mtmaps (OpenMP)

[luciolis]

Hello,

A recent and recurrent crash in warsong gulch

http://pastebin.com/YcFs7P9R

I don't know if it is related with ACE, but the backtrace mention it.

[divisorxx]

Wojta - probably not related to mtmaps. I am using last patch poseted by infinity + mutex around player and event update and haven't any crash at all from few days (700ppl)

Hi one question, what kind of mutex because are you using? Can you show me? Because a try this http://paste2.org/p/688715 and seems to doesn't work

[przemratajczak]

exactly this one ;-) for me it is working, at least i can't complain

[Wojta]

Ive got this crash http://pastebin.com/rYmT4Esb - maybe due to unloadgrid=1?

[caeruleaus]

for 9677+

http://paste2.org/p/759109

it's also been pushed to my github repo

git clone git://github.com/caeruleaus/mangos.git
git checkout mtmaps

or if you already have a clone of the mangos master just make a branch and pull my mtmaps branch into it

git branch mtmaps
git checkout mtmaps
git pull git://github.com/caeruleaus/mangos.git mtmaps

[ilija]

This patch provides any unstability or crashes?

[caeruleaus]

it would be nice if anyone could confirm that my merge worked correctly.... no instability or anything?

I didn't include Wowuer's spell mutex locks but it should patch fine on my repo/patch.

[crashuncle]

Looks good Caeruleaus, but is it also working with the 0.12 branch?

[silvermcd]

lo, i'm stupid how could I forget: (

one big thanks KERO

how to (for debian):

and

and

http://pastebin.com/f66c0e16a

i installed tbb lib with synaptic package manager

would that be the same?

[caeruleaus]

attempt at mtmaps for 0.12

tell me if theres any problems

http://paste2.org/p/765918

[Wowka321]

Just tried patch. Core sturtup fine, but realmd-demon can't start

./mangos-realmd: error while loading shared libraries: libtbb.so.2: cannot open shared object file: No such file or directory

, but in /lib file exis =\\

How related tbb to realmd?

P.S. Ubuntu 8.10 amd64

[silvermcd]

need a little help here, i wish to know how to instal ltbb on debian @ ubuntu

[silvermcd]

admin@ks308761:/root$ sudo mv tbb22_20090809oss_src.tgz tbb

admin@ks308761:/root$ cd tbb/src

admin@ks308761:/root/tbb/src$ make

cd "../build/linux_ia32_gcc_cc4.2.4_libc2.7_kernel2.6.31.5_release" && sh ../../build/generate_tbbvars.sh linux_ia32_gcc_cc4.2.4_libc2.7_kernel2.6.31.5_release

/bin/sh: line 0: cd: ../build/linux_ia32_gcc_cc4.2.4_libc2.7_kernel2.6.31.5_release: No such file or directory

[crashuncle]

attempt at mtmaps for 0.12

tell me if theres any problems

http://paste2.org/p/765918

Anyone tested it yet? A small reply would be helpful.

[AlexM]

I tested on the lastest 0.12 mangos (with scriptdev and relaxed anticheat patch), you need to modify some parts of the patch to have it works.

I posted my diff to caerulueaus because I'm not sure of the quality and the rightness of my code.

[caeruleaus]

yeah sorry I haven't gotten around to looking at the changes yet. Been busy. I'll try to get around to it within the next few days

[zurghh]

Hello all,

I have this error when i try to compil.

../../../src/game/Spell.cpp: In member function ‘void Spell::cast(bool)’:

../../../src/game/Spell.cpp:2892: error: conversion from ‘tbb::internal::vector_iterator<tbb::concurrent_vecto r<Spell::TargetInfo, tbb::cache_aligned_allocator<Spell::TargetInfo> >, Spell::TargetInfo>’ to non-scalar type ‘std::_List_iterator<Spell::TargetInfo>’ requested

../../../src/game/Spell.cpp:2892: error: no match for ‘operator!=’ in ‘ihit != tbb::concurrent_vector<T, A>::e nd() [with T = Spell::TargetInfo, A = tbb::cache_aligned_allocator<Spell::TargetInfo>]()’

/usr/include/c++/4.3/bits/stl_list.h:173: note: candidates are: bool std::_List_iterator<_Tp>::operator!=(cons t std::_List_iterator<_Tp>&) const [with _Tp = Spell::TargetInfo]

/usr/local/include/ace/Time_Value.inl:313: note: bool operator!=(const ACE_Time_Value&, const ACE_Time_Value&)

In advance thx.

