def test_IdpyMethodCU(self):
cu = CUDA()
cu.SetDevice()
tenet = cu.GetTenet()
SwapArrays = self.M_SwapArrays(tenet)
zeros = IdpyMemory.Const(self.n,
dtype=np.int32,
const=0,
tenet=tenet)
ones = IdpyMemory.Const(self.n,
dtype=np.int32,
const=1,
tenet=tenet)
mem_list = [zeros, ones]
print()
print("[0]: ", mem_list[0].D2H(), mem_list[0].dtype)
print("[1]: ", mem_list[1].D2H(), mem_list[1].dtype)
print("Swapping")
SwapArrays.Deploy(mem_list)
print("[0]: ", mem_list[0].D2H(), mem_list[0].dtype)
print("[1]: ", mem_list[1].D2H(), mem_list[1].dtype)
check_1 = np.zeros(self.n, dtype=np.int32)
check_0 = np.zeros(self.n, dtype=np.int32)
check_0.fill(1)
checks = []
checks += [AllTrue(mem_list[0].D2H() == check_0)]
checks += [AllTrue(mem_list[1].D2H() == check_1)]
tenet.End()
self.assertTrue(AllTrue(checks))
def GetTenet(params_dict):
'''
GetTenet:
it looks general enough to be further abstracted
'''
if 'lang' in params_dict and params_dict['lang'] == CUDA_T:
if idpy_langs_sys[CUDA_T]:
cu = CUDA()
device = 0 if 'device' not in params_dict else params_dict['device']
cu.SetDevice(device)
print("CUDA: ", cu.GetDeviceName())
return cu.GetTenet()
else:
raise Exception(
"Selected lang = CUDA_T but CUDA is not found on the system!")
if 'lang' in params_dict and params_dict['lang'] == OCL_T:
if idpy_langs_sys[OCL_T]:
ocl = OpenCL()
cl_type = 'gpu' if 'cl_kind' not in params_dict else params_dict[
'cl_kind']
device = 0 if 'device' not in params_dict else params_dict['device']
ocl.SetDevice(kind=cl_type, device=device)
print("OpenCL: ", ocl.GetDeviceName())
return ocl.GetTenet()
else:
raise Exception(
"Selected lang = OCL_T but openCL is not found on the system!")
def test_IdpyKernelCU(self):
cu = CUDA()
cu.SetDevice()
tenet = cu.GetTenet()
grid, block = ((self.n + self.block_size - 1) // self.block_size,
1, 1), (self.block_size, 1, 1)
myTypes = CustomTypes({'SpinType': 'unsigned int'})
np_c = NpTypes()
SumOne = self.K_SumOne(custom_types=myTypes.Push(),
constants={'DATA_N': self.n})
SumOne_Idea = SumOne(tenet=tenet, grid=grid, block=block)
A = IdpyMemory.Const(self.n,
dtype=np_c.C[myTypes['SpinType']],
const=0,
tenet=tenet)
print()
print("A: ", A.D2H(), A.dtype)
print("SumOne_Idea.Deploy([A])")
SumOne_Idea.Deploy([A])
print("A: ", A.D2H(), A.dtype)
check_array = np.zeros(self.n, dtype=np_c.C[myTypes['SpinType']])
check_array.fill(1)
checks = [AllTrue(A.D2H() == check_array)]
tenet.End()
self.assertTrue(AllTrue(checks))
def test_CUDA_ManageTenet(self):
cuda = CUDA()
cuda.SetDevice()
tenet = cuda.GetTenet()
check_type = isinstance(tenet, CUTenet)
tenet.End()
del cuda
self.assertTrue(check_type)
def test_IdpyMethodLoopCU(self):
cu = CUDA()
cu.SetDevice()
tenet = cu.GetTenet()
zeros = IdpyMemory.Const(self.n,
dtype=np.int32,
const=0,
tenet=tenet)
ones = IdpyMemory.Const(self.n,
dtype=np.int32,
const=1,
tenet=tenet)
mem_dict = {'zeros': zeros, 'ones': ones}
SwapArraysLoop = IdpyLoop(
[mem_dict], [[(self.M_SwapArrays(tenet), ['zeros', 'ones'])]])
print()
print("['zeros']: ", mem_dict['zeros'].D2H(),
mem_dict['zeros'].dtype)
print("['ones']: ", mem_dict['ones'].D2H(), mem_dict['ones'].dtype)
print("SwapArraysLoop(range(1))")
SwapArraysLoop.Run(range(1))
print("['zeros']: ", mem_dict['zeros'].D2H(),
