Mothership-facing changes to support hardware idle.

Config/Orchestrator.ocfg

@@ -8,10 +8,10 @@
 
 [Orchestrator_header]
 // All pretty arbitrary; just copied in and stored
-name        = OrchestratorConfiguration
-author      = "MLV and GMB"
-date        = "2021-01-17"
-version     = "0.0.6"
+name        = "OrchestratorConfiguration"
+author      = "MLV"
+date        = "2021-08-27"
+version     = "0.1.0"
 
 // All these may be overridden by the console "path" command
 [default_paths]
@@ -58,6 +58,11 @@ hardware = "../Config/POETSHardwareOneBox.ocfg"
 // Default flags for the cross-compiler
 // build = "\oink -plop ++wheeee  !. <?> "
 
+// Alter the behaviour of the Orchestrator
+[modes]
+single_app_mode = "true"  // Either 'true' or 'false'. Must be true to support
+                          // hardware idle, for now.
+
 // Elaboration messages passed out to the author for errors in processing THIS
 // file They are all classed as "Unrecoverable", not because they are, but
 // because: If you're not a grown-up you shouldn't be mucking about with it

Config/OrchestratorMessages.ocfg

@@ -133,7 +133,7 @@
 // Loading, typelinking
 
 100(U) : "MLV: Allocated but as yet unused"
-101(S) : "Decoder in %s (PID %s) has dropped a packet from %s(rank %s) to %s(rank %s) with key 0x%s"
+101(S) : "Decoder in %s (rank %s) has dropped a packet from %s (rank %s) with key 0x%s."
 102(I) : "Task graph default file path is ||%s||"
 103(I) : "New path is ||%s||"
 104(W) : "Filename %s does not parse correctly"
@@ -450,7 +450,7 @@
 534(I) : "Mothership (rank %s): All devices on this Mothership for application '%s' have stopped."
 535(I) : "Mothership (rank %s): Recalling application '%s'."
 536(I) : "Mothership (rank %s): Application '%s' recalled. This Mothership has forgotten everything about this application."
-537(U) : "MLV: Allocated but as yet unused"
+537(E) : "Mothership: Error decoding MPI message with key '0x%s': Expected bool in field %s. Ignoring message."
 538(U) : "MLV: Allocated but as yet unused"
 539(U) : "MLV: Allocated but as yet unused"
 540(U) : "GMB: Allocated but as yet unused"
@@ -489,8 +489,10 @@
 573(U) : "MLV: Allocated but as yet unused"
 574(U) : "MLV: Allocated but as yet unused"
 575(U) : "MLV: Allocated but as yet unused"
-578(U) : "MLV: Allocated but as yet unused"
-579(U) : "MLV: Allocated but as yet unused"
+576(U) : "MLV: Allocated but as yet unused"
+577(I) : "Mothership: Calling backend->loadAll."
+578(I) : "Mothership: Calling backend->go."
+579(I) : "Mothership: Calling backend->startAll."
 580(E) : "Mothership: Received a log packet with an invalid device index 0x%s."
 581(I) : "Mothership: Consuming log packet from device address 0x%s with name %s."
 582(I) : "Mothership: Received a message containing packets for a supervisor device for application '%s' that is not running (it may be in the process of stopping). Ignoring these packets."

Source/Common/Decode.cpp

@@ -32,12 +32,9 @@ if (CommonBase::FnMap.find(pPkt->Key())!=CommonBase::FnMap.end()) {
                                        // Nope. Kick.
                                        // Pull out the unknown key and post what
                                        // little we know to the LogServer
-Post(101,Sderived,int2str(pPkt->Src()),pPmap->vPmap[pPkt->Src()].P_class,int2str(pPkt->Tgt()),
-     pPmap->vPmap[pPkt->Tgt()].P_class,hex2str(pPkt->Key()));
+Post(101,Sderived,int2str(pPkt->Tgt()),pPmap->vPmap[pPkt->Src()].P_class,
+     int2str(pPkt->Src()),hex2str(pPkt->Key()));
 return 0;                              // Return "keep going" value
 }
 
