== Progress ==
* Validation: (2/10)
- reached a state where Jenkins "green" actually means no regression found
- submitted patches for review
- confirmed that ABE master branch produces results similar to merge branch
- stopped checking the accuracy of report.sh since a few bugzilla
had been submitted, using ST script instead
* Backports (3/10)
- reviews are getting difficult, with lots of conflicts as the 4.9
branch has diverged from trunk quite a lot
* Sanitizers (1/10)
- trying to build LLVM+run the ASAN tests on AArch32
- tried to reconfigure my Hikey 96board with a 42 VA bits enabled
kernel, but it didn't boot: I still don't have access to such a
platform
* GCC FSF trunk/4.9 monitoring
- stopped tracking random "Interrupted system call"
- builds and validations now run faster, but with more frequent such errors
* Misc (4/10)
- meetings, conf-calls, emails, ...
- Release/maintenance branches policy, respin criteria, validation matrix
== Next ==
* Backports: reviews, and improve review-tool
* Sanitizers
Hi,
I have being trying to persuade gcc to generate the ldp instruction without success. I have tried many combinations, below is an example.
--- cut here ---
#define LDP(x,y,p) { \
struct vec { long x, y; } data; \
data = *(struct vec *)p; p += 2; \
x = data.x; y = data.y; \
}
long testp(long *p) {
long x, y;
LDP(x, y, p);
return x+y;
}
--- cut here ---
$ gcc --version
gcc (GCC) 4.9.1
Copyright (C) 2014 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
$ gcc -S -O3 ldp.c
$ cat ldp.s
.cpu generic+fp+simd
.file "ldp.c"
.text
.align 2
.global testp
.type testp, %function
testp:
ldr x1, [x0,8] ;; why not ldp?????
ldr x0, [x0]
add x0, x1, x0
ret
.size testp, .-testp
.ident "GCC: (GNU) 4.9.1"
What can I do do make it generate ldp?
Ed.
== Progress ==
Bug triage [2/10]
Bug 1199 analysis [4/10]
reproduced, studied IRA dumps for a while, will have to report this upstream
Backports + BackFLiP enhancement [3/10]
Misc [1/10]
meetings, git training presentation, mailing lists
== Plans ==
benchmark 2015.02 release
more bugs
== Progress ==
LLDB development
-- Some code sniffing and created JIRA cards with some task estimation [2/10]
-- Trying AArch64 cross build and update howto [1/10]
-- Trying Arm cross build and update howto [2/10]
GDB patch updates and test cases for ARM record-replay instruction
recording [4/10]
Miscellaneous [1/10]
-- 96board with custom kernel no success with getting ethernet
working on it yet
== Plan ==
Wind up all pending gdb patches
== Issue ==
* none
== Progress ==
* Release and Backports (6/10)
- Supported backporting activity
- Added new feature to Backflip
- Discussed redefinition of release/maintenance policy
* Misc (4/10)
- Various meetings
- Catch up with mails
- Updated GCC release process in collaborate
- HiKey
== Plan ==
* Release and backport support
== This week ==
Short week - 3 days. Took Sat, Sun off due to health issues, severe
ear infection -:(
* Modularization project
a) TCWG-637: Applied TLC to parser in include-remove tool
b) TCWG-621: Began refactoring rtl.h
c) TCWG-628.
* TCWG-619:
- Started going through LTO documentation.
* Bugs
1178: Found a possible fix, prevents the ICE for the test-case.
== Next week ==
* TCWG-619
* Submit fix to 1178 upstream
* TCWG-628
== This week ==
* GCC Modularization project
- TCWG-621: Refactoring rtl.h almost completed.
- TCWG-639: Working on rewriting the tool in python.
- TCWG-637: Working on basing the tool on rtags instead of ctags.
- TCWG-629: Continue target-hook conversion.
* Backports:
- Submitted following backports for review:
r215612, r215722, r219656, r219657, r219659, r219661
== Issues ==
- Building chromium dependencies.
