Updating all kernel examples into new launch parameter syntax

numba_dpex/examples/kernel/atomic_op.py

@@ -5,6 +5,7 @@
 import dpnp as np
 
 import numba_dpex as ndpex
+from numba_dpex.core.kernel_interface.utils import Range
 
 
 @ndpex.kernel
@@ -20,7 +21,7 @@ def main():
     print("Using device ...")
     print(a.device)
 
-    atomic_reduction[N](a)
+    atomic_reduction[Range(N)](a)
     print("Reduction sum =", a[0])
 
     print("Done...")

numba_dpex/examples/kernel/black_scholes.py

@@ -8,6 +8,7 @@
 import dpnp as np
 
 import numba_dpex as ndpx
+from numba_dpex.core.kernel_interface.utils import Range
 
 # Stock price range
 S0L = 10.0
@@ -94,7 +95,9 @@ def main():
     print("Using device ...")
     print(price.device)
 
-    kernel_black_scholes[NOPT](price, strike, t, rate, volatility, call, put)
+    kernel_black_scholes[Range(NOPT)](
+        price, strike, t, rate, volatility, call, put
+    )
 
     print("Call:", call)
     print("Put:", put)

numba_dpex/examples/kernel/device_func.py

@@ -6,6 +6,7 @@
 
 import numba_dpex as ndpex
 from numba_dpex import float32, int32, int64
+from numba_dpex.core.kernel_interface.utils import Range
 
 # Array size
 N = 10
@@ -69,7 +70,7 @@ def test1():
 
     print("A=", a)
     try:
-        a_kernel_function[N](a, b)
+        a_kernel_function[Range(N)](a, b)
     except Exception as err:
         print(err)
     print("B=", b)
@@ -87,7 +88,7 @@ def test2():
 
     print("A=", a)
     try:
-        a_kernel_function_int32[N](a, b)
+        a_kernel_function_int32[Range(N)](a, b)
     except Exception as err:
         print(err)
     print("B=", b)
@@ -105,7 +106,7 @@ def test3():
 
     print("A=", a)
     try:
-        a_kernel_function_int32_float32[N](a, b)
+        a_kernel_function_int32_float32[Range(N)](a, b)
     except Exception as err:
         print(err)
     print("B=", b)
@@ -119,7 +120,7 @@ def test3():
 
     print("A=", a)
     try:
-        a_kernel_function_int32_float32[N](a, b)
+        a_kernel_function_int32_float32[Range(N)](a, b)
     except Exception as err:
         print(err)
     print("B=", b)
@@ -134,7 +135,7 @@ def test3():
 
     print("A=", a)
     try:
-        a_kernel_function_int32_float32[N](a, b)
+        a_kernel_function_int32_float32[Range(N)](a, b)
     except Exception as err:
         print(err)
     print("B=", b)

numba_dpex/examples/kernel/interpolation.py

@@ -7,6 +7,7 @@
 from numpy.testing import assert_almost_equal
 
 import numba_dpex as ndpex
+from numba_dpex.core.kernel_interface.utils import NdRange, Range
 
 # Interpolation domain
 XLO = 10.0
@@ -114,9 +115,13 @@ def main():
 
     print("Using device ...")
     print(xp.device)
-    global_range = (N_POINTS // N_POINTS_PER_WORK_ITEM,)
-    local_range = (LOCAL_SIZE,)
-    kernel_polynomial[global_range, local_range](xp, yp, COEFFICIENTS)
+    global_range = Range(
+        N_POINTS // N_POINTS_PER_WORK_ITEM,
+    )
+    local_range = Range(
+        LOCAL_SIZE,
+    )
+    kernel_polynomial[NdRange(global_range, local_range)](xp, yp, COEFFICIENTS)
 
     # Copy results back to the host
     nyp = np.asnumpy(yp)

numba_dpex/examples/kernel/kernel_private_memory.py

@@ -8,6 +8,7 @@
 from numba import float32
 
 import numba_dpex
+from numba_dpex.core.kernel_interface.utils import NdRange, Range
 
 
 def private_memory():
@@ -39,9 +40,9 @@ def private_memory_kernel(A):
     print("Using device ...")
     device.print_device_info()
 
