Zach Welch --
== Last Week ==
* Continued working on libunwind support. Trying to figure out why my
signal frame detection doesn't work as expected.
* Kept pace with the ltrace tree, testing recent patches on ARM.
== This Week ==
* Continue to work on libunwind signal frame detection.
Julian Brown --
== Linaro GCC ==
* Looked at issues #663198 (double-precision register expected) --
which was already fixed on the linaro branch, but the bug was reported
against a version just prior to that, and #667490 -- which involved a
possible problem with the NEON "load 0.0" patch. Experimented for a
while with the latter, but could not find anything wrong.
Followed up upstream, and requested a stand-alone test case.
* Worked on a proper solution to the VMOVN-in-big-endian-mode problem,
discovering that several other quadword-register operations were
similarly broken in the process. WIP patch sent to linaro-toolchain for
discussion, but it needs a little more work before it can be applied.
Peter Maydell --
Progress:
* qemu: more cleanup of signal handler VFP patchset;
I think I just need to add iwmmx support and it's good
* qemu: VCVT: found yet another bug, did final patchset
cleanup: submitted to upstream list [8 patch series]
* qemu: submitted a trivial patch to fix a problem where
__get_user/_put_user macros had an unnecessary local var
which could clash with a var being used by the macro user
* set up a tree on git.linaro.org which we can use for
a branch to make pull requests for ARM qemu fixes
* did a rough estimate of time to do an Eagle qemu model
(6 months + testing/bug fixing time)
Issues:
* lost some time to a problem where Linux VMs stopped being
able to talk to the LDAP server; however I have a workaround
and IT are investigating
Meetings:
* toolchain, toolchain standup, pdsw-tools, PD doughnuts
Plans
- attend Meego conference in Dublin (Nov 15-18 inc travel)
http://conference2010.meego.com/
- start on qemu consolidation by upstreaming various ARMv7
correctness fixes
Andrew Stubbs --
== GCC 4.5 ==
* Continued working on LP:663939.
* I still have not worked out how best to fix the constant
propagation problem that has been thwarting my optimization patch,
however I think I understand it better now.
* I have started on adding replicated pattern support to the
constant splitting. Initial results were good, but I discovered that I
had to rearrange the code somewhat to get the cost estimation and
negative/inverted constant support working correctly. So far, I have
it successfully using 16-bit replication pattern constants for
set/add/subtract operations. Other operations appear broken at the
moment, but it's almost certainly just a few tweaks required.
* TODO: Add support for 32-bit replicated pattern constants. Adjust
some of the other two-instruction constant generation techniques to
let them fall through to this new code, where it would be beneficial.
* Pushed the latest set of GCC patches into Linaro GCC 4.5.
Chung-Lin Tang --
== Linaro GCC ==
* Linaro #672833, one batch of my backports of Bernd's postreload
patches exposed some varargs regressions for x86-64, was reverted by
Michael. Tested the compiler and found it was fixed on mainline
rev.162384. Backporting this revision plus the postreload patches
fixed the regressions; x86-64 bootstrap also verified okay. There is
however another PR45027 fix that was needed on trunk, but needs a bit
more clarification if needed on a 4.5 compiler.
* Linaro #641397, CS issue #6753: bitfield optimization. Patch tested
without regressions, posted for CS internal review, should soon push
for Linaro merge.
* Started looking further at some GCC DF, IRA internals.
== This week ==
* Look at more Linaro issues.
* Maybe start looking at some GCC bugzilla PRs.
* There is a local ARM technical event in Hsinchu on Thursday, might
go and look around.
Yao Qi --
== Linaro GCC ==
* Mainline patch backport to Linaro 4.5.
** Patch "Fix an if statement in arm_rtx_costs_1". Verified on Linaro
4.5. 0.1% smaller on size, and 0.2% faster on speed. Merged to Linaro
4.5 by Andrew S.
** Try Nathan F's ifcvt-cond-move patch on cortex-a8 with -O2/O3. No
improvements on speed/size for EEMBC.
