fixed issues with new SST core

.gitignore

@@ -37,3 +37,5 @@ bin/compile
 tests/reference/upd
 sstmac/common/config.h.in
 sstmac/common/config.h.in~
+
+*.pyc

acinclude/check_threading.m4

@@ -1,5 +1,9 @@
 
+
 AC_DEFUN([CHECK_THREADING], [
+
+at_least_one_threading=no
+
 # Find out if we want to use pthreads instead of a user-space threads.
 AH_TEMPLATE([HAVE_PTHREAD], [Define to allow use of pthreads])
 AC_ARG_ENABLE(pthread,
@@ -45,6 +49,9 @@ AC_ARG_ENABLE(ucontext,
   ]
 )
 
+if test "X$have_integrated_core" = "Xyes"; then
+AC_MSG_RESULT([pthread virtual thread interface not compatible with unified core])
+else
 if test "$enable_pthread" != no; then
   AC_MSG_CHECKING([whether pthreads is automatically usable])
   AC_LINK_IFELSE(
@@ -86,9 +93,10 @@ if test "$enable_pthread" != no; then
   fi
 
   if test "$enable_pthread" = yes; then
+    at_least_one_threading=yes
     AC_DEFINE(HAVE_PTHREAD)
   fi
-
+fi
 fi
 
 if test "$enable_ucontext" != no; then
@@ -109,6 +117,7 @@ if test "$enable_ucontext" != no; then
       enable_ucontext="yes"
       AC_DEFINE(HAVE_UCONTEXT)
       AM_CONDITIONAL(HAVE_UCONTEXT, true)
+      at_least_one_threading=yes
     ], [
       AC_MSG_RESULT([no])
       AM_CONDITIONAL(HAVE_UCONTEXT, false)
@@ -126,13 +135,11 @@ fi
 # If pth is enabled, but we're not using the builtin version, then
 # test whether or not it works.
 PTH_PATH=none
-if test "$enable_pth" != "no" -a "$enable_pth" != builtin; then
+if test "$enable_pth" != "no"; then
   if test "$enable_pth" != "yes"; then
     CXXFLAGS="$CXXFLAGS -I$enable_pth/include"
     LDFLAGS="$LDFLAGS -L$enable_pth/lib"
     PTH_PATH="$enable_pth"
-  else
-    PTH_PATH="default"
   fi
   dnl First check for a custom pth.
   AC_CHECK_LIB(
@@ -141,6 +148,7 @@ if test "$enable_pth" != "no" -a "$enable_pth" != builtin; then
     [LIBS="-lpth $LIBS"],
     [AC_MSG_ERROR([GNU pth enabled but not found])]
   )
+  at_least_one_threading=yes
   AC_DEFINE(HAVE_PTH)
 fi
 
@@ -166,5 +174,13 @@ elif test "$default_threading" = ucontext -a "$enable_ucontext" != no; then
 elif test "$default_threading" = pthread -a "$enable_pthread" != no; then
   AC_DEFINE(USE_PTHREAD)
 fi
+
+if test "X$at_least_one_threading" = "Xno"; then
+AC_MSG_ERROR([No valid virtual threading interfaces available - must have pth, ucontext, or pthread
+ucontext is not available on Mac OS X
+pthread is not compatible with integrated SST core
+pth must downloaded from https://www.gnu.org/software/pth
+])
+fi
 ])
 

acinclude/sst_core.m4

@@ -22,7 +22,7 @@ AC_ARG_WITH([integrated-core],
       CPPFLAGS="$CPPFLAGS $PY_INCLUDES"
       AC_CHECK_HEADERS([Python.h], [], 
           [AC_MSG_ERROR([Could not locate Python installation needed by SST core])])
-      AC_CHECK_HEADERS([sst/sst_config.h], [],
+      AC_CHECK_HEADERS([sst/core/element.h], [],
           [AC_MSG_ERROR([Could not locate SST core header files at $SST])])
       have_integrated_core="yes"
     ], [

