Tweak test_async_submit to improve test suite stability

The test that expects one of the kernels to still be running adds
more work to each of the dependent kernels.

Author: oleksandr-pavlyk

dpctl/tests/test_sycl_kernel_submit.py

@@ -121,12 +121,22 @@ def test_async_submit():
         pytest.skip("OpenCL queue could not be created")
     oclSrc = (
         "kernel void kern1(global unsigned int *res, unsigned int mod) {"
+        "   size_t unused_sum = 0;"
+        "   size_t i = 0; "
+        "   for (i = 0; i < 4000; i++) { "
+        "       unused_sum += i;"
+        "   } "
         "   size_t index = get_global_id(0);"
         "   int ri = (index % mod);"
         "   res[index] = (ri * ri) % mod;"
         "}"
         " "
         "kernel void kern2(global unsigned int *res, unsigned int mod) {"
+        "   size_t unused_sum = 0;"
+        "   size_t i = 0; "
+        "   for (i = 0; i < 4000; i++) { "
+        "       unused_sum += i;"
+        "   } "
         "   size_t index = get_global_id(0);"
         "   int ri = (index % mod);"
         "   int ri2 = (ri * ri) % mod;"
@@ -137,13 +147,13 @@ def test_async_submit():
         "   global unsigned int *res, global unsigned int *arg1, "
         "   global unsigned int *arg2)"
         "{"
-        "  size_t index = get_global_id(0);"
-        "  size_t i = 0; "
-        "  size_t unused_sum = 0;"
-        "  for (i = 0; i < 4000; i++) { "
+        "   size_t index = get_global_id(0);"
+        "   size_t i = 0; "
+        "   size_t unused_sum = 0;"
+        "   for (i = 0; i < 4000; i++) { "
         "       unused_sum += i;"
-        "  } "
-        "  res[index] = "
+        "   } "
+        "   res[index] = "
         "      (arg1[index] < arg2[index]) ? arg1[index] : arg2[index];"
         "}"
     )