You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This seems like a very nice and useful little library. Definitely fills a need.
Would you be willing to include support for aligned allocation? This is handled a little differently on each platform, and providing a nice, platform independent interface to it that "just works" would be quite useful.
For instance, if you see lines 12-43 in this file, you can see my own attempt at this. Frankly, I have no confidence that I've done this the right way. :-)
(I'm considering using cymem in the library linked to above. Intel's Embree library requires memory to be allocated with alignment.)
I don't have enough experience with it to know when exactly to expect a performance boost. You need to use it in order to take full advantage of SIMD vectorization. I understand your uncertainty... In my case, it's simply a requirement for using Embree---memory that the user allocates must be allocated with a specific alignment.
Activity
honnibal commentedon Nov 25, 2020
Yes I'd be interested in that. We have support for changing the allocation function, so in theory aligned allocation shouldn't be so hard.
The problem I have with aligned allocation is I never feel like I've done it properly =/. So I've never really seen a big performance increase.
sampotter commentedon Nov 25, 2020
I don't have enough experience with it to know when exactly to expect a performance boost. You need to use it in order to take full advantage of SIMD vectorization. I understand your uncertainty... In my case, it's simply a requirement for using Embree---memory that the user allocates must be allocated with a specific alignment.
What can I do to help?
polm commentedon Sep 19, 2021
Sorry for the very late reply here, but if you want to start work on a PR for this I think we'd be happy to look at it.