sstmac/hardware/nic/nic.cc

@@ -230,12 +230,7 @@ nic::handle(const sst_message::ptr& msg)
   } else {
     recv_message(msg);
   }
-  // don't use handle for sending anymore
-  //else if (msg->fromaddr() == msg->toaddr()){
-  //  intranode_send(ptr_safe_cast(network_message, msg));
-  //}
-  //else if (msg->fromaddr() == my_addr_){
-  //  internode_send(ptr_safe_cast(network_message, msg));
+
 }
 
 void

sstmac/hardware/node/node.cc

@@ -232,7 +232,6 @@ node::send_to_nic(const network_message::ptr& netmsg)
   } else {
     nic_->internode_send(netmsg);
   }
-  //SCHEDULE_NOW(nic_, netmsg);
 }
 
 #if SSTMAC_INTEGRATED_SST_CORE
@@ -242,6 +241,8 @@ node::launch()
   std::list<sw::launch_message::ptr>::iterator it, end = launchers_.end();
   for (it=launchers_.begin(); it != end; ++it){
     sw::launch_message::ptr lmsg = *it;
+    node_debug("launching task %d on node %d",
+      int(lmsg->tid()), int(addr()));
     os_->handle_message(lmsg);
   }
 }

sstmac/software/process/operating_system.cc

@@ -91,6 +91,25 @@ static bool you_have_been_warned = false;
 bool operating_system::cxa_finalizing_ = false;
 operating_system::os_thread_context operating_system::cxa_finalize_context_;
 
+
+#if SSTMAC_HAVE_PTH
+#define pth_available "pth,"
+#else
+#define pth_available ""
+#endif
+
+#if SSTMAC_HAVE_PTHREAD
+#define pthread_available "pthread,"
+#else
+#define pthread_available ""
+#endif
+
+#if SSTMAC_HAVE_UCONTEXT
+#define ucontext_available "ucontext,"
+#else
+#define ucontext_available ""
+#endif
+
 void
 operating_system::init_threading()
 {
@@ -127,6 +146,11 @@ operating_system::init_threading()
     threading_string = "ucontext";
     #elif defined(SSTMAC_HAVE_PTHREAD)
     threading_string = "pthread";
+    #else
+    #error no valid thread interfaces available (pth, pthread, ucontext supported)
+    #error ucontext is not available on MAC
+    #error pthread is not compatible with integrated SST core
+    #error pth must be downloaded and installed from GNU site
     #endif
 #endif
   }
@@ -178,8 +202,8 @@ operating_system::init_threading()
   }
   else {
     spkt_throw_printf(sprockit::value_error,
-       "operating_system: invalid value %s for SSTMAC_THREADING environmental variable.\n"
-       "choose one of pth, pthread, ucontext.\n",
+       "operating_system: invalid value %s for SSTMAC_THREADING environmental variable\n"
+       "choose one of " pth_available pthread_available ucontext_available,
        threading_string.c_str());
   }
 
@@ -416,8 +440,6 @@ operating_system::current_os_thread_context()
 {
 #if SSTMAC_USE_MULTITHREAD
   int thr = thread_id();
-  //std::cout << sprockit::printf("OS on node %d is on thread %d\n",
-  //  int(my_addr()), thr);
   return os_thread_contexts_[thr];
 #else
   return os_thread_context_;
@@ -470,7 +492,7 @@ operating_system::switch_to_thread(thread_data_t tothread)
   /** back to main thread */
   current_thread_id_ = old_id;
   ctxt.current_thread = old_thread;
-  os_debug("size of threadstack %d, switching back to thread %d\n",
+  os_debug("size of threadstack %d, switching back to thread %d",
     threadstack_.size(), current_thread_id_);
 }
 
