[GEM-dev] frambuffer and Vertex Texture Fetching

[cyrille henry]

hello,

the good new is that things still can work as it was.
just replace rgb32 with rgb.
the old version does not use the same primitive, so i's possible there is more vertex now. this can explain the 70->130% diference.

we can investigate, i am not sure your card is able to use this acceleration.

with a nvidia 7700, this patch use only few %, so the change is really important!

cyrille
 
marius schebella a écrit :
> hi,
> when I click on type byte, format rgb32 then I get this printout
> [gemframebuffer]: type is BYTE, 5121
> [gemframebuffer]: format is GL_RGB_FLOAT32_ATI, 32992
> but it starts eating up all my cpu.
> with the old version I cud run dim 512 512 at ~70%
> with the new version I am +130%
> this is on os x 10.5.2 with ati X1600.
> marius.
> 
> cyrille henry wrote:
>> hello,
>> i'm currently investigating the use of framebuffer to create a texture 
>> in order to use it as a vertex displacement.
>> this has already been discuss in this list, but new stuff in 
>> framebuffer makes me try again.
>>
>> i'm specially motived because it's the last things that prevent me 
>> trying physical modeling in glsl.
>>
>>
>>
>> having a look at framebuffer, i noticed strange things :
>> -changing type or format does not init the framebuffer, you have to 
>> change dim, or rectangle (mode) in order to init it.
>> (i don't know why)
>>
>> -in void gemframebuffer :: typeMess(char* type), should :
>>
>>     else if (!strcmp(type, "FLOAT")){
>>       post("type is GL_FLOAT, %d",m_type);
>>       m_type = GL_FLOAT;
>>       return;
>>
>> be replaced by :     else if (!strcmp(type, "FLOAT")){
>>       m_type = GL_FLOAT;
>>       post("type is GL_FLOAT, %d",m_type);
>>       return;
>> ???
>> (it's also the same in formatMess)
>>
>> anyway. here are the good news :
>> i add (and commit) a RGB32 format corresponding to        
>> m_internalformat = GL_RGB_FLOAT32_ATI;
>> #ifdef __APPLE__
>>        m_format = GL_BGR;
>> #else        m_format = GL_RGB;
>> #endif
>> (i  don't know if apple definition is ok).
>>
>> but the internal format GL_RGB_FLOAT32_ATI was mandatory in order to 
>> have an efficient implementation of vertex texture fetching.
>>
>> so here is an example. it still need to be cleaned, but i'd like to 
>> know if it work on other computer.
>> (here, it's about 50 time faster than previous example i send to this 
>> list)
>>
>> please update/compile and test!
>>
>> cyrille
>>
>>
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> GEM-dev mailing list
>> GEM-dev at iem.at
>> http://lists.puredata.info/listinfo/gem-dev
> 
> 
>