Closed bostick closed 1 month ago
Thanks for this fix! Are you able to do any performance comparisons versus SIMD disabled on this platform? Just curious.
I have a 32-bit arm7 device I can test on. What comparisons would be good to do?
If you can run the benchmark app that would be great.
https://github.com/erincatto/box2d/blob/2dbb6818cb5e69bd3dcd20b7190cdba3844dd4e8/CMakeLists.txt#L72
OK, I will setup a simple Android app for running the benchmark.
OK @erincatto , I put up a repo with the results of running the benchmark: https://github.com/bostick/Box2DAndroidBenchmark
I have not interpreted the results. Let me know if you'd like anything else
It seems SIMD optimizations are not helping on this CPU. I've heard 32-bit Neon is not very good. I see good results on 64-bit Neon though. This CPU is also quite slow in general. But it does make sense to use around 4 threads on this.
Thanks a lot for running these tests!
Fixed the formatting. You can see other benchmark results here: https://box2d.org/files/benchmark_results.html
Ah, I messed up the formatting when I stripped the \n
in the fprintf
calls when I was replacing all of printf
calls with Android logging. Fixed in the repo now.
Fix https://github.com/erincatto/box2d/issues/813
Emulate vzip1q_f32 for 32-bit