-    global_range = (N,)
-    local_range = (N,)
-    private_memory_kernel[global_range, local_range](arr)
+    global_range = Range(N)
+    local_range = Range(N)
+    private_memory_kernel[NdRange(global_range, local_range)](arr)
 
     arr_out = dpt.asnumpy(arr)
     np.testing.assert_allclose(orig * 2, arr_out)

numba_dpex/examples/kernel/kernel_specialization.py

@@ -11,6 +11,7 @@
     InvalidKernelSpecializationError,
     MissingSpecializationError,
 )
+from numba_dpex.core.kernel_interface.utils import Range
 
 # Similar to Numba, numba-dpex supports eager compilation of functions. The
 # following examples demonstrate the feature for numba_dpex.kernel and presents
@@ -38,7 +39,7 @@ def data_parallel_sum(a, b, c):
 b = dpt.ones(1024, dtype=dpt.int64)
 c = dpt.zeros(1024, dtype=dpt.int64)
 
-data_parallel_sum[1024](a, b, c)
+data_parallel_sum[Range(1024)](a, b, c)
 
 npc = dpt.asnumpy(c)
 npc_expected = np.full(1024, 2, dtype=np.int64)
@@ -65,7 +66,7 @@ def data_parallel_sum2(a, b, c):
 b = dpt.ones(1024, dtype=dpt.int64)
 c = dpt.zeros(1024, dtype=dpt.int64)
 
-data_parallel_sum2[1024](a, b, c)
+data_parallel_sum2[Range(1024)](a, b, c)
 
 npc = dpt.asnumpy(c)
 npc_expected = np.full(1024, 2, dtype=np.int64)
@@ -76,7 +77,7 @@ def data_parallel_sum2(a, b, c):
 b = dpt.ones(1024, dtype=dpt.float32)
 c = dpt.zeros(1024, dtype=dpt.float32)
 
-data_parallel_sum2[1024](a, b, c)
+data_parallel_sum2[Range(1024)](a, b, c)
 
 npc = dpt.asnumpy(c)
 npc_expected = np.full(1024, 2, dtype=np.float32)
@@ -94,7 +95,7 @@ def data_parallel_sum2(a, b, c):
 c = dpt.zeros(1024, dtype=dpt.int32)
 
 try:
-    data_parallel_sum[1024](a, b, c)
+    data_parallel_sum[Range(1024)](a, b, c)
 except MissingSpecializationError as mse:
     print(mse)
 

numba_dpex/examples/kernel/matmul.py

@@ -9,6 +9,7 @@
 import numpy as np
 
 import numba_dpex as dpex
+from numba_dpex.core.kernel_interface.utils import NdRange, Range
 
 
 @dpex.kernel
@@ -30,13 +31,13 @@ def gemm(a, b, c):
 Y = 16
 global_size = X, X
 
-griddim = X, X
-blockdim = Y, Y
+griddim = Range(X, X)
+blockdim = Range(Y, Y)
 
 
 def driver(a, b, c):
     # Invoke the kernel
-    gemm[griddim, blockdim](a, b, c)
+    gemm[NdRange(griddim, blockdim)](a, b, c)
 
 
 def main():

numba_dpex/examples/kernel/pairwise_distance.py

@@ -12,6 +12,7 @@
 import numpy as np
 
 import numba_dpex as dpex
+from numba_dpex.core.kernel_interface.utils import NdRange, Range
 
 parser = argparse.ArgumentParser(
     description="Program to compute pairwise distance"
@@ -25,9 +26,9 @@
 args = parser.parse_args()
 
 # Global work size is equal to the number of points
-global_size = (args.n,)
+global_size = Range(args.n)
 # Local Work size is optional
-local_size = (args.l,)
+local_size = Range(args.l)
 
 X = np.random.random((args.n, args.d)).astype(np.single)
 D = np.empty((args.n, args.n), dtype=np.single)
@@ -65,7 +66,7 @@ def driver():
 
     for repeat in range(args.r):
         start = time()
-        pairwise_distance[global_size, local_size](
+        pairwise_distance[NdRange(global_size, local_size)](
             x_ndarray, d_ndarray, X.shape[0], X.shape[1]
         )
         end = time()

numba_dpex/examples/kernel/scan.py

@@ -7,6 +7,7 @@
 import dpnp as np
 
 import numba_dpex as ndpx
+from numba_dpex.core.kernel_interface.utils import Range
 
