Testing an initial version of the implementation which estimates
register pressure in SMS on libav micro benchmarks.
I see 20% improvements in mjpegenc microbench and 11% on aacsbr-2 with
SMS. However swscale-rgb24ToY_c
still have spills in the final code although it requires maximum 64
VFP_REGS registers out of the available 64 registers so I'm trying to
understand the reason for the spill.
==Progress===
* Off for one day during the week for Diwali.
* Connect preparation - Wrote down areas to look at during connect and
tried to plan what we
want to look at during connect.
* Looked at some of the cases with vcond<float> with Ira and helped
frame blueprint.
* Investigated one of the big performance regressions in the popular
embedded benchmark
and looked at why it wasn't being vectorized only to realize that it
couldn't be. Thanks
Ira. Still don't know why ARM state is 22% faster than Thumb2 state.
* Looked at the issue with fPIC where GCSE appears to remove a label
for sometime
but not much progress.
=== Plans ===
* Connect ! next week and then vacation.
Absences.
* 26 Oct - Diwali
* 31st Oct - 4th Nov - Linaro Summit Orlando - Travel booked -
* 08 Nov - 11 Nov - Vacation booked
* Dec 19 - 31st Dec - Vacation booked
== 64 bit atomics ==
* I've been building and testing membase
* Version 1.7.1.1 source builds OK (after turning off -Werror due to
some of their curious type naming)
* The git version fails to build - it doesn't seem consistent
* 1.7.1.1 passes simple tests, but there are 3 tests in its test
suite that intermittently fail on ARM and
seem to be solid on x86. (There are also some that just require
timeouts increased due to the
relatively slow machine).
* t/issue_163.t turned out to be a timing race in the test itself,
made worse by being on a relatively slow
machine and probably made worse by the Pandas odd idea of timing.
That was reported to them with
a break down of it, and upstream has fixed their test. (
http://code.google.com/p/memcached/issues/detail?id=230 )
* t/issue_67.t is proving tougher; once in a while memcached will
lock up during init in thread_init;
there is one particular point where adding a printf will make it work
apparently reliably. I've got one
or two ideas but I need to check my understanding of pthread_cond_wait first.
* There is an assert I've seen triggered once - not looked at that yet.
== String routines ==
* While I was off last week, my memchr and strlen were accepted into newlib
* Joseph has responded to my eglibc mail, with a couple of small queries.
== Other ==
* Wrote a more detailed test case for bug 873453 (odd timing
behaviour on panda); it's
quite odd - I can get > ~80ms timing discrepency so it's not a clock
granularity issue.
* Replicated a QEMU crash for Peter.
Dave
Hi,
* finished changing libunwind to be more portable
* tested patchset on ARM and X86_64
* now builds on Android without modifications
(Android.mk, config.h and libunwind-common.h are still required)
* verified that the modified debuggerd still works
* discussed backtracing using libunwind on ARM with Harald from the BSC
* they use libunwind in a sampling tool that generates Paraver
tracefiles
* started to upgrade my Linaro Android environment and ran into issues
* need to check:
* why building the toolchain using linaro-build.sh fails
* why repo sync fails due to invalid platform/bionic SHA1
* what happened to LEB-panda.xml
Regards
Ken
RAG:
Red:
Amber:
Green:
Current Milestones:
|| || Planned || Estimate || Actual ||
||a15-usermode-support || 2011-11-10 || 2011-11-10 || 2011-10-27 ||
||upstream-omap3-cleanup || 2011-11-10 || 2011-11-10 || ||
Historical Milestones:
||qemu-linaro-2011-07 || 2011-07-21 || 2011-07-21 || 2011-07-21 ||
||qemu-linaro 2011-08 || 2011-08-18 || 2011-08-18 || 2011-08-18 ||
||qemu-linaro 2011-09 || 2011-09-15 || 2011-09-15 || 2011-09-15 ||
||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 ==
* A15 instruction support patches committed upstream in time for
upstream's 1.0 release
== upstream-omap3-cleanup ==
* some work on restructuring the omap3 patchset -- it's now basically
in the right order and the last 'touches several different
bits of code' jumbo patch has been split
== other ==
* sent some patches upstream which address the main things I
want to get into qemu 1.0 (PL041 audio support and fixing a
regression in handling multithreaded programs in linux-user mode)
* A15 KVM planning work and other preparation for Linaro Connect
* finally tracked down the qemu-on-ARM memory corruption: we
mmap the code generation buffer at 0x1000000 with MAP_FIXED;
unfortunately this is now in the middle of glibc's heap...
