Added libspe2 support contribution by IBM Germany 'Extreme Blue' project:

Thanks to Benjamin Hoeferlin, Minh Cuong Tran,Martina Huellmann,Frederick Roth.

Extras/BulletMultiThreaded/SpuLibspe2Support.cpp

@@ -12,104 +12,82 @@ subject to the following restrictions:
 2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
 3. This notice may not be removed or altered from any source distribution.
 */
+
 #ifdef USE_LIBSPE2
 
 #include "SpuLibspe2Support.h"
 
-#include "SpuCollisionTaskProcess.h"
-#include "SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h"
 
-#include <Windows.h>
 
-///SpuLibspe2Support helps to initialize/shutdown libspe2, start/stop SPU tasks and communication
+
+//SpuLibspe2Support helps to initialize/shutdown libspe2, start/stop SPU tasks and communication
 ///Setup and initialize SPU/CELL/Libspe2
-SpuLibspe2Support::SpuLibspe2Support(spe_program_handle_t *speprog,int numThreads)
+SpuLibspe2Support::SpuLibspe2Support(spe_program_handle_t *speprog, int numThreads)
 {
-	program = speprog
+	this->program = speprog;
-	startSPUs(numThreads);
+	this->numThreads =  ((numThreads <= spe_cpu_info_get(SPE_COUNT_PHYSICAL_SPES, -1)) ? numThreads : spe_cpu_info_get(SPE_COUNT_PHYSICAL_SPES, -1));
-	N = numThreads;
 }
 
 ///cleanup/shutdown Libspe2
 SpuLibspe2Support::~SpuLibspe2Support()
 {
-	stopSPUs();
+	
+	stopSPU();
 }
 
 
 
-#include <stdio.h>
-
-#ifdef WIN32
-DWORD WINAPI Thread_no_1( LPVOID lpParam ) 
-{
-
-	btSpuStatus* status = (btSpuStatus*)lpParam;
-
-	
-	while (1)
-	{
-		WaitForSingleObject(status->m_eventStartHandle,INFINITE);
-		btAssert(status->m_status);
-
-		SpuGatherAndProcessPairsTaskDesc* taskDesc = status->m_taskDesc;
-
-		if (taskDesc)
-		{
-			processCollisionTask(*taskDesc);
-			SetEvent(status->m_eventCompletetHandle);
-		} else
-		{
-			//exit Thread
-			break;
-		}
-		
-	}
-
-	return 0;
-
-}
-#endif
 ///send messages to SPUs
 void SpuLibspe2Support::sendRequest(uint32_t uiCommand, uint32_t uiArgument0, uint32_t uiArgument1)
 {
-	///	gMidphaseSPU.sendRequest(CMD_GATHER_AND_PROCESS_PAIRLIST, (uint32_t) &taskDesc);
+	spe_context_ptr_t context;
-	
-	///we should spawn an SPU task here, and in 'waitForResponse' it should wait for response of the (one of) the first tasks that finished
-	
-
 	
 	switch (uiCommand)
 	{
+	case CMD_SAMPLE_TASK_COMMAND:
+	{
+		//get taskdescription
+		SpuSampleTaskDesc* taskDesc = (SpuSampleTaskDesc*) uiArgument0;
+
+		btAssert(taskDesc->m_taskId<m_activeSpuStatus.size());
+
+		//get status of SPU on which task should run
+		btSpuStatus&	spuStatus = m_activeSpuStatus[taskDesc->m_taskId];
+
+		//set data for spuStatus
+		spuStatus.m_commandId = uiCommand;
+		spuStatus.m_status = Spu_Status_Occupied; //set SPU as "occupied"
+		spuStatus.m_taskDesc.p = taskDesc; 
+		
+		//get context
+		context = data[taskDesc->m_taskId].context;
+		
+		
+		taskDesc->m_mainMemoryPtr = reinterpret_cast<uint64_t> (spuStatus.m_lsMemory.p);
+		
+
+		break;
+	}
 	case CMD_GATHER_AND_PROCESS_PAIRLIST:
 		{
+			//get taskdescription
+			SpuGatherAndProcessPairsTaskDesc* taskDesc = (SpuGatherAndProcessPairsTaskDesc*) uiArgument0;
 
-			SpuGatherAndProcessPairsTaskDesc* taskDesc = (SpuGatherAndProcessPairsTaskDesc*) uiArgument0 ;
-
-//#define SINGLE_THREADED 1
-#ifdef SINGLE_THREADED
-
-			btSpuStatus&	spuStatus = m_activeSpuStatus[0];
-			taskDesc->m_lsMemory = (CollisionTask_LocalStoreMemory*)spuStatus.m_lsMemory;
-			processCollisionTask(*taskDesc);
-			HANDLE handle =0;
-#else
-
-			btAssert(taskDesc->taskId>=0);
 			btAssert(taskDesc->taskId<m_activeSpuStatus.size());
 
