[not a bug]Texture lock() doubles memory?

MolokoTheMole · Post by **MolokoTheMole** » Tue Jun 17, 2008 2:30 pm

I load 7mbs of textures. When I perform a lock() on them they use 14mbs in memory. I checked to be sure that is it and it is. What's up? I'm using Irrlicht 1.3.1

hybrid · Post by **hybrid** » Mon Jun 23, 2008 9:33 am

The texture is copied into a memory location which is accessible by the CPU. So it can indeed double the memory usage. However, an unlock should free that memory region again. I'm not sure if the system will free the memory immediately, though.

MolokoTheMole · Post by **MolokoTheMole** » Wed Jun 09, 2010 10:05 am

Sorry for bumping this but I am curious.
Can I edit the textures pixels directly on the texture memory in vram? This would be mighty fast but I suspect it would need some uber skills.

hybrid · Post by **hybrid** » Wed Jun 09, 2010 4:21 pm

At least we don't have an interface for this, so no, right now it's not possible.

DtD · Post by **DtD** » Wed Jun 09, 2010 6:10 pm

From my experience, copying the texture from the vram to the ram and vice-versa is very fast, I wouldn't worry about it too much.

loki1985 · Post by **loki1985** » Wed Jun 09, 2010 6:22 pm

MolokoTheMole wrote:Can I edit the textures pixels directly on the texture memory in vram?

can't that be done using shaders? a bit non-intuitive maybe, but probably much faster than letting the CPU do it, since that would require a roundtrip of the data on the graphics bus.