* Linaro
Continued trying to understand lower-subreg and its impact on ARM
optimization. After much staring at RTL dumps and trying to minimize
test cases from large code bases I've not found an example of code where
disabling the lowering of pseudo-to-pseudo copies made the code larger
for anything other than unfortunate random knock-on effects (such as no
longer being able to use 16-bit thumb1 opcodes). I posted a message to
this effect on the gcc list, and I'll try benchmarking a patch for this
next week.
In the process of looking at lower-subreg, discovered a silly
missed-optimization bug with many DImode operations, and reported it to
GCC Bugzilla as pr53189.
Hi,
OpenEmbedded-Core/meta-linaro:
* created meta-linaro denzil branch to be used in conjunction with the
oe release
* added a patch that prevents GCC from installing libssp and
libstdc++-v3 to lib64 on X86_64 Linux
* merged patches that use vexpress defconfig only for qemuarmv7a
* built and tested all supported QEMU tarted (ARM, MIPS, PPC, X86,
X86_64) using Linaro GCC 4.6 2012.04
* All images (minimal, sato and Qt) are booting!
* started to work on the automation
Regards,
Ken
Summary:
* arm-linux-gnueabihf and multilib support for linaro toolchain.
* Code size benchmark analysis.
Details:
1. arm-linux-gnueabihf support for linaro toolchain
* Update gcc, libgcc and libstdc++ config to support the triplet.
* Update build scripts to gnueabihf.
* Add sample config for gnueabihf.
2. Multilib support for linaro toolchain
* Merge Terry's multilib patches
(http://gcc.gnu.org/ml/gcc-patches/2012-04/msg00975.htmlhttp://gcc.gnu.org/ml/gcc-patches/2012-05/msg00083.html) and revise
them for linaro.
* Update gcc.c and incpath.c to make multiarch and multilib work together.
* libgcc for armv4t (-mfloat-abi=soft) build fail: ld: error:
armv4t/libgcc_s.so.1.tmp uses VFP register arguments, ... does not.
Root cause: crt*.o from precise sysroot use VFP_args, while other
objects are built with -mfloat-abi=soft. We need additional crt*.o to
build libgcc for armv4t.
3. Code size regression analysis
We find more regression cases introduced by ivopt and loop invariant.
Plans:
* Investigate other code size regressions in 4.7.
* Finalize the arm-linux-gnueabihf and multilib support for linaro toolchain.
Best regards!
-Zhenqiang
I've lost track of the benchmark builds so I've started a manual todo list at:
https://wiki.linaro.org/MichaelHope/Sandbox/Todo
It's been messy with the validation lab being down and some builds
being in my home office and some in Cambridge. Let me know if I'm
missing any.
-- Michael
Hi folks,
We really need to push on with getting the loader path for armhf
standardised. The path that was agreed months ago is
/lib/arm-linux-gnueabihf/ld-linux.so.3
but clearly not everybody is using that yet. Dann has just posted an
updated patch for gcc, and we want to get this reviewed / fixed up /
accepted ASAP. Then we may need to backport it to older gcc releases.
This is *important* so that we can help vendors release binaries that
work on any hard-float distribution. For people who have made binaries
that still use the old, broken location /lib/ld-linux.so.3, we can put
symlinks in place *for now* but in the longer term as many distros
switch to multi-arch the symlink is not an acceptable solution.
I'm working on a more complete spec document for armhf to help us with
this kind of thing, but it's not going as smoothly as I'd hoped and I
don't want to wait for that as a blocker on the linker path.
Cheers,
--
Steve McIntyre steve.mcintyre(a)linaro.org
<http://www.linaro.org/> Linaro.org | Open source software for ARM SoCs
Current Milestones:
|| || Planned || Estimate || Actual ||
||cp15-rework || 2012-01-06 || 2012-06-23 || ||
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 ||
== other ==
* cleaned up and refactored QEMU GIC/NVIC code enough to provide
a solid foundation for the in-kernel irqchip support. Posted
the refactoring bits to qemu-devel. This should be enough to
unblock Marc Z.
* qemu-linaro: LP:978694: added patch from Peter Chubb fixing
beagle bootrom FAT12/FAT16 handling
* finished a trivial bit of cleanup of an omap3 ID register patch
I'd had lying around half-done for a while
* investigating a bug reported by Alex Graf where we get a segfault
gtk-query-immodules-2.0 but only if qemu's 'reserve memory space'
feature is being used: this appears to be an interestingly nasty
case where if the guest does mmap(dll); execute code in dll;
munmap(); mmap(dll 2); execute code in dll 2; and the two mmap()s
happened to pick the same address then we will end up using a
stale cached translation from dll 1 when executing dll 2...