 //==============================================================================
-
-
-

Source/Common/Pglobals.cpp

@@ -40,8 +40,9 @@ const byte Q::SPEC    = 0x50;
 const byte Q::SUPD    = 0x51;
 const byte Q::INIT    = 0x52;
 const byte Q::CNC     = 0x53;
-const byte Q::ACKt     = 0x54;
+const byte Q::ACKt    = 0x54;
 const byte Q::SUPR    = 0x55;
+const byte Q::EMPT    = 0x56;
 // temporary use: for MPI testing ------------------------------------------
 const byte Q::M0      = 0x60;
 const byte Q::M1      = 0x61;

Source/Common/Pglobals.h

@@ -80,8 +80,12 @@ Mothership
 ----------
 EXIT |-    |-    |-    | (None)
 SYST |KILL |-    |-    | (None)
+APP  |EMPT |-    |-    | (0:string)Code path to broadcast
+                       | (1:string)Data path to broadcast
 APP  |SPEC |-    |-    | (0:string)Application name
                          (1:uint32_t)Number of expected distribution messages
+                         (2:uint8_t)Application number
+                         (3:bool)Hardware-idle application?
 APP  |DIST |-    |-    | (0:string)Application name
                          (1:string)Code path for this core
                          (2:string)Data path for this core
@@ -152,6 +156,7 @@ static const byte INIT;
 static const byte CNC;
 static const byte ACK;
 static const byte SUPR;
+static const byte EMPT;
 // temporary use: for MPI testing ------------------------------------------
 static const byte M0;
 static const byte M1;

Source/Mothership/AppDB.cpp

@@ -1,10 +1,12 @@
 #include "AppDB.h"
 
 /* Checks appInfos for an application of a given name. If it doesn't exist,
- * AppDB creates it and returns a pointer to it (passing distCountExpected to
- * it as an argument). If it already exists, returns a pointer to the existing
- * application, and does not use distCountExpected. */
-AppInfo* AppDB::check_create_app(std::string name, uint32_t distCountExpected)
+ * AppDB creates it and returns a pointer to it (passing distCountExpected and
+ * soloApp to it as an argument). If it already exists, returns a pointer to
+ * the existing application, does not use distCountExpected, but sets
+ * soloApp. */
+AppInfo* AppDB::check_create_app(std::string name, uint32_t distCountExpected,
+                                 bool soloApp)
 {
     AppInfoIt appFinder = appInfos.find(name);
 
@@ -21,7 +23,7 @@ AppInfo* AppDB::check_create_app(std::string name, uint32_t distCountExpected)
         else
         {
             appInfos.insert(std::pair<std::string, AppInfo>
-                            (name, AppInfo(name, distCountExpected)));
+                            (name, AppInfo(name, distCountExpected, soloApp)));
         }
 
         return &(appInfos.find(name)->second);
@@ -37,7 +39,7 @@ AppInfo* AppDB::check_create_app(std::string name, uint32_t distCountExpected)
 /* Sub-synonym. */
 AppInfo* AppDB::check_create_app(std::string name)
 {
-    return check_create_app(name, 0);
+    return check_create_app(name, 0, false);  /* soloApp argument not used. */
 }
 
 /* Checks appInfos for an application of a given name, returning true if such

Source/Mothership/AppDB.h

@@ -22,7 +22,7 @@ class AppDB
     std::map<uint32_t, uint32_t> threadToCoreAddr;
     std::map<uint8_t, std::string> numberToApp;
 
-    AppInfo* check_create_app(std::string, uint32_t);
+    AppInfo* check_create_app(std::string, uint32_t, bool);
     AppInfo* check_create_app(std::string);
     bool check_defined_app(std::string);
     void recall_app(AppInfo*);

Source/Mothership/AppInfo.cpp

@@ -1,9 +1,10 @@
 #include "AppInfo.h"
 
