* Goal
Goal of this work is to look for thumb2 code size improvements on FSF
GCC trunk.
* Methodology
** Build FSF GCC trunk w/ and wo/ hardfp, run benchmarks including
eembc, spec2000, and dhrystone, and check asm code to see if there is
any possible improvements on size.
** Get input and suggestion from ARM experts.
** Search open PRs in GCC bugzilla.
* Results
Each item has been tracked on launchpad, and is listed with some elements,
** Cause: cause of this problem is known or unknown
** Difficulty: estimation of implementation difficulty
** Recommendation: Yao's recommendation on that bug for next step
1. LP:633233 Push/pop low register rather than high register when
keeping stack alignment
As Richard E. pointed out, it was implemented in gcc-4.5 on 2009, but
Yao still can see the usage of r8 on FSF GCC trunk.
Cause: Might be a regression if problem disappears on gcc-4.5.
Difficulty: Easy. might not hard to fix a regression.
Recommendations: Fix this regression if it is.
2. LP:633243 Improve regrename to make use of low registers.
Get input from Bernd S. and Julian B. Initial implementation has been
suggested by Bernd S.
Cause: current regrename in gcc treats high and low registers equally.
Difficulty: Medium.
Recommendation: Implement it as Bernd suggested, and do benchmarking
to see how much size is improved.
3. LP:634682 Redundant uxth/sxth insn are generated
Cause: Unknown
Difficulty: Unknown
Recommendation: No recommendation so far.
4. LP:634696 Function is not inlined properly with -Os
In consumer/cjpeg/jmemmgr.c, GCC inlined out_of_memory() with -Os, so
increase code size.
Cause: Unknown.
Difficulty: Unknown
Recommendation: Educate GCC to inline carefully when -Os is turned on.
5. GCC PR40730 LP:634731 Redundant memory load
6. LP:634738 inefficient code to extract least bits from an integer value
GCC PR40697 is for thumb-1. The same problem is in thumb-2.
Cause: Unknown.
Difficulty: Medium.
Recommendation: Fix it the similar way as fixing GCC PR40697.
7. LP:634891 Replace load/store by memcpy more aggressively
Difficulty: Should be easy.
Recommendation: Fix to this problem might be "reduce threshold value
once -Os is turned on".
8. LP:637220 allocate local variables with fewer instructions
GCC PR40657 is about this kind of problem, and was fixed. The similar
prolbme exits on gcc with hardfp.
Cause: Unknown.
Difficulty: Unknown.
Recommendation: No recommendation so far.
9. GCC PR 43721 Failure to optimize (a/b) and (a%b) into single
__aeabi_idivmod call
Difficulty: Medium or easy.
Recommendation: No.
10. LP:637814 Combine add/move to add
LP:637882 Combine ldr/mov to ldr
Possible improvements have been found. No idea how to fix it yet.
Cause: Unknown.
Difficulty: Unknown.
Recommendation: No.
11. LP:638014 Replace memset by memclr when 2nd parameter is zero
Difficulty: Easy.
Recommendation: No recommendation so far.
12. LP:625233 Merge constant pools for small functions
Cause: Unknown.
Difficulty: Medium.
Recommendation: No.
13. LP:638935 Replace multiple vldr by vldm
Some vldr insns accessing consecutive address can be replaced by
single vldm. It is not about thumb2, but related to code size optimization.
Cause: Unknown.
Difficulty: Medium.
Recommendation: No.
--
Yao Qi
CodeSourcery
yao(a)codesourcery.com
(650) 331-3385 x739
Hi there. I've always wanted to mix this:
http://www.futurlec.com/ET-STM32_Stamp.shtml
with some of this:
http://bit.ly/cD0JPS
to control my one of these:
http://www.traxxas.com/products/electric/rustler2006/gallery/3705-3qrtr-Bla…
and it sounds like a good opportunity to dogfood the Linaro toolchain
at the same time. What's the best way to set up a Cortex-M3 toolchain
with an appropriate newlib and libgcc?
A wrapper script works fine but I need a way of recompiling libgcc for
the Cortex-M series. I'd love to get a arm-none-eabi toolchain
package out of this that others could use. Could I re-work the cross
packaging to use newlib and change the configure flags instead? Are
there existing Debianised cross packages that I could reuse?
Ta,
-- Michael
Hi Andrew. Well, the builds are done and they're OK. I've added the
ability to compare against an explicit release to make checking
regressions easier.
