IPP Interface Test.ipynb

{
 "metadata": {
  "language": "Julia",
  "name": "",
  "signature": "sha256:3b2d60d98b839338f5308066590074242994759648c87951cef90b7c08ac1955"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Import [IPPDSP.jl](https://github.com/JayKickliter/IPPDSP.jl) Package"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "import IPPDSP"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 1
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Import [DSP.jl](https://github.com/JuliaDSP/DSP.jl) Package"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "import DSP"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 2
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Prepare Filter Taps and Signal"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "SignalType = Float64\n",
      "TapsType   = Float64\n",
      "taps       = ones( TapsType, 32 )\n",
      "signal     = ones( SignalType, 100_000_000 )\n",
      "buffer     = similar( signal );"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 3
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Setup IPP FIR Filter"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "flt  = IPPDSP.FIRFilter( Float64, taps );"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 4
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Native Julia Time-Domain `fir_filter!` & `fir_filter`\n",
      "\n",
      "### Two Versions\n",
      "* **`fir_filter!`**: Uses a pre-allocated buffer\n",
      "* **`fir_filter`**: Creates an output buffer and calls `fir_filter!`\n",
      "\n",
      "### Special Macros\n",
      "**`@inbounds`** to turn off bound checking. Since the vector lengths are checked before the loop, it is safe to do so. It also speeds up the code.\n",
      "\n",
      "**`@simd`** hints to the compiler that it the code lends itself to SIMD vectorization."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "function fir_filter!{T}( buffer::Vector{T}, taps::Vector{T}, signal::Vector{T} )\n",
      "    signal_len  = length( signal )\n",
      "    taps_len    = length( taps )\n",
      "    buf_len     = length( buffer )\n",
      "    signal_len >= taps_len && buf_len == signal_len || throw()\n",
      "    for n = 1:taps_len-1\n",
      "        for m = 1:n\n",
      "            @inbounds buffer[n] += taps[m] * signal[n-m+1]\n",
      "        end\n",
      "    end\n",
      "    for n = taps_len:signal_len\n",
      "        base_sig_idx = n-taps_len\n",
      "        @simd for m = 1:taps_len\n",
      "            @inbounds buffer[n] += taps[m] * signal[base_sig_idx+m]\n",
      "        end\n",
      "    end\n",
      "    return buffer\n",
      "end"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 5,
       "text": [
        "fir_filter! (generic function with 1 method)"
       ]
      }
     ],
     "prompt_number": 5
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "fir_filter( taps, signal ) = fir_filter!( similar(signal), taps, signal )"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 6,
       "text": [
        "fir_filter (generic function with 1 method)"
       ]
      }
     ],
     "prompt_number": 6
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "code_native( fir_filter!, ( Vector{SignalType}, Vector{TapsType}, Vector{SignalType} ))"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "\t.section\t__TEXT,__text,regular,pure_instructions\n",
        "Filename: In[5]\n",
        "Source line: 2\n",
        "\tpush\tRBP\n",
        "\tmov\tRBP, RSP\n",
        "Source line: 2\n",
        "\tpush\tR15\n",
        "\tpush\tR14\n",
        "\tpush\tR13\n",
        "\tpush\tR12\n",
