add polling example using stdexec (#62)

* add polling example using stdexec

* harmonize printouts between polling examamples

Author: m-fila

examples/CMakeLists.txt

@@ -81,6 +81,8 @@ if(BUILD_CUDA AND BUILD_TBB)
     target_link_libraries(exec_reco_stdexec PRIVATE CoroutineTests CUDA::cudart TBB::tbb STDEXEC::stdexec)
     add_executable(exec_delegate_stdexec exec_delegate.cpp)
     target_link_libraries(exec_delegate_stdexec PRIVATE CoroutineTests CUDA::cudart TBB::tbb STDEXEC::stdexec)
+    add_executable(exec_event_poll_stdexec exec_event_poll.cpp)
+    target_link_libraries(exec_event_poll_stdexec PRIVATE CoroutineTests CUDA::cudart TBB::tbb STDEXEC::stdexec)
   endif()
   if(BUILD_CAPY)
     add_executable(capy_reco capy_reco.cpp)

examples/README.md

@@ -182,6 +182,6 @@ These examples are a modification of [reconstruction examples](#reconstruction)
 
 ## Event poll
 
-Link: [alien_event_poll.cpp](alien_event_poll.cpp), [capy_event_poll.cpp](capy_event_poll.cpp)
+Link: [alien_event_poll.cpp](alien_event_poll.cpp), [exec_event_poll.cpp](exec_event_poll.cpp), [capy_event_poll.cpp](capy_event_poll.cpp)
 
 These examples are a variant of [delegate examples](#delegate) in which awaiting completion of CUDA operations is done by repeatedly querying the event state instead of using a callback as in earlier examples. All the queries are executed by a specific thread.

examples/alien_event_poll.cpp

@@ -50,14 +50,15 @@ struct Retry {
     void await_resume() const noexcept {}
 };
 
-tool::Task<void> poll(cudaEvent_t event) {
+tool::Task<void> poll(cudaEvent_t event, std::string_view parent) {
     auto status = cudaSuccess;
-    log() << "Polling for event completion..." << std::endl;
+    log(parent) << "Polling for event completion..." << std::endl;
     while ((status = cudaEventQuery(event)) == cudaErrorNotReady) {
-        log() << "Event not ready, retrying..." << std::endl;
+        log(parent) << "Event not ready, retrying..." << std::endl;
         co_await Retry{};
     }
     ERROR_CHECK_CUDA(status);
+    log(parent) << "Event completed successfully" << std::endl;
 }
 
 subtool::Task<DeviceBuffer<int>> clusterization(
@@ -83,7 +84,7 @@ subtool::Task<DeviceBuffer<int>> clusterization(
             ERROR_CHECK_CUDA(cudaEventRecord(event, stream));
         }));
 
-    co_await schedule_on(delegation_scheduler, poll(event));
+    co_await schedule_on(delegation_scheduler, poll(event, self));
 
     auto nClusters = 0;
     for (auto v : h_cells)
@@ -136,7 +137,7 @@ subtool::Task<DeviceBuffer<int>> seeding(
             ERROR_CHECK_CUDA(cudaEventRecord(event, stream));
         }));
 
-    co_await schedule_on(delegation_scheduler, poll(event));
+    co_await schedule_on(delegation_scheduler, poll(event, self));
 
     int nSeeds = 0;
     for (auto v : h_clusters)
@@ -200,7 +201,7 @@ tool::Task<tool::StatusCode> reconstruct(
             ERROR_CHECK_CUDA(cudaEventRecord(event, stream));
         }));
 
-    co_await schedule_on(delegation_scheduler, poll(event));
+    co_await schedule_on(delegation_scheduler, poll(event, self));
 
     log(self) << "Finishing reconstruction" << std::endl;
     co_return tool::StatusCode::SUCCESS;

examples/capy_event_poll.cpp

@@ -51,14 +51,15 @@ struct Retry {
     void await_resume() const noexcept {}
 };
 
-boost::capy::task<void> poll(cudaEvent_t event) {
+boost::capy::task<void> poll(cudaEvent_t event, std::string_view parent) {
     auto status = cudaSuccess;
-    log() << "Polling for event completion..." << std::endl;
+    log(parent) << "Polling for event completion..." << std::endl;
     while ((status = cudaEventQuery(event)) == cudaErrorNotReady) {
-        log() << "Event not ready, retrying..." << std::endl;
+        log(parent) << "Event not ready, retrying..." << std::endl;
         co_await Retry{};
     }
     ERROR_CHECK_CUDA(status);
+    log(parent) << "Event completed successfully" << std::endl;
 }
 
 boost::capy::task<DeviceBuffer<int>> clusterization(
@@ -81,7 +82,8 @@ boost::capy::task<DeviceBuffer<int>> clusterization(
         ERROR_CHECK_CUDA(cudaEventRecord(event, stream));
     }));
 
-    co_await boost::capy::run(delegation_thread.get_executor())(poll(event));
+    co_await boost::capy::run(delegation_thread.get_executor())(
+        poll(event, self));
 
     auto nClusters = 0;
     for (auto v : h_cells)
@@ -129,7 +131,8 @@ boost::capy::task<DeviceBuffer<int>> seeding(
         ERROR_CHECK_CUDA(cudaEventRecord(event, stream));
     }));
 
-    co_await boost::capy::run(delegation_thread.get_executor())(poll(event));
+    co_await boost::capy::run(delegation_thread.get_executor())(
+        poll(event, self));
 
     int nSeeds = 0;
     for (auto v : h_clusters)
@@ -188,7 +191,8 @@ boost::capy::task<tools::StatusCode> reconstruct(
         ERROR_CHECK_CUDA(cudaEventRecord(event, stream));
     }));
 
-    co_await boost::capy::run(delegation_thread.get_executor())(poll(event));
+    co_await boost::capy::run(delegation_thread.get_executor())(
+        poll(event, self));
 
     log(self) << "Finishing reconstruction" << std::endl;
     co_return tools::StatusCode::SUCCESS;