nx_gpu = 120 nx_cpu = 80 #nx_gpu = nx_cpu = 100 ny, nz = 300, 64 tmax, tgap = 200, 10 # instances gpu_devices = common_gpu.gpu_device_list(print_info=False) context = cl.Context(gpu_devices) mainf_list = [gpu.Fields(context, device, nx_gpu, ny, nz) for device in gpu_devices] mainf_list.append( cpu.Fields(nx_cpu, ny, nz) ) nodef = NodeFields(mainf_list) core = NodeCore(nodef) exchange = NodeExchange(nodef) pbc = NodePbc(nodef, 'y') pbc = NodePbc(nodef, 'z') pbc_x = NodePbc(nodef, 'x') tfunc = lambda tstep: np.sin(0.1 * tstep) #incident = NodeDirectIncident(nodef, 'ez', (0, 20, 0), (nodef.nx-1, 20, nz-1), tfunc) incident = NodeDirectIncident(nodef, 'ez', (20, 0, 0), (20, ny-1, nz-1), tfunc) getf = NodeGetFields(nodef, 'ez', (0, 0, 2), (nodef.nx-1, ny-1, 2)) # plot import matplotlib.pyplot as plt plt.figure(figsize=(12,8)) plt.ion() for anx in nodef.accum_nx_list[1:]: plt.plot((anx, anx), (0, ny), color='w', linewidth=0.2)
def test_y_pbc_x_exchange(self):
# instance
nx, ny, nz = 40, 50, 60
#nx, ny, nz = 3, 4, 5
gpu_devices = common_gpu.gpu_device_list(print_info=False)
context = cl.Context(gpu_devices)
gpuf = gpu.Fields(context, gpu_devices[0], nx, ny, nz)
cpuf = cpu.Fields(nx, ny, nz)
mainf_list = [gpuf, cpuf]
nodef = NodeFields(mainf_list)
core = NodeCore(nodef)
pbc = NodePbc(nodef, 'y')
exchange = NodeExchange(nodef)
# generate random source
ehs_gpu = common_update.generate_random_ehs(nx, ny, nz, nodef.dtype)
gpuf.set_eh_bufs(*ehs_gpu)
ehs_gpu_dict = dict( zip(['ex', 'ey', 'ez', 'hx', 'hy', 'hz'], ehs_gpu) )
ehs_cpu = common_update.generate_random_ehs(nx, ny, nz, nodef.dtype)
cpuf.set_ehs(*ehs_cpu)
ehs_cpu_dict = dict( zip(['ex', 'ey', 'ez', 'hx', 'hy', 'hz'], ehs_cpu) )
# verify
for mainf in mainf_list:
mainf.update_e()
pbc.update_e()
exchange.update_e()
for mainf in mainf_list:
mainf.update_h()
pbc.update_h()
exchange.update_h()
mainf_list[-1].enqueue_barrier()
getf0, getf1 = {}, {}
# x-axis exchange
getf0['e'] = gpu.GetFields(gpuf, ['ey', 'ez'], (nx-1, 0, 0), (nx-1, ny-2, nz-2))
getf1['e'] = cpu.GetFields(cpuf, ['ey', 'ez'], (0, 0, 0), (0, ny-2, nz-2))
getf0['h'] = gpu.GetFields(gpuf, ['hy', 'hz'], (nx-1, 1, 1), (nx-1, ny-1, nz-1))
getf1['h'] = cpu.GetFields(cpuf, ['hy', 'hz'], (0, 1, 1), (0, ny-1, nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
g0 = getf0[eh].get_fields()
g1 = getf1[eh].get_fields()
norm = np.linalg.norm(g0 - g1)
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'x-axis exchange', eh))
# y-axis pbc gpu
getf0['e'] = gpu.GetFields(gpuf, ['ex', 'ez'], (0, ny-1, 0), (nx-2, ny-1, nz-2))
getf1['e'] = gpu.GetFields(gpuf, ['ex', 'ez'], (0, 0, 0), (nx-2, 0, nz-2))
