notebooks/TrueBinary_Plots.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "\n",
    "# We'll also import seaborn, a Python graphing library\n",
    "import warnings # current version of seaborn generates a bunch of warnings that we'll ignore\n",
    "warnings.filterwarnings(\"ignore\")\n",
    "import seaborn as sns\n",
    "\n",
    "# import matplotlib and set inline\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib\n",
    "\n",
    "from mpl_toolkits.mplot3d import Axes3D\n",
    "from sklearn.decomposition import PCA\n",
    "import numpy as np\n",
    "sns.set()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Bitwidth Comparison"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 256,
   "metadata": {},
   "outputs": [],
   "source": [
    "# unipolar quantization\n",
    "bits = ['1', '2', '3']\n",
    "accuracy = [58.1, 67.2, 70.2]\n",
    "runtime = [243.77, 387.21, 609.97]\n",
    "quant_type_pos = ['unipolar', 'unipolar', 'unipolar']\n",
    "\n",
    "# bipolar quantization\n",
    "runtime_neg = [184.07, 271.42, 423.48]#, 2420]\n",
    "bits_neg = ['1', '2', '3']#, '32']\n",
    "accuracy_neg = [56.2, 64.0, 68.5]#, 72.7]\n",
    "quant_type_neg = ['bipolar', 'bipolar', 'bipolar']#, 'full precision']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 257,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_pos = {'Bitwidth': bits, 'Accuracy (top-1)': accuracy, 'Runtime (ms)': runtime, 'Quantization Type': quant_type_pos}\n",
    "data_neg = {'Bitwidth': bits_neg, 'Accuracy (top-1)': accuracy_neg, 'Runtime (ms)': runtime_neg, 'Quantization Type': quant_type_neg}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 258,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_pos = pd.DataFrame(data_pos)\n",
    "df_neg = pd.DataFrame(data_neg)\n",
    "df = pd.concat([df_pos, df_neg], ignore_index=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 259,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Bitwidth</th>\n",
       "      <th>Accuracy (top-1)</th>\n",
       "      <th>Runtime (ms)</th>\n",
       "      <th>Quantization Type</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>58.1</td>\n",
       "      <td>243.77</td>\n",
       "      <td>unipolar</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>67.2</td>\n",
       "      <td>387.21</td>\n",
       "      <td>unipolar</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>70.2</td>\n",
       "      <td>609.97</td>\n",
       "      <td>unipolar</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>56.2</td>\n",
       "      <td>184.07</td>\n",
       "      <td>bipolar</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2</td>\n",
       "      <td>64.0</td>\n",
       "      <td>271.42</td>\n",
       "      <td>bipolar</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>3</td>\n",
       "      <td>68.5</td>\n",
       "      <td>423.48</td>\n",
       "      <td>bipolar</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  Bitwidth  Accuracy (top-1)  Runtime (ms) Quantization Type\n",
       "0        1              58.1        243.77          unipolar\n",
       "1        2              67.2        387.21          unipolar\n",
       "2        3              70.2        609.97          unipolar\n",
       "3        1              56.2        184.07           bipolar\n",
       "4        2              64.0        271.42           bipolar\n",
       "5        3              68.5        423.48           bipolar"
      ]
     },
     "execution_count": 259,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 260,
   "metadata": {},
   "outputs": [],
   "source": [
    "neg_acc = df.loc[df['Quantization Type'] == 'bipolar']['Accuracy (top-1)']\n",
    "pos_acc = df.loc[df['Quantization Type'] == 'unipolar']['Accuracy (top-1)']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 264,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "#counts = np.array([56000,900,5000,6000,30000,3500,300]).astype(float)\n",
    "#perc =  counts/counts.sum()\n",
    "#df = pd.DataFrame({\"range\": list(\"ABCDEFG\"),\"count\":counts, \"perc\":perc})\n",
    "\n",
    "fig = plt.figure()\n",
    "#ax = df['runtime'].plot(kind=\"bar\", alpha=0.7)\n",
    "ax = sns.barplot(x='Bitwidth', y='Runtime (ms)', hue='Quantization Type', data=df)\n",
    "ax.set_ylim([0,2700])\n",
    "#ax = sns.scatterplot(x='Runtime (ms)', y='Accuracy (top-1)', size='Bitwidth', hue='Quantization Type', data=df)\n",
    "#ax.set(xlabel='Activation Bits', ylabel='Runtime (ms)')\n",
    "ax2 = ax.twinx()\n",
    "ax2.grid(False)\n",
    "ax2 = sns.scatterplot(x='Bitwidth', y='Accuracy (top-1)', hue='Quantization Type', data=df, marker='X', s=60)\n",
    "ax2.get_legend().remove()\n",
    "ax2.set_ylim([50, 80])\n",
    "ax.axhline(y=2420, color='g', xmax=0.2, linestyle='--', label=\"Full Precision\")\n",
    "ax2.axhline(y=72.7, color='g', xmin=0.8, linestyle='--')\n",
    "ax.scatter(x=0, y=0, alpha=1, marker='X', color='black', label='Accuracy', s=50, linewidths=.01)\n",
    "ax.legend(loc='upper center')\n",
    "#ax2.plot(ax.get_xticks(),pos_acc,marker='o', c='navy', linewidth=2)\n",
    "#ax2.plot(ax.get_xticks(),neg_acc,marker='o', c='yellow', linewidth=2)\n",
    "\n",
    "#ax.set_xticklabels(df['Bitwidth'], rotation=0)\n",
    "#ax.set_ylim(0,1.3*df[\"Runtime (ms)\"].max())\n",
    "#ax2.set_ylim(0.8*df[\"Accuracy (top-1)\"].min(),1.3*df[\"Accuracy (top-1)\"].max())\n",
    "#ax2.set(ylabel='Imagenet Top-1 Accuracy')\n",
    "#plt.legend()\n",
    "plt.tight_layout()\n",
    "plt.savefig('/data/jwfromm/bitwidth_comp.png', dpi=900)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Effect of optimizations on speedup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 266,
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizations = ['unoptimized', 'tiling', 'vectorize', 'parallel', 'layer fusion']\n",
    "speedups = [0.5, 0.8, 0.9, 2.0, 10.0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 267,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = {'Optimization': optimizations, 'Relative Speedup': speedups}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 268,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.DataFrame(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 269,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Optimization</th>\n",
       "      <th>Relative Speedup</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>unoptimized</td>\n",
       "      <td>0.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>tiling</td>\n",
       "      <td>0.8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>vectorize</td>\n",
       "      <td>0.9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>parallel</td>\n",
       "      <td>2.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>layer fusion</td>\n",
       "      <td>10.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Optimization  Relative Speedup\n",
       "0   unoptimized               0.5\n",
       "1        tiling               0.8\n",
       "2     vectorize               0.9\n",
       "3      parallel               2.0\n",
       "4  layer fusion              10.0"
      ]
     },
     "execution_count": 269,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 270,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# This plot sucks\n",
    "ax = sns.barplot(x='Optimization', y='Relative Speedup', data=df)\n",
    "ax.axhline(y=1.0, color='y')\n",
    "plt.savefig('/data/jwfromm/optimizations.png', dpi=900)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Layer wise speed-ups on VGG"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 271,
   "metadata": {},
   "outputs": [],
   "source": [
    "layers = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]\n",
    "\n",
    "#N = 1\n",
    "bits = (np.zeros(len(layers)) + 1).astype(np.int32)\n",
    "speedup = [20.1, 16.9, 32.6, 33.7, 16.1, 34.6, 35.6, 10.6, 10.6, 10.8, 20.1, 3.1, 0.7]\n",
    "n1_data = {'Layer': layers, 'Runtime (ms)': speedup, 'Bitwidth': bits}\n",
    "n1_df = pd.DataFrame(n1_data)\n",
    "\n",
    "# N = 2\n",
    "bits = (np.zeros(len(layers)) + 2).astype(np.int32)\n",
    "speedup = [20.1, 31.9, 58.6, 59.8, 28.9, 56.2, 57.0, 16.9, 16.8, 16.9, 20.3, 3.2, 0.8]\n",
    "n2_data = {'Layer': layers, 'Runtime (ms)': speedup, 'Bitwidth': bits}\n",
    "n2_df = pd.DataFrame(n2_data)\n",
    "\n",
    "#N = 3\n",
    "bits = (np.zeros(len(layers)) + 3).astype(np.int32)\n",
    "speedup = [20.3, 53.1, 99.9, 101.2, 46.4, 91.7, 92.5, 26.8, 26.9, 27.1, 20.8, 3.2, 0.9]\n",
    "n3_data = {'Layer': layers, 'Runtime (ms)': speedup, 'Bitwidth': bits}\n",
    "n3_df = pd.DataFrame(n3_data)\n",
    "\n",
    "#N = 32\n",
    "#bits = (np.zeros(len(layers)) + 32).astype(np.int32)\n",
    "#speedup = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]\n",
    "#n32_data = {'Layer': layers, 'Relative Speedup': speedup, 'Bitwidth': bits}\n",
    "#n32_df = pd.DataFrame(n32_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 272,
   "metadata": {},
   "outputs": [],
   "source": [
