Replacement of the pthread call by SwiftShader Thread and Resource objects

src/OpenCL/OpenCL1/OpenCL1.vcxproj

@@ -95,6 +95,7 @@
       <RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
       <PrecompiledHeaderFile />
       <BrowseInformation>true</BrowseInformation>
+      <AdditionalIncludeDirectories>../../Common</AdditionalIncludeDirectories>
     </ClCompile>
     <Link>
       <SubSystem>Windows</SubSystem>
@@ -118,7 +119,7 @@
       <PrecompiledHeaderFile>
       </PrecompiledHeaderFile>
       <BrowseInformation>true</BrowseInformation>
-      <AdditionalIncludeDirectories>C:\Users\mgregoire\Documents\Visual Studio 2013\Projects\SwiftCL\pthread\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>../../Common;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
     </ClCompile>
     <Link>
       <SubSystem>Windows</SubSystem>
@@ -137,6 +138,7 @@
       <IntrinsicFunctions>true</IntrinsicFunctions>
       <PreprocessorDefinitions>WIN32;NDEBUG;_WINDOWS;_USRDLL;OPENCL1_EXPORTS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <SDLCheck>true</SDLCheck>
+      <AdditionalIncludeDirectories>../../Common</AdditionalIncludeDirectories>
     </ClCompile>
     <Link>
       <SubSystem>Windows</SubSystem>
@@ -161,7 +163,7 @@
       <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
       <OmitFramePointers>true</OmitFramePointers>
       <EnableFiberSafeOptimizations>true</EnableFiberSafeOptimizations>
-      <AdditionalIncludeDirectories>C:\Users\mgregoire\Documents\Visual Studio 2013\Projects\SwiftCL\pthread\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>../../Common;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
     </ClCompile>
     <Link>
       <SubSystem>Windows</SubSystem>
@@ -175,6 +177,10 @@
     </Link>
   </ItemDefinitionGroup>
   <ItemGroup>
+    <ClInclude Include="..\..\Common\CPUID.hpp" />
+    <ClInclude Include="..\..\Common\Memory.hpp" />
+    <ClInclude Include="..\..\Common\Resource.hpp" />
+    <ClInclude Include="..\..\Common\Thread.hpp" />
     <ClInclude Include="buffer.h" />
     <ClInclude Include="builtins.h" />
     <ClInclude Include="cl.h" />
@@ -199,6 +205,10 @@
     <ClInclude Include="worker.h" />
   </ItemGroup>
   <ItemGroup>
+    <ClCompile Include="..\..\Common\CPUID.cpp" />
+    <ClCompile Include="..\..\Common\Memory.cpp" />
+    <ClCompile Include="..\..\Common\Resource.cpp" />
+    <ClCompile Include="..\..\Common\Thread.cpp" />
     <ClCompile Include="buffer.cpp" />
     <ClCompile Include="builtins.cpp" />
     <ClCompile Include="commandqueue.cpp" />

src/OpenCL/OpenCL1/commandqueue.cpp

@@ -25,8 +25,8 @@ CommandQueue::CommandQueue(Context *ctx,
 	p_properties(properties), p_flushed(true)
 {
 	// Initialize the locking machinery
-	pthread_mutex_init(&p_event_list_mutex, 0);
-	pthread_cond_init(&p_event_list_cond, 0);
+	p_event_list_mutex = new sw::Resource(0);
+	p_event_list_cond = new sw::Event();
 
 	// Check that the device belongs to the context
 	if(!ctx->hasDevice(device))
@@ -41,8 +41,12 @@ CommandQueue::CommandQueue(Context *ctx,
 CommandQueue::~CommandQueue()
 {
 	// Free the mutex
-	pthread_mutex_destroy(&p_event_list_mutex);
-	pthread_cond_destroy(&p_event_list_cond);
+	p_event_list_mutex->lock(sw::DESTRUCT);
+	p_event_list_mutex->unlock();
+	p_event_list_mutex->destruct();
+	
+	p_event_list_cond->signal();
+	delete p_event_list_cond;
 }
 
 cl_int CommandQueue::info(cl_command_queue_info param_name,
@@ -139,12 +143,12 @@ cl_int CommandQueue::checkProperties() const
 void CommandQueue::flush()
 {
 	// Wait for the command queue to be in state "flushed".
-	pthread_mutex_lock(&p_event_list_mutex);
+	p_event_list_mutex->lock(sw::PRIVATE);
 
 	while(!p_flushed)
-		pthread_cond_wait(&p_event_list_cond, &p_event_list_mutex);
+		p_event_list_cond->wait();
 