getf0['h'] = gpu.GetFields(gpuf, ['hx', 'hz'], (1, ny-1, 1), (nx-1, ny-1, nz-1))
getf1['h'] = gpu.GetFields(gpuf, ['hx', 'hz'], (1, 0, 1), (nx-1, 0, nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
g0 = getf0[eh].get_fields()
g1 = getf1[eh].get_fields()
norm = np.linalg.norm(g0 - g1)
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'y-axis pbc gpu', eh))
# y-axis pbc cpu
getf0['e'] = cpu.GetFields(cpuf, ['ex', 'ez'], (0, ny-1, 0), (nx-2, ny-1, nz-2))
getf1['e'] = cpu.GetFields(cpuf, ['ex', 'ez'], (0, 0, 0), (nx-2, 0, nz-2))
getf0['h'] = cpu.GetFields(cpuf, ['hx', 'hz'], (1, ny-1, 1), (nx-1, ny-1, nz-1))
getf1['h'] = cpu.GetFields(cpuf, ['hx', 'hz'], (1, 0, 1), (nx-1, 0, nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
g0 = getf0[eh].get_fields()
g1 = getf1[eh].get_fields()
norm = np.linalg.norm(g0 - g1)
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'y-axis pbc cpu', eh))
def runTest(self):
nx, ny, nz = self.args
# instance
gpu_devices = common_gpu.gpu_device_list(print_info=False)
context = cl.Context(gpu_devices)
mainf_list = [gpu.Fields(context, device, nx, ny, nz) \
for device in gpu_devices]
mainf_list.append( cpu.Fields(nx, ny, nz) )
nodef = NodeFields(mainf_list)
dtype = nodef.dtype
# buffer instance
nodef.append_buffer_fields(cpu.Fields(3, ny, nz, mpi_type='x-'))
nodef.append_buffer_fields(cpu.Fields(3, nodef.nx, nz, mpi_type='y+'))
nodef.append_buffer_fields(cpu.Fields(3, nodef.nx, nz, mpi_type='y-'))
nodef.append_buffer_fields(cpu.Fields(3, nodef.nx, ny, mpi_type='z+'))
nodef.append_buffer_fields(cpu.Fields(3, nodef.nx, ny, mpi_type='z-'))
exchange = NodeExchange(nodef)
# generate random source
for f in mainf_list[:-1]:
nx, ny, nz = f.ns
ehs = common_update.generate_random_ehs(nx, ny, nz, dtype)
f.set_eh_bufs(*ehs)
for f in nodef.cpuf_dict.values():
nx, ny, nz = f.ns
ehs = common_update.generate_random_ehs(nx, ny, nz, dtype)
f.set_ehs(*ehs)
# verify
exchange.update_e()
exchange.update_h()
getf0, getf1 = {}, {}
# mainf list
self.gpu, self.cpu = gpu, cpu
for f0, f1 in zip(mainf_list[:-1], mainf_list[1:]):
getf0['e'] = getattr(self, f0.device_type).GetFields(f0, ['ey', 'ez'], \
(f0.nx-1, 0, 0), (f0.nx-1, f0.ny-2, f0.nz-2))
getf1['e'] = getattr(self, f1.device_type).GetFields(f1, ['ey', 'ez'], \
(0, 0, 0), (0, f1.ny-2, f1.nz-2))
getf0['h'] = getattr(self, f0.device_type).GetFields(f0, ['hy', 'hz'], \
(f0.nx-1, 1, 1), (f0.nx-1, f0.ny-1, f0.nz-1))
getf1['h'] = getattr(self, f1.device_type).GetFields(f1, ['hy', 'hz'], \
(0, 1, 1), (0, f1.ny-1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm(getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%s, %g, %s, %s, %s' % \
(self.args, norm, 'mainf', \
