Hi folks,
Since we won't be using Pandas for buildbots any more, I decided to put
them in the lab for general use. This is not LAVA or CBuild managed, and
the purpose is to test, and possibly benchmark, small applications, build
systems, native compilation, generated code, etc, in real hardware. For
this purpose, the Pandas are great, so there you have it.
There is no reservation mechanisms, but the etiquette is to "who" and
"top", and if you're the only active user, go for it. Multiple users can
use a single board if only testing, compiling, running code, but if someone
is benchmarking, would be polite to use the other board.
They're sporting a Linaro 13.03 release with kernel 3.5.0. If you need/want
access to them, just let me know.
cheers,
--renato
== Progress ==
* Backports for 2013.09:
- updated list, assigned some to team members
- submitted several requests
- quite a few of them exposed dependencies, making it necessary to
make several attempts.
* Aarch64 frame layout: posted sample code using large frame on
gcc-patches to get feedback.
* Disable-peeling: resumed regression investigation.
* Wiki pages cleanup
* trunk validation:
- continued to work on internal validation of trunk using our
compute farm (build+cross validation) to help catch regressions early.
== Plan ==
* Finish backports, sort out dependencies problems.
* Disable-peeling: investigate regression
* More wiki pages cleanup
* trunk validation: improve reporting
* install Ubuntu on Chromebook
== Progress ==
* Libssp support for AArch64 TCWG 23:
My approach to Cross testing eglibc from host failed since it needs
the build directories to be shared. Mounting the image in build
system and building eglic will not help as the changes will not be
reflected, in the image booted in V8 model dyanamically. The test
system only issues command via ssh but does not copy binary to the
model.
Need to try creating NFS mount from build system into the model.
* Backports
Lots of back porting and merging.
201624 and 201666 can be back ported together. Merge request submitted.
201406 to linaro branch. Small test case comment change. But the test
case fails in linaro 4.8, but passes in trunk.
It needs other revisions to be back ported, since vectorization
improvement in trunk. I tried back porting all relevant versions
r199402,r200394,r200600, r201282. But still fails since interface of
function get_loop has changed in trunk somewhere.
For 201411 look like need to hand pick and apply changes in arm.md
file. since lot of changes gone in that file and simple merge will not
be good. Also it may need some other revision to backport.
* TCWG-209, Looked at wiki pages, Christophe assigned to me and gave
some comments about updating/deleting them.
== Plan ==
* Post RFC patch for Libssp support for AArch64 in GCC
* GCC patch for gprof accepted, need to post patch in glibc.
* Continue testing the patch for Libssp support for AArch64 in GCC
and glibc TCWG-23. Try NFS mouting into the model.
* Backport 201411 201406 from trunk.
== Issues ==
* LTO/PGO work stopped now since libssp support priority is more.
== Issues ==
* None
== Progress ==
* Conditional compare (CCMP)
- Design tens of test cases to verify the codes.
- Handle CCMP in vrp pass. If the compare in CCMP can be simplified,
CCMP will be converted to a copy, compare or a bit operator.
- Investigating to handle CCMP in jump-thread related optimization.
- Investigating bootstrap issue when expanding conditional compare
to optimized RTL.
. It seams that there are illegal combinations of different
compares since bootstrap OK if the two compares of CCMP have the same
CODE.
. Current check bases on function "arm_select_dominance_cc_mode".
Need narrow down the root cause.
== Plan ==
* Continue on conditional compare.
== Panned leaves ==
* Sept. 3-6: internal meeting.
== Progress ==
* spec2k comparison between ARM and x86
- Initiated discussion and addressing the feedback
- Looking at register allocation (IRA + Reload) for reduced test
case and comparing with x86
- Analysed few more spec2k benchmarks
* Back-porting
- All the back ports are checked and reviewed. Thanks Yvan.
== Plan ==
* Continue with spec2k comparison between ARM and x86
* Wiki Cleanup
== Progress ==
* More work on Jenkins/Cbuildv2 integration.
* Finished reviewing all 81 wiki pages I had to review.
* Added a 'dryrun' option to Cbuildv2 that displays all configure
and make command lines without executing them as a learning aid.
* Added 'make check' to toolchain builds via Jenkins/Cbuildv2.
* Added a simple "Hello World" test case in C++ to toolchain builds
in Cbuildv2 to make sure the fresh build minimally works.
* Added several more sections to the Cbuildv2 Usage wiki page.
* Cbuildv2 can now downloads and builds several projects/ (QT, GTK,
etc...) as a toolchain test.
* Got the LAVA team to make many tweaks to Jenkins and other stuff
to support TCWG needs for Cbuildv2. Now Jenkins does a toolchain
build of all supported architectures for each commit to GCC
upstream.
* Two small wildland fires this week, plus 3 motorcycle accidents.
== Plan ==
* More work on binary releases via Cbuildv2.
* Import test results into Jenkins.
* Continued work on Jenkins/Cbuildv2 integration.
== Issues ==
* Libmudflap, needs -ldl, or make check fails when linking with
--as-needed (used on Ubuntu by default).
* On chromebook using a raring chroot, 'make check' for mpfr has
these three test cases: tasinh, tcot, and tcsc which consume all the
memory and hang when run with the other infrastructure libraries via
Cbuildv2. The ttan and tsinh tests segfault, and tgamma return an
incorrect result. These same test cases all pass if I use Cbuildv2
to only build and test mpfr. Weird...
== Progress ==
* GDB Record and Replay (TCWG-197)
Wrote instruction type decoder code for ARM Thumb32 instructions.
Wrote stub handlers for different type of Thumb32 instructions.
Wrote process record handlers for various type of load/store instruction.
Wrote a testing utility to test record and replay instruction decoding code.
* Visa Application
Received email notifcation for UK visa issuance for upcoming trip to Cambridge.
Filled out US visa application form DS-160, gathering information
required to complete my application.
== Plan ==
* GDB Record and Replay
Continue to write code for decoding and recording thumb32 instructions.
Try to find a way to generate test data of various instructions to
test record and replay code.
* Travel and Visas
Go to UK visa application center to collect my passport back.
Apply for US visa and make an appointment for the interview for the
week of 23rd September.
Prepare for UK travel for the week of 14th September.
== Progress ==
* Investigating Cross-Compilation with LLVM
- Gathering all options necessary to cross-compile
- Cross-compiling Clang+LLVM with Clang to ARM
- Writing down all issues found during the process
* Investigating the Clang Driver
- Refactoring some common decisions about targets
- Adding new CPUs, cleaning a few areas
== Plan ==
* Try to upstream some patches to make cross-compilation a bit easier
* Write a document on how to cross-compile with Clang on LLVM website
== Progress ==
* 4 day week (Bank Holiday Monday)
* Wiki cleanup pretty much done
* More malloc implementation work
* Further investigation of ptmalloc to see if it can be improved
* Fixed gdb build for non-Intel architectures
* Found an issue with clone in glibc 2.18 and committed a fix
* Committed strlen code to glibc
== Issues ==
* None
== Plan ==
* Get glibc patches applied upstream
* More malloc
* Look into stack guard and pointer guard stuff for ARM
--
Will Newton
Toolchain Working Group, Linaro