** Bernd's ldm/stm patch. Analyze the reason of regression on Linaro
4.5. Found something wrong in IRA rtl dump, and spend sometime on
understanding IRA rtl dump log. Thanks to Chung-Lin, I realize that IRA
dump is correct, and look back to ARM RTL patterns on ldm/stm. Compared
RTL patterns in 4.5 and 4.6, found some difference. Regenerate
ldmstm.md for Linaro 4.5 after update arm-ldmstm.ml a little bit.
Regressions goes away!
No speed improvement, but code is smaller by 0.2% in EEMBC. Still
prefer to merge to Linaro 4.5.
ocaml is an interesting language, but not easy to learn and read in vim.
* Some discussion on Linaro development process.
* My regrename improvement patch (re. LP:633243). Communicate with
Eric Botcazou back and forth, but current patch is still too
target-dependent to him, as a Middle-End maintainer. Still need some
improvements.
* Build FSF GCC trunk. CLoog requirement in configure is wrong, revert
configure to previous version, and then pass the version checking during
gcc configure.
Hi there. Could everyone in the toolchain working group start sending
their activity reports to this list please? Put [ACTIVITY] at the
start of the subject line so that they can be filtered.
Ta,
-- Michael
Hi there. Attached are the status reports from the Toolchain WG
members for last week.
-- Michael
Ken Werner --
Hi Michael,
* got access to the internal wiki/calendar/email :)
* continued to setup the borrowed vexpress board
* upgraded to the Linaro 10.11 release
* encountered various issues until I found that the /etc/hosts is empty
(#674090)
* learned that the SD card issue is a known problem (#632798)
* the network interface sometimes dies if stressed (Matt was able to
reproduce this)
* the disabled CONFIG_SWAP is being tracked as #672656
* sometimes the entire system hangs (when under heavy load?)
* David noticed that /proc/cpuinfo lacks neon support (but his string
benchmark/testcase ran fine)
* wondering why the kernel reports only about 800 BogoMIPS while it's around
2k on the panda board
* started to work on the atomic memory operations item
* identified the relevant GCC patches
* still looking for a good way to verify the GCC support
* posted a patch on the glibc-ports ml with regard to #643171
David A Gilbert --
I managed to get to try Ken Werner's Versatile Express board with an A9MP
tile; the shape
of the graphs matches that from the Panda, but the raw performance is down
by a factor of about
3 - I'm guessing it's clocked lower for some reason.
It confirms however that the Neon behaviour I was seeing with memset is
not Panda/OMAP4 specific;
no one has replied to my post to linaro-toolchain. It's a difficult
situation in that my fastest memset on
Beagle is with Neon, and my fastest on v9 is without Neon - what would you
select on?
I've just finished writing memchr tests and my first crack at a faster
version; I realised I could use the same
trick that I had used for strlen and it works nicely - it seems to be about
50% faster than the libc version;
I've not tested against any other versions yet.
Paul Mckenney hasn't replied yet about the OSSC stuff, but apparently
he's out travelling and back next
week; so I'll catch him then.
I tried preloading my faster memset into ghostscript, but found it was
blatantly ignoring it - I think the memset
is being called from somewhere inside libc; I managed to get xdeb to cross
build me a libc but haven't yet got my
changes into it.
My order for a USB hard drive for my beagle seems to have been delayed by
the supplier; I'm pushing this but
it's starting to be a bit of a pain.
Richard Sandiford --
== Last Week ==
* Pinged my GAS fix for Thumb PLT branches to locally-defined symbols.
Committed it to binutils trunk and 2.21 branch after approval. This
fixes the libgcc.so build failure that I was seeing with GOLD.
* Worked on a patch to fix GOLD's handling of non-function references
to weak undefined symbols. This ended up touching every backend
(i386, x86_64, ARM, Power and SPARC) and was quite invasive, so it
took a while in the end. Committed to binutils trunk after approval.
* Ran more tests, both with -marm and -mthumb. I'm getting identical
GCC test results (including gfortran and objc) for GOLD and BFD ld, so
I think we're at the stage where GOLD is a viable replacement for the
BFD linker.
== Next Week ==
* I'll start looking at the IFUNC support.
* I'll take another look at launchpad bug 665598.