+			//get status of SPU on which task should run
 			btSpuStatus&	spuStatus = m_activeSpuStatus[taskDesc->taskId];
 
+			//set data for spuStatus
 			spuStatus.m_commandId = uiCommand;
-			spuStatus.m_status = 1;
+			spuStatus.m_status = Spu_Status_Occupied; //set SPU as "occupied"
-			spuStatus.m_taskDesc = taskDesc;
+			spuStatus.m_taskDesc.p = taskDesc; 
-			taskDesc->m_lsMemory = (CollisionTask_LocalStoreMemory*)spuStatus.m_lsMemory;
-			///fire event to start new task
-			SetEvent(spuStatus.m_eventStartHandle);
 			
-#endif //CollisionTask_LocalStoreMemory
+			//get context
+			context = data[taskDesc->taskId].context;
 			
 			
+			taskDesc->m_lsMemory = (CollisionTask_LocalStoreMemory*)spuStatus.m_lsMemory.p;
 			
 			break;
 		}
@@ -122,6 +100,10 @@ void SpuLibspe2Support::sendRequest(uint32_t uiCommand, uint32_t uiArgument0, ui
 	};
 
 	
+	//write taskdescription in mailbox
+	unsigned int event = Spu_Mailbox_Event_Task;
+	spe_in_mbox_write(context, &event, 1, SPE_MBOX_ANY_NONBLOCKING);
+
 }
 
 ///check for messages from SPUs
@@ -131,37 +113,33 @@ void SpuLibspe2Support::waitForResponse(unsigned int *puiArgument0, unsigned int
 	
 	///A possible response can be 'yes, SPU handled it', or 'no, please do a PPU fallback'
 	
-
 	btAssert(m_activeSpuStatus.size());
 
+	
 	int last = -1;
 	
 	//find an active spu/thread
+	while(last < 0)
+	{
 		for (int i=0;i<m_activeSpuStatus.size();i++)
 		{
-		if (m_activeSpuStatus[i].m_status)
+			if ( m_activeSpuStatus[i].m_status == Spu_Status_Free)
 			{
 				last = i;
 				break;
 			}
 		}
+		if(last < 0)
+			sched_yield();
+	}
+
 
 
-#ifndef SINGLE_THREADED
-	btAssert(spuStatus.m_threadHandle);
-	btAssert(spuStatus.m_eventCompletetHandle);
 	btSpuStatus& spuStatus = m_activeSpuStatus[last];
-	WaitForSingleObject(spuStatus.m_eventCompletetHandle, INFINITE);
-	spuStatus.m_status = 0;
 
 	///need to find an active spu
 	btAssert(last>=0);
 
-#else
-	last=0;
-	btSpuStatus& spuStatus = m_activeSpuStatus[last];
-#endif //SINGLE_THREADED
-
 	
 
 	*puiArgument0 = spuStatus.m_taskId;
@@ -171,46 +149,109 @@ void SpuLibspe2Support::waitForResponse(unsigned int *puiArgument0, unsigned int
 }
 
 
-///start the spus (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
+void SpuLibspe2Support::startSPU()
-void SpuLibspe2Support::startSPUs(int numThreads)
+{
+	this->internal_startSPU();
+}
+
+
+
+///start the spus group (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
+void SpuLibspe2Support::internal_startSPU()
 {
-#ifdef WIN32
 	m_activeSpuStatus.resize(numThreads);
-#endif
 	
-	for (int i=0;i<numThreads;i++)
+	
+	for (int i=0; i < numThreads; i++)
 	{
-		printf("starting thread %d\n",i);
 		
-		data[i].context = spe_context_create(0, NULL);
+		if(data[i].context == NULL) 
-		spe_program_load(data[i].context, program);
+		{
-		data[i].entry = SPE_DEFAULT_ENTRY;
-		data[i].flags = 0;
-		data[i].argp = NULL;
-		data[i].envp = NULL;
-		pthread_create(&data[i].pthread, NULL, &ppu_pthread_function, &data[i]);
-		printf("started thread %d with threadHandle %d\n",i,handle);
 					
+			 /* Create context */
+			if ((data[i].context = spe_context_create(0, NULL)) == NULL)
+			{
+			      perror ("Failed creating context");
+		          exit(1);
 			}
 	
