diff --git a/src/lbmpy/_compat.py b/src/lbmpy/_compat.py
index 52760cd2440ea470ebf1a6d4470749ef6b89c068..b41cf60cab9b25231b58dac55bf2771ae461185f 100644
--- a/src/lbmpy/_compat.py
+++ b/src/lbmpy/_compat.py
@@ -13,7 +13,13 @@ if PYSTENCILS_VERSION_MAJOR == 2:
return DEFAULTS.spatial_counters[coord]
def get_supported_instruction_sets():
- return []
+ from pystencils import Target
+ vector_targets = Target.available_vector_cpu_targets()
+ isas = []
+ for target in vector_targets:
+ tokens = target.name.split("_")
+ isas.append(tokens[-1].lower())
+ return isas
else:
from pystencils.backends.simd_instruction_sets import (
diff --git a/tests/full_scenarios/schaefer_turek/scenario_schaefer_turek.py b/tests/full_scenarios/schaefer_turek/scenario_schaefer_turek.py
index da0128180aae2b122cc998949fc9f448742407c0..22d1c02ed23bde4b4b578ef489a6d94448db4e9f 100644
--- a/tests/full_scenarios/schaefer_turek/scenario_schaefer_turek.py
+++ b/tests/full_scenarios/schaefer_turek/scenario_schaefer_turek.py
@@ -10,6 +10,8 @@ import warnings
import numpy as np
import pytest
+from pystencils import Target
+
from lbmpy._compat import get_supported_instruction_sets
from lbmpy.boundaries.boundaryconditions import NoSlip
from lbmpy.geometry import get_pipe_velocity_field
diff --git a/tests/test_vectorization.py b/tests/test_vectorization.py
index efda06fb860f49938141595947eb4d02db2dda3a..dcb01a65a7fbe9ecbe297df10cab3f3ccbe3ce09 100644
--- a/tests/test_vectorization.py
+++ b/tests/test_vectorization.py
@@ -1,8 +1,9 @@
import numpy as np
import pytest
+from dataclasses import replace
import pystencils as ps
-from lbmpy._compat import get_supported_instruction_sets
+from lbmpy._compat import get_supported_instruction_sets, IS_PYSTENCILS_2
from lbmpy.scenarios import create_lid_driven_cavity
from lbmpy.creationfunctions import LBMConfig, LBMOptimisation
@@ -11,7 +12,7 @@ if vector_isas is None:
vector_isas = []
-@pytest.mark.skipif(not vector_isas, reason='cannot detect CPU instruction set')
+@pytest.mark.skipif(not vector_isas, reason="cannot detect CPU instruction set")
def test_lbm_vectorization_short():
print("Computing reference solutions")
size1 = (64, 32)
@@ -21,30 +22,45 @@ def test_lbm_vectorization_short():
ldc1_ref.run(10)
lbm_config = LBMConfig(relaxation_rate=relaxation_rate)
- config = ps.CreateKernelConfig(cpu_vectorize_info={'instruction_set': get_supported_instruction_sets()[-1],
- 'assume_aligned': True,
- 'nontemporal': True,
- 'assume_inner_stride_one': True,
- 'assume_sufficient_line_padding': False,
- })
- ldc1 = create_lid_driven_cavity(size1, lbm_config=lbm_config, config=config,
- fixed_loop_sizes=False)
+ config = ps.CreateKernelConfig(
+ cpu_vectorize_info={
+ "instruction_set": get_supported_instruction_sets()[-1],
+ "assume_aligned": True,
+ "nontemporal": True,
+ "assume_inner_stride_one": True,
+ "assume_sufficient_line_padding": False,
+ }
+ )
+ ldc1 = create_lid_driven_cavity(
+ size1, lbm_config=lbm_config, config=config, fixed_loop_sizes=False
+ )
ldc1.run(10)
-@pytest.mark.skipif(not vector_isas, reason='cannot detect CPU instruction set')
-@pytest.mark.parametrize('instruction_set', vector_isas)
-@pytest.mark.parametrize('aligned_and_padding', [[False, False], [True, False], [True, True]])
-@pytest.mark.parametrize('nontemporal', [False, True])
