r4m0n changed the topic of #kspmodders to: Technical discussion ONLY (KSP related or not), for random shit-talking, join #shitshow (seriously)
VanDisaster has quit [Ping timeout: 186 seconds]
pacbard has quit [Quit: ZNC 1.6.5+deb1 - http://znc.in]
pacbard has joined #kspmodders
VanDisaster has joined #kspmodders
Severian has quit [Ping timeout: 182 seconds]
Severian has joined #kspmodders
Majiir is now known as Snoozee
fcbayerndm has quit [Quit: fcbayerndm]
GlassYuri has joined #kspmodders
m4v has quit [Ping timeout: 186 seconds]
m4v has joined #kspmodders
RandomJeb has quit [Ping timeout: 383 seconds]
Raidernick has quit [Read error: Connection reset by peer]
Raidernick has joined #kspmodders
<Orum>
"subscripted access is not allowed for opencl vectors" why not?! D:
<Orum>
...time to wade into dangerous pointer land :[
<angavrilov>
well, "opencl vectors" seem to be the simd types, not actual vectors
<Orum>
still, one could unload those into an array, is there some function for that?
<angavrilov>
I have no idea about opencl actually, I just googled the error message
<Orum>
yeah I've been looking :\
<angavrilov>
but it is quite obvious why this wont be allowed: the indexing must be known at compile tiome in order to put these 'vectors' in registers
<Orum>
I mean I could write about 1000 LOC to work around this but it's fugly and a horrible idea
<angavrilov>
on the gpu memory is horribly slow but you have a lot of registers, so efficiency requires keeping stuff there
<Orum>
yeah, I get that, the problem is I need to eventually sum those registers
<Orum>
which in-and-of-itself isn't hard, but I actually need to sum 'n' registers first, do something, and then sum the rest
<Orum>
problem is 'n' is kernel instance dependent due to block overlap
<angavrilov>
I only have some small experience with cuda, not opencl
<Orum>
they're similar in most respects, except portability and some naming conventions