== Week of March 24th ==
- STREAM regression (TCWG-388, 5/10)
-- Finished prototype patch. The patch adds modeling of ARM L2 auto-prefetcher hardware to GCC scheduler (the model is very simple as auto-prefetcher is very lightly documented). Half of the patch cleans up and improves GCC scheduler, and the other half implements the auto-prefetcher model.
-- While looking into ARM scheduling support noticed that ARM doesn't use multipass lookahead scheduling, which surprised me. Enabled it (multipass scheduling) in my patches.
- Looked into lll_timed_wait Glibc/uClibc bug upstream (1/10)
-- https://sourceware.org/ml/libc-alpha/2014-03/msg00905.html
- Various discussions and reviews (4/10)
== Week of March 30th ==
- STREAM regression (TCWG-388)
-- Benchmark patches on SPEC2k and find/confirm best values for tuning parameters:
--- dfa_lookahead: should normally be issue_rate-1.
--- L2 auto-prefetcher queue depth: new tuning knob.
-- Investigate any performance regressions from the patches.
- lll_timed_wait Glibc/uClibc bug
-- Make sure it is fixed upstream. Possibly backport to Linaro branches.
--
Maxim Kuvyrkov
www.linaro.org
== Issues ==
* none
== Progress ==
* Launchpad bugs:
o TCWG-422 : ICE in assign_by_spills building linux btrfs module (1/10)
- created blueprint for :
https://bugs.launchpad.net/gcc-linaro/+bug/1296676
- Reported upstream as :
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60650
- Reduced testcase
- Fix committed by Vladimir as rev208876
- still ICE when configured for arm-none-linux-gnueabihf
o Backported "Internal compiler error in push_reload during
bootstrap stage 2" to GCC 4.7 (1/10)
- https://bugs.launchpad.net/gcc-linaro/+bug/1129013
- some testsuite regressions observed. will investigate.
* Backports review: (5/10)
o reviewed backport for pr60264 and rev202663
o cortex-a53 support backport:
- Analysed testsuite regression
- 22K Loc patch under review
* LRA on AArch32:
o TCWG-345 : Analyse performance of LRA for ARM. (1/10)
- looked at the perf tool results
* Misc:
o Various meetings (1/10).
o Various support to team members (1/10)
o Cbuildv1 baby-sitting (Calxedas nodes have to be restarted after
each upgrades !)
== Next ==
- continue cortex-a53 review
- continue on backports.
- continue on TCWG-345.
== Progress ==
* Short Week
-- Monday Day Off: Pakistan Day 23rd March public holiday roll over. [2/10]
-- Short leave on Thursday [1/10]
* Work on gdb testing utility [TCWG-96] [7/10]
-- Writing a new gdb testing utility script that automates gdb
testing in various configurations and compares testsuite results.
-- Support for all kind of native, native-gdbserver and
remote-gdbserver gdb configurations has been added.
-- Interactive and configuration file based user input mode has been added.
-- Testing configurations host alive, ssh possible, board file found
and ability to build sources using user defined configure flags has
been added.
== Plan ==
* Work on gdb testing utility [TCWG-96].
-- Add support to run native and native-gdbserver test via ssh on
remote machines like arm-linux.
-- Bug fix and test gdb testing utility by running it various configurations.
-- Update wiki page with testing utility and how to use it.
== This week ==
a53 support
- Fixed regression found in arm testing and resubmitted build and
merge requests
- arrch64 testing passed with no regressions. Testing with a53
support enabled still required
Merged 202663 (vectorizer bug) into 4.7 and 4.8 branches. Submitted
merge and build requests
== Next week ==
- Test aarch64 with a53 support enabled on qemu64
- Work on bug 60657 - [4.9 Regresssion] ICE: error: insn does not
satisfy its constraints
== Future ==
== Issues ==
* None
== Progress ==
* Create prebuilt sysroot based on Linaro eglibc 2014.04 release
(https://launchpad.net/linaro-toolchain-binaries/support/01/+download/linaro…)
(1/10).
* Enable shrink-wrap for apcs. Patch was out for community review. (1/10)
* Reinvestigate shrink-wrap enhancement (4/10, TCWG-133)
- There was improvement in ira to split_live_ranges_for_shrink_wrap
(http://gcc.gnu.org/bugzilla/show_bug.cgi?id=10474). But it still can
not handle the case in 453.povray.
- Investigate to do a simple copy-forward when prepare_shrink_wrap.
