From fafe98aff669158e5d175ed169c32586ea588135 Mon Sep 17 00:00:00 2001
From: Dominic Kempf <dominic.kempf@iwr.uni-heidelberg.de>
Date: Wed, 21 Dec 2016 11:36:58 +0100
Subject: [PATCH] Detect faces in shape generation regardless of the size being
1
---
python/dune/perftool/sumfact/basis.py | 3 ++-
python/dune/perftool/sumfact/sumfact.py | 6 ++++--
2 files changed, 6 insertions(+), 3 deletions(-)
diff --git a/python/dune/perftool/sumfact/basis.py b/python/dune/perftool/sumfact/basis.py
index 46a639fb..86a9b354 100644
--- a/python/dune/perftool/sumfact/basis.py
+++ b/python/dune/perftool/sumfact/basis.py
@@ -108,6 +108,7 @@ def pymbolic_trialfunction_gradient(element, restriction, component, visitor):
preferred_position=i,
insn_dep=insn_dep,
restriction=restriction,
+ outshape=tuple(mat.rows for mat in a_matrices if mat.face is None),
direct_input=direct_input,
)
buffers.append(var)
@@ -174,7 +175,7 @@ def pymbolic_trialfunction(element, restriction, component, visitor):
1,
preferred_position=None,
insn_dep=frozenset({Writes(inp)}),
- outshape=tuple(mat.rows for mat in a_matrices if mat.rows != 1),
+ outshape=tuple(mat.rows for mat in a_matrices if mat.face is None),
restriction=restriction,
direct_input=direct_input,
)
diff --git a/python/dune/perftool/sumfact/sumfact.py b/python/dune/perftool/sumfact/sumfact.py
index e135fb75..2429da4e 100644
--- a/python/dune/perftool/sumfact/sumfact.py
+++ b/python/dune/perftool/sumfact/sumfact.py
@@ -144,7 +144,7 @@ def generate_accumulation_instruction(visitor, accterm, measure, subdomain_id):
a_matrices, buf, inp, index, padding = get_vectorization_info(a_matrices, (accterm.argument.restriction, restriction))
# Initialize a base storage for this buffer and get a temporay pointing to it
- shape = tuple(mat.cols for mat in a_matrices if mat.cols != 1)
+ shape = tuple(mat.cols for mat in a_matrices if mat.face is None)
dim_tags = ",".join(['f'] * local_dimension())
if index is not None:
shape = shape + (4,)
@@ -434,7 +434,9 @@ def sum_factorization_kernel(a_matrices,
})
if outshape is None:
- outshape = tuple(mat.rows for mat in a_matrices if mat.rows != 1)
+ assert stage == 3
+ outshape = tuple(mat.rows for mat in a_matrices)
+
dim_tags = ",".join(['f'] * len(outshape))
if next(iter(a_matrices)).vectorized:
--
GitLab