mem_dict['zeros'].dtype)
print("['ones']: ", mem_dict['ones'].D2H(), mem_dict['ones'].dtype)
print("SwapArraysLoop(range(4))")
SwapArraysLoop.Run(range(4))
print("['zeros']: ", mem_dict['zeros'].D2H(),
mem_dict['zeros'].dtype)
print("['ones']: ", mem_dict['ones'].D2H(), mem_dict['ones'].dtype)
print("SwapArraysLoop(range(7))")
SwapArraysLoop.Run(range(7))
print("['zeros']: ", mem_dict['zeros'].D2H(),
mem_dict['zeros'].dtype)
print("['ones']: ", mem_dict['ones'].D2H(), mem_dict['ones'].dtype)
checks = []
checks += [
AllTrue(
list(mem_dict['zeros'].D2H() == np.zeros(self.n,
dtype=np.int32)))
]
check_ones = np.zeros(self.n, dtype=np.int32)
check_ones.fill(1)
checks += [AllTrue(list(mem_dict['ones'].D2H() == check_ones))]
tenet.End()
self.assertTrue(AllTrue(checks))
def test_IdpyArray(self):
cu = CUDA()
cu.SetDevice()
tenet = cu.GetTenet()
rand_mem = IdpyMemory.Array(10, dtype=np.int32, tenet=tenet)
on_dev_range = IdpyMemory.OnDevice(np.arange(10, dtype=np.int32),
tenet=tenet)
zeros = IdpyMemory.Zeros(10, dtype=np.float32, tenet=tenet)
i_range = IdpyMemory.Range(10, tenet=tenet)
const = IdpyMemory.Const(10,
dtype=np.int32,
const=self.constant,
tenet=tenet)
print()
print("rand_mem:\t", rand_mem.D2H(), rand_mem.dtype)
print("on_dev_range:\t", on_dev_range.D2H(), on_dev_range.dtype)
print("zeros:\t", zeros.D2H(), zeros.dtype)
print("i_range:\t", i_range.D2H(), i_range.dtype)
print("const:\t", const.D2H(), const.dtype)
int_buffer = np.zeros(10, dtype=np.int32)
const.D2H(int_buffer)
print("int_buffer: ", int_buffer, int_buffer.dtype)
'''
checks
'''
checks = []
checks += [
AllTrue(
list(on_dev_range.D2H() == np.arange(10, dtype=np.int32)))
]
checks += [
AllTrue(list(zeros.D2H() == np.zeros(10, dtype=np.float32)))
]
checks += [
AllTrue(list(i_range.D2H() == np.arange(10, dtype=np.int32)))
]
chk_const = np.zeros(10, dtype=np.int32)
chk_const.fill(self.constant)
checks += [AllTrue(list(const.D2H() == chk_const))]
tenet.End()
self.assertTrue(AllTrue(checks))
def test_IdpyKernelLoopCU(self):
cu = CUDA()
cu.SetDevice()
tenet = cu.GetTenet()
grid, block = ((self.n + self.block_size - 1) // self.block_size,
1, 1), (self.block_size, 1, 1)
myTypes = CustomTypes({'SpinType': 'unsigned int'})
np_c = NpTypes()
SumOne = self.K_SumOne(custom_types=myTypes.Push(),
constants={'DATA_N': self.n})
A = IdpyMemory.Const(self.n,
dtype=np_c.C[myTypes['SpinType']],
const=0,
tenet=tenet)
mem_dict = {'A': A}
SumOne_Loop = IdpyLoop(
[mem_dict],
[[(SumOne(tenet=tenet, grid=grid, block=block), ['A'])]])
print()
print("A: ", A.D2H(), A.dtype)
print("SumOne_Loop.Run(range(1))")
SumOne_Loop.Run(range(1))
print("A: ", A.D2H(), A.dtype)
print("SumOne_Loop.Run(range(8))")
SumOne_Loop.Run(range(8))
print("A: ", A.D2H(), A.dtype)
check_array = np.zeros(self.n, dtype=np_c.C[myTypes['SpinType']])
check_array.fill(9)
checks = []
checks += [AllTrue(A.D2H() == check_array)]
tenet.End()
self.assertTrue(AllTrue(checks))
def test_IdpyKernelLoopConstCU(self):
cu = CUDA()
cu.SetDevice()
tenet = cu.GetTenet()
grid, block = ((self.n + self.block_size - 1) // self.block_size,
1, 1), (self.block_size, 1, 1)
myTypes = CustomTypes({'SpinType': 'unsigned int'})
np_c = NpTypes()
SumConst = self.K_SumConst(custom_types=myTypes.Push(),
constants={'DATA_N': self.n})