- Cannot log in to NX session on tcwg-env-05.
== Next Week ==
- Chromium LTO build (TCWG-619)
- Validate and commit backports.
== Progress ==
* compiler-pass to widen computation (TCWG-547) - (7/10)
- ran into -std=gnu90 issue in compiling spec2k/2006 with the trunk
using the new infrastructure
- ran benchmarking but the results are not much different
- tried with -save-temps but the intermediates are gone. Not sure if
infrastructure cleans it
- setting up spec2k in nx aarch64 chroot for easy comparison of
intermediate values (asm and tree dumps)
- also plan to try benchmarking with widening pass enabled only for
unsigned type where BIT_AND_EXPR (zero_extend) is optimized well as
compared to new middle-end ZEXT_EXPR (which is not handled well yet) to
see if the difference is there.
* Improve register allocation for AArch64 (TCWG-620) - (1/10)
- Looked at the relevant patches and code
* https://bugs.linaro.org/show_bug.cgi?id=1291 (1/10)
* Misc (1/10)
== Plan ==
* TCWG-620 and TCWG-547
* https://bugs.linaro.org/show_bug.cgi?id=1291
* TSAN support for Aarch64 (5/10)
TCWG-642
Discussed on making -pie for -fsanitize=thread as default in GCC.
Learnt that LLVM sanitizers now support both -pie and non -pie mode.
Discussed the same issue with sanitizer's group. They asked me to
do -pie for aarch64 as first version. Pushed GCC changes to
linaro git.
TCWG-654
Discussed with Renato, moved to LLVM tree and woking on that.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64946.
TCWG-634
Ran ASAN tests which 32 bit allocator code now as default on amd-01.
/home/venkataramanan.kumar/LLVM/TSAN_port/Source/projects/compiler-
rt/lib/sanitizer_common/sanitizer_allocator.h:841
((res)) < ((kNumPossibleRegions)) (4193146, 524288)
I got similar error in TSAN and it went off when I switched to 64
bit allocator.
* Bug 869 - Wrote a pattern in match.pd to remove type promotion in
ABS_EXPR . Richard Beiner suggested to add ABSU_EXPR to take care of
signed overflows (3/10)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64946
* Emails, meetings. (2/10)
* Linaro 1-1 with christophe, maxim and Ryan.
* AMD meetings/event, 1-1 with AMD manager, status meeting.
* GCC mailing list.
== Plan ==
* TSAN/ASAN support look if 64 allocator is needed or not .
* LLVM build with local patches.
* Bug869.
* Misc and Meetings (AMD, LInaro)
== Progress ==
* Releases (CARD-1431 2/10)
- Spinning RC4, final, all green
* Buildbots (CARD-1823 1/10)
- Adding new quick Thumb2 non-NEON bot
- Adding new full self-hosting Thumb2 bot
* Libraries (CARD-1831 1/10)
- Trying to build a soft-float compiler-rt test-suite
* Sanitizers for AArch32/64 (CARD-1753 4/10)
- Investigating/fixing broken sanitizer bots (4 failures)
- Investigating other sanitizer failures on Junos
- More planning on TSAN, ASAN
* Background (2/10)
- Code review, meetings, discussions, etc.
- Reviewing the remaining EuroLLVM papers
- More LLDB planning
== Plan ==
* LLDB meeting with Google
* EuroLLVM final paper review meeting
* Continue building test-suite with Compiler-RT, libc++
* Continue help on ASAN/TSAN
* Try our first sprint planning...
# Progress #
* aarch64 gdb, TCWG-652, [5/10]
** TCWG-653: triage fails in aarch64-linux gdb testing, both native and
remote. Nearly done.
** TCWG-670: Clean up GDB testsuite for syscall catchpoint first.
Ongoing.
** TCWG-672: GDB can't unwind from glibc with the expected function name
if glibc is stripped. Nearly done.
** Committed two patches upstream fixing two fails on aarch64-linux
target.
