Thread: [Sablevm-developer] Report: SableVM on various architectures
Brought to you by:
egagnon
Menu
▾
▴
sablevm-developer
|
From: Grzegorz B. P. <ga...@de...> - 2004-02-02 07:00:00
|
Hi all!
During the weekend I put heavy burden on Debian's porting machines
to find out on which architectures SableVM works and to what extent.
SUMMARY:
Fully and surely working inlined: i386, ia64, ppc, sparc
Fully and surely working direct: s390
Needs final test to declare (hopefully) working inlined: m68k
To be tested yet (quickly, but they are promising): arm, hppa
Breaks running BTF on any engine: alpha
MORE DETAILED INFO:
* alpha - broken for anything non-trivial, inlined doesn't work.
- HelloWorld works in non-inlined.
- BTF shows serious problems w/ integer operations in non-inlined.
- In Inlinability Testing Mode even slow versions of two mod
instructions cause errors.
(diagnosis: weird, need to look at assembly level what's going on.
non-inlined probably would be not too hard to fix.)
* arm - jikes brokenness and GNU CP broken --without-java param
prevented me from testing it. Put as "try" on "Soon TODO".
* hppa - libffi didn't make it into .deb yet, got binaries today
and will try it as soon as I merge tausq's hppa changes into
staging (I thought it was already done!)
* i386 - everything works
* ia64 - everything works (now)
- updated inlinability info
- corner cases handling added
* m68k - everything should work (will be known until monday noon)
- m68ks are sloooow - 2-3h to compile one flavor of SableVM!
- updated inlinability info
- inlined-testing mode shows no errors in inlinability nor
any corner cases problems that BTF checks
- it is compiling inlined and inlined-nosig now
(diagnosis: good chances to have it 100% working)
* powerpc - everything works (but I couldn't check)
- as I don't have ppc machine I was unable to run tests on ppc,
but everything worked and I don't think anything got broken since.
(just in case requested installation of sablevm build-deps on
debian ppc porting machine, should be available soon)
* s390 - non-inlined works perfectly (now), serious problems in inlined
- fixed all corner cases where BTF revealed problems
- inlined seems severly broken, ex. GET/PUT-FIELD/STATIC *fast*
versions fail! (similarly as some of them on alpha)
(diagnosis: weird, need to look at assembly level what's going on.)
* sparc - everything works
I'll be sending updates. Cheers,
Grzegorz B. Prokopski
PS: To make such work as easy as possible in future I created a set
of scripts that make it easy to run completly unattended tests
of many flavors of SableVM on many machines at the same time.
To see what I mean go to:
http://gadek.debian.net/testing/massive.png
this was all ran by a single "enter" ;-)
(...these evil shell scripts - I'll learn some more perl next time)
In future this should also allow to easily test SableVM every few
weeks, whether anything on any architecture got broken/unbroken.
Running ex. mauve tests on all these arches also sounds interesting.
--
Grzegorz B. Prokopski <ga...@de...>
Debian GNU/Linux http://www.debian.org
SableVM - LGPLed JVM http://www.sablevm.org
Why SableVM ?!? http://devel.sablevm.org/wiki/WhySableVM
|
|
From: David <db...@cs...> - 2004-02-02 21:11:17
|
On Mon, Feb 02, 2004 at 01:59:52AM -0500, Grzegorz B. Prokopski wrote: > Hi all! >=20 > During the weekend I put heavy burden on Debian's porting machines > to find out on which architectures SableVM works and to what extent. >=20 > SUMMARY: >=20 > Fully and surely working inlined: i386, ia64, ppc, sparc > Fully and surely working direct: s390 > Needs final test to declare (hopefully) working inlined: m68k > To be tested yet (quickly, but they are promising): arm, hppa > Breaks running BTF on any engine: alpha >=20 Good work! ( Note: Code you added in staging cause warnings: instructions_preparation_direct_threaded.c:12822: warning: garbage at end of `#ifdef' argument if #ifdef _SABLEVM_SIGNALS_FOR_EXCEPTIONS && !defined(__s390__) is changed to #if defined(_SABLEVM_SIGNALS_FOR_EXCEPTIONS) && !defined(__s390__) it should work. ) David --- David B=E9langer Graduate Student School of Computer Science McGill University Office: MC226 Web page: http://www.cs.mcgill.ca/~dbelan2/ Public key: http://www.cs.mcgill.ca/~dbelan2/public_key.txt |
