bitfusion-generate-graphs.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "benchlist length is 2 in benchmarks.py\n"
     ]
    }
   ],
   "source": [
    "import pandas\n",
    "import configparser as ConfigParser\n",
    "import os\n",
    "import numpy as np\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "from bitfusion.graph_plot.barchart import BarChart\n",
    "\n",
    "%matplotlib inline\n",
    "import matplotlib\n",
    "\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n",
    "\n",
    "import bitfusion.src.benchmarks.benchmarks as benchmarks\n",
    "from bitfusion.src.simulator.stats import Stats\n",
    "from bitfusion.src.simulator.simulator import Simulator\n",
    "from bitfusion.src.sweep.sweep import SimulatorSweep, check_pandas_or_run\n",
    "from bitfusion.src.utils.utils import *\n",
    "from bitfusion.src.optimizer.optimizer import optimize_for_order, get_stats_fast\n",
    "\n",
    "import dnnweaver2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 16\n",
    "\n",
    "results_dir = './results'\n",
    "if not os.path.exists(results_dir):\n",
    "    os.makedirs(results_dir)\n",
    "\n",
    "fig_dir = './fig'\n",
    "if not os.path.exists(fig_dir):\n",
    "    os.makedirs(fig_dir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "Simulator object\n\tMax supported precision: 8\n\tMin supported precision: 2\n\tSystolic array size: 16 -inputs x 32 -outputs\n\tWbuf size: 65,536 Bytes\n\tIbuf size: 32,768 Bytes\n\tObuf size: 16,384 Bytes\nDouble buffering enabled. Sizes of SRAM are halved\n\n**************************************************\nEnergy costs for BitFusion\nCore dynamic energy : 235.104 pJ/cycle (for entire systolic array)\nWBUF Read energy    : 0.021 pJ/bit\nWBUF Write energy   : 0.029 pJ/bit\nIBUF Read energy    : 0.059 pJ/bit\nIBUF Write energy   : 0.114 pJ/bit\nOBUF Read energy    : 0.033 pJ/bit\nOBUF Write energy   : 0.073 pJ/bit\n\n"
     ]
    }
   ],
   "source": [
    "# BitFusion configuration file\n",
    "config_file = 'bf_e_conf.ini'\n",
    "\n",
    "# Create simulator object\n",
    "verbose = False\n",
    "bf_e_sim = Simulator(config_file, verbose)\n",
    "bf_e_energy_costs = bf_e_sim.get_energy_cost()\n",
    "print(bf_e_sim)\n",
    "\n",
    "energy_tuple = bf_e_energy_costs\n",
    "print('')\n",
    "print('*'*50)\n",
    "print(energy_tuple)"
   ]
  },
  {
   "source": [
    "### Generate "
   ],
   "cell_type": "markdown",
   "metadata": {}
  },
  {
   "source": [
    "sim_sweep_columns = ['N', 'M',\n",
    "        'Max Precision (bits)', 'Min Precision (bits)',\n",
    "        'Network', 'Layer',\n",
    "        'Cycles', 'Memory wait cycles',\n",
    "        'WBUF Read', 'WBUF Write',\n",
    "        'OBUF Read', 'OBUF Write',\n",
    "        'IBUF Read', 'IBUF Write',\n",
    "        'DRAM Read', 'DRAM Write',\n",
    "        'Bandwidth (bits/cycle)',\n",
    "        'WBUF Size (bits)', 'OBUF Size (bits)', 'IBUF Size (bits)',\n",
    "        'Batch size']\n",
    "\n",
    "bf_e_sim_sweep_csv = os.path.join(results_dir, 'bitfusion-eyeriss-sim-sweep.csv')\n",
    "if os.path.exists(bf_e_sim_sweep_csv):\n",
    "    bf_e_sim_sweep_df = pandas.read_csv(bf_e_sim_sweep_csv)\n",
    "else:\n",
    "    bf_e_sim_sweep_df = pandas.DataFrame(columns=sim_sweep_columns)\n",
    "print('Got BitFusion Eyeriss, Numbers')\n",
    "\n",
    "bf_e_results = check_pandas_or_run(bf_e_sim, bf_e_sim_sweep_df,\n",
    "        bf_e_sim_sweep_csv, batch_size=batch_size)\n",
    "bf_e_results = bf_e_results.groupby('Network',as_index=False).agg(np.sum)\n",
    "area_stats = bf_e_sim.get_area()"
   ],
   "cell_type": "code",
   "metadata": {
    "scrolled": false
   },
   "execution_count": 5,
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "Got BitFusion Eyeriss, Numbers\n"
     ]
    }
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_eyeriss_energy(df):\n",
    "    eyeriss_energy_per_mac = 16 * 0.2 * 1.e-3 #energy in nJ\n",
    "    eyeriss_energy_alu = float(df['ALU'])\n",
    "    eyeriss_energy_dram = float(df['DRAM']) * 0.15 #Scaling due to technology\n",
    "    eyeriss_energy_buffer = float(df['Buffer'])\n",
    "    eyeriss_energy_array = float(df['Array'])\n",
    "    eyeriss_energy_rf = float(df['RF'])\n",
    "    eyeriss_energy = eyeriss_energy_alu + eyeriss_energy_dram + eyeriss_energy_buffer + eyeriss_energy_array + eyeriss_energy_rf\n",
    "    eyeriss_energy *= eyeriss_energy_per_mac\n",
    "    return eyeriss_energy\n",
    "\n",
    "def get_eyeriss_energy_breakdown(df):\n",
    "    eyeriss_energy_per_mac = 16 * 0.2 * 1.e-3 #energy in nJ\n",
    "    eyeriss_energy_alu = float(df['ALU'])\n",
    "    eyeriss_energy_dram = float(df['DRAM'])\n",
