I've just submitted a merge request for the vldN and vstN intrinsic
improvements. There are five related patches, so I thought it might
be easier to review the merge if I posted the individual changes here.
See:
http://www.mail-archive.com/linaro-toolchain@lists.linaro.org/msg00969.html
for an example of how this helps.
Richard
I just did `git checkout` for linaro/linux-2.6.38 and tried compiling with
CROSS_COMPILE=arm-linux-gnueabi- and I get this smc #0 error.
The bug filed here is marked as fixed, but it's still broken for me:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/669912
I ran into this error once before, but I can't find my notes for it.
I think the solution was to append something like $(sec) on a line in the
Makefile.
Any ideas?
AJ ONeal
Hi there. I'd like to change and clarify how we make changes inside
Linaro toolchain projects. The write-up is here:
https://wiki.linaro.org/WorkingGroups/ToolChain/MakingChanges
The short story is that now, just like mainline, the developers
themselves do the final commit or merge into bzr. I've volunteered
three of you on a review roster so that someone has the responsibility
for reviewing patches each week. The roster is:
* Ramana week 15, 17, 20, ... starting the 11th of April 2011
* Richard week 16, 18, 21, ... starting the 18th of April 2011
* Ulrich week 17, 19, 22, ... starting the 25th of April 2011
Anyone is welcome to review patches. The more eyes the better.
Thoughts?
-- Michael
I'm just curious as to what "__aeabi_unwind_cpp_pr0" is or means.
I can't find any explanation by googling.
I'm about to post a separate e-mail with my problem concerning it.
AJ ONeal
Reviewed and approved Revital's do-loop patch, and Ira's store sinking
patch. More precisely, I reviewed the test results from Michael's test
system, and cast my eye over the patch to look for anything obvious. I
don't pretend to know exactly what they do.
Attended Ramana's thumb2 optimization discussion.
Continued work on merging useful patches from CodeSourcery. Pushed the
new patch set to Launchpad for testing in Michael's system.
Pointed Bernd at lp:758082 - another probable shrink-wrap failure. Bernd
responded with a new patch, and asked me to test it. I've pushed it to
Launchpad and created a merge request.
Posted an old patch of Dan's upstream:
http://www.mail-archive.com/gcc-patches@gcc.gnu.org/msg03144.html
Pinged my thumb2 constants patch upstream. Richard E responded with some
issues to address. I'll need to change the names of the constraints, at
least.
At Ramana's request, tested the NEON scheduling patch with SPEC2000.
Encountered trouble with time-outs. Fixed those and retested.
Posted an updated version of my EABI half-precision patch:
http://www.mail-archive.com/gcc-patches@gcc.gnu.org/msg03210.html
Merged my backport of Julian's -fstrict-volatile-bitfields bug fix into
Linaro GCC.
The testing of the Android patches came back with a few test
differences, but they appear to be unrelated to the patch, so are
probably environmental. I've merged the changes to Linaro GCC.
[Also spent most of Friday working on internal CS tasks.]
----
Upstream patched requiring review:
* NEON scheduling patch
http://gcc.gnu.org/ml/gcc-patches/2011-02/msg01431.html
* NEON test cases
http://www.mail-archive.com/gcc-patches@gcc.gnu.org/msg03144.html
* ARM EABI half-precision functions (reposted)
http://www.mail-archive.com/gcc-patches@gcc.gnu.org/msg03210.html
== Last week ==
* CoreMark regressions: pushed a merge of my two upstream patches to
Linaro 4.5, some current numbers are here:
http://lists.linaro.org/pipermail/linaro-toolchain/2011-April/001087.html.
* Continued working on another combine patch for improving CoreMark,
hopefully ready to submit this week.
* Committed fix for PR48325 (NEON POST_INC/PRE_DEC load/stores for
struct modes) upstream.
* Committed fix for PR48250 / Launchpad #723185 upstream.
* Launchpad #689887, ICE in get_arm_condition_code(). My prior patch was
tested to cause native bootstrap failure on Linaro 4.5, though retesting
on upstream trunk worked fine. Still investigating.
* Booked travel for Linaro-Budapest event.
== This week ==
* Current combine patch.
* Some unresolved patches, like PR46888.
* Launchpad #689887, hope to figure this out.
I've started up a page with ideas for sessions at next month's summit:
https://wiki.linaro.org/MichaelHope/Sandbox/1111Blueprints
They're a pretty direct map to the TSC technical topics so far which
is nice to see. Feel free to add other ideas to the page. Each
session is around 45 minutes, needs to be fairly well understood by
the drafter beforehand, and should have concrete actions coming out of
it. It's fine to have a few future/blue sky sessions but not too
many.
We'll discuss these at tonights meeting and assign drafters.
-- Michael
I've now submitted the initial vldN and vstN work, so I thought I'd see
how often it triggers for natty's libav package. I've put some initial
results here:
https://wiki.linaro.org/RichardSandiford/Sandbox/NeonLibAv
There are more files to go through, so this isn't complete.
I've also left out cases that were very similar to the ones given.
Some of the code is reasonable, while others are obviously not as good
as they could be. I don't think the problems are really to do with
the vldN and vstN work itself though. They seem to be due to the
underlying interleaved load/store detection, or in the handling
of widening operations.
Richard
== Bug triaging ==
* Bug 745843 (vtk ftbfs) got it down to a bad arm/thumb transition -
identified as a linker error and handed off to RichardS
* Bug 758082 (augeas ftbfs) tracked it down to overwrite of a
parameter in a variadic function before it got stacked; identified by
Ramana as another
instance of the shrink-wrap bug.
* Bug 745861 (petsc ftbfs) isolated the collection of different mpi
related problems this is hitting; really need to find an mpi expert on
this
* Bug 745863 & bug 745891 (ftbfs's) - both were compilations that
timed out; verified this was due to using lots of RAM and also using
lots of RAM on x86
(> ~500MB) - marked as invalid until the build farm grows more RAM
* Bug 757427 gconf seg fault - failed to reproduce under various
tests (although Michael has now managed to catch it in the act)
== Optimisation ==
* neon memcpy tweeking; added prefetches and unrolled the core loop
- now comparable perf to bionic memcpy in most cases (slower on
misaligned destination, faster in other cases)
* tweaked latrace to print address/length of argument strings so I
can get some stats on routine usage.
Dave
