{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div align=\"center\"><img src=\"./images/DLI_Header.png\"></div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Monte Carlo Approximation of $\\pi$ - NVSHMEM with Distributed Work"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this notebook you will use NVSHMEM in the monte-carlo approximation of $\\pi$ program, but this time distributing the work across GPUs."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Objectives"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "By the time you complete this notebook you will:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Be able to distribute work across multiple GPUs using NVSHMEM."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Exercise: Distribute Work Across PEs"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this exercise you will have each GPU divide the $N$   sample points by the number $M$   of PEs. We can use the API `nvshmem_n_pes()` to obtain this:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "```cpp\n",
    "int n_pes = nvshmem_n_pes();\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then it's simple to divide up $N$ by `n_pes`. As an additional step to make this more interesting, also have each GPU do *different* work by choosing the random seed to be unique for each PE:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "```cpp\n",
    "int seed = nvshmem_my_pe();\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Please work in [exercises/nvshmem_pi_step2.cpp](exercises/nvshmem_pi_step2.cpp). As before, deal with any `FIXME` locations, then come back and compile and run. What you should observe is that each PE gets a slightly different answer, and the answer should be less accurate because we're using fewer sample points.\n",
    "\n",
    "Check the [solution](solutions/nvshmem_pi_step2.cpp) if you're stuck."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!nvcc -x cu -arch=sm_70 -rdc=true -I $NVSHMEM_HOME/include -L $NVSHMEM_HOME/lib -lnvshmem -lcuda -o nvshmem_pi_step2 exercises/nvshmem_pi_step2.cpp\n",
    "!nvshmrun -np $NUM_DEVICES ./nvshmem_pi_step2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Next"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the next notebook you will learn about one of the key features on NVSHMEM, *symmetric memory*, which allows for fine-grained inter-GPU communications from device-side code.\n",
    "\n",
    "Please open the next notebook: [_The NVSHMEM Memory Model_](09_MCπ-NVSHMEM-Sym.ipynb)."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}