Re: [Chromium-users] about multipass rendering

[Brian P. <bri...@tu...>]

rn...@ce... wrote:
> Hello Brian,
> I corrected a part of my english sentences.
> 
> Thank you very much for your quick reply.
> 
>>Can't really say without knowing more details.
> 
> I'm using both of "Pack" and "Tilesort" option of Chromium.
> And several parts of my scenegraph use two textures, one is texture mapping and another is environment mapping.
> 
> 
>>Perhaps glCopyTexImage might be the problem since that
>>involves an expensive round-trip operation if you're using 
>>the tilesort or pack SPUs.  But that's just speculation.
> 
> I agree to your opinion.
> I found a lot of glBindTexture in the trace of Chromium debugger.
> I have a question about a texture precess of Chromium. 
> Do slaves of Chromium  refer the memory of masterPC when they draw a texture? 

No.

> Can I avoid this expensive round-trip ?

The round trip occurs for glCopyTex[Sub]Image only.  Remember, with
tilesort, you have a distributed framebuffer.  To get the new texture
image we need to send a glReadPixels to each crserver to get the
relevant framebuffer region.  Then we have to send the texture image
back to the crservers with glTex[Sub]Image.  There's really no way to
avoid this.

-Brian