def test_prescan():
a = np.arange(2048).astype(np.int32)
reference = np.empty_like(a)
ref_sum = MyScan.exprefixsumNumba(a, reference)
a1 = np.arange(1024).astype(np.int32)
a2 = np.arange(1024, 2048).astype(np.int32)
ref1 = np.empty_like(a1)
ref2 = np.empty_like(a2)
ref_sum1 = MyScan.exprefixsumNumba(a1, ref1)
ref_sum2 = MyScan.exprefixsumNumba(a2, ref2)
dAux = cuda.device_array(2, dtype=np.int32)
dA = cuda.to_device(a)
sm_size = 1024 * a.dtype.itemsize
MyScan.prescan[2, 512, 0, sm_size](dA, dAux)
aux = dAux.copy_to_host()
a_gpu = dA.copy_to_host()
print "finish"
def test_exprefixsumNumba_init_random(self):
in_ary = np.random.rand(ARRAY_SIZE)
out_ary = np.empty_like(in_ary)
init = np.random.randint(100000)
output = MyScan.exprefixsumNumba(in_ary, out_ary, init=init)
# check last carry
assert np.isclose(output - init, in_ary.sum()), 'carry return is not sum'
carry = init
for i in xrange(ARRAY_SIZE):
assert out_ary[i] == carry, 'output array not correct'
carry += in_ary[i]
def test_last_block():
MAX_TPB = 512
n = 1024
a = np.arange(n).astype(np.int32)
reference = np.empty_like(a)
start = timer()
MyScan.exprefixsumNumba(a, reference, init=0)
end = timer()
auxidx = -1
elb = a.size
p2elb = np.int(np.ceil(np.log2(elb)))
telb = 2 ** p2elb
tlb = telb / 2
startIdx = 0
sm_size = telb * a.itemsize
aux = np.empty(1, dtype=np.int8)
trash = cuda.device_array(1)
e1, e2 = cuda.event(), cuda.event()
e1.record()
MyScan.last_scan[1, tlb, 0, sm_size](a, aux, -1, elb, startIdx)
e2.record()
print "CPU took: ", (end - start) * 1000, " ms"
print "Kernel took: ", cuda.event_elapsed_time(e1, e2), " ms"
print (a == reference).all()
def test_recursive_big_scan():
print "running recursive scan test"
MAX_TPB = 512
n = 2e6
n = int(n)
a = np.arange(n).astype(np.int32)
reference = np.empty_like(a)
start = timer()
sum_ref = MyScan.exprefixsumNumba(a, reference, init=0)
end = timer()
dA = cuda.to_device(a)
# e1, e2 = cuda.event(), cuda.event()
# e1.record()
# e2.record()
start2 = timer()
total_sum = MyScan.scan_gpu(dA)
end2 = timer()
dA.copy_to_host(ary=a)
sum_gpu = total_sum.copy_to_host()
print "sum_ref = ", sum_ref
print "sum_gpu = ", sum_gpu
print "CPU took: ", (end - start) * 1000, " ms"
print "Kernel took: ", (end2 - start2) * 1000, " ms"
print (a == reference).all()
def test_recursive_step_by_step():
## setup
MAX_TPB = 512
n = 5000
a = np.arange(n).astype(np.int32)
reference = np.empty_like(a)
start = timer()
sum_ref = MyScan.exprefixsumNumba(a, reference, init=0)
end = timer()
dA = cuda.to_device(a)
# e1, e2 = cuda.event(), cuda.event()
# e1.record()
# e2.record()
## scan
in_ary = dA
epb = MAX_TPB * 2
whole_blocks = n // epb
el_last_block = n % epb
n_scans = whole_blocks
if el_last_block != 0:
n_scans += 1
## prescan
dAux = cuda.device_array(shape=n_scans, dtype=np.int32)
sm_size = epb * in_ary.dtype.itemsize
MyScan.prescan[whole_blocks, MAX_TPB, 0, sm_size](in_ary, dAux)
# tIn = in_ary.copy_to_host()
# tAux = dAux.copy_to_host()
p2elb = np.int(np.ceil(np.log2(el_last_block)))
p2_el_last_block = 2 ** p2elb # the smallest number of elements that is power of 2
tlb = p2_el_last_block >> 1 # number of threads in last block
sm_size = p2_el_last_block * in_ary.dtype.itemsize
startIdx = n - el_last_block
auxIdx = n_scans - 1
MyScan.last_scan[1, tlb, 0, sm_size](in_ary, dAux, auxIdx,
el_last_block, startIdx)
in_ary2 = dAux
n2 = in_ary2.size
if n2 < MAX_TPB << 1:
el_last_block2 = n2
p2elb2 = np.int(np.ceil(np.log2(el_last_block2)))
p2_el_last_block2 = 2 ** p2elb # the smallest number of elements that is power of 2
tlb2 = p2_el_last_block2 >> 1 # number of threads in last block
total_sum = cuda.device_array(shape=1, dtype=np.int32)
sm_size2 = p2_el_last_block2 * in_ary2.dtype.itemsize
startIdx2 = 0
auxIdx2 = 0
MyScan.last_scan[1, tlb2, 0, sm_size2](in_ary2, total_sum, auxIdx2, el_last_block2, startIdx2)
MyScan.scan_sum[n_scans, tlb](in_ary, dAux)
tIn = in_ary.copy_to_host()
tAux = dAux.copy_to_host()
tSum = total_sum.copy_to_host()
print "finish"