[SYCL][CUDA] Unoptimized stream regression test (#362)

Previously the CUDA backend would fail due to invalid atomic memory
orders not being optimized out. The use of these invalid memory orders
have been removed in recent changes, so this commit adds a regression
test to make sure they do not resurface.

Signed-off-by: Steffen Larsen <[email protected]>

Author: steffenlarsen

SYCL/Regression/unoptimized_stream.cpp

@@ -0,0 +1,20 @@
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -O0 -o %t.out
+// RUN: %HOST_RUN_PLACEHOLDER %t.out
+// RUN: %CPU_RUN_PLACEHOLDER %t.out
+// RUN: %GPU_RUN_PLACEHOLDER %t.out
+
+// NOTE: The libclc target used by the CUDA backend used to generate atomic load
+//       variants that were unsupported by NVPTX. Even if they were not used
+//       directly, sycl::stream and other operations would keep the invalid
+//       operations in when optimizations were disabled.
+
+#include <sycl/sycl.hpp>
+
+int main() {
+  sycl::queue q;
+  q.submit([&](sycl::handler &cgh) {
+    sycl::stream os(1024, 256, cgh);
+    cgh.single_task([=]() { os << "test"; });
+  });
+  return 0;
+}