Subzero: Fix stats collection and output for multithreading.

src/IceGlobalContext.h

Index: src/IceGlobalContext.h
diff --git a/src/IceGlobalContext.h b/src/IceGlobalContext.h
index 04f08a2dbffa3191633fb39343b9e6e03f1ef1e0..e8e066960b199507f5e98c37b7e26234d764c2e6 100644
--- a/src/IceGlobalContext.h
+++ b/src/IceGlobalContext.h
@@ -73,6 +73,13 @@ class GlobalContext {
     void updateFrameBytes(uint32_t Bytes) { FrameBytes += Bytes; }
     void updateSpills() { ++Spills; }
     void updateFills() { ++Fills; }
+    void add(const CodeStats &Other) {
+      InstructionsEmitted += Other.InstructionsEmitted;
+      RegistersSaved += Other.RegistersSaved;
+      FrameBytes += Other.FrameBytes;
+      Spills += Other.Spills;
+      Fills += Other.Fills;
+    }
     void dump(const IceString &Name, Ostream &Str);
 
   private:
@@ -121,6 +128,7 @@ class GlobalContext {
   public:
     ThreadContext() {}
     CodeStats StatsFunction;
+    CodeStats StatsCumulative;
     TimerList Timers;
   };
 
@@ -210,31 +218,31 @@ public:
     if (!ALLOW_DUMP || !getFlags().DumpStats)
       return;
     ICE_TLS_GET_FIELD(TLS)->StatsFunction.updateEmitted(InstCount);
-    getStatsCumulative()->updateEmitted(InstCount);
+    ICE_TLS_GET_FIELD(TLS)->StatsCumulative.updateEmitted(InstCount);
   }
   void statsUpdateRegistersSaved(uint32_t Num) {
     if (!ALLOW_DUMP || !getFlags().DumpStats)
       return;
     ICE_TLS_GET_FIELD(TLS)->StatsFunction.updateRegistersSaved(Num);
-    getStatsCumulative()->updateRegistersSaved(Num);
+    ICE_TLS_GET_FIELD(TLS)->StatsCumulative.updateRegistersSaved(Num);
   }
   void statsUpdateFrameBytes(uint32_t Bytes) {
     if (!ALLOW_DUMP || !getFlags().DumpStats)
       return;
     ICE_TLS_GET_FIELD(TLS)->StatsFunction.updateFrameBytes(Bytes);
-    getStatsCumulative()->updateFrameBytes(Bytes);
+    ICE_TLS_GET_FIELD(TLS)->StatsCumulative.updateFrameBytes(Bytes);
   }
   void statsUpdateSpills() {
     if (!ALLOW_DUMP || !getFlags().DumpStats)
       return;
     ICE_TLS_GET_FIELD(TLS)->StatsFunction.updateSpills();
-    getStatsCumulative()->updateSpills();
+    ICE_TLS_GET_FIELD(TLS)->StatsCumulative.updateSpills();
   }
   void statsUpdateFills() {
     if (!ALLOW_DUMP || !getFlags().DumpStats)
       return;
     ICE_TLS_GET_FIELD(TLS)->StatsFunction.updateFills();
-    getStatsCumulative()->updateFills();
+    ICE_TLS_GET_FIELD(TLS)->StatsCumulative.updateFills();

[JF, 2015/01/31 05:08:24]

Can you get the field only once per function instead of twice? I'm not sure the
compiler can eliminate redundant TLS fetches in all circumstances.

[Jim Stichnoth, 2015/01/31 05:36:59]

Nice, done.

   }
 
   // These are predefined TimerStackIdT values.
@@ -296,6 +304,13 @@ public:
       for (ThreadContext *TLS : AllThreadContexts)
         Timers->mergeFrom(TLS->Timers);
     }
+    if (ALLOW_DUMP) {
+      // Do a separate loop over AllThreadContexts to avoid holding
+      // two locks at once.
+      auto Stats = getStatsCumulative();
+      for (ThreadContext *TLS : AllThreadContexts)
+        Stats->add(TLS->StatsCumulative);
+    }
   }
 
   // Translation thread startup routine.