    "    eyeriss_energy_buffer = float(df['Buffer'])\n",
    "    eyeriss_energy_array = float(df['Array'])\n",
    "    eyeriss_energy_rf = float(df['RF'])\n",
    "    eyeriss_energy = [eyeriss_energy_alu+eyeriss_energy_array, eyeriss_energy_buffer, eyeriss_energy_rf, eyeriss_energy_dram]\n",
    "    eyeriss_energy = [x * eyeriss_energy_per_mac for x in eyeriss_energy]\n",
    "    return eyeriss_energy\n",
    "\n",
    "def df_to_stats(df):\n",
    "    stats = Stats()\n",
    "    stats.total_cycles = float(df['Cycles'])\n",
    "    stats.mem_stall_cycles = float(df['Memory wait cycles'])\n",
    "    stats.reads['act'] = float(df['IBUF Read'])\n",
    "    stats.reads['out'] = float(df['OBUF Read'])\n",
    "    stats.reads['wgt'] = float(df['WBUF Read'])\n",
    "    stats.reads['dram'] = float(df['DRAM Read'])\n",
    "    stats.writes['act'] = float(df['IBUF Write'])\n",
    "    stats.writes['out'] = float(df['OBUF Write'])\n",
    "    stats.writes['wgt'] = float(df['WBUF Write'])\n",
    "    stats.writes['dram'] = float(df['DRAM Write'])\n",
    "    return stats"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "BitFusion-Eyeriss comparison\nArea budget = 5.871301775147929\n(0.8880907385600001, 0.76747776, 0.242112, 0.24330592)\nWarning: BitFusion Area is outside 10% of eyeriss\ntotal_area = 2.1409864185600003, budget = 5.871301775147929\n**************************************************\nBenchmark: AlexNet\nEyeriss time: 8.785792 ms\nBitFusion time: 4.708311375 ms\nEyeriss power: 373.97237605442973 mWatt\nBitFusion power: 448.3862417181711 mWatt\n**************************************************\n"
     ]
    },
    {
     "output_type": "error",
     "ename": "TypeError",
     "evalue": "cannot convert the series to <class 'float'>",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-14-83164476ebf3>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m     19\u001b[0m     \u001b[0mlookup_dict\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m{\u001b[0m\u001b[0;34m'Benchmark'\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0mbench\u001b[0m\u001b[0;34m}\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     20\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 21\u001b[0;31m     \u001b[0meyeriss_cycles\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfloat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlookup_pandas_dataframe\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0meyeriss_data_bench\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlookup_dict\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'time(ms)'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     22\u001b[0m     \u001b[0meyeriss_time\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0meyeriss_cycles\u001b[0m \u001b[0;34m/\u001b[0m \u001b[0;36m500.e3\u001b[0m \u001b[0;34m/\u001b[0m \u001b[0;36m16\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     23\u001b[0m     \u001b[0meyeriss_energy\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mget_eyeriss_energy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlookup_pandas_dataframe\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0meyeriss_data_bench\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlookup_dict\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/lib/python3.7/site-packages/pandas/core/series.py\u001b[0m in \u001b[0;36mwrapper\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    110\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    111\u001b[0m             \u001b[0;32mreturn\u001b[0m \u001b[0mconverter\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0miloc\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 112\u001b[0;31m         \u001b[0;32mraise\u001b[0m \u001b[0mTypeError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34mf\"cannot convert the series to {converter}\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    113\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    114\u001b[0m     \u001b[0mwrapper\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m__name__\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34mf\"__{converter.__name__}__\"\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mTypeError\u001b[0m: cannot convert the series to <class 'float'>"
     ]
    }
   ],
   "source": [
    "# Eyeriss-Simulator\n",
    "eyeriss_data = pandas.read_csv(os.path.join(results_dir, 'eyeriss_results.csv'))\n",
    "eyeriss_data_bench = eyeriss_data.groupby('Benchmark', as_index=False).agg(np.sum)\n",
    "eyeriss_data_bench['Platform'] = 'Eyeriss (16-bit)'\n",
    "\n",
    "print('BitFusion-Eyeriss comparison')\n",
    "eyeriss_area = 3.5*3.5*45*45/65./65.\n",
