Révision 296 ETSN/MySteps_6.py
| MySteps_6.py (revision 296) | ||
|---|---|---|
| 289 | 289 |
Device=0 |
| 290 | 290 |
Calls=1 |
| 291 | 291 |
Threads=1 |
| 292 |
|
|
| 292 |
Serial=True |
|
| 293 |
|
|
| 293 | 294 |
import getopt |
| 294 | 295 |
|
| 295 |
HowToUse='%s -g <CUDA/OpenCL> -s <SizeOfVector> -d <DeviceId> -c <SillyCalls> -t <Threads>' |
|
| 296 |
HowToUse='%s -n -g <CUDA/OpenCL> -s <SizeOfVector> -d <DeviceId> -c <SillyCalls> -t <Threads>'
|
|
| 296 | 297 |
|
| 297 | 298 |
try: |
| 298 |
opts, args = getopt.getopt(sys.argv[1:],"hg:s:d:c:t:",["gpustyle=","size=","device=","calls=","threads="]) |
|
| 299 |
opts, args = getopt.getopt(sys.argv[1:],"hng:s:d:c:t:",["gpustyle=","size=","device=","calls=","threads="])
|
|
| 299 | 300 |
except getopt.GetoptError: |
| 300 | 301 |
print(HowToUse % sys.argv[0]) |
| 301 | 302 |
sys.exit(2) |
| ... | ... | |
| 347 | 348 |
GpuStyle = arg |
| 348 | 349 |
elif opt in ("-s", "--size"):
|
| 349 | 350 |
SIZE = int(arg) |
| 351 |
elif opt in ("-n"):
|
|
| 352 |
Serial = False |
|
| 350 | 353 |
|
| 351 | 354 |
print("Device Selection : %i" % Device)
|
| 352 | 355 |
print("GpuStyle used : %s" % GpuStyle)
|
| 353 | 356 |
print("Size of complex vector : %i" % SIZE)
|
| 354 | 357 |
print("Number of silly calls : %i" % Calls)
|
| 355 | 358 |
print("Number of Threads : %i" % Threads)
|
| 359 |
print("Serial compute : %i" % Serial)
|
|
| 356 | 360 |
|
| 357 | 361 |
if GpuStyle=='CUDA': |
| 358 | 362 |
try: |
| ... | ... | |
| 391 | 395 |
b_np = np.random.rand(SIZE).astype(np.float32) |
| 392 | 396 |
|
| 393 | 397 |
# Native Implementation |
| 394 |
TimeIn=time.time() |
|
| 395 |
res_np=NativeSillyAddition(a_np,b_np,Calls) |
|
| 396 |
NativeElapsed=time.time()-TimeIn |
|
| 397 |
NativeRate=int(SIZE/NativeElapsed) |
|
| 398 |
print("NativeRate: %i" % NativeRate)
|
|
| 398 |
if Serial: |
|
| 399 |
TimeIn=time.time() |
|
| 400 |
res_np=NativeSillyAddition(a_np,b_np,Calls) |
|
| 401 |
NativeElapsed=time.time()-TimeIn |
|
| 402 |
NativeRate=int(SIZE/NativeElapsed) |
|
| 403 |
print("NativeRate: %i" % NativeRate)
|
|
| 399 | 404 |
|
| 400 | 405 |
# OpenCL Implementation |
| 401 | 406 |
if GpuStyle=='OpenCL' or GpuStyle=='all': |
| ... | ... | |
| 406 | 411 |
OpenCLRate=int(SIZE/OpenCLElapsed) |
| 407 | 412 |
print("OpenCLRate: %i" % OpenCLRate)
|
| 408 | 413 |
# Check on OpenCL with Numpy: |
| 409 |
print(res_cl - res_np) |
|
| 410 |
print(np.linalg.norm(res_cl - res_np)) |
|
| 411 |
try: |
|
| 412 |
assert np.allclose(res_np, res_cl) |
|
| 413 |
except: |
|
| 414 |
print("Results between Native & OpenCL seem to be too different!")
|
|
| 414 |
if Serial: |
|
| 415 |
print(res_cl - res_np) |
|
| 416 |
print(np.linalg.norm(res_cl - res_np)) |
|
| 417 |
try: |
|
| 418 |
assert np.allclose(res_np, res_cl) |
|
| 419 |
except: |
|
| 420 |
print("Results between Native & OpenCL seem to be too different!")
|
|
| 415 | 421 |
|
| 416 |
print("OpenCLvsNative ratio: %f" % (OpenCLRate/NativeRate))
|
|
| 422 |
print("OpenCLvsNative ratio: %f" % (OpenCLRate/NativeRate))
|
|
| 417 | 423 |
|
| 418 | 424 |
# CUDA Implementation |
| 419 | 425 |
if GpuStyle=='CUDA' or GpuStyle=='all': |
| ... | ... | |
| 423 | 429 |
CUDARate=int(SIZE/CUDAElapsed) |
| 424 | 430 |
print("CUDARate: %i" % CUDARate)
|
| 425 | 431 |
# Check on CUDA with Numpy: |
| 426 |
print(res_cuda - res_np) |
|
| 427 |
print(np.linalg.norm(res_cuda - res_np)) |
|
| 428 |
try: |
|
| 429 |
assert np.allclose(res_np, res_cuda) |
|
| 430 |
except: |
|
| 431 |
print("Results between Native & CUDA seem to be too different!")
|
|
| 432 |
if Serial: |
|
| 433 |
print(res_cuda - res_np) |
|
| 434 |
print(np.linalg.norm(res_cuda - res_np)) |
|
| 435 |
try: |
|
| 436 |
assert np.allclose(res_np, res_cuda) |
|
| 437 |
except: |
|
| 438 |
print("Results between Native & CUDA seem to be too different!")
|
|
| 432 | 439 |
|
| 433 |
print("CUDAvsNative ratio: %f" % (CUDARate/NativeRate))
|
|
| 440 |
print("CUDAvsNative ratio: %f" % (CUDARate/NativeRate))
|
|
| 434 | 441 |
|
| 435 | 442 |
|
| 436 | 443 |
|
Formats disponibles : Unified diff