A = IdpyMemory.Const(self.n,
dtype=np_c.C[myTypes['SpinType']],
const=0,
tenet=tenet)
mem_dict = {'A': A, 'const': np.int32(self.in_const)}
# https://stackoverflow.com/questions/5710690/pycuda-passing-variable-by-value-to-kernel
SumOne_Loop = IdpyLoop([mem_dict], [[(SumConst(
tenet=tenet, grid=grid, block=block), ['A', 'const'])]])
print()
print("A: ", A.D2H(), A.dtype)
print("SumOne_Loop.Run(range(1))")
SumOne_Loop.Run(range(1))
print("A: ", A.D2H(), A.dtype)
print("SumOne_Loop.Run(range(8))")
SumOne_Loop.Run(range(8))
print("A: ", A.D2H(), A.dtype)
check_array = np.zeros(self.n, dtype=np_c.C[myTypes['SpinType']])
check_array.fill(self.in_const * 9)
checks = []
checks += [AllTrue(A.D2H() == check_array)]
tenet.End()
self.assertTrue(AllTrue(checks))
def test_IdpyKernelFuncLoopMultStreamCU(self):
cu = CUDA()
cu.SetDevice()
tenet = cu.GetTenet()
grid, block = ((self.n + self.block_size - 1) // self.block_size,
1, 1), (self.block_size, 1, 1)
myTypes = CustomTypes({'SpinType': 'unsigned int'})
np_c = NpTypes()
SumTwoArrConst = self.K_SumTwoArrays(custom_types=myTypes.Push(),
constants={'DATA_N': self.n},
f_classes=[
self.F_SumTwoArraysPtr,
self.F_SumTwoArraysRet,
self.F_SumTwoArraysVal
])
SumConst = self.K_SumConst(custom_types=myTypes.Push(),
constants={'DATA_N': self.n})
A = IdpyMemory.Const(self.n,
dtype=np_c.C[myTypes['SpinType']],
const=0,
tenet=tenet)
B = IdpyMemory.Const(self.n,
dtype=np_c.C[myTypes['SpinType']],
const=1,
tenet=tenet)
C = IdpyMemory.Const(self.n,
dtype=np_c.C[myTypes['SpinType']],
const=2,
tenet=tenet)
D = IdpyMemory.Const(self.n,
dtype=np_c.C[myTypes['SpinType']],
const=3,
tenet=tenet)
print()
print("A: ", A.D2H(), A.dtype)
print("B: ", B.D2H(), B.dtype)
print("C: ", C.D2H(), C.dtype)
print("D: ", D.D2H(), D.dtype)
'''
Checking result
'''
a, b, c, cc = A.D2H()[0], B.D2H()[0], C.D2H()[0], self.in_const
d = D.D2H()[0]
for i in range(2):
'''first stream'''
c += cc
c += a + b
c += a + b
c += a + b + cc
a, c = c, a
'''second stream'''
d += cc
# https://stackoverflow.com/questions/5710690/pycuda-passing-variable-by-value-to-kernel
##
mem_dict_0 = {
'A': A,
'B': B,
'C': C,
'const': np_c.C[myTypes['SpinType']](self.in_const)
}
mem_dict_1 = {
'D': D,
'const': np_c.C[myTypes['SpinType']](self.in_const)
}
SumTwoArrConst_Loop = IdpyLoop(
[mem_dict_0, mem_dict_1],
[[(SumTwoArrConst(tenet=tenet, grid=grid,
block=block), ['A', 'B', 'C', 'const']),
(self.M_SwapArrays(tenet), ['A', 'C'])],
[
(SumConst(tenet=tenet, grid=grid,
block=block), ['D', 'const']),
]])
print()
print("SumTwoArrConst_Loop.Run(range(2))")
SumTwoArrConst_Loop.Run(range(2))
print("A: ", A.D2H(), A.dtype, a)
print("B: ", B.D2H(), B.dtype, b)
print("C: ", C.D2H(), C.dtype, c)
print("D: ", D.D2H(), D.dtype, d)
checks = []
check_array = np.full(self.n, a, dtype=np_c.C[myTypes['SpinType']])
checks += [AllTrue(A.D2H() == check_array)]
check_array = np.full(self.n, b, dtype=np_c.C[myTypes['SpinType']])
checks += [AllTrue(B.D2H() == check_array)]
check_array = np.full(self.n, c, dtype=np_c.C[myTypes['SpinType']])
checks += [AllTrue(C.D2H() == check_array)]
check_array = np.full(self.n, d, dtype=np_c.C[myTypes['SpinType']])
checks += [AllTrue(D.D2H() == check_array)]
tenet.End()
self.assertTrue(AllTrue(checks))