-	pthread_mutex_unlock(&p_event_list_mutex);
+	p_event_list_mutex->unlock();
 }
 
 void CommandQueue::finish()
@@ -155,12 +159,12 @@ void CommandQueue::finish()
 
 	// All the queued events must have completed. When they are, they get
 	// deleted from the command queue, so simply wait for it to become empty.
-	pthread_mutex_lock(&p_event_list_mutex);
+	p_event_list_mutex->lock(sw::PRIVATE);
 
 	while(p_events.size() != 0)
-		pthread_cond_wait(&p_event_list_cond, &p_event_list_mutex);
+		p_event_list_cond->wait();
 
-	pthread_mutex_unlock(&p_event_list_mutex);
+	p_event_list_mutex->unlock();
 }
 
 cl_int CommandQueue::queueEvent(Event *event)
@@ -173,12 +177,12 @@ cl_int CommandQueue::queueEvent(Event *event)
 		return rs;
 
 	// Append the event at the end of the list
-	pthread_mutex_lock(&p_event_list_mutex);
+	p_event_list_mutex->lock(sw::PRIVATE);
 
 	p_events.push_back(event);
 	p_flushed = false;
 
-	pthread_mutex_unlock(&p_event_list_mutex);
+	p_event_list_mutex->unlock();
 
 	// Timing info if needed
 	if(p_properties & CL_QUEUE_PROFILING_ENABLE)
@@ -192,7 +196,7 @@ cl_int CommandQueue::queueEvent(Event *event)
 
 void CommandQueue::cleanEvents()
 {
-	pthread_mutex_lock(&p_event_list_mutex);
+	p_event_list_mutex->lock(sw::PRIVATE);
 
 	std::list<Event *>::iterator it = p_events.begin(), oldit;
 
@@ -218,9 +222,9 @@ void CommandQueue::cleanEvents()
 
 	// We have cleared the list, so wake up the sleeping threads
 	if(p_events.size() == 0)
-		pthread_cond_broadcast(&p_event_list_cond);
+		p_event_list_cond->signal();
 
-	pthread_mutex_unlock(&p_event_list_mutex);
+	p_event_list_mutex->unlock();
 
 	// Check now if we have to be deleted
 	if(references() == 0)
@@ -229,7 +233,7 @@ void CommandQueue::cleanEvents()
 
 void CommandQueue::pushEventsOnDevice()
 {
-	pthread_mutex_lock(&p_event_list_mutex);
+	p_event_list_mutex->lock(sw::PRIVATE);
 	// Explore the events in p_events and push on the device all of them that
 	// are :
 	//
@@ -328,23 +332,23 @@ void CommandQueue::pushEventsOnDevice()
 			// Set the event as completed. This will call pushEventsOnDevice,
 			// again, so release the lock to avoid a deadlock. We also return
 			// because the recursive call will continue our work.
-			pthread_mutex_unlock(&p_event_list_mutex);
+			p_event_list_mutex->unlock();
 			event->setStatus(Event::Complete);
 			return;
 		}
 	}
 
 	if(p_flushed)
-		pthread_cond_broadcast(&p_event_list_cond);
+		p_event_list_cond->signal();
 
-	pthread_mutex_unlock(&p_event_list_mutex);
+	p_event_list_mutex->unlock();
 }
 
 Event **CommandQueue::events(unsigned int &count)
 {
 	Event **result;
 