-- PMM
== GCC ==
* Completed testing and benchmarking of patch to use vld1/vstd1
instead of vldm/vstdm for vector moves. Saw no regressions but
only minor benefits.
* Checked in patch to fix LP #959242 (GCC PR tree-optimization/52633)
to FSF mainline and 4.7 branch; backports to Linaro GCC 4.7 and 4.6
pending.
* Checked in mainline patch to enable -fsched-pressure by default
on s390.
* Ongoing work on improving end-of-loop value computation.
* Started investigation of TSVC vectorizer test kernels.
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 Uli,
While looking into something else I ran into these - I wonder how many
of these GCC manages to vectorize ...
http://www.netlib.org/benchmark/livermorec . These look interesting
from a vectorizer kernels point of view.
The other interesting paper of note was this PACT paper on
vectorization benchmarks comparing icc , xlc and GCC which might
provide some interesting hints / reading.
http://polaris.cs.uiuc.edu/~garzaran/doc/pact11.pdf . The appropriate
benchmarks kernels are linked to below.
http://polaris.cs.uiuc.edu/ ̃maleki1/TSVC.tar.gz.
regards,
Ramana
I've put the agenda for Monday's call at:
https://wiki.linaro.org/WorkingGroups/ToolChain/Meetings/2012-05-07
which is:
* Review action items from last meeting
* Connect sessions:
* GDB for Android
* GCC performance call - Live!
* KVM performance
* Renderscript
* Android benchmarking
* Dalvik improvements
* Hard float switchover status
* multiarch upstreaming
* Next week is release week
* vld1, EEMBC results, and noise
* twolf result variance
* KVM minimum features
* UP/UP, Ubuntu
* [[MichaelHope/Sandbox/KVMUseCase]]
Feel free to edit,
-- Michael
I've done some minor updates to the instructions for working with
gcc-linaro, bzr, and merge requests at:
https://wiki.linaro.org/WorkingGroups/ToolChain/BzrTips
The interesting changes are using:
* bzr commit --fixes=lp:nnnn to link a branch to a bug number
* bzr branch --hardlink to cut down on branch time and disk usage
* bzr-merge-changelog to automatically merge ChangeLog.linaro
-- Michael
Greetings,
I successfully built BusyBox using the 4.5.2 toolchain. When I try to build BusyBox using the same config with the 4.6.3 toolchain I get the following error:
The Linaro 4.5.2 toolchain was installed in my Ubuntu 11.04 distro using aptitude.
The Linaro 4.6.3 toolchain binaries, downloaded via Launchpad, were installed into my own tools directory.
busybox # make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi-
scripts/kconfig/conf -s Config.in
can't find file Config.in
make[2]: *** [silentoldconfig] Error 1
make[1]: *** [silentoldconfig] Error 2
make: *** [include/autoconf.h] Error 2
I get the same error if I try 'make oldconfig' or 'make menuconfig'.
Is there a PATH I need to set? Am I missing a package?
Right now my PATH has following 4.6.3 directories set:
linaro-4.6.3/bin
linaro-4.6.3/arm-linux-gnueabi/libc/usr/include
linaro-4.6.3/arm-linux-gnueabi/libc/lib/arm-linux-gnueabi
linaro-4.6.3/arm-linux-gnueabi/libc/usr/include
Thanks,
Dan
I thought I'd send an update on the SPEC 2000 twolf variance. We're
seeing a high amount of variance in the results for the SPEC 2000
twolf, vpr, and galgel benchmarks. I've run tests on a PandaBoard,
Origen, and IGEPv2 and gotten a coefficent of variance of 0.014,
0.017, and 0.003 which suggests that the problem is Cortex-A9
specific. twolf is hard on the cache so my theory is that it's
something to do with the memory subsystem. I currently have a
PandaBoard running with SMP, heap randomisation, virtual address
randomisation, and the branch predictor off and the CPU down clocked
to the non-overdrive 600 MHz. I'll let this run overnight and see if
the results are tighter.
To solve Andrew's immediate problem, I'm running SPEC on the 64 bit
core register shifts on the OMAP3. The results are tight enough that
we should be able to show any regressions.
-- Michael
I added Ubuntu 12.04 'precise' based sysroots to Jenkins yesterday. They
are built using 'armhf' architecture and available in three flavours:
- alip
- nano
- ubuntu-desktop
So far only -dev ones are present, will work on -dbg ones soon.
We have 'nano' one built and available to fetch at
http://snapshots.linaro.org/precise/images/nano-dev/ - rest will be
built when there will be machine free to do that.
They will also be moved to http://snapshots.linaro.org/precise/sysroots/
to not make people try to boot them ;)
Please take a look do they work as you want and tell me what kind of
changes you would like to have, which images are not needed, which
should be added. Are there any users for -dbg style sysroots?