        "\tpush\tRBX\n",
        "\tpush\tRAX\n",
        "\tmov\tR15, QWORD PTR [RSI]\n",
        "\tmov\tR14, QWORD PTR [RSI + 8]\n",
        "Source line: 3\n",
        "\tmov\tR13, QWORD PTR [R14 + 16]\n",
        "Source line: 2\n",
        "\tmov\tR12, QWORD PTR [RSI + 16]\n",
        "Source line: 2\n",
        "\tmov\tRAX, QWORD PTR [R12 + 16]\n",
        "Source line: 3\n",
        "\tmov\tQWORD PTR [RBP - 48], RAX\n",
        "\tcmp\tR13, RAX\n",
        "\tjg\t14\n",
        "Source line: 4\n",
        "\tmov\tRAX, QWORD PTR [RBP - 48]\n",
        "\tcmp\tQWORD PTR [R15 + 16], RAX\n",
        "\tje\t20\n",
        "\tmovabs\tRAX, 140553122728944\n",
        "Source line: 5\n",
        "\tmov\tRDI, QWORD PTR [RAX]\n",
        "\txor\tESI, ESI\n",
        "\txor\tEDX, EDX\n",
        "\tcall\tQWORD PTR [RDI + 8]\n",
        "Source line: 6\n",
        "\tlea\tR8, QWORD PTR [R13 - 1]\n",
        "\ttest\tR8, R8\n",
        "\tjle\t111\n",
        "\tmov\tECX, 1\n",
        "\txor\tR9D, R9D\n",
        "\ttest\tRCX, RCX\n",
        "\tjle\t79\n",
        "Source line: 8\n",
        "\tmov\tR10, QWORD PTR [R15 + 8]\n",
        "\tlea\tRBX, QWORD PTR [8*R9]\n",
        "\tadd\tRBX, QWORD PTR [R12 + 8]\n",
        "\tvmovsd\tXMM0, QWORD PTR [R10 + 8*RCX - 8]\n",
        "\tmov\tRAX, QWORD PTR [R14 + 8]\n",
        "\txor\tEDX, EDX\n",
        "\tlea\tRSI, QWORD PTR [8*RDX]\n",
        "\tmov\tRDI, RBX\n",
        "\tsub\tRDI, RSI\n",
        "\tvmovsd\tXMM1, QWORD PTR [RAX]\n",
        "\tvmulsd\tXMM1, XMM1, QWORD PTR [RDI]\n",
        "\tvaddsd\tXMM0, XMM0, XMM1\n",
        "\tvmovsd\tQWORD PTR [R10 + 8*RCX - 8], XMM0\n",
        "\tadd\tRAX, 8\n",
        "\tinc\tRDX\n",
        "\tcmp\tRCX, RDX\n",
        "\tjne\t-49\n",
        "Source line: 6\n",
        "\tinc\tRCX\n",
        "Source line: 8\n",
        "\tinc\tR9\n",
        "\tcmp\tRCX, R13\n",
        "\tjne\t-103\n",
        "\tmov\tRAX, QWORD PTR [RBP - 48]\n",
        "\tcmp\tR13, RAX\n",
        "Source line: 11\n",
        "\tcmovle\tR8, RAX\n",
        "\tlea\tRAX, QWORD PTR [R8 + 1]\n",
        "\tcmp\tR13, RAX\n",
        "\tje\t129\n",
        "\txor\tECX, ECX\n",
        "\ttest\tR13, R13\n",
        "Source line: 33\n",
        "\tmov\tR9, R13\n",
        "\tcmovle\tR9, RCX\n",
        "\tlea\tRAX, QWORD PTR [R9 - 1]\n",
        "\tcmp\tR9, 1\n",
        "\tjo\t121\n",
        "\tlea\tRSI, QWORD PTR [RAX + 1]\n",
        "\tadd\tRAX, 1\n",
        "\tjo\t107\n",
        "\ttest\tRSI, RSI\n",
        "\tjle\t64\n",
        "Source line: 48\n",
        "\tlea\tRDI, QWORD PTR [8*RCX]\n",
        "\tadd\tRDI, QWORD PTR [R12 + 8]\n",
        "Source line: 14\n",
        "\tmov\tRAX, QWORD PTR [R15 + 8]\n",
        "\tvmovsd\tXMM0, QWORD PTR [RAX + 8*R13 - 8]\n",
        "\tmov\tRDX, QWORD PTR [R14 + 8]\n",
        "\tvmovsd\tXMM1, QWORD PTR [RDX]\n",
        "\tvmulsd\tXMM1, XMM1, QWORD PTR [RDI]\n",
        "\tvaddsd\tXMM0, XMM0, XMM1\n",
        "\tvmovsd\tQWORD PTR [RAX + 8*R13 - 8], XMM0\n",
        "Source line: 41\n",
        "\tadd\tRDX, 8\n",
        "\tadd\tRDI, 8\n",
        "\tdec\tRSI\n",
        "\tjne\t-36\n",
        "Source line: 48\n",
        "\tinc\tRCX\n",
        "Source line: 11\n",
        "\tcmp\tR13, R8\n",
        "\tlea\tR13, QWORD PTR [R13 + 1]\n",
        "\tjne\t-117\n",
        "Source line: 17\n",
        "\tmov\tRAX, R15\n",
        "\tadd\tRSP, 8\n",
        "\tpop\tRBX\n",
        "\tpop\tR12\n",
        "\tpop\tR13\n",
        "\tpop\tR14\n",
        "\tpop\tR15\n",
        "\tpop\tRBP\n",
        "\tret\n",
        "Source line: 33\n",
        "\tmovabs\tRAX, 4481457312\n",
