Révision 46 Pi/GPU/Pi-GPU.py
| Pi-GPU.py (revision 46) | ||
|---|---|---|
| 304 | 304 |
|
| 305 | 305 |
# Initialisation des variables en les CASTant correctement |
| 306 | 306 |
|
| 307 |
# Je detecte un peripherique GPU dans la liste des peripheriques |
|
| 308 |
# for platform in cl.get_platforms(): |
|
| 309 |
# for device in platform.get_devices(): |
|
| 310 |
# if cl.device_type.to_string(device.type)=='GPU': |
|
| 311 |
# GPU=device |
|
| 312 |
#print "GPU detected: ",device.name |
|
| 313 |
|
|
| 314 |
HasGPU=False |
|
| 315 |
Id=1 |
|
| 316 |
# Primary Device selection based on Device Id |
|
| 317 |
for platform in cl.get_platforms(): |
|
| 318 |
for device in platform.get_devices(): |
|
| 319 |
deviceType=cl.device_type.to_string(device.type) |
|
| 320 |
if Id==Device and not HasGPU: |
|
| 321 |
GPU=device |
|
| 322 |
print "CPU/GPU selected: ",device.name |
|
| 323 |
HasGPU=True |
|
| 324 |
Id=Id+1 |
|
| 325 |
# Default Device selection based on ALU Type |
|
| 326 |
for platform in cl.get_platforms(): |
|
| 327 |
for device in platform.get_devices(): |
|
| 328 |
deviceType=cl.device_type.to_string(device.type) |
|
| 329 |
if deviceType=="GPU" and Alu=="GPU" and not HasGPU: |
|
| 330 |
GPU=device |
|
| 331 |
print "GPU selected: ",device.name |
|
| 332 |
HasGPU=True |
|
| 333 |
if deviceType=="CPU" and Alu=="CPU" and not HasGPU: |
|
| 334 |
GPU=device |
|
| 335 |
print "CPU selected: ",device.name |
|
| 336 |
HasGPU=True |
|
| 307 |
if Device==0: |
|
| 308 |
print "Enter XPU selector based on ALU type: first selected" |
|
| 309 |
HasXPU=False |
|
| 310 |
# Default Device selection based on ALU Type |
|
| 311 |
for platform in cl.get_platforms(): |
|
| 312 |
for device in platform.get_devices(): |
|
| 313 |
deviceType=cl.device_type.to_string(device.type) |
|
| 314 |
if deviceType=="GPU" and Alu=="GPU" and not HasXPU: |
|
| 315 |
XPU=device |
|
| 316 |
print "GPU selected: ",device.name |
|
| 317 |
HasXPU=True |
|
| 318 |
if deviceType=="CPU" and Alu=="CPU" and not HasXPU: |
|
| 319 |
XPU=device |
|
| 320 |
print "CPU selected: ",device.name |
|
| 321 |
HasXPU=True |
|
| 322 |
else: |
|
| 323 |
print "Enter XPU selector based on device number & ALU type" |
|
| 324 |
Id=1 |
|
| 325 |
HasXPU=False |
|
| 326 |
# Primary Device selection based on Device Id |
|
| 327 |
for platform in cl.get_platforms(): |
|
| 328 |
for device in platform.get_devices(): |
|
| 329 |
deviceType=cl.device_type.to_string(device.type) |
|
| 330 |
if Id==Device and Alu==deviceType and HasXPU==False: |
|
| 331 |
XPU=device |
|
| 332 |
print "CPU/GPU selected: ",device.name |
|
| 333 |
HasXPU=True |
|
| 334 |
Id=Id+1 |
|
| 335 |
if HasXPU==False: |
|
| 336 |
print "No XPU #%i of type %s found in all of %i devices, sorry..." % \ |
|
| 337 |
(Device,Alu,Id-1) |
|
| 338 |
return(0,0,0) |
|
| 337 | 339 |
|
| 338 | 340 |
# Je cree le contexte et la queue pour son execution |
| 339 | 341 |
#ctx = cl.create_some_context() |
| 340 |
ctx = cl.Context([GPU])
|
|
| 342 |
ctx = cl.Context([XPU])
