mip mapping question (direct X)
Author
something im not too clear on: how is mipmapping generally handled by directX? is it on a per-triangle basis or a per-draw call basis?
also, ive read auto-mip map generation is available on some cards but not most, so i'm not sure if i want to use it - does it take up a lot of CPU or something to do? I'm also not sure if its something that has to do manually or not to avoid the card doing it; i assume D3DUSAGE_AUTOGENMIPMAP uses the card for automatic mip-mapping, but im not sure about IDirect3DBaseTexture9::GenerateMipSubLevels. Then again it might just be dated information.
Mip-mapping is a rasterization level feature and will be done on a per-pixel level as far as I understand it. In reality there may be various hacks on GPU's to determine coefficients per-primitive or per-block (n*n pixels) but conceptually you'd be fine assuming it's a per-pixel operation.
The resource managers will use algorithms to determine which mip levels are available in VRAM at the time rasterization occurs. I can't actually remember the details here, but they're definitely much coarser decisions based on guesstimates as to actual need versus available VRAM and contention ratios.
Auto-generated mip-maps are useful for dynamically changing resource - just as a render target where you only want to draw to the top-level but want the device to down-sample it for you.
For regular static image data you should stick with load-time or preferrably offline mip-map generation depending on your toolchain and storage requirements.
hth
Jack
The resource managers will use algorithms to determine which mip levels are available in VRAM at the time rasterization occurs. I can't actually remember the details here, but they're definitely much coarser decisions based on guesstimates as to actual need versus available VRAM and contention ratios.
Auto-generated mip-maps are useful for dynamically changing resource - just as a render target where you only want to draw to the top-level but want the device to down-sample it for you.
For regular static image data you should stick with load-time or preferrably offline mip-map generation depending on your toolchain and storage requirements.
hth
Jack
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement