About three months ago, 4.7 stopped being able to optimise things like:
int *__restrict x = ...;
The (libav) loop microbenchmarks that I'd written used this construct
a lot, as an easy way of automatically generating a whole function
from a loop kernel.
I spent a while testing 4.7 with the restrict patch reverted, while
I caught up with my post-holiday email backlog and saw whether the
effect on this code was deliberate. I eventually realised it was,
so implemented a change that Ira had suggested: splitting out a
peak_loop_1 that takes all the restrict pointers as arguments.
I just realised that I never pushed that change back up to bzr,
so I've done it now.
Probably a write-only change, since I doubt anyone's going to be
using the benchmark again, but just in case :-)
Richard
This is my current 4.7 auto-inc-dec.c patch. I submitted an RFC in July:
http://article.gmane.org/gmane.comp.gcc.patches/241779/
and updated the patch in line with the feedback I got. Steven Bosscher
sent some very useful comments in private email, so the update deals
with those as well as Bernd's public ones.
If we do go ahead with this rewrite, it depends on the A9 pipeline
description changes. I submitted some A8 and A9 changes here:
http://article.gmane.org/gmane.comp.gcc.patches/244238/http://article.gmane.org/gmane.comp.gcc.patches/244242/
but because I later noticed that the A9 didn't behave quite as I thought,
I decided not to apply them. Ramana asked around internally about what
the A9 actually does (thanks) and had some ideas.
The patch also relies on the MEM rtx_costs patch that I just posted:
http://lists.linaro.org/pipermail/linaro-toolchain/2011-December/001944.html
Richard
gcc/
* Makefile.in (auto-inc-dec.o): Depends on $(OPTABS_H) and
addresses.h.
* auto-inc-dec.c: Rewrite.
Index: gcc/Makefile.in
===================================================================
--- gcc/Makefile.in 2011-12-07 11:43:29.549238252 +0000
+++ gcc/Makefile.in 2011-12-29 09:24:51.066303201 +0000
@@ -3145,7 +3145,8 @@ alloc-pool.o : alloc-pool.c $(CONFIG_H)
auto-inc-dec.o : auto-inc-dec.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) \
$(TREE_H) $(RTL_H) $(TM_P_H) hard-reg-set.h $(BASIC_BLOCK_H) insn-config.h \
$(REGS_H) $(FLAGS_H) output.h $(FUNCTION_H) $(EXCEPT_H) $(DIAGNOSTIC_CORE_H) $(RECOG_H) \
- $(EXPR_H) $(TIMEVAR_H) $(TREE_PASS_H) $(DF_H) $(DBGCNT_H) $(TARGET_H)
+ $(EXPR_H) $(TIMEVAR_H) $(TREE_PASS_H) $(DF_H) $(DBGCNT_H) $(TARGET_H) \
+ $(OPTABS_H) addresses.h
cfg.o : cfg.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(RTL_H) $(FLAGS_H) \
$(REGS_H) hard-reg-set.h output.h $(DIAGNOSTIC_CORE_H) $(FUNCTION_H) $(EXCEPT_H) $(GGC_H) \
$(TM_P_H) $(TIMEVAR_H) $(OBSTACK_H) $(TREE_H) alloc-pool.h \
Hi,
Thank you all for an interesting and pleasant experience. I am very
grateful to Linaro for the opportunity to meet and work with such an
amazing group of people. I wish you all the best, and hope to meet you
again (at least online).
You can find me at irar(a)il.ibm.com or ira.rsn(a)gmail.com.
Ira
Summary:
* Read armV7-A/R reference manual; crosstool-ng patches and wrapper scripts.
Details:
1. Patches for crosstool-ng:
* Fix symlink issue when CT_USE_SYSROOT is not enabled.
* Update sample/linaro-arm-none-eabi (baremetal) to disable
SYSROOT. So that both include and lib files are in the same dir.
2. Study armV7-A/R reference manual.
3. Validate embedded toolchain Dec. release.
4. Enhance the wrapper to use crosstool-ng for embedded toolchain.
Plan:
* Ramp-up on gcc.
Best regards!
-Zhenqiang
Submitting patch for Bug #879725:
http://gcc.gnu.org/ml/gcc-patches/2011-12/msg01459.html
Looking at the performance results running SMS with automatic testing.