getf0[eh].mainf.device_type, getf1[eh].mainf.device_type) )
# buffer 'x-'
f0, f1 = nodef.cpuf_dict['x-'], mainf_list[0]
getf0['e'] = cpu.GetFields(f0, ['ey', 'ez'], \
(f0.nx-1, 0, 0), (f0.nx-1, f0.ny-2, f0.nz-2))
getf1['e'] = gpu.GetFields(f1, ['ey', 'ez'], \
(1, 0, 0), (1, f1.ny-2, f1.nz-2))
getf0['h'] = cpu.GetFields(f0, ['hy', 'hz'], \
(f0.nx-2, 1, 1), (f0.nx-2, f0.ny-1, f0.nz-1))
getf1['h'] = gpu.GetFields(f1, ['hy', 'hz'], \
(0, 1, 1), (0, f1.ny-1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'x-', eh) )
# buffer 'y+'
anx_list = nodef.accum_nx_list
f1 = nodef.cpuf_dict['y+']
for f0, anx0, anx1 in zip(mainf_list, anx_list[:-1], anx_list[1:]):
getf0['e'] = getattr(self, f0.device_type).GetFields(f0, ['ex', 'ez'], \
(0, f0.ny-1, 0), (f0.nx-2, f0.ny-1, f0.nz-2))
getf1['e'] = cpu.GetFields(f1, ['ey', 'ez'], \
(1, anx0, 0), (1, anx1-1, f1.nz-2))
getf0['h'] = getattr(self, f0.device_type).GetFields(f0, ['hx', 'hz'], \
(1, f0.ny-2, 1), (f0.nx-1, f0.ny-2, f0.nz-1))
getf1['h'] = cpu.GetFields(f1, ['hy', 'hz'], \
(0, anx0+1, 1), (0, anx1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'y+', eh) )
# buffer 'y-'
f0 = nodef.cpuf_dict['y-']
for f1, anx0, anx1 in zip(mainf_list, anx_list[:-1], anx_list[1:]):
getf0['e'] = cpu.GetFields(f0, ['ey', 'ez'], \
(f0.nx-1, anx0, 0), (f0.nx-1, anx1-1, f0.nz-2))
getf1['e'] = getattr(self, f1.device_type).GetFields(f1, ['ex', 'ez'], \
(0, 1, 0), (f1.nx-2, 1, f1.nz-2))
getf0['h'] = cpu.GetFields(f0, ['hy', 'hz'], \
(f0.nx-2, anx0+1, 1), (f0.nx-2, anx1, f0.nz-1))
getf1['h'] = getattr(self, f1.device_type).GetFields(f1, ['hx', 'hz'], \
(1, 0, 1), (f1.nx-1, 0, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'y-', eh) )
# buffer 'z+'
f1 = nodef.cpuf_dict['z+']
for f0, anx0, anx1 in zip(mainf_list, anx_list[:-1], anx_list[1:]):
getf0['e'] = getattr(self, f0.device_type).GetFields(f0, ['ex', 'ey'], \
(0, 0, f0.nz-1), (f0.nx-2, f0.ny-2, f0.nz-1))
getf1['e'] = cpu.GetFields(f1, ['ey', 'ez'], \
(1, anx0, 0), (1, anx1-1, f1.nz-2))
getf0['h'] = getattr(self, f0.device_type).GetFields(f0, ['hx', 'hy'], \
(1, 1, f0.nz-2), (f0.nx-1, f0.ny-1, f0.nz-2))
getf1['h'] = cpu.GetFields(f1, ['hy', 'hz'], \
(0, anx0+1, 1), (0, anx1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'z+', eh) )
# buffer 'z-'
f0 = nodef.cpuf_dict['z-']
for f1, anx0, anx1 in zip(mainf_list, anx_list[:-1], anx_list[1:]):
getf0['e'] = cpu.GetFields(f0, ['ey', 'ez'], \
(f0.nx-1, anx0, 0), (f0.nx-1, anx1-1, f0.nz-2))
getf1['e'] = getattr(self, f1.device_type).GetFields(f1, ['ex', 'ey'], \
(0, 0, 1), (f1.nx-2, f1.ny-2, 1))
getf0['h'] = cpu.GetFields(f0, ['hy', 'hz'], \
(f0.nx-2, anx0+1, 1), (f0.nx-2, anx1, f0.nz-1))
