[llvmlinux] LLVMLinux Meeting 2014-05-07
Behan Webster
behanw at converseincode.com
Fri May 9 22:44:14 UTC 2014
*
LLVMLinux Project: http://llvm.linuxfoundation.org/
This time the meeting was on Thursday, May 07 2014 on Google Plus
Hangout at 15:00 UTC.
-------------------------------------------------------------------------------------------------------------------------
Previous meeting's top 2-3 work items:
*
This is what people said they'd be working on at the end of the last
meeting:
o
Behan
+
Fix VLAIS patches and submit upstream
+
Follow up on netfilter patch
+
fix v3.15 kernel breakage
+
buildbot work
+
Get ready for ELC/ABS
o
Jan-Simon
+
Follow-up with HPA on -no-387 cflags -> cc-option or other
solution
+
mbcache compiler crash !!
+
intermediate bot for linux-next
+
-m16 vs code16gccflags (might be cc-option related)
o
Andrew Wells
+
read/research on static analyzers
+
look through what is already there (sparse, coccinelle)
+
modify timeline on proposal
+
annotation added for sparse, see if valid for clang
-------------------------------------------------------------------------------------------------------------------------
Agenda for this week:
*
(5 mins total) Everyone shared a quick summary of the progress
they've made from the last 4 weeks
o
Behan
+
Pull request into v3.15 was accepted
+
Spoke at Embedded Linux Conference on LLVMLinux project
+
Spoke at Android Builders summit on compiling AOSP with clang
+
Worked on getting more LLVMLinux patches into for-next
(arch/all, arm, and x86_64)
+
A few patch rewrites
+
Fixed beaglebone (white/black) target build
+
Worked on fixing breakage from new kernel merge window
+
Working on new buildbot
o
Jan-Simon
+
linux-next boots
+
New buildbot:
#
brainstorm
#
https://github.com/dl9pf/llvmlinux-buildbot/tree/resetup
#
we will have remote builders
#
main instance will just provide
*
git poller
*
llvm poller
*
buildsteps
o
llvm/clang: configure + build + install
+ record git IDs
o
targets: make, make test-kill (or new make test-bot)
*
4 reference builds (llvm, clang, vexpress, x86_64)
*
a clang build passing all of these tests will be
recorded (git IDs for the remote builders to reuse)
*
remote builder for x86_64-next with test in qemu
*
remote builder for beaglebone with boot test on HW
o
Vinicius Tinti
+
Add panda target for linaro
+
Add vexpress-linaro target for linaro
+
Integrating with LAVA
+
Creating a Vagrant/Puppet machine for buildbot
+
Nexus 7 does not build anymore.
o
Mark
+
Checking new targets
#
cubieboard
+
Added qemu 2.0
+
IFC board still not booting. Probably needs JTAG to resolve
*
(45 mins) General meeting discussions:
o
GSoC
+
Sadly Andrew Wells will not be our GSoC student this year.
He was accepted into multiple organizations and we lost him
during GSoC de-duplication
+
He still would like to work with static analysis checkers
o
Kernel patches
+
9 patches made it into mainline for v3.15
+
Another 2-3 patches were accepted through maintainer trees
+
2 more patches have been sent to fix merge window breakage
#
Turning off 80387 on x86
*
Wrapping -mno-80387 with cc-option
*
Use of FPU is explicitly disabled to make sure no
floating point instructions are ever issued
*
Currently gcc enforces that. Still investigating how
to do the similar thing with clang. Might have to
just turn off all FPUs manually.
#
__builtin_log2
*
mbcache.c uses log2 to calculate hash value
*
In clang it was generating a double, which was then
cast as a unsigned long
*
gcc on the otherhand was generating a long
*
Mark patched it to use the kernel ilog2() macro instead.
*
Using __builtin_log2l also might have solved the problem
+
All the above patches should end up in v3.15
+
Named Registers in Clang
#
Named register support has made it into LLVM, but Renato
is having troubles convincing the clang folks to make
changes to support it in the front end
#
So back to our original named register kernel patches
#
gcc supports many different kinds of allocatable
registers using the same syntax
#
Global allocatable register are much more likely to be
accept rather than local allocatable. Implement first
global that will solve partially the usage on kernel
+
Integrated-as:
#
x86:
*
David Woodhouse is still working on enabling
integrated-as on x86_64sting more of the error
messages to Renato so he can see if th
*
-m16 needs debugging regarding cc-option
#
arm
*
Renato is working with Vinicius in order to get IA
working with ARM. Mostly poey can be fixed.
#
A number of macro expansion issues have been fixed
#
BUG macro is still broken
#
http://llvm.org/bugs/show_bug.cgi?id=19317
+
Example subject lines for patches:
#
Subject: [PATCH no-upstream] USB: LLVMLinux: Remove
VLAIS from gadget driver
o
New buildbot
+
Behan and Jan Simon are working on building the new buildbot
VM Ubuntu Server
#
Master builder will run all the VCS pollers
#
Master builder will build clang, llvm, vexpress, x86-tiny
#
Successful clang builds will be used as known good clang
versions for the rest of the target builds
#
All other targets will be built on other (remote slaves)
#
This allows people to run local builds to allow local HW
testing
#
Some build targets will be pushed to the LAVA test
system at Linaro
+
The old buildbot is broken (distro/gcc/glibc is too old to
compile clang)
+
Vinicius is working on getting builds integrated into LAVA
testing at Linaro
o
Conferences
+
Behan spoke at Embedded Linux Conference on LLVMLinux project
#
http://sched.co/1oqUC2f
+
Behan spoke at Android Builders summit on compiling AOSP
with clang
#
http://sched.co/1or07Oy
+
Behan has submitted talks to LinuxCon US (Chicago) and
Linaro Connect US (SFO)
o
Dev boards
+
New Qualcomm DragonBoard IFC6410 (APQ8064 Quad-core A15 with
GPU)
+
Mark, Behan and Jan-Simon now all have this board
+
The BeagleBone White and Black boards now have a working
target (the BB kernel code is now in mainline) A few patches
are required to support this build.
o
Build system updates
+
You can now build llvm/clang with your system clang (instead
of gcc)
#
make CLANG_SELFHOST=1 clang
+
At Vinicius' request, Behan added the ability to add a
target to allow kernel configs to be dynamically built so
that Vinicius can support the Linaro kernel config which is
built from "config fragments"
#
Tested and working.
*
(5 mins) Each meeting attendee listed the top 2 items to be worked
on for the next 2 weeks
o
Behan
+
Buildbot rework
+
Named register patch work
+
Submit more patches to lkml so that I can send them to
linux-next
o
Jan-Simon
+
buildbot rework
+
testing setup beaglebone black/white (buildslave)
+
x86-64-next buildslave testing with qemu/kvm
o
Vinícius Tinti
+
check integrated assembler
+
buildbot rework
+
lava integration
+
fix nexus7
+
probably n4
o
Mark
+
jtag for IFC
Next meeting in roughly 3 weeks (probably after Embedded Linux
Conference). doodle poll to follow.
*
--
Behan Webster
behanw at converseincode.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/llvmlinux/attachments/20140509/00de12bc/attachment-0001.html>
More information about the LLVMLinux
mailing list