Earning 2,000 stars, the speed is increased by a hundred times, and the high-performance Python compiler Codon is open source

As we all know, Python is a programming language that is easy to learn and has powerful functions. It always ranks at the top of various user usage statistics lists. Accordingly, researchers have developed various convenient tools around Python to better serve this language.

The compiler acts as a translator between high-level languages ​​and machines. Different versions of Python compilers have been developed. Below we will introduce a new high-performance Python to you. Compiler: Codon. The project has been online for just a few days and has already received 2.2k stars.

## Project address: https://github.com/exaloop/codon

As a high-performance Python compiler, Codon compiles Python code into native machine code without any runtime overhead. Typical speedups for Python are around 10-100x or more on a single thread. Codon's performance is generally comparable to C/C's. Unlike Python, Codon supports native multi-threading, which can make it many times faster. Codon is extensible through a plugin infrastructure, which allows users to incorporate new libraries, compiler optimizations and even keywords.

The Codon framework is fully modular and extensible, allowing seamless integration of new modules, compiler optimizations, domain-specific languages, and more, and is actively serving many applications such as bioinformatics and quantitative finance. Develop new Codon extensions in various areas.

Codon Pipeline

What is the effect of this compiler that has been so popular since its release? Let's take a look at some benchmark results.

Benchmarks

The following are results from the Codon benchmark suite, comparing the performance of Python, PyPy, C, and Codon on a range of tasks and applications.

Benchmarks were run on the following settings:

##Comparison of Python, PyPy, and Codon

#Comparison of Python, PyPy, C and Codon

​

##The specific comparison of several languages ​​is as follows:

Codon follows CPython syntax, semantics and API as much as possible, but in some special cases, considering performance reasons, Codon will be somewhat different from CPython. For example, Codon is a 64-bit int, and CPython is arbitrary widthint. In terms of performance, CPython speedups are typically 10-100x speedups.

Although Codon does provide a JIT decorator similar to Numba, Codon is generally an ahead-of-time compiler that can compile end-to-end programs into Native code. It also supports compilation of a wider set of Python constructs and libraries.

PyPy aims to be a simple replacement for CPython, while Codon is different in some places. These differences are mainly reflected in the elimination of dynamic runtime or virtual machine, resulting in better performance.

Codon usually generates the same code as an equivalent C or C program, and can sometimes generate better code than a C/C compiler. There are many reasons, such as better container implementation, Codon not using object files and inlining all library code, or Codon-specific compiler optimizations that are not performed using C or C .

Codon's compilation process is actually closer to C than to Julia. Julia is a dynamically typed language that performs type inference as an optimization, whereas Codon types are checked ahead of time throughout the program. Codon also attempts to circumvent the learning curve of a new language by adopting Python's syntax and semantics.

FAQ

Although Codon supports almost all syntax of Python, it is not a simple replacement, and large code bases may require modifications to compile with Codon The server is running. For example, some Python modules have not been implemented in Codon, and some dynamic features of Python are not allowed. The Codon compiler generates detailed error messages to help identify and resolve any incompatibility issues. Codon supports seamless Python interoperability to handle situations that require specific Python libraries or dynamics.

I want to use Codon, but I have a large Python code base and don’t want to port it, what should I do?

You can use Codon through the @codon.jit decorator, which will only compile annotated functions and automatically handle data conversion to and from Codon. It also allows the use of any Codon-specific modules or extensions, such as multithreading.

How interoperable is it with other languages ​​and frameworks?

Interoperability is a priority at Codon. We don't want to use Codon to prevent users from using other great frameworks and libraries that exist. Codon supports full interoperability with Python and C/C.

Does Codon use garbage collection?

Yes, Codon uses the Boehm garbage collector.

Codon doesn't support Python module X or function Y?

While Codon covers a sizeable subset of the Python standard library, it does not yet cover every function in every module. Note that missing functions can still be called from Python import. Many functions that lack native implementations of Codon (such as I/O or OS-related functions) typically do not achieve substantial speedups from Codon.

Is Codon not faster than Python for my application?

Applications that spend most of their time in library code implemented in C will generally not see substantial performance improvements in Codon. Likewise, applications that are I/O or network bound will experience the same bottlenecks in Codon.

Is Codon slower than Python for my application?

If this is the case, please report Codon being significantly slower than Python to the issue tracker.

Is Codon free?

Codon is always free for non-production use. Users are free to use Codon for personal, academic, or other non-commercial applications.

Is Codon open source?

Codon is licensed under the Business Source License (BSL) and its source code is publicly available and free for non-production use. BSL is not technically an "open source" license, although in many cases you can still treat Codon like any other open source project. Importantly, according to the BSL, every version of Codon is transitioned to an actual open source license (specifically Apache) after 3 years.

The above is the detailed content of Earning 2,000 stars, the speed is increased by a hundred times, and the high-performance Python compiler Codon is open source. For more information, please follow other related articles on the PHP Chinese website!