getf1['h'] = getattr(self, f1.device_type).GetFields(f1, ['hx', 'hy'], \
(1, 1, 0), (f1.nx-1, f1.ny-1, 0))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'z-', eh) )
def test_y_pbc_x_exchange(self):
# instance
nx, ny, nz = 40, 50, 60
#nx, ny, nz = 3, 4, 5
gpu_devices = common_gpu.gpu_device_list(print_info=False)
context = cl.Context(gpu_devices)
gpuf = gpu.Fields(context, gpu_devices[0], nx, ny, nz)
cpuf = cpu.Fields(nx, ny, nz)
mainf_list = [gpuf, cpuf]
nodef = NodeFields(mainf_list)
core = NodeCore(nodef)
pbc = NodePbc(nodef, 'y')
exchange = NodeExchange(nodef)
# generate random source
ehs_gpu = common_update.generate_random_ehs(nx, ny, nz, nodef.dtype)
gpuf.set_eh_bufs(*ehs_gpu)
ehs_gpu_dict = dict(zip(['ex', 'ey', 'ez', 'hx', 'hy', 'hz'], ehs_gpu))
ehs_cpu = common_update.generate_random_ehs(nx, ny, nz, nodef.dtype)
cpuf.set_ehs(*ehs_cpu)
ehs_cpu_dict = dict(zip(['ex', 'ey', 'ez', 'hx', 'hy', 'hz'], ehs_cpu))
# verify
for mainf in mainf_list:
mainf.update_e()
pbc.update_e()
exchange.update_e()
for mainf in mainf_list:
mainf.update_h()
pbc.update_h()
exchange.update_h()
mainf_list[-1].enqueue_barrier()
getf0, getf1 = {}, {}
# x-axis exchange
getf0['e'] = gpu.GetFields(gpuf, ['ey', 'ez'], (nx - 1, 0, 0),
(nx - 1, ny - 2, nz - 2))
getf1['e'] = cpu.GetFields(cpuf, ['ey', 'ez'], (0, 0, 0),
(0, ny - 2, nz - 2))
getf0['h'] = gpu.GetFields(gpuf, ['hy', 'hz'], (nx - 1, 1, 1),
(nx - 1, ny - 1, nz - 1))
getf1['h'] = cpu.GetFields(cpuf, ['hy', 'hz'], (0, 1, 1),
(0, ny - 1, nz - 1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
g0 = getf0[eh].get_fields()
g1 = getf1[eh].get_fields()
norm = np.linalg.norm(g0 - g1)
self.assertEqual(norm, 0,
'%g, %s, %s' % (norm, 'x-axis exchange', eh))
# y-axis pbc gpu
getf0['e'] = gpu.GetFields(gpuf, ['ex', 'ez'], (0, ny - 1, 0),
(nx - 2, ny - 1, nz - 2))
getf1['e'] = gpu.GetFields(gpuf, ['ex', 'ez'], (0, 0, 0),
(nx - 2, 0, nz - 2))
getf0['h'] = gpu.GetFields(gpuf, ['hx', 'hz'], (1, ny - 1, 1),
(nx - 1, ny - 1, nz - 1))
getf1['h'] = gpu.GetFields(gpuf, ['hx', 'hz'], (1, 0, 1),
(nx - 1, 0, nz - 1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
g0 = getf0[eh].get_fields()
g1 = getf1[eh].get_fields()
norm = np.linalg.norm(g0 - g1)
self.assertEqual(norm, 0,
'%g, %s, %s' % (norm, 'y-axis pbc gpu', eh))
# y-axis pbc cpu
getf0['e'] = cpu.GetFields(cpuf, ['ex', 'ez'], (0, ny - 1, 0),
(nx - 2, ny - 1, nz - 2))
getf1['e'] = cpu.GetFields(cpuf, ['ex', 'ez'], (0, 0, 0),
(nx - 2, 0, nz - 2))
getf0['h'] = cpu.GetFields(cpuf, ['hx', 'hz'], (1, ny - 1, 1),
(nx - 1, ny - 1, nz - 1))
getf1['h'] = cpu.GetFields(cpuf, ['hx', 'hz'], (1, 0, 1),
