Hello. During work on packaging new version of PyCUDA and PyOpenCL I have discovered that test_array gives error when testing new reduction code: ============================= test session starts ============================== platform linux2 -- Python 2.6.6 -- pytest-1.3.4 test path 1: test_array.py test_array.py .........................F. =================================== FAILURES =================================== test_scan[ctx_getter=<context getter for <pyopencl.Device 'ION' at 0x19c3a40>>] ctx_getter = <pyopencl.tools.ContextGetter instance at 0x1a2ea28> @pytools.test.mark_test.opencl def test_scan(ctx_getter): context = ctx_getter() queue = cl.CommandQueue(context) from pyopencl.scan import InclusiveScanKernel, ExclusiveScanKernel dtype = np.int32 for cls in [ExclusiveScanKernel, InclusiveScanKernel]: knl = cls(context, dtype, "a+b", "0") for n in [ 10, 2**10-5, 2**10, 2**20-2**18, 2**20-2**18+5, 2**10+5, 2**20+5, 2**20, 2**24 ]: host_data = np.random.randint(0, 10, n).astype(dtype) test_array.py:578: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ def to_device(*args, **kwargs): """Converts a numpy array to a :class:`Array`.""" def _to_device(queue, ary, allocator=None, async=False): result = Array(queue, ary.shape, ary.dtype, allocator, strides=ary.strides) result.set(ary, async=async) return result if isinstance(args[0], cl.Context): from warnings import warn warn("Passing a context as first argument is deprecated. " "This will be continue to be accepted througout " "versions 2011.x of PyOpenCL.", DeprecationWarning, 2) args = args[1:] /usr/lib/python2.6/dist-packages/pyopencl/array.py:717: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ queue = <pyopencl._cl.CommandQueue object at 0x1a3c208> ary = array([9, 8, 7, ..., 1, 7, 7], dtype=int32), allocator = None async = False def _to_device(queue, ary, allocator=None, async=False): result = Array(queue, ary.shape, ary.dtype, allocator, strides=ary.strides) /usr/lib/python2.6/dist-packages/pyopencl/array.py:705: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ self = array([ 0, 8, 8, ..., 57645970, 57645973, 57645979], dtype=int32) ary = array([9, 8, 7, ..., 1, 7, 7], dtype=int32), queue = None, async = False def set(self, ary, queue=None, async=False): assert ary.size == self.size assert ary.dtype == self.dtype assert self.flags.forc if not ary.flags.forc: if async: raise RuntimeError("cannot asynchronously set from " "non-contiguous array") ary = ary.copy() if self.size: cl.enqueue_copy(queue or self.queue, self.data, ary, /usr/lib/python2.6/dist-packages/pyopencl/array.py:360: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ queue = <pyopencl._cl.CommandQueue object at 0x1a3c208> dest = <pyopencl._cl.Buffer object at 0x2650910> src = array([9, 8, 7, ..., 1, 7, 7], dtype=int32) def enqueue_copy(queue, dest, src, **kwargs): if isinstance(dest, Buffer): if isinstance(src, Buffer): if "src_origin" in kwargs: return _cl._enqueue_copy_buffer_rect(queue, src, dest, **kwargs) else: kwargs["dst_offset"] = kwargs.pop("dest_offset", 0) return _cl._enqueue_copy_buffer(queue, src, dest, **kwargs) elif isinstance(src, Image): return _cl._enqueue_copy_image_to_buffer(queue, src, dest, **kwargs) else: # assume from-host if "buffer_origin" in kwargs: return _cl._enqueue_write_buffer_rect(queue, dest, src, **kwargs) else: **kwargs) E MemoryError: clEnqueueWriteBuffer failed: mem object allocation failure /usr/lib/python2.6/dist-packages/pyopencl/__init__.py:525: MemoryError ------------------------------- Captured stderr -------------------------------- /usr/lib/python2.6/dist-packages/pyopencl/__init__.py:163: UserWarning: Build succeeded, but resulted in non-empty logs: Build on <pyopencl.Device 'ION' at 0x19c3a40> succeeded, but said: : Considering profile 'compute_11' for gpu='sm_11' in 'cuModuleLoadDataEx_9c' warn("Build succeeded, but resulted in non-empty logs:\n"+message) /usr/lib/python2.6/dist-packages/pyopencl/__init__.py:163: UserWarning: Build succeeded, but resulted in non-empty logs: Build on <pyopencl.Device 'ION' at 0x19c3a40> succeeded, but said: : Considering profile 'compute_11' for gpu='sm_11' in 'cuModuleLoadDataEx_9d' warn("Build succeeded, but resulted in non-empty logs:\n"+message) /usr/lib/python2.6/dist-packages/pyopencl/cache.py:336: UserWarning: Build succeeded, but resulted in non-empty logs: Build on <pyopencl.Device 'ION' at 0x19c3a40> succeeded, but said: : Considering profile 'compute_11' for gpu='sm_11' in 'cuModuleLoadDataEx_9c' warn("Build succeeded, but resulted in non-empty logs:\n"+message) /usr/lib/python2.6/dist-packages/pyopencl/__init__.py:163: UserWarning: Build succeeded, but resulted in non-empty logs: Build on <pyopencl.Device 'ION' at 0x19c3a40> succeeded, but said: : Considering profile 'compute_11' for gpu='sm_11' in 'cuModuleLoadDataEx_9e' warn("Build succeeded, but resulted in non-empty logs:\n"+message) /usr/lib/python2.6/dist-packages/pyopencl/__init__.py:163: UserWarning: Build succeeded, but resulted in non-empty logs: Build on <pyopencl.Device 'ION' at 0x19c3a40> succeeded, but said: : Considering profile 'compute_11' for gpu='sm_11' in 'cuModuleLoadDataEx_9f' warn("Build succeeded, but resulted in non-empty logs:\n"+message) ===================== 1 failed, 26 passed in 56.81 seconds ===================== Error is caused by lack of memory on my ION. After forcing garbage collection (attached patch) all tests run without problems: ============================= test session starts ============================== platform linux2 -- Python 2.6.6 -- pytest-1.3.4 test path 1: test_array.py test_array.py ........................... ========================== 27 passed in 20.04 seconds ========================== Regards -- Tomasz Rybak <bogomips(a)post.pl&gt; GPG/PGP key ID: 2AD5 9860 Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860 http://member.acm.org/~tomaszrybak