Diffusion cg-project (master)

Edit
cg-project
ActivePublic

cg-project (master)

.gitignore
README.md
cg.c
cg.h
cg_cuda.cu
cg_mpi.c
cuda.h
main.c
makefile
matrices/
mmio.c
mmio.h
mmio_wrapper.h
mpi.c
mpi.h
parameters.h
util.c
util.h
vec.c
vec.h
vec_cuda.cu

Recent Commits

Commit	Author	Details	Committed
54c6e2d33a40	briling	The good mxdotvec for cuda	Jul 3 2020
216384a86bb8	briling	Readme	Jul 3 2020
a7deefc2fd74	briling	vecsadd	Jul 1 2020
274674b3ef59	briling	better naming	Jul 1 2020
7c7fb995d139	briling	kernels to file	Jul 1 2020
587e903d253c	briling	More memory-efficient	Jul 1 2020
73c6b6ff11b7	briling	call smart	Jul 1 2020
b7efb360bc88	briling	cuda 6	Jul 1 2020
77fd4bb4f2b9	briling	cuda 5	Jul 1 2020
d10f18131f5e	briling	cuda 4	Jul 1 2020
c4eb46afba07	briling	cuda 3	Jul 1 2020
91155394e15e	briling	cuda 2	Jul 1 2020
5a6a89d2f68f	briling	cuda	Jul 1 2020
236e37777de8	Ksenia	MPI sparse	Jun 15 2020
1a54a25a068d	Ksenia	start MPI sparse	Jun 15 2020

README.md

CONJUGATE GRADIENT PROJECT

Requirements

for cg: mpicc or mpiicc
for cg_gpu: CUDA® compiler

Build

Compile CPU version on fidis:

module load intel intel-mpi
make cg

Compile GPU version on deneb{1,2}:

slmodules -s x86_E5v2_Mellanox_GPU
module load gcc cuda
make cg_gpu

Run

As an example, we show a sbatch script to run the CPU version:

#!/bin/bash -l
#SBATCH --reservation cours-phpc2020
#SBATCH --account phpc2020
#SBATCH --job-name=p16
#SBATCH --ntasks=16
#SBATCH --mem-per-cpu=1GB
#SBATCH --time=80:00
module purge
module load intel intel-mpi
srun ./cg <matrix file> [mode]

mode can be 0 (dense solver) or 1 (sparse solver). Default value is 0.

The matrix format is Matrix Market format (.mtx extension).