4.4 results are here:
http://ex.seabright.co.nz/helpers/testcompare/gcc-linaro-4.4-2010.09-1/logs…http://ex.seabright.co.nz/helpers/testcompare/gcc-linaro-4.4-2010.09-1/logs…http://ex.seabright.co.nz/helpers/testcompare/gcc-linaro-4.4-2010.09-1/logs…
i686 and x86_64 have not regressed since 2010.08.
On arm, and ignoring the limits test, 2010.09 adds a failure on
gcc.c-torture/compile/991026-2.c. According to the log the run timed
out but I can't reproduce it.
4.5 results are here:
http://ex.seabright.co.nz/helpers/testcompare/gcc-linaro-4.5-2010.09-0/logs…http://ex.seabright.co.nz/helpers/testcompare/gcc-linaro-4.5-2010.09-0/logs…http://ex.seabright.co.nz/helpers/testcompare/gcc-linaro-4.5-2010.09-0/logs…
i686 has not regressed since 2010.08. x86_64 fails on
gcc.target/i386/wmul-1.c, but this is a new tests for new features and
are not a regression against 4.5.1.
arm is messier. The following new failures exist:
Vectoriser related:
* g++.dg/vect/pr36648.cc scan-tree-dump-times vect "vectorized 1 loops" 1
* g++.dg/vect/pr36648.cc scan-tree-dump-times vect "vectorizing
stmts using SLP" 1
* gcc.dg/vect/vect-multitypes-11.c scan-tree-dump-times vect
"vectorized 1 loops" 1
* gcc.dg/vect/vect-multitypes-12.c scan-tree-dump-times vect
"vectorized 1 loops" 1
* gcc.dg/vect/vect-reduc-dot-s16b.c scan-tree-dump-times vect
"vectorized 1 loops" 0
* gcc.dg/vect/vect-reduc-pattern-1a.c scan-tree-dump-times vect
"vectorized 1 loops" 0
* gcc.dg/vect/vect-reduc-pattern-1b.c scan-tree-dump-times vect
"vectorized 1 loops" 0
* gcc.dg/vect/vect-reduc-pattern-1c.c scan-tree-dump-times vect
"vectorized 1 loops" 0
* gcc.dg/vect/vect-reduc-pattern-2a.c scan-tree-dump-times vect
"vectorized 1 loops" 0
* gcc.dg/vect/vect-reduc-pattern-2b.c scan-tree-dump-times vect
"vectorized 1 loops" 0
* gcc.dg/vect/wrapv-vect-reduc-pattern-2c.c scan-tree-dump-times
vect "vectorized 1 loops" 0
Others:
* gcc.target/arm/neon-load-df0.c scan-assembler vmov.i32[
\t]+[dD][0-9]+, #0\n
* gcc.target/arm/synchronize.c scan-assembler __sync_synchronize
neon-load-df0 is a new test. synchronize.c is an incorrect test as
the compiler now correctly uses the dmb instruction.
Your thoughts?
-- Michael
I would like to announce that my work on armel cross toolchain got to the very
nice point - all packages are available from PPA.
What does it mean to you?
1. no "are you sure to install those unverified packages" messages from APT
2. ability to easily rebuild toolchain on own machines
So if you used my repository from people.canonical.com then please switch to
PPA one:
add-apt-repository ppa:hrw/armel-cross-compilers
Old repository will be available for some time but will not get any updates.
Next step: merging those packages into Maverick release.
Regards,
--
JID: hrw(a)jabber.org
Website: http://marcin.juszkiewicz.com.pl/
LinkedIn: http://www.linkedin.com/in/marcinjuszkiewicz
I've been checking over the benchmarks as a lead up to the 2010.09
release. We're in a good way compared to both 4.4.4 and 4.5.1 for
most non-trivial tests.
* pybench is 10.9 % faster than 4.4.4 and 7.7 % faster than 4.5.1.
* linpack is 46.4 % faster than 4.4.4 and the same as 4.5.1.
* ffmpeg h.264 video decode (with hand written assembler versions
turned off) is 15.4 % faster than 4.4.4 and 1.2 % faster than 4.5.1.
All results are statistically invalid and against poor workloads, but
I'll work on that.
See http://ex.seabright.co.nz/helpers/benchcompare for more.
