IntelPython · oleksandr-pavlyk · Sep 5, 2021 · Sep 3, 2021 · Sep 3, 2021 · Sep 5, 2021
@@ -78,7 +78,7 @@ cdef class _SyclQueueManager:
         Returns:
             backend_type: The SYCL backend for the currently selected queue.
         """
-        return self.get_current_queue().get_sycl_backend()
+        return self.get_current_queue().backend
 
     cpdef get_current_device_type(self):
         """
@@ -88,7 +88,7 @@ cdef class _SyclQueueManager:
             device_type: The SYCL device type for the currently selected queue.
             Possible values can be gpu, cpu, accelerator, or host.
         """
-        return self.get_current_queue().get_sycl_device().device_type
+        return self.get_current_queue().sycl_device.device_type
 
     cpdef SyclQueue get_current_queue(self):
         """

@@ -34,6 +34,7 @@ from dpctl._backend cimport (  # noqa: E211
     DPCTLaligned_alloc_host,
     DPCTLaligned_alloc_shared,
     DPCTLContext_Delete,
+    DPCTLDevice_Copy,
     DPCTLEvent_Delete,
     DPCTLEvent_Wait,
     DPCTLfree_with_queue,
@@ -48,6 +49,7 @@ from dpctl._backend cimport (  # noqa: E211
     DPCTLSyclContextRef,
     DPCTLSyclDeviceRef,
     DPCTLSyclEventRef,
+    DPCTLSyclQueueRef,
     DPCTLSyclUSMRef,
     DPCTLUSM_GetPointerDevice,
     DPCTLUSM_GetPointerType,
@@ -138,7 +140,7 @@ cdef class _Memory:
 
     cdef _cinit_alloc(self, Py_ssize_t alignment, Py_ssize_t nbytes,
                       bytes ptr_type, SyclQueue queue):
-        cdef DPCTLSyclUSMRef p
+        cdef DPCTLSyclUSMRef p = NULL
 
         self._cinit_empty()
 
@@ -215,10 +217,12 @@ cdef class _Memory:
             )
 
     def __dealloc__(self):
-        if (self.refobj is None and self.memory_ptr):
-            DPCTLfree_with_queue(
-                self.memory_ptr, self.queue.get_queue_ref()
-            )
+        if (self.refobj is None):
+            if self.memory_ptr:
+                if (type(self.queue) is SyclQueue):
+                    DPCTLfree_with_queue(
+                        self.memory_ptr, self.queue.get_queue_ref()
+                    )
         self._cinit_empty()
 
     cdef _getbuffer(self, Py_buffer *buffer, int flags):
@@ -267,7 +271,7 @@ cdef class _Memory:
     property _queue:
         """
         :class:`dpctl.SyclQueue` with :class:`dpctl.SyclContext` the
-        USM pointer is bound to and :class:`dpctl.SyclDevice` it was
+        USM allocation is bound to and :class:`dpctl.SyclDevice` it was
         allocated on.
         """
         def __get__(self):
@@ -477,8 +481,10 @@ cdef class _Memory:
         cdef DPCTLSyclDeviceRef dref = DPCTLUSM_GetPointerDevice(
             p, ctx.get_context_ref()
         )
-
-        return SyclDevice._create(dref)
+        cdef DPCTLSyclDeviceRef dref_copy = DPCTLDevice_Copy(dref)
+        if (dref_copy is NULL):
+            raise RuntimeError("Could not create a copy of sycl device")
+        return SyclDevice._create(dref_copy)  # deletes the argument
 
     @staticmethod
     cdef bytes get_pointer_type(DPCTLSyclUSMRef p, SyclContext ctx):

@@ -54,7 +54,7 @@ cdef DPCTLSyclQueueRef get_queue_ref_from_ptr_and_syclobj(
         if pycapsule.PyCapsule_IsValid(cap, "SyclQueueRef"):
             q = SyclQueue(cap)
             return _queue_ref_copy_from_SyclQueue(ptr, <SyclQueue> q)
-        elif pycapsule.PyCapsule_IsValid(cap, "SyclContexRef"):
+        elif pycapsule.PyCapsule_IsValid(cap, "SyclContextRef"):
             ctx = <SyclContext>SyclContext(cap)
             return _queue_ref_copy_from_USMRef_and_SyclContext(ptr, ctx)
         else:

@@ -17,85 +17,104 @@
 """Unit test cases for dpctl.tensor.numpy_usm_shared.
 """
 