* Investigate lp:1296942 (pr60663). Patch is sent out for community
review. (4/10)
* Backporting the fix for pr60264 to Linaro 4..8.
== Plans ==
* Continue on shrink-wrap enhancement.
== Planned leaves ==
* April 7: Qingming holiday.
== Progress ==
- TCWG-413 Spec2006 (7/10)
- Investigated compiler error for 481.wrf with FSF 4.8.2. Issue is
due to aarch64_cm<cmp><mode> pattern (fcmle and fcmlt supports only #0
as third val). This is already fixed in trunk and Linaro 4.8.
- Ran profiling to analyse 4.9 regressions. Started looking into
P7Vitterbui which is one of the functions that performs badly.
- TCWG-291 CRC (3/10)
Came up with a patch for improving vrp for test-case. Some c++ test
cases are failing in regression testing with this patch. Looking into it.
- TCWG-394 / PR60034
Patch committed and card closed.
http://gcc.gnu.org/viewcvs?rev=208949&root=gcc&view=rev
== Plan ==
Continue with Spec2006 and crc
== Progress ==
* Android (no card, 10 minutes. ;)
- Implemented __builtin___clear_cache in Clang
* Kernel (TCWG-417) 6/10
- VLAIS in crypto (last place in kernel)
- __builtin__stack_pointer (new builtin)
- Discussion in GCC list, named registers might be a better option
- Discussion in LLVM list about implementing named registers
- Implementing named register variables...
- Planning on LAVA testing LLVM kernels with LLVMLinux
- __aeabi_memset/cpy/move (both Android and Kernel) will have to be fixed
* Libraries (TCWG-125) 2/10
- libc++abi's unwind routines assume (ARM == SjLj)
- libc++ doesn't work without them
- We'll need to teach it about EHABI (later)
* Background 2/10
- Installed ArchLinux on my laptop, took some time to setup
- Re-org of LLVM Jira Cards (TCWG-417)
- Reviewing some GSoC proposals
== Plan ==
* implement named register variables in LLVM, then Clang
* Continue helping LLVMLinux and Milosz to test the LLVM kernel
* Time allowing, check CBuild2 for an LLVM build
* Two long weeks of holidays...
== Progress ==
* Bugfix aarch64 setcontext patch (2/10, TCWG-410)
* malloc requirements wiki page (2/10, TCWG-414)
* Lots of creating and updating and updating JIRA cards (1/10)
* glibc patch review (1/10)
* Assorted small patches - ld relasz, gnulib obstack, glibc strtod
benchmark (2/10)
* Investigate a couple of issues raised by member services and on lists (2/10)
== Issues ==
* None
== Plan ==
* Resurrect glibc malloc benchtest
* More glibc benchmark infrastructure work
* Check status of glibc ARM port build warnings
--
Will Newton
Toolchain Working Group, Linaro
== Week of March 17th ==
- STREAM regression (TCWG-388, 4/10)
-- Investigated how to prioritize memory references instructions in GCC scheduler to take full advantage of L2 autopretch hardware in certain ARM cores.
-- Fixed -fdbg-cnt=sched_insn debug counter along the way. It appear to have been broken since GCC 4.7.
- Discussed reg_pressure instruction scheduling with Charlie. (TCWG-135, 1/10)
- Various discussions about instruction scheduling in GCC. (1/10)
- Together with Michael prepared patch list for GCC contingency plan. (3/10)
- Made first-in-series video about tips-and-trick of GCC development. Your critiques are welcome!
-- Using GCC debug counters (7m34s): https://www.youtube.com/watch?v=IWRYCOkgL04
== Week of March 24th ==
- STREAM regression (TCWG-388)
-- Get a prototype patch.
- Other expected and unexpected tasks that come up.
--
Maxim Kuvyrkov
www.linaro.org
Last week
* one day off [2/10]
* NEON scheduling investigation - TCWG-135 [3/10]
. investigated scheduler register pressure heuristics
. call with Maxim about scheduling algorithm
. still more investigation required
* NEON intrinsics vs assembler libvpx performance difference
investigation [5/10]
. GCC seems to generate poor code for address generation in NEON loads/stores
. maybe some improvements to the intrinsics code would also be possible
Plan:
. write down some conclusions about libvpx performance
. investigate PR60609
. more on TCWG-135
. investigate LP1296676 & LP1296601 ICEs when building the kernel