@@ -942,6 +964,8 @@ operating_system::add_task(const task_id& id)
 void
 operating_system::start_app(app* theapp)
 {
+  os_debug("starting app %d:%d on thread %d",
+    int(theapp->tid()), int(theapp->aid()), thread_id());
   //this should be called from the actual thread running it
   init_threading();
 #if SSTMAC_HAVE_GRAPHVIZ

sstmac/software/threading/threading_pthread.cc

@@ -10,16 +10,42 @@
  */
 
 #include <sprockit/errors.h>
+#include <sprockit/debug.h>
 #include <sstmac/common/thread_info.h>
 #include <sstmac/common/thread_lock.h>
 #include <sstmac/software/threading/threading_pthread.h>
 
+MakeDebugSlot(pth);
+#define pthread_ctx_debug(ctx,...) \
+  debug_printf(sprockit::dbg::pth, "context=%p,thread=%p: %s", \
+    ctx, pthread_self(), sprockit::printf(__VA_ARGS__).c_str())
+#define pthread_debug(...) \
+   pthread_ctx_debug(&context_, __VA_ARGS__)
+
 namespace sstmac {
 namespace sw {
 #ifdef SSTMAC_HAVE_PTHREAD
 
 std::vector<pthread_mutex_t> threading_pthread::context_switch_mutexes;
 
+void
+send_signal(threadcontext_t* context)
+{
+  context->waiting = false;
+  pthread_cond_signal(&context->ready);
+}
+
+void
+wait_signal(threadcontext_t* context)
+{
+  do {
+    pthread_cond_wait(&context->ready, context->context_switch_lock);
+  }
+  while (context->waiting);
+  //go back to wait mode
+  context->waiting = true;
+}
+
 threading_pthread::threading_pthread(int thread_id, int nthread) :
   thread_id_(thread_id),
   nthread_(nthread)
@@ -29,7 +55,9 @@ threading_pthread::threading_pthread(int thread_id, int nthread) :
   if (context_switch_mutexes.size() == 0){
     //not yet done
     context_switch_mutexes.resize(nthread);
+    pthread_debug("initializing pthread app stack for %d real threads", nthread);
     for (int i=0; i < nthread; ++i){
+      pthread_debug("init context switch mutex %d", i);
       pthread_mutex_init(&context_switch_mutexes[i], NULL);
     }
   }
@@ -45,20 +73,28 @@ pthreadfunc(void*args)
   void* thread_args = thread_info->args;
   void (*func)(void*) = thread_info->func;
 
+  pthread_ctx_debug(thread_info->context, "launched new pthread application stack");
+
   // The following code handles the initial context switch
   // starting this thread. After that, swap_context
   // performs all context switches.
 
   // Grab the context lock.
   pthread_mutex_lock(thread_info->context->context_switch_lock);
 
+  pthread_ctx_debug(thread_info->context,
+    "locked new pthread application stack");
+
   // Put this thread to sleep, if it has not
   // already been started.
   if (!thread_info->context->started) {
-    pthread_cond_wait(&thread_info->context->ready,
-                      thread_info->context->context_switch_lock);
+    pthread_ctx_debug(thread_info->context,
+      "putting thread to sleep");
+    wait_signal(thread_info->context);
   }
 
+  pthread_ctx_debug(thread_info->context, "thread awoken - unlocking mutex");
+
   pthread_mutex_unlock(thread_info->context->context_switch_lock);
 
   //delete the thread info here
@@ -69,7 +105,8 @@ pthreadfunc(void*args)
   (*func)(thread_args);
 
   spkt_throw_printf(sprockit::illformed_error,
-                   "pthreadfunc arrived at end of function. complete_context should have terminated this with pthread_exit");
+    "pthreadfunc arrived at end of function\n"
+    "complete_context should have terminated this with pthread_exit");
 
   return NULL;
 }
@@ -82,6 +119,7 @@ threading_pthread::init_context_common(threadcontext_t &context)
   context_.context_switch_lock = &context_switch_mutexes[thread_id_];
   pthread_cond_init(&context.ready, NULL);
   context.started = false;
+  context.waiting = true;
 }
 
 /// Initialize the context to be that for the currently running thread.
@@ -116,6 +154,8 @@ threading_pthread::start_context(int physical_thread_id,
   init_context_common(context_);
 