    "print('Area budget = {}'.format(eyeriss_area))\n",
    "\n",
    "\n",
    "print(area_stats)\n",
    "if abs(sum(area_stats)-eyeriss_area)/eyeriss_area > 0.1:\n",
    "    print('Warning: BitFusion Area is outside 10% of eyeriss')\n",
    "print('total_area = {}, budget = {}'.format(sum(area_stats), eyeriss_area))\n",
    "bf_e_area = sum(area_stats)\n",
    "\n",
    "baseline_data = []\n",
    "for bench in benchmarks.benchlist:\n",
    "    lookup_dict = {'Benchmark': bench}\n",
    "    eyeriss_cycles = float(lookup_pandas_dataframe(eyeriss_data_bench, lookup_dict)['time(ms)'])\n",
    "    eyeriss_time = eyeriss_cycles / 500.e3 / 16\n",
    "    eyeriss_energy = get_eyeriss_energy(lookup_pandas_dataframe(eyeriss_data_bench, lookup_dict))\n",
    "    eyeriss_power = eyeriss_energy / eyeriss_time * 1.e-9\n",
    "\n",
    "    eyeriss_speedup = eyeriss_time / eyeriss_time\n",
    "    eyeriss_energy_efficiency = eyeriss_energy / eyeriss_energy\n",
    "\n",
    "    eyeriss_ppa = eyeriss_speedup / eyeriss_area / (eyeriss_speedup / eyeriss_area)\n",
    "    eyeriss_ppw = eyeriss_speedup / eyeriss_power / (eyeriss_speedup / eyeriss_power)\n",
    "    \n",
    "    bf_e_stats = df_to_stats(bf_e_results.loc[bf_e_results['Network'] == bench])\n",
    "    bf_e_cycles = bf_e_stats.total_cycles * (batch_size / 16.)\n",
    "    bf_e_time = bf_e_cycles / 500.e3 / 16\n",
    "    bf_e_energy = bf_e_stats.get_energy(bf_e_sim.get_energy_cost()) * (batch_size / 16.)\n",
    "    bf_e_power = bf_e_energy / bf_e_time * 1.e-9\n",
    "\n",
    "    bf_e_speedup = eyeriss_time / bf_e_time\n",
    "    bf_e_energy_efficiency = eyeriss_energy / bf_e_energy\n",
    "\n",
    "    bf_e_ppa = bf_e_speedup / bf_e_area / (eyeriss_speedup / eyeriss_area)\n",
    "    bf_e_ppw = bf_e_speedup / bf_e_power / (eyeriss_speedup / eyeriss_power)\n",
    "\n",
    "    baseline_data.append(['Performance', bench, bf_e_speedup])\n",
    "    baseline_data.append(['Energy reduction', bench, bf_e_energy_efficiency])\n",
    "    baseline_data.append(['Performance-per-Watt', bench, bf_e_ppw])\n",
    "    baseline_data.append(['Performance-per-Area', bench, bf_e_ppa])\n",
    "    \n",
    "    print('*'*50)\n",
    "    print('Benchmark: {}'.format(bench))\n",
    "    print('Eyeriss time: {} ms'.format(eyeriss_time))\n",
    "    print('BitFusion time: {} ms'.format(bf_e_time))    \n",
    "    print('Eyeriss power: {} mWatt'.format(eyeriss_power*1.e3*16))\n",
    "    print('BitFusion power: {} mWatt'.format(bf_e_power*1.e3*16))\n",
    "    print('*'*50)\n",
    "    \n",
    "eyeriss_comparison_df = pandas.DataFrame(baseline_data, columns=['Metric', 'Network', 'Value'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "output_type": "error",
     "ename": "NameError",
     "evalue": "name 'eyeriss_comparison_df' is not defined",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-13-202fb51e82ac>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m     14\u001b[0m \u001b[0mbc\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mLEGEND_FONTSIZE\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;36m16\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     15\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 16\u001b[0;31m \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0meyeriss_comparison_df\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     17\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     18\u001b[0m \u001b[0;31m# print('Generating Figure for GPU performance comparison')\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'eyeriss_comparison_df' is not defined"
     ]
    }
   ],
   "source": [
    "bc = BarChart()\n",
    "bc.LABEL_BAR = True\n",
    "bc.BAR_WIDTH = None\n",
    "bc.YAXIS_MAX = 16\n",
    "bc.ISTIMES = True\n",
    "bc.TOP_ROTATE = True\n",
    "bc.LEGEND_LOCATION = \"upper center\"\n",
    "bc.LEGEND_NCOLUMN = 5\n",
    "bc.MV_LEGEND_OUTSIDE_Y = 1.4\n",
    "bc.MV_LEGEND_OUTSIDE_X = 0.5\n",
    "bc.LOG_SCALE = False\n",
    "bc.FIG_WIDTH = 16\n",
    "bc.TOP_LABEL_FONTSIZE = 12\n",
    "bc.LEGEND_FONTSIZE = 16\n",
    "\n",
    "print(eyeriss_comparison_df)\n",
    "\n",
    "# print('Generating Figure for GPU performance comparison')\n",
    "bc.yaxis = 'Performance-over-Eyeriss'\n",
    "bc.BAR_WIDTH = None\n",
    "legend_list = ['Performance', 'Energy reduction']\n",
    "fig_name = os.path.join(fig_dir, 'eyeriss-comparision.pdf')\n",
    "bc.plot(fig_name, eyeriss_comparison_df, x_plot='Network', y_plot='Value', bar_plot='Metric',\n",
    "        bar_list=legend_list, add_geomean=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Energy breakdown for Eyeriss and BitFusion"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "breakdown_data = []\n",
