{
 "metadata": {
  "name": "memview_bench_2"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Memoryview Benchmarks: Round 2"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%load_ext cythonmagic"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 1
    },
    {
     "cell_type": "heading",
     "level": 2,
     "metadata": {},
     "source": [
      "Inlined Memoryview"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%%cython\n",
      "\n",
      "import numpy as np\n",
      "cimport numpy as np\n",
      "cimport cython\n",
      "\n",
      "@cython.boundscheck(False)\n",
      "@cython.wraparound(False)\n",
      "cdef inline double inner_func(double[:, ::1] X):\n",
      "    return X[0, 0]\n",
      "\n",
      "def loop_1(int N, switch=True):\n",
      "    cdef double[:, ::1] X = np.zeros((100, 100))\n",
      "    cdef int i\n",
      "    for i in range(N):\n",
      "        # this should be inlined by the compiler\n",
      "        inner_func(X)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "missing cimport: /home/vanderplas/.config/ipython/cython/_cython_magic_aad7a6b46ed24a8baf2373c6b8784ac3.pyx\n",
        "cython\n"
       ]
      }
     ],
     "prompt_number": 2
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "timeit loop_1(1E6)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "100000 loops, best of 3: 10.1 us per loop\n"
       ]
      }
     ],
     "prompt_number": 3
    },
    {
     "cell_type": "heading",
     "level": 2,
     "metadata": {},
     "source": [
      "Inlined ndarray"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Now we'll repeat, but make a dummy function such that we\n",
      "can guarantee that `inner_func` will not be inlined"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%%cython\n",
      "\n",
      "import numpy as np\n",
      "cimport numpy as np\n",
      "cimport cython\n",
      "\n",
      "ctypedef double (*inner_func_ptr)(double[:, ::1])\n",
      "\n",
      "@cython.boundscheck(False)\n",
      "@cython.wraparound(False)\n",
      "cdef double inner_func_1(double[:, ::1] X):\n",
      "    return X[0, 0]\n",
      "\n",
      "@cython.boundscheck(False)\n",
      "@cython.wraparound(False)\n",
      "cdef double inner_func_2(double[:, ::1] X):\n",
      "    return X[0, 0]\n",
      "\n",
      "def loop_2(int N, switch=True):\n",
      "    # use a switch to ensure that inlining can't happen: compilers\n",
      "    # are usually smart enough to figure it out otherwise.\n",
      "    cdef inner_func_ptr func\n",
      "    if switch:\n",
      "        func = inner_func_1\n",
      "    else:\n",
      "        func = inner_func_2\n",
      "        \n",
      "    cdef double[:, ::1] X = np.zeros((100, 100))\n",
      "    cdef int i\n",
      "    for i in range(N):\n",
      "        func(X)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "missing cimport: /home/vanderplas/.config/ipython/cython/_cython_magic_ac4b8923a5786ceee8b7f92131a08998.pyx\n",
        "cython\n"
       ]
      }
     ],
     "prompt_number": 4
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%timeit loop_2(1E6)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "10 loops, best of 3: 22.9 ms per loop\n"
       ]
      }
     ],
     "prompt_number": 5
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "In this case, inlining improves the computation speed by a factor of 2000!"
     ]
    },
    {
     "cell_type": "heading",
     "level": 2,
     "metadata": {},
     "source": [
      "Attempting to inline ndarrays"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Here we'll replicate the fast method above, but with a typed numpy array\n",
      "rather than a typed memoryview:"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "%%cython\n",
      "\n",
      "import numpy as np\n",
      "cimport numpy as np\n",
      "cimport cython\n",
      "\n",
      "@cython.boundscheck(False)\n",
      "@cython.wraparound(False)\n",
      "cdef inline double inner_func(np.ndarray[double, ndim=2, mode='c'] X):\n",
      "    return X[0, 0]\n",
      "\n",
      "def loop_3(int N, switch=True):\n",
      "    cdef np.ndarray[double, ndim=2, mode='c'] X = np.zeros((100, 100))\n",
      "    cdef int i\n",
      "    for i in range(N):\n",
      "        inner_func(X)\n",
      "        "
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "missing cimport: /home/vanderplas/.config/ipython/cython/_cython_magic_0523a248c0a6af0427879e879eda0095.pyx\n",
        "cython\n"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stderr",
       "text": [
        "warning: /home/vanderplas/.config/ipython/cython/_cython_magic_0523a248c0a6af0427879e879eda0095.pyx:8:30: Buffer unpacking not optimized away.\n",
        "warning: /home/vanderplas/.config/ipython/cython/_cython_magic_0523a248c0a6af0427879e879eda0095.pyx:8:30: Buffer unpacking not optimized away.\n"
       ]
      }
     ],
     "prompt_number": 6
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "These warnings indicate a problem: buffer unpacking\n",
      "cannot be optimized for the numpy array dtype.\n",
      "Let's see how this compares to the other implementations:"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "print \"inlined memview:\"\n",
      "%timeit loop_1(1E6)\n",
      "print \"non-inlined memview:\"\n",
      "%timeit loop_2(1E6)\n",
      "print \"inlined ndarray:\"\n",
      "%timeit loop_3(1E6)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "inlined memview:\n",
        "100000 loops, best of 3: 10.1 us per loop"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n",
        "non-inlined memview:\n",
        "10 loops, best of 3: 22.9 ms per loop"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n",
        "inlined ndarray:\n",
        "1 loops, best of 3: 617 ms per loop"
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\n"
       ]
      }
     ],
     "prompt_number": 7
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "The result for the ndarray is many times slower than\n",
      "either the inlined or the non-inlined example above!"
     ]
    }
   ],
   "metadata": {}
  }
 ]
}