diff --git a/examples/dagrt-fusion.py b/examples/dagrt-fusion.py
index 85ddda7db414d314022817bcd2232a4a79bdc46b..2a285ad533b0d1ffb545da65f69efc8665072a10 100755
--- a/examples/dagrt-fusion.py
+++ b/examples/dagrt-fusion.py
@@ -448,7 +448,6 @@ def _get_source_term(dims):
def get_wave_op_with_discr(actx, dims=2, order=4):
from meshmode.mesh.generation import generate_regular_rect_mesh
-
mesh = generate_regular_rect_mesh(
a=(-0.5,)*dims,
b=(0.5,)*dims,
@@ -949,10 +948,7 @@ def test_stepper_timing(ctx_factory, use_fusion):
properties=cl.command_queue_properties.PROFILING_ENABLE)
actx = PyOpenCLArrayContext(queue)
- if os.environ.get("GITHUB_ACTIONS") == "true":
- dims = 2
- else:
- dims = 3
+ dims = 3
sym_operator, discr = get_wave_op_with_discr(
actx, dims=dims, order=3)
@@ -1192,26 +1188,27 @@ def scalar_assignment_percent_of_total_mem_ops_table():
actx = PyOpenCLArrayContext(queue)
result2d = mem_ops_results(actx, 2)
- do3d = os.environ.get("GITHUB_ACTIONS") != "true"
- if do3d:
- result3d = mem_ops_results(actx, 3)
+ result3d = mem_ops_results(actx, 3)
with open_output_file("scalar-assignments-mem-op-percentage.tex") as outf:
if not PAPER_OUTPUT:
print("==== Scalar Assigment % of Total Mem Ops ====", file=outf)
- rows = (
- ("2D: Baseline",
- "%.1f" % (
- 100 * result2d["nonfused_bytes_total_by_scalar_assignments"]
- / result2d["nonfused_bytes_total"])),
- ("2D: Inlined",
- "%.1f" % (
- 100 * result2d["fused_bytes_total_by_scalar_assignments"]
- / result2d["fused_bytes_total"])),
- )
- if do3d:
- rows = rows + (
+ print(
+ table(
+ "lr",
+ ("Operator",
+ r"\parbox{1in}{\centering \% Memory Ops. "
+ r"Due to Scalar Assignments}"),
+ (
+ ("2D: Baseline",
+ "%.1f" % (
+ 100 * result2d["nonfused_bytes_total_by_scalar_assignments"]
+ / result2d["nonfused_bytes_total"])),
+ ("2D: Inlined",
+ "%.1f" % (
+ 100 * result2d["fused_bytes_total_by_scalar_assignments"]
+ / result2d["fused_bytes_total"])),
("3D: Baseline",
"%.1f" % (
100 * result3d["nonfused_bytes_total_by_scalar_assignments"]
@@ -1220,15 +1217,7 @@ def scalar_assignment_percent_of_total_mem_ops_table():
"%.1f" % (
100 * result3d["fused_bytes_total_by_scalar_assignments"]
/ result3d["fused_bytes_total"])),
- )
- print(
- table(
- "lr",
- ("Operator",
- r"\parbox{1in}{\centering \% Memory Ops. "
- r"Due to Scalar Assignments}"),
- rows
- ),
+ )),
file=outf)
logger.info("Wrote '%s'", outf.name)
@@ -1239,37 +1228,39 @@ def scalar_assignment_effect_of_fusion_mem_ops_table():
queue = cl.CommandQueue(cl_ctx)
result2d = mem_ops_results(queue, 2)
- do3d = os.environ.get("GITHUB_ACTIONS") != "true"
- if do3d:
- result3d = mem_ops_results(queue, 3)
+ result3d = mem_ops_results(queue, 3)
with open_output_file("scalar-assignments-fusion-impact.tex") as outf:
if not PAPER_OUTPUT:
print("==== Scalar Assigment Inlining Impact ====", file=outf)
- rows = (
- ("2D: Baseline",
- r"\num{%d}" % (
- result2d["nonfused_bytes_read_by_scalar_assignments"]),
- r"\num{%d}" % (
- result2d["nonfused_bytes_written_by_scalar_assignments"]),
- r"\num{%d}" % (
- result2d["nonfused_bytes_total_by_scalar_assignments"]),
- "100"),
- ("2D: Inlined",
- r"\num{%d}" % (
- result2d["fused_bytes_read_by_scalar_assignments"]),
- r"\num{%d}" % (
- result2d["fused_bytes_written_by_scalar_assignments"]),
- r"\num{%d}" % (