    "joint_df = pd.concat([n1_df, n2_df, n3_df], ignore_index=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 273,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZIAAAESCAYAAADXMlMiAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvOIA7rQAAIABJREFUeJzt3XtcVHX+P/DXXJhBhGEAEVFSk8qo3USh7KqFFVbcNA2XMC+VlVtifr2Sys1UkC3Zn1q4bfr1sva1tURpW21TSy1dS61lwctDyTBAYLijXGbm/P4wp8gBB885c8HX8/HgkTNn5vN5n6E5L87nnPM5CkEQBBAREV0npaMLICIi18YgISIiURgkREQkCoOEiIhEYZAQEZEoDBIiIhKFQUJERKIwSIiISBQGCRERicIgISIiURgkREQkCoOEiIhEYZAQEZEoakcXILeamiaYzZzgmIjIFkqlAj4+Pbv0nm4fJGazwCAhIpIRh7aIiEgUBgkREYnS7Ye2iIikJAgCamoq0draDMBVh80V0Gjc4ePjD4VCIbo1BgkRURc0NtZBoVAgICAICoVrDuoIghm1tVVobKyDl5dedHuu+SkQETnIpUuN8PLSu2yIAIBCoYSXlw8uXWqUpD3X/SSIiBzAbDZBpXL9wRyVSg2z2SRJW3YJkszMTERERGDw4ME4deqU5fni4mLEx8cjMjIS8fHx+OGHH2xaRkTkSFIcV3A0KdfBLkEyatQobN68Gf369Wv3fEpKChISErBr1y4kJCRg8eLFNi0j56Xz1sLf38vqj85b6+jyiGTx0EN3Y/LkBEya9Ac891w8vvxyn2XZ7NkzUFdXCwBYv/69LrW7ffvf8cknO6wui4i4HwDQ0NCAbdv+z/J8WVkpEhKe7uIaiGOX/bPw8PCrnjMYDCgsLMS6desAAFFRUcjIyEB1dTUEQehwma+vrz1Kpuuk1WgweV2S1WXrp+QAaLFvQUR24ObmhvXr/wYAOHfuByQlvYIRIx4GAGRn/9nyug0b3sfkyS/Y3G5c3LhrvqaxsQHbtm3F00/Hd61oCTnsGElZWRkCAgKgUqkAACqVCr1790ZZWVmny4iInFljYyO8vLwsj8eNi4bBUIXVq3PQ1taGyZMTMHv2DHzxxV5kZi4BAPzrX7vw6KMPwmg0wmg0YsyYJ2EymfDXv+Zi48b1AIDS0p/w8stTMXHiM1i9OsfS/urVOSgvL8PkyQlYvjwDwOUZPbKzl+O55+IxffoLqK+vk3WdXf+I0TX4+Xk6ugSn0Wpsg0btZvPzcvD397r2i4icWEWFEmp1+7/B29raMGVKAoxGI0pLS5GWtqTda9RqJZKSXse2bf+HTZs+AADU1tZg7drVUKuV+O67Yxg0KBinThVBpVLh1ltvg1brBqVSAZVKAbVaiT//+U+IjY1DdHQcduzYjtbWVqjVSsyY8TqKi89Y2i0tLUVp6XlERy/B/PnJyMx8E59+mo9nn5141boolUpJvpMOC5LAwEBcuHABJpMJKpUKJpMJFRUVCAwMhCAIHS7rKoOhkXNt/czf38vqsNP6KTmorGyQrI/OSNUPkaOYzWYYjeZ2z7m5uWHdustDWz/+eA5//OOLGDz4Dvj79wYAGI2/vOfKfz09vaFUKlFRUYkTJ4owfvwf8O2330ClUuOuu0JhNJphNgswmQQYjWZ8991xpKcvh9FoxqOPjsayZRkwGs0wmcwQBMHSrslkRkBAHwwefAeMRjMGDw7BqVMnr6r5yrr89jupVCq6/Ae4w4a2/Pz8EBISgvz8fABAfn4+QkJC4Ovr2+kyIiJn1r//APTpE4j//Of7a752yJCh2Lv3c+h03ggPvxvHjx/Dd98dRWhomKga3Nx+GWFQKi//MS4nuwTJkiVLMGLECJSXl2PKlCl46qmnAACpqanYtGkTIiMjsWnTJqSlpVne09kyIiJnVV1tQEnJjwgKCrpqmZubG1pafjnhZMiQodi8+X8RGjoUvr5+qKurxenTp3D77SFXvfeuu4bgX//aBQD47LN/QhAuj7R4eHigqalJprWxjV2GthYuXIiFCxde9XxwcDA+/PBDq+/pbBkRkTO5chBdEAS0tbVi0qTncdttt1/1urFjn8GUKQno27cfsrP/jNDQYaiouIAhQ4YCAAYPvh0//XQeavXVm+akpNlIS1uIDz7YhHvvvR8ajQYA4O2tR3j43Zg48RnceefvMWnS8/KurBUK4UqsdVM8RvILex0j6ez0Xx4jIVdXXn4OffoMcHQZkrC2Li51jISIiLoHBgkREYnCICEiIlEYJEREJAqDhIiIRGGQEBGRKAwSIiISpdtP2khEZA9eOne4a6Wf/LS5pQ0N9c3XfN2qVSvxxRd7UFZWig0bPsCgQbdIXktHGCRERBJw17ohYe5mydv9W9azaMC1g+Shhx7G+PET8Mc/vih5DdfCICFyAjpvLbQ/T3nxay2traiv483A6NqGDAl1WN8MEiIn0NGdJXlXSXIFDBKiTnS0pwBwb4HoCgYJUSd4D3qia+Ppv0REJAr3SIiIJNDc0oa/ZT0rS7u2WLlyBb74Yi+qqw2YOfOP0Om8sWnTVsnrsYZBQkQkgYb6ZptO05XLzJlzMHPmHIf0zaEtIiIShUFCRESiMEiIiEgUBgkREYnCICEiIlEYJEREJApP/yUikoCPtwZqjVbydo2tLaipa73m6+rqapGRsRg//XQebm5uCArqjzlzkuHj4yN5Tb/FICEikoBao8W3WS9I3m7Y3PcAXDtIFAoFEhKew7Bh4QCA1atz8O67/w8LFiyWvKbf4tAWEVE3oNN5W0IEAO6883coLy+3S98MEiKibsZsNuPjj7fhwQdH2KU/BgkRUTfz9tsr4OHRA08//Yxd+uMxEiKibmTVqpU4f/5HZGa+DaXSPvsKDBIiom4iN3c1Tp4swooVOdB0cEM2OTBIiIi6gbNnz2DjxnW46ab+ePnlqQCAwMC+WLYsW/a+GSRERBIwtrb8fKqu9O3aYtCgYBw48I3k/dvCKYJk7969yMnJgSAIEAQBr776Kh5//HEUFxdj/vz5qK2thV6vR2ZmJgYOHOjocomIrnL5osFrX+/RHTk8SARBwNy5c7F582bcdtttOHHiBP7whz/g0UcfRUpKChISEhAbG4u8vDwsXrwYGzZscHTJRET0K05x+q9SqURDQwMAoKGhAb1790ZNTQ0KCwsRFRUFAIiKikJhYSGqq6sdWSoREf2Gw/dIFAoFVq5cienTp8PDwwNNTU1Yu3YtysrKEBAQAJVKBQBQqVTo3bs3ysrK4Ovr6+CqiYjoCocHidFoRG5uLtasWYOwsDB8++23mDlzJrKysiRp38/PU5J2ujt/f69u1Y+92GN9uttn5uoqKpRQq51iMEc0pVIpyf9fDg+SoqIiVFRUICwsDAAQFhaGHj16QKvV4sKFCzCZTFCpVDCZTKioqEBgYGCX2jcYGmE2C3KU7nI6+x+msrJB9j6k7Mde7LU+9vjdkDTMZjOMRrOjy5CE2Wy+6v8vpVLR5T/AHR6rffr0QXl5Oc6ePQsAOHPmDAwGAwYMGICQkBDk5+cDAPLz8xESEsJhLSIiJ+PwPRJ/f3+kpqYiKSkJCoUCALB06VLo9XqkpqZi/vz5WLNmDXQ6HTIzMx1cLRGRdTpvLbQyXE3e0tqK+jrbriVZsOB/UFpaCqVSgR49PPD663Nw662DJa/ptxweJAAQExODmJiYq54PDg7Ghx9+6ICKiIi6RqvRYPK6JMnbXT8lB4BtQfLGG2nw9Lw8LLV//z4sW5aO99/fLHlNv+XwoS0iIpLGlRABgMbGRigUnLSRiIi6aPnyDPz734cAANnZf7ZLn9wjISLqRubPX4SPPvoE06ZNx5o1OXbpk0FCRNQNjR79FI4e/RZ1dbWy98UgISLqBi5evIgLF365R/uBA19Cp9NBp/OWvW8eIyEikkBLa+vPZ1hJ364tmpsvYdGi+WhuvgSlUvXzJRNvWy6rkBODhIhIApev9bDtNF05+Pr6Ye3a9Q7pm0NbREQkCoOEiIhEYZAQEXWRILj+RLBSrgODhIioC5RKFUwmo6PLEM1kMkKpVEnSVqcH26urq5GXl4d9+/bhxIkTaGxshKenJ26//XaMGDECY8aM4Wy8RHRD6dHDEw0NtdDr/ew2BYnUBMGMhoYa9Oghzf2aOgyS7Oxs7Ny5EyNHjsS4ceMQHByMnj17oqmpCWfOnMGRI0cwZswYREdHY/bs2ZIUQ0Tk7Dw9vVFTU4kLF84DcNUhLgU0Gnd4ekpzjUmHQdKnTx989tln0FiZFvmOO+5AdHQ0WlpaODsvEd1QFAoFfH17O7oMp9JhkCQmJl7zzVqt1qbXERFR92XTAN+hQ4dQUlICAKioqMC8efOwYMECVFZWylocERE5P5uCJC0tDSrV5aP7mZmZMBqNUCgUWLRokazFERGR87NpipQLFy6gb9++MBqNOHDgAPbs2QM3Nzc89NBDctdHREROzqYg8fT0RFVVFU6fPm05e6u1tRVGo+ufS01EROLYFCSJiYkYN24c2trakJycDAA4evQoBg0aJGtxRETk/GwKkmnTpuGxxx6DSqVC//79AQABAQFYsmSJrMUREZHzs3ka+ZtvvrnTx0REdGOyKUhOnDiBpUuX4sSJE7h48SKAyxN+KRQKFBQUyFogERE5N5uCZNasWXj88cexcOFCuLu7y10TERG5EJuCpKqqCklJSXa5ZSMREbkWmy5IjIuLw86dO+WuhYiIXJDNZ23Fx8cjNzcXfn5+7ZZt2LBBlsKIiMg12BQkM2bMQFBQEB577DFotVq5ayIiIhdiU5AUFRXh8OHDVqeUJyKiG5tNx0jCw8Nx5swZuWshIiIXZNMeSVBQEKZOnYrHHnvsqmMkSUlJshRGRESuwaYgaW5uxsMPP4y2tjaUl5fLXRMREbkQm4Jk2bJlshbR0tKCpUuX4uuvv4ZWq0VoaCgyMjJQXFyM+fPno7a2Fnq9HpmZmRg4cKCstRARUdd0GCRVVVXo1avXNRuw9XWdWbFiBbRaLXbt2gWFQoGqqioAQEpKChISEhAbG4u8vDwsXryYpxsTETmZDg+2T5o0CampqTh27BjMZnO7ZWazGcePH0dqaiomT54sqoCmpiZs37693ZXzvXr1gsFgQGFhIaKiogAAUVFRKCwsRHV1taj+iIhIWh3ukXz88cfYunUrFi1ahPPnz+Omm25Cz5490dTUhPPnz6N///6Ij4+33J/kepWUlECv12PVqlU4fPgwevbsiaSkJLi7uyMgIMByi1+VSoXevXujrKwMvr6+ovokIiLpdBgkGo0GiYmJSExMRFlZGU6dOoX6+nrodDrcfvvtCAgIkKQAk8mEkpIS3HHHHZg3bx6+++47vPzyy8jJyZGkfT8/T0na6e78/b1sfq3Z2Aal2k32flyBPdanu31m1P3YdLA9MDAQgYGBshQQGBgItVptGcIaMmQIfHx84O7ujgsXLsBkMkGlUsFkMqGioqLLdRgMjTCbBTlKdzmdbZAqKxu61M63WS9YXRY2971O39uVfpzBtTbiUq2PVL8bIrGUSkWX/wC36YJEOfn6+mL48OE4ePAgAKC4uBgGgwEDBw5ESEgI8vPzAQD5+fkICQnhsNY1+Hhr4O/vZfWHiEgONt8hUU5paWlITk5GZmYm1Go1srKyoNPpkJqaivnz52PNmjXQ6XTIzMx0dKlOT63RXvfeAhHR9XCKILnpppuwcePGq54PDg7Ghx9+6ICKiIjIVl0a2jKbzaioqJCrFiIickE27ZHU19cjLS0Nu3btglqtxvHjx/H555/j+++/x+uvvy53jUTdgo+3BmoNb8NA3Y9NQZKSkgKdToc9e/bgqaeeAgAMHToUmZmZDBJyefbawPP4FXVXNgXJ119/jf3798PNzc1y9bmvry8MBoOsxRHZAzfwROLYdIzEy8sLNTU17Z4rLS2Fv7+/LEUREZHrsClIxo8fjxkzZuDQoUMwm804duwY5s2bhwkTJshdHxEROTmbhrZefPFFaLVapKenw2g0Ijk5GfHx8Zg0aZLc9RERkZOzKUgUCgUmTZrE4CAioqvYfEHiTz/9hBMnTuDixYvtno+Ojpa8KCIich02BUlubi7WrFmD4OBguLu7W55XKBQMEgnovLXQajRWl7W0tqK+rsXOFRER2c6mIHn//fexbds23HLLLXLXc0PSajSYvC7J6rL1U3IAMEiIyHnZdNaWXq9Hv3795K6FiIhckE17JMnJyVi0aBEmTZoEPz+/dsv69u0rS2FEROQabAqStrY2HDx40HJvkCsUCgWKiopkKYyIiFyDTUGSlpaGWbNm4cknn2x3sJ2IiMimIDGZTBg7dixUKpXc9RARkYux6WD71KlTsXbtWggC731ORETt2bRHsnHjRlRVVSE3Nxd6vb7dsn379slRFxERuQibgmTFihVy10FERC7KpiC555575K6DiIhcVIdB8s477+CVV14BAOTk5HTYQFKS9SuyiYjoxtBhkJSXl1v9NxER0a91GCRpaWmWfy9btswuxRDZihNdEjkPm4+R/Pvf/77q+fvuuw9ff/215EURXQsnuiRyHjZdR9LW1mb1ObPZLHlBRETkWjrdI0lISIBCoUBrayueffbZdsvKy8sxdOhQWYsjIiLn12mQjB8/HoIg4D//+Q/GjRtneV6hUMDPzw/33nuv7AUSEZFz6zRIxowZAwAYMmQIgoOD7VIQERG5FpsOtgcHB+PAgQMoKiq66p7tvI6EiOjGZlOQpKen49NPP8Xw4cPRo0cPuWsiIiIXYlOQ5OfnIy8vD4GBgXLXQ0RELsam0399fHzg5eUldy1EROSCbAqSKVOmYPbs2Th27BhKSkra/Uhp1apVGDx4ME6dOgUAOH78OGJiYhAZGYmpU6fCYDBI2h8REYln09BWamoqgKvvPSLlPdv/+9//4vjx4+jXrx8AwGw2Y86cOVi2bBnCw8OxZs0aZGdnc7oWIiInY1OQnDhxQtYiWltbkZ6ejj/96U947rnnAAAFBQXQarUIDw8HAEyYMAGjRo1ikBARORmbgkRuOTk5iImJQVBQkOW5srIy9O3b1/LY19cXZrMZtbW1V92lsTN+fp6S1uoI/v7yH5+yRx/sx3n7IBLDpiC5MlWKNZs3bxZVwLFjx1BQUIDZs2eLaqcjBkMjzGbnvtf8tTYUlZUNkrXlyD662o+YGpxxfezxuyESS6lUdPkPcJuCZPz48e0eV1ZWYtu2bYiOju5SZ9YcOXIEZ86cwahRowBcnsPr+eefx8SJE1FaWmp5XXV1NZRKZZf2RoiISH42BcmVqVJ+LTIyEgsWLMCrr74qqoBp06Zh2rRplscRERF49913ccstt2Dr1q345ptvEB4ejg8++ACjR48W1RcREUnvuo+RBAQE4OTJk1LW0o5SqURWVhZSUlLQ0tKCfv36YcWKFbL1R0RE18emIPn73//e7nFzczN2796N0NBQyQvas2eP5d/Dhg3Dzp07Je+DiIikY1OQ5OXltXvs4eGBoUOHYvLkyXLURERELsSmINm4caPcdRARkYuyaYoUa06ePIkZM2ZIWQsREbmgTvdILl26hNzcXJw4cQIDBgzAa6+9hpqaGixfvhxfffUV4uLi7FUnkUvw0rnDXevm6DKI7KrTIElPT0dhYSEefPBBfPnllzh16hTOnj2LuLg4ZGRkwNfX1151ErkEd60bEuZav0j3b1nP2rkaIvvoNEj279+PvLw8+Pn5YeLEiXj44YexadMmy/xXRK6CewpE8uk0SC5evAg/Pz8AQJ8+feDh4cEQIZfEPQUi+XQaJCaTCYcOHYIg/DJX1W8f33ffffJVRw7Fv+KJyBadBomfnx+Sk5Mtj/V6fbvHCoUCn3/+uXzVkUN19Fc8/4Inol/rNEh+fZU5ERGRNU5xPxLqGg45EZEzYZDYiY+3BmqNVpK2eOCYiJwJg8RO1Botvs16weqysLnv2bkaIiLpXPcUKURERACDhIiIRGKQEBGRKAwSIiIShQfbyWlJeaYbEcmHQUJOi2e6EbkGDm0REZEoDBIiIhKFQ1tENwidtxZajcbqspbWVtTXtbAfui4MEqIbhFajweR1SVaXrZ+SA0CaDW9364eujUNbREQkCoOEiIhEYZAQEZEoDBIiIhKFQUJERKIwSIiISBQGCRERicIgISIiURgkREQkisOvbK+pqcHcuXPx448/QqPRYMCAAUhPT4evry+OHz+OxYsXo6WlBf369cOKFSvg5+fn6JKJiOhXHL5HolAo8MILL2DXrl3YuXMnbrrpJmRnZ8NsNmPOnDlYvHgxdu3ahfDwcGRnZzu6XCIi+g2HB4ler8fw4cMtj0NDQ1FaWoqCggJotVqEh4cDACZMmIB//vOfjiqTiG5QOm8t/P29rvrRefOma1c4fGjr18xmM7Zs2YKIiAiUlZWhb9++lmW+vr4wm82ora2FXq+3uU0/P0+0tpmgcVNZXW5qa4XKzfoMoq3GVmjUHS1rg0btZnMdYvj7e3WLPtiP8/ZxI/RzvdsBAFYnh1w/JQf+/h2/50biVEGSkZEBDw8PJCYm4rPPPpOkTYOhEX5+nkiYu9nq8r9lPdvpXfg6m120srLB5jrEfHl+248cX0Rr62Kvfjri7J+ZvfrpymfWmWvV1t378ff3uu7tgK19dAdKpQJ+fp5deo/TBElmZibOnTuHd999F0qlEoGBgSgtLbUsr66uhlKp7NLeCBERyc/hx0gA4K233kJBQQFWr14Nzc83qvnd736H5uZmfPPNNwCADz74AKNHj3ZkmUREZIXD90hOnz6N3NxcDBw4EBMmTAAABAUFYfXq1cjKykJKSkq703+JiMi5ODxIbr31Vpw8edLqsmHDhmHnzp12roiIiLrCKYa2iIjIdTl8j6Q78dK5w11rn1OCiazx8dZArZH/+obu1g+JwyCRkLvWrdPTC+lqDF9pqTXa6zqN9Ubvh8RhkJBDMXyJXB+PkRARkSgMEiIiEoVBQkREojBIiIhIFAYJERGJwiAhIiJRGCRERCQKg4SIiERhkBARkSi8sp2IHIrT5Lg+BgmRC7LHxtdeG3hOk+P6GCRELqijja+UG15u4MlWPEZCRESiMEiIiEgUBgkREYnCICEiIlEYJEREJAqDhIiIRGGQEBGRKAwSIiIShUFCRESiMEiIiEgUBgkREYnCICEiIlEYJEREJApn/yUiklBH0++b2lqhctNYfU+rsRUatfVlLa2tqK9rkbRGqTFIiIgk1NkU/99mvWD1PWFz38PkdUlWl62fkgPAuYOEQ1tERCSK0wdJcXEx4uPjERkZifj4ePzwww+OLomIiH7F6Ye2UlJSkJCQgNjYWOTl5WHx4sXYsGGDo8siInI5Om8ttBrpj8U4dZAYDAYUFhZi3bp1AICoqChkZGSguroavr6+NrWhVCoAAL18enb4Go3Or8NlvTw77udK2+1e76L9WOujs36ceV26Wz/83ThvP4763Xh6aqG1ckAf6PygPgDM/jDN6vPZ41OgVLZ2uE6dUQiCIHT5XXZSUFCAefPm4ZNPPrE89+STT2LFihW48847HVgZERFd4fTHSIiIyLk5dZAEBgbiwoULMJlMAACTyYSKigoEBgY6uDIiIrrCqYPEz88PISEhyM/PBwDk5+cjJCTE5uMjREQkP6c+RgIAZ86cwfz581FfXw+dTofMzEwMGjTI0WUREdHPnD5IiIjIuTn10BYRETk/BgkREYnCICEiIlEYJEREJAqD5Gf2mBwyMzMTERERGDx4ME6dOiV5+wBQU1ODF198EZGRkYiOjsarr76K6upqWfqaPn06YmJiEBcXh4SEBBQVFcnSzxWrVq2S9bOLiIjA6NGjERsbi9jYWOzfv1/yPlpaWpCSkoLHH38c0dHRWLRokeR9nD9/3rIOsbGxiIiIwD333CN5PwCwd+9exMXFITY2FjExMdi9e7cs/ezbtw9jxoxBdHQ0EhMTUVJSIrrNjr6PUm8LOupH6u2Btfbstj0QSBAEQZg4caKwfft2QRAEYfv27cLEiRMl7+PIkSNCaWmp8MgjjwgnT56UvH1BEISamhrh0KFDlsfLly8XFixYIEtf9fX1ln9/9tlnQlxcnCz9CIIgFBQUCM8//7ysn52cbV+RkZEhvPnmm4LZbBYEQRAqKytl7U8QBGHJkiVCWlqa5O2azWYhPDzc8pkVFRUJoaGhgslkkrSf2tpa4Z577hHOnj0rCMLl7+fUqVNFt9vR91HqbUFH/Ui9PbDWnr22B9wjwS+TQ0ZFRQG4PDlkYWGh5MkdHh4u+1X5er0ew4cPtzwODQ1FaWmpLH15eXlZ/t3Y2AiFouuTvdmitbUV6enpSE1NlaV9e2lqasL27duRlJRk+ax69eola5+tra3YuXMnnn76aVnaVyqVaGhoAAA0NDSgd+/eUCql3aycO3cOvXr1ws033wwAGDlyJA4cOCD6+2nt+yjHtqCj773U2wNr7dlre+DUs//aS1lZGQICAqBSqQAAKpUKvXv3RllZmUtfRW82m7FlyxZERETI1scbb7yBgwcPQhAEvPfee7L0kZOTg5iYGAQFBcnS/q/Nnj0bgiAgLCwMs2bNgk6nk6ztkpIS6PV6rFq1CocPH0bPnj2RlJSE8PBwyfr4rT179iAgIECWSU4VCgVWrlyJ6dOnw8PDA01NTVi7dq3k/dx8882oqqrC999/j7vuugs7d+4EAFm+n911WwDIuz3gHkk3lpGRAQ8PDyQmJsrWx5tvvol9+/bh9ddfR1ZWluTtHzt2DAUFBUhISJC87d/avHkzduzYgW3btkEQBKSnp0vavslkQklJCe644w589NFHmD17Nl577TU0NjZK2s+vbdu2Tba9EaPRiNzcXKxZswZ79+7FO++8g5kzZ6KpqUnSfry8vPD2229j2bJlGDt2LAwGA3Q6nWVjT7aRc3vAIEH3nBwyMzMT586dw8qVKyUfarAmLi4Ohw8fRk1NjaTtHjlyBGfOnMGoUaMQERGB8vJyPP/88zhw4ICk/QCw/L41Gg0SEhJw9OhRydtXq9WWYZMhQ4bAx8cHxcXFkvZikzkAAAAEsUlEQVRzxYULF3DkyBFER0fL0n5RUREqKioQFhYGAAgLC0OPHj1w5swZyfu6//77sWXLFnz00UdITExEc3Mz+vfvL3k/3XFbAMi/PWCQoPtNDvnWW2+hoKAAq1evhqaDu6GJ1dTUhLKyMsvjPXv2wNvbG3q9XtJ+pk2bhgMHDmDPnj3Ys2cP+vTpg7/+9a948MEHJe3n4sWLlrF+QRDwj3/8AyEhIZL24evri+HDh+PgwYMALp8dZDAYMGDAAEn7ueLjjz/GyJEj4ePjI0v7ffr0QXl5Oc6ePQvg8rx4BoNBlg18ZWUlgMvDM2+99RYmTJgADw8PyfvpbtsCwD7bA8619TN7TA65ZMkS7N69G1VVVfDx8YFer2930y4pnD59GlFRURg4cCDc3d0BAEFBQVi9erWk/VRVVWH69Om4dOkSlEolvL29MW/ePNlvOBYREYF3330Xt912m6TtlpSU4LXXXoPJZILZbEZwcDAWLlyI3r17S95PcnIyamtroVarMXPmTIwcOVLSPq6IjIzEG2+8gREjRsjSPgDs2LEDf/nLXywnD8yYMQOPPvqo5P288cYbOHr0KNra2vDAAw8gOTkZWq1WVJsdfR+l3hZ01I/U2wNr7a1cudIu2wMGCRERicKhLSIiEoVBQkREojBIiIhIFAYJERGJwiAhIiJRGCRERCQKg4ToOkREROCrr75ydBlEToFBQuRijEajo0sgaodBQiSRuro6vPTSS7j33ntx991346WXXkJ5eTkA4NNPP8XYsWPbvX7dunV45ZVXAFye7j0zMxMPP/ww7r//fixevBjNzc0AgMOHD2PEiBFYu3YtHnjgASxYsMC+K0Z0DQwSIomYzWaMHTsWe/fuxd69e6HVai0zCI8aNQrnz59vN6FhXl4e4uLiAADZ2dkoLi7G9u3bsXv3blRUVLSbxqKqqgp1dXXYu3cvMjIy7LtiRNfAICGSiI+PDyIjI9GjRw94enrilVdewZEjRwBcnlH4iSeewI4dOwBcnhPtp59+wiOPPAJBELB161YkJydDr9fD09MTL730Urt5l5RKJWbMmAGNRmOZM4nIWfDGVkQSuXTpEpYtW4b9+/ejrq4OwOVZkk0mE1QqFcaMGYNZs2Zh5syZyMvLwxNPPAGNRgODwYBLly61G/oSBAFms9ny2MfHR/QkhURyYZAQSeT9999HcXExtm7dCn9/fxQVFSEuLg5X5kUNDQ2Fm5sbvvnmG+Tn5yM7OxvA5ZBwd3fHJ598goCAAKtty3UbYyIpcGiL6Dq1tbWhpaXF8lNfXw+tVgudTofa2lqsWrXqqvfExcUhPT0darXacotdpVKJ8ePHY+nSpTAYDAAu35Rq//79dl0fouvFICG6TtOmTcNdd91l+amvr0dLSwvuvfdexMfH46GHHrrqPbGxsTh9+jRiYmLaPT9nzhwMGDAAzzzzDIYNG4bJkyfLdudEIqnxfiREdtTc3Iz77rsPH3/8MQYOHOjocogkwT0SIjvasmULfv/73zNEqFvhwXYiO4mIiIAgCJLf5pTI0Ti0RUREonBoi4iIRGGQEBGRKAwSIiIShUFCRESiMEiIiEgUBgkREYny/wHyyv9M3nJeJwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "ax = sns.barplot(x='Layer', y='Runtime (ms)', hue='Bitwidth', data=joint_df)\n",
    "plt.savefig('/data/jwfromm/layerwise_speed.png', dpi=900)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Polarity Comparison"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 292,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Here we consider 3x3 conv of fire layer 1.\n",
    "# Input shape: [1, 56, 56, 32], Filter shape: [3, 3, 64, 64]\n",
    "# Processor max throughput is 4.8 GHz\n",
    "# Op count (standard): 56*56*3*3*64*64 = 115.6 MFlops -> 24 ms for perfect execution (actual is 26ms)\n",
    "# Op count (bipolar): standard / 32 = 8 MBops -> .8 ms (actual is 3 ms)\n",
    "# Op count (unipolar): standard / 16 = 16 MBops -> 1.6 ms (actual is 4 ms)\n",
    "# For bitserial ops its obvious that memory movement dominates. Unipolar requires no additional memory movement over bipolar\n",
    "# so its only marginally slower. Thus, in most cases unipolar quantization is desirable.\n",
    "standard_ops = 56*56*3*3*64*64\n",
    "bipolar_ops = standard_ops / 32\n",
    "unipolar_ops = standard_ops / 16\n",
    "\n",
    "standard_runtime = 26.2\n",
    "bipolar_runtime = 3.0\n",
    "unipolar_runtime = 4.0\n",
    "\n",
    "standard_hyp = 24\n",
    "bipolar_hyp = .8\n",
    "unipolar_hyp = 1.6\n",
    "\n",
    "standard = pd.DataFrame([{'class': 'baseline', 'ops': standard_ops, 'runtime': standard_runtime}])\n",
    "bipolar = pd.DataFrame([{'class': 'bipolar', 'ops': bipolar_ops, 'runtime': bipolar_runtime}])\n",
    "unipolar = pd.DataFrame([{'class': 'unipolar', 'ops': unipolar_ops, 'runtime': unipolar_runtime}])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 293,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axs = plt.subplots(ncols=3)\n",
    "fig.tight_layout()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 294,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1, 'Baseline')"
      ]
     },
     "execution_count": 294,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "s = sns.barplot(data=standard, y='runtime', ax=axs[0])\n",
    "s.axhline(y=standard_hyp, color='y')\n",
    "s.set_ylabel(\"Runtime (ms)\")\n",
    "s.set_title(\"Baseline\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 295,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(0, 5)]"
      ]
     },
     "execution_count": 295,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "u = sns.barplot(data=unipolar, y='runtime', ax=axs[1])\n",
    "u.axhline(y=unipolar_hyp, color='y')\n",
    "u.set_ylabel(\"\")\n",
    "u.set_title(\"Unipolar\")\n",
    "u.set(ylim=(0, 5))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 296,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(0, 5)]"
      ]
     },
     "execution_count": 296,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "b = sns.barplot(data=bipolar, y='runtime', ax=axs[2])\n",
    "b.axhline(y=bipolar_hyp, color='y')\n",
    "b.set_ylabel(\"\")\n",
    "b.set_title(\"Bipolar\")\n",
    "b.set(ylim=(0, 5))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 297,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 3 Axes>"
      ]
     },
     "execution_count": 297,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "fig"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 298,
   "metadata": {},
   "outputs": [],
   "source": [
    "fig.savefig('/data/jwfromm/polarity.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Unfused Cost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 421,
   "metadata": {},
   "outputs": [],
   "source": [
    "# In this plot we want to show the per layer breakdown of inference time including interlayer ops. Maybe we do bar chart for ms of layer\n",
    "# on one y axis and other axis is % of runtime that increases from left to right.\n",
    "names = ['0', '1', '1g', '2', '2g', '3', '3g', '4', '4g', '5', '5g', '6', '6g', '7', '7g', '8', '8g']\n",
    "runtime = [9.3, 6.7, 6.5, 7, 6.5, 10.7, 11.4, 3.4, 3.0, 4.8, 4.6, 5.7, 4.66, 8.1, 5.9, 1.9, 1.5]\n",
    "glue = ['g' in n for n in names]\n",
    "glue = ['Glue Layer' if g else 'Bitserial Layer' for g in glue]\n",
    "glue_effect = []\n",
    "delta_sum = 0\n",
    "for i in range(len(names)):\n",
    "    name, time = names[i], runtime[i]\n",
    "    if 'g' in name:\n",
    "        delta = time / sum(runtime)\n",
    "        delta_sum += delta\n",
    "        glue_effect.append(delta_sum)\n",
    "            \n",
    "pure_name = []\n",
    "for i in range(len(names)):\n",
    "    if 'g' in names[i]:\n",
    "        pure_name.append(pure_name[i-1])\n",
    "    else:\n",
    "        pure_name.append(names[i])\n",
    "        \n",
    "last_val = glue_effect[-1]\n",
    "for i in range(len(np.unique(pure_name)) - len(glue_effect)):\n",
    "    glue_effect.append(last_val)        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 422,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Make dataframe\n",
    "unfused_df = pd.DataFrame({'name': pure_name, 'runtime': runtime, 'Layer Type': glue})\n",
    "cumul_df = pd.DataFrame({'name': range(len(glue_effect)), 'cumul': glue_effect})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 424,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib.lines import Line2D\n",
    "fig = plt.figure()\n",
    "g = sns.barplot(x='name', y='runtime', hue='Layer Type', data=unfused_df)\n",
    "plt.tight_layout()\n",
    "g.set_xticklabels(g.get_xticklabels(), fontsize=9)\n",
    "g.set_xlabel('Layer')\n",
    "g.set_ylabel('Runtime (ms)')\n",
    "g.legend_.remove()\n",
    "\n",
    "g2 = g.twinx()\n",
    "g2.grid(False)\n",
    "g2 = sns.lineplot(x='name', y='cumul', data=cumul_df, label='Cumulative Glue Fraction')\n",
    "g2.set_ylim(0, 1)\n",
    "g2.lines[0].set_linestyle('--')\n",
    "g2.lines[0].set_c(sns.xkcd_rgb[\"faded orange\"])\n",
    "g2.set_ylabel(\"Glue Compute Fraction\")\n",
    "g2.legend_.remove()\n",
    "\n",
    "fig.legend(bbox_to_anchor=[.975, .960])\n",
    "fig.savefig('/data/jwfromm/unfused_cost.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Glue Breakdown"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 323,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f16a643fd30>"
      ]
     },
     "execution_count": 323,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Here we break down the costs in the first fire layer of squeezenet.\n",
    "squeeze_1x1_conv = 2.36\n",
    "squeeze_1x1_cast_fp = .27\n",
    "squeeze_1x1_weight_scale = .3\n",
    "squeeze_1x1_bn = .35\n",
    "squeeze_1x1_quantize = 1.23\n",
    "squeeze_1x1_cast_int = .19\n",
    "\n",
    "expand_1x1_conv = .92\n",
    "expand_1x1_cast_fp = .83\n",
    "expand_1x1_weight_scale = .92\n",
    "expand_1x1_bn = 1.0\n",
    "expand_1x1_quantize = 2.49\n",
    "expand_1x1_cast_int = .74\n",
    "\n",
    "expand_3x3_conv = 4.23\n",
    "expand_3x3_cast_fp = .79\n",
    "expand_3x3_weight_scale = .81\n",
    "expand_3x3_bn = .97\n",
    "expand_3x3_quantize = 2.47\n",
    "expand_3x3_cast_int = .73\n",
    "\n",
    "concat = 1.43\n",
    "\n",
    "times = [squeeze_1x1_conv, squeeze_1x1_cast_fp, squeeze_1x1_weight_scale, squeeze_1x1_bn, squeeze_1x1_quantize, squeeze_1x1_cast_int,\n",
    "         expand_1x1_conv, expand_1x1_cast_fp, expand_1x1_weight_scale, expand_1x1_bn, expand_1x1_quantize, expand_1x1_cast_int,\n",
    "         expand_3x3_conv, expand_3x3_cast_fp, expand_3x3_weight_scale, expand_3x3_bn, expand_3x3_quantize, expand_3x3_cast_int]\n",
    "layer = ['Squeeze (1x1)'] * 6 + ['Expand (1x1)'] * 6 + ['Expand (3x3)'] * 6\n",
    "op_type = ['Convolution', 'Cast FP', 'WeightScale', 'BatchNorm', 'Quantize', 'Cast Int'] * 3\n",
    "\n",
    "glue_df = pd.DataFrame({'Runtime (ms)': times, 'Layer': layer, 'Type': op_type})\n",
    "\n",
    "g = sns.barplot(x='Layer', y=\"Runtime (ms)\", data=glue_df, hue='Type')\n",
    "g.legend(title=\"\", ncol=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 324,
   "metadata": {},
   "outputs": [],
   "source": [
    "g.get_figure().savefig('/data/jwfromm/glue_breakdown.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Scheduling Effect"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 406,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Here we examine the speedups of a fused bitserial squeezenet with various optimizations applied.\n",