|
|
| 341 | 343 |
queue = cl.CommandQueue(ctx, |
| 342 | 344 |
properties=cl.command_queue_properties.PROFILING_ENABLE) |
| 343 | 345 |
|
| ... | ... | |
| 493 | 495 |
# Set defaults values |
| 494 | 496 |
# Alu can be CPU or GPU |
| 495 | 497 |
Alu='CPU' |
| 496 |
# Id of GPU : 0 is for first find !
|
|
| 498 |
# Id of GPU : 1 is for first find !
|
|
| 497 | 499 |
Device=0 |
| 498 | 500 |
# GPU style can be Cuda (Nvidia implementation) or OpenCL |
| 499 | 501 |
GpuStyle='OpenCL' |
| ... | ... | |
| 522 | 524 |
for opt, arg in opts: |
| 523 | 525 |
if opt == '-h': |
| 524 | 526 |
print '%s -o (Out of Core Metrology) -c (Print Curves) -a <CPU/GPU> -d <DeviceId> -g <CUDA/OpenCL> -p <Threads/Hybrid/Blocks> -i <Iterations> -s <JobStart> -e <JobEnd> -r <RedoToImproveStats>' % sys.argv[0] |
| 527 |
|
|
| 528 |
print "\nInformations about devices detected under OpenCL:" |
|
| 529 |
# For PyOpenCL import |
|
| 530 |
import pyopencl as cl |
|
| 531 |
Id=1 |
|
| 532 |
for platform in cl.get_platforms(): |
|
| 533 |
for device in platform.get_devices(): |
|
| 534 |
deviceType=cl.device_type.to_string(device.type) |
|
| 535 |
print "Device #%i of type %s : %s" % (Id,deviceType,device.name) |
|
| 536 |
Id=Id+1 |
|
| 537 |
|
|
| 525 | 538 |
sys.exit() |
| 526 | 539 |
elif opt == '-o': |
| 527 | 540 |
OutMetrology=True |
| ... | ... | |
| 599 | 612 |
start=time.time() |
| 600 | 613 |
if GpuStyle=='CUDA': |
| 601 | 614 |
try: |
| 602 |
MetropolisCuda(circle,Iterations,1,Jobs,ParaStyle) |
|
| 615 |
a,m,s=MetropolisCuda(circle,Iterations,1,Jobs,ParaStyle)
|
|
| 603 | 616 |
except: |
| 604 | 617 |
print "Problem with %i // computations on Cuda" % Jobs |
| 605 | 618 |
elif GpuStyle=='OpenCL': |
| 606 | 619 |
try: |
| 607 |
MetropolisOpenCL(circle,Iterations,1,Jobs,ParaStyle,Alu,Device) |
|
| 620 |
a,m,s=MetropolisOpenCL(circle,Iterations,1,Jobs,ParaStyle,Alu,Device)
|
|
| 608 | 621 |
except: |
| 609 | 622 |
print "Problem with %i // computations on OpenCL" % Jobs |
| 610 | 623 |
duration=numpy.append(duration,time.time()-start) |
| 611 |
avg=numpy.mean(duration) |
|
| 612 |
med=numpy.median(duration) |
|
| 613 |
std=numpy.std(duration) |
|
| 624 |
if (a,m,s) != (0,0,0): |
|
| 625 |
avg=numpy.mean(duration) |
|
| 626 |
med=numpy.median(duration) |
|
| 627 |
std=numpy.std(duration) |
|
| 628 |
else: |
|
| 629 |
print "Values seem to be wrong..." |
|
| 614 | 630 |
else: |
| 615 | 631 |
if GpuStyle=='CUDA': |
| 616 | 632 |
try: |
| ... | ... | |
| 632 | 648 |
else: |
| 633 | 649 |
print "Values seem to be wrong..." |
| 634 | 650 |
#THREADS*=2 |
| 635 |
numpy.savez("Pi_%s_%s_%s_%s_%i_%.8i_Device%i_%s_%s" % (Alu,GpuStyle,ParaStyle,JobStart,JobEnd,Iterations,Device,Metrology,gethostname()),(ExploredJobs,average,median,stddev))
|
|
| 636 |
numpy.savetxt("Pi_%s_%s_%s_%s_%i_%.8i_Device%i_%s_%s" % (Alu,GpuStyle,ParaStyle,JobStart,JobEnd,Iterations,Device,Metrology,gethostname()),(ExploredJobs,average,median,stddev))
|
|
| 651 |
if len(average)!=0: |
|
| 652 |
numpy.savez("Pi_%s_%s_%s_%s_%i_%.8i_Device%i_%s_%s" % (Alu,GpuStyle,ParaStyle,JobStart,JobEnd,Iterations,Device,Metrology,gethostname()),(ExploredJobs,average,median,stddev))
|
|
| 653 |
numpy.savetxt("Pi_%s_%s_%s_%s_%i_%.8i_Device%i_%s_%s" % (Alu,GpuStyle,ParaStyle,JobStart,JobEnd,Iterations,Device,Metrology,gethostname()),(ExploredJobs,average,median,stddev))
|
|
| 637 | 654 |
Jobs+=1 |
| 638 | 655 |
|
| 639 | 656 |
FitAndPrint(ExploredJobs,median,Curves) |
Formats disponibles : Unified diff