   //thread_info::register_kernel_space_virtual_thread(physical_thread_id, &context_.thread, &thread_attr);
+  pthread_debug("starting pthread %p with stack %p of size %lu on physical thread %d",
+       &context_, &context_.thread, stack, stacksize, physical_thread_id);
 
   threadargs *targs = new threadargs(func, args, &context_);
   pthread_create(&context_.thread, &thread_attr, &pthreadfunc, (void*) targs);
@@ -136,7 +176,6 @@ threading_pthread::complete_context(threading_interface *to)
   else {
     spkt_throw_printf(sprockit::illformed_error,
                      "received non-pthread context on complete_context");
-    //pthread_cond_signal(NULL);
   }
   pthread_exit(0);
 }
@@ -161,9 +200,6 @@ threading_pthread::swap_context(threading_interface *to)
   // (lock) c2 signals c1 to start
   // c2 begins waiting on signals (unlock)
 
-  // Grab the context lock.
-  pthread_mutex_lock(context_.context_switch_lock);
-
 #if SSTMAC_SANITY_CHECK
   threading_pthread *casted = dynamic_cast<threading_pthread*>(to);
   if (!casted){
@@ -174,14 +210,21 @@ threading_pthread::swap_context(threading_interface *to)
   threading_pthread *casted = static_cast<threading_pthread*>(to);
 #endif
 
+  // Grab the context lock.
+  pthread_debug("locking context switch to start swap to thread");
+  pthread_mutex_lock(context_.context_switch_lock);
+
   casted->context_.started = true;
 
   // Wake up the to context.
-  pthread_cond_signal(&casted->context_.ready);
+  pthread_debug("cond signal to cause swap context");
+  send_signal(&casted->context_);
 
   // Put the from thread to sleep.
-  pthread_cond_wait(&context_.ready, context_.context_switch_lock);
+  pthread_debug("wait on cond in swap context");
+  wait_signal(&context_);
 
+  pthread_debug("resuming, unlocking context switch");
   pthread_mutex_unlock(context_.context_switch_lock);
 }
 

sstmac/software/threading/threading_pthread.h

@@ -29,13 +29,13 @@ namespace sw {
 #ifdef SSTMAC_HAVE_PTHREAD
 
 
-class threadcontext_t
+struct threadcontext_t
 {
- public:
   pthread_t thread;
   pthread_cond_t ready;
   pthread_mutex_t* context_switch_lock;
   bool started;
+  bool waiting;
 };
 
 class threading_pthread : public threading_interface
@@ -78,7 +78,6 @@ class threading_pthread : public threading_interface
 
   static std::vector<pthread_mutex_t> context_switch_mutexes;
 
-
   virtual
   threading_interface* copy() {
     return new threading_pthread(thread_id_, nthread_);

tests/reference/test_core_apps_ping_all_crossbar_valiant.sst-out

@@ -0,0 +1,3 @@
+Runtime = 5000.1002ms
+*** Event queue empty, exiting simulation... ***
+Simulation is complete, simulated time: 18.4467 Ms

tests/reference/test_core_apps_ping_all_dragonfly_valiant.sst-out

@@ -1,3 +1,3 @@
-Runtime = 5000.1292ms
+Runtime = 5000.1283ms
 *** Event queue empty, exiting simulation... ***
 Simulation is complete, simulated time: 18.4467 Ms

tests/reference/test_core_apps_ping_all_null_arb.sst-out

@@ -0,0 +1,3 @@
+Runtime = 5000.0631ms
+*** Event queue empty, exiting simulation... ***
+Simulation is complete, simulated time: 18.4467 Ms

tests/reference/test_core_apps_ping_all_torus_pflow.sst-out

@@ -1,3 +1,3 @@
-Runtime = 5000.0635ms
+Runtime = 5000.0631ms
 *** Event queue empty, exiting simulation... ***
 Simulation is complete, simulated time: 18.4467 Ms