Merge branch 'threaded' - but will be hidden

under a cheat code for now

src/engine/engine.cpp

@@ -23,6 +23,7 @@
 #include "engine.h"
 #include "instrument.h"
 #include "safeReader.h"
+#include "workPool.h"
 #include "../ta-log.h"
 #include "../fileutils.h"
 #ifdef HAVE_SDL2
@@ -3123,6 +3124,10 @@ bool DivEngine::switchMaster(bool full) {
     quitDispatch();
     initDispatch();
   }
+  if (renderPool!=NULL) {
+    delete renderPool;
+    renderPool=NULL;
+  }
   if (initAudioBackend()) {
     for (int i=0; i<song.systemLen; i++) {
       disCont[i].setRates(got.rate);
@@ -3314,6 +3319,7 @@ bool DivEngine::initAudioBackend() {
   midiOutMode=getConfInt("midiOutMode",DIV_MIDI_MODE_NOTE);
   if (metroVol<0.0f) metroVol=0.0f;
   if (metroVol>2.0f) metroVol=2.0f;
+  renderPoolThreads=getConfInt("renderPoolThreads",0);
 
   if (lowLatency) logI("using low latency mode.");
 

src/engine/engine.h

@@ -39,6 +39,8 @@
 #include <unordered_map>
 #include <deque>
 
+class DivWorkPool;
+
 #define addWarning(x) \
   if (warnings.empty()) { \
     warnings+=x; \
@@ -485,6 +487,9 @@ class DivEngine {
 
   size_t totalProcessed;
 
+  unsigned int renderPoolThreads;
+  DivWorkPool* renderPool;
+
   // MIDI stuff
   std::function<int(const TAMidiMessage&)> midiCallback=[](const TAMidiMessage&) -> int {return -2;};
 
@@ -1260,6 +1265,8 @@ class DivEngine {
       metroAmp(0.0f),
       metroVol(1.0f),
       totalProcessed(0),
+      renderPoolThreads(0),
+      renderPool(NULL),
       curOrders(NULL),
       curPat(NULL),
       tempIns(NULL),

src/engine/playback.cpp

@@ -22,6 +22,7 @@
 #define _USE_MATH_DEFINES
 #include "dispatch.h"
 #include "engine.h"
+#include "workPool.h"
 #include "../ta-log.h"
 #include <math.h>
 
@@ -1788,6 +1789,10 @@ void DivEngine::nextBuf(float** in, float** out, int inChans, int outChans, unsi
 
   std::chrono::steady_clock::time_point ts_processBegin=std::chrono::steady_clock::now();
 
+  if (renderPool==NULL) {
+    renderPool=new DivWorkPool(renderPoolThreads);
+  }
+
   // process MIDI events (TODO: everything)
   if (output) if (output->midiIn) while (!output->midiIn->queue.empty()) {
     TAMidiMessage& msg=output->midiIn->queue.front();
@@ -2061,20 +2066,28 @@ void DivEngine::nextBuf(float** in, float** out, int inChans, int outChans, unsi
         // 5. tick the clock and fill buffers as needed
         if (cycles<runLeftG) {
           for (int i=0; i<song.systemLen; i++) {
-            int total=(cycles*disCont[i].runtotal)/(size<<MASTER_CLOCK_PREC);
-            disCont[i].acquire(disCont[i].runPos,total);
-            disCont[i].runLeft-=total;
-            disCont[i].runPos+=total;
+            renderPool->push([this,size](void* d) {
+              DivDispatchContainer* dc=(DivDispatchContainer*)d;
+              int total=(cycles*dc->runtotal)/(size<<MASTER_CLOCK_PREC);
+              dc->acquire(dc->runPos,total);
+              dc->runLeft-=total;
+              dc->runPos+=total;
+            },&disCont[i]);
           }
+          renderPool->wait();
           runLeftG-=cycles;
           cycles=0;
         } else {
           cycles-=runLeftG;
           runLeftG=0;
           for (int i=0; i<song.systemLen; i++) {
-            disCont[i].acquire(disCont[i].runPos,disCont[i].runLeft);
-            disCont[i].runLeft=0;
+            renderPool->push([](void* d) {
+              DivDispatchContainer* dc=(DivDispatchContainer*)d;
+              dc->acquire(dc->runPos,dc->runLeft);
+              dc->runLeft=0;
+            },&disCont[i]);
           }
+          renderPool->wait();
         }
       }
     }

src/engine/workPool.cpp

@@ -21,14 +21,17 @@
 #include "../ta-log.h"
 #include <thread>
 
+#include <SDL.h>
+
 void* _workThread(void* inst) {
   ((DivWorkThread*)inst)->run();
   return NULL;
 }
 
 void DivWorkThread::run() {
-  std::unique_lock<std::mutex> unique(selfLock);
+  //std::unique_lock<std::mutex> unique(selfLock);
   DivPendingTask task;
+  bool setFuckingPromise=false;
 
   logV("running work thread");
 
