diff --git a/pystencils/datahandling/serial_datahandling.py b/pystencils/datahandling/serial_datahandling.py index a3e6820f0c28deb57cd75083e6f5a99f23abd0dd..931a7df30c2413712f2fd4ec1898a11db7239462 100644 --- a/pystencils/datahandling/serial_datahandling.py +++ b/pystencils/datahandling/serial_datahandling.py @@ -23,6 +23,7 @@ class SerialDataHandling(DataHandling): periodicity: Union[bool, Sequence[bool]] = False, default_target: str = 'cpu', opencl_queue=None, + opencl_ctx=None, array_handler=None) -> None: """ Creates a data handling for single node simulations. @@ -44,6 +45,8 @@ class SerialDataHandling(DataHandling): self.custom_data_cpu = DotDict() self.custom_data_gpu = DotDict() self._custom_data_transfer_functions = {} + self._opencl_queue = opencl_queue + self._opencl_ctx = opencl_ctx if array_handler: self.array_handler = array_handler @@ -315,11 +318,15 @@ class SerialDataHandling(DataHandling): result.append(get_periodic_boundary_functor(filtered_stencil, ghost_layers=gls)) else: from pystencils.gpucuda.periodicity import get_periodic_boundary_functor as boundary_func + target = 'gpu' if not isinstance(self.array_handler, PyOpenClArrayHandler) else 'opencl' result.append(boundary_func(filtered_stencil, self._domainSize, index_dimensions=self.fields[name].index_dimensions, index_dim_shape=values_per_cell, dtype=self.fields[name].dtype.numpy_dtype, - ghost_layers=gls)) + ghost_layers=gls, + target=target, + opencl_queue=self._opencl_queue, + opencl_ctx=self._opencl_ctx)) if target == 'cpu': def result_functor(): diff --git a/pystencils/gpucuda/periodicity.py b/pystencils/gpucuda/periodicity.py index e94a1796e3a66da626fc4e13fcc0413f5f93961d..080ef44ebd995e1d8dcf4dbe1f0839e0a218fde6 100644 --- a/pystencils/gpucuda/periodicity.py +++ b/pystencils/gpucuda/periodicity.py @@ -1,7 +1,8 @@ import numpy as np +import pystencils.gpucuda +import pystencils.opencl from pystencils import Assignment, Field -from pystencils.gpucuda import make_python_function from pystencils.gpucuda.kernelcreation import create_cuda_kernel from pystencils.slicing import get_periodic_boundary_src_dst_slices, normalize_slice @@ -25,16 +26,22 @@ def create_copy_kernel(domain_size, from_slice, to_slice, index_dimensions=0, in update_eqs.append(eq) ast = create_cuda_kernel(update_eqs, iteration_slice=to_slice, skip_independence_check=True) - return make_python_function(ast) + return ast def get_periodic_boundary_functor(stencil, domain_size, index_dimensions=0, index_dim_shape=1, ghost_layers=1, - thickness=None, dtype=float): + thickness=None, dtype=float, target='gpu', opencl_queue=None, opencl_ctx=None): + assert target in ['gpu', 'opencl'] src_dst_slice_tuples = get_periodic_boundary_src_dst_slices(stencil, ghost_layers, thickness) kernels = [] index_dimensions = index_dimensions + for src_slice, dst_slice in src_dst_slice_tuples: - kernels.append(create_copy_kernel(domain_size, src_slice, dst_slice, index_dimensions, index_dim_shape, dtype)) + ast = create_copy_kernel(domain_size, src_slice, dst_slice, index_dimensions, index_dim_shape, dtype) + if target == 'gpu': + kernels.append(pystencils.gpucuda.make_python_function(ast)) + else: + kernels.append(pystencils.opencl.make_python_function(ast, opencl_queue, opencl_ctx)) def functor(pdfs, **_): for kernel in kernels: