From 63357a4c89c7554ff8a8a6456d218dd2ebb65a76 Mon Sep 17 00:00:00 2001
From: Christoph Rettinger <christoph.rettinger@fau.de>
Date: Thu, 18 Jun 2020 18:57:57 +0200
Subject: [PATCH] Allowed different field layouts in LBM codegen
---
python/lbmpy_walberla/walberla_lbm_generation.py | 11 +++++++----
1 file changed, 7 insertions(+), 4 deletions(-)
diff --git a/python/lbmpy_walberla/walberla_lbm_generation.py b/python/lbmpy_walberla/walberla_lbm_generation.py
index 7e32867b0..e3523dc1f 100644
--- a/python/lbmpy_walberla/walberla_lbm_generation.py
+++ b/python/lbmpy_walberla/walberla_lbm_generation.py
@@ -122,7 +122,7 @@ def __lattice_model(generation_context, class_name, lb_method, stream_collide_as
generation_context.write_file("{}.cpp".format(class_name), source)
-def generate_lattice_model(generation_context, class_name, collision_rule, refinement_scaling=None,
+def generate_lattice_model(generation_context, class_name, collision_rule, field_layout='fzyx', refinement_scaling=None,
**create_kernel_params):
# usually a numpy layout is chosen by default i.e. xyzf - which is bad for waLBerla where at least the spatial
@@ -138,8 +138,11 @@ def generate_lattice_model(generation_context, class_name, collision_rule, refin
if create_kernel_params['target'] == 'gpu':
raise ValueError("Lattice Models can only be generated for CPUs. To generate LBM on GPUs use sweeps directly")
- src_field = ps.Field.create_generic('pdfs', dim, dtype, index_dimensions=1, layout='fzyx', index_shape=(q,))
- dst_field = ps.Field.create_generic('pdfs_tmp', dim, dtype, index_dimensions=1, layout='fzyx', index_shape=(q,))
+ if field_layout != 'fzyx' and create_kernel_params['cpu_vectorize_info']['instruction_set'] != None:
+ raise ValueError("Vectorization is not available for given field layout, use fzyx instead")
+
+ src_field = ps.Field.create_generic('pdfs', dim, dtype, index_dimensions=1, layout=field_layout, index_shape=(q,))
+ dst_field = ps.Field.create_generic('pdfs_tmp', dim, dtype, index_dimensions=1, layout=field_layout, index_shape=(q,))
stream_collide_update_rule = create_lbm_kernel(collision_rule, src_field, dst_field, StreamPullTwoFieldsAccessor())
stream_collide_ast = create_kernel(stream_collide_update_rule, **create_kernel_params)
@@ -149,7 +152,7 @@ def generate_lattice_model(generation_context, class_name, collision_rule, refin
collide_ast = create_kernel(collide_update_rule, **create_kernel_params)
collide_ast.function_name = 'kernel_collide'
- stream_update_rule = create_stream_pull_only_kernel(lb_method.stencil, None, 'pdfs', 'pdfs_tmp', 'fzyx', dtype)
+ stream_update_rule = create_stream_pull_only_kernel(lb_method.stencil, None, 'pdfs', 'pdfs_tmp', field_layout, dtype)
stream_ast = create_kernel(stream_update_rule, **create_kernel_params)
stream_ast.function_name = 'kernel_stream'
__lattice_model(generation_context, class_name, lb_method, stream_collide_ast, collide_ast, stream_ast,
--
GitLab