From 7b4424f2a47465ba7fb402f5a974b632bdad7ca1 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Mon, 26 Mar 2012 11:51:25 -0400
Subject: [PATCH] Fix precompute/prefetch bugs that kept the n-body test from
working.
---
loopy/__init__.py | 1 -
loopy/cse.py | 12 +++++++++---
test/test_nbody.py | 4 ++--
test/test_sem.py | 2 ++
4 files changed, 13 insertions(+), 6 deletions(-)
diff --git a/loopy/__init__.py b/loopy/__init__.py
index a35060141..78342b9e9 100644
--- a/loopy/__init__.py
+++ b/loopy/__init__.py
@@ -502,7 +502,6 @@ def add_prefetch(kernel, var_name, sweep_inames=[], dim_arg_names=None,
kernel = extract_subst(kernel, rule_name, uni_template, parameters)
-
if footprint_subscripts is not None:
if not isinstance(footprint_subscripts, (list, tuple)):
footprint_subscripts = [footprint_subscripts]
diff --git a/loopy/cse.py b/loopy/cse.py
index 640a69bb4..54618ed71 100644
--- a/loopy/cse.py
+++ b/loopy/cse.py
@@ -671,12 +671,18 @@ def precompute(kernel, subst_use, dtype, sweep_inames=[],
found = True
break
- if not invdesc.is_in_footprint:
+ if footprint_generators is None:
+ # We only have a right to find the expression if the
+ # invocation descriptors if they were generated by a scan
+ # of the code in the first place. If the user gave us
+ # the footprint generators, that isn't true.
+
+ assert found, expr
+
+ if not found or not invdesc.is_in_footprint:
left_unused_subst_rule_invocations[0] = True
return expr
- assert found, expr
-
else:
# The current subsitution *was* found inside another substitution
# rule. We can't dig up the corresponding invocation descriptor,
diff --git a/test/test_nbody.py b/test/test_nbody.py
index 4f888ef09..f1641256f 100644
--- a/test/test_nbody.py
+++ b/test/test_nbody.py
@@ -1,4 +1,3 @@
-
from __future__ import division
import numpy as np
@@ -38,6 +37,7 @@ def test_nbody(ctx_factory):
return knl, []
def variant_cpu(knl):
+ knl = lp.expand_subst(knl)
knl = lp.split_dimension(knl, "i", 1024,
outer_tag="g.0", slabs=(0,1))
knl = lp.add_prefetch(knl, "x[i,k]", ["k"], default_tag=None)
@@ -56,7 +56,7 @@ def test_nbody(ctx_factory):
n = 3000
- for variant in [variant_gpu]:
+ for variant in [variant_1, variant_cpu, variant_gpu]:
variant_knl, loop_prio = variant(knl)
kernel_gen = lp.generate_loop_schedules(variant_knl,
loop_priority=loop_prio)
diff --git a/test/test_sem.py b/test/test_sem.py
index a83a74c90..8a93d11c0 100644
--- a/test/test_sem.py
+++ b/test/test_sem.py
@@ -10,6 +10,8 @@ from pyopencl.tools import pytest_generate_tests_for_pyopencl \
+1/0 # not ready
+
def test_laplacian(ctx_factory):
1/0 # not adapted to new language
--
GitLab