global memory debug through command line flag

extras/Hadrons/Environment.hpp

@@ -216,14 +216,21 @@ void Environment::createDerivedObject(const std::string name,
     if (!object_[address].data)
     {
         MemoryStats memStats;
-
+    
-        MemoryProfiler::stats    = &memStats;
+        if (!MemoryProfiler::stats)
+        {
+            MemoryProfiler::stats = &memStats;
+        }
+        size_t initMem           = MemoryProfiler::stats->currentlyAllocated;
         object_[address].storage = storage;
         object_[address].Ls      = Ls;
         object_[address].data.reset(new Holder<B>(new T(std::forward<Ts>(args)...)));
-        object_[address].size    = memStats.maxAllocated;
+        object_[address].size    = MemoryProfiler::stats->maxAllocated - initMem;
         object_[address].type    = &typeid(T);
-        MemoryProfiler::stats    = nullptr;
+        if (MemoryProfiler::stats == &memStats)
+        {
+            MemoryProfiler::stats = nullptr;
+        }
     }
     else
     {

extras/Hadrons/Global.cc

@@ -39,33 +39,6 @@ HadronsLogger Hadrons::HadronsLogMessage(1,"Message");
 HadronsLogger Hadrons::HadronsLogIterative(1,"Iterative");
 HadronsLogger Hadrons::HadronsLogDebug(1,"Debug");
 
-// pretty size formatting //////////////////////////////////////////////////////
-std::string Hadrons::sizeString(long unsigned int bytes)
-
-{
-    constexpr unsigned int bufSize = 256;
-    const char             *suffixes[7] = {"", "K", "M", "G", "T", "P", "E"};
-    char                   buf[256];
-    long unsigned int      s     = 0;
-    double                 count = bytes;
-    
-    while (count >= 1024 && s < 7)
-    {
-        s++;
-        count /= 1024;
-    }
-    if (count - floor(count) == 0.0)
-    {
-        snprintf(buf, bufSize, "%d %sB", (int)count, suffixes[s]);
-    }
-    else
-    {
-        snprintf(buf, bufSize, "%.1f %sB", count, suffixes[s]);
-    }
-    
-    return std::string(buf);
-}
-
 // type utilities //////////////////////////////////////////////////////////////
 constexpr unsigned int maxNameSize = 1024u;
 

extras/Hadrons/Global.hpp

@@ -138,9 +138,6 @@ public:\
 private:\
     name(void) = default;
 
-// pretty size formating
-std::string sizeString(long unsigned int bytes);
-
 // type utilities
 template <typename T>
 const std::type_info * typeIdPt(const T &x)

lib/allocator/AlignedAllocator.cc

@@ -97,4 +97,29 @@ void check_huge_pages(void *Buf,uint64_t BYTES)
 #endif
 }
 
+std::string sizeString(const size_t bytes)
+{
+  constexpr unsigned int bufSize = 256;
+  const char             *suffixes[7] = {"", "K", "M", "G", "T", "P", "E"};
+  char                   buf[256];
+  size_t                 s     = 0;
+  double                 count = bytes;
+  
+  while (count >= 1024 && s < 7)
+  {
+      s++;
+      count /= 1024;
+  }
+  if (count - floor(count) == 0.0)
+  {
+      snprintf(buf, bufSize, "%d %sB", (int)count, suffixes[s]);
+  }
+  else
+  {
+      snprintf(buf, bufSize, "%.1f %sB", count, suffixes[s]);
+  }
+  
+  return std::string(buf);
+}
+
 }

lib/allocator/AlignedAllocator.h

@@ -64,6 +64,8 @@ namespace Grid {
 
   };
   
+  std::string sizeString(size_t bytes);
+
   struct MemoryStats
   {
     size_t totalAllocated{0}, maxAllocated{0}, 
@@ -77,15 +79,20 @@ namespace Grid {
     static bool        debug;
   };
 
+  #define memString(bytes) std::to_string(bytes) + " (" + sizeString(bytes) + ")"
   #define profilerDebugPrint \
   if (MemoryProfiler::stats)\
   {\
     auto s = MemoryProfiler::stats;\
-    std::cout << "[Memory debug] Stats " << MemoryProfiler::stats << std::endl;\
+    std::cout << GridLogDebug << "[Memory debug] Stats " << MemoryProfiler::stats << std::endl;\
-    std::cout << "[Memory debug] Total  : " << s->totalAllocated << "B" << std::endl;\
+    std::cout << GridLogDebug << "[Memory debug] total  : " << memString(s->totalAllocated) \
-    std::cout << "[Memory debug] Max    : " << s->maxAllocated << "B" << std::endl;\
+              << std::endl;\
-    std::cout << "[Memory debug] Current: " << s->totalAllocated << "B" << std::endl;\
+    std::cout << GridLogDebug << "[Memory debug] max    : " << memString(s->maxAllocated) \
-    std::cout << "[Memory debug] Freed  : " << s->totalFreed << "B" << std::endl;\
+              << std::endl;\
+    std::cout << GridLogDebug << "[Memory debug] current: " << memString(s->currentlyAllocated) \
+              << std::endl;\
+    std::cout << GridLogDebug << "[Memory debug] freed  : " << memString(s->totalFreed) \
+              << std::endl;\
   }
 
   #define profilerAllocate(bytes)\
@@ -98,7 +105,7 @@ namespace Grid {
   }\
   if (MemoryProfiler::debug)\
   {\
-    std::cout << "[Memory debug] allocating " << bytes << "B" << std::endl;\
+    std::cout << GridLogDebug << "[Memory debug] allocating " << memString(bytes) << std::endl;\
     profilerDebugPrint;\
   }
 
@@ -111,7 +118,7 @@ namespace Grid {
   }\
   if (MemoryProfiler::debug)\
   {\
-    std::cout << "[Memory debug] freeing " << bytes << "B" << std::endl;\
+    std::cout << GridLogDebug << "[Memory debug] freeing " << memString(bytes) << std::endl;\
     profilerDebugPrint;\
   }
 

lib/util/Init.cc

@@ -204,7 +204,7 @@ std::string GridCmdVectorIntToString(const std::vector<int> & vec){
 // Reinit guard
 /////////////////////////////////////////////////////////
 static int Grid_is_initialised = 0;
-
+static MemoryStats dbgMemStats;
 
 void Grid_init(int *argc,char ***argv)
 {
@@ -251,6 +251,11 @@ void Grid_init(int *argc,char ***argv)
     assert(fp!=(FILE *)NULL);
   }
 
+  if( GridCmdOptionExists(*argv,*argv+*argc,"--debug-mem") ){
+    MemoryProfiler::debug = true;
+    MemoryProfiler::stats = &dbgMemStats;
+  }
+
   ////////////////////////////////////
   // Banner
   ////////////////////////////////////
@@ -324,6 +329,7 @@ void Grid_init(int *argc,char ***argv)
     std::cout<<GridLogMessage<<"  --decomposition : report on default omp,mpi and simd decomposition"<<std::endl;    
     std::cout<<GridLogMessage<<"  --debug-signals : catch sigsegv and print a blame report"<<std::endl;
     std::cout<<GridLogMessage<<"  --debug-stdout  : print stdout from EVERY node"<<std::endl;
+    std::cout<<GridLogMessage<<"  --debug-mem     : print Grid allocator activity"<<std::endl;
     std::cout<<GridLogMessage<<"  --notimestamp   : suppress millisecond resolution stamps"<<std::endl;
     std::cout<<GridLogMessage<<std::endl;
     std::cout<<GridLogMessage<<"Performance:"<<std::endl;