 /* This constructor is used by SPEC messages. */
-AppInfo::AppInfo(std::string name, uint32_t distCountExpected):
+AppInfo::AppInfo(std::string name, uint32_t distCountExpected, bool soloApp):
     name(name),
-    distCountExpected(distCountExpected)
+    distCountExpected(distCountExpected),
+    soloApp(soloApp)
 {
     pendingCommands = 0;
     distCountCurrent = 0;

Source/Mothership/AppInfo.h

@@ -39,11 +39,12 @@ enum AppState{UNDERDEFINED,  /* We're still receiving DIST messages. */
 class AppInfo
 {
 public:
-    AppInfo(std::string nameArg, uint32_t distCountExpected);
+    AppInfo(std::string nameArg, uint32_t distCountExpected, bool soloApp);
     AppInfo(std::string nameArg);
 
     std::string name;
     uint32_t distCountExpected;
+    bool soloApp;
     AppState state;
     std::map<uint32_t, CoreInfo> coreInfos;
     std::set<uint32_t> coresLoaded;

Source/Mothership/AppTransitions.cpp

@@ -58,44 +58,62 @@ void Mothership::initialise_application(AppInfo* app)
                                  meshX, meshY, coreId);
     }
 
-    /* 2: For each core, kick off the threads (mode=false).
-     * 3: For each core, start execution (mode=true). */
-    mode = false;
-    do
+    /* 2: For each core, kick off the threads.
+     * 3: For each core, start execution. */
+
+    /* In solo-app mode, use backend's available global start and go
+     * methods. */
+    if (app->soloApp)
     {
-        for (coreIt = app->coreInfos.begin(); coreIt != app->coreInfos.end();
-             coreIt++)
-        {
-            backend->fromAddr(coreIt->first, &meshX, &meshY, &coreId,
-                              &threadId);
+        debug_post(579, 0);
+        backend->startAll();  /* 2 */
+        debug_post(578, 0);
+        backend->go();  /* 3 */
+    }
 
-            if (!mode)  /* 2 */
-            {
-                debug_post(
-                    587, 4, hex2str(meshX).c_str(), hex2str(meshY).c_str(),
-                    hex2str(coreId).c_str(),
-                    uint2str(coreIt->second.threadsExpected.size()).c_str());
-                /* Note that startOne can hang for the Tinsel backend if the
-                 * number of threads expected is greater than the number of
-                 * threads that the core will start - this is because startOne
-                 * waits for an acknowledgement message from the core that
-                 * varies as a function of the number of threads. If you find
-                 * the above 587 being the last message you see from the MPI
-                 * CNC Resolver thread (for example), this is most likely your
-                 * issue. */
-                backend->startOne(meshX, meshY, coreId,
-                                  coreIt->second.threadsExpected.size());
-            }
-            else  /* 3 */
+    /* Otherwise, things get more complicated... */
+    else
+    {
+        mode = false;  /* 2 when false, 3 when true. */
+        do  /* Simple loop to reduce code duplication (see comment accompanying
+             * this function definition). */
+        {
+            for (coreIt = app->coreInfos.begin();
+                 coreIt != app->coreInfos.end(); coreIt++)
             {
-                debug_post(586, 3, hex2str(meshX).c_str(),
-                           hex2str(meshY).c_str(), hex2str(coreId).c_str());
-                backend->goOne(meshX, meshY, coreId);
+                backend->fromAddr(coreIt->first, &meshX, &meshY, &coreId,
+                                  &threadId);
+
+                if (!mode)  /* 2 */
+                {
+                    debug_post(
+                        587, 4, hex2str(meshX).c_str(), hex2str(meshY).c_str(),
+                        hex2str(coreId).c_str(),
+                        uint2str(coreIt->second.threadsExpected.size())
+                        .c_str());
+                    /* Note that startOne can hang for the Tinsel backend if
+                     * the number of threads expected is greater than the
+                     * number of threads that the core will start - this is
+                     * because startOne waits for an acknowledgement message
+                     * from the core that varies as a function of the number of
+                     * threads. If you find the above 587 being the last
+                     * message you see from the MPI CNC Resolver thread (for
+                     * example), this is most likely your issue. */
+                    backend->startOne(meshX, meshY, coreId,
+                                      coreIt->second.threadsExpected.size());
+                }
+                else  /* 3 */
+                {
+                    debug_post(586, 3, hex2str(meshX).c_str(),
+                               hex2str(meshY).c_str(),
+                               hex2str(coreId).c_str());
+                    backend->goOne(meshX, meshY, coreId);
+                }
             }
-        }
 
