fyi, carbon vm on mac intel Carbon VM 3.8.11beta5U, versus unix out of the box non-gnuified and unix out of the box gnuified with fixes to avoid crash on startup. Thanks to Todd for doing this work, I'll work with a bit more to see if we can improve the poor raw performance.
Begin forwarded message:
From: Todd Blanchard tblanchard@mac.com Date: March 31, 2006 11:33:22 PM PST To: johnmci@smalltalkconsulting.com Subject: Re: Building Unix Squeak VM On Intel Mac
Ok - if I do that:
Squeak-3.8a1 Unix VM compiled using INTERP=interp (no gnuify):
From Squeak3.8 of '5 May 2005' [latest update: #6665] on 31 March 2006 at 11:04:41 pm
Benchmark #1: 9713ms
uptime 9.7s full 0 totalling 0ms (0.0% uptime) incr 1634 totalling 345ms (4.0% uptime), avg 0.0ms tenures 4 (avg 408 GCs/tenure)
Benchmark #2: 55247ms
uptime 55.3s full 15 totalling 1,256ms (2.0% uptime), avg 84.0ms incr 29544 totalling 6,133ms (11.0% uptime), avg 0.0ms tenures 2,143 (avg 13 GCs/tenure)
Benchmark #3: 19737ms
uptime 19.8s full 1 totalling 103ms (1.0% uptime), avg 103.0ms incr 1954 totalling 571ms (3.0% uptime), avg 0.0ms tenures 122 (avg 16 GCs/tenure)
Benchmark #4: 0ms
uptime 0.0s full 0 totalling 0ms (0.0% uptime) incr 1 totalling 0ms (0.0% uptime), avg 0.0ms tenures 0
Benchmark #5: 0ms
uptime 0.0s full 0 totalling 0ms (0.0% uptime) incr 0 totalling 0ms (0.0% uptime), avg tenures 0
Benchmark #6: 8958ms
uptime 9.0s full 1 totalling 79ms (1.0% uptime), avg 79.0ms incr 2810 totalling 591ms (7.0% uptime), avg 0.0ms tenures 41 (avg 68 GCs/tenure)
Benchmark #7: 10402ms
uptime 10.4s full 0 totalling 0ms (0.0% uptime) incr 767 totalling 265ms (3.0% uptime), avg 0.0ms tenures 18 (avg 42 GCs/tenure)
Total time = 114459 milliseconds.
Carbon VM 3.8.11beta5U
From Squeak3.8 of '5 May 2005' [latest update: #6665] on 31 March 2006 at 11:09:27 pm
Benchmark #1: 7077ms
uptime 7.1s full 0 totalling 0ms (0.0% uptime) incr 1636 totalling 380ms (5.0% uptime), avg 0.0ms tenures 3 (avg 545 GCs/tenure)
Benchmark #2: 44388ms
uptime 44.4s full 15 totalling 1,336ms (3.0% uptime), avg 89.0ms incr 26721 totalling 6,170ms (14.0% uptime), avg 0.0ms tenures 2,156 (avg 12 GCs/tenure)
Benchmark #3: 15368ms
uptime 15.4s full 1 totalling 112ms (1.0% uptime), avg 112.0ms incr 1830 totalling 549ms (4.0% uptime), avg 0.0ms tenures 123 (avg 14 GCs/tenure)
Benchmark #4: 0ms
uptime 0.0s full 0 totalling 0ms (0.0% uptime) incr 1 totalling 0ms (0.0% uptime), avg 0.0ms tenures 0
Benchmark #5: 0ms
uptime 0.0s full 0 totalling 0ms (0.0% uptime) incr 0 totalling 0ms (0.0% uptime), avg tenures 0
Benchmark #6: 6947ms
uptime 7.0s full 1 totalling 86ms (1.0% uptime), avg 86.0ms incr 2785 totalling 683ms (10.0% uptime), avg 0.0ms tenures 42 (avg 66 GCs/tenure)
Benchmark #7: 2872ms
uptime 2.9s full 0 totalling 0ms (0.0% uptime) incr 701 totalling 220ms (8.0% uptime), avg 0.0ms tenures 16 (avg 43 GCs/tenure)
Total time = 79524 milliseconds.