-import unittest
-
 import numpy
 
 from dpctl.tensor import numpy_usm_shared as dparray
 
 
-class Test_dparray(unittest.TestCase):
-    def setUp(self):
-        self.X = dparray.ndarray((256, 4), dtype="d")
-        self.X.fill(1.0)
+def get_arg():
+    X = dparray.ndarray((256, 4), dtype="d")
+    X.fill(1.0)
+    return X
+
+
+def test_dparray_type():
+    X = get_arg()
+    assert isinstance(X, dparray.ndarray)
+
+
+def test_dparray_as_ndarray_self():
+    X = get_arg()
+    Y = X.as_ndarray()
+    assert type(Y) == numpy.ndarray
+
+
+def test_dparray_as_ndarray():
+    X = get_arg()
+    Y = dparray.as_ndarray(X)
+    assert type(Y) == numpy.ndarray
+
+
+def test_dparray_from_ndarray():
+    X = get_arg()
+    Y = dparray.as_ndarray(X)
+    dp1 = dparray.from_ndarray(Y)
+    assert isinstance(dp1, dparray.ndarray)
+
+
+def test_multiplication_dparray():
+    C = get_arg() * 5
+    assert isinstance(C, dparray.ndarray)
+
+
+def test_inplace_sub():
+    X = get_arg()
+    X -= 1
 
-    def test_dparray_type(self):
-        self.assertIsInstance(self.X, dparray.ndarray)
 
-    def test_dparray_as_ndarray_self(self):
-        Y = self.X.as_ndarray()
-        self.assertEqual(type(Y), numpy.ndarray)
+def test_dparray_through_python_func():
+    def func_operation_with_const(dpctl_array):
+        return dpctl_array * 2.0 + 13
 
-    def test_dparray_as_ndarray(self):
-        Y = dparray.as_ndarray(self.X)
-        self.assertEqual(type(Y), numpy.ndarray)
+    C = get_arg() * 5
+    dp_func = func_operation_with_const(C)
+    assert isinstance(dp_func, dparray.ndarray)
 
-    def test_dparray_from_ndarray(self):
-        Y = dparray.as_ndarray(self.X)
-        dp1 = dparray.from_ndarray(Y)
-        self.assertIsInstance(dp1, dparray.ndarray)
 
-    def test_multiplication_dparray(self):
-        C = self.X * 5
-        self.assertIsInstance(C, dparray.ndarray)
+def test_dparray_mixing_dpctl_and_numpy():
+    dp_numpy = numpy.ones((256, 4), dtype="d")
+    X = get_arg()
+    res = dp_numpy * X
+    assert isinstance(X, dparray.ndarray)
+    assert isinstance(res, dparray.ndarray)
 
-    def test_inplace_sub(self):
-        self.X -= 1
 
-    def test_dparray_through_python_func(self):
-        def func_operation_with_const(dpctl_array):
-            return dpctl_array * 2.0 + 13
+def test_dparray_shape():
+    X = get_arg()
+    res = X.shape
+    assert res == (256, 4)
 
-        C = self.X * 5
-        dp_func = func_operation_with_const(C)
-        self.assertIsInstance(dp_func, dparray.ndarray)
 
-    def test_dparray_mixing_dpctl_and_numpy(self):
-        dp_numpy = numpy.ones((256, 4), dtype="d")
-        res = dp_numpy * self.X
-        self.assertIsInstance(self.X, dparray.ndarray)
-        self.assertIsInstance(res, dparray.ndarray)
+def test_dparray_T():
+    X = get_arg()
+    res = X.T
+    assert res.shape == (4, 256)
 
-    def test_dparray_shape(self):
-        res = self.X.shape
-        self.assertEqual(res, (256, 4))
 
-    def test_dparray_T(self):
-        res = self.X.T
-        self.assertEqual(res.shape, (4, 256))
+def test_numpy_ravel_with_dparray():
+    X = get_arg()
+    res = numpy.ravel(X)
+    assert res.shape == (1024,)
 
-    def test_numpy_ravel_with_dparray(self):
-        res = numpy.ravel(self.X)
-        self.assertEqual(res.shape, (1024,))
 
-    def test_numpy_sum_with_dparray(self):
-        res = numpy.sum(self.X)
-        self.assertEqual(res, 1024.0)
+def test_numpy_sum_with_dparray():
+    X = get_arg()
+    res = numpy.sum(X)
+    assert res == 1024.0
 
-    def test_numpy_sum_with_dparray_out(self):
-        res = dparray.empty((self.X.shape[1],), dtype=self.X.dtype)
-        res2 = numpy.sum(self.X, axis=0, out=res)
-        self.assertTrue(res is res2)
-        self.assertIsInstance(res2, dparray.ndarray)
 
