Fix STF compilation with nvc++ as host compiler (#8230)

Author: davebayer

cudax/include/cuda/experimental/__stf/internal/launch.cuh

@@ -181,16 +181,16 @@ public:
   template <typename Fun>
   void operator->*(Fun&& f)
   {
-#  if __NVCOMPILER
+#  if _CCCL_CUDA_COMPILER(NVHPC)
     // With nvc++, all lambdas can run on host and device.
     static constexpr bool is_extended_host_device_lambda_closure_type = true,
                           is_extended_device_lambda_closure_type      = false;
-#  else
+#  else // ^^^ _CCCL_CUDA_COMPILER(NVHPC) ^^^ / VVV !_CCCL_CUDA_COMPILER(NVHPC) VVV
     // With nvcpp, dedicated traits tell how a lambda can be executed.
     static constexpr bool is_extended_host_device_lambda_closure_type =
                             __nv_is_extended_host_device_lambda_closure_type(Fun),
                           is_extended_device_lambda_closure_type = __nv_is_extended_device_lambda_closure_type(Fun);
-#  endif
+#  endif // ^^^ !_CCCL_CUDA_COMPILER(NVHPC) ^^^
 
     static_assert(is_extended_host_device_lambda_closure_type || is_extended_device_lambda_closure_type,
                   "Cannot run launch() on the host");
@@ -306,16 +306,16 @@ public:
   template <typename Fun>
   void operator->*(Fun&& f)
   {
-#  if __NVCOMPILER
+#  if _CCCL_CUDA_COMPILER(NVHPC)
     // With nvc++, all lambdas can run on host and device.
     static constexpr bool is_extended_host_device_lambda_closure_type = true,
                           is_extended_device_lambda_closure_type      = false;
-#  else
+#  else // ^^^ _CCCL_CUDA_COMPILER(NVHPC) ^^^ / VVV !_CCCL_CUDA_COMPILER(NVHPC) VVV
     // With nvcpp, dedicated traits tell how a lambda can be executed.
     static constexpr bool is_extended_host_device_lambda_closure_type =
                             __nv_is_extended_host_device_lambda_closure_type(Fun),
                           is_extended_device_lambda_closure_type = __nv_is_extended_device_lambda_closure_type(Fun);
-#  endif
+#  endif // ^^^ !_CCCL_CUDA_COMPILER(NVHPC) ^^^
 
     static_assert(is_extended_device_lambda_closure_type || is_extended_host_device_lambda_closure_type,
                   "Cannot run launch() on the host");

cudax/include/cuda/experimental/__stf/internal/parallel_for_scope.cuh

@@ -615,16 +615,16 @@ public:
 
     static constexpr bool need_reduction = (deps_ops_t::does_work || ...);
 
-#  if __NVCOMPILER
+#  if _CCCL_CUDA_COMPILER(NVHPC)
     // With nvc++, all lambdas can run on host and device.
     static constexpr bool is_extended_host_device_lambda_closure_type = true,
                           is_extended_device_lambda_closure_type      = false;
-#  else
+#  else // ^^^ _CCCL_CUDA_COMPILER(NVHPC) ^^^ / vvv !_CCCL_CUDA_COMPILER(NVHPC)
     // With nvcpp, dedicated traits tell how a lambda can be executed.
     static constexpr bool is_extended_host_device_lambda_closure_type =
                             __nv_is_extended_host_device_lambda_closure_type(Fun),
                           is_extended_device_lambda_closure_type = __nv_is_extended_device_lambda_closure_type(Fun);
-#  endif
+#  endif // ^^^ !_CCCL_CUDA_COMPILER(NVHPC) ^^^
 
     // TODO redo cascade of tests
     if constexpr (need_reduction)