(nx - 1, 0, nz - 1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
g0 = getf0[eh].get_fields()
g1 = getf1[eh].get_fields()
norm = np.linalg.norm(g0 - g1)
self.assertEqual(norm, 0,
'%g, %s, %s' % (norm, 'y-axis pbc cpu', eh))
def runTest(self):
axis, nx, ny, nz = self.args
self.gpu, self.cpu = gpu, cpu
# instance
gpu_devices = common_gpu.gpu_device_list(print_info=False)
context = cl.Context(gpu_devices)
mainf_list = [gpu.Fields(context, device, nx, ny, nz) \
for device in gpu_devices]
mainf_list.append(cpu.Fields(nx, ny, nz))
nodef = NodeFields(mainf_list)
dtype = nodef.dtype
pbc = NodePbc(nodef, axis)
exchange = NodeExchange(nodef)
# generate random source
for f in mainf_list[:-1]:
nx, ny, nz = f.ns
ehs = common_update.generate_random_ehs(nx, ny, nz, dtype)
f.set_eh_bufs(*ehs)
for f in nodef.cpuf_dict.values():
nx, ny, nz = f.ns
ehs = common_update.generate_random_ehs(nx, ny, nz, dtype)
f.set_ehs(*ehs)
# verify
for mainf in mainf_list:
mainf.update_e()
pbc.update_e()
exchange.update_e()
for mainf in mainf_list:
mainf.update_h()
pbc.update_h()
exchange.update_h()
mainf_list[-1].enqueue_barrier()
getf0, getf1 = {}, {}
if axis == 'x':
f0, f1 = mainf_list[0], mainf_list[-1]
getf0['e'] = getattr(self, f0.device_type).GetFields(f0, ['ey', 'ez'], \
(0, 0, 0), (0, f0.ny-2, f0.nz-2))
getf1['e'] = getattr(self, f1.device_type).GetFields(f1, ['ey', 'ez'], \
(f1.nx-1, 0, 0), (f1.nx-1, f1.ny-2, f1.nz-2))
getf0['h'] = getattr(self, f0.device_type).GetFields(f0, ['hy', 'hz'], \
(0, 1, 1), (0, f0.ny-1, f0.nz-1))
getf1['h'] = getattr(self, f1.device_type).GetFields(f1, ['hy', 'hz'], \
(f1.nx-1, 1, 1), (f1.nx-1, f1.ny-1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm(getf0[eh].get_fields() -
getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'x', eh))
elif axis == 'y':
for f in mainf_list:
getf0['e'] = getattr(self, f.device_type).GetFields(f, ['ex', 'ez'], \
(0, 0, 0), (f.nx-2, 0, f.nz-2))
getf1['e'] = getattr(self, f.device_type).GetFields(f, ['ex', 'ez'], \
(0, f.ny-1, 0), (f.nx-2, f.ny-1, f.nz-2))
getf0['h'] = getattr(self, f.device_type).GetFields(f, ['hx', 'hz'], \
(1, 0, 1), (f.nx-1, 0, f.nz-1))
getf1['h'] = getattr(self, f.device_type).GetFields(f, ['hx', 'hz'], \
(1, f.ny-1, 1), (f.nx-1, f.ny-1, f.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(
norm, 0,
'%g, %s, %s, %s' % (norm, 'y', eh, f.device_type))
elif axis == 'z':
for f in mainf_list:
getf0['e'] = getattr(self, f.device_type).GetFields(f, ['ex', 'ey'], \
(0, 0, f.nz-1), (f.nx-2, f.ny-2, f.nz-1))
getf1['e'] = getattr(self, f.device_type).GetFields(f, ['ex', 'ey'], \
(0, 0, 0), (f.nx-2, f.ny-2, 0))
getf0['h'] = getattr(self, f.device_type).GetFields(f, ['hx', 'hy'], \