-	pthread_mutex_lock(&p_event_list_mutex);
+	p_event_list_mutex->lock(sw::PRIVATE);
 
 	count = p_events.size();
 	result = (Event **)std::malloc(count * sizeof(Event *));
@@ -365,7 +369,7 @@ Event **CommandQueue::events(unsigned int &count)
 	// Now result contains an immutable list of events. Even if the events
 	// become completed in another thread while result is used, the events
 	// are retained and so guaranteed to remain valid.
-	pthread_mutex_unlock(&p_event_list_mutex);
+	p_event_list_mutex->unlock();
 
 	return result;
 }
@@ -384,8 +388,8 @@ Event::Event(CommandQueue *parent,
 	p_status(status), p_device_data(0)
 {
 	// Initialize the locking machinery
-	pthread_cond_init(&p_state_change_cond, 0);
-	pthread_mutex_init(&p_state_mutex, 0);
+	p_state_change_cond = new sw::Event();
+	p_state_mutex = new sw::Resource(0);
 
 	std::memset(&p_timing, 0, sizeof(p_timing));
 
@@ -461,8 +465,13 @@ Event::~Event()
 	if(p_event_wait_list)
 		std::free((void *)p_event_wait_list);
 
-	pthread_mutex_destroy(&p_state_mutex);
-	pthread_cond_destroy(&p_state_change_cond);
+	// Free the mutex
+	p_state_mutex->lock(sw::DESTRUCT);
+	p_state_mutex->unlock();
+	p_state_mutex->destruct();
+
+	p_state_change_cond->signal();
+	delete p_state_change_cond;
 }
 
 bool Event::isDummy() const
@@ -486,10 +495,10 @@ bool Event::isDummy() const
 void Event::setStatus(Status status)
 {
 	// TODO: If status < 0, terminate all the events depending on us.
-	pthread_mutex_lock(&p_state_mutex);
+	p_state_mutex->lock(sw::PRIVATE);
 	p_status = status;
 
-	pthread_cond_broadcast(&p_state_change_cond);
+	p_state_change_cond->signal();
 
 	// Call the callbacks
 	std::multimap<Status, CallbackData>::const_iterator it;
@@ -504,7 +513,7 @@ void Event::setStatus(Status status)
 		data.callback((cl_event)this, p_status, data.user_data);
 	}
 
-	pthread_mutex_unlock(&p_state_mutex);
+	p_state_mutex->unlock();
 
 	// If the event is completed, inform our parent so it can push other events
 	// to the device.
@@ -586,12 +595,12 @@ void Event::updateTiming(Timing timing)
 	if(timing >= Max)
 		return;
 
-	pthread_mutex_lock(&p_state_mutex);
+	p_state_mutex->lock(sw::PRIVATE);
 
 	// Don't update more than one time (NDRangeKernel for example)
 	if(p_timing[timing])
 	{
-		pthread_mutex_unlock(&p_state_mutex);
+		p_state_mutex->unlock();
 		return;
 	}
 
@@ -610,7 +619,7 @@ void Event::updateTiming(Timing timing)
 
 	p_timing[timing] = rs;
 
-	pthread_mutex_unlock(&p_state_mutex);
+	p_state_mutex->unlock();
 }
 
 Event::Status Event::status() const
@@ -618,25 +627,25 @@ Event::Status Event::status() const
 	// HACK : We need const qualifier but we also need to lock a mutex
 	Event *me = (Event *)(void *)this;
 
-	pthread_mutex_lock(&me->p_state_mutex);
+	me->p_state_mutex->lock(sw::PRIVATE);
 
 	Status ret = p_status;
 
-	pthread_mutex_unlock(&me->p_state_mutex);
+	me->p_state_mutex->unlock();
 
 	return ret;
 }
 
 void Event::waitForStatus(Status status)
 {
-	pthread_mutex_lock(&p_state_mutex);
+	p_state_mutex->lock(sw::PRIVATE);
 
 	while(p_status != status && p_status > 0)
 	{
-		pthread_cond_wait(&p_state_change_cond, &p_state_mutex);
+		p_state_change_cond->signal();
 	}
 
-	pthread_mutex_unlock(&p_state_mutex);
+	p_state_mutex->unlock();
 }
 
 void *Event::deviceData()
@@ -659,13 +668,13 @@ void Event::setCallback(cl_int command_exec_callback_type,
 	data.callback = callback;
 	data.user_data = user_data;
 
