sklam

Prime factorization is useful for reshaping an array into the highest dimension form where non of the dimension has size of 1.

sklam

Numba + Radare2

Information on seeing source info of Numba jit code in radare2.

Getting object file from numba

@njit(debug=True)
def foo():

sklam

/*
Compile this with:

    $ clang -O3 do_no_evil.c && ./a.out

Reference http://blog.llvm.org/2011/05/what-every-c-programmer-should-know_14.html
*/

#include <stdio.h>
#include <stdlib.h>

sklam

from numba import jitclass
from numba import types
from numba.typed import Dict


@jitclass([('d', types.DictType(types.intp, types.float64))])
class DictWrapper(object):
    def __init__(self):
        d = Dict()
        d[1] = 1.2

sklam

from numba import cuda, typeof
import numpy as np
import pickle


@cuda.jit
def foo(x, v):
    x[0] = v

sklam

from numba import njit


@njit
def foo(functor, a, b):
    return bar(functor, functor(a, b), b)


@njit
def bar(functor, a, b):

sklam

sklam

sklam

sklam

Usage:

conda create -n volcano_bench python=3.6 numba scipy tbb
conda activate volcano_bench
pip install pythran
pythran -O3 -fopenmp -march=native -DUSE_BOOST_SIMD shade_pythran.py
python driver_pythran.py
python driver_numba.py