stencil_plot.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Mon Jan 13 14:19:59 2020

@author: shahrzad
"""

import pandas
import numpy as np
from matplotlib import pyplot as plt
from matplotlib.backends.backend_pdf import PdfPages
import itertools
import glob
import matplotlib.tri as mtri
from mpl_toolkits import mplot3d
import math
from scipy.optimize import curve_fit
from collections import Counter
from scipy.optimize import nnls
import csv
from scipy.optimize import nnls


def create_dict_stencil(directories,to_csv=False):
    thr=[]
    data_files=[]
    
    for directory in directories:
        [data_files.append(i) for i in glob.glob(directory+'/*.dat')]

    nps=[]
    nxs=[]
    nodes=[]
    gps=[]
    if to_csv:
        f_csv=open('/home/shahrzad/repos/Blazemark/data/stencil_data_perf_all.csv','w')
        f_writer=csv.writer(f_csv, delimiter=',', quotechar='"', quoting=csv.QUOTE_MINIMAL)
        f_writer.writerow(['node','grid_points','num_threads','num_partitions','partition_size','grain_size','work_per_core','num_tasks','execution_time'])

    for filename in data_files:            
        (node, _, th, npo, nx) = filename.replace('_old','').split('/')[-1].replace('.dat','').split('_')                 
        th=int(th)
        npo=int(npo)
        nx=int(nx)
       
        if node not in nodes:
            nodes.append(node)
        if npo not in nps:
            nps.append(npo)              
        if nx not in nxs:
            nxs.append(nx)
        if th not in thr:
            thr.append(th)
        if nx*npo not in gps:
            gps.append(nx*npo)
            
    nodes.sort()
    nxs.sort()
    nps.sort()
    gps.sort()
    thr.sort()              
                                                           
    data_files.sort()   
    
    for filename in data_files:   
        f=open(filename, 'r')
                 
        result=f.readlines()
        if len(result)!=0:
            print(filename)
            (node, _, th, npo, nx) = filename.replace('_old','').split('/')[-1].replace('.dat','').split('_')                 
            th=float(th)
            npo=float(npo)
            nx=float(nx)
           
            r=result[1].split(',')
            r=[rr.strip() for rr in r]
            execution_time=float(r[1])*1e6
            num_tasks=npo
            grain_size=nx        
            w_c=math.ceil(num_tasks/th)*grain_size
            f_writer.writerow([node,npo*nx,th,npo,nx,grain_size,w_c,num_tasks,execution_time])

    if to_csv:
        f_csv.close()
#    return (data, d, thr, iter_lengths, num_iterations)  

marvin_dir='/home/shahrzad/repos/Blazemark/data/stencil/marvin_old/100000000'
create_dict_stencil([marvin_dir],1)


titles=['node','grid_points','num_threads','num_partitions','partition_size','grain_size','work_per_core','num_tasks','execution_time']
filename='/home/shahrzad/repos/Blazemark/data/stencil_data_perf_all.csv'


dataframe = pandas.read_csv(filename, header=0,index_col=False,dtype=str,names=titles)
for col in titles[1:]:
    dataframe[col] = dataframe[col].astype(float)

nodes=dataframe['node'].drop_duplicates().values
nodes.sort()

grid_points=dataframe['grid_points'].drop_duplicates().values
grid_points.sort()

            
for node in nodes:
    node_selected=dataframe['node']==node
    df_n_selected=dataframe[node_selected][titles[1:]]
    
    thr=df_n_selected['num_threads'].drop_duplicates().values
    thr.sort()

    grid_points=dataframe['grid_points'].drop_duplicates().values
    grid_points.sort()

    array=df_n_selected[['num_threads','grain_size','execution_time']].values
    array=array.astype(float)
    j=1
    for th in range(1,9): 
        plt.figure(j)
        array_t=array[array[:,0]==th]
        plt.scatter(array_t[:,1],array_t[:,2],marker='.',label=str(th))
        plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)
        plt.xscale('log')
        j=j+1