    "baseline = 719.9\n",
    "no_schedule = 4604.4\n",
    "fully_optimized = 67.6\n",
    "# Speedups. Note that popcount here means using an instruction that utilizes the full SIMD register size on rpi.\n",
    "parallel = 3.3\n",
    "vectorize = 1.6\n",
    "popcount = 3.3\n",
    "tile = 1.5\n",
    "bitpack_fusion = 1.3\n",
    "autotune = 2.0\n",
    "\n",
    "# Relative runtime\n",
    "unopt_rel = baseline / no_schedule\n",
    "tile_rel = (unopt_rel * tile * autotune)\n",
    "vectorize_rel = tile_rel * vectorize\n",
    "parallel_rel = vectorize_rel * parallel\n",
    "popcount_rel = parallel_rel * popcount\n",
    "fusion_rel = popcount_rel * bitpack_fusion\n",
    "\n",
    "names = ['Unoptimized', 'Tiling', 'Vectorize', 'Parallelize', 'Fast Popcount', 'Bitpack Fusion']\n",
    "rel_speedup = [unopt_rel, tile_rel, vectorize_rel, parallel_rel, popcount_rel, fusion_rel]\n",
    "fast = [r > 1.0 for r in rel_speedup]\n",
    "fast = [sns.xkcd_rgb[\"faded green\"] if r else sns.xkcd_rgb[\"faded red\"] for r in fast]\n",
    "\n",
    "sched_df = pd.DataFrame({'Optimization': names, 'Relative Speedup': rel_speedup, 'Fast': fast})\n",
    "\n",
    "g = sns.barplot(x='Optimization', y='Relative Speedup', data=sched_df, palette=fast)\n",
    "\n",
    "g.axhline(y=1.0, color=sns.xkcd_rgb[\"faded red\"], linestyle='--')\n",
    "for item in g.get_xticklabels():\n",
    "    item.set_rotation(25)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 408,
   "metadata": {},
   "outputs": [],
   "source": [
    "g.get_figure().savefig('/data/jwfromm/sched_speed.png', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Bitwidth runtime"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 413,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# This figure is just a per layer runtime with 1, 2, and 3 bit activations. Nothing fancy.(all unipolar)\n",
    "fig = plt.figure()\n",
    "onebit = [9.9, \n",
    "          sum([3.28, 1.88, 4.04]),\n",
    "          sum([2.89, 1.03, 4.02]), \n",
    "          sum([3.95, 1.37, 8.8]), \n",
    "          sum([1.11, .38, 2.0]), \n",
    "          sum([1.34, .65, 3.24]), \n",
    "          sum([2.18, .62, 3.14]),\n",
    "          sum([4.2, 1.49, 4.04]),\n",
    "          sum([.72, .20, 1.07])]\n",
    "\n",
    "onebit_cumul = []\n",
    "for i, val in enumerate(onebit):\n",
    "    if len(onebit_cumul) == 0:\n",
    "        onebit_cumul.append(val)\n",
    "    else:\n",
    "        onebit_cumul.append(val + onebit_cumul[i-1])\n",
    "\n",
    "twobit = [10.5, \n",
    "          sum([4.51, 1.84, 8.12]),\n",
    "          sum([3.54, 1.76, 8.18]),\n",
    "          sum([6.59, 3.58, 19.4]),\n",
    "          sum([1.87, .68, 3.87]),\n",
    "          sum([2.02, 1.08, 5.9]),\n",
    "          sum([2.08, 1.09, 6.0]),\n",
    "          sum([2.07, 1.25, 7.82]),\n",
    "          sum([1.0, .33, 2.05])]\n",
    "\n",
    "twobit_cumul = []\n",
    "for i, val in enumerate(twobit):\n",
    "    if len(twobit_cumul) == 0:\n",
    "        twobit_cumul.append(val)\n",
    "    else:\n",
    "        twobit_cumul.append(val + twobit_cumul[i-1])\n",
    "\n",
    "threebit = [10.74, \n",
    "            sum([5.97, 2.22, 11.69]),\n",
    "            sum([4.74, 2.23, 11.7]),\n",
    "            sum([4.74, 3.49, 22.44]),\n",
    "            sum([2.37, .91, 5.68]),\n",
    "            sum([2.97, 1.41, 8.89]),\n",
    "            sum([4.0, 1.37, 8.55]),\n",
    "            sum([3.99, 1.7, 11.14]),\n",
    "            sum([1.29, .46, 2.85])]\n",
    "\n",
    "threebit_cumul = []\n",
    "for i, val in enumerate(threebit):\n",
    "    if len(threebit_cumul) == 0:\n",
    "        threebit_cumul.append(val)\n",
    "    else:\n",
    "        threebit_cumul.append(val + threebit_cumul[i-1])\n",
    "\n",
    "layer_index = range(len(onebit))\n",
    "\n",
    "one_df = pd.DataFrame({'Runtime (ms)': onebit, 'Layer': layer_index, 'Bitwidth': '1 Bit'})\n",
    "two_df = pd.DataFrame({'Runtime (ms)': twobit, 'Layer': layer_index, 'Bitwidth': '2 Bit'})\n",
    "three_df = pd.DataFrame({'Runtime (ms)': threebit, 'Layer': layer_index, 'Bitwidth': '3 Bit'})\n",
    "\n",
    "merge_df = pd.concat([one_df, two_df, three_df])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 414,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "g = sns.barplot(x='Layer', y='Runtime (ms)', hue='Bitwidth', data=merge_df)\n",
    "g.set_ylabel(\"Layer Runtime (ms)\")\n",
    "g.legend(loc='upper left', title='Bitwidth')\n",
    "\n",
    "one_cumul_df = pd.DataFrame({\"Runtime (ms)\": onebit_cumul, 'Layer': layer_index, 'Bitwidth': 'a'})\n",
    "two_cumul_df = pd.DataFrame({\"Runtime (ms)\": twobit_cumul, 'Layer': layer_index, 'Bitwidth': 'b'})\n",
    "three_cumul_df = pd.DataFrame({\"Runtime (ms)\": threebit_cumul, 'Layer': layer_index, 'Bitwidth': 'c'})\n",
    "\n",
    "cumul_df = pd.concat([one_cumul_df, two_cumul_df, three_cumul_df])\n",
    "\n",
    "g2 = g.twinx()\n",
    "g2.grid(False)\n",
    "g2 = sns.lineplot(x='Layer', y='Runtime (ms)', data=cumul_df, hue='Bitwidth')\n",
    "#g2.set_ylim(0, 1)\n",
    "g2.lines[0].set_linestyle('--')\n",
    "g2.lines[1].set_linestyle('--')\n",
    "g2.lines[2].set_linestyle('--')\n",
    "#g2.lines[0].set_c('orange')\n",
    "g2.set_ylabel(\"Cumulative Runtime (ms)\")\n",
    "g2.legend_.remove()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 415,
   "metadata": {},
   "outputs": [],
   "source": [
    "g.get_figure().savefig('/data/jwfromm/quant_layer_speed.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 416,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Add dual y axis of previous graph to this one then only keep it.\n",
    "baseline = [10.5,\n",
    "            sum([8.08, 5.13, 21.79]),\n",
    "            sum([10.9, 5.11, 21.79]),\n",
    "            sum([11.73, 17.35, 104.53]),\n",
    "            sum([5.21, 30.7, 28.4]),\n",
    "            sum([9.03, 8.5, 87.67]),\n",
    "            sum([13.84, 8.77, 88.12]),\n",
    "            sum([13.8, 11.22, 115.25]),\n",
    "            sum([4.93, 3.26, 23.2])]\n",
    "\n",
    "one_df = pd.DataFrame({'Relative Speedup': np.asarray(baseline) / np.asarray(onebit), 'Layer': layer_index, 'Bitwidth': '1 Bit'})\n",
    "two_df = pd.DataFrame({'Relative Speedup': np.asarray(baseline) / np.asarray(twobit), 'Layer': layer_index, 'Bitwidth': '2 Bit'})\n",
    "three_df = pd.DataFrame({'Relative Speedup': np.asarray(baseline) / np.asarray(threebit), 'Layer': layer_index, 'Bitwidth': '3 Bit'})\n",
    "baseline_df = pd.DataFrame({'Relative Speedup': np.asarray(baseline) / np.asarray(baseline), 'Layer': layer_index, 'Bitwidth': 'FP32'})\n",
    "\n",
    "merge_df = pd.concat([one_df, two_df, three_df, baseline_df])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 418,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAb8AAAEMCAYAAABUTbipAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjAsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+17YcXAAAgAElEQVR4nOydd3hUVdrAf3daJj2BBAKEXg4dKbKoqJFdC7jYEAuCYvlE1l2X/RRdXdeuq6ALttXsuordT1SwYq+ooICKUg4dAiEQIL1Nuff7407ikHonbSbJ+T1Pnrlz6nsnyX3nnPMWzTAMFAqFQqHoSNjCLYBCoVAoFK2NUn4KhUKh6HAo5adQKBSKDodSfgqFQqHocCjlp1AoFIoOhyPcArQS8cBMYDPgDbMsCoVC0VZwAoOBF4CiMMvSrHQU5TcT+Fe4hVAoFIo2zBPhFqA56SjKbzNAYWEZfr8eblkUCoWiTWC320hIiIbAM7Q90VGUnxfA79fx+ZTyUygUihCxdFwkhHgQmAb0AUZIKX+pVn87cEdwnRBiApAJRAO7gJlSyoPNJXhdKIMXhUKhUDQXy4GTgN3VK4QQY4AJwXVCCBvmeeK1UspBwJfA/a0hqFJ+CoVCoWgWpJQrpZRZ1cuFEFHA48DcalVjgXIp5crA+yeBC1pWSpOOsu1ZJ4ZhkJeXi8dTDnTEOKcaLpeb5ORUNE0LtzAKhSICWbFiRdq8efP6VCvOl1LmWxziLuAFKeUuIURweS+CVoJSykNCCJsQopOU8kiThG6ADq/8iosL0DSNrl3T0bSOtxA2DJ38/EMUFxcQH58UbnEUCkUEsnDhwqW1FN+JeX5XL0KI44BxwF+bWawm0fGe9tUoKysmPj6pQyo+AE2zER+fTFlZcbhFUSgUEcr8+fOnA32r/Sy22P1kYAiwUwixC0gHPhBCnAbsAXpXNhRCpAB6S6/6oJVWfkKIzsDzQH/AA2wF5kgpc61a+gghYoBnMPeIfcANUsp3miqbrvux2zv2Athud6Dr/nCLoVAoIpTJkyfnTJ48eVdj+kop7yfIiCWgAH8vpfwlYPASLYSYGDj3uwaobZXZ7LTWcscAFkgphZRyBLAduD9ES58bgEIp5QBgKvCUECKuOYTr6GddHf3+FYpIwzAMcksPs/bAj7yx9R0WrXuCx358KtxiNYgQ4hEhxF7M1d3HQogN9bWXUurALOAJIcRWzFViq2yPtsqSJ7CE/TyoaBWm1U9tlj67gCtqGeZC4LLAeFuFEGuAyVT7liCESAKOOrzKzMxMy8jIaOptVHHiicfSv/9ADMPAMHSuumouJ51kjn/DDdfx97/fRWJiEkuWPMXs2VdZHnf58tdwOl2ceeZZNeomTTqeTz/9hqKiIj788D2mTbsQgP37s7n++j/x0kuvN8u9KdoH8Qlu3FHOWuvKK7wUFZa3skSK+sivKGB3YRbdYtPoEpPChsObeWL9MwA4bA7S47rTJ6FXmKVsGCnldcB1DbTpU+39N8CIFhSrVlp9vy+w2psLvEVolj5HtcXcK+5ZyxTzgNuDCzIzM8nIyCA5ObZG44MHbTgcoS2AnU4nL7zwCgC7du3kT3+6hkmTJgGwePFjVe2ee+5prrrqasvjnn9+/Ra+DoeNsrIS3nhjKRdeeDFgRmDQNC3ke6iOzWYjNTW+SWMoIosZN75Ya/lLCy7BnVq7YlS0DqXeMt7b8hk7juxm+5Hd5JUXADBz1LkM692XcQlDudo1g/6d+tAzsTsOmz3MErc/wnHY9ShQDDwGnNsC4y8GlgQXzJkzZxywNC+vpEaEF11vXNSXyj4FBUXExcVXvT///KlkZj7DK6+8iNfrZebMi0hJSWHq1HNZteprbrrpVj7++APuv/9u3n//cwCmTz+L1157myVLnsLlimLWrNlkZ+/jrrv+TklJMRMmnFA15yOPLGL//mxmzryIwYOHcNllV+L369x//32sX/8DcXHx3H//QyQkJIZ0P7quk5vbruLWdmga+iKjftetQ5mvnKyifewuzGJ30V56x6dzau8MvLqPNzauoLO7EwMS+9O7Zzq9E3qSHte96nczKuEY8EHe4dKwye9w2GpdNLQHWlX5BULfDASmSil1IUQolj6VbXMD73sBn1VvFPA7qe57kt4M4lfh9XqZPXsGPp+P/fv3cdtt99Roc+21f+b11/+PJUteAiA/P59///txAH78cR19+/Zj06YN2O12BgwYiN1+9De7hx9+kN///mx+//uzeeed5Xg8nqpxd+zYVjXu/v3ZZGfvZcqUu7jhhr+ycOF9vPvu21x88czmvGWFQtEAHr+XgopCUmM6A7Bo3RNsz9+FEfAf7uxOpltsVwCcNgcLTryDKLsrbPJ2dFpN+Qkh7sM84ztTSlkRKF6LdUufpcAcYI0QYiBwLHBxC4tdK06ns0r57Nmzm2uv/R+GDh1GamqXOvskJSVhs9k4cuQwUm5m+vQZ/PjjOux2B6NGja7Rfv36n7j77gcAOO20KTzwwL11jt21axpDhw4HYMiQoWzZIptyewqFwgI5JQfYlr+T3YV72VO0l+ySHFKjO3PbhPkAiOQBDE4eSK+EnvSK70G862j7PKX4wktruToMA24GtgDfBDz8d0opzxVCzAIyhRBuAq4OQf1+BKZIKbOBhcASIcQ2wA9cLaUM+95Nr169SUvrxs8/r2fSpN/V23bUqNF89tknJCQkMm7csdx77504HHZmzarNvsc6Tuev5zc2mx2/X7ktKBTNhW7oHCzNZXfhXvYV7+fcAWeiaRof7fmCVfvXEOOIpndCT07tnEHvhF/NEKb0PTWMUisaorWsPTcAtdrT12fpI6U8Jui6BJjeIgI2gSNHDpOVtYf09Jo7q06nk4qKCqKiogBT+T3xxKOce+75dOrUmYKCfI4cOczgwUNq9B05chQff/wBU6ZM5aOP3scwzK2TmJgYSkpKWvamFAoFft3Pe7s+5vOslZT7zc0ql93FpF4nkhSVyOm9J3FG79+SEt1JuQu1QTq2d3cjqTzzMwwDr9fDZZddyaBBg2u0O++8C7j88hl0796DBx98hGOOGcPBgweqtjmFGMy+fXtxOGr+Gv785xu4885beeWVF5gw4XhcLnOLJDExiXHjjmXWrAsYNmwEl112ZcverELRQdmct5X3d33CmC4jGdZ5ML0TetI1JhVbIBpUl5iUMEuoaApa5YqinTMR+Ko