Add test for dry runs

include/runtime.h

@@ -79,6 +79,8 @@ namespace detail {
 
 		bool is_dry_run() const { return cfg->is_dry_run().value_or(false); }
 
+		int get_command_count() const;
+
 	  private:
 		inline static bool mpi_initialized = false;
 		inline static bool mpi_finalized = false;

src/runtime.cc

@@ -283,5 +283,7 @@ namespace detail {
 		if(done) { active_flushes.pop_front(); }
 	}
 
+	int runtime::get_command_count() const { return cdag->command_count(); }
+
 } // namespace detail
 } // namespace celerity

test/runtime_tests.cc

@@ -992,5 +992,38 @@ namespace detail {
 		});
 	}
 
+	void dry_run_with_nodes(const int nodes) {
+		const std::string dryrun_envvar_name = "CELERITY_DRYRUN_NODES";
+		setenv(dryrun_envvar_name.c_str(), std::to_string(nodes).c_str(), 1);
+
+		distr_queue q;
+
+		auto& rt = runtime::get_instance();
+		auto& tm = rt.get_task_manager();
+		tm.set_horizon_step(2);
+
+		CHECK(rt.is_dry_run());
+
+		buffer<int, 1> buff{nodes * 2};
+		q.submit([=](handler& cgh) {
+			accessor acc{buff, cgh, celerity::access::one_to_one{}, write_only};
+			cgh.parallel_for<class UKN(dryrun)>(buff.get_range(), [=](item<1> item) { acc[item.get_id()] = item.get_id(); });
+		});
+		q.slow_full_sync();
+
+		// (intial epoch + task + sync epoch) per node.
+		CHECK(rt.get_command_count() == 3 * nodes);
+		test_utils::maybe_print_graph(tm);
+	}
+
+	TEST_CASE_METHOD(test_utils::runtime_fixture, "dry run", "[dryrun]") {
+		SECTION("with 2 nodes") { dry_run_with_nodes(2); }
+		SECTION("with 8 nodes") { dry_run_with_nodes(8); }
+		SECTION("with 16 nodes") { dry_run_with_nodes(16); }
+		SECTION("with 32 nodes") { dry_run_with_nodes(32); }
+		SECTION("with 64 nodes") { dry_run_with_nodes(64); }
+		SECTION("with 128 nodes") { dry_run_with_nodes(128); }
+	}
+
 } // namespace detail
 } // namespace celerity