-- Michael
Loïc Minier wrote:
> I see you moved the wiki page to the public space, thanks
>
> Couple of notes:
> * make sure you use the rename action on the page, I think this will
> preserver history (I didn't check whether you did or not, but I think
> not)
No, I didn't. I use copy and paste. I'll use rename action.
> * add a page at the old location with "#redirect NewPage" or
> "#refresh 0 http://newurl/"
OK, got it. Thanks for your help on wiki.
...are available here:
https://wiki.linaro.org/WorkingGroups/ToolChain/Meetings/2010-09-06
A copy and activity reports are included below.
-- Michael
Attendees
• Name Email IRC Nick
Andrew Stubbs ams(a)codesourcery.com ams_cs
Chung-Lin Tang cltang(a)codesourcery.com cltang
Julian Brown julian(a)codesourcery.com jbrown
Loïc Minier lool(a)linaro.org lool
Marcin Juszkiewicz marcin.juszkiewicz(a)linaro.org hrw
Matthias Klose doko(a)canonical.com doko
Michael Hope michael.hope(a)linaro.org michaelh
Peter Maydell peter.maydell(a)linaro.org pm215
Richard Earnshaw richard.earnshaw(a)arm.com rearnshaw
Ulrich Weigand ulrich.weigand(a)linaro.org uweigand
Yao Qi yao(a)codesourcery.com yao
Agenda
• Licensing of string routines
• State of valgrind
• State of GDB
• Open tickets
□ 600298, 616141, 604753: SMP/sync related
□ 605059 4.4.5
□ 629671 ICE in reload_cse_simplify_operands in thumb-1 mode
□ 590696 Wrong use of objdump during cross build
• Upcoming release
• Creating blueprints
Blueprint Assignee
Initial delivery of Linaro GCC 4.4 ams
Cross Compiler Packages hrw
Action Items from this Meeting
• ACTION: Richard to check with the legal department on string licensing
issues
• ACTION: Peter to talk with valgrind upstream re: Linaro releasing a
ARM-focused version
• ACTION: Michael to organise an 'experimental' PPA that toolchain output can
go into
• ACTION: Michael to talk with Cody Somerville re: building on ARM
• ACTION: Michael to set up a GDB 7.2 based off the release tarball
• ACTION: Andrew to pull sync changes back into 4.4 for this release
• ACTION: Michael to assign appropriate sync ticket to Andrew to track the
backport
• ACTION: Andrew to merge the current post 4.4.4 release branch into our 4.4
for this release
• ACTION: Julian to do a basic investigation into 629671
• ACTION: Andrew to merge the cross-compile objdump ticket into this release
and re-kick upstream process
Action Items from Previous Meeting
• ACTION: Michael to re-check with TSC that we can assign copyright but keep
ability to relicense
• DONE: Yao to continue on GDB for a week then switch to investigation
• ACTION: Peter to check into the state and progress of valgrind for the
meeting on the 30th.
• ACTION: Chung-Lin to shift the CSL backport list out onto the Linaro wiki
• ACTION: Michael to see about doing an archive rebuild with 4.5
• DONE: Michael to send IBM's list to Yao
Minutes
String routines:
• Michael asked Richard about getting the current str* routines by ARM
transferred to Linaro
• Linaro will then get these into other C libraries
• FSF prefers LGPL and copyright for glibc
• Linaro prefers MIT/X11 everywhere so that fixes and improvements can be
shared
• Richard is concerned about the copyright assignment and any patent grant
• ACTION: Richard to check with the legal department on string licensing
issues
• Extreme fallback is to re-write the routines to all be under Linaro
copyright. memcpy() and similar may need this
Valgrind:
• Peter has been looking at how it works on the ARM platform
• Upstream is very responsive to issues
• Now works on Firefox and OO.org
• Upstrem doesn't have any particular release cycle
• ARM changes are pretty extensive and can't be extracted
• Peter suggested making valgrind available in a PPA to start with
• NEON detection at startup is remaining issue
• What next?