-	pthread_mutex_lock(&p_state_mutex);
+	p_state_mutex->lock(sw::PRIVATE);
 
 	p_callbacks.insert(std::pair<Status, CallbackData>(
 		(Status)command_exec_callback_type,
 		data));
 
-	pthread_mutex_unlock(&p_state_mutex);
+	p_state_mutex->unlock();
 }
 
 cl_int Event::info(cl_event_info param_name,

src/OpenCL/OpenCL1/commandqueue.h

@@ -5,7 +5,7 @@
 
 #include "object.h"
 #include "opencl.h"
-#include "pthread.h"
+#include "Resource.hpp"
 
 #include <map>
 #include <list>
@@ -148,8 +148,8 @@ namespace Devices
 		cl_command_queue_properties p_properties;
 
 		std::list<Event *> p_events;
-		pthread_mutex_t p_event_list_mutex;
-		pthread_cond_t p_event_list_cond;
+		sw::Resource *p_event_list_mutex;
+		sw::Event *p_event_list_cond;
 		bool p_flushed;
 	};
 
@@ -359,8 +359,8 @@ namespace Devices
 		cl_uint p_num_events_in_wait_list;
 		const Event **p_event_wait_list;
 
-		pthread_cond_t p_state_change_cond;
-		pthread_mutex_t p_state_mutex;
+		sw::Event *p_state_change_cond;
+		sw::Resource *p_state_mutex;
 
 		Status p_status;
 		void *p_device_data;

src/OpenCL/OpenCL1/device.cpp

@@ -6,13 +6,13 @@
 #include "program.h"
 #include "worker.h"
 #include "builtins.h"
-
 #include "propertylist.h"
 #include "commandqueue.h"
 #include "events.h"
 #include "memobject.h"
-#include "kernel.h"
-#include "program.h"
+
+#include "Thread.hpp"
+#include "CPUID.hpp"
 
 #include <cstring>
 #include <cstdlib>
@@ -24,84 +24,48 @@
 using namespace Devices;
 
 CPUDevice::CPUDevice()
-	: DeviceInterface(), p_cores(0), p_num_events(0), p_workers(0), p_stop(false),
-	p_initialized(false)
+	: DeviceInterface(), p_num_events(0), p_stop(false)
 {
 
 }
 
 void CPUDevice::init()
 {
-	if(p_initialized)
-		return;
-
 	// Initialize the locking machinery
-	pthread_cond_init(&p_events_cond, 0);
-	pthread_mutex_init(&p_events_mutex, 0);
-
-	//TODO
-	// Get info about the system
-	/*p_cores = sysconf(_SC_NPROCESSORS_ONLN);
-	p_cpu_mhz = 0.0f;
-
-	std::filebuf fb;
-	fb.open("/proc/cpuinfo", std::ios::in);
-	std::istream is(&fb);
-
-	while(!is.eof())
-	{
-		std::string key, value;
-
-		std::getline(is, key, ':');
-		is.ignore(1);
-		std::getline(is, value);
-
-		if(key.compare(0, 7, "cpu MHz") == 0)
-		{
-			std::istringstream ss(value);
-			ss >> p_cpu_mhz;
-			break;
-		}
-	}*/
+	p_events_cond = new sw::Event();
+	eventListResource = new sw::Resource(0);
 
 	//TODO CHANGE pcore value to real
-	p_cores = 1;
 	p_cpu_mhz = 3200;
-	// Create worker threads
-	p_workers = (pthread_t *)std::malloc(numCPUs() * sizeof(pthread_t));
 
-	for(unsigned int i = 0; i<numCPUs(); ++i)
+	// Create worker threads
+	for(unsigned int i = 0; i < sw::CPUID::coreCount(); ++i)
 	{
-		pthread_create(&p_workers[i], 0, &worker, this);
+		p_workers[i] = new sw::Thread(worker, this);
 	}
-
-	p_initialized = true;
 }
 
