From c29aa746719862dbc6bc433bc865aa161b77626c Mon Sep 17 00:00:00 2001
From: Dominic Kempf <dominic.kempf@iwr.uni-heidelberg.de>
Date: Mon, 5 Dec 2016 17:44:09 +0100
Subject: [PATCH] Fix gradvec strategy
---
python/dune/perftool/sumfact/basis.py | 40 +++++++++++++++++--------
python/dune/perftool/sumfact/sumfact.py | 18 +++++------
2 files changed, 36 insertions(+), 22 deletions(-)
diff --git a/python/dune/perftool/sumfact/basis.py b/python/dune/perftool/sumfact/basis.py
index 7221e89d..62077833 100644
--- a/python/dune/perftool/sumfact/basis.py
+++ b/python/dune/perftool/sumfact/basis.py
@@ -27,6 +27,7 @@ from dune.perftool.sumfact.quadrature import quadrature_inames
from dune.perftool.loopy.buffer import initialize_buffer
from dune.perftool.pdelab.driver import FEM_name_mangling
from dune.perftool.pdelab.restriction import restricted_name
+from dune.perftool.tools import maybe_wrap_subscript
from pytools import product
@@ -67,6 +68,8 @@ def pymbolic_trialfunction_gradient(element, restriction, component, visitor):
formdata = get_global_context_value('formdata')
dim = formdata.geometric_dimension
buffers = []
+ insn_dep = None
+ ret = False
for i in range(dim):
a_matrices = [theta_matrix] * dim
a_matrices[i] = dtheta_matrix
@@ -86,6 +89,8 @@ def pymbolic_trialfunction_gradient(element, restriction, component, visitor):
).get_temporary(shape=shape,
name=input,
)
+ if insn_dep is None:
+ insn_dep = frozenset({Writes(input)})
# Setup the input!
setup_theta(input, element, restriction, component, index)
@@ -94,22 +99,31 @@ def pymbolic_trialfunction_gradient(element, restriction, component, visitor):
# evaluation of the gradients of basis functions at quadrature
# points (stage 1)
if index:
- var, _ = sum_factorization_kernel(a_matrices,
- buffer,
- 1,
- preferred_position=i,
- insn_dep=frozenset({Writes(input)}),
- output_name=name,
- )
+ assert len(visitor.indices) == 1
+ var, insn_dep = sum_factorization_kernel(a_matrices,
+ buffer,
+ 1,
+ preferred_position=i,
+ insn_dep=insn_dep,
+ )
+ ret = True
else:
- var, _ = sum_factorization_kernel(a_matrices,
- buffer,
- 1,
- preferred_position=i,
- insn_dep=frozenset({Writes(input)}),
- )
+ var, insn_dep = sum_factorization_kernel(a_matrices,
+ buffer,
+ 1,
+ preferred_position=i,
+ insn_dep=insn_dep,
+ )
buffers.append(var)
+ # Check whether we want to return early with something that has the indexing
+ # already handled! This happens with vectorization when the index coincides
+ # with the position in the vector register.
+ if ret:
+ indices = visitor.indices
+ visitor.indices = None
+ return maybe_wrap_subscript(var, tuple(prim.Variable(i) for i in quadrature_inames()) + indices)
+
# TODO this should be quite conditional!!!
for i, buf in enumerate(buffers):
# Write solution from sumfactorization to gradient variable
diff --git a/python/dune/perftool/sumfact/sumfact.py b/python/dune/perftool/sumfact/sumfact.py
index 059488f9..5787913e 100644
--- a/python/dune/perftool/sumfact/sumfact.py
+++ b/python/dune/perftool/sumfact/sumfact.py
@@ -11,6 +11,7 @@ from dune.perftool.generation import (backend,
built_instruction,
domain,
function_mangler,
+ generator_factory,
get_counter,
get_global_context_value,
globalarg,
@@ -120,10 +121,7 @@ def generate_accumulation_instruction(visitor, accterm, measure, subdomain_id):
else:
buffers.append(name_test_function_contribution(accterm.argument))
- # Collect a list of instruction IDs of the contributions
- contribution_ids = []
-
- # TODO covers only 2D
+ insn_dep = None
for i, buf in enumerate(buffers):
# Get the a matrices needed for this accumulation term
rows = basis_functions_per_direction()
@@ -188,14 +186,16 @@ def generate_accumulation_instruction(visitor, accterm, measure, subdomain_id):
tags=frozenset({"quadvec"}),
depends_on=frozenset({deps})
)
- contribution_ids.append(contrib_dep)
+
+ if insn_dep is None:
+ insn_dep = frozenset({contrib_dep})
# Add a sum factorization kernel that implements the multiplication
# with the test function (stage 3)
result, insn_dep = sum_factorization_kernel(a_matrices,
buffer,
3,
- insn_dep=frozenset({contrib_dep}),
+ insn_dep=insn_dep,
additional_inames=frozenset(visitor.inames),
preferred_position=pref_pos,
)
@@ -223,7 +223,7 @@ def generate_accumulation_instruction(visitor, accterm, measure, subdomain_id):
expr = Call(PDELabAccumulationFunction(accum, rank),
(ansatz_lfs.get_args() +
test_lfs.get_args() +
- (Subscript(result, tuple(Variable(i) for i in inames)),)
+ (Subscript(result, tuple(Variable(i) for i in inames) + index),)
)
)
instruction(assignees=(),
@@ -237,7 +237,8 @@ def generate_accumulation_instruction(visitor, accterm, measure, subdomain_id):
transform(nest_quadrature_loops, visitor.inames)
-def sum_factorization_kernel(a_matrices, buf, stage, insn_dep=frozenset({}), additional_inames=frozenset({}), preferred_position=None, output_name=None):
+@generator_factory(item_tags=("sumfactkernel",), context_tags=("kernel",), cache_key_generator=lambda a, b, s, **kw: (a,b,s))
+def sum_factorization_kernel(a_matrices, buf, stage, insn_dep=frozenset({}), additional_inames=frozenset({}), preferred_position=None):
"""
Calculate a sum factorization matrix product.
@@ -340,7 +341,6 @@ def sum_factorization_kernel(a_matrices, buf, stage, insn_dep=frozenset({}), add
out = get_buffer_temporary(buf,
shape=out_shape,
dim_tags=dim_tags,
- name=output_name,
)
silenced_warning('read_no_write({})'.format(out))
--
GitLab