    "for bench in benchmarks.benchlist:\n",
    "    lookup_dict = {'Benchmark': bench}\n",
    "    eyeriss_energy = get_eyeriss_energy_breakdown(lookup_pandas_dataframe(eyeriss_data_bench, lookup_dict))\n",
    "    total_energy = sum(eyeriss_energy)\n",
    "    eyeriss_breakdown = [x / float(total_energy) for x in eyeriss_energy]\n",
    "    \n",
    "    breakdown_data.append(['Eyeriss', bench] + eyeriss_breakdown)\n",
    "    \n",
    "    bf_e_stats = df_to_stats(bf_e_results.loc[bf_e_results['Network'] == bench])\n",
    "    bf_breakdown = bf_e_stats.get_energy_breakdown(bf_e_sim.get_energy_cost())\n",
    "    total = sum(bf_breakdown)\n",
    "    bf_breakdown = [x / float(total) for x in bf_breakdown]\n",
    "    \n",
    "    breakdown_data.append(['BitFusion', bench] + bf_breakdown)\n",
    "    \n",
    "breakdown_df = pandas.DataFrame(breakdown_data, columns = ['Platform', 'Network', 'Compute', 'Register File', 'Buffers', 'DRAM'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x252 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "bc = BarChart()\n",
    "bc.LABEL_BAR = False\n",
    "bc.BAR_LEFT_MARGIN = 1\n",
    "bc.BAR_GAP = 0.1\n",
    "bc.YAXIS_MAX = 1\n",
    "bc.ISTIMES = True\n",
    "bc.TOP_ROTATE = True\n",
    "bc.LEGEND_LOCATION = \"upper center\"\n",
    "bc.LEGEND_NCOLUMN = 5\n",
    "bc.MV_LEGEND_OUTSIDE_Y = 1.2\n",
    "bc.MV_LEGEND_OUTSIDE_X = 0.5\n",
    "bc.LOG_SCALE = False\n",
    "bc.FIG_WIDTH = 16\n",
    "bc.TOP_LABEL_FONTSIZE = 12\n",
    "bc.LEGEND_FONTSIZE = 16\n",
    "\n",
    "# print(eyeriss_comparison_df)\n",
    "\n",
    "# print('Generating Figure for GPU performance comparison')\n",
    "bc.yaxis = 'Energy breakdown'\n",
    "bc.BAR_WIDTH = None\n",
    "legend_list = ['Eyeriss', 'BitFusion']\n",
    "fig_name = os.path.join(fig_dir, 'bitfusion-eyeriss-energy-breakdown.pdf')\n",
    "bc.plot(fig_name, breakdown_df, x_plot='Network', y_plot=['Compute', 'Register File', 'Buffers', 'DRAM'], bar_plot='Platform',\n",
    "        bar_list=legend_list)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## GPU numbers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "           Name    Network            fps  Batch size\n",
      "0   TitanX-fp32    AlexNet    5555.555556          16\n",
      "1   TitanX-int8    AlexNet   10884.353741          16\n",
      "2      TX2-fp32    AlexNet     459.188049          16\n",
      "3      TX2-int8    AlexNet     899.633729          16\n",
      "4   TitanX-fp32       SVHN   27118.644068          16\n",
      "5   TitanX-int8       SVHN   39506.172840          16\n",
      "6      TX2-fp32       SVHN    1818.575156          16\n",
      "7      TX2-int8       SVHN    2649.282326          16\n",
      "8   TitanX-fp32    CIFAR10    9142.857143          16\n",
      "9   TitanX-int8    CIFAR10   19207.683073          16\n",
      "10     TX2-fp32    CIFAR10     654.911590          16\n",
      "11     TX2-int8    CIFAR10    1375.864685          16\n",
      "12  TitanX-fp32    LeNet-5   87912.087912          16\n",
      "13  TitanX-int8    LeNet-5  115107.913669          16\n",
      "14     TX2-fp32    LeNet-5    4230.188957          16\n",
      "15     TX2-int8    LeNet-5    5538.808563          16\n",
      "16  TitanX-fp32      VGG-7    9411.764706          16\n",
      "17  TitanX-int8      VGG-7   20278.833967          16\n",
      "18     TX2-fp32      VGG-7     660.348364          16\n",
      "19     TX2-int8      VGG-7    1422.803827          16\n",
      "20  TitanX-fp32  RESNET-18    2946.593002          16\n",
      "21  TitanX-int8  RESNET-18    6837.606838          16\n",
      "22     TX2-fp32  RESNET-18     220.442505          16\n",
      "23     TX2-int8  RESNET-18     511.539660          16\n",
      "24  TitanX-fp32        RNN   80000.000000          16\n",
      "25  TitanX-int8        RNN   84210.526316          16\n",
      "26     TX2-fp32        RNN   11662.860633          16\n",
      "27     TX2-int8        RNN   12276.695403          16\n",
      "28  TitanX-fp32       LSTM   88888.888889          16\n",
      "29  TitanX-int8       LSTM   94117.647059          16\n",
      "30     TX2-fp32       LSTM   13976.207556          16\n",
      "31     TX2-int8       LSTM   14798.337412          16\n"
     ]
    }
   ],
   "source": [
    "gpu_data = []\n",
    "gpu_list = ['TX2-fp32', 'TitanX-fp32', 'TitanX-int8']\n",
    "config = ConfigParser.ConfigParser()\n",
    "config.read('results/output.ini')\n",
    "for bench in benchmarks.benchlist:\n",
    "    # Titan-X Pascal numbers\n",
    "    titanx_fp32_sec = '{}-performance-batch-{}'.format('TitanX-fp32', batch_size)    \n",