+			/* Load program into context */
+			if(spe_program_load(data[i].context, this->program))
+			{
+			      perror ("Failed loading program");
+		          exit(1);
+			}
+			
+			m_activeSpuStatus[i].m_status = Spu_Status_Startup; 
+			m_activeSpuStatus[i].m_taskId = i; 
+			m_activeSpuStatus[i].m_commandId = 0; 
+			m_activeSpuStatus[i].m_lsMemory.p = NULL; 
+			
+			
+			data[i].entry = SPE_DEFAULT_ENTRY;
+			data[i].flags = 0;
+			data[i].argp.p = &m_activeSpuStatus[i];
+			data[i].envp.p = NULL;
+			
+		    /* Create thread for each SPE context */
+			if (pthread_create(&data[i].pthread, NULL, &ppu_pthread_function, &(data[i]) ))
+			{
+			      perror ("Failed creating thread");
+		          exit(1);
+			}
+			/*
+			else
+			{
+				printf("started thread %d\n",i);
+			}*/
+		}		
+	}
+	
+	
+	for (int i=0; i < numThreads; i++)
+	{
+		if(data[i].context != NULL) 
+		{
+			while( m_activeSpuStatus[i].m_status == Spu_Status_Startup)
+			{
+				// wait for spu to set up
+				sched_yield();
+			}
+			printf("Spu %d is ready\n", i);
+		}
+	}
 }
 
 ///tell the task scheduler we are done with the SPU tasks
-void SpuLibspe2Support::stopSPUs()
+void SpuLibspe2Support::stopSPU()
 {
 	// wait for all threads to finish 
-	for ( i=0; i<N; i++ ) { 
+	int i;
+	for ( i = 0; i < this->numThreads; i++ ) 
+	{ 
+		
+		unsigned int event = Spu_Mailbox_Event_Shutdown;
+		spe_context_ptr_t context = data[i].context;
+		spe_in_mbox_write(context, &event, 1, SPE_MBOX_ALL_BLOCKING);
 		pthread_join (data[i].pthread, NULL); 
+		
 	} 
 	// close SPE program 
 	spe_image_close(program); 
 	// destroy SPE contexts 
-	for ( i=0; i<N; i++ ) { 
+	for ( i = 0; i < this->numThreads; i++ ) 
+	{ 
+		if(data[i].context != NULL)
+		{
 			spe_context_destroy (data[i].context);
 		}
+	} 
 	
+	m_activeSpuStatus.clear();
 	
 }
 
-#endif// USE_LIBSPE2
+
+
+#endif //USE_LIBSPE2
 

Extras/BulletMultiThreaded/SpuLibspe2Support.h

@@ -13,119 +13,161 @@ subject to the following restrictions:
 3. This notice may not be removed or altered from any source distribution.
 */
 
-///USE_LIBSPE2: this define should be in the build system, or in <LinearMath/btScalar.h>
-//#define USE_LIBSPE2 1
-#ifdef USE_LIBSPE2
 
 #ifndef SPU_LIBSPE2_SUPPORT_H
 #define SPU_LIBSPE2_SUPPORT_H
 
-#include "LinearMath/btAlignedObjectArray.h"
-#include <libspe2.h>
-#include <pthread.h>
-
-
-/**
- * Note:
- * The order of elements in this enum are important, that's why each one is explicitly
- * given a value.  They will correspond to the .elf names/addresses that will be
- * loaded into Libspe2.
- * Mixing up these values will cause the wrong code to execute, for instance, the
- * solver may be asked to do a collision detection job.
- */
-#ifdef WIN32 // original enum, but for libspe2 will pass the SPE program handle ptr directly
-typedef enum {
-	SPU_ELF_COLLISION_DETECTION=0,
-	SPU_ELF_SAMPLE,
-//SPU_ELF_INTEGRATION,
-//SPU_ELF_SOLVER,
-	SPU_ELF_LAST,
-} SpuLibspe2ElfId_t;
-#endif
-#ifdef WIN32
-#include <malloc.h>
-#define memalign(alignment, size) malloc(size);
-#else
-#include <stdlib.h>
-#endif // WIN32
-
-#define MAXSPUS 16
-typedef struct ppu_pthread_data {
-spe_context_ptr_t context;
-pthread_t pthread;
-unsigned int entry;
-unsigned int flags;
-void *argp;
-void *envp;
-spe_stop_info_t stopinfo;
-} ppu_pthread_data_t;
-
-void *ppu_pthread_function(void *arg)
-{
-    ppu_pthread_data_t datap = *(ppu_pthread_data_t *)arg;
-    int rc;
-    do {
-        rc = spe_context_run(datap->context, &datap->entry, datap->flags, datap->argp, datap->envp, &datap->stopinfo);
-    } while (rc > 0); // loop until exit or error, and while any stop & signal
-    pthread_exit(NULL);
-}
-
-
-
 #include <LinearMath/btScalar.h> //for uint32_t etc.
 
