GPGPU sampler

Hi,

Right, try again. This time the changes file included:

http://celeriac.net/gpgpu25122009.zip

(17MB - thanks Josh)

-hope it works for you :-)

Cheers,
Matthew

Better, but I'm still having trouble with the changes file.  When trying to browse any of the new-ish code (including the Cobalt code), I get "MessageNotUnderstood: UndefinedObject>>isSeparator" because the image is trying to read data from past the end of the changes file.  The changes file is only 229KB, which is definitely on the small side.  

Josh Gargus wrote:
>
> Did you perhaps condense changes and then not save the image?  (just
> guessing)


that's exactly what i did, after looking up .changes files (i don't know
what it's for). i thought it sounded like a good idea to save space.

- will try again, but without that step...

Matthew Chadwick wrote:

Josh Gargus wrote:

Did you perhaps condense changes and then not save the image?  (just guessing)

that's exactly what i did, after looking up .changes files (i don't know what it's for). i thought it sounded like a good idea to save space.

- will try again, but without that step...

ok new one uploading now, should be there in about 20 mins...same url as before

Matthew Chadwick wrote:

Matthew Chadwick wrote:

Josh Gargus wrote:

Did you perhaps condense changes and then not save the image?  (just guessing)

that's exactly what i did, after looking up .changes files (i don't know what it's for). i thought it sounded like a good idea to save space.

- will try again, but without that step...

uploaded a new image + changes (i think .changes file is ok this time)
http://celeriac.net/gpgpu25122009.zip

also added just the montcello files: http://celeriac.net/gpgl.zip

There's a terse explanation at:  http://wiki.squeak.org/squeak/49 ... I'll elaborate.

The .changes file is a change log of code that has changed in the image.  Whenever you change a method, that code is appended to the .changes file (so, for example, if you crash Squeak then your changes aren't lost forever... they can be recovered from the .changes file... see http://wiki.squeak.org/squeak/2168).  The actual source code for each method isn't actually stored in the image.  Instead, each method contains an index into the .changes file to look up the source code.  Some implications for our situation:

- if there is no .changes file, then there is no source code.  However, Squeak can decompile the bytecodes in a CompiledMethod object to recover the source code.  Awesome!  Except that it doesn't know the names of temporary variables and method arguments, so instead of meaningful names, you see "t1", "t2", "t3", etc.

- if you've been programming in an image for a while, there are probably many methods that you have changed several times.  All of these versions are in the .changes file.  Condensing changes rewrites the .changes file, keeping only the most recent version of each changed method; the resulting changes file is much smaller.  At the same time, the methods in the image must be updated with new indices into the .changes file in order to point to the new location of their source code.  If you condense changes but don't save the image, then your image will try to look for the source code in the old location in the new, smaller .changes file.  Often, the old location will be past the end of the new file, leading to the error that I mentioned.

Cheers,
Josh





On Dec 25, 2009, at 12:44 PM, Matthew Chadwick wrote:

> Josh Gargus wrote:
>>
>> Did you perhaps condense changes and then not save the image?  (just guessing)
>
>
> that's exactly what i did, after looking up .changes files (i don't know what it's for). i thought it sounded like a good idea to save space.
>
> - will try again, but without that step...
>

Matthew Chadwick wrote:

Matthew Chadwick wrote:

Josh Gargus wrote:

Did you perhaps condense changes and then not save the image?  (just guessing)

that's exactly what i did, after looking up .changes files (i don't know what it's for). i thought it sounded like a good idea to save space.

- will try again, but without that step...

uploaded a new image + changes (i think .changes file is ok this time)
http://celeriac.net/gpgpu25122009.zip

also added just the montcello files: http://celeriac.net/gpgl.zip

Do you have any information on the versions/settings for OpenGL and graphics drivers that are required to run this?

It uses features from OpenGL 1.5 so any card from the last couple of years should run it. Windows shouldn't be an issue, nor drivers (just make sure they're the latest I guess). It could be that your machines have crappy integrated graphics - try it on a machine with a recent discrete card & it should run.

Are you getting any errors BTW ?

Do you have any information on the versions/settings for OpenGL and graphics drivers that are required to run this?

It uses features from OpenGL 1.5 so any card from the last couple of years should run it. Windows shouldn't be an issue, nor drivers (just make sure they're the latest I guess). It could be that your machines have crappy integrated graphics - try it on a machine with a recent discrete card & it should run.

Are you getting any errors BTW ?

yeh I've not had much joy with intel chipsets either. the gpu needs FBO and PBO, and 128/256 fp bit precision.

BTW when you do find a machine that runs it, there's an error in the Mandelbrot example - i used mac option key when i should have used yellowbuttondown (you'll see what i mean - correct it in debug)

Baveco, Hans wrote:

Thanks for the quick reply! The machine I am using now apparently supports only openGL 1.4 (says OpenGL Extensions Viewer) after updating to the latest driver. Probably this is a crappy Intel thing:-).

 

Starting the Mandelbrot example leads to the MessageNotUnderstood: GPGPU>>handleError:

where handleError: is send from GPGPU>>checkForFBO

 

checkForFBO
 "(adapted from David Fraught's code) Check for FBO support"
 "this is a pretty minimal check"
 | status |

 

 status := ogl extensions includes: #'GL_EXT_framebuffer_object'.
 status ifFalse:  [ self handleError: 'FBO is not supported\Try running:\GPUInfo new chkGLSLSupport' withCRs. ]

 

Clearly, 'GL_EXT_framebuffer_object' is not available on my system. (handleError: is not implemented in GPGPU, nor is the class GPUInfo available; maybe these were part of David's code?)

 

Hans

---

From: Matthew Chadwick [[hidden email]]
Sent: vrijdag 29 januari 2010 10:52
To: Baveco, Hans
Cc: [hidden email]
Subject: Re: [croquet-dev] GPGPU sampler

Do you have any information on the versions/settings for OpenGL and graphics drivers that are required to run this?

It uses features from OpenGL 1.5 so any card from the last couple of years should run it. Windows shouldn't be an issue, nor drivers (just make sure they're the latest I guess). It could be that your machines have crappy integrated graphics - try it on a machine with a recent discrete card & it should run.

Are you getting any errors BTW ?

Matthew Chadwick wrote:

128/256 fp bit precision.

sorry that should have been 128/256 bit precision - but that's over the whole RGBA vector, meaning 32/64bit floats - it usually says somewhere in the card's spec.