    "    titanx_int8_sec = '{}-performance-batch-{}'.format('TitanX-int8', batch_size)\n",
    "    titanx_fp32_time = config.getfloat(titanx_fp32_sec, bench.lower())\n",
    "    titanx_int8_time = config.getfloat(titanx_int8_sec, bench.lower())    \n",
    "    titanx_fp32_fps = batch_size / titanx_fp32_time * 1.e3\n",
    "    titanx_int8_fps = batch_size / titanx_int8_time * 1.e3\n",
    "    int8_speedup = titanx_int8_fps/titanx_fp32_fps\n",
    "    \n",
    "    gpu_data.append(['TitanX', bench, 'fp32', titanx_fp32_fps, batch_size])\n",
    "    gpu_data.append(['TitanX', bench, 'int8', titanx_int8_fps, batch_size])\n",
    "    \n",
    "    # Jetson-TX2 numbers\n",
    "    tx2_fp32_sec = '{}-performance-batch-{}'.format('TX2-fp32', batch_size)    \n",
    "    tx2_int8_sec = '{}-performance-batch-{}'.format('TX2-int8', batch_size)\n",
    "    tx2_fp32_time = config.getfloat(tx2_fp32_sec, bench.lower())\n",
    "    try:\n",
    "        tx2_int8_time = config.getfloat(tx2_int8_sec, bench.lower())\n",
    "    except:\n",
    "        # ID4A speedup\n",
    "        tx2_int8_time = tx2_fp32_time / int8_speedup\n",
    "    tx2_fp32_fps = batch_size / tx2_fp32_time * 1.e3 * 2\n",
    "    tx2_int8_fps = batch_size / tx2_int8_time * 1.e3 * 2\n",
    "    \n",
    "    gpu_data.append(['TX2', bench, 'fp32', tx2_fp32_fps, batch_size])\n",
    "    gpu_data.append(['TX2', bench, 'int8', tx2_int8_fps, batch_size])\n",
    "    \n",
    "#     print(v100_fp32_time, v100_int8_time)\n",
    "#     print(titanx_fp32_time, titanx_int8_time)\n",
    "    \n",
    "gpu_data = pandas.DataFrame(gpu_data, columns=['Platform', 'Network', 'Precision', 'fps', 'Batch size'])\n",
    "\n",
    "gpu_fps_df = gpu_data\n",
    "gpu_fps_df['Name'] = gpu_fps_df['Platform'] + '-' + gpu_fps_df['Precision']\n",
    "gpu_fps_df = gpu_fps_df[['Name', 'Network', 'fps', 'Batch size']]\n",
    "\n",
    "print(gpu_fps_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## BitFusion numbers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "**************************************************\n",
      "AlexNet\n",
      "\tAchieved FPS = 212.39 frames/sec\n",
      "\tBitfusion speedup over TX2: 0.462534622567\n",
      "\tTitanX speedup over TX2: 23.7034778206\n",
      "**************************************************\n",
      "**************************************************\n",
      "SVHN\n",
      "\tAchieved FPS = 6072.00 frames/sec\n",
      "\tBitfusion speedup over TX2: 3.33887782483\n",
      "\tTitanX speedup over TX2: 21.7236954489\n",
      "**************************************************\n",
      "**************************************************\n",
      "CIFAR10\n",
      "\tAchieved FPS = 2428.88 frames/sec\n",
      "\tBitfusion speedup over TX2: 3.70870887357\n",
      "\tTitanX speedup over TX2: 29.3286656703\n",
      "**************************************************\n",
      "**************************************************\n",
      "LeNet-5\n",
      "\tAchieved FPS = 19495.46 frames/sec\n",
      "\tBitfusion speedup over TX2: 4.60864934288\n",
      "\tTitanX speedup over TX2: 27.2110571964\n",
      "**************************************************\n",
      "**************************************************\n",
      "VGG-7\n",
      "\tAchieved FPS = 2834.38 frames/sec\n",
      "\tBitfusion speedup over TX2: 4.29224313535\n",
      "\tTitanX speedup over TX2: 30.7092968804\n",
      "**************************************************\n",
      "**************************************************\n",
      "RESNET-18\n",
      "\tAchieved FPS = 163.00 frames/sec\n",
      "\tBitfusion speedup over TX2: 0.739418400454\n",
      "\tTitanX speedup over TX2: 31.0176425511\n",
      "**************************************************\n",
      "**************************************************\n",
      "RNN\n",
      "\tAchieved FPS = 20274.57 frames/sec\n",
      "\tBitfusion speedup over TX2: 1.73838726018\n",
      "\tTitanX speedup over TX2: 7.22040063455\n",
      "**************************************************\n",
      "**************************************************\n",
      "LSTM\n",
      "\tAchieved FPS = 23726.55 frames/sec\n",
      "\tBitfusion speedup over TX2: 1.69763874776\n",
      "\tTitanX speedup over TX2: 6.73413346794\n",
      "**************************************************\n"
     ]
    }
   ],
   "source": [
    "# Constant for a particular accelerator configuration\n",
    "accelerator_constants = ['N', 'M',\n",
    "                         'Max Precision (bits)', 'Min Precision (bits)',\n",
    "                         'WBUF Size (bits)', 'OBUF Size (bits)', 'IBUF Size (bits)',\n",
