Add test balloon for custom type array kernels

inducer · Aug 17, 2016 · e499c28 · e499c28
1 parent 40dda54
commit e499c28
Show file tree

Hide file tree

Showing 3 changed files with 34 additions and 1 deletion.
diff --git a/pyopencl/elementwise.py b/pyopencl/elementwise.py
@@ -36,7 +36,7 @@
 import pyopencl as cl
 from pytools import memoize_method
 from pyopencl.tools import (dtype_to_ctype, VectorArg, ScalarArg,
-        KernelTemplateBase)
+        KernelTemplateBase, dtype_to_c_struct)
 
 
 # {{{ elementwise kernel code generator
@@ -740,6 +740,7 @@ def get_fill_kernel(context, dtype):
                 "tp": dtype_to_ctype(dtype),
                 },
             "z[i] = a",
+            preamble=dtype_to_c_struct(context.devices[0], dtype),
             name="fill")
 
 

diff --git a/pyopencl/tools.py b/pyopencl/tools.py
@@ -646,6 +646,9 @@ def calc_field_type():
 
 @memoize
 def dtype_to_c_struct(device, dtype):
+    if dtype.fields is None:
+        return ""
+
     matched_dtype, c_decl = match_dtype_to_c_struct(
             device, dtype_to_ctype(dtype), dtype)
 

diff --git a/test/test_array.py b/test/test_array.py
@@ -260,6 +260,35 @@ def test_custom_type_zeros(ctx_factory):
 
     assert np.array_equal(np.zeros(n, dtype), z)
 
+
+def test_custom_type_fill(ctx_factory):
+    context = ctx_factory()
+    queue = cl.CommandQueue(context)
+
+    from pyopencl.characterize import has_struct_arg_count_bug
+    if has_struct_arg_count_bug(queue.device):
+        pytest.skip("device has LLVM arg counting bug")
+
+    dtype = np.dtype([
+        ("cur_min", np.int32),
+        ("cur_max", np.int32),
+        ("pad", np.int32),
+        ])
+
+    from pyopencl.tools import get_or_register_dtype, match_dtype_to_c_struct
+
+    name = "mmc_type"
+    dtype, c_decl = match_dtype_to_c_struct(queue.device, name, dtype)
+    dtype = get_or_register_dtype(name, dtype)
+
+    n = 1000
+    z_dev = cl.array.empty(queue, n, dtype=dtype)
+    z_dev.fill(np.zeros((), dtype))
+
+    z = z_dev.get()
+
+    assert np.array_equal(np.zeros(n, dtype), z)
+
 # }}}