|
From: Grzegorz B. P. <ga...@de...> - 2004-02-03 06:22:15
|
Update: ARM also breaks running any engine, it looks just like on Alpha. HPPA is having problems w/ inlined [*], direct works fully. M68K works fully, inlined inlcluding. So the list of arches w/ fully working inlined is now: i386, ia64, ppc, sparc, m68k Fully working direct: hppa, s390 I've just finished updating my testing scripts to run some benchmarks and I am going to throw them at these architectures (w/ exception of ppc as I don't have any and I am not sure what about m68k, as the compilation of sablevm takes 2-3 hours, then running benchmarks may take a week... or two ;-) [*] I have to search for this mail tomorrow but I guess that while working w/ tausq we got inlined to at least run HelloWorld. And now it severly segfaults :-( That'd be the first thing to check. Alpha, ARM and s390 all seem to have quite similar symptoms. But I have to go down the machine code to find out what it is. Looks like I've just made plans for this weekend... I also found out that libffi IS available on MIPS (not 100% sure about MIPSEL) though it doesn't have all the features. But it should be good enough for what SableVM needs. Debian-mips ML contacted, no reply yet. Cheers, Grzegorz B. Prokopski -- Grzegorz B. Prokopski <ga...@de...> Debian GNU/Linux http://www.debian.org SableVM - LGPLed JVM http://www.sablevm.org Why SableVM ?!? http://devel.sablevm.org/wiki/WhySableVM |
|
From: David <db...@cs...> - 2004-02-03 15:08:12
|
On Tue, Feb 03, 2004 at 01:22:08AM -0500, Grzegorz B. Prokopski wrote: > I've just finished updating my testing scripts to run some benchmarks > and I am going to throw them at these architectures (w/ exception of > ppc as I don't have any and I am not sure what about m68k, as the > compilation of sablevm takes 2-3 hours, then running benchmarks may > take a week... or two ;-) Hi Greg, You can be pretty confident that it works fine on Linux/powerpc, I am using it as my main development platforms. Runs all the SPECjvm98 benchmarks no problem. David --- David B=E9langer Graduate Student School of Computer Science McGill University Office: MC226 Web page: http://www.cs.mcgill.ca/~dbelan2/ Public key: http://www.cs.mcgill.ca/~dbelan2/public_key.txt |
|
From: Grzegorz B. P. <ga...@de...> - 2004-02-09 08:40:14
|
Hi all! Another update: alpha works w/ inlined now. We have 6 arches w/ inlined! Remaining to get (inlined) working: s390 - haven't managed to get around it yet, direct works, arm - broken w/ any engine, but port machine is unreachable currently hppa - direct works, tausq offered help in trying inlined on his machine (Debian port machine behaves strangely, and we HAD it working on tausq's hppa) mips/mipsel - At first I didn't know that there's libffi for them, and currently the port machines are unaccessible Next victims / goals: arm (if machine is back), s390. About alpha: The arch is weird, I had to detect some magic value 128 of info->si_code (in signal handler) that I don't find any #define for in /usr/include/*. We get this sig code on division by 0. I guess this may be a kernel bug. PUTSTATIC_*/PUTFIELD_* aren't inlinable at all (I haven't looked at the assembly, but I think it'd be interesting to take a look), while GETFIELD_* are inlinable only w/ signals. I think it's the only such arch so far. Besides these, only LDC_STRING, FREM, DREM are noninlinable. svn diff -r 1554:1555 svn+ssh://svn.sablevm.org/sablevm/branches/staging will tell you the story in details. Cheers, Grzegorz B. Prokopski -- Grzegorz B. Prokopski <ga...@de...> Debian GNU/Linux http://www.debian.org SableVM - LGPLed JVM http://www.sablevm.org Why SableVM ?!? http://devel.sablevm.org/wiki/WhySableVM |
×
Want the latest updates on software, tech news, and AI?
Get latest updates about software, tech news, and AI from SourceForge directly in your inbox once a month.
Thanks for helping keep SourceForge clean.
X