[crashuncle]

I tested on the lastest 0.12 mangos (with scriptdev and relaxed anticheat patch), you need to modify some parts of the patch to have it works.

I posted my diff to caerulueaus because I'm not sure of the quality and the rightness of my code.

Could you post it anyway? It is currently better then no patch at a ll.

[j4r0d]

There's the patch updated for Multithreaded Maps, required Mangos 9976+

I have included some missing things, and also included compatibility for VC100 project

http://paste2.org/p/852526

Enjoy

EDIT : Patch updated again due to fix on gameobject problems, like base capture in Arathi Basin, opening chests and some other

[HSC_Dev3]

ty jarod this is exactly what i came in looking for!

[Thyros]

Please update this patch for last rev!

[j4r0d]

Please update this patch for last rev!

Patch updated for rev 10003+ remember that this patch need test first

http://paste2.org/p/856193

[przemratajczak]

you didn't include body of those 4 new files ;-)

[j4r0d]

you didn't include body of those 4 new files ;-)

omg sorry

diff --git a/src/game/MapUpdater.cpp b/src/game/MapUpdater.cpp
new file mode 100644
index 0000000..1a37d7e
--- /dev/null
+++ b/src/game/MapUpdater.cpp
@@ -0,0 +1,140 @@
+#include "MapUpdater.h"
+
+#include "DelayExecutor.h"
+#include "Map.h"
+#include "Database/DatabaseEnv.h"
+
+#include <ace/Guard_T.h>
+#include <ace/Method_Request.h>
+
+//the reason this things are here is that i want to make
+//the netcode patch and the multithreaded maps independant
+//once they are merged 1 class should be used
+class  WDBThreadStartReq1 : public ACE_Method_Request
+{
+public:
+  WDBThreadStartReq1 () { }
+  virtual int
+  call (void)
+  {
+    WorldDatabase.ThreadStart ();
+    return 0;
+  }
+};
+
+class  WDBThreadEndReq1 : public ACE_Method_Request
+{
+public:
+  WDBThreadEndReq1 () { }
+  virtual int
+  call (void)
+  {
+    WorldDatabase.ThreadEnd ();
+    return 0;
+  }
+};
+
+class MapUpdateRequest : public ACE_Method_Request
+{
+public:
+  Map& m_map;
+  MapUpdater& m_updater;
+  ACE_UINT32 m_diff;
+  MapUpdateRequest (Map& m,MapUpdater& u,ACE_UINT32 d) : m_map(m),m_updater(u),m_diff(d) { }
+  virtual int
+  call (void)
+  {
+    m_map.Update (m_diff);
+    m_updater.update_finished ();
+    return 0;
+  }
+};
+
+MapUpdater::MapUpdater () :
+m_mutex (),
+m_condition (m_mutex),
+m_executor (),
+pedning_requests (0)
+{
+  return;
+}
+
+MapUpdater::~MapUpdater ()
+{
+  this->deactivate ();
+}
+
+int
+MapUpdater::activate (size_t num_threads)
+{
+  return this->m_executor.activate (static_cast<int> (num_threads),
+                                    new WDBThreadStartReq1,
+                                    new WDBThreadEndReq1);
+}
+
+int
+MapUpdater::Deactivate (void)
+{
+  this->wait ();
+
+  return this->m_executor.deactivate ();
+}
+
+int
+MapUpdater::wait ()
+{
+  ACE_GUARD_RETURN(ACE_Thread_Mutex,guard,this->m_mutex,-1);
+
+  while(this->pedning_requests > 0)