(1, 1, f.nz-1), (f.nx-1, f.ny-1, f.nz-1))
getf1['h'] = getattr(self, f.device_type).GetFields(f, ['hx', 'hy'], \
(1, 1, 0), (f.nx-1, f.ny-1, 0))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'z', eh))
def runTest(self):
nx, ny, nz = self.args
# instance
gpu_devices = common_gpu.gpu_device_list(print_info=False)
context = cl.Context(gpu_devices)
mainf_list = [gpu.Fields(context, device, nx, ny, nz) \
for device in gpu_devices]
mainf_list.append(cpu.Fields(nx, ny, nz))
nodef = NodeFields(mainf_list)
dtype = nodef.dtype
# buffer instance
nodef.append_buffer_fields(cpu.Fields(3, ny, nz, mpi_type='x-'))
nodef.append_buffer_fields(cpu.Fields(3, nodef.nx, nz, mpi_type='y+'))
nodef.append_buffer_fields(cpu.Fields(3, nodef.nx, nz, mpi_type='y-'))
nodef.append_buffer_fields(cpu.Fields(3, nodef.nx, ny, mpi_type='z+'))
nodef.append_buffer_fields(cpu.Fields(3, nodef.nx, ny, mpi_type='z-'))
exchange = NodeExchange(nodef)
# generate random source
for f in mainf_list[:-1]:
nx, ny, nz = f.ns
ehs = common_update.generate_random_ehs(nx, ny, nz, dtype)
f.set_eh_bufs(*ehs)
for f in nodef.cpuf_dict.values():
nx, ny, nz = f.ns
ehs = common_update.generate_random_ehs(nx, ny, nz, dtype)
f.set_ehs(*ehs)
# verify
exchange.update_e()
exchange.update_h()
getf0, getf1 = {}, {}
# mainf list
self.gpu, self.cpu = gpu, cpu
for f0, f1 in zip(mainf_list[:-1], mainf_list[1:]):
getf0['e'] = getattr(self, f0.device_type).GetFields(f0, ['ey', 'ez'], \
(f0.nx-1, 0, 0), (f0.nx-1, f0.ny-2, f0.nz-2))
getf1['e'] = getattr(self, f1.device_type).GetFields(f1, ['ey', 'ez'], \
(0, 0, 0), (0, f1.ny-2, f1.nz-2))
getf0['h'] = getattr(self, f0.device_type).GetFields(f0, ['hy', 'hz'], \
(f0.nx-1, 1, 1), (f0.nx-1, f0.ny-1, f0.nz-1))
getf1['h'] = getattr(self, f1.device_type).GetFields(f1, ['hy', 'hz'], \
(0, 1, 1), (0, f1.ny-1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm(getf0[eh].get_fields() -
getf1[eh].get_fields())
self.assertEqual(norm, 0, '%s, %g, %s, %s, %s' % \
(self.args, norm, 'mainf', \
getf0[eh].mainf.device_type, getf1[eh].mainf.device_type) )
# buffer 'x-'
f0, f1 = nodef.cpuf_dict['x-'], mainf_list[0]
getf0['e'] = cpu.GetFields(f0, ['ey', 'ez'], \
(f0.nx-1, 0, 0), (f0.nx-1, f0.ny-2, f0.nz-2))
getf1['e'] = gpu.GetFields(f1, ['ey', 'ez'], \
(1, 0, 0), (1, f1.ny-2, f1.nz-2))
getf0['h'] = cpu.GetFields(f0, ['hy', 'hz'], \
(f0.nx-2, 1, 1), (f0.nx-2, f0.ny-1, f0.nz-1))
getf1['h'] = gpu.GetFields(f1, ['hy', 'hz'], \
(0, 1, 1), (0, f1.ny-1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'x-', eh))
# buffer 'y+'
anx_list = nodef.accum_nx_list
f1 = nodef.cpuf_dict['y+']
for f0, anx0, anx1 in zip(mainf_list, anx_list[:-1], anx_list[1:]):