 # 1D array size
 N = 64
@@ -56,7 +57,7 @@ def main():
 
     print("Using device ...")
     print(arr.device)
-    kernel_hillis_steele_scan[N](arr)
+    kernel_hillis_steele_scan[Range(N)](arr)
 
     # the output should be [0, 1, 3, 6, ...]
     arr_np = np.asnumpy(arr)

numba_dpex/examples/kernel/select_device_for_kernel.py

@@ -9,6 +9,7 @@
 import numpy as np
 
 import numba_dpex
+from numba_dpex.core.kernel_interface.utils import NdRange, Range
 
 """
 We support passing arrays of two types to a @numba_dpex.kernel decorated
@@ -86,7 +87,7 @@ def select_device_ndarray(N):
     default_device = dpctl.select_default_device()
 
     with numba_dpex.offload_to_sycl_device(default_device.filter_string):
-        sum_kernel[(N,), (1,)](a, b, got)
+        sum_kernel[NdRange(Range(N), Range(1))](a, b, got)
 
     expected = a + b
 
@@ -110,7 +111,7 @@ def select_device_SUAI(N):
 
     # Users don't need to specify where the computation will
     # take place. It will be inferred from data.
-    sum_kernel[(N,), (1,)](da, db, dc)
+    sum_kernel[NdRange(Range(N), Range(1))](da, db, dc)
 
     dc.usm_data.copy_to_host(got.reshape((-1)).view("|u1"))
 

numba_dpex/examples/kernel/sum_reduction_ocl.py

@@ -7,6 +7,7 @@
 from numba import int32
 
 import numba_dpex as dpex
+from numba_dpex.core.kernel_interface.utils import NdRange, Range
 
 
 @dpex.kernel
@@ -49,9 +50,9 @@ def sum_reduce(A):
 
     partial_sums = dpt.zeros(nb_work_groups, dtype=A.dtype, device=A.device)
 
-    gs = (global_size,)
-    ls = (work_group_size,)
-    sum_reduction_kernel[gs, ls](A, partial_sums)
+    gs = Range(global_size)
+    ls = Range(work_group_size)
+    sum_reduction_kernel[NdRange(gs, ls)](A, partial_sums)
 
     final_sum = 0
     # calculate the final sum in HOST

numba_dpex/examples/kernel/vector_sum.py

@@ -6,6 +6,7 @@
 import numpy.testing as testing
 
 import numba_dpex as ndpx
+from numba_dpex.core.kernel_interface.utils import Range
 
 
 # Data parallel kernel implementing vector sum
@@ -18,7 +19,7 @@ def kernel_vector_sum(a, b, c):
 # Utility function for printing and testing
 def driver(a, b, c, global_size):
 
-    kernel_vector_sum[global_size](a, b, c)
+    kernel_vector_sum[Range(global_size)](a, b, c)
 
     a_np = dpnp.asnumpy(a)  # Copy dpnp array a to NumPy array a_np
     b_np = dpnp.asnumpy(b)  # Copy dpnp array b to NumPy array b_np

numba_dpex/examples/kernel/vector_sum2D.py

@@ -23,19 +23,14 @@ def data_parallel_sum(a, b, c):
 
 
 def driver(a, b, c, global_size):
-    data_parallel_sum[global_size, dpex.DEFAULT_LOCAL_SIZE](a, b, c)
-
-
-def driver_with_range(a, b, c, global_size):
-    ranges = Range(*global_size)
-    data_parallel_sum[ranges](a, b, c)
+    data_parallel_sum[global_size](a, b, c)
 
 
 def main():
     # Array dimensions
     X = 8
     Y = 8
-    global_size = X, Y
+    global_size = Range(X, Y)
 
     a = np.arange(X * Y, dtype=np.float32).reshape(X, Y)
     b = np.arange(X * Y, dtype=np.float32).reshape(X, Y)
@@ -59,11 +54,6 @@ def main():
     c_out = dpt.asnumpy(c_dpt)
     assert np.allclose(c, c_out)
 
-    print("Running kernel with the new lanuch parameter syntax ...")
-    driver_with_range(a_dpt, b_dpt, c_dpt, global_size)
-    c_out = dpt.asnumpy(c_dpt)
-    assert np.allclose(c, c_out)
-
     print("Done...")