2a8+cnN2kpfWuvVcHQn0O7YvU1Ph6XR2UtWfdFHqKSHCZ1rI7CnbTL7Hj/l8EWXueCKxsoHmbomMGtFQoFIpq6IbOOzs+4PZv7ie7OAegQyu+9o7a9lQoFB2egopCntnwElvzdzAhbRydozuFWyRFC6OUn0Kh6NBsPrKVJRtfptxXwcwhF3Bct3HhFknRCijlp1AoOjQbj0hiHDFcd8zVdI9LC7c4ilZCKT+FQtHhKPIUU1BRSHp8d87uN5kpfU7F7YgKt1iKVkQZvCgUig7F1rzt/OO7Rfx3wwvoho7dZleKrwOiVn4KhaJFiZT0Srqh8+Huz3hnx4ekxnTmquGzqnz2FB0PpfyqUd8/alOw+k/+2GOL+eKLT9m/P5vnnnuFfv0G1Nruvffe5pFHHiItrTsANpvGtdfOY+zYYwGYPXsGmZlPExXl5tVXX+LUU88gOVlZsClaH3eUs16fwyJaXvmV+8p56pcX2HRkC2O7jGLG4Gm4He4Wn1cRuSjlV436/lGbgtV/8hNPzGD69Iu49tr/abDtuHHjueeeBQB8++1K/vnPB3jxxdcAqgJvA7z66suMGzdeKT9Fh8Vld+Gw2blInMfE7r9R4cgUSvlFGqNGHdNwo1ooLi4mPj6h6v3EieP48MMvWbr0ZQ4dyuXWW2/C5Yri9tvvoW/ffs0lrqIakbLFpzC3OT/PWsmYrqNIikpkzojZSukpqlDKrw2zZs13zJ49g7KyUvLz81iw4OEabS677Erefns599zzQJ1bqIrmIxK2+BRQ7C3huY3/x4bDm6nwe5nc97dK8SmOQim/Nkzwtue6dWu4445bePnlN3C71VmGouOyo2A3T//yIkWeIi4cdA4n9jgu3CIpIhBl6tROGDNmHD6fj507t4dbFIUibPx8aCOL1j2BTbNx/dhrOSn9eLXiU9SKWvm1E7Zv30ZpaUmV9WcwsbGxFBcXh0EqhaJ1GZDUlxN7TOD3fU8nxhkdbnEUEUyrKT8hxIPANKAPMEJK+YsQog+wPKhZEpAgpaxhliiEuAP4A5AdKPpaSnltc8tZXuHlpQWXNPewlFd4LbVbvHghX3zxGUeOHGbevGtJSEjkhRderbVt5ZmfmZbK4JZb7iA5OblGu/PPv4j77rsLt9utDF4U7Y5dhXv4cPfnXD70YqId0Vww6Jxwi9RhqeM53xl4HugPeICtwBwpZW6gzwQgE4gGdgEzpZQH6xh/DHAmMApTX+QDPwErpJRrQpG1NVd+y4GHga8qC6SUu4Aq80YhxOIGZHpOSnlDSwkIUFRYHlajhHnz5jNv3vwG202ZMpUpU6bWWb9y5a9/B1OnnsPUqeqBoGhfGIbBZ3tXsnzbeyRGJZBXUaASzIafGs95wAAWSCk/BxBCLATuB64UQtiAF4DZUsqVQohbA3VXBA8qhDgduBeIB74AvgaKAu+HAC8KIYqAW6WU71sRtNWUn5RyJYAQotZ6IYQLuAQ4vbVkUigUbZNSbxkvbF7KT7m/MDJlGLOGTCfGGRNusTo8tT3npZRHgM+Dmq0C5gauxwLllf2AJzFXf0cpP+B/gLlSyu/rmlsIcSxwExBZys8CZwH7pJTr6mlzkRDiNCAHuF1K+W31BkKIJMzlcBWZmZlpGRkZzSmrQqEII89tMt0Yzhvweyb1PFEZtbQwK1asSJs3b16fasX5Usr8UMYJrPTmAm8FinoBuyvrpZSHhBA2IUSngNKsLD+/obEDirHBdpVEkvK7Ani6nvongXullF4hxKnAm0KIIVLKw9XazQNuDy7IzMwkIyOD5OTYGoMePGjD4VBGrzabjdTU+HCL0e6JlM84UuQA67IYhoFf9+OwO7jy2OkUe0oZlKLOr1uDhQsXLq2l+E7gjhCHehQoBh5rrCxCiFSgTEpZLISwA5cCOvC8lFK3Ok5EKD8hRA/gZGBWXW2klDlB1x8JIbKA4Zj7v8EsBpYEF8yZM2ccsDQvrwSf7+jPRtf1GmUdEV3Xyc0tCrcYbZ6GHuSt9RlHihzQPLKU+cp4cfPr2DUbs4dejFOLJZlY9TfbwjgcNpKTY5k/f/70efPmVTcoCXXV9yAwEJgapKT2AL2D2qQAevCqrxbeAa4BfsA8B5wKeDHtR/5iVZ6IUH7AZcC7taziqhBC9JBS7gtcH4NpTSSrtwssw6v/UtKbT1SFQtGaZBXt46lfXuBIeR5n9Tsj3OJ0SCZPnpwzefLkXY3tL4S4D/N870wpZUVQ1VogWggxMXDudw1Q2yozmEHAj4HrmcDxmKvJDUSi8hNCPAKcB6QBHwshDksphwWqZwPX1dLnPeC2gAnrfUKIsYAf01x2VvBqUKFQtC8Mw2Bl9ipe2/o2cc5Y5o2+hv5JfcItlqIeanvOAxcANwNbgG8CxjA7pZTnSil1IcQsIFMI4Sbg6tDANH7AJYQYBBRIKfcEzhLjQpG1Na09r6MWBReoG1RH+ZSg68taSLSjSE504XA1f2JLn6eCvAJPvW0KCvK5++7b2LdvL06nk/T0Xsyff0utvnv//W8my5a9RkpKKoZh4Ha7ueGGmxk4cBCHDuVy55238uijmVVtL730CpzO5k/VpFC0FMXeEt7a/j6Dkvpz2dCLiHPVPLNXRBb1POfrtEiSUn4DjAhhmhXAq0Bn4JVA2VBgXwhjRMy2Z8TgcEWxdsFVzT7u2Bufwlyw1o2macyYcSljxowD4PHHH+bJJx/l5ptvq7X9GWecyR//OA+AN95YypNPPsZDDz1CSkpqleIDeOaZ/3DxxbOU8lO0CXYV7qFXfDrxrjhuGPdHUqM7q6SzimCuwjwq82I6zwOkEKLxjVJ+EURCQmKV4gMYNmw4y5a9bqlvSUkx8fGmYcH+/dlcddUs3n33Ex566AEA5s69Ak2z8eijmVXtFIpIothbwrJt77Jq/xouGXw+x3cfT9eY1HCLpYgwAmeG/65W9nmo4yjlF6Hous6yZa8zceJJdbZ5//13WbPmOwoLC/D7fTz66L9rtLn++ptYtmwpTzzxNDExyglYEXkYhsHqnLUs2/Yupb4yTut9CuO6Ni6vpaL9I4RIxNxaHU21cz4p5WlWx1HKL0JZtGghMTHRTJt2QZ1tgrc9V6x4h9tvv5lnnnmpzvYKRSTy0ubX+Wb/d/RL7M3FYhrd49LCLZIislkK2IFlQFljB1HKLwJ57LHF7N27hwceWITNZu2sY9Kk33HvvXeQl5fXwtIpFE3H0HTMkI9wbNpoeiekc3z38epsT2GFCUCKlLJ+I4oGUMovwsjMfBwpN7Fw4cO4XC7L/datW0NiYiKJiYmUlx/9ZSgmJpaSkmK17amICPS4fPTe29HyzeQtg5L7Myi5f5ilUrQhVgKDgfVNGUQpvwhix47tPP/8M/Ts2YtrrjHjunbr1p1//OPBWttXnvkZhoHD4eDuux+odaV40UWXcN111xAV5VYGL4qwYTi86Ok7MTofhHI3WmFSw50UiprMBt4TQqwGDgRXSCnvsjqIUn7V8HkqAm4JzT9uQ/Tr1/+oVET1ceWVc7jyyjm11nXr1p133/2k6v0VV1zNFVdcbU1QhaIF0BOOoPfdAjY/2v6e2Panoxn2cIulaJvcC/TEdIhPCCo3QhlEKb9qmI7oTdpKVigUAcxEy6BVuNFK47Bl9UMrV9vviiZxETBISrm/KYOo02WFQtHsePxe3t7xAUs2vgyAVhGDfetwpfgUzcEOTAf3JqFWfgqFolnZdGQLr8hlHCo7zPi0Mfh0f7hFUrQvngfeEkI8Ss0zv0+tDqKUn0KhaBaKPSUs3fomaw78SJfoFK475mpEpwE4bOpsT9GsXBt4va9auQFYTvColJ9CoWg2tuZtZ0qf33Fa71Nw2lUsWUXzI6Xs2xzjKOWnUCgaTXZxDl/s+4YLB51DnCuWO477Ky6l9BokPsGNO6r2z6m8wktRYXkrS9TxUMpPoVCEjMfvYcWuT/h4zxdEO9yckn4CabFdleKziDvKyYwbX6y17qUFl1CEUn7BCCG+BxYAb9YW2UUI4QLOAa6XUv7GypghKT8hhIaZOuKQlDIkn4q2QkJiFFEhRFaxSoXHQ2FBw75+N998PdnZ2dhsGtHRMfzlL/MZOFDUaPfee2/zyCMPkZbWHQCbTePaa+cxduyxAMyePYPMzKeJinLz6qsvceqpZ5Cc3Kl5b0rRIdlwWPJ/chmHy48wIW0c5w44U+XaU7Q0lwF3AU8IIdYBEigC4jEzu48BPsV0gLeEJeUnhEgCHgWmAy7AI4RYCvxZSnkkhBuIeKJcLmY/8+dmH3fJ5Q8DDSu/v/3tTuLizEDlX331Of/4x108/XTt3xDHjRvPPfcsAODbb1fyz38+wIsvvmbOt+TXANevvvoy48aNV8pP0WT8up/Xtr6Jw2bnz6PnqLBkilZBSrkROF8IkQacipn8NgXIA54DZkkpD4YyptWV3zOYqeNHA7uB3sCdwNOYS80GEUI8CEwD+gAjpJS/BMp3AeWBH4CbpJQf1NI/JiDHWMAH3CClfMei/G2GSsUHUFxcjGYx0G9xcTHx8b8GO5g4cRwffvglS5e+zKFDudx66024XFHcfvs99O1r2SBKoUA3dFbtX8vYrqOIsruYO/IKkt1JOG3q1KSlMAwDTdMwfB48P3+IXrAfzRGFe+Kl4RYtrEgpc/g1gW2TsPrXOwlIk1JWRkzeJISYDWSHMNdy4GHgq1rqzq9UhvVwA1AopRwghBgIfCWEGCClLA5BhjbB/fffzXffrQLgwQcfqbPdmjXfMXv2DMrKSsnPz2PBgodrtLnssit5++3l3HPPA/TrN6DFZFa0T/YWZfOyfINdhXvw6T5OSj+OLjEp4RarXdHXnkuqvZAUWzEptiL2PLYSrccI3BNngc2BZ+1yNHcc9rSB4Ra1XWFV+W3GXLFtCirrhbnvagkp5UoAIWqeX1nkQsx9X6SUW4UQa4DJmLmd2hV//evfATNw9b/+9XCdCjB423PdujXcccctvPzyG7jd7laTVdE+qfB7eG/nR3ya9RUxjmguG3oRx3YdHW6x2iSGpwy9IAc9f3/gNYdDnVOBKADOi1lLsq2UCsPOIT2eqB7D8Hcxd2c0m4242Y+jOaLCeAftE6vK7xPgQyHE80AWZlDRmcDzQogrKhtJKZ9upBwvBoxpVgK3SCnza2nTC3PLtZI9ATmOInA+eVS4+MzMzLSMjIxGihY+zjjjTBYsuI+CgnwSE+uPgD9mzDh8Ph87d25nyJBhrSShor3y8ubX+f7ADxzfbTznDJhCrFOFJasPQ/djFOWi5+egF+zH8HqIGns2AGXvL8Kfs8VsqGlo8akYCQlUKr8XSyZQYkRRZLgBjZfOvYTc3KKqsZXiaxmsKr/jgG2B1+MCZduB4wM/YHrXN0b5nSilzBJCRAGLgccwFWtjmQfcHlyQmZlJRkYGyck1LdIOHrThcLROiNOG5iktLaWoqJCuXc1M1l999QWJiQl06pSMpmlHtbXZNDRNqxpz27atlJaWkp6eXlXmcJj3FhsbS1lZab3z22w2UlNVqqOWJlI+49rkOFKaj03TSIpO5JKxZzO1fBKDU1t+qzxSPpOGMAwDvbQQ75FsvHk5xI88BYDDHz1DwZoVEBTGzZGQQsrpl6BpGqUnT8fweXF27o4zOQ3NEXAH+cI0ZMvRa36xbSufSVvGkvKTUp7SUgJIKbMCrxVCiH8Bb9XRdA+moU1u4H0v4LNa2i0GlgQXzJkzZxywNC+vBJ9PP6qxrutHlVV4PAHLzOalwuOpMXd1iotLuPnmGykvL8Nms5OQkMD99y/C7zeonq1D1w2+/341M2deFIicb3DLLbcTH59YNY/PZ97b+edfxN1334Hb7a7T4EXX9aO+bSoaR0MPrdb6jEORw6f7+CxrJe/t+piRKUO5fNgMnMTQmZhmkTdSPpPG4t22KmB0kgOe0qrysuTBaO44vPE9cY08A1timvmT1A3NHcehQwFzhMSgs7o807avrXwmDoet1kVDXdRj2DgIeBboDBwGLpVSbm2oro45NOAq4GLMjO4jhRAnYdqlvGr53izeUJ1LBill/U/0+seNBRxSyoLADV0E/FhH86XAHGBNwODlWMybry5PPlB92zTdqkymL17DLgktQadOnfn3v5dYajtlylSmTJlaZ31wXsCpU89h6lRLRrmKDsamI1tYuuVNDpTmMiJlKFP7nR5ukcKOYRj4d/+ILaU3trhOoNnQnFE4B0wIKLc0bIndIMrcCnb2/w30t+RX3RGoy7DxSeBxKeULQoiZQCamIWVDdbVxF6a7w+JAX4C9wCKgeZUfpmtBXU7tlqLWCiEeAc4D0oCPhRCHganA60IIe2CcjcAfgvr8CEyRUmYDC4ElQohtmG4XV0spI+PrkULRBvly77f835ZlpEZ3Zu7IyxmeMiTcIoUVwzDw7f4Bz9o30Q/vxjV6KlHHTsPZfzzO/uPDLV6boDbDRiFEF0wn9FMDRS8DjwkhUgGtrjopZS61MxsYLaU8JIR4IlC2kxCCWoN15Vc9kGg34K/A21YnklJeB1xXS1WdJmRSymOCrkswnewVCkUjMTSdQyVHACeju4ygwl9BRs+JHd5nz7f7ByrWLEc/vBstoQvujKtwDDiu4Y4dhBUrVqTNmzevT7Xi/DqME6vTE9gnpfQDSCn9QojsQLlWT11dys8OVLq4VS7K4oLKLGH1zG93taLdQojLgO+B/4YyoUKhCA964mH0njv55zfZzBs1l3hXHKf2zgi3WGGj0pEcwLtjDYa3vErpaSoN01EsXLiwNpeyO4E7WlkUgPeAfwoh/gJVZ4B3E8JiDJqWyT0BSG1Cf4VC0QoYUWX4+29AH7AJdI2LR55dw3q4IxCf4CY1NZ6UlFhiDm/E89ZdJPhyzbLTryD2gvtwDpqoFF8tzJ8/fzrmDmDwz2KL3bOAHoHjLQKv3QPl9dXVxf9i7j4WAImYK77ewE2h3JNVg5fnOfrMLwY4CXghlMkUCkXrYsQV4B/4Cxg2bFl90Q52Y8TswRFjTdiaRLns3H3LAia5N9PNXsAhfxxPPPIWO/2pvLTgEko8He8zscrkyZNzJk+evKsxfaWUBwP2Gxdj6oyLgR8qz/Tqq6tjvELgXCFEV0yr/6xA2LOQsLrRv63a+xLgSSnlx6FOqFAoWhbDMDhSFjiKKYlHy+2GLScdzdf82UraCoZhsP/527gkdhO5/jiWlo7jZ286epM2vxTVqc2wUUo5DLgGeFYIcRtmMOrgIKX11dVHGbAPsAkhugMEjCMtYfXM706rA7Z1kuJdON3NH1HBW15BflGNNFQ1OP/8qbhcLlwuU4YxY8YSGxvHsmWvkZKSisdTwYgRo7jhhptxOp11pkAqKMjn7rtvY9++vTidTtLTezF//i0kJyc3+70pIoeckgO8uuVNDlUcxrANRtPt2Pd2zEDmhqHj37sBe/pwNE0jdugJPC9jldJrQeoybJRSbgZq9Qepr642hBC/A/6NudUZvH9vYNH7AOpRfsFhy+qjCSHNIhKnO4r3Lr282ced8twzYEH5ATWCUP/3v5mcccaZ/PGP8/B4PPzpT3NYvvx1pk+/qM4USJqmMWPGpYwZMw6Axx9/mCeffJSbb76t2e9NEX7KfOWs2Pkxn+1dSZQ9iotHnsWzX+0Nt1hhwTB0fDvX4ln3JvqRvURP/l8cPUeSOG4yP73arjKwdVT+i2ng8grm6q9R1LfymxV0rQEnADn8GtszDTMWZ7tSfpGOy+Vi1Khj2LPHNMCtKwVSQkJileIDGDZsOMuWvd66wipahcNleTy49jEKPUUc3+1Yzuo/mX49uvEcteeBbK9UKb21b6Ln7UVLTMN9ytXYewwPt2iK5sUNPFPpHtFY6lR+wSHNhBCPAsullAabj2AAACAASURBVIuDyv4MqEyWLUBl7j2AuXP/dFRdcXEx3323mmnTLqgqaygFkq7rLFv2OhMnntSCUitam1JvGTHOaDq5kxjdZQTj08bQJ6FXuMUKH7pOxbcvozlcuCfNwdHvN2g2tb3ZDlkE3CiEuF9KWVfwlQaxavAyEzNrbjCPAYeo3XFd0QSqb3v+8st63n//Xdas+Q6bTeP44yceFdqsoRRIixYtJCYm+iiFqWi7lHpLeWfnh3yXs45bf3M9SVGJXDCo44WvM1d6a/Bu/pLo0/+MZncSM/WvaHEpSum1b14HPgBuFkIcCq6QUlo+4Laq/HKAs4BlQWVTgZDSxisaT+WZX0NtqqdAeuyxxezdu4cHHliETT0Q2jRmRvU1vLl9BSXeUk5KPw6XreNZcFYqPXN7cx+2pG4YRYfRktKwJXQJt3iKluc1zNihS2mhM79grsOMwTkf88yvFzAUFW4srFRPgbRy5ZckJCSQkJAIQGbm40i5iYULH8bl6ngPyfaE1+9l8Q+Z7CrcQ//EPlww6BzS47uHW6xWRy8toOzdBQGl1x33pGtw9BuvVnodi76YsT0bnVQBrLs6fCSE6IeZOb078C7wrpTycFMmVzSN8vIy/v73vx6VAumBBxahaRo7dmzn+eefoWfPXlxzjWm4261bd/7xjwfDLLUiFDx+Ly67E6fdyYCkvpycfjzHdh3doSK0GLqOnr8Pe6eeaNEJ2Dr3wjV6qlJ6HZc3MbM+NMnP3HI020AE7c+BHlLKVU2ZNJLxlleYbgktMK4VXnutZni6K6+cU2vb+lIg9evX/6i0Roq2hW7orNy3ind2fsgfj7mKXvHpnDvgzHCL1aoYuo5v5/emy0LxEWIvXojNHU/0pNr/HxQdhijgLSHEV8CB4AoppVUHecvhzXphppo4BtORME4IcT5whpTyKssitwHyizyW/fEUipZge/4uXt2ynL3F2QxKHkBUBzvXM3Qfvh3f41n3Nnp+Nrbk7rhPugItynpSVUW7ZkPgp0lYXfllYm51noiZaRfgI+ChpgqgUCh+5eXNr7MyezXJUUlcOXwmo1NHdJgtTsPQ0TQbesFByj/NxJbcA/fv/oCj77gq/1WForkijllVfuOBM6WUuhDCCAhQIIRIbA4hFIqOjF/3Y9NsaJpGJ3cyp/eexOl9JhFlb/8rPsPQ8WWtx7vxM3C6iZ40B3tyd6LP+hv2rv2V0lMAIIQ4SUr5ZeC6zizvUspPrY5pVfkdAAYAW4KEGQrssTpRJBOc16sjYhiN9hNVNAGP38PaAz/xcdaXnNXvDEalDuP0PnX+X7crYrQKxjh3k/XEn/Dl5aC543EO/fXeHWkDwyidIgL5F1AZqqeuHLIGIWRzt6r8HgTeEUL8A3AIIS4GbgHutzqREOJBYBrQBxghpfxFCNEZeB4zUowH2ArMqS2dhRBiCfA7TMd6gKVSynutzl8XDoeLkpJCYmMTOqQCNAyDkpJCHI72v8qIFPYV72flvtV8f2AdZb5y0mK74rI7wy1WK1D5JUvjeNc2MtwSR9wQHKPPwdF3LFqH+AwUjUFKOTzoum9zjGnV1eFpIcRhYA6mn9+lwN+llMtDmGs58DCmc2IlBrBASvk5gBBiIaZCvbKOMe6XUj4WwpwNkpycSl5eLsXF+c05bJvC4XCRnKzyErckRuDBbxgGT/3yPEfK8xmdOoKJPSbQP7FPu/7ipVeUcaxrB+NdO/m4fCjS141Vnv6s96az+NI/dMjcgorGI4R4U0p5di3lb0gpz7M6TiiuDm9i+lc0CinlSgAhRHDZEeDzoGargLmNnaMx2O0OUlK6teaUig6E4S5BT83BSDyCx3cBmqZx+dAZdIpOJs7Zvq0X/Uf24t34Kbu3fcvZ0WXs9yeiBzLQFBtuig13mCVUtFFOqaM8I5RBrLo6aMBVwEVAqpRypBDiJCBNSvlqKBPWM4cNU/G9VU+z/xVCzAG2AzdLKTfVMk4SkBRclpmZmZaRkdEcYioUDWJofozkQ+ipORBXBLqGlp9CqbcMsNErIT3cIrYYlefnhqFT9v4ijLIC4oaewEPfQpa/E0enX4scdJ+X1NT4Wut8ngryCpT7U7gRQtwVuHQFXVfSD9gdynhWV353AacCi4EnA2V7MaNrN4vyAx4FijEDZtfG34D9AYvTS4H3hRD9aklrMQ+4PbggMzOTjIwMkpPb9zdtRXjx6YE/xehS9L