-        mode = !mode;
-    } while (mode);
+            mode = !mode;
+        } while (mode);
+    }
 
     /* Good stuff. Now the cores will spin up and send BARRIER messages to the
      * Mothership. */

Source/Mothership/MPIHandlers.cpp

@@ -54,8 +54,8 @@ unsigned Mothership::handle_msg_cnc(PMsg_p* message)
 {
     #if ORCHESTRATOR_DEBUG
     std::string key = "Unknown";
-    if (message->Key() == PMsg_p::KEY(Q::APP,Q::SPEC))
-        key = "Q::APP,Q::SPEC";
+    if (message->Key() == PMsg_p::KEY(Q::APP,Q::EMPT))
+        key = "Q::APP,Q::EMPT";
     else if (message->Key() == PMsg_p::KEY(Q::APP,Q::SPEC))
         key = "Q::APP,Q::SPEC";
     else if (message->Key() == PMsg_p::KEY(Q::APP,Q::DIST))
@@ -79,6 +79,28 @@ unsigned Mothership::handle_msg_cnc(PMsg_p* message)
     return 0;
 }
 
+unsigned Mothership::handle_msg_app_empt(PMsg_p* message)
+{
+    /* Pull message contents. */
+    std::string codePath;
+    std::string dataPath;
+    if (!decode_app_empt_message(message, &codePath, &dataPath))
+    {
+        debug_post(597, 3, "Q::APP,Q::EMPT", hex2str(message->Key()).c_str(),
+                   "Failed to decode.");
+        return 0;
+    }
+
+    debug_post(597, 3, "Q::APP,Q::EMPT", hex2str(message->Key()).c_str(),
+               dformat("codePath=%s, dataPath=%s",
+                       codePath.c_str(), dataPath.c_str()).c_str());
+
+    /* gogogo */
+    debug_post(577, 0);
+    backend->loadAll(codePath.c_str(), dataPath.c_str());
+    return 0;
+}
+
 unsigned Mothership::handle_msg_app_spec(PMsg_p* message)
 {
     AppInfo* appInfo;
@@ -87,21 +109,23 @@ unsigned Mothership::handle_msg_app_spec(PMsg_p* message)
     std::string appName;
     uint32_t distCount;
     uint8_t appNumber;
+    bool soloApp;
     if (!decode_app_spec_message(message, &appName, &distCount,
-                                 &appNumber))
+                                 &appNumber, &soloApp))
     {
         debug_post(597, 3, "Q::APP,Q::SPEC", hex2str(message->Key()).c_str(),
                    "Failed to decode.");
         return 0;
     }
 
     debug_post(597, 3, "Q::APP,Q::SPEC", hex2str(message->Key()).c_str(),
-               dformat("appName=%s, distCount=%u, appNumber=%u",
-                       appName.c_str(), distCount, appNumber).c_str());
+               dformat("appName=%s, distCount=%u, appNumber=%u, soloApp=%s",
+                       appName.c_str(), distCount, appNumber,
+                       soloApp ? "true" : "false").c_str());
 
     /* Ensure application existence idempotently (it might have been created by
      * an AppDist message). */
-    appInfo = appdb.check_create_app(appName, distCount);
+    appInfo = appdb.check_create_app(appName, distCount, soloApp);
 
     /* If the application is not in the UNDERDEFINED state, post bossily and do
      * nothing else. */