getf0['e'] = getattr(self, f0.device_type).GetFields(f0, ['ex', 'ez'], \
(0, f0.ny-1, 0), (f0.nx-2, f0.ny-1, f0.nz-2))
getf1['e'] = cpu.GetFields(f1, ['ey', 'ez'], \
(1, anx0, 0), (1, anx1-1, f1.nz-2))
getf0['h'] = getattr(self, f0.device_type).GetFields(f0, ['hx', 'hz'], \
(1, f0.ny-2, 1), (f0.nx-1, f0.ny-2, f0.nz-1))
getf1['h'] = cpu.GetFields(f1, ['hy', 'hz'], \
(0, anx0+1, 1), (0, anx1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'y+', eh))
# buffer 'y-'
f0 = nodef.cpuf_dict['y-']
for f1, anx0, anx1 in zip(mainf_list, anx_list[:-1], anx_list[1:]):
getf0['e'] = cpu.GetFields(f0, ['ey', 'ez'], \
(f0.nx-1, anx0, 0), (f0.nx-1, anx1-1, f0.nz-2))
getf1['e'] = getattr(self, f1.device_type).GetFields(f1, ['ex', 'ez'], \
(0, 1, 0), (f1.nx-2, 1, f1.nz-2))
getf0['h'] = cpu.GetFields(f0, ['hy', 'hz'], \
(f0.nx-2, anx0+1, 1), (f0.nx-2, anx1, f0.nz-1))
getf1['h'] = getattr(self, f1.device_type).GetFields(f1, ['hx', 'hz'], \
(1, 0, 1), (f1.nx-1, 0, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'y-', eh))
# buffer 'z+'
f1 = nodef.cpuf_dict['z+']
for f0, anx0, anx1 in zip(mainf_list, anx_list[:-1], anx_list[1:]):
getf0['e'] = getattr(self, f0.device_type).GetFields(f0, ['ex', 'ey'], \
(0, 0, f0.nz-1), (f0.nx-2, f0.ny-2, f0.nz-1))
getf1['e'] = cpu.GetFields(f1, ['ey', 'ez'], \
(1, anx0, 0), (1, anx1-1, f1.nz-2))
getf0['h'] = getattr(self, f0.device_type).GetFields(f0, ['hx', 'hy'], \
(1, 1, f0.nz-2), (f0.nx-1, f0.ny-1, f0.nz-2))
getf1['h'] = cpu.GetFields(f1, ['hy', 'hz'], \
(0, anx0+1, 1), (0, anx1, f1.nz-1))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'z+', eh))
# buffer 'z-'
f0 = nodef.cpuf_dict['z-']
for f1, anx0, anx1 in zip(mainf_list, anx_list[:-1], anx_list[1:]):
getf0['e'] = cpu.GetFields(f0, ['ey', 'ez'], \
(f0.nx-1, anx0, 0), (f0.nx-1, anx1-1, f0.nz-2))
getf1['e'] = getattr(self, f1.device_type).GetFields(f1, ['ex', 'ey'], \
(0, 0, 1), (f1.nx-2, f1.ny-2, 1))
getf0['h'] = cpu.GetFields(f0, ['hy', 'hz'], \
(f0.nx-2, anx0+1, 1), (f0.nx-2, anx1, f0.nz-1))
getf1['h'] = getattr(self, f1.device_type).GetFields(f1, ['hx', 'hy'], \
(1, 1, 0), (f1.nx-1, f1.ny-1, 0))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ['e', 'h']:
norm = np.linalg.norm( \
getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, '%g, %s, %s' % (norm, 'z-', eh))
def runTest(self):
axis, nx, ny, nz = self.args
self.gpu, self.cpu = gpu, cpu
# instance
gpu_devices = common_gpu.gpu_device_list(print_info=False)
context = cl.Context(gpu_devices)
mainf_list = [gpu.Fields(context, device, nx, ny, nz) for device in gpu_devices]
mainf_list.append(cpu.Fields(nx, ny, nz))
nodef = NodeFields(mainf_list)
dtype = nodef.dtype
