mypyc-aot 1.0.3

A performance optimization library based on mypyc, similar to numba but supporting accelerating general Python code.

[English | 简体中文]

mypyc_aot is a performance optimization library based on mypyc, supporting the acceleration of function/class performance using decorators, which is similar to the numba library but capable of accelerating general Python code. It enables leveraging modern performance from mypyc without the need to write setup.py, making it suitable for rapid prototyping, Jupyter notebooks, and similar scenarios.

Installation

pip install mypyc_aot

After installation, ensure a C compiler (such as gcc, clang or msvc) is installed.

Quick Start

Basic Usage

Using mypyc_aot in Python scripts is straightforward:

from mypyc_aot import Compiler

# Create a compiler instance
compiler = Compiler(globals())

# Use decorator to mark functions for optimization
@compiler.aot
def compute_sum(n: int) -> int:
    total = 0
    for i in range(n):
        total += i
    return total

# Start compilation
compiler.compile()

# Call the compiled function
result = compute_sum(10000000)

Jupyter Integration

Using mypyc_aot in Jupyter notebooks:

1. Firstly load the extension:

%load_ext mypyc_aot

2. Use cell magic to mark functions for optimization:

%%mypyc_aot
def process_data(data: list[float]) -> float:
    result = 0.0
    for value in data:
        result += value * value
    return result

3. The function will be compiled and ready for subsequent calls.

API Reference

mypyc_aot_nocache() and mypyc_aot()

---

mypyc_aot_nocache() Function

Performs non-cached AOT compilation. Takes source code as input and returns the corresponding compiled module object.

def mypyc_aot_nocache(
    pycode: str, 
    prefix: str = "mypyc_aot", 
    cache_dir: str | None = None, 
    quiet: bool = True, 
    compiler: str | None = None, 
    opt_level: str = "3", 
    strict_dunder_typing: bool = True, 
    experimental_features: bool = False
) -> module

Parameters

• pycode (str): Python source code string to be compiled
• prefix (str, optional): Prefix for generated module names, defaults to "mypyc_aot"
• cache_dir (str | None, optional): Cache directory path. If None, uses a temporary directory
• quiet (bool, optional): Silent mode. When True, suppresses compilation output, defaults to True
• compiler (str | None, optional): Specifies the C compiler name. If None, uses the default compiler
• opt_level (str, optional): Optimization level, defaults to "3" (maximum optimization)
• strict_dunder_typing (bool, optional): Whether to apply strict type checking to dunder methods, defaults to True
• experimental_features (bool, optional): Whether to enable experimental features, defaults to False

mypyc_aot() Function

Performs cached AOT compilation. If a cache exists, it is loaded directly; otherwise, compilation is performed and cached. Returns a compiled module object corresponding to the code.

def mypyc_aot(
    pycode: str, 
    prefix: str = "mypyc_aot", 
    cache_dir: str = CACHE_DIR, 
    compression_method: str = DEFAULT_COMP_METHOD, 
    **kw
) -> module

Parameters

• pycode (str): Python source code string to be compiled
• prefix (str, optional): Prefix for generated module names, defaults to "mypyc_aot"
• cache_dir (str, optional): Cache directory path, defaults to ~/.mypyc_aot_cache
• compression_method (str, optional): Cache compression method. Supports "zstandard", "zlib", or None (no compression). Defaults to DEFAULT_COMP_METHOD (automatically detects available compression libraries)
• kw: Additional keyword arguments passed to the mypyc_aot_nocache() function

Compiler Class

Compiler is the core class of mypyc_aot, used to manage compilation environments and function optimization.

Initialization Parameters:

• scope (dict): Global namespace, typically globals()
• ignored_vars (list[str] | None): List of variable names to ignore
• cache_dir (str): Cache directory path, defaults to .mypyc_aot_cache in user home directory
• quiet (bool): Whether to use silent mode, controlling compilation output
• compiler (str | None): Specified C compiler
• no_symbol_warnings (bool): Whether to disable symbol warnings
• ignore_import_not_found (bool): Whether to ignore import not found errors
• ignore_self (bool): Whether to ignore the mypyc_aot module itself
• **kw: Other mypyc compilation options

Main Methods:

The aot Decorator

@compiler.aot
def func(...):
    ...

Marks a function for AOT compilation optimization.
Note that you cannot use fn = compiler.aot(func) instead of @compiler.aot since the compiler modifies global() scope when compilation is completed.

add_func_or_class(source: str)

Adds function or class source code string for compilation.

add_source(source: str)

Adds arbitrary source code string.

compile()

Executes the compilation process, writing compiled functions back to the scope.

start_compilation_thread()

Begins compilation in a background thread to avoid blocking the main thread.

use_custom_compiler(cc: str, cxx: str)

Specifies custom C/C++ compiler, for example:

compiler.use_custom_compiler("gcc", "g++")

get_source_code() -> str

Retrieves all source code to be compiled.

Notes

1. Type Annotations: For optimal optimization results, it is recommended to add type annotations for function parameters and return values.
2. First-Run Delay: Initial compilation requires time, but subsequent runs will use cache and be significantly faster.
3. Compatibility: Not all Python features support compilation optimization; complex dynamic features may not be optimizable.
4. Cache Cleaning: To force recompilation, delete the cache directory (default: ~/.mypyc_aot_cache).

Troubleshooting

There are a few directions to consider for troubleshooting:

1. Check if code complies with mypyc's type requirements.
2. Check the generated codes and intermediate files in the cache directory (e.g. ~/.mypyc_aot_cache/cache).

Version

Current mypyc_aot version: 1.0.3