Dmitry Babokin
Dmitry Babokin
> Could you wrap your test code into a trivial microbenchmark, similar to what we have here: [ISPC Benchmarks](https://github.com/ispc/ispc/tree/main/benchmarks/01_trivial)? This will allow us to retain and integrate your testing code...
Also I measured the impact of the changes on Apple M3 chip with `neon-i32x4` target. `exp` is not affected. `log` is slower approximately 50%.
To be more clear, we should do something like https://github.com/llvm/torch-mlir-release project, which has a "release" with nightly builds. But we should probably have just the latest, not the history with...
Once it's done, we should not forget to update the links in: - [ ] README.md - [ ] in the downloads section on the website - [ ] fix...
I did an experiment and all mac job completed in time, but the longer one - LTO `mac-build` (i.e. non Release one). For others time was: 1. LTO ON `mac-build-release`...
Do you mean supporting fat binary for x86 and M1? We do not support ARM64 on macOS at the moment. Though it might be working out of the box using...
I have the same problem. Looks like the timeout for upload is set for 5 minutes. All of my failing uploading steps run 5m1s.
You are welcome to submit patches for tasksys.cpp, which fix memory leaks. Or at least please be more specific about what you have found. But patches are better!
Good catch. I've checked how we handle result of fopen() and looks like we need some clean up.
@DeepakRajendrakumaran Please have a look.