□ Packaging is straight forward
□ Don't want to steal upstream's thunder or release something
inappropriate
□ ACTION: Peter to talk with valgrind upstream re: Linaro releasing a
ARM-focused version
• Could bring into the Linaro overlay PPA
• ACTION: Michael to organise an 'experimental' PPA that toolchain output can
go into
• ACTION: Michael to talk with Cody Somerville re: building on ARM
GDB:
• 7.2 is now available
• Time to start up a gdb-linaro based on that
• Matthias mentioned that we will have GDB 7.2 on Maverick
• How should we manage the source
□ QEMU is over git
□ Could use bzr or git
□ bzr with Launchpad can't handle multiple branches when pulling from git
□ GDB is unique in how it's mixed in with the rest of the projects hosted
on sourceare
□ Branches as such are trucky
□ Could just base off tarballs
□ ACTION: Michael to set up a GDB 7.2 based off the release tarball
Tickets:
• ACTION: Andrew to pull sync changes back into 4.4 for this release
• ACTION: Michael to assign appropriate sync ticket to Andrew to track the
backport
• ACTION: Andrew to merge the current post 4.4.4 release branch into our 4.4
for this release
• ACTION: Julian to do a basic investigation into 629671
• ACTION: Andrew to merge the cross-compile objdump ticket into this release
and re-kick upstream process
Patch tracker:
• Andrew noted that it is now fully populated with the GCC data
• Has assigned various patches that still need to go upstream to Yao and
Julian
Next meeting is on 2010-09-08 on the public code.
--- Chung-Lin Tang
== Linaro Toolchain ==
* Google ARM patch sets: committed a second set to SG++ 4.5 trunk on
Tues. AndrewS pushed both sets to Linaro. Worked on a third set, those
related to PR42235, but this time regression test results were not so
clean. Will look into, but considering whether to stop the backports
here.
* LP:628526, submitted a patch to gcc-patches for explicitly turning
off stack protection in libgcc build flags, awaiting response.
* LP:601030, eglibc 2.11/12 problem with ___longjmp_chk on x86-64.
Problem seems to be clear, fix quite simple, but so far cannot seem to
reproduce and verify. Also unclear if I should send the fix to eglibc
or glibc, the idea of the latter making me a bit nervous... :P
== libffi ==
* Got an acknowledgement from the libffi maintainer that he'll review
the VFP hard-float support patch soon.
== This week ==
* Look into remaining Google approved patches, mainly those related to
PR42235 and PR42575.
* Try to reproduce LP:601030 and send patch soon.
* Linaro GCC investigations.
--- Andrew Stubbs
== Linaro GCC ==
* Michael has get the new patch tracker into a usable state. I've
transferred all the data from the old wiki tracker, and looked up the
remaining data as far as I can. The new tracker should now be fully
populated with data. It's here, for the moment:
http://ex.seabright.co.nz/helpers/patchtrack
* Start Yao and Julian on the optimization investigation tasks.
* Continue trawl through the CS bugs looking for candidates to push
to the Linaro tracker.
== Other ==
* Public holiday on Monday.
* Attended the monthly CS/Linaro sync meeting.
--- Yao Qi
== Linaro GDB wrap up ==
* LP:615993 gdb.base/sigstep.exp failures
Patch was committed to gdb mainline and 7.2 branch.
* LP:615995 gdb.base/watch-vfork.exp failures
Discussed with Pedro, create a patch, which fixed failures on ARM,
but can't fix failures on x86(they are caused by different problems).
Leave the x86 failures there, and patch is being reviewed in
gdb-patches.
== Linaro GCC ==
* CS306:Investigate on thumb2 improvement
Read/understand previous effort related on code size
improvement from CSL wiki pages.
Experiment with CSL scripts for size benchmarking. With Dan's
help, run benchmarking in a correct/reasonable way.
'Reproduce' some inefficient code mentioned by Julian. Some of them
are still there.
== Misc ==
* LP:605042
Revert one patch, and rebuild it. No seg fault is found.
== This Week ==
* Continue my work on CS306.
--- Peter Maydell
RAG:
Red:
Amber: virtio-system writeup not going as fast as expected
Green: ARM legal OK now received
Milestones:
| Planned | Estimate | Actual |
finish virtio-system | 2010-08-27 | ? | |
I need to replan this (no forward progress this week
because more important stuff intervened)
Progress:
virtio-system:
- actually trying a SATA disk revealed that the PB926 PCI
interrupt mapping was wrong; now fixed after consulting
the schematics and a round or two of patch testing with Arnd
- I have a PB1176 board but it doesn't seem to talk to
the serial port on poweron. Will try a firmware reflash
but it might just be broken...