 CPUDevice::~CPUDevice()
 {
-	if(!p_initialized)
-		return;
-
-	// Terminate the workers and wait for them
-	pthread_mutex_lock(&p_events_mutex);
-
 	p_stop = true;
 
-	pthread_cond_broadcast(&p_events_cond);
-	pthread_mutex_unlock(&p_events_mutex);
-
-	//TODO
-	for(unsigned int i = 0; i<numCPUs(); ++i)
+	for(int thread = 0; thread < sw::CPUID::coreCount(); thread++)
 	{
-		pthread_join(p_workers[i], 0);
+		if(p_workers[thread])
+		{
+			p_workers[thread]->join();
+
+			delete p_workers[thread];
+			p_workers[thread] = 0;
+		}
 	}
 
-	// Free allocated memory
-	std::free((void *)p_workers);
-	pthread_mutex_destroy(&p_events_mutex);
-	p_events_cond = NULL;
-	//pthread_cond_destroy(&p_events_cond);
+	p_events_cond->signal();
+	delete p_events_cond;
+
+	eventListResource->lock(sw::DESTRUCT);
+	eventListResource->unlock();
+	eventListResource->destruct();
 }
 
 DeviceBuffer *CPUDevice::createDeviceBuffer(MemObject *buffer, cl_int *rs)
@@ -203,28 +167,27 @@ void CPUDevice::freeEventDeviceData(Event *event)
 
 void CPUDevice::pushEvent(Event *event)
 {
-	// Add an event in the list
-	pthread_mutex_lock(&p_events_mutex);
+	eventListResource->lock(sw::PRIVATE);
 
 	p_events.push_back(event);
-	p_num_events++;                 // Way faster than STL list::size() !
+	p_num_events++;
+	p_events_cond->signal();
 
-	pthread_cond_broadcast(&p_events_cond);
-	pthread_mutex_unlock(&p_events_mutex);
+	eventListResource->unlock();
 }
 
 Event *CPUDevice::getEvent(bool &stop)
 {
+	eventListResource->lock(sw::PRIVATE);
+
 	// Return the first event in the list, if any. Remove it if it is a
 	// single-shot event.
-	pthread_mutex_lock(&p_events_mutex);
-
 	while(p_num_events == 0 && !p_stop)
-		pthread_cond_wait(&p_events_cond, &p_events_mutex);
+		p_events_cond->wait();
 
 	if(p_stop)
 	{
-		pthread_mutex_unlock(&p_events_mutex);
+		eventListResource->unlock();
 		stop = true;
 		return 0;
 	}
@@ -247,16 +210,11 @@ Event *CPUDevice::getEvent(bool &stop)
 		p_events.pop_front();
 	}
 
-	pthread_mutex_unlock(&p_events_mutex);
+	eventListResource->unlock();
 
 	return event;
 }
 
-unsigned int CPUDevice::numCPUs() const
-{
-	return p_cores;
-}
-
 float CPUDevice::cpuMhz() const
 {
 	return p_cpu_mhz;
@@ -309,7 +267,7 @@ cl_int CPUDevice::info(cl_device_info param_name,
 		break;
 
 	case CL_DEVICE_MAX_COMPUTE_UNITS:
-		SIMPLE_ASSIGN(cl_uint, numCPUs());
+		SIMPLE_ASSIGN(cl_uint, sw::CPUID::coreCount());
 		break;
 
 	case CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS:

src/OpenCL/OpenCL1/device.h

@@ -4,11 +4,13 @@
 #ifndef __CPU_DEVICE_H__
 #define __CPU_DEVICE_H__
 
+#define MAX_THREAD_AMOUNT 16
+
 #include <list>
 
 #include "opencl.h"
 #include "device_interface.h"
-#include "pthread.h"
+#include "Resource.hpp"
 
 namespace Devices
 {
@@ -43,18 +45,16 @@ public:
 	void pushEvent(Event *event);
 	Event *getEvent(bool &stop);
 