-@pytest.mark.parametrize('double_precision', [False, True])
-@pytest.mark.parametrize('fixed_loop_sizes', [False, True])
+@pytest.mark.skipif(not vector_isas, reason="cannot detect CPU instruction set")
+@pytest.mark.xfail(reason="Loop splitting is not available yet")
+@pytest.mark.parametrize("instruction_set", vector_isas)
+@pytest.mark.parametrize(
+ "aligned_and_padding", [[False, False], [True, False], [True, True]]
+)
+@pytest.mark.parametrize("nontemporal", [False, True])
+@pytest.mark.parametrize("double_precision", [False, True])
+@pytest.mark.parametrize("fixed_loop_sizes", [False, True])
@pytest.mark.longrun
-def test_lbm_vectorization(instruction_set, aligned_and_padding, nontemporal, double_precision, fixed_loop_sizes):
- vectorization_options = {'instruction_set': instruction_set,
- 'assume_aligned': aligned_and_padding[0],
- 'nontemporal': nontemporal,
- 'assume_inner_stride_one': True,
- 'assume_sufficient_line_padding': aligned_and_padding[1]}
+def test_lbm_vectorization(
+ instruction_set,
+ aligned_and_padding,
+ nontemporal,
+ double_precision,
+ fixed_loop_sizes,
+):
+ vectorization_options = {
+ "instruction_set": instruction_set,
+ "assume_aligned": aligned_and_padding[0],
+ "nontemporal": nontemporal,
+ "assume_inner_stride_one": True,
+ "assume_sufficient_line_padding": aligned_and_padding[1],
+ }
time_steps = 100
size1 = (64, 32)
size2 = (666, 34)
@@ -57,20 +73,40 @@ def test_lbm_vectorization(instruction_set, aligned_and_padding, nontemporal, do
ldc2_ref.run(time_steps)
lbm_config = LBMConfig(relaxation_rate=relaxation_rate)
- config = ps.CreateKernelConfig(data_type="float64" if double_precision else "float32",
- default_number_float="float64" if double_precision else "float32",
- cpu_vectorize_info=vectorization_options)
+ config = ps.CreateKernelConfig(
+ data_type="float64" if double_precision else "float32",
+ cpu_vectorize_info=vectorization_options,
+ )
+
+ if not IS_PYSTENCILS_2:
+ config = replace(
+ config,
+ default_number_float="float64" if double_precision else "float32",
+ )
+
lbm_opt_split = LBMOptimisation(cse_global=True, split=True)
lbm_opt = LBMOptimisation(cse_global=True, split=False)
- print(f"Vectorization test, double precision {double_precision}, vectorization {vectorization_options}, "
- f"fixed loop sizes {fixed_loop_sizes}")
- ldc1 = create_lid_driven_cavity(size1, fixed_loop_sizes=fixed_loop_sizes,
- lbm_config=lbm_config, lbm_optimisation=lbm_opt, config=config)
+ print(
+ f"Vectorization test, double precision {double_precision}, vectorization {vectorization_options}, "
+ f"fixed loop sizes {fixed_loop_sizes}"
+ )
+ ldc1 = create_lid_driven_cavity(
+ size1,
+ fixed_loop_sizes=fixed_loop_sizes,
+ lbm_config=lbm_config,
+ lbm_optimisation=lbm_opt,
+ config=config,
+ )
ldc1.run(time_steps)
np.testing.assert_almost_equal(ldc1_ref.velocity[:, :], ldc1.velocity[:, :])
- ldc2 = create_lid_driven_cavity(size2, fixed_loop_sizes=fixed_loop_sizes,
- lbm_config=lbm_config, lbm_optimisation=lbm_opt_split, config=config)
+ ldc2 = create_lid_driven_cavity(
+ size2,
+ fixed_loop_sizes=fixed_loop_sizes,
+ lbm_config=lbm_config,
+ lbm_optimisation=lbm_opt_split,
+ config=config,
+ )
ldc2.run(time_steps)
np.testing.assert_almost_equal(ldc2_ref.velocity[:, :], ldc2.velocity[:, :])