symjit 2.19

[Shahriar Iravanian]

Hi All,

I just released version 2.19.1 of symjit (a JIT-compiler for SymPy expressions). The code is available at https://github.com/siravan/symjit/, and the binary can be installed as `conda install symjit -c conda-forge`. 

There have been many modifications and improvements in the last year. The major ones are

1. Symjit can now generate code for complex expressions. In fact, considering that one of its main use cases has been in solving Feynman integrals, the generated code for complex expressions is highly optimized. 

2. It works very well for creating huge straight line functions (10^5-10^7 terms), compiling to tens of MB. 

3. As of version 2.19, it has a Composer interface ( https://github.com/siravan/symjit/blob/v219/docs/COMPOSER.md), which allows for defining functions directly and at a low level without going through SymPy expressions. This is similar to llvmlite Builder. At one time, Oscar was interested in such applications regarding the protosym project. If this project is still active, Symjit may be a useful component. One advantage of Symjit compared to llvmlite for this purpose is the ease of generating vectorized/SIMD code.

I'm planning to release version 3.0 soon, which will have some breaking changes. I appreciate comments, especially regarding the Composer API, to have a stable interface before moving to version 3.0.

Shahriar Iravanian

[Jason Moore]

Dear Shahriar,

Thanks for the update. The latest release of PyDy now depends on symjit (optionally) and works nicely. I hope you try to maintain a public API without breaking changes or at least plenty of time with deprecation warnings.

Jason

moorepants.info
+01 530-601-9791

--
You received this message because you are subscribed to the Google Groups "sympy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sympy+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/sympy/000d0ba4-0f08-4404-a9a1-0c304e078f5cn%40googlegroups.com.

[Shahriar Iravanian]

Dear Jason,

Thanks a lot. My hope is not to break anyone's workflow! The main features flagged to be deprecated are:

1. Python-only backend, which will be released as a separate package. This hasn't been updated since version 1, and I doubt it is used by anyone. 

2. The minimum CPU support for x64 CPUs will be raised to x86-64-v3 (it is currently x86-64-v2). This roughly covers the majority of Intel and AMD processors since 2015, with the exception of Celeron processors. Many Linux distros are already moving to x86-64-v3, although numpy still supports x86-64-v2.   

However, there is no rush to do any of these. I agree with your recommendation to add deprecation warnings in the next version. Also, we may go through a release candidate phase before fully releasing version 3.0 (I think conda-forge has a mechanism to mark releases as rc).

Shahriar

You received this message because you are subscribed to a topic in the Google Groups "sympy" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sympy/grpQzWVg71E/unsubscribe.
To unsubscribe from this group and all its topics, send an email to sympy+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/sympy/CAP7f1AioBsRCibBOcbM2fZxN7766_J%2Bjaj8Yt5EGiur9NEYt4w%40mail.gmail.com.