-	unsigned int numCPUs() const;   /*!< \brief Number of logical CPU cores on the system */
 	float cpuMhz() const;           /*!< \brief Speed of the CPU in Mhz */
 
 private:
-	unsigned int p_cores, p_num_events;
+	unsigned int p_num_events;
 	float p_cpu_mhz;
-	pthread_t *p_workers;
-
+	sw::Thread *p_workers[MAX_THREAD_AMOUNT];
+	sw::Resource *eventListResource;
+	sw::Event *p_events_cond;
 	std::list<Event *> p_events;
-	pthread_cond_t p_events_cond;
-	pthread_mutex_t p_events_mutex;
-	bool p_stop, p_initialized;
+	bool p_stop;
 };
 
 //class GPUDevice : public DeviceInterface

src/OpenCL/OpenCL1/dllmain.cpp

 // dllmain.cpp : Defines the entry point for the DLL application.
+#include "dllmain.h"
+
 #include "windows.h"
 #include "opencl.h"
+#include "Thread.hpp"
 #include "debug.h"
+
 #include <windows.h>
 #include <intrin.h>
 #include <WinUser.h>
-#include "dllmain.h"
-
-#if defined(_WIN32)
-typedef DWORD LocalStorageKey;
-#else
-typedef pthread_key_t LocalStorageKey;
-#endif
 
-static LocalStorageKey currentTLS = TLS_OUT_OF_INDEXES;
+static sw::Thread::LocalStorageKey currentTLS = TLS_OUT_OF_INDEXES;
 
 #if defined(_WIN32)
 #define IDD_DIALOG1                     101

src/OpenCL/OpenCL1/kernel.cpp

@@ -28,7 +28,7 @@ CPUKernel::CPUKernel(CPUDevice *device, Kernel *kernel, llvm::Function *function
 	: DeviceKernel(), p_device(device), p_kernel(kernel), p_function(function),
 	p_call_function(0)
 {
-	pthread_mutex_init(&p_call_function_mutex, 0);
+	p_call_function_mutex = new sw::Resource(0);
 }
 
 CPUKernel::~CPUKernel()
@@ -37,7 +37,9 @@ CPUKernel::~CPUKernel()
 	//if(p_call_function)
 		//p_call_function->eraseFromParent();
 
-	pthread_mutex_destroy(&p_call_function_mutex);
+	p_call_function_mutex->lock(sw::DESTRUCT);
+	p_call_function_mutex->unlock();
+	p_call_function_mutex->destruct();
 }
 
 size_t CPUKernel::workGroupSize() const
@@ -76,7 +78,7 @@ T k_exp(T base, unsigned int e)
 size_t CPUKernel::guessWorkGroupSize(cl_uint num_dims, cl_uint dim,
 	size_t global_work_size) const
 {
-	unsigned int cpus = p_device->numCPUs();
+	unsigned int cpus = sw::CPUID::coreCount();
 
 	// Don't break in too small parts
 	if(k_exp(global_work_size, num_dims) > 64)
@@ -148,13 +150,13 @@ size_t CPUKernel::typeOffset(size_t &offset, size_t type_len)
 
 llvm::Function *CPUKernel::callFunction()
 {
-	pthread_mutex_lock(&p_call_function_mutex);
+	p_call_function_mutex->lock(sw::PRIVATE);
 
 	// If we can reuse the same function between work groups, do it
 	if(p_call_function)
 	{
 		llvm::Function *rs = p_call_function;
-		pthread_mutex_unlock(&p_call_function_mutex);
+		p_call_function_mutex->unlock();
 
 		return rs;
 	}
@@ -257,11 +259,11 @@ llvm::Function *CPUKernel::callFunction()
 	//// Retain the function if it can be reused
 	//p_call_function = stub_function;
 
-	pthread_mutex_unlock(&p_call_function_mutex);
+	p_call_function_mutex->unlock();
 
 
 	llvm::Function *rs = p_call_function;
-	pthread_mutex_unlock(&p_call_function_mutex);
+	p_call_function_mutex->unlock();
 	return rs;
 
 	//return stub_function;
@@ -275,7 +277,7 @@ CPUKernelEvent::CPUKernelEvent(CPUDevice *device, KernelEvent *event)
 	p_kernel_args(0)
 {
 	// Mutex
-	pthread_mutex_init(&p_mutex, 0);
+	p_mutex = new sw::Resource(0);
 
