...are available here: https://wiki.linaro.org/WorkingGroups/ToolChain/Meetings/2010-08-09
A copy follows, along with activity reports for the different members.
-- Michael
Attendees
• Name Email IRC Nick Yao Qi yao.qi@linaro.org yao Ulrich Weigand ulrich.weigand@linaro.org uweigand Scott Bambrough scott.bambrough@linaro.org scottb Richard Earnshaw richard.earnshaw@arm.com rearnshaw Peter Maydell peter.maydell@linaro.org pm215 Michael Hope michael.hope@linaro.org michaelh Marcin Juszkiewicz marcin.juszkiewicz@linaro.org hrw Loïc Minier loic.minier@linaro.org lool Julian Brown julian@codesourcery.com jbrown Dave Martin dave.martin@linaro.org dmart Chung-Lin Tang cltang@codesourcery.com cltang Andrew Stubbs andrew.stubbs@linaro.org ams
Agenda
• Review action items from last meeting • New public call-in number: code 263 441 7169 • Release status □ 4.4 at http://ex.seabright.co.nz/build/gcc-linaro-4.4-2010.08-0/ □ 4.5 at http://ex.seabright.co.nz/build/gcc-linaro-4.5-2010.08-0/ □ Ubuntu toolchain would prefer 4.5.1 • Next focus □ Merging patches upstream □ Prioritising GDB features • Future focus □ Emphasise performance, neutral on correctness □ Topics to investigate ☆ Vectoriser ☆ Pipeline ☆ PGO ☆ Upstream research topics □ Topics for UDS □ Benchmarks
Blueprint Assignee Initial delivery of Linaro GCC 4.4 ams Cross Compiler Packages hrw
Action Items from this Meeting
• TBD
Action Items from Previous Meeting
• DONE: Michael to organise and spread next Monday's call in number • ACTION: Richard to ask the GCC developers on IRC what the status of 4.4.5 is • DONE: 500524 Ulrich to backport to Linaro 4.4 • DONE: Test failures: Michael to build FSF 4.4.4 as a baseline • DONE: Test failures: Michael to compare FSF and Linaro GCC failures • DONE: Test failures: All Linaro GCC failures to be ticketed • DONE: LTO check on ARM: Michael to reproduce. check-lto fails with no Makefile target • DONE: Ulrich to write up known GDB work • DONE: 598147: Michael to reproduce the test failures • DONE: 398403: Andrew to reproduce on his IGEP board (has more RAM) • DONE: Michael to make sure merge requests are present for work that has been done
Minutes
• Release status □ Talked about the upcoming 2010.08 release □ Andrew has posted tarballs which Michael has grabbed and built □ Andrew is cross-testing • Discussed 4.5.1 □ 4.5.1 came out, depending on where you look, 11, 8, or two days ago □ Rolling in 4.5.1 and the Firefox exposed alignment bug would have more value □ Discussed time based vs feature based releases and a pragmatic middle ground □ Discussed releasing 4.5.0 based now vs 4.5.1 at next months release vs something inbetween □ Decided to delay the Linaro 4.5 release until at most 2010-08-17 □ Roll in 4.5.1 and the Firefox fix □ Gives something more useful to Ubuntu and others □ ACTION: Andrew to merge 4.5.1 and the Firefox fix by 2010-08-17 • Discussed what goes in next month □ Richard wants the sync primitives fixes □ CSL is still merging their patchset • Discussed what can be done in the future □ Please start thinking of topics for UDS □ Backporting the new costing infrastructure (Richard) ☆ Works by changing how costs are done and making them more generic ☆ Backend provides options for a certain segment ☆ Cost of an option is target dependent ☆ Removes a lot of hard-coded values for particular architectures ☆ Good for core-next ☆ Infrastructure matches trunk ☆ Might pull in ldm changes □ Richard's wishlist ☆ ACLE ☆ v6 SIMD, v5 (limited) □ Conditional instructions ☆ Some are best done at the expand phase where tree goes to RTL ☆ Not on Roger' slist ☆ Check on possible gains □ NEON intrinsics ☆ Don't generate very good code at the moment due to load/stores ☆ Might need internal intrinsics to compose these out of ☆ Developers currently use straight asm, perhaps out of personal preference ☆ Intrinsics aren't well documented/known ☆ Worthwhile as people coming from RVDS and using intrinsics will see big drop □ GDB ☆ Ulrich sent through his list ☆ To categorise into features and fixes ☆ ACTION: Ulrich to ticket items ☆ Work on critical fixes first ☆ Yao, Ulrich to work on list ☆ ACTION: Michael to understand whiteboards as a way of organising features
Next meeting is a standup meeting on 2010-08-11 on the public code.