- result2d["fused_bytes_total_by_scalar_assignments"]),
- r"%.1f" % (
- 100 * result2d["fused_bytes_total_by_scalar_assignments"]
- / result2d["nonfused_bytes_total_by_scalar_assignments"])),
- )
-
- if do3d:
- rows = rows + (
+ print(
+ table(
+ "lrrrr",
+ ("Operator",
+ r"Bytes Read",
+ r"Bytes Written",
+ r"Total",
+ r"\% of Baseline"),
+ (
+ ("2D: Baseline",
+ r"\num{%d}" % (
+ result2d["nonfused_bytes_read_by_scalar_assignments"]),
+ r"\num{%d}" % (
+ result2d["nonfused_bytes_written_by_scalar_assignments"]),
+ r"\num{%d}" % (
+ result2d["nonfused_bytes_total_by_scalar_assignments"]),
+ "100"),
+ ("2D: Inlined",
+ r"\num{%d}" % (
+ result2d["fused_bytes_read_by_scalar_assignments"]),
+ r"\num{%d}" % (
+ result2d["fused_bytes_written_by_scalar_assignments"]),
+ r"\num{%d}" % (
+ result2d["fused_bytes_total_by_scalar_assignments"]),
+ r"%.1f" % (
+ 100 * result2d["fused_bytes_total_by_scalar_assignments"]
+ / result2d["nonfused_bytes_total_by_scalar_assignments"])),
("3D: Baseline",
r"\num{%d}" % (
result3d["nonfused_bytes_read_by_scalar_assignments"]),
@@ -1288,16 +1279,7 @@ def scalar_assignment_effect_of_fusion_mem_ops_table():
r"%.1f" % (
100 * result3d["fused_bytes_total_by_scalar_assignments"]
/ result3d["nonfused_bytes_total_by_scalar_assignments"])),
- )
- print(
- table(
- "lrrrr",
- ("Operator",
- r"Bytes Read",
- r"Bytes Written",
- r"Total",
- r"\% of Baseline"),
- rows),
+ )),
file=outf)
logger.info("Wrote '%s'", outf.name)
diff --git a/grudge/execution.py b/grudge/execution.py
index 3afe569b7bc1330f82b673b1fb5ced76ac085455..102fa108a94e7c3ec0f0999ce46ac02f10da4b10 100644
--- a/grudge/execution.py
+++ b/grudge/execution.py
@@ -525,19 +525,14 @@ class ExecutionMapper(mappers.Evaluator,
if in_grp.nelements == 0:
continue
- # Cache operator
- cache_key = "diff_batch", in_grp, out_grp, tuple(insn.operators),\
- field.dtype
- try:
- matrices_ary_dev = self.bound_op.operator_data_cache[cache_key]
- except KeyError:
- matrices = repr_op.matrices(out_grp, in_grp)
- matrices_ary = np.empty((
- noperators, out_grp.nunit_dofs, in_grp.nunit_dofs))
- for i, op in enumerate(insn.operators):
- matrices_ary[i] = matrices[op.rst_axis]
- matrices_ary_dev = self.array_context.from_numpy(matrices_ary)
- self.bound_op.operator_data_cache[cache_key] = matrices_ary_dev
+ matrices = repr_op.matrices(out_grp, in_grp)
+
+ # FIXME: Should transfer matrices to device and cache them
+ matrices_ary = np.empty((
+ noperators, out_grp.nunit_dofs, in_grp.nunit_dofs))
+ for i, op in enumerate(insn.operators):
+ matrices_ary[i] = matrices[op.rst_axis]
+ matrices_ary_dev = self.array_context.from_numpy(matrices_ary)
self.array_context.call_loopy(
prg(noperators),
diff --git a/test/test_grudge.py b/test/test_grudge.py
index 755ae3040cce1f695640e35262621e60ac0bef27..9fd6ca3d1cc7935ef25aa4db0b180ef16fe3920b 100644
--- a/test/test_grudge.py
+++ b/test/test_grudge.py
@@ -805,10 +805,6 @@ def test_convergence_advec(actx_factory, mesh_name, mesh_pars, op_type, flux_typ
def test_convergence_maxwell(actx_factory, order):
"""Test whether 3D Maxwell's actually converges"""
- import os
- if os.environ.get("GITHUB_ACTIONS") == "true" and order >= 4:
- pytest.skip("not enough memory on Github CI")
-
actx = actx_factory()
from pytools.convergence import EOCRecorder