AymenFJA

import shutil
import argparse
import os, os.path
from mpi4py import MPI
from queue import Queue
import iwp_divideimg as divide

parser = argparse.ArgumentParser()
parser.add_argument('imgs_path', help='Path of the dataset')
parser.add_argument('worker_root', help='main work dir')

AymenFJA

#include <stdio.h>
#include <stdlib.h>
#include <math.h>
 
// CUDA kernel. Each thread takes care of one element of c
__global__ void vecAdd(double *a, double *b, double *c, int n)
{
    // Get our global thread ID
    int id = blockIdx.x*blockDim.x+threadIdx.x;
 

AymenFJA

from myproxy.client import MyProxyClient
myproxy_clnt = MyProxyClient(hostname="myproxy.somewhere.ac.uk")
cert, private_key = myproxy_clnt.logon(username, password, bootstrap=True)

AymenFJA

import os
import radical.pilot as rp
from radical.entk import Pipeline, Stage, Task, AppManager


crop_size = int(360)              #covnvert this to argument later
worker_root = r"/pylon5/mc3bggp/aymen/local_dir/datasets/polygon/"                                                                 #convert this to argument later
weights_path = r"/pylon5/mc3bggp/aymen/local_dir/datasets/logs/ice_wedge_polygon20180823T1403/mask_rcnn_ice_wedge_polygon_0008.h5" #convert this to argument later
imgs_path = r"/pylon5/mc3bggp/aymen/local_dir/datasets/polygon/input_img/"                                                         #convert this to argument later

AymenFJA

AymenFJA

#!/bin/bash
NGPUS=2 # Number of gpus with compute_capability 3.5 per server
# Start the MPS server for each GPU
for ((i=0; i< $NGPUS; i++))
do
mkdir /home/aymen/mps_$i
mkdir /home/aymen/mps_log_$i
export CUDA_VISIBLE_DEVICES=$i
export CUDA_MPS_PIPE_DIRECTORY=/home/aymen/mps_$i
export CUDA_MPS_LOG_DIRECTORY=/home/aymen/mps_log_$i

AymenFJA

#!/bin/bash
export CUDA_VISIBLE_DEVICES=0
lrank=$OMPI_COMM_WORLD_LOCAL_RANK
case ${lrank} in
[0])
 export CUDA_MPS_PIPE_DIRECTORY=/home/aymen/mps_0; ./vector_add
 ;;
[1])
 export CUDA_MPS_PIPE_DIRECTORY=/home/aymen/mps_1; ./vector_add
 ;;

AymenFJA

#!/bin/bash
# Stop the MPS control daemon for each GPU and clean up /tmp
NGPUS=2 # Number of gpus with compute_capability 3.5 per server
for ((i=0; i< $NGPUS; i++))
do
echo $i
export CUDA_MPS_PIPE_DIRECTORY=/home/aymen/mps_$i
echo "quit" | nvidia-cuda-mps-control
rm -rf /home/aymen/mps_$i
rm -rf /home/aymen/mps_log_$i

AymenFJA

1. Login to the VM.

• From your terminal ssh to the following IP:

  ssh [email protected]

2. Generate XSEDE certificate.

• In order to be able to login to Bridges from your VM you need to generate XSEDE certificate using the following command:

  myproxy-logon -s myproxy.xsede.org -l user_name -t 72

AymenFJA

from __future__ import print_function
'''
Basic Multi GPU computation example using TensorFlow library.
Single/Multi-GPU non-MPI Author: Aymeric Damien
Multi-GPU Large scale/Multi-node MPI Author: Aymen Alsaadi
Project: https://github.com/aymericdamien/TensorFlow-Examples/
'''

'''
This tutorial requires your machine to have 2 GPUs