This is my last week in Linaro so I would also like to thank you all
for the interesting year -- it was a great experience for me to work
in this project. I wish you all good luck and happy holidays!
Revital
Hi,
OpenEmbedded-Core:
* No response on the CSL patches I posted to the ml yet
* khem says someone (other than me) needs to try them
* Linaro binary toolchain
* Runs on Oneiric-X86_64 after installing lsb-core
(interpreter: /lib/ld-lsb.so.3)
* The do_rootfs tasks fails with runtine dependecy issues when
using the external-linaro-toolchain_arm-2011.11.bb recipe.
When re-using my CSL 2011.03 recipe with the linaro toolchain
the error doesn't show up - strange.
* OE-Core build gets confused by the (arm-linux-gnueabi-)pkg-config
of the external linaro toolchain. As a workaround I just renamed
this script.
* The qemuarm MACHINE configuration uses "-march=armv5te -mno-thumb"
Since the linaro toolchain defaults to thumb and -mno-thumb has no
effect some inline assemblies are failing (i.e. on the umull insn).
GNU #47930 suggests using -marm instead -> OE-Core patch posted.
* Got the core-image-minimal to build, but it doesn't run yet
(I suspect some basic runtime dependencies like libc again)
* The build of the sato image fails
(seems libtool and/or C++ related - need to investigate)
Regards
Ken
Hi,
* Continued with comparison of eembc results for gcc4.4 and gcc 4.6 (FSF
and Linaro). Collecting results for 4.6 with loop-unrolling turned off.
* Working on a plotbench.py script that will use matplotlib for plotting
the results. Right now the script plots the geomean value, for instance for
eembc. I now try to make it plot all subtest as well. Then it should also
show relative improvements instead of just the numbers, and then also
sorted from best to worse. This script depends on Michaels script
libtabulate.py for transforming the tabulated file back to python records.
* Will be back January 9
/Regards
Åsa
== GDB ==
* Ongoing work on remote support for "info proc" and core file
generation. Completed implementation of latest solution
via accessing arbitrary files on the remote site, only to
run into a fundamental design problem ... so it's probably
back to the previous approach. Discussion on the list is
ongoing.
* Fixed a GDB 7.4 test suite failure on ARM: PR tdep/12797
* Fixed another GBD 7.4 test suite failure on ARM, by enabling
pthread_t thread debugging on core files.
== GCC ==
* Patch review week.
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 there. I've looked further into the intermittent
gcc/testsuite/g++.dg/cdce3.C test failures. Taking Ira's
vectoriser-only fix-pr51301-4.6 branch and comparing it with it's
predecessor r106845:
* cdce3.o itself is identical across compilers
* Fault occurs in a parallel test run as part of the normal auto build
* Fault occurs every time
* Fault occurs with a manual 'make check-gcc RUNTESTFLAGS="dg.exp=cdce*'
* Fault doesn't occur when building from the command line
* Fault doesn't occur after updating binutils
I'm suspicious of the linker. The auto builders are Natty based and
come with ld 2.21.0.20110327. Updating them to Oneiric's
2.21.53.20110810 clears the problem.
I've saved the build trees. I see no reason not to commit
~ramana/gcc-linaro/fix-lp-900426 and ~irar/gcc-linaro/fix-pr51301-4.6.
-- Michael
== GDB ==
* Ongoing work on remote support for "info proc" and core file
generation. Implemented initial version of latest solution
via accessing arbitrary files on the remote site.
== GCC ==
* Started familiarizing myself with current status of various
performance patches in programm, in preparation of my taking
on GCC performance work next year.
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
Summary
* make check-gcc on windows.
* crosstool-ng patches
Details:
1. Two patches for crosstool-ng:
* Fix the compile error when CT_USE_SYSROOT is not "y". With this
fix, we can config crosstool-ng to remove the symbol link for windows
build.
* Add scripts to build manual for newlib.
2. make check-gcc on Windows:
* Wrap gcc/g++ for windows test. testglue.c should be compiled with
gcc not g++.
* Enhance scripts to convert path using "cygpath -w"
3. Analyze and root cause the pseudo new failed cases on windows.
* gcc fail cases (gcc.dg/cpp/assert3.c, gcc.dg/cpp/include7.c and
gcc.dg/cpp/trad/assert3.c)
Root cause: " in options are removed in the test scripts. e.g.
