Add synchronization points in example

Author: oleksandr-pavlyk

examples/cython/usm_memory/blackscholes/blackscholes.pyx

@@ -106,11 +106,15 @@ def black_scholes_price(c_dpt.usm_ndarray option_params_arr):
         call_put_prices = dpt.empty((n_opts, 2), dtype='d', sycl_queue=q)
         dp1 = <double *>option_params_arr.get_data()
         dp2 = <double *>call_put_prices.get_data()
+        # ensure content of dp1 and dp2 is no longer worked on
+        exec_q_ptr[0].wait()
         cpp_blackscholes[double](exec_q_ptr[0], n_opts, dp1, dp2)
     elif (typenum_ == c_dpt.UAR_FLOAT):
         call_put_prices = dpt.empty((n_opts, 2), dtype='f', sycl_queue=q)
         fp1 = <float *>option_params_arr.get_data()
         fp2 = <float *>call_put_prices.get_data()
+        # ensure content of fp1 and fp2 is no longer worked on
+        exec_q_ptr[0].wait()
         cpp_blackscholes[float](exec_q_ptr[0], n_opts, fp1, fp2)
     else:
         raise ValueError("Unsupported data-type")
@@ -196,11 +200,13 @@ def populate_params(
 
     if (typenum_ == c_dpt.UAR_DOUBLE):
         dp = <double *>option_params_arr.get_data()
+        exec_q_ptr[0].wait()
         cpp_populate_params[double](
             exec_q_ptr[0], n_opts, dp, pl, ph, sl, sh, tl, th, rl, rh, vl, vh, seed
         )
     elif (typenum_ == c_dpt.UAR_FLOAT):
         fp = <float *>option_params_arr.get_data()
+        exec_q_ptr[0].wait()
         cpp_populate_params[float](
             exec_q_ptr[0], n_opts, fp, pl, ph, sl, sh, tl, th, rl, rh, vl, vh, seed
         )

examples/cython/usm_memory/src/sycl_blackscholes.hpp

@@ -47,7 +47,7 @@ constexpr int CALL = 0;
 constexpr int PUT = 1;
 
 template <typename T>
-void cpp_blackscholes(sycl::queue q, size_t n_opts, T *params, T *callput)
+void cpp_blackscholes(sycl::queue &q, size_t n_opts, T *params, T *callput)
 {
     using data_t = T;
 
@@ -57,8 +57,7 @@ void cpp_blackscholes(sycl::queue q, size_t n_opts, T *params, T *callput)
         data_t half = one / two;
 
         cgh.parallel_for<class black_scholes_kernel<T>>(
-		sycl::range<1>(n_opts),
-		[=](sycl::id<1> idx)
+            sycl::range<1>(n_opts), [=](sycl::id<1> idx)
         {
             const size_t i = n_params * idx[0];
             const data_t opt_price = params[i + PRICE];
@@ -106,7 +105,7 @@ void cpp_blackscholes(sycl::queue q, size_t n_opts, T *params, T *callput)
             const size_t callput_i = n_prices * idx[0];
             callput[callput_i + CALL] = call_price;
             callput[callput_i + PUT] = put_price;
-        });
+            });
     });
 
     e.wait_and_throw();

examples/pybind11/external_usm_allocation/external_usm_allocation/_usm_alloc_example.cpp

@@ -114,6 +114,9 @@ py::dict construct_sua_iface(DMatrix &m)
     iface["typestr"] = "|f8";
     iface["syclobj"] = syclobj;
 
+    // ensure that content of array is flushed out
+    m.get_queue().wait();
+
     return iface;
 }
 

examples/pybind11/onemkl_gemv/solve.py

@@ -127,12 +127,14 @@ def cg_solve(A, b):
     converged is False if solver has not converged, or the iteration number
     """
     exec_queue = A.sycl_queue
+    exec_queue.wait()
+
     x = dpt.zeros_like(b)
     Ap = dpt.empty_like(x)
 
     all_host_tasks = []
-    r = dpt.copy(b)  # synchronous copy
-    p = dpt.copy(b)  # synchronous copy
+    r = dpt.copy(b)
+    p = dpt.copy(b)
 
     rsold = sycl_gemm.norm_squared_blocking(exec_queue, r)
     if rsold < 1e-20:

examples/pybind11/onemkl_gemv/sycl_gemm/_onemkl.cpp

@@ -41,7 +41,7 @@ namespace py = pybind11;
 using dpctl::utils::keep_args_alive;
 
 std::pair<sycl::event, sycl::event>
-py_gemv(sycl::queue q,
+py_gemv(sycl::queue &q,
         dpctl::tensor::usm_ndarray matrix,
         dpctl::tensor::usm_ndarray vector,
         dpctl::tensor::usm_ndarray result,

examples/pybind11/use_dpctl_sycl_kernel/tests/test_user_kernel.py

@@ -64,6 +64,7 @@ def test_kernel_submit_through_extension():
     x = dpt.arange(0, stop=13, step=1, dtype="i4", sycl_queue=q)
     y = dpt.zeros_like(x)
 
+    q.wait()
     uk.submit_custom_kernel(q, krn, x, y, [])
 
     assert np.array_equal(dpt.asnumpy(y), np.arange(0, 26, step=2, dtype="i4"))

examples/pybind11/use_dpctl_sycl_kernel/use_kernel/_example.cpp

@@ -35,8 +35,8 @@
 
 namespace py = pybind11;
 
-void submit_custom_kernel(sycl::queue q,
-                          sycl::kernel krn,
+void submit_custom_kernel(sycl::queue &q,
+                          sycl::kernel &krn,
                           dpctl::tensor::usm_ndarray x,
                           dpctl::tensor::usm_ndarray y,
                           const std::vector<sycl::event> &depends = {})