- no progress on writeup because other things intervened.
valgrind:
- went through the motions of getting a valgrind svn snapshot
into the ubuntu packaging
- tested on pegatron (A8, maverick, thumb2), found four bugs:
+ BX PC not implemented (fixed upstream)
https://bugs.kde.org/show_bug.cgi?id=249775
+ RBIT not implemented (fixed upstream)
https://bugs.kde.org/show_bug.cgi?id=249924
+ pwrite64 syscall not implemented (fixed upstream)
https://bugs.kde.org/show_bug.cgi?id=249996
+ test for presence of neon wrong
https://bugs.kde.org/show_bug.cgi?id=249775
With a bodge for the last and the fixes for the first 3,
valgrind now successfully runs openoffice and firefox.
other:
- Investigated https://bugs.launchpad.net/bugs/628471 : qemu-maemo
doesn't work with new linaro beagleboard kernels. It looks
like we now try to probe for NAND (which failed earlier
for other reasons which I suspect are a now-fixed bug),
and qemu-maemo's NAND implementation doesn't map anything at
the address the nand code is trying to poll for a status bit.
- first post to qemu-devel :-) (review of somebody's
patch to not confuse SMC with BKPT in the arm decoder)
Plans:
virtio-system:
- hoping to get the qemu patches into the ubuntu qemu-maemo
package, which will avoid the need to talk about patching qemu
- finish the writeup and put it on the wiki
- test PCI patches on PB1176
valgrind:
- respin a valgrind with proper fixes for everything and
put it in a PPA somewhere
other:
- come up with some fix or workaround for #628471
- put the rebased ubuntu qemu-maemo work up onto gitorious
so other people can see it
Absences:
Friday 5 November and 20 other days in this calendar year
Looks good. I've created a real project, added a README/LICENSE, and
merged your changes. See:
https://launchpad.net/tcwg-web
There was a funny render difference between Firefox and Chromium -
revisions with no bugs lead to a rowspan of zero which Firefox doesn't
like. I also pulled some common code out into a function and used the
built-in variable 'loop'. 'loop' is quite nice as it provides values
like .index, .first, .odd, and so on based on your position in the
loop.
-- Michael
On Fri, Sep 3, 2010 at 11:02 PM, Andrew Stubbs <ams(a)codesourcery.com> wrote:
> Hi Michael,
>
> I've been playing with you patch tracker, and come up with this:
>
> https://code.launchpad.net/~ams-codesourcery/+junk/tcwg-web
>
> I don't seem to be able to propose an official merge request to your branch,
> but it's just a quick implementation anyway, and could probably be cleaned
> up.
>
> The patch renders each ticket in it's own row (without changing the way the
> first two columns are rendered). This means they can have their own colour
> and we can maybe see better what status goes with what bug.
>
> To see an example of what it does, see revision 4.4:93544
>
> Andrew
>
I want to share status of my cross compiler packages work with all of you.
Some time ago I did a split of them into two:
- armel-cross-toolchain-base (1.36 now)
- armel-cross-toolchain (1.29 now)
Where first one provides binutils, linux headers, libc6 and libgcc
packages. Second provides final gcc.
Today I got a-c-t-base to a moment when it builds fine on PPA [1]. 1.36 got
sent for rebuild to fix missing gcc-4.5-arm-linux-gnueabi-base package. When
it will build then a-c-toolchain package will get uploaded for build.
Result will deprecate my current repository at people.canonical.com [2]
because PPA gives signed repository.
On Monday I will probably have to update both components because there was
gcc-4.4 upload so probably gcc-4.5 will follow (so I will be able to drop one
patch).
Additionally I made 'gcc-defaults-armel-cross' package (available in [1])
which makes installing of cross compilers a bit easier (no need to worry which
version to install - just "apt-get install gcc-arm-linux-gnueabi" is enough).
Selection of cross gcc version is done in other way then native one. Native is
using "gcc" package which contains /usr/bin/gcc as symlink to /usr/bin/gcc-4.4
file. Cross gcc uses "update-alternatives" to setup /usr/bin/arm-linux-
gnueabi-gcc file. I want to fix it in 11.04 so cross gcc will use same method
as native one.
1. https://edge.launchpad.net/~hrw/+archive/arm-cross-compiler
2. http://people.canonical.com/~hrw/ubuntu-maverick-armel-cross-compilers/
Regards,
--
JID: hrw(a)jabber.org
Website: http://marcin.juszkiewicz.com.pl/
LinkedIn: http://www.linkedin.com/in/marcinjuszkiewicz
