Révision 55 Pi/GPU/Pi-GPU.py
| Pi-GPU.py (revision 55) | ||
|---|---|---|
| 1 | 1 |
#!/usr/bin/env python |
| 2 | 2 |
|
| 3 | 3 |
# |
| 4 |
# Pi-by-MC using PyCUDA/PyOpenCL
|
|
| 4 |
# Pi-by-MonteCarlo using PyCUDA/PyOpenCL
|
|
| 5 | 5 |
# |
| 6 | 6 |
# CC BY-NC-SA 2011 : <emmanuel.quemener@ens-lyon.fr> |
| 7 | 7 |
# |
| ... | ... | |
| 19 | 19 |
import sys |
| 20 | 20 |
import getopt |
| 21 | 21 |
import time |
| 22 |
import matplotlib.pyplot as plt |
|
| 23 | 22 |
import math |
| 24 |
from scipy.optimize import curve_fit |
|
| 25 | 23 |
from socket import gethostname |
| 26 | 24 |
|
| 27 | 25 |
# find prime factors of a number |
| ... | ... | |
| 600 | 598 |
|
| 601 | 599 |
def FitAndPrint(N,D,Curves): |
| 602 | 600 |
|
| 601 |
from scipy.optimize import curve_fit |
|
| 602 |
import matplotlib.pyplot as plt |
|
| 603 |
|
|
| 603 | 604 |
try: |
| 604 | 605 |
coeffs_Amdahl, matcov_Amdahl = curve_fit(Amdahl, N, D) |
| 605 | 606 |
|
| ... | ... | |
| 672 | 673 |
if __name__=='__main__': |
| 673 | 674 |
|
| 674 | 675 |
# Set defaults values |
| 675 |
# Alu can be CPU or GPU |
|
| 676 |
|
|
| 677 |
# Alu can be CPU, GPU or ACCELERATOR |
|
| 676 | 678 |
Alu='CPU' |
| 677 | 679 |
# Id of GPU : 1 is for first find ! |
| 678 | 680 |
Device=0 |
| ... | ... | |
| 695 | 697 |
Metrology='InMetro' |
| 696 | 698 |
# Curves is True to print the curves |
| 697 | 699 |
Curves=False |
| 700 |
# Fit is True to print the curves |
|
| 701 |
Fit=False |
|
| 698 | 702 |
# DoublePrecision on FP calculus |
| 699 | 703 |
DoublePrecision=False |
| 700 | 704 |
|
| 701 | 705 |
try: |
| 702 |
opts, args = getopt.getopt(sys.argv[1:],"hocla:g:p:i:s:e:t:r:d:",["alu=","gpustyle=","parastyle=","iterations=","jobstart=","jobend=","jobstep=","redo=","device="]) |
|
| 706 |
opts, args = getopt.getopt(sys.argv[1:],"hoclfa:g:p:i:s:e:t:r:d:",["alu=","gpustyle=","parastyle=","iterations=","jobstart=","jobend=","jobstep=","redo=","device="])
|
|
| 703 | 707 |
except getopt.GetoptError: |
| 704 |
print '%s -o (Out of Core Metrology) -c (Print Curves) -l (Double Precision) -a <CPU/GPU/ACCELERATOR> -d <DeviceId> -g <CUDA/OpenCL> -p <Threads/Hybrid/Blocks> -i <Iterations> -s <JobStart> -e <JobEnd> -t <JobStep> -r <RedoToImproveStats> ' % sys.argv[0] |
|
| 708 |
print '%s -o (Out of Core Metrology) -c (Print Curves) -l (Double Precision) -f (Fit to Amdahl Law) -a <CPU/GPU/ACCELERATOR> -d <DeviceId> -g <CUDA/OpenCL> -p <Threads/Hybrid/Blocks> -i <Iterations> -s <JobStart> -e <JobEnd> -t <JobStep> -r <RedoToImproveStats> ' % sys.argv[0]
|
|
| 705 | 709 |
sys.exit(2) |
| 706 | 710 |
|
| 707 | 711 |
for opt, arg in opts: |
| 708 | 712 |
if opt == '-h': |
| 709 |
print '%s -o (Out of Core Metrology) -c (Print Curves) -l (Double Precision) -a <CPU/GPU/ACCELERATOR> -d <DeviceId> -g <CUDA/OpenCL> -p <Threads/Hybrid/Blocks> -i <Iterations> -s <JobStart> -e <JobEnd> -t <JobStep> -r <RedoToImproveStats>' % sys.argv[0] |
|
| 713 |
print '%s -o (Out of Core Metrology) -c (Print Curves) -l (Double Precision) -f (Fit to Amdahl Law) -a <CPU/GPU/ACCELERATOR> -d <DeviceId> -g <CUDA/OpenCL> -p <Threads/Hybrid/Blocks> -i <Iterations> -s <JobStart> -e <JobEnd> -t <JobStep> -r <RedoToImproveStats>' % sys.argv[0]
|
|
| 710 | 714 |
|
| 711 | 715 |
print "\nInformations about devices detected under OpenCL:" |
| 712 | 716 |
# For PyOpenCL import |
| 713 |
import pyopencl as cl |
|
| 714 |
Id=1 |
|
| 715 |
for platform in cl.get_platforms(): |
|
| 716 |
for device in platform.get_devices(): |
|
| 717 |
deviceType=cl.device_type.to_string(device.type) |
|
| 718 |
print "Device #%i of type %s : %s" % (Id,deviceType,device.name) |
|
| 719 |
Id=Id+1 |
|
| 717 |
try: |
|
| 718 |
import pyopencl as cl |
|
| 719 |
Id=1 |
|
| 720 |
for platform in cl.get_platforms(): |
|
| 721 |
for device in platform.get_devices(): |
|
| 722 |
deviceType=cl.device_type.to_string(device.type) |
