def execute_sum_on_gpu_via_CU(a, b):
assert a.shape == b.shape
cu = CU(target='gpu')
result = np.zeros_like(a)
d_result = cu.output(result)
cu.enqueue(sum, ntid=result.size, args=(a, b, result))
cu.wait()
cu.close()
return result
def main():
target = sys.argv[1]
cu = CU(target)
width = 500 * 40
height = 750 * 40
with closing(cu):
image = np.zeros((width, height), dtype=np.uint8)
d_image = cu.output(image)
s = timer()
create_fractal(cu, -2.0, 1.0, -1.0, 1.0, d_image, 20)
cu.wait()
e = timer()
print(e - s)
def main():
target = sys.argv[1]
cu = CU(target)
width = 500 * 40
height = 750 * 40
with closing(cu):
image = np.zeros((width, height), dtype=np.uint8)
d_image = cu.output(image)
s = timer()
create_fractal(cu, -2.0, 1.0, -1.0, 1.0, d_image, 20)
cu.wait()
e = timer()
print(e - s)
def execute_sum_on_gpu_via_CU(a, b):
assert a.shape == b.shape
cu = CU(target='gpu')
result = np.zeros_like(a)
d_result = cu.output(result)
cu.enqueue(sum, ntid=result.size, args=(a, b, result))
cu.wait()
cu.close()
return result
def monte_carlo_pricer(paths, dt, interest, volatility):
n = paths.shape[0]
cu = CU('gpu')
with closing(cu): # <--- auto closing the cu
# seed the cuRAND RNG
cu.configure(builtins.random.seed, 1234)
c0 = interest - 0.5 * volatility ** 2
c1 = volatility * np.sqrt(dt)
# Step 1. prepare data
# -------- Exercise --------
# allocate scratchpad memory on the device for noises
d_noises = # fill in the RHS
# allocate a in-out memory on the device for the initial prices
# "paths" is a 2-D array with the 1st dimension as number of paths
# the 2nd dimension as the number of time step.
d_last_paths = cu.inout(paths[:, 0])
# -- Step 2. simulation loop --
# compute one step for all paths in each iteration
for i in range(1, paths.shape[1]):
# Allocate a in-out memory for the next batch of simulated prices
d_paths = cu.inout(paths[:, i])
# Use builtin kernel "builtins.random.normal"
# to generate a sequence of normal distribution.
cu.enqueue(builtins.random.normal, # the kernel
ntid=n, # number of threads
args=(d_noises,)) # arguments
# -------- Exercise --------
# Enqueue the "step" kernel
# Hints: The "tid" argument is automatically inserted.
# prepare for next step
d_last_paths = d_paths
# wait the the task to complete
cu.wait()
def main():
try:
target = sys.argv[1]
except:
target = 'cpu'
cu = CU(target)
width = 500 * 10
height = 750 * 10
with closing(cu):
image = np.zeros((width, height), dtype=np.uint8)
d_image = cu.output(image)
s = timer()
create_fractal(cu, -2.0, 1.0, -1.0, 1.0, d_image, 20)
cu.wait()
e = timer()
print('time: %f' % (e - s, ))
# print(image)
imshow(image)
show()
def main():
try:
target = sys.argv[1]
except:
target = 'cpu'
cu = CU(target)
width = 500 * 10
height = 750 * 10
with closing(cu):
image = np.zeros((width, height), dtype=np.uint8)
d_image = cu.output(image)
s = timer()
create_fractal(cu, -2.0, 1.0, -1.0, 1.0, d_image, 20)
cu.wait()
e = timer()
print('time: %f' % (e - s,))
# print(image)
imshow(image)
show()