-///placeholder, until libspe2 support is there
+#ifdef USE_LIBSPE2
+
+#include <stdlib.h>
+#include <stdio.h>
+//#include "SpuNarrowPhaseCollisionTask/SpuGatheringCollisionTask.h"
+#include "PlatformDefinitions.h"
+
+
+//extern struct SpuGatherAndProcessPairsTaskDesc;
+
+enum
+{
+	Spu_Mailbox_Event_Nothing = 0,
+	Spu_Mailbox_Event_Task = 1,
+	Spu_Mailbox_Event_Shutdown = 2,
+	
+	Spu_Mailbox_Event_ForceDword = 0xFFFFFFFF
+	
+};
+
+enum
+{
+	Spu_Status_Free = 0,
+	Spu_Status_Occupied = 1,
+	Spu_Status_Startup = 2,
+	
+	Spu_Status_ForceDword = 0xFFFFFFFF
+	
+};
+
+
 struct btSpuStatus
 {
 	uint32_t	m_taskId;
 	uint32_t	m_commandId;
 	uint32_t	m_status;
 
-	struct SpuGatherAndProcessPairsTaskDesc* m_taskDesc;
+	addr64 m_taskDesc;
+	addr64 m_lsMemory;
+	
+}
+__attribute__ ((aligned (128)))
+;
+
+
+
+#ifndef __SPU__
+
+#include "LinearMath/btAlignedObjectArray.h"
+#include "SpuCollisionTaskProcess.h"
+#include "SpuSampleTaskProcess.h"
+#include "btThreadSupportInterface.h"
+#include <libspe2.h>
+#include <pthread.h>
+#include <sched.h>
+
+#define MAX_SPUS 4 
+
+typedef struct ppu_pthread_data 
+{
+	spe_context_ptr_t context;
+	pthread_t pthread;
+	unsigned int entry;
+	unsigned int flags;
+	addr64 argp;
+	addr64 envp;
+	spe_stop_info_t stopinfo;
+} ppu_pthread_data_t;
+
+
+static void *ppu_pthread_function(void *arg)
+{
+    ppu_pthread_data_t * datap = (ppu_pthread_data_t *)arg;
+    /*
+    int rc;
+    do 
+    {*/
+        spe_context_run(datap->context, &datap->entry, datap->flags, datap->argp.p, datap->envp.p, &datap->stopinfo);
+        if (datap->stopinfo.stop_reason == SPE_EXIT) 
+        {
+           if (datap->stopinfo.result.spe_exit_code != 0) 
+           {
+             perror("FAILED: SPE returned a non-zero exit status: \n");
+             exit(1);
+           }
+         } 
+        else 
+         {
+           perror("FAILED: SPE abnormally terminated\n");
+           exit(1);
+         }
+        
+        
+    //} while (rc > 0); // loop until exit or error, and while any stop & signal
+    pthread_exit(NULL);
+}
+
+
 
-	void*	m_threadHandle;
-	void*	m_lsMemory;
-
-	void*	m_eventStartHandle;
-	char	m_eventStartHandleName[32];
-
-	void*	m_eventCompletetHandle;
-	char	m_eventCompletetHandleName[32];
 
 
-};
 
 ///SpuLibspe2Support helps to initialize/shutdown libspe2, start/stop SPU tasks and communication
-class SpuLibspe2Support {
+class SpuLibspe2Support : public btThreadSupportInterface
+{
 
 	btAlignedObjectArray<btSpuStatus>	m_activeSpuStatus;
 	
 public:
-	///Setup and initialize SPU/CELL/Libspe2
+	//Setup and initialize SPU/CELL/Libspe2
 	SpuLibspe2Support(spe_program_handle_t *speprog,int numThreads);
+	
 	// SPE program handle ptr.
 	spe_program_handle_t *program;
+	
 	// SPE program data
 	ppu_pthread_data_t data[MAX_SPUS];
-	// num SPE Threads
+	
-	unsigned int N;
+	//cleanup/shutdown Libspe2
-	///cleanup/shutdown Libspe2
 	~SpuLibspe2Support();
 
 	///send messages to SPUs
 	void sendRequest(uint32_t uiCommand, uint32_t uiArgument0, uint32_t uiArgument1=0);
 
-	///check for messages from SPUs
+	//check for messages from SPUs
 	void waitForResponse(unsigned int *puiArgument0, unsigned int *puiArgument1);
 
-	///start the spus (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
+	//start the spus (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
-	void startSPUs(int numThreads);
+	virtual void startSPU();
 
-	///tell the task scheduler we are done with the SPU tasks
+	//tell the task scheduler we are done with the SPU tasks
-	void stopSPUs();
+	virtual void stopSPU();
+
+private:
+	
+	///start the spus (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
+	void internal_startSPU();
+	
+	
+	int numThreads;
 
 };
 
-#endif //SPU_LIBSPE2_SUPPORT_H
+#endif // NOT __SPU__
 
 #endif //USE_LIBSPE2
+
+#endif //SPU_LIBSPE2_SUPPORT_H
+
+
+
+