How to use

Install mimalloc

Install mimalloc in ~/local/lib.

$ git clone https://github.com/microsoft/mimalloc.git

$ cd mimalloc

$ vim include/mimalloc-types.h # enable MI_STAT=2

$ mkdir build

$ cd build

$ cmake .. -D CMAKE_INSTALL_PREFIX=$HOME/local

...

$ make -j8

...

$ make install

-- Install configuration: "Release"

-- Installing: /home/inada-n/local/lib/mimalloc-1.0/libmimalloc.so

-- Installing: /home/inada-n/local/lib/mimalloc-1.0/libmimalloc.a

-- Installing: /home/inada-n/local/lib/mimalloc-1.0/include/mimalloc.h

-- Installing: /home/inada-n/local/lib/mimalloc-1.0/cmake/mimalloc-config.cmake

-- Installing: /home/inada-n/local/lib/mimalloc-1.0/cmake/mimalloc-config-version.cmake

-- Installing: /home/inada-n/local/lib/mimalloc-1.0/cmake/mimalloc.cmake

-- Installing: /home/inada-n/local/lib/mimalloc-1.0/cmake/mimalloc-release.cmake

-- Installing: /home/inada-n/local/lib/libmimalloc.so

-- Installing: /home/inada-n/local/lib/mimalloc-1.0/mimalloc.o

Using mimalloc on CPython

link

The easiest way to link mimalloc is LD_PRELOAD=$HOME/local/lib/libmimalloc.so

You can statically link the mimalloc with LDFLAGS=$HOME/local/lib/mimalloc-1.0/libmimalloc.a too.

Use

mimalloc overrides malloc. So you can just use PYTHONMALLOC=malloc.

Performance

Benchmark (pyperformance)

$ ./python -m pyperformance run -o pymalloc.json

$ LD_PRELOAD=$HOME/local/lib/libmimalloc.so PYTHONMALLOC=malloc ./python -m pyperformance run -o mimalloc.json --inherit-environ=LD_PRELOAD,PYTHONMALLOC

The mimalloc outperforms the pymalloc on allocation heavy benchmarks.

$ ./python -m pyperf compare_to pymalloc.json mimalloc.json -G --min-speed=3

Faster (10):

- spectral_norm: 199 ms +- 1 ms -> 182 ms +- 1 ms: 1.10x faster (-9%)

- mako: 23.2 ms +- 0.2 ms -> 21.6 ms +- 0.3 ms: 1.07x faster (-7%)

​​How to use

​​Install mimalloc

​​Using mimalloc on CPython

​​Performance

​​Benchmark (pyperformance)

How to use

Install mimalloc

Using mimalloc on CPython

Performance

Benchmark (pyperformance)