When reading gcc.log, you can find “-Aabc = jkl” in "Executing
on host" as:
Executing on host: …/cpp/assert3.c -A abc=def -A abc(ghi)
"-Aabc = jkl" …
But in spawn: the “” are removed.
spawn …/cpp/assert3.c -A abc=def -A abc(ghi) -Aabc = jkl
* g++ fail cases (dwarf2/lineno-simple1.C, dwarf2/pr44641.C and
dwarf2/pr46527.C)
The assembler codes generated from windows g++ and linux g++ are
same except the PATH string. And all PASS on linux test.
It seams the scripts can not grep the expected string on windows.
* Tests on windows are not stable. For each test, there will have
random fail cases (pass when retesting separately).
Plans:
* Create Makefile for embedded toolchain in linaro crosstool-ng.
Best regards!
-Zhenqiang
Hi,
I was just wondering if anyone knows of any current or future dependencies with the Linaro toolchain (2011.11) and the Linaro release of GDB and the Linux Kernel?
Is it considered safe to use the toolchain with the upstream releases of GDB and the Kernel, assuming that the versions of each are suitably compatible?
Or are there potential dependencies on work that has been done in the toolchain? For example, new instructions supported in the compiler/assembler, or enhancements to the kernel/runtime library on the Linaro branch that would depend on them being in sync.
Thanks,
Dave.
Hi,
OpenEmbedded-Core:
* the CSL 2011.03 recipe works with localization support disabled
* got the OE-Core sato image to built (~250 source packages)
* also built the Qt4 demo image (~100 source packages) to stress the
C++ part of the toolchain
* both are booting using qemu and seem to work just fine.
* all of the Linaro members approved the request to contribute to
OpenEmbedded
* started to post patches onto the mailing list
* briefly looked at the Linaro binary toolchain
* the most recent one is dynamically linked
* while the old one has old binutils (.21) that causes issue with
--gc-sections
* Currently the build is using the OE qemuarm machine configuration that
uses a Yocto kernel and targets armv5te. This is something I'd like to
look at too.
Regards
Ken
Continued work on 64-bit shifts.
- Improved 64-bit shifts without NEON (should benefit all cases).
- Fixed bugs in constant shift code.
- Rewrote 64-bit neon patch to take advantage of the new non-neon
code, in the fall-back case.
- Titied up the code, in general.
- Rewrote SImode shift amount patch for new neon patches.
The code produced now seems pretty good, but it still seems to choose
which mode to use slightly haphazardly. The next step is to figure that
out and benchmark the results.
Had a few more attempts at getting the LAVA system to do something
useful for me. I'm getting closer, but keep hitting new problems. Some
of them my fault, and some are bugs in the system. Paul Larson has been
very kindly helping me out and swatting the bugs.
Didn't get much further with benchmarking for the generic tuning. This
has been put on the back-burner while I work on the Neon shifts, and my
test runs on my IGEPv2 A8 board have all been interrupted by power cuts
or rendered useless by my forgetting to kill background tasks (such as
Xorg).
---- Next weeks
On vacation December 19th - January 3rd (returning January 4th).
== General ==
* Tidying things up and updating my list of statuses
== String routines ==
* Adding strchr and strlen to eglibc; tests running at the moment.
Dave
[short week, three days]
RAG:
Red:
Amber:
Green:
Current Milestones:
|| || Planned || Estimate || Actual ||
||upstream-omap3-cleanup || 2011-11-10 || 2011-12-15 || 2011-12-12 ||
||cp15-rework || 2012-01-06 || 2012-01-17 || ||
||initial-a15-system-model || 2012-01-27 || 2012-01-27 || ||
||qemu-kvm-getting-started || 2012-03-04?|| 2012-03-04?|| ||
(for blueprint definitions: https://wiki.linaro.org/PeterMaydell/QemuKVM)
Historical Milestones:
||add-omap3-networking || 2011-10-13 || 2011-10-13 || 2011-10-13 ||
||a15-systemmode-planning || 2011-10-13 || 2011-10-13 || 2011-09-22 ||
||a15-usermode-support || 2011-11-10 || 2011-11-10 || 2011-10-27 ||
== cp15-rework ==
* estimate pushed back a bit because I've ended up doing this in
parallel with the other blueprints. Also exynos4210 review has
taken some time.