    "                         'Bandwidth (bits/cycle)'\n",
    "                        ]\n",
    "# Constant for a particular network run\n",
    "algorithm_constants = ['Batch size']\n",
    "\n",
    "\n",
    "# Variables (to be summed) for a particular network run\n",
    "variable_columns = ['Cycles', 'Memory wait cycles',\n",
    "                    'WBUF Read', 'WBUF Write',\n",
    "                    'OBUF Read', 'OBUF Write',\n",
    "                    'IBUF Read', 'IBUF Write',\n",
    "                    'DRAM Read', 'DRAM Write',\n",
    "                   ]\n",
    "\n",
    "sim_sweep_columns = accelerator_constants + \\\n",
    "                        algorithm_constants + \\\n",
    "                        variable_columns + \\\n",
    "                        ['Network', 'Layer']\n",
    "\n",
    "\n",
    "# CSV for sweeps\n",
    "bf_e_sim_sweep_csv = os.path.join(results_dir, 'bitfusion-eyeriss-sim-sweep.csv')\n",
    "if os.path.exists(bf_e_sim_sweep_csv):\n",
    "    bf_e_sim_sweep_df = pandas.read_csv(bf_e_sim_sweep_csv)\n",
    "else:\n",
    "    bf_e_sim_sweep_df = pandas.DataFrame(columns=sim_sweep_columns)\n",
    "# print('Got BitFusion Eyeriss Numbers')\n",
    "# print(bf_e_sim_sweep_df)\n",
    "\n",
    "bf_e_results = check_pandas_or_run(bf_e_sim, bf_e_sim_sweep_df,\n",
    "                                   bf_e_sim_sweep_csv, batch_size=batch_size)\n",
    "bf_e_results = bf_e_results.groupby(['Network', 'Batch size']+accelerator_constants,\n",
    "                                    as_index=False).agg(np.sum)\n",
    "frequency = bf_e_sim.accelerator.frequency #500 MHz\n",
    "bf_e_results['Time (ms)'] = bf_e_results['Cycles'] * 1.e3 / frequency # in ms\n",
    "bf_e_results['fps'] = batch_size / bf_e_results['Time (ms)'] * 1.e3\n",
    "\n",
    "\n",
    "for bench in benchmarks.benchlist:\n",
    "    graph = benchmarks.get_bench_nn(bench)\n",
    "    bf_e_datarow = lookup_pandas_dataframe(bf_e_results, {'Network': bench})\n",
    "    cycles = bf_e_datarow.Cycles\n",
    "    mem_cycles = bf_e_datarow['Memory wait cycles']\n",
    "            \n",
    "    print('*'*50)\n",
    "    print(bench)\n",
    "    bf_e_fps = float(bf_e_datarow['fps'])\n",
    "    print('\\tAchieved FPS = {:.2f} frames/sec'.format(bf_e_fps))\n",
    "    tx2_fps = float(lookup_pandas_dataframe(gpu_fps_df, {'Name': 'TX2-fp32', 'Network': bench})['fps'])\n",
    "    titanx_fps = float(lookup_pandas_dataframe(gpu_fps_df, {'Name': 'TitanX-int8', 'Network': bench})['fps'])\n",
    "    \n",
    "    titanx_speedup = titanx_fps / tx2_fps\n",
    "    bf_e_speedup = bf_e_fps / tx2_fps\n",
    "    \n",
    "    print('\\tBitfusion speedup over TX2: {}'.format(bf_e_speedup))\n",
    "    print('\\tTitanX speedup over TX2: {}'.format(titanx_speedup))\n",
    "    \n",
    "    print('*'*50)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generate performance comparisons"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "      Network            Name            fps\n",
      "0     AlexNet  BitFusion-45nm     212.390371\n",
      "1     CIFAR10  BitFusion-45nm    2428.876426\n",
      "2        LSTM  BitFusion-45nm   23726.551494\n",
      "3     LeNet-5  BitFusion-45nm   19495.457558\n",
      "4   RESNET-18  BitFusion-45nm     162.999245\n",
      "5         RNN  BitFusion-45nm   20274.568342\n",
      "6        SVHN  BitFusion-45nm    6072.000261\n",
      "7       VGG-7  BitFusion-45nm    2834.375734\n",
      "0     AlexNet     TitanX-fp32    5555.555556\n",
      "1     AlexNet     TitanX-int8   10884.353741\n",
      "2     AlexNet        TX2-fp32     459.188049\n",
      "3     AlexNet        TX2-int8     899.633729\n",
      "4        SVHN     TitanX-fp32   27118.644068\n",
      "5        SVHN     TitanX-int8   39506.172840\n",
      "6        SVHN        TX2-fp32    1818.575156\n",
      "7        SVHN        TX2-int8    2649.282326\n",
      "8     CIFAR10     TitanX-fp32    9142.857143\n",
      "9     CIFAR10     TitanX-int8   19207.683073\n",
      "10    CIFAR10        TX2-fp32     654.911590\n",
      "11    CIFAR10        TX2-int8    1375.864685\n",
      "12    LeNet-5     TitanX-fp32   87912.087912\n",
      "13    LeNet-5     TitanX-int8  115107.913669\n",
      "14    LeNet-5        TX2-fp32    4230.188957\n",
      "15    LeNet-5        TX2-int8    5538.808563\n",
      "16      VGG-7     TitanX-fp32    9411.764706\n",
      "17      VGG-7     TitanX-int8   20278.833967\n",
      "18      VGG-7        TX2-fp32     660.348364\n",
      "19      VGG-7        TX2-int8    1422.803827\n",
      "20  RESNET-18     TitanX-fp32    2946.593002\n",
      "21  RESNET-18     TitanX-int8    6837.606838\n",
