Skip to content

Latest commit

 

History

History
63 lines (49 loc) · 2.73 KB

README.md

File metadata and controls

63 lines (49 loc) · 2.73 KB

Copyright (c) 2022, Advanced Micro Devices, Inc. All rights reserved.

This training example is released under the MIT license as listed in the top-level directory. If this example is separated from the main directory, include the LICENSE file with it.

Contributions from Suyash Tandon, Noel Chalmers, Nick Curtis, Justin Chang, and Gina Sitaraman.

Description document for the GPU-based Jacobi solver

Contents:

  1. Application overview
  2. Prerequisites
  3. Build instructions
  4. Run instructions

Application overview

This is a distributed Jacobi solver, using GPUs to perform the computation and MPI for halo exchanges. It uses a 2D domain decomposition scheme to allow for a better computation-to-communication ratio than just 1D domain decomposition.

The flow of the application is as follows:

  • The MPI environment is initialized
  • The command-line arguments are parsed, and a MPI grid and mesh are created
  • Resources (including host and device memory blocks, streams etc.) are initialized
  • The Jacobi loop is executed; in every iteration, the local block is updated and then the halo values are exchanged; the algorithm converges when the global residue for an iteration falls below a threshold, but it is also limited by a maximum number of iterations (irrespective if convergence has been achieved or not)
  • Run measurements are displayed and resources are disposed

The application uses the following command-line arguments:

  • -g x y - mandatory argument for the process topology, x denotes the number of processes on the X direction (i.e. per row) and y denotes the number of processes on the Y direction (i.e. per column); the topology size must always match the number of available processes (i.e. the number of launched MPI processes must be equal to x * y)
  • -m dx dy - optional argument indicating the size of the local (per-process) domain size; if it is omitted, the size will default to DEFAULT_DOMAIN_SIZE as defined in defines.h
  • -h | --help - optional argument for printing help information; this overrides all other arguments

Prerequisites

To build and run the jacobi application on A+A hardware, the following dependencies must be installed first:

  • an MPI implementation (openMPI, MPICH, etc.)
  • ROCm 2.1 or later.

Build Instructions

A Makefile is included along with the source files that configures and builds multiple objects and then stitches them together to build the binary for the application Jacobi_hip. To build, simply run:

make

An alternative cmake build system is also include

mkdir build && cd build
cmake ..
make

Run instructions

To run use:

mpirun -np 2 ./Jacobi_hip -g 2 1