5boTwaW1YftMNd0fxOkqITITq8MlZS14O+sXjzD1C47kNKt60j/coFaHYn5effgDO5G/aYeLJW1p1eqbllaQw2h5O1C2p3WR5741OkpjZ/guv6iITPJALpGXi1BV2DeXyWBdwRymBWld9szFhqh4QQTwTKdhKCZU19BIxhBgJT64rXJqXcF3T9nBBiEZBOTW2/GFgSXDBnzpxxwNK8vBJ8viaFg1MojsLj9/LDwfWszF5Nt9gu/PnEy6E0DvvmkVASj1ZtpdNa51sNPTybQw5D1/Fnrcez8VP8WT+DBo5ex3Bwbw62mCRwdYMSSI2pf5xI+Uzqo7llbI3fT3PgcNgiZtEgpbwcQAjxjZTyP00dz6rys2OuyuBXk624oLJGI4S4DxiL6UdYZwwwIUSPSgUohDgd8AP7qreTUuYD1a1X2u8+kyIs5JQcYOW+1azOWUupr4zU6M6M63oMgKnwShLCLGHLUbm16T+4jbIPFqPFJOEaMxXn4JOxxXUOt3iKdo6U8j8BH3OBqYeC65rdz+894J9CiL9A1Rng3UDNQJR1IIR4BDgPMwP8xwHr0QuAmzH9B78JGMPslFKeG+jzIzBFSpkNPCuE6AroQCFwlpTSZ3V+haKpeP1eHDYHmqbxxd5v+Tp7NcekDueE7r9hYHI/bO3YIdswDPz7Jd6Nn6LFdcI94SLsXQcSfcY87OnD0WyWbecUiiYhhJgNPI65+CoNqmoRP7//BZ4FCgBnYNIPMV0eLCGlvI7aE9/WeQIupTwm6Pp3VudSKJqTnJKDfJ29mtX713L1yMsYkNSXM/pMYkrf3xHvimt4gDaM4SnFu+VrvJs+Q8/LBlcMruGnAqBpGo5exzQwQv0oQxNFI7gXOF9KuaIpg1j18ysEzhVCdAF6A1lSypymTKxQRDJ+3c+6g+tZmb2Kbfk7sWk2RqUOJ9phmucnRrXfbc1gKla/infT59hS++I++Uoc/cejOZrP+KMhQxMz9oVCcRQOzMVXkwexRMCF4FTMfH7ZQoj3pJR5TRVAoYgkyn3luB1uDAxe2/oWbnsUZ/efzIRu40hwtW8LPMPnwbfjOzwbP8V9/EzsXfrhGjkZ5+CTsac2S1ANRTtHCPF7zCMxLfBzp5TyDSHEIMzdw86YyREulVJubeQ0DwC3CiHubkpCW6t+fpOANwCJaV3ZC3hcCDFNSvlJYydXKCIBr+7jp9xfWLlvFXnl+dx+3I04bA5uGPtHOkcnt+uzPAB/fjbejZ/j3fo1VJRgS0zD8JhHKbbErmGWTtFWCNiCPA+cGAhfORL4WgixHNNF7nEp5QtCiJmYmYIaG8j2L5i2IzcGbEeqkFL2sjqI1ZXfY8DVwQ7tQojpmIeOg61OplBEEofLjvDlvm9ZtX8Nxd4SOrs7cUL38WaWBbuN1Jj2abmo+7ykpMSZzuh+L7ufuw/dU06sGE/siFMoTRzQrsOtKVoUHajM9pME7AdSgDGYO4dg5oZ9TAiRWlscZwvMbLKUWFd+3YHXq5UtA5rsa6FQhIvskhw+zfqKkSlDmdh9AqLTgHa/yutsKyLv85c4/N27FNnjQdNw6OC3xZK3bRNjz7uesgjxMVNEDitWrEibN29en2rF+QHXMgCklIYQ4gLgTSFECRAPTMF0SN9XGZBESukXQmQHykNWflLKLxp5G0dhVfk9D1wLPBJUNhd4rjmEUChak7zyfJLdSQztJLjn+FvavfGKHZ0hzmyOde2kvyOXgu9s6Nir6n02lU1BUT8LFy5cWkvxnQRFVRFCODBd186WUn4thDgBMwLYrOaUpZbQZlVIKW+zOo7Vr7mjgYeEEHuFEKuFEHsxs7iPFkJ8WfljdVKFIlzsLNjNHd8+wNoDP2K32du54jPjUQjHfi6K+Y5OthI+Kh9Krz/9mxJHHKitTYVF5s+fPx3oW+1ncbVmxwDdpZRfAwReS4ByoIcQwg4QeO2OGZKsMfSs9nMscANmajzLWF35/Qe1xalo45T5ynhmw0skRCUwpJNouEMbxNB9+Hb/yP6Pv+LkKC9fVAxG+rrxbMnxbPN1xUBjTnxyuMVUtDEmT56cM3ny5F0NNNsLpAshhJRSCiGGAF0x87T+CFwMvBB4/aGR531VYc6CEUKcERjXMlb9/J4NZVCFItIwDIOXN79BXkUBfxkzlxhnhESYbib0okN4N3+Bd/OXGGUF2OM7U2aYUf382NjqSwuzhIr2jpQyRwgxF3hNCFHpgnCFlPKIEOIazChdtwF5hBAgxSIfAv8XSod6lZ8QYixQIaX8JfA+FXOpOxz4FrhBStnk+J4KRUuzav8a1h78ian9zqBfYu9wi9MsGIaOFjDQqfjmRXy7f8TeaySuIaeQNuZ4vvvrK2GWUNHRkFK+CNRI4SGl3Az8pjnmEEJUD2EWA8wgxG3UhlZ+izEPNX8JvH8Kc6/235hLzAXAH0KZUKEIBz7Dx9BOgtN6Z4RblCajFx8xV3nyK2Km/hVbQhdcv5lO1AkzqwJLazZ7A6MoFG2WbZgH2pWH1qXAD8BloQzSkPIbAnwFVRFeJgPDpZRbhBBvAd+glJ+iDXBij+OY2H1Cm/VfM3Qd/971eDZ+jj/rJzDAnj4Mw2cmQrEndQ+zhApF6yClbBZ/pIaUn4Nfg+tNAHKklFsCAmQFFKJCEbG8vf19esR3Z0yXkW1S8Rm6jmazYVQUU/bho2hRsbhGnWmmD0pIDbd4CkVEIIRwAf8jpXzcap+GlN8GYDqmr8ZFwMdBk/XAzPKgUEQkPx/ayPu7P+WU9ImM6TIy3OJYxjB0/Ps24t34GYanlJjf34QtOoGYqTdjS+mDZlfpgxQdEyHEbzFdKrZJKd8M+Bb+AbgJOIIZdcwSDf0X3QS8LYR4EjN57MSguguBr0MRXKFoLfIrCnh+06ukx3Xn7AFTwi2OJfTSArxbvsK76QuMolw0dzyOQROrVn/2rgPCLaJCETaEEDcBf8dclA0TQvwLyAAqMMNvvhvKePUqPynlSiFEL2AQsEVKGRz36F1AmZMpIg7d0Hl2wyt4/V6uGDYDZwQnWjV0HXQfmsOFb+f3eL57DXs3gfPYaTj6jkWzq+grCkWAOcDJUsq1QogJmIuv66WU1Z3tLdHgUyGg8NbWUi4bM6FC0dL8cmgTW/K3M3PwdLrGdgm3ODXQywrxZ/2ML2s9vr2/EHXsNFxDJ+EceAL2HkOV8YpCUTspUsq1AFLKVUKICuDhxg7WKl+JhRAPAtOAPsCIIL9BSzmeAuFwHgHOwDRxvV9K+VRryK5oe4xMHcZfxsylf2KfcItyFIbfS8myu9BzdwCgRSfg6D0aWyfTGV1zRWN3tS/ne4WiOQmkTar8KQ+UVVl/hpLfr7X2g5ZjauivqpVbzfF0CTAAGIipKH8QQnwspdzVciIr2hql3jLyKwroHpfGgKTwJl+N0SoY6DjAQMcBDDReLxuHZndi75yOo/cxOHqNxNa5V5WTukKhaJA4wBf0Xgt6r2EujCw7uLaK8pNSrgQQ4td4ikKILljP8XQh8J+AVs8NJEecDiysPlfA/eIoF4zMzMy0jIyM5rkZRURiGAYvydfZdFhy1/E3E+uMCYscRes/Y07sZ/Sw52HToER3sdH76zam+6QrwiKXQtEOaNZvtJaVnxCiM2Zupm5SygVCiO6ATUq5t5Fzh5LjqRdmBvlK9gTa1cY84PbggszMTDIyMkhOjm2kqIpI55PtK/nh4HpmjDyHPt1bJ/u4v6yIsh0/UbrjR1JOvxKbK5o8aSaW/qxiCFt8Xcn2J2MEAlHoPi+pqfE1xtF9XmyOyDFsqU3GcBEpsrS2HJFy35GElHJ3w62sY0n5CSFOxkxmuwY4ATOs2UDMNBJTm1OgZmAxsCS4YM6cOeOApXl5Jfh8lreEFW2EnJIDPL3uVQYnD+S4zhPIbcFkrHpJHt4tK/HvWY//4DYwDIiKRR9wMvaUPqScMI3Mtypq7WtzOFm74Koa5WNvfKrZZW7Kw7O9yhIpckDDsrTk33AoOBy2drtosLryWwxcKKX8RAiRFyhbDYxvwtxZBHI8BVZ99eV42gP0Br4PvK++EqwikFk4v1pxehPkVEQwXr+Xpze8RJTdxaVDL2z2TOyGpxTf3g3Y4lOxp/bBKCvE8/3r2FL64Bo9FUfPkdhS+6HZzHnbYhQZhaIjYlX59ZFSfhK4NgKvnhD610BKeVAIYTXH01Lgf4QQb2AavJwDnNjYuRXtB03TGJw8ENFpQLMkpjUMA/3IXnxZ6/FnrcefsxUMHefQSdhT+2Dr3IvYmYuxxajIfoqWobzCx8H8MgpLPBSVeHA4bIzo1zncYrU7rCqvjUKI06WUHwSV/Q742UpnIcQjwHlAGvCxEOKwlHIYUGeOJyHEe8BtUso1wPOY6TAq3SDuklLutCi7op1iGAYOm4PzBv6+aeN4ytALD2BP6QNA2fuLMEqOYOvcC9eoKdh7jcTexUwSrWkamlJ8ihAwDKNqR2DPgSIO5pWhy1wqdFfgPNjAbTNDKJf6o5l+y9GBSvqkxSvlVwtCiJ5ADynlqsb0t6r8rgfeEUK8C0QLITIxz/rOttJZSnkdcF0t5XXmeJJSTgm69gNzLcqq6ADkleeTuX4JFw+eRu+Eumyf6kYvOoRvx/fmCi9nC1pUHLEzF6FpNty/nYstPgVbrMp4rqgd3TAoLvNSWOKhuNTL4N7m38qqjTls3JlHYamHghIPhSWmUnvo2hMAePvrXazdUrm55UTDwMavdggOzcslk4djxyAx1kVCrIukuKhWvbdIJxB17GXMGJ8GECeEOB84Q0pZ81C9Dqxmcl8lhBiF6W/3NOa53PgmWHoqFI1GN3Se3fgKB8oO4Xa4iU9w446q3VqyvMJLUWH50f0LcihZdhd4SrF1Ssc14nTsPX8NfO1IG9ii8ivaBoYBOzwp/PT2Bk4b24Mop533V+/hg+/2UFTqRTeMqraZN5yM02Fnd04Rv+w8TEJAcaWnxJIYF1W1+jvv5H6cNbEvfXsl84e7Xqf6EbHL5mP6bwdFjMFLhJKJGV7zRMzgKAAfAQ+FMohVa89jpJQ/Ylp5KhRh5YNdn7I1fwezhlxA15hU3FFOZtxYI3k0AC8tuIQijlZ+RkUptrhOuH/7N+zJPVpDZEUbwmPYWV/Wg9Wlfcn1J+BcuYPjh3QhKtFOalI0owakkBDrJCHGVHCJsa6qbc0LJw3kwkl1f3nq1tm0nEyOd9dQfArLjAfOlFLqQggDQEpZIIRIDGUQq9ueHwohcjGXmi+q8zZFuNiev4t3d37EsV1H85u0sSH1rfz2be/Sj5hpd6noKooaHPTF8cyREygzXKQ5Cjgn4Qcu/dvfyc8rBWCsSGWsaNk8inX5gwL4PBXkFXhqretAHMCM+LWlskAIMRTTK8AyVpVfN8y4mhcDPwkhNgAvAf8npTwYyoQKRVP4Jvs7Okd34kJxbshuBZ61y8HvxTX+fKX4FFX4DRtfr8/GDaTYixnq3s8I9156O4+gaeB0WI6Y1SzU5Q8Kpk/or/nFOywPYtqg/ANwCCEuBm4B7g9lEKtnfn7MPdZ3hRDRmIYucwNCqNNYRatxyZDzKfQUEe1wh9TPu+N7POvexDHoREDtN3V0DAN8hgOP4UTHzvPvbeQqDWwaTE1YH27xFPUgpXxaCHEYM8VRFqaXwN+llMtDGSckPz0hhBv4PWaszXHUDFStULQIPx/aSM/4HiRFJZIUFdLWPv4jWZR//hS2Lv1wn3ipckTv4HgNOxV6FAY2NHSitAoW/eVMNj6s0pO2BQKBUd4E3mzKOFYNXqYAM4CzgI2YSWznSilzmjK5QmGF7OIc/vvLC4xMGcYVwy8Jqa+/tIiyDx5Bc0UTfeqfVHLYDsre3GJwmI+7SvcCl60CO340DaKjIjfhcVsisEBahOkHXg58K6W82mr6OovkCCGWYtqffN1YWa3+xh/ENHYZLaXc3tjJFIpQ8fi9PLPhJdx2N9MGnhVy/4r92zDKi4mZcr3y2+tg6IbB+u2H+ej7LDbtzmPqif0AcGg6Dnt5A70VjWQBptIbJKU0hBCVUeatpq+zwmmY9icvCyH8mIuxl6SUloKuVGL1zG9o6PIpFE3njW3vkF2Sw7WjriQxKvTAxDH9RxM340G0qPYZnFdRO5//uI/3V+/hYF4ZyfFRTDu5H+f9VvD5N+o8r6UQQsRhnr+lSykrXRAOhJi+rkGklD8APwA3BpIuXAx8KoTYL6UcWX/vX6lT+Qkh/ialvDdwfVc9gtxmXWyFwjo/H9rIV/u+5be9TmJoZ9FwhyBGOfdUpRJSiq9jkFdUQXK8aX+3I7uQ+Ggn553UjzGDUnHYbSTEusIsYdtlxYoVafPmzetTrTg/kEigkv6YW5q3CyFOAYqBW4EyrKevC5XNwCZMN4eQolPUt/ILzoQQevwohaKJ9E/sy+m9JzGl7+9C6tfDfoRzoteR5e+MYagUVu0ZwzDYkpXPh99n8eO2Q/xt1jj6dU9g1mkCp0O5szQXCxcuXFpL8Z3AHUHv7UA/zAQF84UQvwHexkw83mwEEpZPw7RDmQB8CDwAvBXKOHUqPynl3KDryxsnpkIROn7dj45BjDOas/qfEVLfOK2MS2JWUWy4eaV0PL9T/nztEp9fZ9WGA3y0Jousg8XERTuZMqE3nRPMlV9zKj5PPU7nFR4PhQW1529sT8yfP3/6vHnz1lQrrp46bg/gw9zWREq5WghxCHPlZzV9nRWygW8wfc2nVVt9WsaqtecRKWWnWsoPSim7NGZihaIu3t/1CT8f2si8MdfgDsGfz46fGTGrcWteMoszKDWUC2p7w6/r2G02fH6dlz/ZSqeEKGZPHsyEoV1xOVvGGd3lcDL7mT/XWrfk8oeB9q/8Jk+enDN58uRd9bWRUh4SQnyGebb3YcDCswtmJBar6eus0F9Kub+Rfauwau1Zwz5cCOHEXOYqFM3G1rwdrNj1CcemjQ5J8QEMduynl+MIL5eM54Aemi+gIrLZlVPIR9/vZW9uMbdffixul4PbZ48jNSla+W1GFtcATwshHgK8wCwpZb4Qos70dVYQQpwkpfwy8HaIEGJIbe2klJ9aHbNe5SeE+AozZYRbCPFltep0zKWnQtEslHhLWbLxZVKiO3HhoHNC7r/Bl85jRXHk6CrfXnvA79dZs/kgH63JYuveAqJcdiaO6IbXqxPlstMlOSbcIiqqIaXcAWTUUl5n+jqL/AsYHrj+bx1tDMwzR0s0tPJ7CjMW1LHVJjQwg4ta1rIKRX0YhsGLm1+jyFPMDWOvDWnVV7ZnI93teWT7k5Xia+MYBviw4dR0Vq7fy7+W/0LXTjFcedZwTh3fi9hocxOqo5y1KUyklMODrvs2x5j1Kj8p5bMAQohVAc3d7Agh+gDBMdmSgITqZ4xCiDuAP2AedgJ8LaW8tiVkUrQ+Jb5ScksPcXb/yfRKSG+4QwC96BAH3lzIOdEaTxRPqnJvUDSOcBh3FPjd7PCksMOTyg5PCmOj9zApTnLCyHQyez2LN/4Qy3fA8h2/9ukoZ22Kmggh3pRS1kikLoR4Q0p5ntVxrDq5bw546o8HUgiKDCylfNrqZHWMvQszIy8AQojF9cj1nJTyhqbMp4hM4pyx3DjuT9ht1o+RDV8FZR8+An4fr5aeoBRfM9Aaxh2GAZpmvv7v4i/Yesj0fY61VdDPlUu6Mw8Ah92GK+FQk+dTtDtOqaM8I5RBrFp7noNppbMVGAZswNx/XYmZ2b1ZEEK4MLPFn95cYyoiG4/fw4pdn3B671NC2uo0DIPyz/+LfjiLtAtv5tC/W2RjQtEMeH06PsOG33DgM8wvN7H2MjQNxg3pSt+8r+nnyqWLowib+v6iqIOgYCuuWgKv9AN2hzKeVWvPe4DLpZRLhRB5UsrRQojLMRVhc3IWZiSAdXXUXySEOA3IAW6XUn5bvUHAAfKog5/MzMy0jIyMZhZV0Ry8vvVtVmavZnDyQESnAZb7+bavxrfjO1zjpxMzYCxmoAdFJFCZNBjg3W938c63u6nQYyAQUNqh+atWfzNOH8zan3bUO55CEaAy2IqNowOvGJg+g3eEMphV5ddLSlndw/9ZTCXUnNuQV1D3SvJJ4F4ppVcIcSrwphBiiJTycLV284DbgwsyMzPJyMggOVmFuYokVmWtY2X2as4afBoTxeiQ+hqdJ1Ec5yJu+EkNtq3rDCuSaAsyQt1yHswr5actufy4NZf1Ww/x0J9PokunGAb26cypPoOPv92II5BBoaVlaW1aW45Iue/WpjLYihDiGynlf5o6nlXld1AI0VVKeQDYJYQ4DjhEM/r5CSF6ACcDs2qrD06fJKX8SAiRhbn1+kW1pouBJcEFc+bMGQcszcsrwedT4a4igcNleTzx/Qv0ju/Jb9MyyM0tstRPLzgAzihsMUmQNobyQ8UNPgysjt0cNPbB1NwyttQDsrqcO7IL+c/bGziQVwZAYqyLoX2SOZhbhOb3M6hbPCeM7M4Xq0IKuN8oWRqitT4TKzRFltb8e3Y4bBG3aKhUfEKIeGraoFjeRrCq/P4DTARex8zV9BmgAw9ZncgClwHv1rKSA0zlKKXcF7g+BugDyOrtAqFuqoe7sW4+qGgVXt2yHMPQuWL4DBw2a3+GhqeUsvcXgcNFzHl3KufmVsTQbfy0NZdvf9rHxl1HOGFENyaNSadTQhRdO8UwaUw6Q/sk0z0lVv1eFC1KwMH9JWAU5panFniFEBZkvwaAJwAAIABJREFUVq09Hwi6fk4I8TkQK6XcZHUiC8wGrgsuEEK8B9wmpVwD3CeEGAv4AQ9m5ACVTLeNcqE4hwMluaREd7bU3tB1yj7NRC/MJfrM+e3qARvJsSMNA4p2j8RbksytG7/Bpmn065FArNv0t0uKi2Le9FFhk0/RIXkCcwF2CrATcyH0D0IMutKo9MVSyj2N6dfAmINqKZsSdH1Zc8+paH3yyvNJikqkkzuZTm7ryWU9a97Av+cnok6YhaP74BaUsPWJhNiRhgFH/DF8sGoXRXuGgQbxPTegaaDZfbg7ZTP/nPNJS4hSWc8V4WYUcGrA/kOTUhYIIf6/vTuPr6q8Ez/+OXfJvdk3AgkJ+/LIJggiuCFQRXG0alvFcbQVa8FlatGOdto6VWd+7TjajrVVR6xFUKwLtbVWi1ZrbUHFqoisfkXZ95CFkPXm3nt+f5wbDfEC2e6S3O/79cqL3HPOfc43JDnfPOc8z/e5FViPMyuhXY61nt9OPu9KHpWIDGzvyVRqqw3U8dP3H+TEPmOYY9pfviy47QMCa17Ee8I0vKM7u/izai1s89m0gr/VjuAfDYOpC/th2Ye4PLl4sz9/+pA9YCMAp4wujuvzJqWOohGn3nQzcNAYMxCnXmj7biNFHOtPuCs7H5tSRzpQf5CH1y6mNlDLaf0nd+i97pKRpI0/n7STL+lVtzvjybYhhJuQ7eb2h9/ko/LzuLnPa/hdQdJdAYallTPQW8n519/M7S/d3q2jMpXqZiuAy3AGNv4WWI5ze6RD5TaPtZ5f21GUSnXK5qpP+dW6JwD41wnXMiC7tF3vs5vqwJOG5cvEN+WyWIbY61TWNH42l6457KHR9tEyLuBwfTPj/TsJ2s6ad6dkbOeUyPzgAf2yNfElsWR+PhwvItL6YvADnKIrWcDjHWmnvRVefMCPcNZhKhSR3Mhk85Ei8kBHTqhSS1MowKPrl5KdlsV1J86lKKO9A1yCNLz6ANhh0i/4d+3xHYNt2+yrrGfzrkNs3lnN5l2HOFDdgN/y4LWCuK0QaTTjtkK4CXH/LRfz/j3tfjSikkgyPB9OJiISBp7ozHvb++T6PqAUp/TY8si2DZHtmvzUF4TtMBYWPnca14+fS7+MItI96e1+f9OqZwjt2YR/+rc08bURDIXZtu8wXreLQcXZVNQ08sNfvQNAdoaXEWV5XDhtKE+/+A8AXJaNzwokMmSlusQY8wTtG4PS7nUC25v8LgGGi0idMSYcOcnuyMR0pY7QEGzksQ2/YUTeUM4ZNJ3BOR0bE9UsK2he/yrecefiHXl6jKLsWdZvqUAivbqte2toDoaZMrof8788hsIcP9+6cDRDS3Lom+8s7lpUlM2zL72T6LCV6i6fdHeD7U1+gbbHGmOKgKgT0lXqOthQycNrH2N/fTnj+ozu8PtD+z+hccUS3KVjUvY5XyjgI1ifRzjk/Wzbs3/9lD0H6xhUnMWMk0oZUZbL8DKnhK1lWZw6pjhR4SoVcyJyV3e32d7ktwxnCfqbAYwxJThlxJ7u7oBUz/VJ9VZ+te5xQnaYG8d/kxMKRnS8EV8G7tLRpM+Yh9WB5Y16ukBNH5oO9XWSXrOzuoXL04htO3d6brhkLPlZPnxpqfN/olQ0xpijzncSkXaP+Gxv8vsB8D/AOiADZ2mjXwFtl5VQKepQ02EeWPMo+f5crjvRecbXEXY4BJYLd15/MmbfEqMoE8e2YdPWSlY3DOBgMMv5CGUxr2AFAMGGHJrr8vFmVOPpswNvRjVufx2WNQeA4oKMRIavVDL5dZvXRUAasAtnaaN2aW95swBwM3Bz5HbnQRE57sNHlTpyfdlcPeafGZk3lAxvxy7Utm3T+PfHAPCf9c0eO8AlHLbZe7COoO0mbLsI4yLNCuCybJptD7c9sAKYgJsQhZ46ij01BCLr26UXbSW97xadZqDUcYjIkNavjTFu4HagQxUYOlynSETKIycch1N389KOtqF6h8ZgI49vepbTSiYzts8oJhSN7VQ7zRteI/jxStImXtQjEl9jIMj+ygb2VtQxpCSHfgUZbNpexX3PfkgwFAacUa0WYTyWhQsbjxXie9dMofr5n5Dnrv/Coq2WS/+WVKozRCRkjPkxTs/vf9v7vmMmP2NMBvB9YALOrc47cZaQ+BlwDs6afioFVTRUsXDdYvbU7mNUwRfKsrZbcPdGmt5+Cs+gk0ibdFE3Rtg1tm1TXRvAZUFulo+qw00semkjeyvrqaz5fC7VFWePoF9BBv3y0zl7UhkjBhfw69++hYvwEb04l2UzeXQx779Yn4CvRqle7xyclYba7Xg9vweBk4BXgNnAOOAEnKT3LRE52IkgVQ+35dB2Hlm7hKAd5Mbx32RUYeeSX7imnIbXHsSVV4x/xjwsy9XNkR5fy6rjoXCY5at2sLeinn2VdeytqKcxEGL2lIFcOmM4GT4PdY1BzIA8igszKSnIoKQwg775zi3eghw/l80cTlFRNoufWxn3r0OpZGKMuQOnszRORNYbY6YCC3Fui2wDrhSRA51su23d6QzAD9zQkXaOl/zOBSaIyAFjzC+BHcBZIrKiIydRvcfeuv3cv/ph8vx5LDjxOooz+3a6rXDtQay0dNJnfQcrrf0T4Luqpi7A7/6+BdlRxbDSXK69YDQuy+LP7+4kzeuipCCD08eVUFyQwYiyXAB8aW5+dHXHapIqlYqMMROBqeDUzDPGuHBWW7haRFYaY24H7gau6eQp2tadrgM+FpGajjRyvOSX1ZKdRWSXMaZWE19qK87oy0XDZnNKySSyvF1b4dnTfxSZc+7Gauditl1l2zZvrtvHM69vpjEQYsLwPgwrdZKbZVn87MbT8Hp0KoFSnRUphfkgTinMNyKbJwGNItJyS+RhnN5fp5Jfd9WdPt5Vx2OMmUGrZeLbvu7IvArVMzWFAjwjv2fWoOkUZ/Zj5sBpXWtvzYtYLg/ecefGLfEBvPreLp7+y2ZGlOXy9fNOoLTPkclbE59S0S1fvrx4wYIFg9tsrhaR6jbb/hNYKiLbjDEt2wYS6QUCiMhBY4zLGFMgIpUdjcUYk4uz8PlJOAWtPyMis9rbzvGuPAeARa1eV7R5bdOBeRWq56lqrObhtYvZXbuXEfnDKM7s16X2gtvXEPjHc3iGT8F7/MO7rDkYpuJQI4W5fs4YV0J6mpvTTyzB1QNGlSqVLO69995lUTbfhfNcDwBjzKnAycC/xzicZYAb+D3Q0NlGjpn8RGRwZxvuCGPMNpwFChsjm74nIq+0OSYDeAynCx0E/k1EXoxHfKlqW80OFq5dQiDUzPXj5zKmsGsrqIcOfErD6w/j6jMQ/7RrYj6tIWS7WHDfG4RCYe6cO5kMv4czx/eP6TmV6o1uvfXWSxcsWPBem81te31nAaOArZFeXxnOYMlfAINaDjLG9AHCnen1RUwF+kTmn3da/O45Hd/XRGT9Mfb/G1AjIsONMSOAFcaY4SJSG6f4UsqWQ9v4xQePkJuWw00nzaOkiz2+pn/8lsCal7Ay80mfdROWJ62bIv0i24YmO41m20t9Y5ArvjQCtyv+I0mV6i1mz569b/bs2duOdYyI3I0zkAX4rFNzAbARmGeMOSPy3O86nN5bZ63EmXWwtgttJFXyO545wDcARGSzMeY9nOkXXflPVEcxILuMM/pP5bzBXyIrrXMDW2w7DDZYLheuvGK8Y8/Bd/IlMR3ZGbYt6sPp2Fh4rWYevHUGdYcbj/9GpVRMiEjYGHMVsNAY4ycy1aELTV4N/MkY8w6wv8252l1yM5mS35PGGAsnq/8gyoPUIx6a4ky7GNC2EWNMHpDXetvChQuLp0+f3r3R9kKBUIA/bnmF2YO/RIY3g6+N/HKn2wod3E7jm0/gHT6VtDFn4x15Bt7Oz4U/rpZVyy1sZ/FWqxm3FSbD79Xkp1QCtH5sJiJv4cwT7w4/xrn2bwNyWm3vUJmkZEl+Z4rIzsgw2Z/jLJDb2b8MFgB3tN6wcOFCpk+fTn5+14bm92aV9dX8cuUjbK3ayfgBhlP7T+pUO6HGOqr+9hT177+CKz2L3KIisoqyuznaVucL27y4cgt14QwyXfVYFqRbR65mXRTD88dDMsWvsXxRssQByRVLDF0OjBSRvV1pJCmSn4jsjPzbZIx5CHghymE7cB6alkdeDwT+GuW4nwOLW2+YP3/+ycCyqqo6gsEOVcBJCdtrdrJw7WIaQ03MP/EbDPePpLy8QzViAQhu/4DGvz+G3XgY76iZ+CZ/hQZfJg2daKs9dh6oZcnLH7FlTw1uwthYWG3++AsHm3F5oo8rDQaaqDrUvSucx+Li05nvRawugj05lmSJA5IrlmPxeFzJ2GnYAjR3tZGEJz9jTCbgEZFDkduelwNrohy6DJgPvBcZ8DIZZyLlESK3S9veMi3r3qh7j00VH7Nw3RKyvJl8d9KNlGaVtOt92Tl+/L4jk0pDXT6VBcVkz/w+Tenta6czQuEwz6/Yysvv7CDd5+G7V0xk4dN/j7oigsvj5f17ro3azqTbHsVZp1kp1YM8AbwQqTrW9plft6/nF0v9gOciy1K4cUYG3QBgjFkDnC8ie4B7gcXGmE+AEDBPRGLTpUgh/bOKGdtnFHNGXkx2Wtbx3xDh93mZe9tiZvo3EbJd/LmpZUWHMfzm6s71HNvLZVnsPFDL1DH9mDNzBEMGFvDIMzE7nVIqudwY+fcnbbZ3aN55wpOfiGzBmakfbd+EVp/XAbp8UjcIhJr52643mTngTHJ9OVw7tmOPV23bpnbDCr6T/SpZViP/CAzF+bmzaFX8p1sdrg/w2zc+5YLTBlOUl86/fmUcHrdOX1Aq1bRdz6+zEp78VHwdaqph4dol7Di8iwHZpZxQMKJD7w8f2k/jisXU7tnE4XAeTzZOZXeoIEbROon27Q37ePovn9DQFMQMzKMoL10Tn1KqSzT5pZCdh3fz8NrF1Acb+Na4qzqc+ACww4SrdtPnvHn8xzPl2DHq6QEcqKrn8VeEjduqGFaawzfOO4Gyovbfmj2eQLD5qAMPmgIBag41Rd2nlEqcKEsafUZEBra3HU1+KWLdwY0sWv8kmd5Mbpl4AwOy21fmy7ZtglvfI7TnI/xnXIUrr4TMK35GTnEB9jNPxjTmV9/dxda9NVw1ayRnnVTa7fU40zxern7sO1H3LZ57P6DJT6kk1PY5TQnwHeDpjjSiyS9F5KRlMzhnIFePuYJcX/uGWYer99L45lJCuzfgKhyIHWjASkvHcseuJPWWPTW4XDC4OIdLpg3l/FMHkZ/ti9n5lFI9S7QljYwxbwAvA/e3tx1Nfr1YKBxifcVHjC8aw6CcAdx00rx2FZO2g00EVv+RwNrl4EnDd9qVeEfPwHLFbsmfhqYgv/v7Fl5/fxdjhhRwy5wJZPg9ZPj1R1QpdVxNQIcGwuiVpZcKhAIs2vAk6w5u4t8m3ciQ3EHtXkXBDgZo3vQGnmFT8E2ZgysjN6axfrC5nKV//pjqw03MnFTGV6bpKllKqeiMMW3rd2YA5wPLO9KOJr9e6HCglofXLmZ7zU7mjLyYIbmDjvuecM0BAhv+4iQ7fzYZc/4blz/2pZLelwM8+Pv1lBVlcsMlYxnWP7aJVinV47Wt6VwH/C/O5Pd20+TXw0SrrNKisamZrft38+CaX1PVVM21465iQtHYqMe2sIMBAmteIvDhS+Dy4B15Bu7CATFNfGHbpry6gX75GYwf3oevn2s448QSnb6glDouEZnbHe1o8uth/D4vV9wWfZTlb+75F7bX7KIuWM9NJ81jaO7gY7YV3L6GxreexD5cjmfYVHxT5+DKzI9B1J/bXV7LkpeFA9UN/Pe8qaT7PEw/qTSm51RK9XzGmNOBL4vI96Lsuxt4XkRWtbc9TX69hO126rxO6jeeUQUjyfAee808OxykadXTWG4v/n+6DU/p6JjFFgyF2bitklUb9/PupgOk+zzMmTkcf1rsBtAopXqdHwAPHWXfG8APgQvb25gmv14gXLCf8MBP+fjgNPIpOmris0PNNG94De+oGVheP+mzb8HKLMByd/+PgW3bNDWH8Kd52FVey8+XrSXT7+GsCf358hlDyMmI3UruSqleaQLOdIZoXgMWdaQxTX49mI2NXbyLcOl2rJpcynJLqKsORj02uHMdjW8txT60H8ufg3fk6bhy+nZ7TCHbYunyTbz+3g5OGJjP3PNHMahfNjdfNp5Rg/L1uZ5SqrNygDSgIco+L9ChgQqa/HooG5vwgE+x++7DqijCtX0EGd506jhyNYVwXRVNbz1JcOt7WLn9SJ/9XTwDumtB5c81hz0EbC9h3Cz7y8ecMCifMUOcmp+WZTFuaGG3n1MplVI+AmYBf4iyb1Zkf7tp8uuh7IIDTuLbV4Zr9yCso9TYbFyxhNDuDaSd/BXSxs/utuos9Y3NNIc9eKwglgUhnB6dz2rikR99mVBTl9eaVEqp1u4DFkaWv3teRMLGGBdwMfAgcEtHGtPk10NZlX1xBdNw1XxxdGa45gB40nBl5OE/7QoIh3HlFXf5nMFQmHWfVvD2xv2s/eQgAdtPhlWPmzA+K4AVuaNZkOOnvFyTn1Kq+4jIb4wxxcASwGeMOQj0wanucoeIPNWR9jT59SAVDZU89NdHsdNysAJ+rDaJzw6HCKx9haZ3n8MzZCLpM6/rtud6uw7U8j+/WU1dY5DsDC+zpgzi9bc34CIMEHUVdaWU6k4i8r/GmEeBU4FCoAJ4W0RqOtqWJr8eYufh3Tz04SJCdhC8IyDgP2J/X9ch9iz5IU17NuMeOB7fKV1b93fXgVre3riPvEwf50weQHFhBpNMERNH9mX04HxKinP526p1XTqHUkp1VCTRvdLVdhKe/IwxhThlaYYBAWAzMF9Eytsctxg4GzgY2bRMRH4cx1ATZlPlxzy67gnSPencOfNmblv1+hH7R3n2MCfjHZqrs/HPvA7PsCntruPZWmVNI6s27mfVhn3sKq/DZVlMG18CgMft4urZo9rVTvgY6+QFA01UHQp0ODalVHI71rXcGDMVWAikA9uAK0XkQKJihSRIfjiLEt4jIm8AGGPuBe4Gvhnl2LtF5IE4xpZwH1Vu5qEPF1GS2Y8bxl9DWW7JZ/tchAnjYnuokNWBwVx6839QWd+xqQQNTUHSfc6PwdN/2cx7Us6w0hz+5ZyRTB7Vt1Pz8VweL+/fc23UfZNuexTn90Ip1ctEvZYbY74FLAWuFpGVxpjbca7x1yQsUpIg+YlIJc7s/BargOsTE03yGZwzkDNLT+XCobNI9ziT170EOdu/kQHuSh6tm0a97eOFxpO4PDMX6g8fp0UINIdY88lBVm3Yz7otFfy/a6fQryCDS6YN5WszhtM379jVYZRSqq1jXMsnAY0isjKy/WGc3l9qJ7/WIsNWrwdeOMohtxhj5gOfAt8XkU1R2sgD8lpvW7hwYfH06dO7OdrYCdthXt+5gjP6T8Xv8XHZyIs+29ewdS3fzn6NAlc97zQNwR3p/bVHZU0jv1+xhfelnMZAiNysNL40qQyvx3l/SWFmTL4epVTPtnz58uIFCxYMbrO5WkSqox3f5lo+ENjesk9EDhpjXMaYgkjCTIikSn7AL4FaINqtzR8CeyNzO74OvGyMGSoioTbHLQDuaL1h4cKFTJ8+nfz85L+4B4IBfrHqMf6xew3F+QXMGHoaAOGmBipefYy9H/6FsJ3Fo7XT2Bbqc8R7oz1r23XgMHUNzYwozSY908f6rZWcMb6U6RPLGDu8D25X/IdpHu15YLJJljiTJQ7QWKJJljggdrHce++9y6Jsvgu48yhvaX0tvyQmQXVR0iQ/Y8xPgRHAhSISbrtfRHa3+vxxY8x9QBmt/qKI+DmwuPWG+fPnnwwsq6qqIxj8QtNJo665nofXLmbroe18dcSFjM0eR3m5cxvTDgao3/ERuadezJ0v2wT5YlHolmdtIdtCmop5t2EQWwNFlHkr+b+759JQdZif3nAabpfT06usqO10rF35JWv5mrpLrH7hOxNnLGJJljigZ8eSLHFAcsVyLB6Pi/z8TG699dZLFyxY8F6b3Ufr9R1xLTfG7AAGtdrfBwgnstcHSZL8jDE/wbkv/E8i0nSUY0pbEqAx5lwgBOxue1ykG972m1LWvRF3v4qGKh768NccbKhg7pgrmNRvPOH6QwRWv4BvyqVYXj8ZX7mTwuICgi9HX9IIYHXDAP5aazgcTifXVc/MrE1M9O8AnCWwWhKfUkq11+zZs/fNnj172/GOO8q1/H0g3RhzRuS533VAtJ5kXCU8+RljxgDfBz4G3jLGAGwVkUuMMWuA80VkD7DEGNMPCAM1OOs6Ra/i3APZhLGx+dcJ1zI8byjNH6+k8e2noLkJz+CT8JSNjVqazLYhhBs3n9/97ec5zAUZ6xiRtp8E3NVUSqWg41zLr8IpTeYnMtUhYYFGJDz5icgGiF6YUkQmtPr87LgFFUe7a/dSktmPPumF3D7lu1BbQcPynxHatR53vxH4zpqLO6//F95n29BsewnYXmxc+Cznj6yJ6TuZmL4z3l9GuwWOMQewKRCg5lDUjr9SKskd51r+FtD9FfW7IOHJL5W9u+8Dntj0LBcPm83MgdNwWS7q31xKaP8n+E6/Eu/omVjWkbcpg6EwjWEfzbYHsHARwmc14rF6Ric4zePl6se+E3Xf4rn345TpU0qp2NLklwC2bfPajr/x/Kd/YkTeUCanlxGuq8KVmY//tCvBsnBlfz6SM9AcYsueGk6IrIcXti28VhCv1YzbSt4BPEoplaw0+cVZ2A7z3OY/8sauN5lYNI7LA9mE//ATmoadQvqMebhyij47dl9lPW98sJs31+2lMRDiZzeeTlERpLsaO1xIWm83KqXU5zT5xdme2n2s2L2K6YUncp58RLhyF56hp+CbMuezY3YdqOXp1zezcVsVbpfFxJFFzJxYSnaGM+ClMyso6O1GpZT6nCa/OAmFQ7hdbsqy+3NbyZfIXfEkVnouvlk34R08karDTTRW1FFSmIkvzc3+ygYumTaUaSeWkJvlS3T4SinVq2jyi4PKxioe+nAR5w6YxuT+kykdPJWmygOkTbwI2dfE679fxwcfH2Ts0AIWXDqeorx0/uf6U3HpInlKKRUTmvxibHftXh5c8yhNgTrS3vsj9gUTsfxZrM6ZwYuPr2NfZT2Zfg+zJg/grJM+n9KgiU8ppWJHk18MfVz1CQs/fIy05gDX7akit2waweYgXp+bypomMv0evvlPo5h8Ql/SvF8sV6aUUio2NPnFyIHqnTyw+hEKA0GuqPby1/DFrHo3g/kl1UwZ3Y/zpw7igtMGJzpMpZRKSZr82iE7x4/f98XSYgCNTc0crmn8wvZcXxGzKqGufCD31Iyhb2E2V5xdyrihhQC4tO6YUkoljCa/dvD7vFxxW/Ri0kt/clnU+XO2ncXquq9SNiSX/zptCGOHFWK1eo4XCgRwp0VfJb25sYnqw/Fb7TwUOPocwHjGkixxJFMsyRJHT4klWeJI9Vh6Ak1+XdSyjFA09333Ea59/GY2rQBWHLlv8dz7+dPX50Z93/mPPwZx/EF1p3mTIpZkiSOZYkmWOHpKLMkSR6rH0hPo+jYx5HHrf69SSiUjvTorpZRKOZr8lFJKpRxNfkoppVKOJj+llFIpJylGexpjRgJLgEKgAvi6iGxuc4wb+AVwHmADd4vIo/GOVSmlVM+XLD2/h4EHRWQk8CCwMMox/wIMB0YApwJ3GmMGxy1CpZRSvUbCe37GmL7AROCcyKangAeMMUUiUt7q0DnAr0QkDJQbY54HLgXubdNeHpDXetuiRYsGnH766bi7MPVg5OCio+7LLB1+1H0j+g456r78EUd/n8dz9FiTJZZkiaM3xJIscfSGWJIljp4Uy9G0umZGL3HVg1m2bSc0AGPMJOBxERnTattG4EoRWd1q2zrgGhF5N/L6NqBMRG5q096dwB2tt11++eXcddddsfsilFKqd7sB+L9EB9GdEt7zi4GfA4tbb6ipqcmqqKg4t7CwcDXQnJColFKq5/ECJwBLEx1Id0uG5LcTKDXGuEUkFBnY0j+yvbUdwCDg3cjrgcD2to2JSDVQHeU867svZKWUShl/TXQAsZDwAS8icgBYA/xzZNM/Ax+0ed4HsAz4ljHGZYwpAi4Gfhu/SJVSSvUWCU9+EdcB3zbGfAx8O/IaY8yfjDEnR455AtgCbAZWAf8pIlsTEaxSSqmeLeEDXpRSSql4S5aen1JKKRU3mvyUUkqlHE1+SimlUo4mP6WUUiknGeb59TrtKdQdpzh+CnwVGAyME5GEzXU0xhTijNgdBgRwRu3OjzKlJR6xPA8MAcJALfBtEVkT7zjaxHQHcCcJ+j4ZY7YBjZEPgO+JyCvxjiMSix+4Dzg7Es/bIjIvAXEMBp5vtSkPyBGRggTEcgHwX4AV+bhLRH4X7zh6E01+sdFSqHupMeZKnELdMxMQx/PA/cCKBJy7LRu4R0TeADDG3AvcDXwzAbF8Q0QOReK4CFiEU182IYwxE4GpRCnaEGdfS+QfSK3cg5P0RoqIbYzpl4ggRGQbMKHltTHm5yTgmmmMsXD+cDxTRNYbY04E3jTGPB+pdaw6QW97drNWhbqfimx6CpgYmZgfVyKyUkTaVspJCBGpbEl8EatwKvYkIpZDrV7m4vQAE8IY48NZyeT6RMWQTIwxWcDXgf8QERtARPYnNiowxqThrCyzKEEhhHF+VsHpge7VxNc12vPrfgOA3SISAoiUbNsT2R73W3zJyBjjwrnYv5DAGB4FZuHcQjovUXEA/wksFZFtxpgEhgHAk5FexkrgB5FSgfE2DOdRwR3GmBk4t6VvF5GVCYiltS/j/F6vPu6R3SzS+70M+IMxpg7IBs6Pdxy9jfb8VCL8Euei9kCiAhCRa0VkIPAD2iyLFS/GmFOBk4GHEnH+Ns4UkfHAZJw/CBL1vXEQ/WGNAAADxUlEQVQDQ3FKHJ4MfA/4nTEmJ0HxtLiGBPX6jDEe4PvARSIyCLgQeDbSS1adpMmv+31WqBs+W4E+WqHulBQZhDMCmJMMt21E5AlgRmRATrydBYwCtkYGnJQBrxhjZsU7kJbb4yLShJOMT493DBE7gCCRxwYi8g5wEBiZoHgwxpTifK+eTFAIE4D+IvImQOTfOpyfHdVJmvy6WQcKdaccY8xPgEnAxZGLbCJiyDLGDGj1+kKgMvIRVyJyt4j0F5HBIjIY2AWcKyJ/jmccxphMY0xu5HMLuBznZzjuROQgzioC50TiGQn0BT5JRDwR3wBeEpGKBJ1/F1BmIvfFjTGjgH7ApwmKp1fQZ36xcR2wxBjzI6AK5wF+3BljfgF8BSgGXjPGVLReNDjOsYzBuXXzMfBW5Pd4q4hcEudQMoFlxphMIIST9C5sGVyRovoBz0XuUriBjTiLlybKdcAiY8zPcNbfvCpBzx9bXA3cdLyDYkVE9hljrgd+a4xpuVtyjYjE/Q+23kQLWyullEo5ettTKaVUytHkp5RSKuVo8lNKKZVyNPkppZRKOZr8lFJKpRxNfkoppVKOJj+luoExZpsx5uxEx6GUah9Nfkr1cJHaj0qpDtBfGqVixBiTj7MO2xSc37U3getEZJcx5lLg30VkUqvjbwHOEpGLIksd/Ri4DPABvwduFpEGY8x0YClOgfCbgVeBq+L3lSnV82nPT6nYcQGP4axbOBBo4PPVEl4AhkTqNLa4Cng88vndOMWcJwDDgVLgR62OLQYKIm3HfZVzpXo67fkpFSORQsjPtbw2xvwYp2gzItJkjHkGuBL4YaT26WDgxUhx6XnAiS31GyNFwX+DUx8VnMVN70hUgXClejpNfkrFiDEmA7gPZ7Hc/MjmbGOMO7LY8RLgKWPM7Ti9vmcjSbEvkAG832qBWwun6HSLchFpjMfXoVRvpMlPqdj5LmCAKZHK/BOAD3ASGSKyyhgTAM4Eroh8gLN+XQMwRkR2H6VtrUivVBfoMz+luo/XGONv+cDp7TUA1caYAuCOKO95HOc5YLOIrASILPL7K+C+SC8QY0ypMebcuHwVSqUATX5KdZ8/4SS7lo88IB2nJ7cKeDnKe54AxuKM3mztezgLuK4yxtQAr+H0IpVS3UDX81MqgYwx6cABYKKIbE50PEqlCu35KZVY1wPvauJTKr50wItSCWKM2YYz+OXixEaiVOrR255KKaVSjt72VEoplXI0+SmllEo5mvyUUkqlHE1+SimlUo4mP6WUUilHk59SSqmU8/8BbsAhqhX9vsoAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "g = sns.barplot(x='Layer', y='Relative Speedup', hue='Bitwidth', data=merge_df)\n",
