{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "import time\n",
    "\n",
    "import torch\n",
    "import torch_scatter as ts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_uniform_data(n, type):\n",
    "    output = type(n).fill_(0)\n",
    "    index = torch.arange(0, n, out=torch.LongTensor())\n",
    "    input = type(n).fill_(1)\n",
    "    return output, index, input\n",
    "\n",
    "def get_even_data(n, type):\n",
    "    output = type(n).fill_(0)\n",
    "    index = torch.LongTensor(n).fill_(0)\n",
    "    input = type(n).fill_(1)\n",
    "    return output, index, input\n",
    "\n",
    "n = 1000000\n",
    "num_runs = 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "types = [torch.FloatTensor, torch.DoubleTensor, torch.ByteTensor, torch.CharTensor,\n",
    "         torch.ShortTensor, torch.IntTensor, torch.LongTensor]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "def benchmark(name, output, index, input):\n",
    "    func = getattr(ts, name)\n",
    "    runtimes = []\n",
    "    for type in types:\n",
    "        runtime = 0\n",
    "        for i in range(num_runs):\n",
    "            t = time.process_time()\n",
    "            func(output, index, input, 0)\n",
    "            runtime += time.process_time() - t\n",
    "        runtime /= num_runs\n",
    "        runtimes.append(runtime)\n",
    "    return runtimes\n",
    "\n",
    "def benchmark_pytorch_scatter_add(output, index, input):\n",
    "    runtimes = []\n",
    "    for type in types:\n",
    "        runtime = 0\n",
    "        for i in range(num_runs):\n",
    "            t = time.process_time()\n",
    "            output.scatter_add_(0, index, input)\n",
    "            runtime += time.process_time() - t\n",
    "        runtime /= num_runs\n",
    "        runtimes.append(runtime)\n",
    "    return runtimes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# CPU Benchmark"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare to PyTorch `scatter_add_`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.0013608833600000025, 0.0010018577800000373, 0.001011737080000028, 0.0010304318000000912, 0.0010575933199999277, 0.001046791089999779, 0.001026226260000076]\n",
      "[0.0024379739199999763, 0.0020916335999999623, 0.0021022859400000016, 0.0020662273700000535, 0.002074936339999951, 0.002058829469999992, 0.002079201120000054]\n"
     ]
    }
   ],
   "source": [
    "output, index, input = get_uniform_data(n, type=torch.FloatTensor)\n",
    "runtimes = benchmark_pytorch_scatter_add(output, index, input)\n",
    "print(runtimes)\n",
    "\n",
    "output, index, input = get_even_data(n, type=torch.FloatTensor)\n",
    "runtimes = benchmark_pytorch_scatter_add(output, index, input)\n",
    "print(runtimes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.00494133797000007, 0.004897300620000032, 0.0045747565699999805, 0.004568737810000112, 0.004554663379999937, 0.004549899970000002, 0.004568819980000019]\n",
      "[0.0049159168400001365, 0.004562161230000079, 0.004574221990000033, 0.004564846730000127, 0.004551143499999953, 0.00455917282999998, 0.004575252730000017]\n"
     ]
    }
   ],
   "source": [
    "output, index, input = get_uniform_data(n, type=torch.FloatTensor)\n",
    "runtimes = benchmark('scatter_add_', output, index, input)\n",
    "print(runtimes)\n",
    "\n",
    "output, index, input = get_even_data(n, type=torch.FloatTensor)\n",
    "runtimes = benchmark('scatter_add_', output, index, input)\n",
    "print(runtimes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare internally"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# GPU Benchmark"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare to PyTorch `scatter_add_`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare internally"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}