-    def test_frexp_with_out(self):
-        X = dparray.array([0.5, 4.7])
-        mant = dparray.empty((2,), dtype="d")
-        exp = dparray.empty((2,), dtype="i4")
-        res = numpy.frexp(X, out=(mant, exp))
-        self.assertTrue(res[0] is mant)
-        self.assertTrue(res[1] is exp)
+def test_numpy_sum_with_dparray_out():
+    X = get_arg()
+    res = dparray.empty((X.shape[1],), dtype=X.dtype)
+    res2 = numpy.sum(X, axis=0, out=res)
+    assert res is res2
+    assert isinstance(res2, dparray.ndarray)
 
 
-if __name__ == "__main__":
-    unittest.main()
+def test_frexp_with_out():
+    X = dparray.array([0.5, 4.7])
+    mant = dparray.empty((2,), dtype="d")
+    exp = dparray.empty((2,), dtype="i4")
+    res = numpy.frexp(X, out=(mant, exp))
+    assert res[0] is mant
+    assert res[1] is exp
@@ -18,60 +18,55 @@
 """
 
 import ctypes
-import unittest
 
 import numpy as np
+import pytest
 
 import dpctl
 import dpctl.memory as dpctl_mem
 import dpctl.program as dpctl_prog
 
-from ._helper import has_gpu
 
+def test_create_program_from_source():
+    try:
+        q = dpctl.SyclQueue("opencl", property="enable_profiling")
+    except dpctl.SyclQueueCreationError:
+        pytest.skip("OpenCL queue could not be created")
+    oclSrc = "                                                             \
+    kernel void axpy(global int* a, global int* b, global int* c, int d) { \
+        size_t index = get_global_id(0);                                   \
+        c[index] = d*a[index] + b[index];                                  \
+    }"
+    prog = dpctl_prog.create_program_from_source(q, oclSrc)
+    axpyKernel = prog.get_sycl_kernel("axpy")
 
-@unittest.skipUnless(has_gpu(), "No OpenCL GPU queues available")
-class Test1DKernelSubmit(unittest.TestCase):
-    def test_create_program_from_source(self):
-        oclSrc = "                                                             \
-        kernel void axpy(global int* a, global int* b, global int* c, int d) { \
-            size_t index = get_global_id(0);                                   \
-            c[index] = d*a[index] + b[index];                                  \
-        }"
-        q = dpctl.SyclQueue("opencl:gpu", property="enable_profiling")
-        prog = dpctl_prog.create_program_from_source(q, oclSrc)
-        axpyKernel = prog.get_sycl_kernel("axpy")
+    n_elems = 1024 * 512
+    bufBytes = n_elems * np.dtype("i").itemsize
+    abuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
+    bbuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
+    cbuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
+    a = np.ndarray((n_elems,), buffer=abuf, dtype="i")
+    b = np.ndarray((n_elems,), buffer=bbuf, dtype="i")
+    c = np.ndarray((n_elems,), buffer=cbuf, dtype="i")
+    a[:] = np.arange(n_elems)
+    b[:] = np.arange(n_elems, 0, -1)
+    c[:] = 0
+    d = 2
+    args = []
 
-        n_elems = 1024 * 512
-        bufBytes = n_elems * np.dtype("i").itemsize
-        abuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
-        bbuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
-        cbuf = dpctl_mem.MemoryUSMShared(bufBytes, queue=q)
-        a = np.ndarray((n_elems,), buffer=abuf, dtype="i")
-        b = np.ndarray((n_elems,), buffer=bbuf, dtype="i")
-        c = np.ndarray((n_elems,), buffer=cbuf, dtype="i")
-        a[:] = np.arange(n_elems)
-        b[:] = np.arange(n_elems, 0, -1)
-        c[:] = 0
-        d = 2
-        args = []
+    args.append(a.base)
+    args.append(b.base)
+    args.append(c.base)
+    args.append(ctypes.c_int(d))
 
-        args.append(a.base)
-        args.append(b.base)
-        args.append(c.base)
-        args.append(ctypes.c_int(d))
+    r = [
+        n_elems,
+    ]
 
-        r = [
-            n_elems,
-        ]
-
-        timer = dpctl.SyclTimer()
-        with timer(q):
-            q.submit(axpyKernel, args, r)
-            ref_c = a * d + b
-        host_dt, device_dt = timer.dt
-        self.assertTrue(host_dt > device_dt)
-        self.assertTrue(np.allclose(c, ref_c))
-
-
-if __name__ == "__main__":
-    unittest.main()
+    timer = dpctl.SyclTimer()
+    with timer(q):
+        q.submit(axpyKernel, args, r)
+        ref_c = a * d + b
+    host_dt, device_dt = timer.dt
+    assert host_dt > device_dt
+    assert np.allclose(c, ref_c)