Smalltalk › Squeak › Squeak VM

Re: [Pharo-project] newDelta >0 failed in sqUnixMemory

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

4 messages Options

johnmci

Re: [Pharo-project] newDelta >0 failed in sqUnixMemory

Yes, I can't recall it was what 50% of the performance of a 3.x
compile out of the box, and after *much*
tweaking of compiler flags it was mmm just 20% slower.. Well
certainly the performance impact was measured in 10 of percent.

On 30-Jan-09, at 4:33 AM, Michael Rueger wrote:

>
> John McIntosh and I actually spent some time a few ago compiling VMs
> on
> Windows and MacOS with the newest GCCs and ran into a bunch of
> problems,
> especially concerning performance. I had to put the work on the back
> burner for now, but switching to newer compilers will take a bit of
> extra work at some point.
>
> Michael

--
=
=
=
========================================================================
John M. McIntosh <[hidden email]>
Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com
=
=
=
========================================================================

Andreas.Raab

Re: Re: [Pharo-project] newDelta >0 failed in sqUnixMemory

John M McIntosh wrote:
> Yes, I can't recall it was what 50% of the performance of a 3.x compile
> out of the box, and after *much*
> tweaking of compiler flags it was mmm just 20% slower.. Well certainly
> the performance impact was measured in 10 of percent.

So what do you use today? gcc 4? And how does it stack up
performance-wise against a Windows VM on the same box? I personally
wouldn't mind moving forward but the results in the past were so
godawful that I haven't tried in a long time.

Cheers,
- Andreas

> On 30-Jan-09, at 4:33 AM, Michael Rueger wrote:
>>
>> John McIntosh and I actually spent some time a few ago compiling VMs on
>> Windows and MacOS with the newest GCCs and ran into a bunch of problems,
>> especially concerning performance. I had to put the work on the back
>> burner for now, but switching to newer compilers will take a bit of
>> extra work at some point.
>>
>> Michael
>
> --
> ===========================================================================
> John M. McIntosh <[hidden email]>
> Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com
> ===========================================================================
>
>
>

johnmci

Re: Re: [Pharo-project] newDelta >0 failed in sqUnixMemory

mmm, let me clarify. I currently use Apple's GCC 4.0.1 <i686-apple-
darwin9-gcc-4.0.1>
that works fine for macintel, 3.3 is required for decent performance
on powerpc.

The version that I worked with Michael on was the LLVM GCC 4.2 and or
GCC 4.2. Michael I believe tried GCC 4.3
Michael said:
"after managing to build the VM with the recommended setup I tried
with newer GCCs (3.45 and 4.3), both significantly slower.
GCC 3.45 about 50%, GCC 4.3 about 80% of GCC 2.95."

With 4.0.1 the compiler optimizations tweaks in the macintosh carbon
VM build give it a good 10%? better than the defaults optimizations.
However that presupposes -mtune=prescott -march=pentium-m which one
can get away with on the macintosh platform, versus the
bazillion cpu choices & clones on the Wintel side.

On 30-Jan-09, at 9:37 AM, Andreas Raab wrote:

> John M McIntosh wrote:
>> Yes, I can't recall it was what 50% of the performance of a 3.x
>> compile out of the box, and after *much*
>> tweaking of compiler flags it was mmm just 20% slower.. Well
>> certainly the performance impact was measured in 10 of percent.
>
> So what do you use today? gcc 4? And how does it stack up
> performance-wise against a Windows VM on the same box? I personally
> wouldn't mind moving forward but the results in the past were so
> godawful that I haven't tried in a long time.
>
> Cheers,
> - Andreas
>
>> On 30-Jan-09, at 4:33 AM, Michael Rueger wrote:
>>>
>>> John McIntosh and I actually spent some time a few ago compiling
>>> VMs on
>>> Windows and MacOS with the newest GCCs and ran into a bunch of
>>> problems,
>>> especially concerning performance. I had to put the work on the back
>>> burner for now, but switching to newer compilers will take a bit of
>>> extra work at some point.
>>>
>>> Michael
>> --
>> =
>> =
>> =
>> =
>> =
>> =
>> =====================================================================
>> John M. McIntosh <[hidden email]>
>> Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com
>> =
>> =
>> =
>> =
>> =
>> =
>> =====================================================================

Andreas.Raab

Re: Re: [Pharo-project] newDelta >0 failed in sqUnixMemory

John M McIntosh wrote:
> With 4.0.1 the compiler optimizations tweaks in the macintosh carbon VM
> build give it a good 10%? better than the defaults optimizations.
> However that presupposes -mtune=prescott -march=pentium-m which one
> can get away with on the macintosh platform, versus the
> bazillion cpu choices & clones on the Wintel side.

I think it would be okay to optimize by default for the more modern CPU
variants. One thing that would be interesting is if you could run
benchmark comparison of your latest Mac production VMs against a Windows
VM running under Parallels or (even better) Bootcamp on the same hardware.

Cheers,
- Andreas

> On 30-Jan-09, at 9:37 AM, Andreas Raab wrote:
>
>> John M McIntosh wrote:
>>> Yes, I can't recall it was what 50% of the performance of a 3.x
>>> compile out of the box, and after *much*
>>> tweaking of compiler flags it was mmm just 20% slower.. Well
>>> certainly the performance impact was measured in 10 of percent.
>>
>> So what do you use today? gcc 4? And how does it stack up
>> performance-wise against a Windows VM on the same box? I personally
>> wouldn't mind moving forward but the results in the past were so
>> godawful that I haven't tried in a long time.
>>
>> Cheers,
>> - Andreas
>>
>>> On 30-Jan-09, at 4:33 AM, Michael Rueger wrote:
>>>>
>>>> John McIntosh and I actually spent some time a few ago compiling VMs on
>>>> Windows and MacOS with the newest GCCs and ran into a bunch of
>>>> problems,
>>>> especially concerning performance. I had to put the work on the back
>>>> burner for now, but switching to newer compilers will take a bit of
>>>> extra work at some point.
>>>>
>>>> Michael
>>> --
>>> ===========================================================================
>>>
>>> John M. McIntosh <[hidden email]>
>>> Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com
>>> ===========================================================================
>>>
>
> --
> ===========================================================================
> John M. McIntosh <[hidden email]>
> Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com
> ===========================================================================
>
>
>