped with the 13.08 arm-linux-gnueabihf
toolchain do not define them, implying kernel headers older than October 2012.
The aarch64-linux-gnu toolchain on the other hand does have them defined. If
this analysis is correct, can I expect the kernel headers in the 32-bit
toolchain to be updated in the near future or should I really be generating
and referencing my own kernel headers?
Thanks,
Christopher
--
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by the Linux Foundation.
== Progress ==
* GDB Record and Replay (TCWG-197)
Debugged code for an exception coming up in gdb code after applying
Thumb32 patch.
* Travel to UK on Wednesday night stopped in UAE for Thursday and Friday.
* Holiday on Thursday and Friday.
== Plan ==
* Working from Linaro Cambridge from Monday till Friday
* GDB Record and Replay (TCWG-197)
Fix issues Submit patch upstream.
* Meetings with team mates in Cambridge and visit to ARM.
== Progress ==
* 4.8 2013.09 backports:
- Finished ongoing reviews.
* LRA:
- Fixed and refactored rtl analyser patch, waiting fro upstream green light
- Reduced AArch64 testcase, and find the root cause of the issue:
Only 64 bits of callee saved registered v8-v15 are preserved, larger values
has to be handle by the caller, which not the case with LRA. Working on
the issue with LRA maintainer.
* LCU'13
- Hotel and flight booked
== Next ==
* Continue on LRA
== Issues ==
* None
== Progress ==
* Prepare Linaro toolchain binary 2013.09 release
- Create linaro-prebuilt-sysroot-2013.09.
- Update local patches and all Linaro samples to use
eglibc-linaro-2.18-2013.09.
* Update Conditional compare (CCMP) according to community comments.
Current method is:
- In fold-const, if HAVE_conditional_compare, set
LOGICAL_OP_NON_SHORT_CIRCUIT to true.
- Identify CCMP during expanding. A CCMP candidate is a BIT_AND_EXPR
or BIT_IOR_EXPR, whose operators are compares.
- Add a new op in optab to expand the CCMP to RTL, e.g.
and_scc_scc/ior_scc_scc in ARM.
- Testing is ongoing.
== Plan ==
* Send out the updated conditional compare patches for review.
== Panned leaves ==
* Sept. 19-21: Mid-Autumn Festival
* Oct. 1-7: National Day of China holiday
== Progress ==
* Set up a GCC git mirror on git.linaro.org. now at
ssh://LOGIN@git.linaro.org/srv/git.linaro.org/git/toolchain/gcc.git. The
two Linaro branches most people are interested in are now
'linaro-4.7-branch' (updated every 30 minutes from launchpad),
and 'linaro-4.8-branch' updated every 10 minutes from subversion.
* Tweaked Cbuildv2 & Jenkins to work with GIT repositories and
branches as well as source tarballs.
* Jenkins now "bootstraps" a cross compiler by building the native
GCC, and then using that to compile the cross GCC.
* Got aarch64-none-elf building with newlib via Cbuildv2.
* Source and binary tarball release support in Cbuildv2 now
works. I've had to change the naming convention to handle daily
automated builds, and added the commit ID of HEAD for git based
builds.
* Almost got Canadian crosses building via Cbuildv2.
* Continued work on Jenkins/Cbuildv2 integration.
* Cbuildv2 now works on Fedora as well as Ubuntu.
* Survived major 100+ year flood. Got to watch a dam overflow, kindof
cool, all the roads in and out of this area have been closed for days,
most of the bridges are gone. Been busy all weekend evacuating
people, car accidents, normal disaster area stuff. I worked Katrina
relief, never thought I'd see a flood in the mountains.
== Plan ==
* Continuing with Jenkins/Cbuildv2/Git integration.
* Get testing results from Jenkins builds imported via Junit conversion.
* Compare automated tarballs with the current release tarballs.
* Get Jenkins to upload binary & source tarballs to
snapshots.linaro.org.
* Add support for using specific commit IDs with Git and Cbuildv2.
* Finish getting Canadian Cross builds working in Cbuildv2.
* Odroids should arrive (when the roads open...) Monday, install
Ubuntu and do some tests to see if it's a good platform.
== Issues ==
* Flooding...
== Progress ==
* Addressing patch review
- Woking on vrp based zero/sign extension elimination based on feedback
- Reworked the patch based and posted the first patch
- Speck2k benchmarking in progress for second patch and will post
once finished.
* spec2k comparison between ARM and x86
- More benchmarks analysed.
- Started with the report. Plan to send in next couple of days.
- Created more reduced test cases.
== Plan ==
* Continue with spec2k comparison between ARM and x86
* Start divmod work
== Progress ==
* Committed glibc realloc test
* Committed glibc malloc integer overflow patches and requested CVE number
* More work on glibc malloc benchmarking
* Investigated AArch64 gdb issue and submitted a patch
* Investigated gas NEON issue and submitted a patch
== Issues ==
* None
== Plan ==
* Visit Linaro in Cambridge from Tuesday to Thursday (see you there!)
* Submit glibc malloc benchmark
* Get benchmark numbers for tweaked memcpy
* Start glibc pointer guard work
--
Will Newton
Toolchain Working Group, Linaro
== Progress ==
* Patches
- Committed lzma fix for cross-compilation
- Split ARMv7 fixes in three patches, committed
- Reviewed many ARM patches
* ImageMagick benchmark
- Using IM's convert to test compiler performance
- LLVM seems pretty good on x86_64, par on A15 and behind on A9
when compared to gcc 4.7
- http://people.linaro.org/~rengolin/llvm/scripts/
* Profiling
- Found a serious bug in LLVM when emitting profile information
- mcount segfaults on prelude, luckily only on x86. :D
* Phoronix
- Found out that our problems with Phoronix were not unique
- Might need a bit more of work to get the build scripts changed
- And A bit more to get the data out of the reports
== Plan ==
* Profile IM's convert on A9 (after finding out why it's not generating
gmon.out)
* Hoping to find low-hanging SLP-vectorizer fruits... fingers crossed!
* Try something on Jenkins, hopefully it'll be easy to get Phoronix and
ImageMagick running on it
