Make adios_iotest work with SST again: add CurrentStep() to SstWriter…

docs/user_guide/source/engines/sst.rst

@@ -157,7 +157,7 @@ the underlying network communication mechanism to use for exchanging
 data in SST.  Generally this is chosen by SST based upon what is
 available on the current platform.  However, specifying this engine
 parameter allows overriding SST's choice.  Current allowed values are
-**"UCX"**,**"MPI"**, **"RDMA"**, and **"WAN"**.  (**ib** and **fabric** are accepted as
+**"UCX"**, **"MPI"**, **"RDMA"**, and **"WAN"**.  (**ib** and **fabric** are accepted as
 equivalent to **RDMA** and **evpath** is equivalent to **WAN**.)
 Generally both the reader and writer should be using the same network
 transport, and the network transport chosen may be dictated by the
@@ -219,7 +219,7 @@ plane, this parameter controls the libfabric interface choice.
 12. ``FirstTimestepPrecious``: Default **FALSE**.
 FirstTimestepPrecious is a boolean parameter that affects the queueing
 of the first timestep presented to the SST Writer engine. If
-FirstTimestepPrecious is **TRUE***, then the first timestep is
+FirstTimestepPrecious is **TRUE**, then the first timestep is
 effectively never removed from the output queue and will be presented
 as a first timestep to any reader that joins at a later time.  This
 can be used to convey run parameters or other information that every
@@ -235,7 +235,7 @@ This value is interpreted by only by the SST Writer engine.
 13. ``AlwaysProvideLatestTimestep``: Default **FALSE**.
 AlwaysProvideLatestTimestep is a boolean parameter that affects what
 of the available timesteps will be provided to the reader engine.  If
-AlwaysProvideLatestTimestep is **TRUE***, then if there are multiple
+AlwaysProvideLatestTimestep is **TRUE**, then if there are multiple
 timesteps available to the reader, older timesteps will be skipped and
 the reader will see only the newest available upon BeginStep.
 This value is interpreted by only by the SST Reader engine.

source/adios2/engine/sst/SstWriter.cpp

@@ -183,6 +183,8 @@ StepStatus SstWriter::BeginStep(StepMode mode, const float timeout_sec)
     return StepStatus::OK;
 }
 
+size_t SstWriter::CurrentStep() const { return m_WriterStep; }
+
 void SstWriter::MarshalAttributes()
 {
     PERFSTUBS_SCOPED_TIMER_FUNC();

source/adios2/engine/sst/SstWriter.h

@@ -38,6 +38,7 @@ class SstWriter : public Engine
 
     StepStatus BeginStep(StepMode mode,
                          const float timeoutSeconds = -1.0) final;
+    size_t CurrentStep() const final;
     void PerformPuts() final;
     void EndStep() final;
     void Flush(const int transportIndex = -1) final;

source/adios2/toolkit/format/bp/bp4/BP4Serializer.tcc

@@ -339,7 +339,8 @@ BP4Serializer::GetBPStats(const bool singleValue,
     if (blockInfo.MemSpace == MemorySpace::CUDA)
     {
         const size_t size = helper::GetTotalSize(blockInfo.Count);
-        helper::CUDAMinMax(blockInfo.Data, size, stats.Min, stats.Max);
+        if (!std::is_same<T, long double>::value)
+            helper::CUDAMinMax(blockInfo.Data, size, stats.Min, stats.Max);
         return stats;
     }
 #endif

source/adios2/toolkit/format/bp5/BP5Serializer.cpp

@@ -606,8 +606,9 @@ static void GetMinMax(const void *Data, size_t ElemCount, const DataType Type,
              Type == helper::GetDataType<T>())                                 \
     {                                                                          \
         const T *values = (const T *)Data;                                     \
-        helper::CUDAMinMax(values, ElemCount, MinMax.MinUnion.field_##N,       \
-                           MinMax.MaxUnion.field_##N);                         \
+        if (!std::is_same<T, long double>::value)                              \
+            helper::CUDAMinMax(values, ElemCount, MinMax.MinUnion.field_##N,   \
+                               MinMax.MaxUnion.field_##N);                     \
     }
     ADIOS2_FOREACH_MINMAX_STDTYPE_2ARGS(pertype)
 #undef pertype

source/utils/adios_iotest/adios_iotest.cpp

@@ -18,23 +18,27 @@
 
 int main(int argc, char *argv[])
 {
+    Settings settings;
+
+    /* Check input arguments. Quit if something is wrong. */
+    if (settings.processArguments(argc, argv))
+    {
+        return 1;
+    }
+
     int provided;
     int threadSupportLevel = MPI_THREAD_SINGLE;
-
-    if (std::string(argv[1]) == "SST")
+    if (settings.multithreadedMPI)
     {
         threadSupportLevel = MPI_THREAD_MULTIPLE;
     }
-
     MPI_Init_thread(&argc, &argv, threadSupportLevel, &provided);
 
-    Settings settings;
+    settings.initDecomp(MPI_COMM_WORLD);
 
-    /* Check input arguments. Quit if something is wrong. */
-    if (settings.processArguments(argc, argv, MPI_COMM_WORLD) ||
-        settings.extraArgumentChecks())
+    // MPI-dependent argument checks
+    if (settings.extraArgumentChecks())
     {
-        MPI_Finalize();
         return 1;
     }
 

source/utils/adios_iotest/settings.cpp

@@ -31,12 +31,13 @@ struct option options[] = {{"help", no_argument, NULL, 'h'},
                            {"weak-scaling", no_argument, NULL, 'w'},
                            {"timer", no_argument, NULL, 't'},
                            {"fixed", no_argument, NULL, 'F'},
+                           {"multithreaded-mpi", no_argument, NULL, 'T'},
 #ifdef ADIOS2_HAVE_HDF5_PARALLEL
                            {"hdf5", no_argument, NULL, 'H'},
 #endif
                            {NULL, 0, NULL, 0}};
 
-static const char *optstring = "-hvswtFHa:c:d:D:x:p:";
+static const char *optstring = "-hvswtTFHa:c:d:D:x:p:";
 
 size_t Settings::ndigits(size_t n) const
 {
@@ -74,6 +75,7 @@ void Settings::displayHelp()
         << "  -v         increase verbosity\n"
         << "  -h         display this help\n"
         << "  -F         turn on fixed I/O pattern explicitly\n"
+        << "  -T         turn on multi-threaded MPI (needed by SST/MPI)\n"
         << "  -p         specify the path of the output explicitly\n"
         << "  -t         print and dump the timing measured by the I/O "
            "timer\n\n";
@@ -206,6 +208,9 @@ int Settings::processArgs(int argc, char *argv[])
         case 'F':
             fixedPattern = true;
             break;
+        case 'T':
+            multithreadedMPI = true;
+            break;
         case 'h':
             if (!myRank)
             {
@@ -318,13 +323,27 @@ int Settings::processArgs(int argc, char *argv[])
     return 0;
 }
 
-int Settings::processArguments(int argc, char *argv[], MPI_Comm worldComm)
+int Settings::processArguments(int argc, char *argv[])
 {
     int retval = 0;
     try
     {
         retval = processArgs(argc, argv);
+    }
+    catch (std::exception &e) // command-line argument errors
+    {
+        std::cout << "ERROR : " << e.what() << std::endl;
+        displayHelp();
+        retval = 1;
+    }
+    return retval;
+}
 
+int Settings::initDecomp(MPI_Comm worldComm)
+{
+    int retval = 0;
+    try
+    {
         int wrank;
         MPI_Comm_rank(worldComm, &wrank);
         MPI_Comm_split(worldComm, static_cast<int>(appId), wrank, &appComm);

source/utils/adios_iotest/settings.h

@@ -42,6 +42,7 @@ class Settings
     bool ioTimer = false;        // used to measure io time
     bool fixedPattern = false;   // should Lock definitions?
     bool isRatioDecomp = false;
+    bool multithreadedMPI = false; // turn on MT-enabled MPI
     IOLib iolib = IOLib::ADIOS;
     //   process decomposition
     std::vector<size_t> processDecomp = {1, 1, 1, 1, 1, 1, 1, 1,
@@ -56,7 +57,8 @@ class Settings
 
     Settings() = default;
     ~Settings() = default;
-    int processArguments(int argc, char *argv[], MPI_Comm worldComm);
+    int processArguments(int argc, char *argv[]);
+    int initDecomp(MPI_Comm worldComm);
     int extraArgumentChecks();
     size_t stringToNumber(const std::string &varName, const char *arg) const;
     int parseCSDecomp(const char *arg);