      "22  RESNET-18        TX2-fp32     220.442505\n",
      "23  RESNET-18        TX2-int8     511.539660\n",
      "24        RNN     TitanX-fp32   80000.000000\n",
      "25        RNN     TitanX-int8   84210.526316\n",
      "26        RNN        TX2-fp32   11662.860633\n",
      "27        RNN        TX2-int8   12276.695403\n",
      "28       LSTM     TitanX-fp32   88888.888889\n",
      "29       LSTM     TitanX-int8   94117.647059\n",
      "30       LSTM        TX2-fp32   13976.207556\n",
      "31       LSTM        TX2-int8   14798.337412\n"
     ]
    }
   ],
   "source": [
    "bf_e_fps_df = bf_e_results[['Network', 'fps']]\n",
    "bf_e_fps_df['Name'] = 'BitFusion-45nm'\n",
    "\n",
    "all_fps_df = pandas.concat([bf_e_fps_df, gpu_fps_df])[['Network', 'Name', 'fps']]\n",
    "print(all_fps_df)\n",
    "platform_list = ['TX2-fp32', 'TitanX-fp32', 'TitanX-int8', 'BitFusion-45nm']\n",
    "all_fps_df = all_fps_df[all_fps_df['Name'].isin(platform_list)]\n",
    "# print(all_fps_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## GPU performance comparison figure"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x252 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "bc = BarChart()\n",
    "bc.LABEL_BAR = True\n",
    "bc.BAR_WIDTH = None\n",
    "bc.YAXIS_MAX = 5\n",
    "bc.ISTIMES = True\n",
    "bc.TOP_ROTATE = True\n",
    "bc.LEGEND_LOCATION = \"upper center\"\n",
    "bc.LEGEND_NCOLUMN = 5\n",
    "bc.MV_LEGEND_OUTSIDE_Y = 1.4\n",
    "bc.MV_LEGEND_OUTSIDE_X = 0.5\n",
    "bc.LOG_SCALE = False\n",
    "bc.FIG_WIDTH = 16\n",
    "bc.TOP_LABEL_FONTSIZE = 12\n",
    "bc.LEGEND_FONTSIZE = 16\n",
    "\n",
    "# print('Generating Figure for GPU performance comparison')\n",
    "bc.yaxis = 'Performance-over-TX2'\n",
    "bc.BAR_WIDTH = None\n",
    "platform_list = ['TX2-fp32', 'TitanX-fp32', 'TitanX-int8', 'BitFusion-45nm']\n",
    "fig_name = os.path.join(fig_dir, 'performance-gpu.pdf')\n",
    "bc.plot(fig_name, all_fps_df, x_plot='Network', y_plot='fps', bar_plot='Name',\n",
    "        bar_list=platform_list, add_geomean=True, baseline='TX2-fp32')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Get 2x, 4x, and 8x area performance for bitfusion"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Can't find binary file /Users/hardik/workspace/projects/fraqnn-kaist/bitfusion/bitfusion/src/simulator/../../sram/cacti/cacti. Please clone and compile cacti first\n",
      "Can't find binary file /Users/hardik/workspace/projects/fraqnn-kaist/bitfusion/bitfusion/src/simulator/../../sram/cacti/cacti. Please clone and compile cacti first\n",
      "Can't find binary file /Users/hardik/workspace/projects/fraqnn-kaist/bitfusion/bitfusion/src/simulator/../../sram/cacti/cacti. Please clone and compile cacti first\n",
      "Can't find binary file /Users/hardik/workspace/projects/fraqnn-kaist/bitfusion/bitfusion/src/simulator/../../sram/cacti/cacti. Please clone and compile cacti first\n",
      "              Name    Network            fps\n",
      "0   BitFusion-45nm    AlexNet     212.390371\n",
      "1   BitFusion-45nm    CIFAR10    2428.876426\n",
      "2   BitFusion-45nm       LSTM   23726.551494\n",
      "3   BitFusion-45nm    LeNet-5   19495.457558\n",
      "4   BitFusion-45nm  RESNET-18     162.999245\n",
      "5   BitFusion-45nm        RNN   20274.568342\n",
      "6   BitFusion-45nm       SVHN    6072.000261\n",
      "7   BitFusion-45nm      VGG-7    2834.375734\n",
      "0      TitanX-fp32    AlexNet    5555.555556\n",
      "1      TitanX-int8    AlexNet   10884.353741\n",
      "2         TX2-fp32    AlexNet     459.188049\n",
      "4      TitanX-fp32       SVHN   27118.644068\n",
      "5      TitanX-int8       SVHN   39506.172840\n",
      "6         TX2-fp32       SVHN    1818.575156\n",
      "8      TitanX-fp32    CIFAR10    9142.857143\n",
      "9      TitanX-int8    CIFAR10   19207.683073\n",
      "10        TX2-fp32    CIFAR10     654.911590\n",
      "12     TitanX-fp32    LeNet-5   87912.087912\n",
      "13     TitanX-int8    LeNet-5  115107.913669\n",
      "14        TX2-fp32    LeNet-5    4230.188957\n",
      "16     TitanX-fp32      VGG-7    9411.764706\n",
      "17     TitanX-int8      VGG-7   20278.833967\n",
      "18        TX2-fp32      VGG-7     660.348364\n",
      "20     TitanX-fp32  RESNET-18    2946.593002\n",
      "21     TitanX-int8  RESNET-18    6837.606838\n",
      "22        TX2-fp32  RESNET-18     220.442505\n",
      "24     TitanX-fp32        RNN   80000.000000\n",
      "25     TitanX-int8        RNN   84210.526316\n",
      "26        TX2-fp32        RNN   11662.860633\n",
      "28     TitanX-fp32       LSTM   88888.888889\n",