        "\tmov\tRDI, QWORD PTR [RAX]\n",
        "\tmovabs\tRAX, 4468907616\n",
        "\tmov\tESI, 33\n",
        "\tcall\tRAX\n"
       ]
      }
     ],
     "prompt_number": 7
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "## Precompile Functions\n",
      "\n",
      "Not necessary, but it to get accurate first-run @time results. Using `taps` as a small signal-vector."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "IPPDSP.filt( flt, signal )\n",
      "IPPDSP.filt!( flt, buffer, signal )\n",
      "filt( taps, one(TapsType), signal )\n",
      "DSP.firfilt( taps, signal )\n",
      "fir_filter( taps, signal );"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 8
    },
    {
     "cell_type": "heading",
     "level": 2,
     "metadata": {},
     "source": [
      "Times"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "### Julia's Built-in `filt()`\n",
      "\n",
      "Internally implemented with FFT convolution (I think)."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "@time filt( taps, one(taps[1]), signal );"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "elapsed time: "
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "5.367116163 seconds (800017656 bytes allocated, 0.31% gc time)\n"
       ]
      }
     ],
     "prompt_number": 9
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "DSP.jl `firfilt()`"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "@time DSP.firfilt( taps, signal );"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "elapsed time: "
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "1.672305065 seconds (804912080 bytes allocated, 3.34% gc time)\n"
       ]
      }
     ],
     "prompt_number": 10
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "### IPPDSP `filt()`\n",
      "\n",
      "Uses multiple dispatch to call the correct IPP function name."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "@time IPPDSP.filt( flt, signal );"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "elapsed time: "
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "0.74280871 seconds (800073668 bytes allocated, 9.43% gc time)\n"
       ]
      }
     ],
     "prompt_number": 11
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "### IPPDSP `filt!()`\n",
      "\n",
      "IPP has depreciated its in-place functions. `filt!` calls the same same IPP function as above, but with a pre-allocated buffer. It may be possible to pass the IPP function the same pointer for both source and destination; more testing needed."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "@time IPPDSP.filt!( flt, buffer, signal );"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "elapsed time: "
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "0.301538253 seconds (112 bytes allocated)\n"
       ]
      }
     ],
     "prompt_number": 12
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "Native `fir_filter()`"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "@time fir_filter( taps, signal );"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "elapsed time: "
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "3.451685513 seconds (800000128 bytes allocated)\n"
       ]
      }
     ],
     "prompt_number": 13
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "Native `fir_filter!()`"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "@time fir_filter!( buffer, taps, signal);"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "elapsed time: "
       ]
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "2.851588211 seconds (80 bytes allocated)\n"
       ]
      }
     ],
     "prompt_number": 14
    }
   ],
   "metadata": {}
  }
 ]
}