diff --git a/python/dune/perftool/generation/cache.py b/python/dune/perftool/generation/cache.py
index 58b67988800a8be0cd2200ce340862b81b150764..e4c3fb592dd3eb4e32a51e44fc377d49e969aa10 100644
--- a/python/dune/perftool/generation/cache.py
+++ b/python/dune/perftool/generation/cache.py
@@ -119,6 +119,9 @@ class _RegisteredFunction(object):
# Return the result for immediate usage
return self._get_content(cache_key)
+ def remove_by_value(self, val):
+ self._memoize_cache = {k:v for k, v in self._memoize_cache.items() if v != val}
+
def generator_factory(**factory_kwargs):
""" A function decorator factory
diff --git a/python/dune/perftool/pdelab/localoperator.py b/python/dune/perftool/pdelab/localoperator.py
index c1455f35f71a8cb82103df1662870cc39b4992a3..8610952bbcf299f948b50f5227fde362f4601119 100644
--- a/python/dune/perftool/pdelab/localoperator.py
+++ b/python/dune/perftool/pdelab/localoperator.py
@@ -488,10 +488,8 @@ def generate_kernel(integrals):
def extract_kernel_from_cache(tag):
# Preprocess some instruction!
- from dune.perftool.sumfact.sumfact import expand_sumfact_kernels, filter_sumfact_instructions
- instructions = [i for i in retrieve_cache_items("{} and instruction".format(tag))]
- expand_sumfact_kernels(instructions)
- filter_sumfact_instructions()
+ from dune.perftool.sumfact.sumfact import expand_sumfact_kernels
+ expand_sumfact_kernels(tag)
# Now extract regular loopy kernel components
from dune.perftool.loopy.target import DuneTarget
diff --git a/python/dune/perftool/sumfact/sumfact.py b/python/dune/perftool/sumfact/sumfact.py
index e3042690f55007928716e0baeb806fa4e0ba9030..c6bc625490cb24361ed198a84da3c978eee652b2 100644
--- a/python/dune/perftool/sumfact/sumfact.py
+++ b/python/dune/perftool/sumfact/sumfact.py
@@ -16,6 +16,7 @@ from dune.perftool.generation import (backend,
globalarg,
iname,
instruction,
+ retrieve_cache_items,
silenced_warning,
temporary_variable,
transform,
@@ -139,6 +140,12 @@ def default_resolution(insns):
depends_on=frozenset(*deps)
)
)
+ if isinstance(insn, lp.Assignment):
+ from dune.perftool.generation.loopy import expr_instruction_impl
+ expr_instruction_impl.remove_by_value(insn)
+ if isinstance(insn, lp.CallInstruction):
+ from dune.perftool.generation.loopy import call_instruction_impl
+ call_instruction_impl.remove_by_value(insn)
def apply_sumfact_grad_vectorization(insns, stage):
@@ -153,7 +160,7 @@ def apply_sumfact_grad_vectorization(insns, stage):
# Now apply some heuristics when to vectorize...
if len(set(sumfact_kernels)) < 3:
- default_resolution(insns)
+ return
else:
# Vectorize!!!
sumfact_kernels = sorted(sumfact_kernels, key=lambda s: s.preferred_interleaving_position)
@@ -167,6 +174,7 @@ def apply_sumfact_grad_vectorization(insns, stage):
# Maybe initialize the input buffer
if sumfact_kernels[0].setup_method:
+ assert stage == 1
shape = product(mat.cols for mat in sumfact_kernels[0].a_matrices)
shape = (shape, 4)
initialize_buffer(buffer, base_storage_size=4 * shape[0])
@@ -178,6 +186,7 @@ def apply_sumfact_grad_vectorization(insns, stage):
func, args = sumf.setup_method
insn_dep = insn_dep.union({func(inp, *args, additional_indices=(i,))})
else:
+ assert stage == 3
# No setup method defined. We need to make sure the input is correctly setup
shape = tuple(mat.cols for mat in sumfact_kernels[0].a_matrices) + (4,)
initialize_buffer(buffer, base_storage_size=4 * shape[0])
@@ -189,9 +198,11 @@ def apply_sumfact_grad_vectorization(insns, stage):
for insn in insns:
if isinstance(insn, lp.Assignment):
if get_pymbolic_basename(insn.assignee) == sumf.input_temporary:
- built_instruction(insn.copy(assignee=prim.Subscript(prim.Variable(inp), insn.assignee.index + (i,))))
+ built_instruction(insn.copy(assignee=prim.Subscript(prim.Variable(inp), insn.assignee.index + (i,)),
+ id=insn.id + "__{}".format(i)))
+ insn_dep = insn_dep.union(insn.depends_on)
from dune.perftool.generation.loopy import expr_instruction_impl
- expr_instruction_impl._memoize_cache = {k: v for k, v in expr_instruction_impl._memoize_cache.items() if v.id != insn.id}
+ expr_instruction_impl.remove_by_value(insn)
# Determine the joined AMatrix
large_a_matrices = []
@@ -225,19 +236,19 @@ def apply_sumfact_grad_vectorization(insns, stage):
depends_on=dep,
))
-def expand_sumfact_kernels(insns):
- if get_option("vectorize_grads"):
- apply_sumfact_grad_vectorization(insns, 1)
- apply_sumfact_grad_vectorization(insns, 3)
- else:
- default_resolution(insns)
+ if isinstance(insn, lp.Assignment):
+ from dune.perftool.generation.loopy import expr_instruction_impl
+ expr_instruction_impl.remove_by_value(insn)
+ if isinstance(insn, lp.CallInstruction):
+ from dune.perftool.generation.loopy import call_instruction_impl
+ call_instruction_impl.remove_by_value(insn)
-def filter_sumfact_instructions():
- """ Remove all instructions that contain a SumfactKernel node """
- from dune.perftool.generation.loopy import expr_instruction_impl, call_instruction_impl
- expr_instruction_impl._memoize_cache = {k: v for k, v in expr_instruction_impl._memoize_cache.items() if not find_sumfact(v.expression)}
- call_instruction_impl._memoize_cache = {k: v for k, v in call_instruction_impl._memoize_cache.items() if not find_sumfact(v.expression)}
+def expand_sumfact_kernels(tag):
+ if get_option("vectorize_grads"):
+ apply_sumfact_grad_vectorization([i for i in retrieve_cache_items("{} and instruction".format(tag))], 1)
+ apply_sumfact_grad_vectorization([i for i in retrieve_cache_items("{} and instruction".format(tag))], 3)
+ default_resolution(retrieve_cache_items("{} and instruction".format(tag)))
@iname
@@ -389,7 +400,6 @@ def generate_accumulation_instruction(visitor, accterm, measure, subdomain_id):
(Subscript(result, tuple(Variable(i) for i in inames)),)
)
)
-
instruction(assignees=(),
expression=expr,
forced_iname_deps=frozenset(inames + visitor.inames),