@@ -37,11 +40,20 @@ void DivWorkThread::run() {
     if (tasks.empty()) {
       lock.unlock();
       isBusy=false;
-      parent->notify.notify_one();
+      if (setFuckingPromise) {
+        parent->notify.set_value();
+        setFuckingPromise=false;
+        std::this_thread::yield();
+      }
       if (terminate) {
         break;
       }
-      notify.wait(unique);
+      std::future<void> future=notify.get_future();
+      future.wait();
+      lock.lock();
+      notify=std::promise<void>();
+      promiseAlreadySet=false;
+      lock.unlock();
       continue;
     } else {
       task=tasks.front();
@@ -50,8 +62,13 @@ void DivWorkThread::run() {
 
       task.func(task.funcArg);
 
-      parent->busyCount--;
-      parent->notify.notify_one();
+      int busyCount=--parent->busyCount;
+      if (busyCount<0) {
+        logE("oh no PROBLEM...");
+      }
+      if (busyCount==0) {
+        setFuckingPromise=true;
+      }
     }
   }
 }
@@ -80,8 +97,8 @@ bool DivWorkThread::busy() {
 void DivWorkThread::finish() {
   lock.lock();
   terminate=true;
+  notify.set_value();
   lock.unlock();
-  notify.notify_one();
   thread->join();
 }
 
@@ -117,19 +134,38 @@ bool DivWorkPool::busy() {
 
 void DivWorkPool::wait() {
   if (!threaded) return;
-  std::unique_lock<std::mutex> unique(selfLock);
+
+  if (busyCount==0) {
+    return;
+  }
+
+  std::future<void> future=notify.get_future();
 
   // start running
   for (unsigned int i=0; i<count; i++) {
-    workThreads[i].notify.notify_one();
+    if (!workThreads[i].promiseAlreadySet) {
+      try {
+        workThreads[i].lock.lock();
+        workThreads[i].promiseAlreadySet=true;
+        workThreads[i].notify.set_value();
+        workThreads[i].lock.unlock();
+      } catch (std::exception& e) {
+        SDL_ShowSimpleMessageBox(SDL_MESSAGEBOX_ERROR,"EXCEPTION ON WAIT",e.what(),NULL);
+        abort();
       }
+    }
+  }
+  std::this_thread::yield();
 
   // wait
-  while (busyCount!=0) {
-    if (notify.wait_for(unique,std::chrono::milliseconds(100))==std::cv_status::timeout) {
-      logW("DivWorkPool: wait() timed out!");
-    }
-  }
+  //SDL_ShowSimpleMessageBox(SDL_MESSAGEBOX_ERROR,"Error","waiting on future.",NULL);
+  future.wait();
+  //SDL_ShowSimpleMessageBox(SDL_MESSAGEBOX_ERROR,"Error","waited - reset promise.",NULL);
+
+  notify=std::promise<void>();
+  //SDL_ShowSimpleMessageBox(SDL_MESSAGEBOX_ERROR,"Error","YES",NULL);
+
+  pos=0;
 }
 
 DivWorkPool::DivWorkPool(unsigned int threads):

src/engine/workPool.h

@@ -24,7 +24,7 @@
 #include <mutex>
 #include <atomic>
 #include <functional>
-#include <condition_variable>
+#include <future>
 
 #include "fixedQueue.h"
 
@@ -44,12 +44,12 @@ struct DivPendingTask {
 struct DivWorkThread {
   DivWorkPool* parent;
   std::mutex lock;
-  std::mutex selfLock;
   std::thread* thread;
-  std::condition_variable notify;
+  std::promise<void> notify;
   FixedQueue<DivPendingTask,32> tasks;
   std::atomic<bool> isBusy;
   bool terminate;
+  bool promiseAlreadySet;
 
   void run();
   bool assign(const std::function<void(void*)>& what, void* arg);
@@ -61,7 +61,8 @@ struct DivWorkThread {
   DivWorkThread():
     parent(NULL),
     isBusy(false),
-    terminate(false) {}
+    terminate(false),
+    promiseAlreadySet(false) {}
 };
 