|
| 723 |
print "Device #%i of type %s : %s" % (Id,deviceType,device.name) |
|
| 724 |
Id=Id+1 |
|
| 720 | 725 |
|
| 721 |
sys.exit() |
|
| 726 |
|
|
| 727 |
sys.exit() |
|
| 728 |
except ImportError: |
|
| 729 |
print "Your platform does not seem to support OpenCL" |
|
| 730 |
|
|
| 722 | 731 |
elif opt == '-o': |
| 723 | 732 |
OutMetrology=True |
| 724 | 733 |
Metrology='OutMetro' |
| ... | ... | |
| 726 | 735 |
DoublePrecision=True |
| 727 | 736 |
elif opt == '-c': |
| 728 | 737 |
Curves=True |
| 738 |
elif opt == '-f': |
|
| 739 |
Fit=True |
|
| 729 | 740 |
elif opt in ("-a", "--alu"):
|
| 730 | 741 |
Alu = arg |
| 731 | 742 |
elif opt in ("-d", "--device"):
|
| ... | ... | |
| 765 | 776 |
print "Double Precision in Kernels : %r" % DoublePrecision |
| 766 | 777 |
|
| 767 | 778 |
if GpuStyle=='CUDA': |
| 768 |
# For PyCUDA import |
|
| 769 |
import pycuda.driver as cuda |
|
| 770 |
import pycuda.gpuarray as gpuarray |
|
| 771 |
import pycuda.autoinit |
|
| 772 |
from pycuda.compiler import SourceModule |
|
| 779 |
try: |
|
| 780 |
# For PyCUDA import |
|
| 781 |
import pycuda.driver as cuda |
|
| 782 |
import pycuda.gpuarray as gpuarray |
|
| 783 |
import pycuda.autoinit |
|
| 784 |
from pycuda.compiler import SourceModule |
|
| 785 |
except ImportError: |
|
| 786 |
print "Platform does not seem to support CUDA" |
|
| 773 | 787 |
|
| 774 | 788 |
if GpuStyle=='OpenCL': |
| 775 |
# For PyOpenCL import |
|
| 776 |
import pyopencl as cl |
|
| 777 |
Id=1 |
|
| 778 |
for platform in cl.get_platforms(): |
|
| 779 |
for device in platform.get_devices(): |
|
| 780 |
deviceType=cl.device_type.to_string(device.type) |
|
| 781 |
print "Device #%i of type %s : %s" % (Id,deviceType,device.name) |
|
| 782 |
if Id == Device: |
|
| 783 |
# Set the Alu as detected Device Type |
|
| 784 |
Alu=deviceType |
|
| 785 |
Id=Id+1 |
|
| 786 |
|
|
| 789 |
try: |
|
| 790 |
# For PyOpenCL import |
|
| 791 |
import pyopencl as cl |
|
| 792 |
Id=1 |
|
| 793 |
for platform in cl.get_platforms(): |
|
| 794 |
for device in platform.get_devices(): |
|
| 795 |
deviceType=cl.device_type.to_string(device.type) |
|
| 796 |
print "Device #%i of type %s : %s" % (Id,deviceType,device.name) |
|
| 797 |
if Id == Device: |
|
| 798 |
# Set the Alu as detected Device Type |
|
| 799 |
Alu=deviceType |
|
| 800 |
Id=Id+1 |
|
| 801 |
except ImportError: |
|
| 802 |
print "Platform does not seem to support CUDA" |
|
| 803 |
|
|
| 787 | 804 |
average=numpy.array([]).astype(numpy.float32) |
| 788 | 805 |
median=numpy.array([]).astype(numpy.float32) |
| 789 | 806 |
stddev=numpy.array([]).astype(numpy.float32) |
| ... | ... | |
| 828 | 845 |
except: |
| 829 | 846 |
print "Problem with %i // computations on Cuda" % Jobs |
| 830 | 847 |
elif GpuStyle=='OpenCL': |
| 831 |
# try: |
|
| 832 |
# avg,med,std=MetropolisOpenCL(circle,Iterations,Redo,Jobs,ParaStyle,Alu,Device) |
|
| 833 |
# except: |
|
| 834 |
# print "Problem with %i // computations on OpenCL" % Jobs |
|
| 835 |
avg,med,std=MetropolisOpenCL(circle,Iterations,Redo,Jobs,ParaStyle,Alu,Device,DoublePrecision) |
|
| 848 |
try: |
|
| 849 |
avg,med,std=MetropolisOpenCL(circle,Iterations,Redo,Jobs,ParaStyle,Alu,Device,DoublePrecision) |
|
| 850 |
except: |
|
| 851 |
print "Problem with %i // computations on OpenCL" % Jobs |
|
| 836 | 852 |
|
| 837 | 853 |
if (avg,med,std) != (0,0,0): |
| 838 | 854 |
print "jobs,avg,med,std",Jobs,avg,med,std |
| ... | ... | |
| 852 | 868 |
numpy.savetxt("Pi%s_%s_%s_%s_%s_%i_%.8i_Device%i_%s_%s" % (Precision,Alu,GpuStyle,ParaStyle,JobStart,JobEnd,Iterations,Device,Metrology,gethostname()),numpy.transpose(ToSave))
|
| 853 | 869 |
Jobs+=JobStep |
| 854 | 870 |
|
| 855 |
FitAndPrint(ExploredJobs,median,Curves) |
|
| 871 |
if Fit: |
|
| 872 |
FitAndPrint(ExploredJobs,median,Curves) |
|
| 856 | 873 |
|
Formats disponibles : Unified diff