We have a mathematical algorithm that has a relatively unbound number of inputs. We allocate memory before running the jobs, run the jobs, then deallocate that memory.
What we’ve noticed, particularly on a large batch of work, is that the TempJob allocator appears to run out, and we end up using a fallback allocator. This fallback allocator seems to be a lot slower than the TempJob one
Is there any way to control how big the tempjob allocator space is? is it dynamic? any information on how this works would be valuable
Thanks!