does that mean it's taking the difference between the centre and the downsampled texel representing the local area?
is that quicker than each output pixel looking up the vicinity (i.e. a render and a filtered downsample?)
super interesting approach
i guess since tex lookups are so heavy,
and downsample is an order fewer than per output pixel
any reasons for the terse code?
it feels like obfuscation (or other share protection)
obfuscated look is kind of essential for sizecoding (demoscene) :P got used to it...
this is mostly just a hack, to get fast and "wide-range" blur in a single pass (but didnt work in one pass here, had to re-render to make sure mipmaps are present)
ahah! that make sense
it is possible to switch on texture info to read if mipmaps available?
next more to make texturefx more modular internally? (e.g. that would in itself be a texturefx and would be embedded in here 'CreateMipMaps' which creates mip maps to specification where not already available)
generating mipmaps for a texture, which are required for this shader to work correctly
does that mean it's taking the difference between the centre and the downsampled texel representing the local area?
is that quicker than each output pixel looking up the vicinity (i.e. a render and a filtered downsample?)
super interesting approach
i guess since tex lookups are so heavy,
and downsample is an order fewer than per output pixel
any reasons for the terse code?
it feels like obfuscation (or other share protection)
obfuscated look is kind of essential for sizecoding (demoscene) :P got used to it...
this is mostly just a hack, to get fast and "wide-range" blur in a single pass (but didnt work in one pass here, had to re-render to make sure mipmaps are present)
ahah! that make sense
it is possible to switch on texture info to read if mipmaps available?
next more to make texturefx more modular internally? (e.g. that would in itself be a texturefx and would be embedded in here 'CreateMipMaps' which creates mip maps to specification where not already available)
or just wait for DX11?
lovely work of course.