Hi, As part of my new role, I'm still trying to compile and understand all the building process... and testing with a 5.x produced I found that the tiny benchmarks for my vm and for the 5.x series was significantly worse than it's predecessors (4.x). The difference is significative (32 million send in 5.x/44 million send in 4.x). Trying several optimizations with John, he found that changing the declaration:
struct foo { ... } fum;
to
static struct foo { ... } fum;
speeds the execution, up to 48 million send, which is very reasonable.
So, I want to change the default definition of fum variable (to "static struct foo"). What do you think?
Cheers, Esteban
On 2 December 2010 23:30, Esteban Lorenzano estebanlm@gmail.com wrote:
Hi, As part of my new role, I'm still trying to compile and understand all the building process... and testing with a 5.x produced I found that the tiny benchmarks for my vm and for the 5.x series was significantly worse than it's predecessors (4.x). The difference is significative (32 million send in 5.x/44 million send in 4.x). Trying several optimizations with John, he found that changing the declaration:
struct foo { ... } fum;
to
static struct foo { ... } fum;
speeds the execution, up to 48 million send, which is very reasonable.
So, I want to change the default definition of fum variable (to "static struct foo"). What do you think?
+1 it could change depending on compiler.. some could compile 'static' to be more slower than auto. and that's what we should live with.
Cheers, Esteban
On Fri, Dec 03, 2010 at 12:12:20AM +0100, Igor Stasenko wrote:
On 2 December 2010 23:30, Esteban Lorenzano estebanlm@gmail.com wrote:
Hi, As part of my new role, I'm still trying to compile and understand all the building process... and testing with a 5.x produced I found that the tiny benchmarks for my vm and for the 5.x series was significantly worse than it's predecessors (4.x). The difference is significative (32 million send in 5.x/44 million send in 4.x). Trying several optimizations with John, he found that changing the declaration:
struct foo { ... } fum;
to
static struct foo { ... } fum;
speeds the execution, up to 48 million send, ??which is very reasonable.
So, I want to change the default definition of fum variable (to "static struct foo"). What do you think?
This change sounds fine to me. Esteban, please feel free to make the update to VMMaker on SqueakSource (or I will be happy to do it on your behalf if you prefer, just let me know). Eliot will probably want to make the corresponding change in the oscog branch of VMMaker on SqueakSource.
+1 it could change depending on compiler.. some could compile 'static' to be more slower than auto. and that's what we should live with.
If it turns out to be an issue on some platforms, we can alway change it later, maybe using a macro that would be set at compile time. So no reason not to make the change now.
Thanks!
Dave
On 12/2/2010 3:33 PM, David T. Lewis wrote:
So, I want to change the default definition of fum variable (to "static struct foo"). What do you think?
While you're at it, could you *please* give these variables reasonable names? I mean 'struct foo fum'? WTF?
Cheers, - Andreas
It was "Fe Fi Fo(o) Fum" and building things (aka standing) on the shoulders of Giants for this minor change to all the VM work that came before.
On 2010-12-02, at 3:54 PM, Andreas Raab wrote:
On 12/2/2010 3:33 PM, David T. Lewis wrote:
So, I want to change the default definition of fum variable (to "static struct foo"). What do you think?
While you're at it, could you *please* give these variables reasonable names? I mean 'struct foo fum'? WTF?
Cheers,
- Andreas
-- =========================================================================== John M. McIntosh johnmci@smalltalkconsulting.com Twitter: squeaker68882 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ===========================================================================
On Fri, Dec 03, 2010 at 11:22:39AM +0100, Craig Latta wrote:
It was "Fe Fi Fo(o) Fum" and building things (aka standing) on the shoulders of Giants for this minor change to all the VM work that came before.
Cute, but inexcusably horrid. :) Reasonable names, please!
Would this be reasonable?
static struct localVars { ... } localVars; struct localVars * interpVars = &localVars;
(usqInt) interpVars->memoryLimit
Dave
On 12/3/2010 5:25 AM, David T. Lewis wrote:
On Fri, Dec 03, 2010 at 11:22:39AM +0100, Craig Latta wrote:
It was "Fe Fi Fo(o) Fum" and building things (aka standing) on the shoulders of Giants for this minor change to all the VM work that came before.
Cute, but inexcusably horrid. :) Reasonable names, please!Would this be reasonable?
static struct localVars { ... } localVars; struct localVars * interpVars =&localVars;
(usqInt) interpVars->memoryLimit
+1.
Cheers, - Andreas
On Thu, Dec 02, 2010 at 03:54:30PM -0800, Andreas Raab wrote:
On 12/2/2010 3:33 PM, David T. Lewis wrote:
So, I want to change the default definition of fum variable (to "static struct foo"). What do you think?
While you're at it, could you *please* give these variables reasonable names? I mean 'struct foo fum'? WTF?
Andreas,
Esteban was asking if it is OK to change the declaration from "struct foo" to "static struct foo", and the answer is yes, he should proceed with that change (or I'll commit it to VMMaker if Esteban prefers to handle it that way).
With respect to changing the names to something less whimsical, that's a fair point. However, it effects two classes and a dozen methods in VMMaker, as well as the external gnuify.awk script that has to be in sync with these. Given that we have two significant branches of VMMaker and two branches of the platform sources in need of convergence, this may not be a good time to introduce additional changes that require coordination. Can we do it a few months from now?
Change set attached, but I vote for waiting until next year to apply it.
Dave
On 12/2/2010 9:07 PM, David T. Lewis wrote:
With respect to changing the names to something less whimsical, that's a fair point. However, it effects two classes and a dozen methods in VMMaker, as well as the external gnuify.awk script that has to be in sync with these. Given that we have two significant branches of VMMaker and two branches of the platform sources in need of convergence, this may not be a good time to introduce additional changes that require coordination. Can we do it a few months from now?
Sure can. I wasn't aware that it affected that much stuff; I had hoped that it was a simple renaming.
Cheers, - Andreas
Hi David, please, go ahead... my vmmaker image right now is polluted with alien stuff (which overrides things in vmmaker), and it will be faster if you do it :)
Cheers, Esteban
El 02/12/2010, a las 8:33p.m., David T. Lewis escribió:
On Fri, Dec 03, 2010 at 12:12:20AM +0100, Igor Stasenko wrote:
On 2 December 2010 23:30, Esteban Lorenzano estebanlm@gmail.com wrote:
Hi, As part of my new role, I'm still trying to compile and understand all the building process... and testing with a 5.x produced I found that the tiny benchmarks for my vm and for the 5.x series was significantly worse than it's predecessors (4.x). The difference is significative (32 million send in 5.x/44 million send in 4.x). Trying several optimizations with John, he found that changing the declaration:
struct foo { ... } fum;
to
static struct foo { ... } fum;
speeds the execution, up to 48 million send, ??which is very reasonable.
So, I want to change the default definition of fum variable (to "static struct foo"). What do you think?
This change sounds fine to me. Esteban, please feel free to make the update to VMMaker on SqueakSource (or I will be happy to do it on your behalf if you prefer, just let me know). Eliot will probably want to make the corresponding change in the oscog branch of VMMaker on SqueakSource.
+1 it could change depending on compiler.. some could compile 'static' to be more slower than auto. and that's what we should live with.
If it turns out to be an issue on some platforms, we can alway change it later, maybe using a macro that would be set at compile time. So no reason not to make the change now.
Thanks!
Dave
vm-dev@lists.squeakfoundation.org
-
Andreas Raab -
Craig Latta -
David T. Lewis -
Esteban Lorenzano -
Igor Stasenko -
John M McIntosh