def main():
import sys
from utils import load_csv, get_stencil_num
raw_data = load_csv(sys.argv[1])
k_l = set()
for k in raw_data:
k_l.add((get_stencil_num(k), k['Global NX']))
k_l = list(k_l)
bsz_l = set()
for k in raw_data:
if k['Multi-wavefront updates'] == '0': continue
bsz_l.add(k['Multi-wavefront updates'])
bsz_l = sorted(list(bsz_l))
for k, N in k_l:
for bsz in bsz_l:
gen_res(raw_data, int(k), int(bsz), int(N))
def main():
import sys
from utils import load_csv, get_stencil_num
raw_data = load_csv(sys.argv[1])
k_l = set()
for k in raw_data:
k_l.add((get_stencil_num(k), k['Global NX']))
k_l = list(k_l)
bsz_l = set()
for k in raw_data:
if k['Multi-wavefront updates']=='0': continue
bsz_l.add(k['Multi-wavefront updates'])
bsz_l = sorted(list(bsz_l))
for k, N in k_l:
for bsz in bsz_l:
gen_res(raw_data, int(k), int(bsz), int(N))
def gen_res(raw_data, stencil_kernel, bsz, N):
from operator import itemgetter
import matplotlib.pyplot as plt
import pylab
from csv import DictWriter
from operator import itemgetter
from utils import get_stencil_num
#fig_width = 8.588*0.393701 # inches
fig_width = 4.0 * 0.393701 # inches
fig_height = 0.60 * fig_width #* 210.0/280.0#433.62/578.16
fig_size = [fig_width, fig_height]
params = {
'axes.labelsize': 6,
'axes.linewidth': 0.5,
'lines.linewidth': 1,
'font.size': 5,
'legend.fontsize': 6,
'xtick.labelsize': 6,
'ytick.labelsize': 6,
'lines.markersize': 5,
'text.usetex': True,
'figure.figsize': fig_size
}
pylab.rcParams.update(params)
req_fields = [('Total cache block size (kiB)', int),
('MStencil/s MAX', float), ('Time stepper orig name', str),
('Stencil Kernel semi-bandwidth', int),
('Stencil Kernel coefficients', str), ('Precision', str),
('Time unroll', int), ('Number of time steps', int),
('Number of tests', int), ('Local NX', int),
('Local NY', int), ('Local NZ', int),
('Total Memory Transfer', float), ('Thread group size', int),
('Intra-diamond prologue/epilogue MStencils', int),
('Multi-wavefront updates', int),
('Intra-diamond width', int)]
data = []
for k in raw_data:
# Use single field to represent the performance
if 'Total RANK0 MStencil/s MAX' in k.keys():
if (k['Total RANK0 MStencil/s MAX'] != ''):
k['MStencil/s MAX'] = k['MWD main-loop RANK0 MStencil/s MAX']
# temporary for deprecated format
if 'RANK0 MStencil/s MAX' in k.keys():
if k['RANK0 MStencil/s MAX'] != '':
k['MStencil/s MAX'] = k['RANK0 MStencil/s MAX']
tup = dict()
# add the general fileds
for f in req_fields:
try:
tup[f[0]] = map(f[1], [k[f[0]]])[0]
except:
print f[0]
# add the stencil operator
tup['Kernel'] = get_stencil_num(k)
data.append(tup)
# add precision information
tup['word size'] = 8 if k['Precision'] in 'DP' else 4
#for i in data: print i
for tup in data:
tup['Actual Bytes/LUP'] = actual_BpU(tup)
tup['Model'] = models(tup)
# model error
tup['Err %'] = 100 * (
tup['Model'] - tup['Actual Bytes/LUP']) / tup['Actual Bytes/LUP']
tup['D_width'] = tup['Intra-diamond width']
tup['bsz'] = tup['Multi-wavefront updates']
tup['Performance'] = tup['MStencil/s MAX']
# tup['Cache block'] = get_bs(Dw=tup['D_width'], Nd=get_nd(tup['Kernel']), Nf=tup['Multi-wavefront updates'], Nx=tup['Local NX'], WS=tup['word size'], R=tup['Stencil Kernel semi-bandwidth'])
tup['Cache block'] = tup['Total cache block size (kiB)'] / 1024.0