 /**
@@ -70,12 +71,11 @@ struct DivWorkThread {
  */
 class DivWorkPool {
   bool threaded;
-  std::mutex selfLock;
   unsigned int count;
   unsigned int pos;
   DivWorkThread* workThreads;
   public:
-    std::condition_variable notify;
+    std::promise<void> notify;
     std::atomic<int> busyCount;
     
     /**

src/gui/gui.h

@@ -1576,6 +1576,7 @@ class FurnaceGUI {
     int classicChipOptions;
     int wasapiEx;
     int chanOscThreads;
+    int renderPoolThreads;
     unsigned int maxUndoSteps;
     String mainFontPath;
     String headFontPath;
@@ -1753,6 +1754,7 @@ class FurnaceGUI {
       classicChipOptions(0),
       wasapiEx(0),
       chanOscThreads(0),
+      renderPoolThreads(0),
       maxUndoSteps(100),
       mainFontPath(""),
       headFontPath(""),

src/gui/settings.cpp

@@ -892,6 +892,36 @@ void FurnaceGUI::drawSettings() {
           ImGui::EndTable();
         }
 
+        bool renderPoolThreadsB=(settings.renderPoolThreads>0);
+        if (ImGui::Checkbox("Multi-threaded (EXPERIMENTAL)",&renderPoolThreadsB)) {
+          if (renderPoolThreadsB) {
+            settings.renderPoolThreads=2;
+          } else {
+            settings.renderPoolThreads=0;
+          }
+        }
+        if (ImGui::IsItemHovered()) {
+          ImGui::SetTooltip("runs chip emulation on separate threads.\nmay increase performance on multi-core CPUs.\n\nwarnings:\n- experimental! currently broken/not working well.\n- only useful on multi-chip songs.");
+        }
+
+        if (renderPoolThreadsB) {
+          pushWarningColor(settings.renderPoolThreads>cpuCores,settings.renderPoolThreads>cpuCores);
+          if (ImGui::InputInt("Number of threads",&settings.renderPoolThreads)) {
+            if (settings.renderPoolThreads<2) settings.renderPoolThreads=2;
+            if (settings.renderPoolThreads>32) settings.renderPoolThreads=32;
+          }
+          if (settings.renderPoolThreads>=DIV_MAX_CHIPS) {
+            if (ImGui::IsItemHovered()) {
+              ImGui::SetTooltip("that's the limit!");
+            }
+          } else if (settings.renderPoolThreads>cpuCores) {
+            if (ImGui::IsItemHovered()) {
+              ImGui::SetTooltip("it is a VERY bad idea to set this number higher than your CPU core count (%d)!",cpuCores);
+            }
+          }
+          popWarningColor();
+        }
+
         bool lowLatencyB=settings.lowLatency;
         if (ImGui::Checkbox("Low-latency mode",&lowLatencyB)) {
           settings.lowLatency=lowLatencyB;
@@ -3286,6 +3316,7 @@ void FurnaceGUI::syncSettings() {
   settings.classicChipOptions=e->getConfInt("classicChipOptions",0);
   settings.wasapiEx=e->getConfInt("wasapiEx",0);
   settings.chanOscThreads=e->getConfInt("chanOscThreads",0);
+  settings.renderPoolThreads=e->getConfInt("renderPoolThreads",0);
 
   clampSetting(settings.mainFontSize,2,96);
   clampSetting(settings.headFontSize,2,96);
@@ -3436,6 +3467,7 @@ void FurnaceGUI::syncSettings() {
   clampSetting(settings.classicChipOptions,0,1);
   clampSetting(settings.wasapiEx,0,1);
   clampSetting(settings.chanOscThreads,0,256);
+  clampSetting(settings.renderPoolThreads,0,256);
 
   if (settings.exportLoops<0.0) settings.exportLoops=0.0;
   if (settings.exportFadeOut<0.0) settings.exportFadeOut=0.0;
@@ -3693,6 +3725,7 @@ void FurnaceGUI::commitSettings() {
   e->setConf("classicChipOptions",settings.classicChipOptions);
   e->setConf("wasapiEx",settings.wasapiEx);
   e->setConf("chanOscThreads",settings.chanOscThreads);
+  e->setConf("renderPoolThreads",settings.renderPoolThreads);
 
   // colors
   for (int i=0; i<GUI_COLOR_MAX; i++) {