Peter Maydell --
Progress:
* qemu: more cleanup of signal handler VFP patchset;
I think I just need to add iwmmx support and it's good
* qemu: VCVT: found yet another bug, did final patchset
cleanup: submitted to upstream list [8 patch series]
* qemu: submitted a trivial patch to fix a problem where
__get_user/_put_user macros had an unnecessary local var
which could clash with a var being used by the macro user
* set up a tree on git.linaro.org which we can use for
a branch to make pull requests for ARM qemu fixes
* did a rough estimate of time to do an Eagle qemu model
(6 months + testing/bug fixing time)
Issues:
* lost some time to a problem where Linux VMs stopped being
able to talk to the LDAP server; however I have a workaround
and IT are investigating
Meetings:
* toolchain, toolchain standup, pdsw-tools, PD doughnuts
Plans
- attend Meego conference in Dublin (Nov 15-18 inc travel)
http://conference2010.meego.com/
- start on qemu consolidation by upstreaming various ARMv7
correctness fixes
Ira Rosen --
Here is this week report:
1. BeagleBoard installed, now "playing" with it
2. Continued to work on auto-detection of vector size
3. Looked into mixed vector sizes
4. Learning about vld and vst instructions
It looks like I won't be able to participate in Wed calls, since I am alone
with the kids on Wednesday evenings.
Hi all,
I've hit a probable assembler bug trying to build a Thumb-2 kernel:
Trying to assemble the attached file, I get:
arch/arm/kernel/relocate_kernel.S: Assembler messages:
arch/arm/kernel/relocate_kernel.S:10: Error: invalid offset, value too
big (0xFFFFFFFFFFFFFFFC)
arch/arm/kernel/relocate_kernel.S:11: Error: invalid offset, value too
big (0xFFFFFFFFFFFFFFFC)
arch/arm/kernel/relocate_kernel.S:58: Error: invalid offset, value too
big (0xFFFFFFFFFFFFFFFC)
arch/arm/kernel/relocate_kernel.S:59: Error: invalid offset, value too
big (0xFFFFFFFFFFFFFFFC)
The code appears correct and resonable, except that there should be a
.align directive before the data words at the end of the file (but
adding this doesn't fix the error)
Assembling in ARM (i.e., without -mthumb), or deleting the .globl
lines associated with the affected target symbols, the problem goes
away.
I believe this may be already by tracked by CodeSourcery as is issue #8775 (?)
Has anyone hit this issue before? Is it fixed upstream?
Any help much appreciated.
Cheers
---Dave
Hi,
I've been looking at some basic libc routine optimisation and have a
curious problem with memset and wondered if
anyone can offer some insights.
Some graphs and links to code are on
https://wiki.linaro.org/WorkingGroups/ToolChain/Benchmarks/InitialMemset
I've written a simple memset in both a with and without Neon variety and
tested them on a Beagle(C4) and a Panda
board and I'm finding that the Neon version is faster than the non-neon
version (a bit) on the Beagle but a LOT slower on the
Panda - and I'd like to understand why it's slower than the non-neon version
- I'm guessing it's some form of cache interaction.
The graphs on that page are all generated by timing a loop that repeatedly
memsets the same area of memory; the X axis
is the size of the memset. Prior to the test loop the area is read into
cache (I came to the conclusion the A8 didn't write
allocate?). There are two variants of the graphs - absolute in MB/s on Y,
and a relative set (below the absolute) that
are relative to the performance of the libc routines. (The ones below those
pairs are just older versions).
if you look at the top left graph on that page you can see that on the
Beagle (left) my Neon routine beats my Thumb routine
a bit (both beating libc). If you look on the top right you see the Panda
performance with my Thumb code being the fastest and generally
following libc, but the Neon code (red line) topping out at about 2.5GB/s
which is substantially below the peak of the libc and ARM code.
The core loop of the Neon code (see the bzr link for the full thing) is:
4:
subs r4,r4,#32
vst2.8 {d0,d1,d2,d3}, [ r3:256 ]!
bne 4b
while the core of the non-Neon version is:
4:
subs r4,r4,#16
stmia r3!,{r1,r5,r6,r7}
bne 4b
I've also tried vst1 and vstm in the neon loop and it still won't match the
non-Neon version.