+    this->m_condition.wait ();
+
+  return 0;
+}
+
+int
+MapUpdater::schedule_update(Map& map, ACE_UINT32 diff)
+{
+  ACE_GUARD_RETURN(ACE_Thread_Mutex,guard,this->m_mutex,-1);
+
+  ++this->pedning_requests;
+
+  if( this->m_executor.execute (new MapUpdateRequest(map,*this,diff)) == -1)
+    {
+      ACE_DEBUG ((LM_ERROR,
+                  ACE_TEXT ("(%t) \\n"),
+                  ACE_TEXT ("Failed to schedule Map Update")));
+
+      --this->pedning_requests;
+      return -1;
+    }
+
+  return 0;
+}
+
+bool
+MapUpdater::activated ()
+{
+  return m_executor.activated();
+}
+
+void
+MapUpdater::update_finished ()
+{
+  ACE_GUARD (ACE_Thread_Mutex, guard, this->m_mutex);
+
+  if (this->pedning_requests == 0)
+    {
+      ACE_ERROR ((LM_ERROR,
+                  ACE_TEXT ("(%t)\\n"),
+                  ACE_TEXT ("MapUpdater::update_finished BUG, report to devs")));
+
+      return;
+    }
+
+  --this->pedning_requests;
+
+  //TODO can more than one thread call wait (), it shouldnt happen
+  //however I ensure if in future more than 1 thread call it by
+  //using broadcast instead of signal ()
+  this->m_condition.broadcast ();
+}
diff --git a/src/game/MapUpdater.h b/src/game/MapUpdater.h
new file mode 100644
index 0000000..d4d4be5
--- /dev/null
+++ b/src/game/MapUpdater.h
@@ -0,0 +1,46 @@
+#ifndef _MAP_UPDATER_H_INCLUDED
+#define _MAP_UPDATER_H_INCLUDED
+
+
+#include <ace/Thread_Mutex.h>
+#include <ace/Condition_Thread_Mutex.h>
+
+#include "DelayExecutor.h"
+
+class Map;
+
+class MapUpdater
+{
+public:
+  MapUpdater ();
+  virtual ~MapUpdater ();
+
+  friend class MapUpdateRequest;
+
+  /// schedule update on a map, the update will start
+  /// as soon as possible ,
+  /// it may even start before the call returns
+  int schedule_update(Map& map, ACE_UINT32 diff);
+
+  /// Wait untill all pending updates finish
+  int wait ();
+
+  /// Start the worker threads
+  int activate (size_t num_threads);
+
+  /// Stop the worker threads
+  int deactivate (void);
+
+  bool activated ();
+
+private:
+  /// hook called by worker threads
+  void update_finished ();
+
+  DelayExecutor m_executor;
+  ACE_Condition_Thread_Mutex m_condition;
+  ACE_Thread_Mutex m_mutex;
+  size_t pedning_requests;
+};
+
+#endif //_MAP_UPDATER_H_INCLUDED
diff --git a/src/shared/DelayExecutor.cpp b/src/shared/DelayExecutor.cpp
new file mode 100644
index 0000000..d567ceb
--- /dev/null
+++ b/src/shared/DelayExecutor.cpp
@@ -0,0 +1,130 @@
+#include <ace/Singleton.h>
+#include <ace/Thread_Mutex.h>
+
+#include "DelayExecutor.h"
+
+#include <ace/Log_Msg.h>
+
+DelayExecutor*
+DelayExecutor::instance ()
+{
+  return ACE_Singleton<DelayExecutor, ACE_Thread_Mutex>::instance ();
+}
+
+DelayExecutor::DelayExecutor () :
+activated_ (false),
+pre_svc_hook_ (0),
+post_svc_hook_ (0) { }
+
+DelayExecutor::~DelayExecutor ()
+{
+  if (pre_svc_hook_)
+    delete pre_svc_hook_;
+
+  if (post_svc_hook_)
+    delete post_svc_hook_;
+
+  this->deactivate ();
+
+  //todo probably free the queue ??