 	// Set current work group to (0, 0, ..., 0)
 	std::memset(p_current_work_group, 0, event->work_dim() * sizeof(size_t));
@@ -294,7 +296,9 @@ CPUKernelEvent::CPUKernelEvent(CPUDevice *device, KernelEvent *event)
 
 CPUKernelEvent::~CPUKernelEvent()
 {
-	pthread_mutex_destroy(&p_mutex);
+	p_mutex->lock(sw::DESTRUCT);
+	p_mutex->unlock();
+	p_mutex->destruct();
 
 	if(p_kernel_args)
 		std::free(p_kernel_args);
@@ -303,7 +307,7 @@ CPUKernelEvent::~CPUKernelEvent()
 bool CPUKernelEvent::reserve()
 {
 	// Lock, this will be unlocked in takeInstance()
-	pthread_mutex_lock(&p_mutex);
+	p_mutex->lock(sw::PRIVATE);
 
 	// Last work group if current == max - 1
 	return (p_current_wg == p_num_wg - 1);
@@ -312,23 +316,23 @@ bool CPUKernelEvent::reserve()
 bool CPUKernelEvent::finished()
 {
 	bool rs;
-
-	pthread_mutex_lock(&p_mutex);
+	
+	p_mutex->lock(sw::PRIVATE);
 
 	rs = (p_finished_wg == p_num_wg);
 
-	pthread_mutex_unlock(&p_mutex);
+	p_mutex->unlock();
 
 	return rs;
 }
 
 void CPUKernelEvent::workGroupFinished()
 {
-	pthread_mutex_lock(&p_mutex);
+	p_mutex->lock(sw::PRIVATE);
 
 	p_finished_wg++;
 
-	pthread_mutex_unlock(&p_mutex);
+	p_mutex->unlock();
 }
 
 CPUKernelWorkGroup *CPUKernelEvent::takeInstance()
@@ -343,7 +347,7 @@ CPUKernelWorkGroup *CPUKernelEvent::takeInstance()
 	p_current_wg += 1;
 
 	// Release event
-	pthread_mutex_unlock(&p_mutex);
+	p_mutex->unlock();
 
 	return wg;
 }

src/OpenCL/OpenCL1/kernel.h

@@ -2,11 +2,12 @@
 #define __CPU_KERNEL_H__
 
 #include "device_interface.h"
+#include "CPUID.hpp"
+#include "Resource.hpp"
 
 //#include <llvm/ExecutionEngine/GenericValue.h>
 #include <vector>
 #include <string>
-#include "pthread.h"
 
 #include <stdint.h>
 
@@ -103,7 +104,7 @@ namespace Devices
 		CPUDevice *p_device;
 		Kernel *p_kernel;
 		llvm::Function *p_function, *p_call_function;
-		pthread_mutex_t p_call_function_mutex;
+		sw::Resource *p_call_function_mutex;
 	};
 
 	class CPUKernelEvent;
@@ -286,7 +287,7 @@ namespace Devices
 		size_t p_current_work_group[MAX_WORK_DIMS],
 			p_max_work_groups[MAX_WORK_DIMS];
 		size_t p_current_wg, p_finished_wg, p_num_wg;
-		pthread_mutex_t p_mutex;
+		sw::Resource *p_mutex;
 		void *p_kernel_args;
 	};
 

src/OpenCL/OpenCL1/worker.cpp

@@ -17,7 +17,7 @@
 
 using namespace Devices;
 
-void *worker(void *data)
+void worker(void *data)
 {
 	CPUDevice *device = (CPUDevice *)data;
 	bool stop = false;
@@ -242,6 +242,4 @@ void *worker(void *data)
 
 	/*if(mapped_data)
 		munmap(mapped_data, mapped_size);*/
-
-	return 0;
 }

src/OpenCL/OpenCL1/worker.h

@@ -10,6 +10,6 @@
 * system. As explained by \ref events , this function waits until there
 * are \c Coal::Event objects to process and handle them.
 */
-void *worker(void *data);
+void worker(void *data);
 
 #endif