All suggestions welcome, plus I'd appreciate if anyone can suggest which
particular limit it's hitting - does
anyone have figures for the theoretical bus and L1 and L2 write bandwidths
for a Panda (and Beagle) ?
Thanks in advance,
Dave
Hi there. I've uploaded a draft of the slides and notes for next
weeks public review at:
http://bazaar.launchpad.net/~linaro-toolchain-wg/+junk/publicreview1105/fil…
'Toolchain Public Review 11.05.odp' is a set of slides I'll talk to.
The first 15-20 minutes will go through these to describe our focus
and goals and how they tie together the blueprints and priorities.
The rest of the session will go through the current blueprints and
priorities. See:
Toolchain Blueprints (short).pdf
for the summary version and:
Toolchain Blueprints (long).pdf
for the long version. The long version is interesting if you can't
find a particular tool or technology. It may be small enough to be
called out as a single work item.
These are only a draft, but I realised I haven't shared the plans with
the rest of the group very well and Monday's meeting won't be the
best.
I'm on holiday tomorrow but feel free to send me any comments,
-- Michael
Hi,
I started to look into mixed vector sizes (in the same loop). My main reason
for this was to allow widening and narrowing instructions, that have
different vector sizes for src and dest, to work properly. My example was
widen_mult (int = short * short), I thought its implementation was not
optimal. But now that I have a working GCC mainline for ARM, I see that it
works just fine.
short ub[], uc[];
int c[];
for (i = 0; i < n; i++)
c[i] = ub[i] * ua[i];
is compiled as:
.L11:
add r1, r1, #1
vldmia r4!, {d18-d19}
cmp r5, r1
vldmia ip!, {d16-d17}
vmull.s16 q10, d18, d16
vstr d20, [r3, #-32]
vstr d21, [r3, #-24]
vmull.s16 q8, d19, d17
vstr d16, [r3, #-16]
vstr d17, [r3, #-8]
add r3, r3, #32
bhi .L11
which looks good to me at least from the vmull point of view.
Does anyone have an example when mixed vector size instructions are not used
properly?
Another reason for mixed sizes could be cases where only part of the loop
can be vectorized with the wider vectors. I don't know how common this is.
Are there any other reasons to implement mixed vector sizes? I understand
that this can be a useful feature, I am just not sure it's the most
important one.
Thanks,
Ira
I've been going through the ChangeLog for the release and am having
trouble justifying some of the changes brought in. In particular:
* -fstrict-volatile-bitfields, which is more appropriate for bare
metal/kernel code
* Cortex-M4 support
* C locale support in libstdc++-v3
The march/mcpu clean up is OK but marginal.
Our focus is time based performance on the Cortex-A series with an
implied applications over kernel/bare metal. This is a very narrow
view, but every non-performance line of code we bring in can also
bring in a bug.
Any thoughts? For those who are looking at using our toolchain, is
earlier access to other toolchain improvements interesting?
-- Michael
Hi all,
As you may or may not know, upstream GCC has now entered 'stage 3' of
it's development cycle. This will last until spring.
This means that they are only accepting bug fixes and documentation
improvements. New features and any performance improvements must wait
until GCC 4.6 branches, prior to release, and GCC 4.7 development opens.
During this process, our usual preferred work flow (upstream first) will
not work, so we'll have to do something else.
Here's my proposal:
* Create a new Launchpad branch for GCC 4.6.
* Synchronize this branch with upstream regularly
* once per week, perhaps.
* Try to get upstream approval for all new patches in the usual way
* on the understanding that they won't be applied until stage 1
* bug fixes are unaffected and may commit as usual.
* Commit all pending patches to our own 4.6 branch
* and backport them to our 4.5, branch, of course.
* Usual "no test regressions" policy applies to our own patches
* but beware regressions from merges from upstream.
* we may want to track the clean 4.6 test results for comparison
This is little different to what we do with the 4.5 release branch now.
Thoughts?
Andrew
