The Python sum function may return a float64 array even if the argume…

…nts are float32, so we need to force the result's dtype. Adding more explicit error messages.
inducer · Nov 22, 2013 · 9b74b8b · 9b74b8b
1 parent e08c7e0
commit 9b74b8b
Show file tree

Hide file tree

Showing 3 changed files with 5 additions and 5 deletions.
diff --git a/hedge/backends/cuda/diff_shared_fld.py b/hedge/backends/cuda/diff_shared_fld.py
@@ -194,7 +194,7 @@ def __call__(self, op_class, field):
 
         block, func = self.get_kernel(op_class, elgroup)
 
-        assert field.dtype == given.float_type
+        assert field.dtype == given.float_type, "Wrong types: %s: %s, %s: %s" % (field, field.dtype, given, given.float_type)
 
         use_debugbuf = set(["cuda_diff", "cuda_debugbuf"]) <= discr.debug
         if use_debugbuf:

diff --git a/hedge/backends/cuda/fluxgather.py b/hedge/backends/cuda/fluxgather.py
@@ -283,7 +283,7 @@ def make_kernel(self, discr, executor, fluxes):
 
 
 
-def make_plan(discr, eg, given, tune_for, dofs_per_face, quadrature_tag, 
+def make_plan(discr, eg, given, tune_for, dofs_per_face, quadrature_tag,
         given_mbs_per_block=None):
     from hedge.backends.cuda.execute import Executor
     if tune_for is not None:
@@ -466,7 +466,7 @@ def __call__(self, eval_dependency, lift_plan):
                 else:
                     dep_field = discr.volume_zeros()
 
-            assert dep_field.dtype == given.float_type
+            assert dep_field.dtype == given.float_type, "Wrong types: %s: %s, %s: %s" % (dep_expr, dep_field.dtype, given, given.float_type)
             dep_field.bind_to_texref_ext(texref_map[dep_expr],
                     allow_double_hack=True)
 

diff --git a/hedge/vector_primitives.py b/hedge/vector_primitives.py
@@ -49,10 +49,10 @@ def __call__(self, *args):
 
 class UnoptimizedLinearCombiner(object):
     def __init__(self, result_dtype, scalar_dtype):
-        self.result_type = result_dtype.type
+        self.result_dtype = result_dtype
 
     def __call__(self, *args):
-        return sum(vec*self.result_type(fac) for fac, vec in args)
+        return sum(vec*self.result_dtype.type(fac) for fac, vec in args).astype(self.result_dtype)
 
 
 class NumpyLinearCombiner(object):