    "\n",
    "g2 = g.twinx()\n",
    "g2.grid(False)\n",
    "g2 = sns.lineplot(x='Layer', y='Runtime (ms)', data=cumul_df, hue='Bitwidth')\n",
    "#g2.set_ylim(0, 1)\n",
    "g2.lines[0].set_linestyle('--')\n",
    "g2.lines[1].set_linestyle('--')\n",
    "g2.lines[2].set_linestyle('--')\n",
    "#g2.lines[0].set_c('orange')\n",
    "g2.set_ylabel(\"Cumulative Runtime (ms)\")\n",
    "g2.legend_.remove()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 420,
   "metadata": {},
   "outputs": [],
   "source": [
    "g.get_figure().savefig('/data/jwfromm/quant_layer_rel.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Shiftnorm Fusion"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 312,
   "metadata": {},
   "outputs": [],
   "source": [
    "# In this figure we compare the runtime of squeezenet when shiftnorm is fused into the preceding convolution (giving memory benefits)\n",
    "# and wehen its left unfused. (unipolar)\n",
    "\n",
    "onebit_fused = 67.6\n",
    "twobit_fused = 111.9\n",
    "threebit_fused = 155.1\n",
    "\n",
    "onebit_unfused = 86.04\n",
    "twobit_unfused = 122.6\n",
    "threebit_unfused = 158.8\n",
    "\n",
    "fused = [onebit_fused, twobit_fused, threebit_fused]\n",
    "unfused = [onebit_unfused, twobit_unfused, threebit_unfused]\n",
    "names = [\"1 Bit\", \"2 Bit\", \"3 Bit\"]\n",
    "\n",
    "fused_df = pd.DataFrame({\"Runtime (ms)\": fused, \"Quantization\": names, \"Type\": \"Fused\"})\n",
    "unfused_df = pd.DataFrame({\"Runtime (ms)\": unfused, \"Quantization\": names, \"Type\": \"Unfused\"})\n",
    "\n",
    "merged_df = pd.concat([fused_df, unfused_df])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 313,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f16a6636ac8>"
      ]
     },
     "execution_count": 313,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "g = sns.barplot(x=\"Quantization\", y=\"Runtime (ms)\", hue=\"Type\", data=merged_df)\n",
    "g.legend(title=\"\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 314,
   "metadata": {},
   "outputs": [],
   "source": [
    "g.get_figure().savefig('/data/jwfromm/shiftnorm_fusion.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# One-Off Ablation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 409,
   "metadata": {},
   "outputs": [],
   "source": [
    "# In this plot we look at the relative speedup of our system minus one optimization at a time.\n",
    "baseline = 719.92\n",
    "\n",
    "# 1-bit binary glue\n",
    "opt = 67.65\n",
    "pack_fusion = 86.04\n",
    "tiling = 101.21\n",
    "vectorize = 95.21\n",
    "binary_glue = 122.94\n",
    "parallel = 182.14\n",
    "SIMD = 252.6\n",
    "\n",
    "binglue_vals = [opt, pack_fusion, binary_glue, tiling, vectorize, parallel, SIMD]\n",
    "binglue_rel = baseline / np.asarray(binglue_vals)\n",
    "\n",
    "# 1-bit fp glue\n",
    "opt = 122.94\n",
    "pack_fusion = 158.6\n",
    "parallel = 201.46\n",
    "vectorize = 129.47\n",
    "SIMD = 437.43\n",
    "\n",
    "fpglue_vals = [opt, pack_fusion, parallel, vectorize, SIMD]\n",
    "fpglue_rel = baseline / np.asarray(fpglue_vals)\n",
    "\n",
    "names = ['Optimized', \"-Bitpack Fusion\", \"-Fused Glue\", \"-Tiling\", \"-Vectorize\", \"-Parallelize\", \"-Fast Popcount\"]\n",
    "\n",
    "binglue_df = pd.DataFrame({'One-Off Optimization': names, 'Relative Speedup': binglue_rel, 'Type': 'Binary Glue'})\n",
    "#fpglue_df = pd.DataFrame({'One-Off Optimization': names, 'Relative Speedup': fpglue_rel, 'Type': \"FP32 Glue\"})\n",
    "\n",
    "#oneoff_df = pd.concat([binglue_df, fpglue_df])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 431,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "g = sns.barplot(x='One-Off Optimization', y='Relative Speedup', hue='Type', data=binglue_df)\n",
    "g.axhline(y=1.0, color=sns.xkcd_rgb[\"faded red\"], linestyle='--', label='Baseline')\n",
    "g.legend_.remove()\n",
    "g.set_ylim(0, 12)\n",
    "for item in g.get_xticklabels():\n",
    "    item.set_rotation(25)\n",
    "    \n",
    "for p in g.patches:\n",
    "    g.annotate(format(p.get_height(), '.2f'), (p.get_x() + p.get_width() / 2., p.get_height()), ha = 'center', va = 'center', xytext = (0, 10), textcoords = 'offset points')    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 433,
   "metadata": {},
   "outputs": [],
   "source": [
    "g.get_figure().savefig('/data/jwfromm/opt_ablation.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# End2End Speeds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 318,
   "metadata": {},
   "outputs": [],
   "source": [
    "# In this figure we're just gonna slam down all the end2end speedups at every bitwidth for every model.\n",
    "\n",
    "bitwidths = ['1', '2', '3']\n",
    "\n",
    "sq_baseline = 719.92\n",
    "sq_1_uni = 67.65\n",
    "sq_1_bi = 56.79\n",
    "sq_2_uni = 111.89\n",
    "sq_2_bi = 87.1\n",
    "sq_3_uni = 155.05\n",
    "sq_3_bi = 111\n",
    "\n",
    "sq_uni = sq_baseline / np.asarray([sq_1_uni, sq_2_uni, sq_3_uni])\n",
    "sq_bi = sq_baseline / np.asarray([sq_1_bi, sq_2_bi, sq_3_bi])\n",
    "\n",
    "sq_uni_df = pd.DataFrame({'Relative Speedup': sq_uni, 'Bitwidth': bitwidths, 'Polarity': 'Unipolar', 'Model': 'Squeezenet'})\n",
    "sq_bi_df = pd.DataFrame({'Relative Speedup': sq_bi, 'Bitwidth': bitwidths, 'Polarity': 'Bipolar', 'Model': 'Squeezenet'})\n",
    "sq_df = pd.concat([sq_uni_df, sq_bi_df])\n",
    "\n",
    "vgg_baseline = 2420\n",
    "vgg_1_uni = 243.8\n",
    "vgg_1_bi = 184.1\n",
    "vgg_2_uni = 387.2\n",
    "vgg_2_bi = 271.4\n",
    "vgg_3_uni = 610.0\n",
    "vgg_3_bi = 423.5\n",
    "\n",
    "vgg_uni = vgg_baseline / np.asarray([vgg_1_uni, vgg_2_uni, vgg_3_uni])\n",
    "vgg_bi = vgg_baseline / np.asarray([vgg_1_bi, vgg_2_bi, vgg_3_bi])\n",
    "\n",
    "vgg_uni_df = pd.DataFrame({'Relative Speedup': vgg_uni, 'Bitwidth': bitwidths, 'Polarity': 'Unipolar', 'Model': 'Vggnet'})\n",
    "vgg_bi_df = pd.DataFrame({'Relative Speedup': vgg_bi, 'Bitwidth': bitwidths, 'Polarity': 'Bipolar', 'Model': 'Vggnet'})\n",
    "vgg_df = pd.concat([vgg_uni_df, vgg_bi_df])\n",
    "\n",
    "alexnet_baseline = 1260\n",
    "alexnet_1_uni = 150.37\n",
    "alexnet_1_bi = 122.68\n",
    "alexnet_2_uni = 196.80\n",
    "alexnet_2_bi = 154.6\n",
    "alexnet_3_uni = 282.84\n",
    "alexnet_3_bi = 207.02\n",
    "\n",
    "alexnet_uni = alexnet_baseline / np.asarray([alexnet_1_uni, alexnet_2_uni, alexnet_3_uni])\n",
    "alexnet_bi = alexnet_baseline / np.asarray([alexnet_1_bi, alexnet_2_bi, alexnet_3_bi])\n",
    "\n",
    "alexnet_uni_df = pd.DataFrame({'Relative Speedup': alexnet_uni, 'Bitwidth': bitwidths, 'Polarity': 'Unipolar', 'Model': 'Alexnet'})\n",
    "alexnet_bi_df = pd.DataFrame({'Relative Speedup': alexnet_bi, 'Bitwidth': bitwidths, 'Polarity': 'Bipolar', 'Model': 'Alexnet'})\n",
    "alexnet_df = pd.concat([alexnet_uni_df, alexnet_bi_df])\n",
    "\n",
    "merged_df = pd.concat([sq_df, vgg_df, alexnet_df])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 319,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.lines.Line2D at 0x7f16a6590518>"
      ]
     },
     "execution_count": 319,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 771.3x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "g = sns.factorplot(x='Model', y='Relative Speedup', hue='Bitwidth', data=merged_df, col='Polarity', kind=\"bar\")\n",
    "g1, g2 = g.axes[0]\n",
    "g1.axhline(y=1.0, color=sns.xkcd_rgb[\"faded red\"], linestyle='--', label='Baseline')\n",
    "g2.axhline(y=1.0, color=sns.xkcd_rgb[\"faded red\"], linestyle='--', label='Baseline')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 320,
   "metadata": {},
   "outputs": [],
   "source": [
    "g.savefig('/data/jwfromm/all_speedups.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Estimated Glue Operations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 353,
   "metadata": {},
   "outputs": [],
   "source": [
    "# In this figure we consider the hypothetical impact of glue layers at varying assumed binary speedups. Y axis is estimated amount of runtime spent in glue operations\n",
    "\n",
    "Speedups = ['43X', '20X', '10X', '5X', '2X', '1X']\n",
    "conv_ops = 786057216\n",
    "glue_ops = 51806720\n",
    "\n",
    "glue_fraction = 100 * np.asarray([glue_ops / (glue_ops + conv_ops / 43), glue_ops / (glue_ops + conv_ops / 20), glue_ops / (glue_ops + conv_ops / 10), glue_ops / (glue_ops + conv_ops / 5), glue_ops / (glue_ops + conv_ops / 2), glue_ops / (glue_ops + conv_ops / 1)])\n",
    "\n",
    "glue_df = pd.DataFrame({'Assumed Conv Layer Speedup from Binarization': Speedups, 'Glue % of Total Runtime': glue_fraction, 'Type': 1})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 354,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "g = sns.barplot(x='Assumed Conv Layer Speedup from Binarization', y='Glue % of Total Runtime', data=glue_df, hue='Type')\n",
    "g.legend_.remove()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 356,
   "metadata": {},
   "outputs": [],
   "source": [
    "g.get_figure().savefig('/data/jwfromm/glue_estimate.pdf', dpi=900, bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}