[SYCL][CUDA] Fix incorrectly dispatched PI methods (#2726)

Author: Alexander Batashev

sycl/plugins/cuda/pi_cuda.cpp

@@ -4529,8 +4529,9 @@ pi_result piPluginInit(pi_plugin *PluginInit) {
   _PI_CL(piProgramGetBuildInfo, cuda_piProgramGetBuildInfo)
   _PI_CL(piProgramRetain, cuda_piProgramRetain)
   _PI_CL(piProgramRelease, cuda_piProgramRelease)
-  _PI_CL(piextMemGetNativeHandle, cuda_piextMemGetNativeHandle)
-  _PI_CL(piextMemCreateWithNativeHandle, cuda_piextMemCreateWithNativeHandle)
+  _PI_CL(piextProgramGetNativeHandle, cuda_piextProgramGetNativeHandle)
+  _PI_CL(piextProgramCreateWithNativeHandle,
+         cuda_piextProgramCreateWithNativeHandle)
   // Kernel
   _PI_CL(piKernelCreate, cuda_piKernelCreate)
   _PI_CL(piKernelSetArg, cuda_piKernelSetArg)

sycl/test/regression/cuda_program_interop.cpp

@@ -0,0 +1,23 @@
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
+// RUN: %GPU_RUN_PLACEHOLDER %t.out
+// REQUIRES: cuda
+
+#include <CL/sycl.hpp>
+#include <CL/sycl/backend/cuda.hpp>
+
+int main() {
+  sycl::queue Queue{sycl::default_selector{}};
+  Queue.submit(
+      [](sycl::handler &CGH) { CGH.single_task<class TestKernel>([] {}); });
+
+  sycl::program Prog{Queue.get_context()};
+  Prog.build_with_kernel_type<class TestKernel>();
+
+  auto NativeProgram = Prog.get_native<sycl::backend::cuda>();
+
+  assert(NativeProgram != 0);
+
+  // TODO check program interop constructor, once it is available.
+
+  return 0;
+}