Open MattFerraro opened 4 months ago
I wouldn't say I've really started looking at it yet, but I did the easy thing, I turned on --release
, I did get a 10x or so performance boost
cargo run --example performance-issue 29.26s user 0.20s system 286% cpu 10.295 total
cargo run --profile release --example performance-issue 2.60s user 0.17s system 211% cpu 1.311 total
Here is an example script:
This takes 13 seconds of compute time on my M1 macbook air. Is it possible to cut the compute time down by a factor of 10, or even 100?
For anyone who might want to dive in a test things, here is a simpler example focusing just on OR, and it takes 15 seconds to run on my laptop:
This example is slower because the third parameter to
or()
is the tolerance to mesh to. The run time is strongly dependent on that tolerance. If I use a tolerance of0.9
it runs in just 4 seconds. But if I use a tolerances of1.0
it fails to solve and we get a panic instead.