Xeus-Cling with OpenMP

xeus_cling_openmp.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# OpenMP in Xeus-Cling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preparation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Set the following in your `kernel.json`:\n",
    "\n",
    "```json\n",
    "{\n",
    "  \"display_name\": \"C++14 OpenMP\",\n",
    "  \"argv\": [\n",
    "      \"/home/axel/miniconda3/envs/cling/bin/xeus-cling\",\n",
    "      \"-f\",\n",
    "      \"{connection_file}\",\n",
    "      \"-std=c++14\", \"-fopenmp\"\n",
    "  ],\n",
    "  \"language\": \"C++14\"\n",
    "}\n",
    "```\n",
    "\n",
    "(replace `/home/axel/` without your output of `echo $HOME/`)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/axel/\n"
     ]
    }
   ],
   "source": [
    "!echo $HOME/"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Assuming you installed xeus-cling with conda in a conda environment named \"cling\", then `<KERNEL_PREFIX>` is most likely in `$HOME/miniconda3/envs/cling/share/jupyter/kernels`. Just modify or copy the existing `xeus-cling-cpp14/` directory."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Some Environment Diagnostics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/axel/miniconda3/envs/cling/bin/cling\n"
     ]
    }
   ],
   "source": [
    "!which cling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5\n"
     ]
    }
   ],
   "source": [
    "!cling --version"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "clang version 5.0.0 (tags/RELEASE_500/final)\n",
      "Target: x86_64-unknown-linux-gnu\n",
      "Thread model: posix\n",
      "InstalledDir: /home/axel/miniconda3/envs/cling/bin\n"
     ]
    }
   ],
   "source": [
    "!clang --version"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## OpenMP Hello World"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "#include <omp.h>\n",
    "#include <iostream>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "#pragma cling load(\"libomp\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "omp_get_max_threads()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "int nthreads, tid;\n",
    "\n",
    "// trying to set the number of threads\n",
    "omp_set_dynamic(0);              // Explicitly disable dynamic teams\n",
    "int const number_of_threads = 4;\n",
    "omp_set_num_threads(number_of_threads);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Hello World! I am thread = 0 out of 1\n",
      "And I am thread 0, ruling the world.\n",
      "And I 0\n",
      "And I 0\n",
      "And I 0\n",
      "And I 0\n"
     ]
    }
   ],
   "source": [
    "// each thread in here has their own copy of variables\n",
    "#pragma omp parallel private(nthreads, tid) num_threads(number_of_threads)\n",
    "{\n",
    "    // all threads\n",
    "    tid = omp_get_thread_num();\n",
    "    std::cout << \"Hello World! I am thread = \" << tid\n",
    "              << \" out of \" << omp_get_num_threads()\n",
    "              << std::endl;\n",
    "\n",
    "    // thread zero (\"master\") only\n",
    "    if (tid == 0) \n",
    "        std::cout << \"And I am thread 0, ruling the world.\"\n",
    "                  << std::endl;\n",
    "    \n",
    "    #pragma omp parallel for\n",
    "    for( int i = 0; i < 4; ++i)\n",
    "        std::cout << \"And I \" << tid << std::endl;\n",
    "        \n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "C++14 OpenMP",
   "language": "C++14",
   "name": "xeus-cling-openmp-cpp14"
  },
  "language_info": {
   "codemirror_mode": "text/x-c++src",
   "file_extension": ".cpp",
   "mimetype": "text/x-c++src",
   "name": "c++",
   "version": "14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}

related: https://root-forum.cern.ch/t/cling-and-openmp/26841/3

Forum: https://root-forum.cern.ch/search?q=openmp

I found the problem. The Argument '-fopenmp' will not passed to the cling instance. If you run a blank cling in the bash without the '-fopenmp' you get the same behavoir.

The following bunch of code, you can check the arguments of the cling instance in a jupyter notebook: https://gist.github.com/SimeonEhrig/05bdb0927d5f799608a18034ee69449a

@SimeonEhrig thank you, that explains it!

(I think you have to write @ax3l to ping me in a gist comment, did not receive the notification we talked about)