** Clean up arm and aarch64 bug reports on sourceware.org bugzilla.
* fsf gdb patch review [4/10]
** Reviewed Omair's two process record GDB patches.
** Reviewed one patch may cause fails on remote testing.
* misc [1/10]
** 1x1 with Ryan.
** With Bernard's help, set up ssh access to linaro servers.
# Plan #
* All cards under TCWG-652.
* Move out of temporary apartment to a rented one on Friday.
--
Yao
Go away where ? ABE just uses whatever triplet you give it. I commonly use just arm-linux-gnueabi for example. If you don't specify 'none', it shouldn't appear anywhere. If it does, let me know and I'll fix it.
- rob -
-------- Original message --------
From: Christopher Covington <cov(a)codeaurora.org>
Date: 02/26/2015 2:52 AM (GMT+07:00)
To: linaro-toolchain(a)lists.linaro.org
Subject: Debian Multiarch Tuple Style Naming With ABE
Hi,
I recently built with ABE for the first time. It's pretty slick.
Is there a way to make the vendor string (the "none" in
"arm-none-linux-gnueabihf-*") go away? It breaks my scripts without appearing
to add value.
Thanks,
Chris
--
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
_______________________________________________
linaro-toolchain mailing list
linaro-toolchain(a)lists.linaro.org
http://lists.linaro.org/mailman/listinfo/linaro-toolchain
Holiday Friday [2/10]
catomics - TCWG-436 [2/10]
* Fixed broken benchmark-against-sysroot code in scripts
* Built a lot of sysroots
Juno cache effects - LDTS-1238 [4/10]
* Quite a bit of reading and learning some more perf and other tools
* Formed a few hypothesis, started testing
* Looking promising, but I find these things often go wrong, so we'll see
Misc [2/10]
* Featuring first successful Jenkins run of all of SPEC2006
=Plan=
Make sure I've really fixed the last sysroot bug, actually run
benchmarks on catomics
Attempt to close out the cache effects issue
Carry on with Jenkins
Hi,
I recently built with ABE for the first time. It's pretty slick.
Is there a way to make the vendor string (the "none" in
"arm-none-linux-gnueabihf-*") go away? It breaks my scripts without appearing
to add value.
Thanks,
Chris
--
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
Hi,
I am trying to port the Boost.Context library (from www.boost.org) to aarch64 gcc and have come across a gnarly problem.
Boost.Context essentially does co-routine style context switching. It has a structure f_context which it uses to save and restore contexts. The structure f_context contains both integer (x0..xN) and floating point VFP (d8..d15) context.
The function jump_context switches contexts
typedef struct f_context *f_context_t
extern void jump_context(fcontext_t *old, f_context_t new, bool save_fp);
So this jumps to a new context returning the old context in *old. If save_fp is set the floating point context must be saved because the application uses floating point. Otherwise the save and restore of the floating point context may be ignored.
So, essentially, to save the old context it does
jump_fcontext:
# prepare stack for GP + FPU
sub sp, sp, #0xb0
# test if fpu env should be preserved
cmp w3, #0
b.eq 1f
# save d8 - d15
stp d8, d9, [sp, #0x00]
stp d10, d11, [sp, #0x10]
stp d12, d13, [sp, #0x20]
stp d14, d15, [sp, #0x30]
1:
# save x19-x30
stp x19, x20, [sp, #0x40]
stp x21, x22, [sp, #0x50]
stp x23, x24, [sp, #0x60]
stp x25, x26, [sp, #0x70]
stp x27, x28, [sp, #0x80]
stp x29, x30, [sp, #0x90]
However, there is a problem with this because gcc may store integer value in floating point registers around a function call.
So, I have no way of knowing whether it is actually necessary to save/restore floating point context.
Even worse applications using Boost.Context may be completely borken if they assume it is safe to call jump_context with save_fp == 0.
Any suggestions?
Ed.