I did see gcc-4.7 fail to build for an sf/hf multilib configuration. The reason
was that gcc -print-multi-directory didn't print anything for the non-default,
and gcc -print-multi-lib only prints `.'. The reason is that set_multilib_dir
in gcc.c only consults MULTILIB_DEFAULTS (only defined in linux-elf.h), but not
the default configure options in configargs.h. This proposed patch updates
MULTILIB_DEFAULTS depending on the configure options. An alternative approach
would be to update set_multilib_dir and print_multilib_info to lookup the
configure_default_options in configargs.h as well.
Note that this didn't fail to build in gcc-4.6, but I can't see yet what change
did cause the build failure.
Matthias
* Linaro
Continued looking at the lower-subreg problem (that the lowering is only
really intended for 32-bit targets that don't have 64-bit operation).
Without NEON this is only really a problem for "ldrd/strd" 64-bit loads
and stores: the patterns are all (or should be all) written such that
they only access DImode regs via SUBREG, so it all works well. When
adding NEON 64-bit this becomes less clear: many operations must remain
in DImode without SUBREG until after register allocation. The
lower-subreg passes mostly cope with this well enough, but it has some
features that attempt to lower zero_extends, certain shifts, and
pseudo-reg copies, unconditionally. I've been investigating what happens
if I disable the pseudo-reg copy "optimization" in the first
lower-subreg pass. As I expected, in many cases it actually leads to
smaller code (more use of LDRD/STRD) without NEON, and even better
results with NEON. Unfortunately I have found so counter-examples, so
I'm trying to figure out what's going on there: for some reason reload
goes crazy and starts spilling things to the stack, even though I can't
see that more registers are required. More investigation required.
Richard E approved my NEON-immediates patch for upstream. I don't have
time to commit it with proper care this week, so that's delayed to next
week.
* Other
Vacation Monday and Tuesday. Had a fun long weekend in Cornwall with my
family.
Progress
Away last week - nothing to report. Will be in BST + 4:30 timezone
this week.
Plans
* Finish off the VFP addressing modes patch.
* Follow up on iterations idiom patch upstream
* Pursue backporting gnu_unique_object upstream.
* Look at some of the existing blueprints and start discussions around
prioritizing this.
* Investigate some of the SEGVs with h-c partitioning in the future.
Hi Zhenqiang. Ubuntu Precise is now out and has switched to hard
float by default. I want to do the same for the next binaries
release. Here's the work that needs to be done:
* Bring in the new sysroot
* Change the triplet to arm-linux-gnueabihf
* Change GCC's configure so it recognises the new triplet
* Change the default float ABI to hard
We should include a soft float (not softfp) multlib libgcc for those
who use the binary toolchain to build bare metal programs like u-boot
or the kernel. They don't use floating point, but the linker will
complain about mixed calling conventions.
I've updated make-sysroot.sh and spun an experimental sysroot at
http://people.linaro.org/~michaelh/incoming/precise-sysroot-armhf-r0.tar.bz2.
Hopefully we'll use Marcin's ones instead.
Matthias has patches for many of these changes. Let's talk about it
at tonight's meeting.
Could you start on these? I'd like the changes done within two weeks
so we have plenty of time to test.
-- Michael
Summary:
* Linaro binary toolchain 2012.04 release.
* Code size benchmark analysis.
Details:
1. Validate and bug fix for linaro binary toolchain 2012.04 release.
2. Investigate code size regressions in 4.7
Find more regression cases due to loop invariant hoisting, and tests
should we can reduce some codesize with option
-fno-move-loop-invariants. Some regressions are due to
pass_reorder_blocks, which is disabled for -Os in 4.7. For some cases,
ivopt will introduce to more codes and function inline might lead to
more spilling. We also try linaro 2012.04 baremetal build. But there
is a few bytes regression compared with 4.7 trunk.
3. Setup the qemu env following Michael’s instructions
(https://wiki.linaro.org/MichaelHope/Sandbox/QEMUCrossTest) and tests
show it work.
4. Try to build gcc-linaro-2012.04 configured with "--with-fpu=neon
--with-float=hard" based on a precise sysroot
(precise-sysroot-armhf-r0.tar.bz2) from
http://people.linaro.org/~michaelh/incoming
* Linux build is OK without any change.
* Mingw32 build reports error. After removing the code, the build PASS.
[ERROR] .../libc/usr/include/arm-linux-gnueabihf/sys/types.h:117:19:
error: two or more data types in declaration specifiers
Plans:
* Investigate other code size regressions in 4.7.
Planed leaves:
* Labor Day holiday: April 30 and May 1.
Best regards!