(filed as LP:883133)
* qemu now has a coroutine implementation which defaults to using
makecontext() if it is present. Unfortunately ARM eglibc provides
an implementation which always returns ENOSYS, which is a bit
tricky to detect with a compile time configure check (without
breaking cross-compilation support).
* these two things (and some other known bugs) mean that QEMU on
ARM hosts is basically broken, and will probably continue to be
since we don't have the spare resource to test and fix bugs
(beyond those which we need to fix for KVM-on-ARM)
* Looked at how to configure Firefox and how to build different parts of the
program. Usage of .mozconfig, myrules.mk and myconfig.mk.
* Tested the Talos framework. https://wiki.mozilla.org/Buildbot/Talos. I
think it would be good to use Talos for the browsing benchmarks. We can
discuss it further at connect.
* Preparing for connect.
Best Regards
Åsa
Summary:
* Exercise crosstool-ng and summarize the gaps.
Details:
* Exercise crosstool-ng
(1) Sync with lp:~linaro-toolchain-dev/crosstool-ng/linaro.
(2) Try to config linux-host-baremental-target an
mingw32-host-baremental-target.
(3) Try to build the toolchain for both embedded toolchain and
linaro-gcc-4.6-2011.10 with the config.
. C compiler for linux and mingw32 hosts and c++ compiler for
linux host can be built without any change.
. C++ compiler for mingw32 host can be built after PCH is disabled.
. GDB-cross build fail due to dependence packages.
* Gaps in crosstool-ng
(1) Improve GDB-cross scripts to download and build the dependence
packages: expat and ncurses. Or put expat and ncurses as
companion_libraries.
(2) To remove dependence, embedded toolchain requires more
prerequisites like zlib.
New config and scripts are required to support the packages.
(3) Currently, the embedded toolchain source packages are released
as a tarball, which includes gcc, gmp, etc. New scripts are required
to support it.
(4) To make sure the toolchain can run with lower version glibc like
redhat4/5, the embedded toolchain requires lower version native
gcc4.3.6 to build it.
To support it,
. Users can build the native gcc manually, or
. Enhance the scripts to add one step to build native gcc.
(5) All the default package configurations are different from
embedded toolchain internal build scripts.
Since the configurations in embedded toolchain had been tuned
and tested, we will change the configurations in crosstool-ng if they
do not match and not configurable.
The same rule will apply for linaro toolchain.
Plans:
* Write scripts to re-pack the embedded toolchain source packages.
* Add the supports for all prerequisites in crosstool-ng menuconfig.
Thanks!
-Zhenqiang
Posted a patch upstream to fix big-endian for generic tuning. This was a
simple omission from my previous patches.
Merged GCC 4.6.2 to Linaro GCC. It's still in testing now, so I'll have
to commit it sometime over the weekend or next week.
Looked at the benchmark results from Spec2000 running on both A8 and A9
systems, with and with NEON, and with various compiler options. Posted
the results in a spreadsheet (visible within Linaro only).
Begun making adjustments to generic tuning and started new spec2k runs
to see if they are beneficial. First, I'm trying A9 prefetch settings on
A8 to see how much damage it does. Next I'll try enabling the A8 NEON
tuning settings on A9 to see what happens there.
Prepared for travel next week.
Vacation Friday
Hi,
- Merged to gcc-linaro:
- widening shifts
- SLP features: support loads with different offsets and swap
operands if necessary
- Started rewriting SLP analysis to support operations with more than
two operands (towards SLP of conditions)
- Updated NEON presentation following Ramana's suggestions (thanks!)
- Suggested to Ramana to implement vcond with mixed types, created a
blueprint: https://blueprints.launchpad.net/gcc-linaro/+spec/vcond-with-mixed-types
- Vectorizer:
- updated vectorizer's webpage
- updated vectorizer's wiki page
- the usual maintenance
- Committed upstream two SLP data-ref analysis improvements: PR 50730
and PR 50819
Ira