== This week ==
* Released Linaro 4.8 and 4.9 February 2015 release (2/10)
* Vector Extensions Project (7/10)
- Design work on C++ classes
- Reviewed Boost.simd as alternative implementation
- Review of libvpx benchmark
* Misc. (1/10)
- Conference calls
- ARM required online training
== Next week ==
- ARM CodeGen conference in Cambridge
--
Michael Collison
Linaro Toolchain Working Group
michael.collison(a)linaro.org
== Progress ==
LLDB development
-- LLDB Call graph analysis for understanding some development nits [1/10]
-- Trying out LLDB server arm and aarch64 cross compilation. [2/10]
-- List down arm-lldb task list with some debugging [2/10]
Miscellaneous [1/10]
-- 96board setup
Holiday [4/10]
-- Monday 16th and Tuesday 17th Feb 2015
== Plan ==
Convert LLDB Arm task list to JIRA cards
Try to fix LLDB server arm and aarch64 cross build issues.
Some GDB patch updates
Connect detox [2/10]
ABE benchmarking automation - TCWG-360 [5/10]
* Backport benchmarking
** Harder than expected, have to build the backport into a binary,
then benchmark that binary, passing information between jobs.
** AFAIK I'm blazing the Jenkins-chainging trail for us.
* Release benchmarking
** New LAVA release broke my image (but quickly fixed by the LAVA folks)
** Now seems to be running into issues with Jenkins/builders/an
anonymous job-slayer
* Misc fixes and improvements
** Main highlight being support for building with separated-out sysroot
Misc - [3/10]
=Plan=
Kick off catomics benchmark runs
(Attempt to) close out cache effects on Juno issue
Carry on with Jenkins
=Issues=
* Some of the Jenkins builders seem unreliable
* My lightly hacked version of the BinaryRelease job is unreliable
** Is the BinaryRelease job itself is reliable
* TSAN support for Aarch64 (5/10)
Fixed some kernel size issues in my local working tree. (TCWG-581) .
GCC does not default to -fPIE or -pie. Ran into address layout
issues when I used -fsanitize=thread. Some test cases passed with the
mapping I used for 42 bit VA when -pie and -fPIE are added.
* Bug 869 - Looking at match.pd to remove type promotion in ABS_EXPR (1/10)
* Emails, meetings. (2/10)
* Linaro 1-1 with maxim and Ryan.
* AMD meetings/event, 1-1 with AMD manager, status meeting.
* GCC mailing list.
Leave on 20-Feb-2015 (2/10)
== Plan ==
* TSAN support for Aarch64 work fixing GCC -fsanitize=thread option.
* Break tasks under TCWG-581.
* LLVM build with local patches.
* Bug869.
* Misc and Meetings (AMD, LInaro)
== Progress ==
* compiler-pass to widen computation (TCWG-547) - (5/10)
- Analysed core-mark code size regression
- fixed constant handling
- Added VRP support for ZEXT_EXPR
* Improve register allocation for AArch64 (TCWG-620) - (1/10)
- Looked at the relevant patches
* Misc (4/10)
- Setting up ABE benchmarking scripts
- Ran into some LAVA related issues for ABE benchmarking
- Recovering from connect
- Started with 69 board setup
== Plan ==
* TCWG-620 and TCWG-547
* ABE benchmarking
== Progress ==
* GCC trunk/4.9 monitoring
* Neon intrinsics tests
- committed the last patch of the 2nd series.
- about another 1/3 of the tests remain to be converted: will wait
for GCC stage1
* AArch64 sanitizers (2/10)
- Renato committed my patches to enable ASAN in LLVM/AArch64
- the official aarch64 buildbot (Juno) had problems after these, so
the asan tests are disabled again until the problem is understood.
Probable board configuration problem, since the tests passed on our
Juno.
- preparing a patch to solve the long standing issue with dependency
on aarch64 kernel version (kernel_old_[ug]id)
* Backports (4/10)
- reviews
- respawned validations under MergeFarm
* ABE:
- improving reporting
* Misc (4/10)
- meetings, conf-calls, emails, ....
== Next ==
Connect, then holidays
