diff --git a/python/dune/perftool/sumfact/accumulation.py b/python/dune/perftool/sumfact/accumulation.py
index 6b1972a201e43320102211691a64292d53057174..d24e100c29725f2beb27aca5fa6483afeb553aeb 100644
--- a/python/dune/perftool/sumfact/accumulation.py
+++ b/python/dune/perftool/sumfact/accumulation.py
@@ -59,12 +59,6 @@ def accum_iname(restriction, bound, i):
return sumfact_iname(bound, "accum")
-def name_test_function_contribution(test):
- count = get_counter('sumfact_contribution_counter')
- grad = "grad_" if test.reference_grad else ""
- return restricted_name("contrib_{}phi_{}".format(grad, str(count)), test.restriction)
-
-
@backend(interface="accum_insn", name="sumfact")
def generate_accumulation_instruction(visitor, accterm, measure, subdomain_id):
# When doing sumfactorization we want to split the test function
@@ -148,7 +142,7 @@ def generate_accumulation_instruction(visitor, accterm, measure, subdomain_id):
temp = get_buffer_temporary(buffer,
shape=vsf.quadrature_shape,
dim_tags=vsf.quadrature_dimtags,
- name=vsf.input,
+ name="input_{}".format(buffer),
)
# Those input fields, that are padded need to be set to zero
diff --git a/python/dune/perftool/sumfact/realization.py b/python/dune/perftool/sumfact/realization.py
index de5e61f055683fbf081b3f2225004dda6189aa5f..4ff4dfb4c4d4873641a67090d637fcd7a912252f 100644
--- a/python/dune/perftool/sumfact/realization.py
+++ b/python/dune/perftool/sumfact/realization.py
@@ -64,12 +64,11 @@ def _realize_sum_factorization_kernel(sf):
# Set up the input for stage 1
if sf.stage == 1 and not get_option("fastdg"):
assert sf.coeff_func
- assert sf.input is not None
# Get the input temporary!
input_setup = get_buffer_temporary(sf.buffer,
shape=sf.flat_input_shape,
- name=sf.input
+ name="input_{}".format(sf.buffer)
)
def _write_input(inputsf, index=0):
@@ -113,7 +112,7 @@ def _realize_sum_factorization_kernel(sf):
else:
_write_input(sf)
- insn_dep = insn_dep.union(frozenset({lp.match.Writes(sf.input)}))
+ insn_dep = insn_dep.union(frozenset({lp.match.Writes("input_{}".format(sf.buffer))}))
# Construct the direct_input for the FastDG case
direct_input = None
diff --git a/python/dune/perftool/sumfact/symbolic.py b/python/dune/perftool/sumfact/symbolic.py
index e4188107250743b123baf0a14417dea86e95a0b2..b254d5c66904566353c55c49ab23d5dbb9b2292b 100644
--- a/python/dune/perftool/sumfact/symbolic.py
+++ b/python/dune/perftool/sumfact/symbolic.py
@@ -26,7 +26,6 @@ class SumfactKernel(SumfactKernelBase, ImmutableRecord, prim.Variable):
preferred_position=None,
restriction=0,
within_inames=(),
- input=None,
insn_dep=frozenset(),
coeff_func=None,
coeff_func_index=None,
@@ -94,7 +93,6 @@ class SumfactKernel(SumfactKernelBase, ImmutableRecord, prim.Variable):
preferred_position: Will be used in the dry run to order kernels
when doing vectorization e.g. (dx u,dy u,dz u, u).
restriction: Restriction for faces values.
- input: The name to use for the input temporary
padding: a set of slots in the input temporary to be padded
index: The slot in the input temporary dedicated to this kernel
coeff_func: The function to call to get the input coefficient
@@ -291,7 +289,6 @@ class VectorizedSumfactKernel(SumfactKernelBase, ImmutableRecord, prim.Variable)
horizontal_width=1,
vertical_width=1,
buffer=None,
- input=None,
insn_dep=frozenset(),
):
# Assert the input data structure
@@ -321,7 +318,6 @@ class VectorizedSumfactKernel(SumfactKernelBase, ImmutableRecord, prim.Variable)
kernels=kernels,
horizontal_width=horizontal_width,
buffer=buffer,
- input=input,
insn_dep=insn_dep,
vertical_width=vertical_width,
)
@@ -329,14 +325,14 @@ class VectorizedSumfactKernel(SumfactKernelBase, ImmutableRecord, prim.Variable)
prim.Variable.__init__(self, "VecSUMFAC")
def __getinitargs__(self):
- return (self.kernels, self.horizontal_width, self.vertical_width, self.buffer, self.input, self.insn_dep)
+ return (self.kernels, self.horizontal_width, self.vertical_width, self.buffer, self.insn_dep)
def stringifier(self):
return lp.symbolic.StringifyMapper
mapper_method = "map_vectorized_sumfact_kernel"
- init_arg_names = ("kernels", "horizontal_width", "vertical_width", "buffer", "input", "insn_dep")
+ init_arg_names = ("kernels", "horizontal_width", "vertical_width", "buffer", "insn_dep")
#
# Some cache key definitions
diff --git a/python/dune/perftool/sumfact/vectorization.py b/python/dune/perftool/sumfact/vectorization.py
index 6900885499385a62fc9dbca9d1e9adac7e444551..bdaa5bd76b93666b2d2fb4cdf87fb80a516c0f1f 100644
--- a/python/dune/perftool/sumfact/vectorization.py
+++ b/python/dune/perftool/sumfact/vectorization.py
@@ -36,8 +36,7 @@ def attach_vectorization_info(sf):
def no_vec(sf):
- return sf.copy(buffer=get_counted_variable("buffer"),
- input=get_counted_variable("input"))
+ return sf.copy(buffer=get_counted_variable("buffer"))
def no_vectorization(sumfacts):
@@ -75,10 +74,8 @@ def vertical_vectorization_strategy(sumfact, depth):
if isinstance(sumfact, SumfactKernel):
buffer = get_counted_variable("vertical_buffer")
- input = get_counted_variable("vertical_input")
result[sumfact] = VectorizedSumfactKernel(kernels=tuple(newkernels),
buffer=buffer,
- input=input,
vertical_width=depth,
)
else:
@@ -118,14 +115,12 @@ def horizontal_vectorization_strategy(sumfacts, width, allow_padding=1):
kernels = tuple(kernels)
buffer = get_counted_variable("joined_buffer")
- input = get_counted_variable("joined_input")
for sumf in kernels:
if len(kernels) in range(width - allow_padding, width + 1):
result[sumf] = VectorizedSumfactKernel(kernels=kernels,
horizontal_width=width,
buffer=buffer,
- input=input,
)
return result