#
✕ nick:karolherbst
#asahi-gpu · 2022-10-02 17:02
karolherbstuhhh that "On jerks and institutional acceptance of them" part reminds me of some...
#asahi-gpu · 2022-10-02 16:57
karolherbstin the end what works decides what's the best thing, not some abstract reasoning
#asahi-gpu · 2022-10-02 16:56
karolherbstnot even wanting to say that it's a bad thing, because most of the time "technical points" are just mindsets/culture put into something abstract and ticked of with a "yep, that's how it must be"
#asahi-gpu · 2022-10-02 16:53
karolherbstor well.. the more relevant reason why things turned out how they did
#asahi-gpu · 2022-10-02 16:52
karolherbstcoporate culture is more important than "technical reasoning" for most things
#asahi-gpu · 2022-10-02 16:52
karolherbstyeah...
#asahi-gpu · 2022-10-02 16:52
karolherbstand never will
#asahi-gpu · 2022-10-02 16:52
karolherbstif you can use avx/sse all the way, it can be super fast, but normal code doesn't give you that
#asahi-gpu · 2022-10-02 16:52
karolherbstthe main problem is, that for different problems you have to use different mindeset. Normal code written for C doesn't optimize well, and normal C devs never had to think much about data paralization and all of that
#asahi-gpu · 2022-10-02 16:51
karolherbstthough I am not sure myself if I read the full thing, might have stopped somewhere in the middle. The first part is in itself already a good story about intel here :D
#asahi-gpu · 2022-10-02 16:45
karolherbstit's fun, isn't it?
#asahi-gpu · 2022-10-02 15:51
karolherbstalyssa: sure.. but I could also imagine that nvidia or AMD are more aggressive about killing of stages, but maybe they are simply less concerned about space and keep their stuff. At least for nvidia I know that they already just use L1 or L2 cache for all those stage special memory things
#asahi-gpu · 2022-10-02 15:33
karolherbstI am actually wondering if we'll see more hardware like that even in the discrete space, where stages will be more like that
#asahi-gpu · 2022-10-02 15:11
karolherbstdoes something like helper invocations even exist?
#asahi-gpu · 2022-10-02 15:10
karolherbstmaybe I should try to use metal once and then I have data and can decide to hate or love it or something
#asahi-gpu · 2022-10-02 15:09
karolherbstcompared to GL or vulkan?
#asahi-gpu · 2022-10-02 15:08
karolherbstis metal really that great/terrible how people say?
#asahi-gpu · 2022-10-02 14:54
karolherbstlet's see how that all goes
#asahi-gpu · 2022-10-02 14:54
karolherbstoh well...
#asahi-gpu · 2022-10-02 14:54
karolherbstyeah... so my plan was to try to stay awak over the night and sleep on my flight to the US
#asahi-gpu · 2022-10-02 14:51
karolherbstI had the smart idea of taking a flight at 6 am
#asahi-gpu · 2022-10-02 14:50
karolherbstgoes into panic mode because flight preparations
#asahi-gpu · 2022-10-02 14:42
karolherbstanyway.. hw design matters for writing good driver impls, and some care less than others
#asahi-gpu · 2022-10-02 14:42
karolherbst:D
#asahi-gpu · 2022-10-02 14:41
karolherbstwhat's that about broadcom?
#asahi-gpu · 2022-10-02 14:35
karolherbstanyway, that blog post sums it all up nicely
#asahi-gpu · 2022-10-02 14:35
karolherbstpeople don't write scalar applications, they make have use of vectors and it sucks
#asahi-gpu · 2022-10-02 14:34
karolherbstthat's also one reason why CL is so inherently bad most of the times
#asahi-gpu · 2022-10-02 14:34
karolherbstfun read
#asahi-gpu · 2022-10-02 14:34
karolherbsthttps://pharr.org/matt/blog/2018/04/18/ispc-origins
#asahi-gpu · 2022-10-02 14:33
karolherbstand asked "how did nvidia solved this issue" - "they didn't, they don't use vectors" - "sure, but how did they solve auto vectorization?"
#asahi-gpu · 2022-10-02 14:33
karolherbstthere is even this joke story about intel engineers trying to beat cuda by using auto vectorization
#asahi-gpu · 2022-10-02 14:32
karolherbstalyssa: yeah.. very good thread
#asahi-gpu · 2022-10-02 14:32
karolherbstit's really idiotic
#asahi-gpu · 2022-10-02 14:32
karolherbstbut intel went this insane way where compute and something was scalar, but the other stages vectorized
#asahi-gpu · 2022-10-02 14:32
karolherbstoh sure.. for memory
#asahi-gpu · 2022-10-02 14:31
karolherbstnot sure when intel went full scalar though.. gfx9?
#asahi-gpu · 2022-10-02 14:31
karolherbstsome ditched them 15 years ago
#asahi-gpu · 2022-10-02 14:31
karolherbstthat's the reason they got vectorized ISAs for that long
#asahi-gpu · 2022-10-02 14:30
karolherbstauto vectorization is a religion at intel
#asahi-gpu · 2022-10-02 14:30
karolherbstit is
#asahi-gpu · 2022-10-02 14:29
karolherbstlina: yeah.. until you hear the reason
#asahi-gpu · 2022-10-02 14:29
karolherbst:D
#asahi-gpu · 2022-10-02 14:29
karolherbstthere is probably some good reason behind it once you hear the reason, but a totaly mystery for anybody else
#asahi-gpu · 2022-10-02 14:28
karolherbstlina: usually by hw engineers getting too nifty
#asahi-gpu · 2022-10-02 14:27
karolherbstat least modern intel GPUs are all scalar, finally
#asahi-gpu · 2022-10-02 14:27
karolherbstbut I also never talked to actual hw engineers at Intel
#asahi-gpu · 2022-10-02 14:27
karolherbstI don't think I ever heard anybody at Intel saying that even
#asahi-gpu · 2022-10-02 14:26
karolherbstuhhh
#asahi-gpu · 2022-10-02 14:26
karolherbstthough implementing geometry shaders sounds like pain on agx?

plus ›