For fun, I made the changes in the register declarations you mentioned in gnu-interp and built another copy of the Unix VM - it gets
From Squeak3.8 of '5 May 2005' [latest update: #6665] on 31 March 2006 at 11:22:34 pm
Benchmark #1: 8900ms
uptime 8.9s full 0 totalling 0ms (0.0% uptime) incr 1635 totalling 371ms (4.0% uptime), avg 0.0ms tenures 3 (avg 545 GCs/tenure)
Benchmark #2: 52675ms
uptime 52.7s full 15 totalling 1,264ms (2.0% uptime), avg 84.0ms incr 28943 totalling 6,033ms (11.0% uptime), avg 0.0ms tenures 2,148 (avg 13 GCs/tenure)
Benchmark #3: 18877ms
uptime 18.9s full 1 totalling 96ms (1.0% uptime), avg 96.0ms incr 2008 totalling 544ms (3.0% uptime), avg 0.0ms tenures 123 (avg 16 GCs/tenure)
Benchmark #4: 0ms
uptime 0.0s full 0 totalling 0ms (0.0% uptime) incr 1 totalling 0ms (0.0% uptime), avg 0.0ms tenures 0
Benchmark #5: 0ms
uptime 0.0s full 0 totalling 0ms (0.0% uptime) incr 0 totalling 0ms (0.0% uptime), avg tenures 0
Benchmark #6: 8328ms
uptime 8.3s full 1 totalling 80ms (1.0% uptime), avg 80.0ms incr 2417 totalling 512ms (6.0% uptime), avg 0.0ms tenures 41 (avg 58 GCs/tenure)
Benchmark #7: 10367ms
uptime 10.4s full 0 totalling 0ms (0.0% uptime) incr 760 totalling 261ms (3.0% uptime), avg 0.0ms tenures 18 (avg 42 GCs/tenure)
Total time = 109514 milliseconds.
On Mar 31, 2006, at 10:33 PM, John M McIntosh wrote:
Well I take a Squeak3.8-6665.image from the squeak.org site, load the benchmarks stuff from squeakMap and do MacroBenchmarks run
mmm ya error
k, so alter things a bit to avoid the error in macroBenchmark3
MacroBenchmarks>>macroBenchmark3 true ifTrue: [^self].
On 31-Mar-06, at 10:02 PM, Todd Blanchard wrote:
If its the Benchmarks package on squeakmap you're talking about, then I can't find a one that runs to completion in my 3.8 based image. Where would I find an image where the benchmarks work?
On Mar 31, 2006, at 10:41 AM, John M McIntosh wrote:
Ah yes, well the unix maintainer can figure out why it crashes, it's a compiler bug exposed I think due to a choice he made. If someone is doing unix maintenance then we could undo that choice.
I'm curious what type of tinybenchmark and macrobenchmarks numbers you get then as compared to the carbon mac vm universal.
On 31-Mar-06, at 8:00 AM, Todd Blanchard wrote:
The VM crashes on launch in the big switch statement with a memory access violation.
Of course, the Intel Macs are much faster than the PPC ones - so even with this "handicap" it seems quite a bit faster than the PPC one.
On Mar 31, 2006, at 2:32 AM, Bert Freudenberg wrote:
Am 31.03.2006 um 09:29 schrieb Todd Blanchard: > make INTERP=interp > make INTERP=interp Squeak.app
Ugh, does the gnuify script not work? This will severely effect performance.
- Bert -
--
======== John M. McIntosh johnmci@smalltalkconsulting.com 1-800-477-2659 Corporate Smalltalk Consulting Ltd. http:// www.smalltalkconsulting.com =================================================================== ========
--
====== John M. McIntosh johnmci@smalltalkconsulting.com 1-800-477-2659 Corporate Smalltalk Consulting Ltd. http:// www.smalltalkconsulting.com ===================================================================== ======
-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com 1-800-477-2659 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===
vm-dev@lists.squeakfoundation.org