def cumsum(x_gpu):
"""
Cumulative sum.
Return the cumulative sum of the elements in the specified array.
Parameters
----------
x_gpu : pycuda.gpuarray.GPUArray
Input array.
Returns
-------
c_gpu : pycuda.gpuarray.GPUArray
Output array containing cumulative sum of `x_gpu`.
Notes
-----
Higher dimensional arrays are implicitly flattened row-wise by this function.
Examples
--------
>>> import pycuda.autoinit
>>> import pycuda.gpuarray as gpuarray
>>> import misc
>>> x_gpu = gpuarray.to_gpu(np.random.rand(5).astype(np.float32))
>>> c_gpu = misc.cumsum(x_gpu)
>>> np.allclose(c_gpu.get(), np.cumsum(x_gpu.get()))
True
"""
try:
func = cumsum.cache[x_gpu.dtype]
except KeyError:
func = scan.InclusiveScanKernel(
x_gpu.dtype, 'a+b', preamble='#include <pycuda-complex.hpp>')
cumsum.cache[x_gpu.dtype] = func
return func(x_gpu)
a_gpu,
np.int32(size),
block=(BLOCK_S, 1, 1),
grid=((size - 1) // BLOCK_S + 1, 1, 1))
time_inefficient.append(time.time() - start)
start = time.time()
scan_efficient(b_gpu2,
a_gpu2,
np.int32(size),
block=(BLOCK_S, 1, 1),
grid=((size - 1) // BLOCK_S + 1, 1, 1))
time_efficient.append(time.time() - start)
knl_gpu = gpuarray.to_gpu(a)
scan_knl = scan.InclusiveScanKernel(np.float32, "a + b", "0")
start = time.time()
scan_knl(knl_gpu)
time_scan_knl.append(time.time() - start)
'''
print a
print "\n"
print b_gpu.get()
print "\n"
print scan_cpu(a, size)
'''
MAKE_PLOT = True
if MAKE_PLOT:
import matplotlib as mpl
mpl.use('agg')