== upstream-omap3-cleanup ==
* split up the last handful of patches in the stack which were doing
several things at once
* this blueprint is now complete, meaning that the next stage of omap3
upstreaming will be cleaning up individual subsets of functionality
to send upstream. This is all backburner level priority, though.
== other ==
* reviewed most of Samsung's exynos4210 model
* completed a conflict-heavy rebase of qemu-linaro (the result of
MemoryRegion conversions for omap devices landing upstream)
* LP:903239 : added linux-user support for some missing xattr syscalls
that were causing build problems for apparmor
-- PMM
Hi,
After learning how to control MEM_ALIGN and, therefore, alignment
hints from the vectorizer, I was able to generate 64-bit hints (with
the help of Ramana's patches). I saw a 16% improvement on a benchmark
with stack variables, for which we now force alignment to 64 bits and
create alignment hints, instead of using peeling.
Ira
Hi,
* Finished an across-compilers report for benchmarks over the latest in FSF
and Linaro series. Will start storing results in the
linaro-toolchain-benchmarks bzr repository.
* Looking closer at eembc results, especially regressions between gcc-4.4
and gcc-4.6. Did runs with gcc-linaro-4.4 with -fno-unroll-loop. Will
continue analyze and try to present the result in a good way.
* Reviewed Michael's geomean implementation.
* I will be on Christmas holiday w52 and w01, will be back 9/1.
/Regards
Åsa
Hi there. Could the toolchain team please have a look through the
current GCC blueprints and update them? You can see a list and states
at:
http://apus.seabright.co.nz/helpers/backlog
and for gcc-linaro only at:
http://apus.seabright.co.nz/helpers/backlog/project/gcc-linaro
Please check for any that:
* are on your short-term todo list but aren't against your name
* have been started but are stuck in the backlog or todo
* are finished but not marked as such
* are blocked
* are duplicates or too undefined
* or are obsolete
I'm especially interested in:
* "slp-supported-ops"
* "sms-register-scheduling"
* "better-block-operations"
* "libraries-for-backlog"
* "backport-conditional-execution"
* "improve-peeling"
* "64-bit-sync-pimitives"
* "neon-strided-load-extract"
If you've finished a significant amount of work on one blueprint then
let me know. We can split that work out and push the rest back into
the backlog.
Also, let me know if you're blocked on final benchmarking. We can now
easily benchmark a merge request and see the difference.
-- Michael
Continued work on 64-bit neon operations. The negdi2 seems to be more
difficult than previously thought - vneg won't do it, and there's no way
to encode either "0-reg" or "not(reg)+1", so I'm shelving that idea for
the moment, and moving on to one_compldi2_neon, which ought to be
straight forward.
Did the entire Linaro GCC release process, in the absence of Michael
Hope, from source to announcement. The process didn't go as smoothly as
I'd have liked, but I got through it, mostly. Hopefully Michael won't be
travelling next time ...
Tried to figure out how to do 64-bit shifts using a QImode shift amount.
This promised to eliminate the unnecessary zero-extends, but it doesn't
work because neither iwmmxt or neon registers are permitted to hold
QImode values (presumably changing this would have consequences
elsewhere?). Annoyingly, it's also not possible to put SImode values in
(most) neon registers, so I'm not sure quite how to optimize the values.
More investigation required.
Hi!
This week was spent doing internal ST-E work, but related to the Linaro
tcwg so I will give a short summary anyway.
I have taken the Linaro toolchain (prebuilt by the Android working group)
and used it in our internal Android build.
There were several build errors, as expected when going from gcc-4.4.3,
which is the default compiler in Android (Gingerbread) to gcc-4.6.2. Many
errors were solved with patches from the Linaro Android distribution.
Did some benchmarking related to web browsing:
ARMBBench (load and rendering of web pages) - gave me 4-6% improvement with
the Linaro toolchain.
Sunspider and BroserMark (JavaScript) - gave me ~6% overall regression with
the Lianaro toolchain. However, when zooming in to individual test cases -
SunSpider consist of ~25 tests in 9 categories - the results are really
scattered. A few tests are mainly contributing to the regression. I try to
narrow things down to understand which code parts in v8 (the JavaScript
engine) that causes the slowdown.
Best regards
Åsa
Continue working on the patch to estimate register pressure on SMS:
Addressing the comments received from Richard and Ayal.
Testing the patch on libav micro benchmarks.