pbc = NodePbc(nodef, axis)
exchange = NodeExchange(nodef)
# generate random source
for f in mainf_list[:-1]:
nx, ny, nz = f.ns
ehs = common_update.generate_random_ehs(nx, ny, nz, dtype)
f.set_eh_bufs(*ehs)
for f in nodef.cpuf_dict.values():
nx, ny, nz = f.ns
ehs = common_update.generate_random_ehs(nx, ny, nz, dtype)
f.set_ehs(*ehs)
# verify
for mainf in mainf_list:
mainf.update_e()
pbc.update_e()
exchange.update_e()
for mainf in mainf_list:
mainf.update_h()
pbc.update_h()
exchange.update_h()
mainf_list[-1].enqueue_barrier()
getf0, getf1 = {}, {}
if axis == "x":
f0, f1 = mainf_list[0], mainf_list[-1]
getf0["e"] = getattr(self, f0.device_type).GetFields(f0, ["ey", "ez"], (0, 0, 0), (0, f0.ny - 2, f0.nz - 2))
getf1["e"] = getattr(self, f1.device_type).GetFields(
f1, ["ey", "ez"], (f1.nx - 1, 0, 0), (f1.nx - 1, f1.ny - 2, f1.nz - 2)
)
getf0["h"] = getattr(self, f0.device_type).GetFields(f0, ["hy", "hz"], (0, 1, 1), (0, f0.ny - 1, f0.nz - 1))
getf1["h"] = getattr(self, f1.device_type).GetFields(
f1, ["hy", "hz"], (f1.nx - 1, 1, 1), (f1.nx - 1, f1.ny - 1, f1.nz - 1)
)
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ["e", "h"]:
norm = np.linalg.norm(getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, "%g, %s, %s" % (norm, "x", eh))
elif axis == "y":
for f in mainf_list:
getf0["e"] = getattr(self, f.device_type).GetFields(f, ["ex", "ez"], (0, 0, 0), (f.nx - 2, 0, f.nz - 2))
getf1["e"] = getattr(self, f.device_type).GetFields(
f, ["ex", "ez"], (0, f.ny - 1, 0), (f.nx - 2, f.ny - 1, f.nz - 2)
)
getf0["h"] = getattr(self, f.device_type).GetFields(f, ["hx", "hz"], (1, 0, 1), (f.nx - 1, 0, f.nz - 1))
getf1["h"] = getattr(self, f.device_type).GetFields(
f, ["hx", "hz"], (1, f.ny - 1, 1), (f.nx - 1, f.ny - 1, f.nz - 1)
)
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ["e", "h"]:
norm = np.linalg.norm(getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, "%g, %s, %s, %s" % (norm, "y", eh, f.device_type))
elif axis == "z":
for f in mainf_list:
getf0["e"] = getattr(self, f.device_type).GetFields(
f, ["ex", "ey"], (0, 0, f.nz - 1), (f.nx - 2, f.ny - 2, f.nz - 1)
)
getf1["e"] = getattr(self, f.device_type).GetFields(f, ["ex", "ey"], (0, 0, 0), (f.nx - 2, f.ny - 2, 0))
getf0["h"] = getattr(self, f.device_type).GetFields(
f, ["hx", "hy"], (1, 1, f.nz - 1), (f.nx - 1, f.ny - 1, f.nz - 1)
)
getf1["h"] = getattr(self, f.device_type).GetFields(f, ["hx", "hy"], (1, 1, 0), (f.nx - 1, f.ny - 1, 0))
for getf in getf0.values() + getf1.values():
getf.get_event().wait()
for eh in ["e", "h"]:
norm = np.linalg.norm(getf0[eh].get_fields() - getf1[eh].get_fields())
self.assertEqual(norm, 0, "%g, %s, %s" % (norm, "z", eh))