Feature/memory queue=none (#382)

* Modified queue=None behavior for Memory objects

queue=None is now understood as use `dpctl.SyclQueue()` queue.

* use raise newError from oldError syntax

* Use explicit queue= argument in memory buffer constructors, now that it no longer uses get_current_queue

Author: oleksandr-pavlyk

dpctl/_sycl_device.pyx

@@ -762,7 +762,7 @@ cdef class SyclDevice(_SyclDevice):
                 partition = int(partition)
                 return self.create_sub_devices_equally(partition)
             except Exception as e:
-                raise TypeError("Unsupported type of sub-device argument")
+                raise TypeError("Unsupported type of sub-device argument") from e
 
     @property
     def parent_device(self):

dpctl/memory/_memory.pyx

@@ -28,7 +28,6 @@ from dpctl._backend cimport *
 from .._sycl_context cimport SyclContext
 from .._sycl_device cimport SyclDevice
 from .._sycl_queue cimport SyclQueue
-from .._sycl_queue_manager cimport get_current_queue
 
 from cpython cimport Py_buffer
 from cpython.bytes cimport PyBytes_AS_STRING, PyBytes_FromStringAndSize
@@ -222,7 +221,7 @@ cdef class _Memory:
 
         if (nbytes > 0):
             if queue is None:
-                queue = get_current_queue()
+                queue = dpctl.SyclQueue()
 
             if (ptr_type == b"shared"):
                 if alignment > 0:
@@ -493,7 +492,7 @@ cdef class MemoryUSMShared(_Memory):
     USM shared memory.
 
     Non-positive alignments are not used (malloc_shared is used instead).
-    The queue=None the current `dpctl.get_current_queue()` is used to allocate memory.
+    For the queue=None cast the `dpctl.SyclQueue()` is used to allocate memory.
 
     MemoryUSMShared(usm_obj) constructor create instance from `usm_obj` expected to
     implement `__sycl_usm_array_interface__` protocol and exposing a contiguous block of
@@ -525,7 +524,7 @@ cdef class MemoryUSMHost(_Memory):
     USM host memory.
 
     Non-positive alignments are not used (malloc_host is used instead).
-    The queue=None the current `dpctl.get_current_queue()` is used to allocate memory.
+    For the queue=None case `dpctl.SyclQueue()` is used to allocate memory.
 
     MemoryUSMDevice(usm_obj) constructor create instance from `usm_obj` expected to
     implement `__sycl_usm_array_interface__` protocol and exposing a contiguous block of
@@ -557,7 +556,7 @@ cdef class MemoryUSMDevice(_Memory):
     USM device memory.
 
     Non-positive alignments are not used (malloc_device is used instead).
-    The queue=None the current `dpctl.get_current_queue()` is used to allocate memory.
+    For the queue=None cast the `dpctl.SyclQueue()` is used to allocate memory.
 
     MemoryUSMDevice(usm_obj) constructor create instance from `usm_obj` expected to
     implement `__sycl_usm_array_interface__` protocol and exposing a contiguous block of

dpctl/tests/test_sycl_kernel_submit.py

@@ -34,32 +34,32 @@ def test_create_program_from_source(self):
             size_t index = get_global_id(0);                                   \
             c[index] = d*a[index] + b[index];                                  \
         }"
-        with dpctl.device_context("opencl:gpu:0"):
-            q = dpctl.get_current_queue()
-            prog = dpctl_prog.create_program_from_source(q, oclSrc)
-            axpyKernel = prog.get_sycl_kernel("axpy")
+        q = dpctl.SyclQueue("opencl:gpu")
+        prog = dpctl_prog.create_program_from_source(q, oclSrc)
+        axpyKernel = prog.get_sycl_kernel("axpy")
 
-            abuf = dpctl_mem.MemoryUSMShared(1024 * np.dtype("i").itemsize)
-            bbuf = dpctl_mem.MemoryUSMShared(1024 * np.dtype("i").itemsize)
-            cbuf = dpctl_mem.MemoryUSMShared(1024 * np.dtype("i").itemsize)
-            a = np.ndarray((1024), buffer=abuf, dtype="i")
-            b = np.ndarray((1024), buffer=bbuf, dtype="i")
-            c = np.ndarray((1024), buffer=cbuf, dtype="i")
-            a[:] = np.arange(1024)
-            b[:] = np.arange(1024, 0, -1)
-            c[:] = 0
-            d = 2
-            args = []
+        bufBytes = 1024 * 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((1024), buffer=abuf, dtype="i")
+        b = np.ndarray((1024), buffer=bbuf, dtype="i")
+        c = np.ndarray((1024), buffer=cbuf, dtype="i")
+        a[:] = np.arange(1024)
+        b[:] = np.arange(1024, 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 = [1024]
+        r = [1024]
 
-            q.submit(axpyKernel, args, r)
-            self.assertTrue(np.allclose(c, a * d + b))
+        q.submit(axpyKernel, args, r)
+        self.assertTrue(np.allclose(c, a * d + b))
 
 
 if __name__ == "__main__":

dpctl/tests/test_sycl_queue_memcpy.py

@@ -51,7 +51,7 @@ def test_memcpy_copy_usm_to_usm(self):
     )
     def test_memcpy_type_error(self):
         mobj = self._create_memory()
-        q = dpctl.get_current_queue()
+        q = mobj._queue
 
         with self.assertRaises(TypeError) as cm:
             q.memcpy(None, mobj, 3)