[SYCL][CUDA] Add tests for exceeding maximum number of work groups

SYCL/Basic/cuda_max_wgs_error.cpp

@@ -0,0 +1,52 @@
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -fno-sycl-id-queries-fit-in-int
+// RUN: %GPU_RUN_PLACEHOLDER %t.out
+//
+// REQUIRES: cuda
+
+#include <numeric>
+#include <sycl/sycl.hpp>
+
+using namespace sycl;
+
+const size_t lsize = 32;
+const std::string expected_msg =
+    "Number of work-groups exceed limit for dimension ";
+
+template <int N>
+void check(range<N> global, range<N> local, bool expect_fail = false) {
+  queue q;
+  try {
+    q.submit([&](handler &cgh) {
+      cgh.parallel_for(nd_range<N>(global, local), [=](nd_item<N> item) {});
+    });
+  } catch (nd_range_error e) {
+    if (expect_fail) {
+      std::string msg = e.what();
+      assert(msg.rfind(expected_msg, 0) == 0);
+    } else {
+      throw e;
+    }
+  }
+}
+
+int main() {
+  queue q;
+  device d = q.get_device();
+  id<1> max_1 = d.get_info<sycl::info::device::ext_oneapi_max_work_groups_1d>();
+  check(range<1>(max_1[0] * lsize), range<1>(lsize));
+  check(range<1>((max_1[0] + 1) * lsize), range<1>(lsize), true);
+
+  id<2> max_2 = d.get_info<sycl::info::device::ext_oneapi_max_work_groups_2d>();
+  check(range<2>(1, max_2[1] * lsize), range<2>(1, lsize));
+  check(range<2>(1, (max_2[1] + 1) * lsize), range<2>(1, lsize), true);
+  check(range<2>(max_2[0] * lsize, 1), range<2>(lsize, 1));
+  check(range<2>((max_2[0] + 1) * lsize, 1), range<2>(lsize, 1), true);
+
+  id<3> max_3 = d.get_info<sycl::info::device::ext_oneapi_max_work_groups_3d>();
+  check(range<3>(1, 1, max_3[2] * lsize), range<3>(1, 1, lsize));
+  check(range<3>(1, 1, (max_3[2] + 1) * lsize), range<3>(1, 1, lsize), true);
+  check(range<3>(1, max_3[1] * lsize, 1), range<3>(1, lsize, 1));
+  check(range<3>(1, (max_3[1] + 1) * lsize, 1), range<3>(1, lsize, 1), true);
+  check(range<3>(max_3[0] * lsize, 1, 1), range<3>(lsize, 1, 1));
+  check(range<3>((max_3[0] + 1) * lsize, 1, 1), range<3>(lsize, 1, 1), true);
+}