modified sycl_buffer Cython example

Author: oleksandr-pavlyk

examples/cython/sycl_buffer/README.md

@@ -21,9 +21,9 @@ CC=clang CXX=dpcpp python setup.py build_ext --inplace
 #2 Running
 
 ```
-# SYCL_BE=PI_OPENCL sets SYCL backend to OpenCL to avoid a
+# SYCL_DEVICE_FILTER=opencl sets SYCL backend to OpenCL to avoid a
 # transient issue with MKL's using the default Level-0 backend
-(idp) [08:16:12 ansatnuc04 simple]$ SYCL_BE=PI_OPENCL ipython
+(idp) [08:16:12 ansatnuc04 simple]$ SYCL_DEVICE_FILTER=opencl ipython
 Python 3.7.7 (default, Jul 14 2020, 22:02:37)
 Type 'copyright', 'credits' or 'license' for more information
 IPython 7.17.0 -- An enhanced Interactive Python. Type '?' for help.
@@ -67,7 +67,7 @@ Times for NumPy
 Running run.py:
 
 ```
-(idp) [09:14:53 ansatnuc04 sycl_buffer]$ SYCL_BE=PI_OPENCL python run.py
+(idp) [09:14:53 ansatnuc04 sycl_buffer]$ SYCL_DEVICE_FILTER=opencl python run.py
 Result computed by NumPy
 [  0.27170187 -23.36798583   7.31326489  -1.95121928]
 Result computed by SYCL extension

examples/cython/sycl_buffer/_buffer_example.pyx

@@ -24,14 +24,19 @@ cdef extern from "use_sycl_buffer.h":
     int c_columnwise_total(c_dpctl.DPCTLSyclQueueRef q, size_t n, size_t m, double *m, double *ct) nogil
     int c_columnwise_total_no_mkl(c_dpctl.DPCTLSyclQueueRef q, size_t n, size_t m, double *m, double *ct) nogil
 
-def columnwise_total(double[:, ::1] v, method='mkl'):
+def columnwise_total(double[:, ::1] v, method='mkl', queue=None):
     cdef cnp.ndarray res_array = np.empty((v.shape[1],), dtype='d')
     cdef double[::1] res_memslice = res_array
     cdef int ret_status
     cdef c_dpctl.SyclQueue q
     cdef c_dpctl.DPCTLSyclQueueRef q_ref
 
-    q = c_dpctl.get_current_queue()
+    if (queue is None):
+        q = c_dpctl.SyclQueue()
+    elif isinstance(queue, dpctl.SyclQueue):
+        q = <c_dpctl.SyclQueue> queue
+    else:
+        q = c_dpctl.SyclQueue(queue)
     q_ref = q.get_queue_ref()
 
     if method == 'mkl':

examples/cython/sycl_buffer/bench.py

@@ -24,19 +24,19 @@
 print("=" * 10 + " Executing warm-up " + "=" * 10)
 print("NumPy result: ", X.sum(axis=0))
 
-dpctl.set_global_queue("opencl:cpu")
+q = dpctl.SyclQueue("opencl:cpu")
 print(
     "SYCL({}) result: {}".format(
-        dpctl.get_current_queue().sycl_device.name,
-        sb.columnwise_total(X),
+        q.sycl_device.name,
+        sb.columnwise_total(X, queue=q),
     )
 )
 
-dpctl.set_default_queue("opencl:gpu")
+q = dpctl.SyclQueue("opencl:gpu")
 print(
     "SYCL({}) result: {}".format(
-        dpctl.get_current_queue().sycl_device.name,
-        sb.columnwise_total(X),
+        q.sycl_device.name,
+        sb.columnwise_total(X, queue=q),
     )
 )
 
@@ -45,9 +45,9 @@
 print("Times for 'opencl:cpu'")
 print(
     timeit.repeat(
-        stmt="sb.columnwise_total(X)",
-        setup='dpctl.set_global_queue("opencl:cpu"); '
-        "sb.columnwise_total(X)",  # ensure JIT compilation is not counted
+        stmt="sb.columnwise_total(X, queue=q)",
+        setup='q = dpctl.SyclQueue("opencl:cpu"); '
+        "sb.columnwise_total(X, queue=q)",  # ensure JIT compilation is not counted
         number=100,
         globals=globals(),
     )
@@ -56,8 +56,8 @@
 print("Times for 'opencl:gpu'")
 print(
     timeit.repeat(
-        stmt="sb.columnwise_total(X)",
-        setup='dpctl.set_default_queue("opencl:gpu"); sb.columnwise_total(X)',
+        stmt="sb.columnwise_total(X, queue=q)",
+        setup='q = dpctl.SyclQueue("opencl:gpu"); sb.columnwise_total(X, queue=q)',
         number=100,
         globals=globals(),
     )

examples/cython/sycl_buffer/run.py

@@ -16,23 +16,24 @@
 
 import syclbuffer as sb
 import numpy as np
+import dpctl
 
 X = np.random.randn(100, 4)
 
 print("Result computed by NumPy")
 print(X.sum(axis=0))
-print("Result computed by SYCL extension")
+print("Result computed by SYCL extension using default offloading target")
 print(sb.columnwise_total(X))
 
 
 print("")
+
 # controlling where to offload
-import dpctl
 
-with dpctl.device_context("opencl:gpu"):
-    print("Running on: ", dpctl.get_current_queue().sycl_device.name)
-    print(sb.columnwise_total(X))
+q = dpctl.SyclQueue("opencl:gpu")
+print("Running on: ", q.sycl_device.name)
+print(sb.columnwise_total(X, queue=q))
 
-with dpctl.device_context("opencl:cpu"):
-    print("Running on: ", dpctl.get_current_queue().sycl_device.name)
-    print(sb.columnwise_total(X))
+q = dpctl.SyclQueue("opencl:cpu")
+print("Running on: ", q.sycl_device.name)
+print(sb.columnwise_total(X, queue=q))