-Zhenqiang
Hi,
OpenEmbedded-Core/meta-linaro:
* pushed support for Linaro GCC 4.6.4 2012.04 and for the
2012.03-20120326 binary toolchain
* updated the wiki
* created a branch to support GCC 4.7
* built several images using several GCC 4.7 based toolchains (OE,
linaro 4.7.1, binary toolchain 2012-04)
* minimal images are working
* sato image shows splash screen forever (won't bring up the
matchbox window manager)
* only when using the binary toolchain 2012-04 ? -> needs to be
analyzed
* Qt 4.8 requires small patches (posted and merged into
oe-core/master-next)
Libunwind:
* User reports issues when backtracing through signal frame
* reason: the lib falls back on APCS frame parsing which usually
segfaults on EABI systems
* provided debugging hints and insights on how libunwind handles
signal frames
* posted a reduced testcase
Regards,
Ken
== GCC ==
* Fixed regression (ICE when building EEMBC) with patch to use vld1/vstd1
instead of vldm/vstdm for vector moves. Updated merge request and
restarted testing.
* Implemented patch to fix LP #959242; backported to Linaro GCC 4.7 and
created merge request for regression testing and benchmarking.
* Ongoing discussions on -fsched-pressure; worked on patch to enable it
by default on s390.
* 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
Current Milestones:
|| || Planned || Estimate || Actual ||
||cp15-rework || 2012-01-06 || 2012-06-23 || ||
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 ==
* have had no review feedback on this patchset yet, and we're now
into QEMU 1.1 hardfreeze. I'm hoping to get review during the
month we're in freeze and then commit early June when 1.2 opens.
However since that coincides with me being on holiday I've set the
Estimate date to allow for that plus a week or so of buffer.
Actual further work required is probably 1 week max, most of this
is waiting-for-other-people or being-away.
== kvm-boot-wrapper ==
* pushed dtb support changes to git repo
== other ==
* basic QEMU side support for the VGIC kernel implementation
Marc Z has been working on. I have something that seems to
work but it's still a bit prototype and missing features.
* investigated why my guest kernel was causing kvm to exit:
turns out that we haven't implemented the kernel support for
gracefully not using KVM if not booted in hyp mode, so trying
to boot a KVM-aware kernel as a KVM guest doesn't work yet.
* sent off the final few patches which I want in QEMU 1.1 before
hardfreeze at the start of next week
* tested a beagle board linaro snapshot image (it panics on
bootup, LP:989737)
* trying to clarify the KVM todo list...
-- PMM
Greetings,
I successfully built and booted Linux 3.1 for the beaglebone (TI am335x) using the 4.5.2 toolchain. I rebuild the same kernel using the same config with the 4.6.3 toolchain, but the board hangs at "Uncompressing Linux... done, booting the kernel."
I halt the beaglebone at the U-Boot prompt and have it download and run uImage from an NFS mounted RFS.
The Linaro 4.5.2 toolchain was installed in my Ubuntu 11.04 distro using aptitude.
The Linaro 4.6.3 toolchain binaries, downloaded via Launchpad, were installed into my own tools directory.
4.5.2 INFO:
./gcc-linaro-arm-linux-gnueabi-2012.01-20120125_linux/bin/arm-linux-gnueabi-gcc --version arm-linux-gnueabi-gcc (crosstool-NG linaro-1.13.1-2012.01-20120125 - Linaro GCC 2012.01) 4.6.3 20120105 (prerelease) Copyright (C) 2011 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
Built with: $ make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- uImage
4.6.3 INFO:
./bin/arm-linux-gnueabi-gcc --version arm-linux-gnueabi-gcc (crosstool-NG linaro-1.13.1-2012.01-20120125 - Linaro GCC 2012.01) 4.6.3 20120105 (prerelease) Copyright (C) 2011 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
Built with: $ make ARCH=arm CROSS_COMPILE=<mytoolsdir>/gcc-linaro-arm-linux-gnueabi-2012.01-20120125_linux/bin/arm-linux-gnueabi- uImage
Could it have anything to do with my binutils and/or U-Boot version? Any debugging tips?
U-Boot INFO:
U-Boot# version
U-Boot 2011.09-00010-g81c8c79 (Feb 13 2012 - 14:48:03) arm-angstrom-linux-gnueabi-gcc (GCC) 4.5.4 20111126 (prerelease) GNU ld (GNU Binutils) 2.20.1.20100303
Thanks for reading this far.
We use QEMU to test programs built by the toolchain binary release for
correctness. I've written up the instructions for spinning up your
own at:
https://wiki.linaro.org/MichaelHope/Sandbox/QEMUCrossTest
It's focused on simplicity - getting a running, SSH only Cortex-A9 up
and going as soon as possible. It's not the latest, not graphical,
and doesn't replace the deeper documentation at:
https://wiki.linaro.org/Resources/HowTo/Qemu
-- Michael
