== This week ==
* Looked at why the fix for #721531 wasn't working on the Linaro branches.
Wrote follow-up patches for both. Fix now committed to 4.5 and 4.6.
* Looked at #736007. Submitted and committed patch upstream.
* Looked at some "odd" ivopts behaviour. It turned out that this was
working mostly as expected. I'm still wondering about a couple of tweaks.
* Looked briefly at a miscompilation of vector code that turned out to
occur during predictive commoning. I haven't yet checked whether the bug
is there or not. For the time being, I'm using -fno-predictive-commoning
so that I can get on with other stuff.
* Looked at why the auto-inc-dec stuff wasn't as effective with current
mainline. It turns out that the new misaligned load/store patterns
are using overly weak predicates, so it appears to the RTL optimisers
as though we support reg+offset addresses.
* Reviewed the shrink-wrap patch.
* More auto inc/dec.
== Next week ==
* Backport the fix for #736007.
* More loop stuff.
Richard
Achieved:
* HW in place. Borrowed another Panda board that I can use permanently.
* linaro-media-create installed and working. Flashed sd-card with 11.05 for
Panda board.
* Played around with the Panda board. Networking is on the way.
I can ssh to my computer from the Panda but not the other way around yet.
Issues:
* Serial log from Panda board not working yet. Mike suggested I try a
straight-through cable instead in of a null-modem cable. (Have not done this
yet.)
* The ST-E firewall (proxy) is causing me some headache. But at least
everyone here is facing the same issues, so there are people to talk to
about the problems.
Good to know:
* I will be on vacation 18th July - 15 August
Best Regards
Åsa
Hi,
- continued working on prevention of over-widening in vectorization -
finalizing the patch
- improvement of vectorizer peeling heuristic - merged to gcc-linaro-4.6
- vectorization of widen-mult with over-promoted operands - proposed
for merge to gcc-linaro-4.6
- fixed PR 49610
- patch reviews
I am on vacation tomorrow and on Sunday.
Ira
Hello,
I'm interested in LLVM correct performance on ARM hardware and it looks
like LLVM is kind of sensitive to what GCC version is used for its
compilation. I tested LLVM 2.9 as a reference point and LLVM
HEAD as of June 29 on ARMv7 (two boards with two different Ubuntu
versions) compiled by GCC 4.3.4, 4.4.1, 4.4.5, 4.5.2, 4.6.1 Linaro
2011.05 and 4.6.1 Linaro 2011.06. Please see
http://ghcarm.wordpress.com/2011/07/03/llvm-on-arm-testing/
It looks like LLVM HEAD does have
about 28 regressions in comparison with LLVM 2.9. But also Linaro's GCC
4.6.1s do have some regressions in comparison with older GCC 4.3.4 and
4.4.1. Also what is really interesting with LLVM is how much tests fails
when compiled with -O2 or default -O3 compilation option. I don't know
if the culprit here is LLVM code or just GCC
miscompilation/overoptimization?
Is there any testing I may do to help you fix those regressions?
Thanks,
Karel
Hi,
* continued to look into how to add remote support for libunwind using
ptrace
* reworked the lookup of the ARM specific unwind tables for local
unwinding
* re-use the existent (dwarf related) infrastructure to find the ARM
specific unwind tables rather than doing it on our own
* removes some code and the limitation of only supporting a certain
amount of unw tables
* (hopefully) smooths the way for remote unwinding via ptrace
* submitted a patch that fixes the syntax of inline assemblies for
some GCC versions
Regards
Ken
(This is a combined report having realized that I didn't manage to
send this out on Monday because of some power issues).
== GCC ==
== Progress ==
* Cleared out some backlog of backports from mainline.
* Investigating some of the BRANCH_COST results.
* Small patches to fix VFP constraints being tested and fixed on trunk
upstream.
* Upstream bugzilla triaging.
* Upstream bugzilla triaging.
* Looked at Neon 64 bit arithmetic and sent out report.
* Reviewed the unaligned access patch and the EPILOGUE_USES regression
with coremark.
* Issues with my panda board means my benchmark runs aren't happening.
Need to urgently find a way of running them. The panda board dies
mysteriously once a while. Playing around with the kernel to get
something going . Will look at it again next week.
* Merged neon shift immediates patch into linaro-4.6 but needs some rework.
* Merged the fix for LP744754 into linaro 4.5 and 4.6
== Plans ==
* Back on to BRANCH_COST .
* Resubmit neon shift immediates patch for merging.
* Look at some of the perf regressions between a8 and A9.
* Get a working panda board again. !
* Backport arith_shiftsi patch to 4.6 branch upstream.
Meetings:
* 1-1s
* TCWG calls.
== This week ==
* More on the address-of-main() bug. The original patch caused
regressions on x86_64, so I submitted a different approach,
which has now been applied upstream.
* Worked on the libnih bug. It turned out to be a problem in the
libc start routine.
* A bit of patch review.
* More on auto inc/dec.
== Next week ==
* General bug-fixing.
* More auto inc/dec.
Richard
== 64 bit atomics ==
* Submitted patches to gcc patch list
- One comment back already asking if we should really change ARM
to have a VDSO to make checks of the user helper version easier
* Added thumb ldrexd/strexd to valgrind; patch added to bug in their
bugtracker (KDE bug 266035)
* Came to the conclusion eglibc doesn't actually need any changes
- It's got a place holder for a 64bit routine, but it's unused and
isn't exposed to the libc users
- Note that neither eglibc or glibc built cleanly from the trunk on ARM.
* Started digging into QEmu a bit more to find out how to solve the
helper problem
== String routines ==
* Added SPEC2k6 string routine results to my charts; while most
stuff is in the noise it seems
the bionic routine is a bit slower overall than everything else, and
my absolutely trivially simple
~5 instruction loop is a tie for the fastest with my smarter 4
byte/loop using uadd.
== Next week ==
* Sleep, Rest, Relaxation, getting older
* (Will be polling email for any more follow ups on my gcc patches)
Dave