# print "cache block C:", tup['Total cache block size (kiB)']/1024.0, " Python:", tup['Cache block']
# try: print "%6.3f %6.3f %6.3f" % (tup['Cache block'], tup['Total cache block size (kiB)']/1024.0,tup['Cache block']- tup['Total cache block size (kiB)']/1024.0)
# except: pass
data = sorted(data, key=itemgetter('Kernel', 'Local NX', 'D_width'))
#for i in data: print i['Kernel'], i['Local NX'], i['D_width'], i['bsz']
fig, ax = plt.subplots()
cs = []
cb = []
cb_meas = []
Dw = []
for k in data:
if k['Kernel'] == stencil_kernel and k['Local NX'] == N and (
k['bsz'] == bsz
or k['Time stepper orig name'] == 'Spatial Blocking'):
cs.append(k['Cache block'])
cb.append(k['Model'])
cb_meas.append(k['Actual Bytes/LUP'])
Dw.append(k['D_width'])
#for i in range(len(cs)):
# print Dw[i], cs[i], cb_meas[i], cb[i]
if Dw == []: return
ax.plot(cs, cb, marker='+', linestyle='-', color='k', label="Model")
ax.plot(cs,
cb_meas,
marker='x',
linestyle='--',
color='b',
label="Measured")
# show the usable cache size limits
# ax.plot([22.5, 22.5], [0, 0.7*cb[0]], linestyle='-', color='r', label="Usable cache size")
if (stencil_kernel == 1):
ax.set_ylabel('Code balance (Bytes/LUP)')
ax.set_xlabel('Block size (MiB) PER THREAD')
ax.set_ylim([0, max(cb_meas + cb) + 1])
ax.set_xlim([0, max(cs) + 0.5])
ax2 = ax.twiny()
ax2.set_xticks(cs)
ax2.set_xlabel('Diamond width')
ax2.set_xlim(ax.get_xlim())
if stencil_kernel == 1:
Dw = map(str, Dw)
Dw[0] = ''
Dw[1] = ''
Dw[3] = ''
Dw[4] = ''
Dw[5] = ''
if stencil_kernel == 5:
Dw = map(str, Dw)
Dw[1] = ''
ax2.set_xticklabels(Dw)
# for i, d in enumerate(Dw):
#if ((d+4)%8 == 0):
# ax.annotate(d, (cs[i], cb[i]))
title = '_code_balance_vs_cache_size_N%d_bsz%d' % (N, bsz)
if stencil_kernel == 0:
title = '25_pt_const' + title
elif stencil_kernel == 1:
title = '7_pt_const' + title
elif stencil_kernel == 4:
title = '25_pt_var' + title
elif stencil_kernel == 5:
title = '7_pt_var' + title
elif stencil_kernel == 6:
title = 'solar' + title
if (bsz == 1):
if (stencil_kernel == 1):
ax.legend(loc='best')
ax.grid()
pylab.savefig(title + '.pdf',
format='pdf',
bbox_inches="tight",
pad_inches=0)
plt.clf()
def gen_res(raw_data, stencil_kernel, bsz, N):
from operator import itemgetter
import matplotlib.pyplot as plt
import pylab
from csv import DictWriter
from operator import itemgetter
from utils import get_stencil_num
#fig_width = 8.588*0.393701 # inches
fig_width = 4.0*0.393701 # inches
fig_height = 0.60*fig_width #* 210.0/280.0#433.62/578.16
fig_size = [fig_width,fig_height]
params = {
'axes.labelsize': 6,
'axes.linewidth': 0.5,
'lines.linewidth': 1,
'font.size': 5,
'legend.fontsize': 6,
'xtick.labelsize': 6,
'ytick.labelsize': 6,
'lines.markersize': 5,
'text.usetex': True,
'figure.figsize': fig_size}
pylab.rcParams.update(params)
req_fields = [('Total cache block size (kiB)', int), ('MStencil/s MAX', float), ('Time stepper orig name', str), ('Stencil Kernel semi-bandwidth', int), ('Stencil Kernel coefficients', str), ('Precision', str), ('Time unroll',int), ('Number of time steps',int), ('Number of tests',int), ('Local NX',int), ('Local NY',int), ('Local NZ',int), ('Total Memory Transfer', float), ('Thread group size' ,int), ('Intra-diamond prologue/epilogue MStencils',int), ('Multi-wavefront updates', int), ('Intra-diamond width', int)]
data = []
for k in raw_data:
# Use single field to represent the performance
if 'Total RANK0 MStencil/s MAX' in k.keys():
if(k['Total RANK0 MStencil/s MAX']!=''):
k['MStencil/s MAX'] = k['MWD main-loop RANK0 MStencil/s MAX']
# temporary for deprecated format
if 'RANK0 MStencil/s MAX' in k.keys():
if k['RANK0 MStencil/s MAX']!='':
k['MStencil/s MAX'] = k['RANK0 MStencil/s MAX']
tup = dict()
# add the general fileds
for f in req_fields:
try:
tup[f[0]] = map(f[1], [k[f[0]]] )[0]
except:
print f[0]
# add the stencil operator
tup['Kernel'] = get_stencil_num(k)
data.append(tup)
# add precision information
tup['word size'] = 8 if k['Precision'] in 'DP' else 4
#for i in data: print i
for tup in data:
tup['Actual Bytes/LUP'] = actual_BpU(tup)
tup['Model'] = models(tup)
# model error
tup['Err %'] = 100 * (tup['Model'] - tup['Actual Bytes/LUP'])/tup['Actual Bytes/LUP']
tup['D_width'] = tup['Intra-diamond width']
tup['bsz'] = tup['Multi-wavefront updates']
tup['Performance'] = tup['MStencil/s MAX']
# tup['Cache block'] = get_bs(Dw=tup['D_width'], Nd=get_nd(tup['Kernel']), Nf=tup['Multi-wavefront updates'], Nx=tup['Local NX'], WS=tup['word size'], R=tup['Stencil Kernel semi-bandwidth'])
tup['Cache block'] = tup['Total cache block size (kiB)']/1024.0
# print "cache block C:", tup['Total cache block size (kiB)']/1024.0, " Python:", tup['Cache block']
# try: print "%6.3f %6.3f %6.3f" % (tup['Cache block'], tup['Total cache block size (kiB)']/1024.0,tup['Cache block']- tup['Total cache block size (kiB)']/1024.0)
# except: pass
data = sorted(data, key=itemgetter('Kernel', 'Local NX', 'D_width'))
#for i in data: print i['Kernel'], i['Local NX'], i['D_width'], i['bsz']
fig, ax = plt.subplots()
cs=[]
cb=[]
cb_meas=[]
Dw=[]
for k in data:
if k['Kernel']==stencil_kernel and k['Local NX']==N and (k['bsz']==bsz or k['Time stepper orig name'] == 'Spatial Blocking'):
cs.append(k['Cache block'])
cb.append(k['Model'])
cb_meas.append(k['Actual Bytes/LUP'])
Dw.append(k['D_width'])
#for i in range(len(cs)):
# print Dw[i], cs[i], cb_meas[i], cb[i]
if Dw==[]: return
ax.plot(cs, cb , marker='+', linestyle='-', color='k', label="Model")
ax.plot(cs, cb_meas, marker='x', linestyle='--', color='b', label="Measured")
# show the usable cache size limits
# ax.plot([22.5, 22.5], [0, 0.7*cb[0]], linestyle='-', color='r', label="Usable cache size")
if(stencil_kernel == 1):
ax.set_ylabel('Code balance (Bytes/LUP)')
ax.set_xlabel('Block size (MiB) PER THREAD')
ax.set_ylim([0, max(cb_meas+cb)+1])
ax.set_xlim([0, max(cs)+0.5])
ax2 = ax.twiny()
ax2.set_xticks(cs)
ax2.set_xlabel('Diamond width')
ax2.set_xlim(ax.get_xlim())
if stencil_kernel==1:
Dw = map(str,Dw)
Dw[0]=''
Dw[1]=''
Dw[3]=''
Dw[4]=''
Dw[5]=''
if stencil_kernel==5:
Dw = map(str,Dw)
Dw[1]=''
ax2.set_xticklabels(Dw)
# for i, d in enumerate(Dw):
#if ((d+4)%8 == 0):
# ax.annotate(d, (cs[i], cb[i]))
title = '_code_balance_vs_cache_size_N%d_bsz%d'%(N, bsz)
if stencil_kernel == 0:
title = '25_pt_const' + title
elif stencil_kernel == 1:
title = '7_pt_const' + title
elif stencil_kernel == 4:
title = '25_pt_var' + title
elif stencil_kernel == 5:
title = '7_pt_var' + title
elif stencil_kernel == 6:
title = 'solar' + title
if(bsz==1):
if(stencil_kernel == 1):
ax.legend(loc='best')
ax.grid()
pylab.savefig(title+'.pdf', format='pdf', bbox_inches="tight", pad_inches=0)
plt.clf()
