Subzero: Refine the memory usage report with -szstats .

src/IceGlobalContext.cpp

 //===- subzero/src/IceGlobalContext.cpp - Global context defs -------------===//
 //
 //                        The Subzero Code Generator
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 ///
 /// \file
@@ skipping 226 matching lines @@
   }
   if (BuildDefs::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);
   }
 }
 
-void GlobalContext::CodeStats::dump(const std::string &Name,
-                                    GlobalContext *Ctx) {
+void GlobalContext::CodeStats::dump(const Cfg *Func, GlobalContext *Ctx) {
   if (!BuildDefs::dump())
     return;
   OstreamLocker _(Ctx);
   Ostream &Str = Ctx->getStrDump();
+  const std::string Name =
+      (Func == nullptr ? "_FINAL_" : Func->getFunctionNameAndSize());
 #define X(str, tag)                                                            \
   Str << "|" << Name << "|" str "|" << Stats[CS_##tag] << "\n";
   CODESTATS_TABLE
 #undef X
   Str << "|" << Name << "|Spills+Fills|"
       << Stats[CS_NumSpills] + Stats[CS_NumFills] << "\n";
   Str << "|" << Name << "|Memory Usage|";
   if (ssize_t MemUsed = llvm::TimeRecord::getCurrentTime(false).getMemUsed())
     Str << MemUsed;
   else
     Str << "(requires '-track-memory')";
   Str << "\n";
   Str << "|" << Name << "|CPool Sizes ";
   {
     auto Pool = Ctx->getConstPool();
     Str << "|f32=" << Pool->Floats.size();
     Str << "|f64=" << Pool->Doubles.size();
     Str << "|i1=" << Pool->Integers1.size();
     Str << "|i8=" << Pool->Integers8.size();
     Str << "|i16=" << Pool->Integers16.size();
     Str << "|i32=" << Pool->Integers32.size();
     Str << "|i64=" << Pool->Integers64.size();
     Str << "|Rel=" << Pool->Relocatables.size();
     Str << "|ExtRel=" << Pool->ExternRelocatables.size();
   }
   Str << "\n";
+  if (Func != nullptr) {
+    Str << "|" << Name << "|Cfg Memory  |" << Func->getTotalMemoryMB()
+        << " MB\n";
+  }
 }
 
 GlobalContext::GlobalContext(Ostream *OsDump, Ostream *OsEmit, Ostream *OsError,
                              ELFStreamer *ELFStr)
     : Strings(new StringPool()), ConstPool(new ConstantPool()), ErrorStatus(),
       StrDump(OsDump), StrEmit(OsEmit), StrError(OsError), IntrinsicsInfo(this),
       ObjectWriter(), OptQ(/*Sequential=*/Flags.isSequential(),
                            /*MaxSize=*/
                            (Flags.getParseParallel() && Flags.getBuildOnRead())
                                ? MaxOptQSize
@@ skipping 84 matching lines @@
     } else {
       Func->getAssembler<>()->setInternal(Func->getInternal());
       switch (getFlags().getOutFileType()) {
       case FT_Elf:
       case FT_Iasm: {
         Func->emitIAS();
         // The Cfg has already emitted into the assembly buffer, so
         // stats have been fully collected into this thread's TLS.
         // Dump them before TLS is reset for the next Cfg.
         if (BuildDefs::dump())
-          dumpStats(Func->getFunctionNameAndSize());
+          dumpStats(Func.get());
         auto Asm = Func->releaseAssembler();
         // Copy relevant fields into Asm before Func is deleted.
         Asm->setFunctionName(Func->getFunctionName());
         Item = makeUnique<EmitterWorkItem>(Func->getSequenceNumber(),
                                            std::move(Asm));
         Item->setGlobalInits(Func->getGlobalInits());
       } break;
       case FT_Asm:
         // The Cfg has not been emitted yet, so stats are not ready
         // to be dumped.
@@ skipping 233 matching lines @@
         lowerGlobalsIfNoCodeHasBeenSeen();
         accumulateGlobals(Item->getGlobalInits());
 
         assert(getFlags().getOutFileType() == FT_Asm);
         std::unique_ptr<Cfg> Func = Item->getCfg();
         // Unfortunately, we have to temporarily install the Cfg in TLS
         // because Variable::asType() uses the allocator to create the
         // differently-typed copy.
         CfgLocalAllocatorScope _(Func.get());
         Func->emit();
-        dumpStats(Func->getFunctionNameAndSize());
+        dumpStats(Func.get());
       } break;
       }
     }
     // Update the start index for next shuffling queue
     ShuffleStartIndex = ShuffleEndIndex;
   }
 
   // In case there are no code to be generated, we invoke the conditional
   // lowerGlobals again -- this is a no-op if code has been emitted.
   lowerGlobalsIfNoCodeHasBeenSeen();
@@ skipping 289 matching lines @@
   }
   if (getFlags().isSequential())
     emitItems();
 }
 
 std::unique_ptr<EmitterWorkItem> GlobalContext::emitQueueBlockingPop() {
   TimerMarker _(TimerStack::TT_qEmitPop, this);
   return EmitQ.blockingPop();
 }
 
-void GlobalContext::dumpStats(const std::string &Name, bool Final) {
+void GlobalContext::dumpStats(const Cfg *Func) {
   if (!getFlags().getDumpStats())
     return;
-  if (Final) {
-    getStatsCumulative()->dump(Name, this);
+  if (Func == nullptr) {
+    getStatsCumulative()->dump(Func, this);
   } else {
-    ICE_TLS_GET_FIELD(TLS)->StatsFunction.dump(Name, this);
+    ICE_TLS_GET_FIELD(TLS)->StatsFunction.dump(Func, this);
   }
 }
 
 void GlobalContext::dumpTimers(TimerStackIdT StackID, bool DumpCumulative) {
   if (!BuildDefs::timers())
     return;
   auto Timers = getTimers();
   assert(Timers->size() > StackID);
   OstreamLocker L(this);
   Timers->at(StackID).dump(getStrDump(), DumpCumulative);
@@ skipping 34 matching lines @@
   Ctx = Func->getContext();
   Active =
       Func->getFocusedTiming() || Ctx->getFlags().getSubzeroTimingEnabled();
   if (Active)
     Ctx->pushTimer(ID, StackID);
 }
 
 ICE_TLS_DEFINE_FIELD(GlobalContext::ThreadContext *, GlobalContext, TLS);
 
 } // end of namespace Ice