--- ams: == Linaro GCC 4.4 ==
* Reviewed various merge requests.
* Created and tested GCC Linaro 4.4-2010.08-0 release tarball.
== Linaro GCC 4.5 ==
* Continued pushing CS patches to Linaro GCC 4.5. I've done quite a lot of these this week, but I've had to omit some patches that caused regressions.
* Created and tested GCC Linaro 4.5-2010.08-0 release tarball.
== Other ==
* Set up EEMBC for use with Linaro GCC so Chung-Lin can compare hard-float and soft-float ABIs.
* Attended the monthly CS/Linaro sync call.
--- yao: == Linaro GCC 4.4 bug fix ==
* LP:602190 & LP:602285 They are caused by a patch for performance improvement of generated code. Team suggests keep this patch, and fix test case a little bit to fix failures. Fix to LP:602285 is committed and merged to linaro 4.4. Thinking about how to fix LP:602190 from test case side. An option to force to turn 'cunroll' off is helpful.
* LP:612011 & GCC PR45094 Update test case pr45094.c from a compilation test to an execute test. Sent to gcc-patches again. Waiting for ARM backend maintainer's comment. Merge this patch to linaro 4.4 first, because of release. Plan to backport to linaro gcc 4.5 once patch is approved by upstreams.
* LP:612405 & LP:612406 Fix them by changing ARM target triplet from "arm-*-*" to "arm*-*-*", which is more general. Merged to linaro gcc 4.4. Create a patch against FSF GCC trunk to fix this in other test cases.
* LP:612402 Same bug was reported in GCC Bugz PR41501/PR34999, which have been fixed on FSF GCC 4.5.0. No plan to backport to FSF 4.4.x. Leave this bug as Won't Fix on linaro gcc 4.4 and Fix Released on linaro gcc 4.5.
* LP:602174 Can't reproduce it on arm-none-linux-gnueabi. Latest result shows obj-c++.dg/bitfield* only failes on x86-64.
* LP:602168 g++.dg/eh/pr42859.C test failure Analyze this bug a little bit, and my initial result is that two BBs in exception catch are *not* marked as UNREACHABLE in linaro 4.4, while these two BBs are UNREACHABLE FSF GCC 4.4 trunk.
== This Week ==
* Ping ARM backend maintainer for my patch to GCC PR45094.
* Submit my patch on ARM target triplet in gcc patches.
* Update status of LP bugs. Mark them as Fix Released if failure go away in gcc-linaro-4.4-2010.08-0.
* Other linaro 4.4 bugs to look or other new work to do. Wrap up LP:602190. Analyze deeply on LP:602168/LP:602174.
--- pmaydell:
RAG: Red: None Amber: waiting final ARM legal OK for qemu contributions Green: Completed all the linaro new-starter admin tasks
Milestones: none as yet
Progress:
- Miscellaneous setup/configure/getting started.
virtio-system: - got qemu to boot linaro alpha-2 with random ubuntu versatile kernel - I have a set up that lets me rebuild kernels now
Issues: - maemo-qemu supports beagle but the stock linaro alpha 3 takes a div-by-zero exception and panics on startup
Plans:
general: - flesh out some of the blueprints assigned to me virtio-system: - investigate the div-by-zero panic - find out the easiest way to replace the kernel in the shipped alpha 3 with one I've compiled - make sure virtio is enabled in kernel and try to get it working
Absences: None planned.
--- uweigand: == GCC ==
* Prepared Linaro GCC 4.5 backport for bug #604874 (Firefox fails to build) * Prepared Linaro GCC 4.4 backport for bug #500525 (bootstrap failure in stage3) * Investigated bug #600209 (getfem++ test failures) * Investigated bug #602168 (g++.dg/eh/pr42859.C failure) * Investigated bug #506538 (GENERAL_REGS spill failure) * Determined root cause of bug #608125 (gnat build failure) * Determined root cause of Debian bug #591075 (glib2.0 segfault on ARM / i386)
== GDB ==
* Started planning future GDB feature / bugfix work
== Miscellaneous ==
* Converted multiarch toolchain implications document to Wiki page
--- cltang: == libffi VFP hard-float support ==
Mostly finished and doing testing. Using libffi's testsuite, currently has three regressions, two of them related to testing of variadic functions, which are pretty unsolvable due to AAPCS' rule of reverting to the base calling convention for all variadic functions (the libffi code currently has no way of detecting variadic function type from the user, so cannot correctly do this ABI revert). May solve by marking the testcases to be skipped under -mfloat-abi=hard. Will work on a few code optimizations before trying to submit it upstream.
== Other ==
Looking at some Linaro GCC bugs.