      "..             ...        ...            ...\n",
      "2   BitFusion (1x)       LSTM   77622.425119\n",
      "3   BitFusion (1x)    LeNet-5   35489.466283\n",
      "4   BitFusion (1x)  RESNET-18     202.331251\n",
      "5   BitFusion (1x)        RNN   61031.896795\n",
      "6   BitFusion (1x)       SVHN   10249.209530\n",
      "7   BitFusion (1x)      VGG-7    7748.919510\n",
      "0   BitFusion (2x)    AlexNet     520.874874\n",
      "1   BitFusion (2x)    CIFAR10    9025.230031\n",
      "2   BitFusion (2x)       LSTM  153689.508770\n",
      "3   BitFusion (2x)    LeNet-5   65992.443865\n",
      "4   BitFusion (2x)  RESNET-18     404.647825\n",
      "5   BitFusion (2x)        RNN  121928.915442\n",
      "6   BitFusion (2x)       SVHN   20436.941816\n",
      "7   BitFusion (2x)      VGG-7   15475.923332\n",
      "0   BitFusion (4x)    AlexNet    1041.574715\n",
      "1   BitFusion (4x)    CIFAR10   17967.272613\n",
      "2   BitFusion (4x)       LSTM  301375.023545\n",
      "3   BitFusion (4x)    LeNet-5   58504.764482\n",
      "4   BitFusion (4x)  RESNET-18     808.023838\n",
      "5   BitFusion (4x)        RNN  243368.216111\n",
      "6   BitFusion (4x)       SVHN   27094.580407\n",
      "7   BitFusion (4x)      VGG-7   27365.396456\n",
      "0   BitFusion (8x)    AlexNet    1508.588583\n",
      "1   BitFusion (8x)    CIFAR10   22846.763899\n",
      "2   BitFusion (8x)       LSTM  544328.774580\n",
      "3   BitFusion (8x)    LeNet-5   47186.226341\n",
      "4   BitFusion (8x)  RESNET-18    1111.650570\n",
      "5   BitFusion (8x)        RNN  457875.457875\n",
      "6   BitFusion (8x)       SVHN   25897.678273\n",
      "7   BitFusion (8x)      VGG-7   32261.186566\n",
      "\n",
      "[64 rows x 3 columns]\n"
     ]
    }
   ],
   "source": [
    "bf_fps_df = pandas.DataFrame(columns=['Name', 'Network', 'fps'])\n",
    "df_list = []\n",
    "for factor in [1, 2, 4, 8]:\n",
    "    config_file = 'bf_e_conf.ini'\n",
    "    verbose = False\n",
    "    bf_scaled = Simulator(config_file, verbose)\n",
    "    bf_scaled.accelerator.M *= factor\n",
    "    bf_scaled.accelerator.sram['act'] *= factor\n",
    "    bf_scaled.accelerator.sram['wgt'] *= factor\n",
    "    bf_scaled.accelerator.sram['out'] *= factor\n",
    "#     if bf_scaled.accelerator.mem_if_width * factor > 1024:\n",
    "#         bf_scaled.accelerator.mem_if_width = 1024\n",
    "#     else:\n",
    "#         bf_scaled.accelerator.mem_if_width = bf_scaled.accelerator.mem_if_width * factor\n",
    "#     bf_scaled.accelerator.mem_if_width = bf_scaled.accelerator.mem_if_width * factor\n",
    "    bf_scaled.accelerator.mem_if_width = 4096\n",
    "    frequency = bf_scaled.accelerator.frequency\n",
    "#     print(bf_scaled)\n",
    "    bf_e_scaled_results = check_pandas_or_run(bf_scaled, bf_e_sim_sweep_df,\n",
    "                                       bf_e_sim_sweep_csv, batch_size=batch_size)\n",
    "    bf_e_scaled_results = bf_e_scaled_results.groupby(['Network', 'Batch size']+accelerator_constants,\n",
    "                                    as_index=False).agg(np.sum)\n",
    "    bf_e_scaled_results['fps'] = batch_size * frequency / (bf_e_scaled_results.Cycles)\n",
    "    bf_e_scaled_results['Name'] = 'BitFusion ({}x)'.format(factor)\n",
    "    bf_e_scaled_results = bf_e_scaled_results[['Name', 'Network', 'fps']]\n",
    "    df_list.append(bf_e_scaled_results)\n",
    "    \n",
    "bf_fps_df = pandas.concat(df_list)\n",
    "all_fps_df = pandas.concat([all_fps_df, bf_fps_df])\n",
    "print(all_fps_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x252 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "bc = BarChart()\n",
    "bc.LABEL_BAR = True\n",
    "bc.BAR_WIDTH = None\n",
    "bc.YAXIS_MAX = 16\n",
    "bc.FIG_WIDTH = 16\n",
    "bc.TOP_LABEL_FONTSIZE = 12\n",
    "bc.LEGEND_FONTSIZE = 16\n",
    "bc.ISTIMES = True\n",
    "bc.TOP_ROTATE = True\n",
    "bc.LEGEND_LOCATION = \"upper center\"\n",
    "bc.LEGEND_NCOLUMN = 5\n",
    "bc.MV_LEGEND_OUTSIDE_Y = 1.6\n",
    "bc.MV_LEGEND_OUTSIDE_X = 0.5\n",
    "bc.LOG_SCALE = False\n",
    "\n",
    "# print('Generating Figure for GPU performance comparison')\n",
    "bc.yaxis = 'Performance'\n",
    "bc.BAR_WIDTH = None\n",
    "fig_name = os.path.join(fig_dir, 'performance-eyeriss-comparision.pdf')\n",
    "platform_list = ['TX2-fp32', 'TitanX-fp32', 'TitanX-int8', 'BitFusion (8x)']\n",
    "bc.plot(fig_name, all_fps_df, x_plot='Network', y_plot='fps', bar_plot='Name',\n",
    "        bar_list=platform_list, add_geomean=True, baseline='TX2-fp32')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "3.7.6-final"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}