+}
+
+int
+DelayExecutor::Deactivate ()
+{
+  if (!this->activated ())
+    return -1;
+
+  this->activated (false);
+
+  this->queue_.queue ()->deactivate ();
+
+  this->wait ();
+
+  return 0;
+}
+
+int
+DelayExecutor::svc (void)
+{
+  if (pre_svc_hook_)
+    pre_svc_hook_->call ();
+
+  for (;
+    {
+      ACE_Method_Request* rq = this->queue_.dequeue ();
+
+      if (!rq)
+        break;
+
+      rq->call ();
+
+      delete rq;
+    }
+
+  if (post_svc_hook_)
+    post_svc_hook_->call ();
+
+  return 0;
+}
+
+int
+DelayExecutor::activate (int num_threads,
+                         ACE_Method_Request* pre_svc_hook,
+                         ACE_Method_Request* post_svc_hook)
+{
+  if (this->activated ())
+    return -1;
+
+  if (num_threads < 1)
+    return -1;
+
+  if (pre_svc_hook_)
+    delete pre_svc_hook_;
+
+  if (post_svc_hook_)
+    delete post_svc_hook_;
+
+  this->pre_svc_hook_ = pre_svc_hook;
+  this->post_svc_hook_ = post_svc_hook;
+
+  this->queue_.queue ()->activate ();
+
+  if( ACE_Task_Base::activate (THR_NEW_LWP | THR_JOINABLE | THR_INHERIT_SCHED,
+                                  num_threads) == -1)
+    return -1;
+
+  this->activated(true);
+
+  return true;
+}
+
+int
+DelayExecutor::execute (ACE_Method_Request* new_req)
+{
+  if(new_req == NULL)
+    return -1;
+
+  if(this->queue_.enqueue (new_req,
+                           (ACE_Time_Value*)&ACE_Time_Value::zero) == -1)
+    {
+      delete new_req;
+      ACE_ERROR_RETURN ((LM_ERROR,
+                         ACE_TEXT ("(%t) %p\\n"),
+                         ACE_TEXT ("DelayExecutor::execute enqueue")),
+                        -1);
+    }
+  return 0;
+}
+
+bool
+DelayExecutor::activated ()
+{
+  return this->activated_;
+}
+
+void
+DelayExecutor::activated (bool s)
+{
+  this->activated_ = s;
+}
diff --git a/src/shared/DelayExecutor.h b/src/shared/DelayExecutor.h
new file mode 100644
index 0000000..18c6a1b
--- /dev/null
+++ b/src/shared/DelayExecutor.h
@@ -0,0 +1,39 @@
+#ifndef _M_DELAY_EXECUTOR_H
+#define    _M_DELAY_EXECUTOR_H
+
+#include <ace/Task.h>
+#include <ace/Activation_Queue.h>
+#include <ace/Method_Request.h>
+
+class DelayExecutor : protected ACE_Task_Base
+{
+public:
+  DelayExecutor ();
+  virtual ~DelayExecutor ();
+
+  static DelayExecutor* instance ();
+
+  /// example
+  /// DelayExecutor::instance ()->execute(new MyRequest(myarg));
+  /// returns -1 on failures
+  int execute (ACE_Method_Request* new_req);
+
+  int activate (int num_threads = 1,
+                ACE_Method_Request* pre_svc_hook = 0,
+                ACE_Method_Request* post_svc_hook = 0);
+
+  int deactivate ();
+
+  bool activated ();
+
+  virtual int svc (void);
+private:
+  ACE_Activation_Queue queue_;
+  ACE_Method_Request* pre_svc_hook_;
+  ACE_Method_Request* post_svc_hook_;
+
+  void activated (bool s);
+  bool activated_;
+};
+
+#endif // _M_DELAY_EXECUTOR_H

[crashuncle]

Someone tested it with 0.12 yet? Else I have a look at it.