@cmuratori @webmarco (1/2) Assuming texture data is > cache size and given typical min mem bursts are, 32~128 bytes, you could waste
-
-
Replying to @Simon_Fe1
@cmuratori@webmarco (2/2) a lot of bandwidth with unused data in the planar mode unless very careful with access patterns.1 reply 0 retweets 0 likes -
Replying to @Simon_Fe1
@Simon_Fe1@webmarco Thinking about this based on@nothings discussion, it lead me to realize something else I hadn't thought about...1 reply 0 retweets 0 likes -
Replying to @cmuratori
@Simon_Fe1@webmarco@nothings ... which is that if you have "dependent" texture reads (rgb not read if a is zero) ...1 reply 0 retweets 0 likes -
Replying to @cmuratori
@Simon_Fe1@webmarco@nothings ... it might be a performance improvement to split the dependent part into another texture.1 reply 0 retweets 0 likes -
Replying to @cmuratori
@Simon_Fe1@webmarco@nothings Not sure if that's true in practice or not, but it seems at least theoretically possible?1 reply 0 retweets 0 likes -
Replying to @cmuratori
@Simon_Fe1@webmarco@nothings Basically it becomes like "early Z" at that point, and saves the subsequent fetch.2 replies 0 retweets 0 likes -
Replying to @cmuratori
@Simon_Fe1@webmarco@nothings I think I had been thinking about GPU texel fetching wrong, I was thinking it was like the lanes.1 reply 0 retweets 0 likes -
Replying to @cmuratori
@Simon_Fe1@webmarco@nothings So I was assuming that the channels were independent, but obviously that's incorrect.1 reply 0 retweets 0 likes -
Replying to @cmuratori
@Simon_Fe1@webmarco@nothings So the channels are independent _for the computations_, but they are coupled for the texel fetch.1 reply 0 retweets 0 likes
@Simon_Fe1 @webmarco @nothings So if you need them decoupled for performance, you have to ensure that you do that manually...
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.