Merge pull request #30 from echoedit/dev

Refactor some perf model to OO

9 files changed, 1072 insertions, 762 deletions

diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc
index 716e297..5cefe60 100644
--- a/libcuda/cuda_runtime_api.cc
+++ b/libcuda/cuda_runtime_api.cc
@@ -150,9 +150,6 @@
 #endif
 
 
-extern void synchronize();
-extern void exit_simulation();
-
 /*DEVICE_BUILTIN*/
 struct cudaArray
 {
@@ -176,8 +173,6 @@ struct cudaArray
 
 cudaError_t g_last_cudaError = cudaSuccess;
 
-//extern stream_manager *g_stream_manager();
-
 void register_ptx_function( const char *name, function_info *impl )
 {
 	// no longer need this
@@ -199,8 +194,7 @@ void register_ptx_function( const char *name, function_info *impl )
 
 struct _cuda_device_id *gpgpu_context::GPGPUSim_Init()
 {
-	//static _cuda_device_id *the_device = NULL;
-	_cuda_device_id *the_device = GPGPUsim_ctx_ptr()->the_cude_device;
+	_cuda_device_id *the_device = the_gpgpusim->the_cude_device;
 	if( !the_device ) {
 		gpgpu_sim *the_gpu = gpgpu_ptx_sim_init_perf();
 
@@ -245,22 +239,21 @@ struct _cuda_device_id *gpgpu_context::GPGPUSim_Init()
 		prop->maxThreadsPerMultiProcessor = the_gpu->threads_per_core();
 #endif
 		the_gpu->set_prop(prop);
-		GPGPUsim_ctx_ptr()->the_cude_device = new _cuda_device_id(the_gpu);
-		the_device = GPGPUsim_ctx_ptr()->the_cude_device;
+		the_gpgpusim->the_cude_device = new _cuda_device_id(the_gpu);
+		the_device = the_gpgpusim->the_cude_device;
 	}
 	start_sim_thread(1);
 	return the_device;
 }
 
-static CUctx_st* GPGPUSim_Context()
+CUctx_st* GPGPUSim_Context(gpgpu_context * ctx)
 {
 	//static CUctx_st *the_context = NULL;
-	gpgpu_context *cur_ctx = GPGPU_Context();
-	CUctx_st *the_context = GPGPUsim_ctx_ptr()->the_context;
+	CUctx_st *the_context = ctx->the_gpgpusim->the_context;
 	if( the_context == NULL ) {
-		_cuda_device_id *the_gpu = cur_ctx->GPGPUSim_Init();
-		GPGPUsim_ctx_ptr()->the_context = new CUctx_st(the_gpu);
-		the_context = GPGPUsim_ctx_ptr()->the_context;
+		_cuda_device_id *the_gpu = ctx->GPGPUSim_Init();
+		ctx->the_gpgpusim->the_context = new CUctx_st(the_gpu);
+		the_context = ctx->the_gpgpusim->the_context;
 	}
 	return the_context;
 }
@@ -276,9 +269,9 @@ gpgpu_context* GPGPU_Context()
 
  void ptxinfo_data::ptxinfo_addinfo()
 {
+	CUctx_st *context = GPGPUSim_Context(gpgpu_ctx);
 	 if(!get_ptxinfo_kname()){
 		 /* This info is not per kernel (since CUDA 5.0 some info (e.g. gmem, and cmem) is added at the beginning for the whole binary ) */
-		CUctx_st *context = GPGPUSim_Context();
 		print_ptxinfo();
 		context->add_ptxinfo(get_ptxinfo());
 		clear_ptxinfo();
@@ -289,7 +282,6 @@ gpgpu_context* GPGPU_Context()
 		clear_ptxinfo();
 		return;
 	}
-	CUctx_st *context = GPGPUSim_Context();
 	print_ptxinfo();
 	context->add_ptxinfo( get_ptxinfo_kname(), get_ptxinfo() );
 	clear_ptxinfo();
@@ -585,7 +577,7 @@ void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx =
 	printf("GPGPU-Sim PTX: ERROR ** this version of GPGPU-Sim requires CUDA 2.1 or higher\n");
 	exit(1);
 #endif
-	CUctx_st *context = GPGPUSim_Context();
+	CUctx_st *context = GPGPUSim_Context(ctx);
 	static unsigned next_fat_bin_handle = 1;
 	if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) {
 		// The following workaround has only been verified on 64-bit systems. 
@@ -732,7 +724,7 @@ void cudaRegisterFunctionInternal(
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	CUctx_st *context = GPGPUSim_Context();
+	CUctx_st *context = GPGPUSim_Context(ctx);
 	unsigned fat_cubin_handle = (unsigned)(unsigned long long)fatCubinHandle;
 	printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n",
 			deviceFun, hostFun, fat_cubin_handle);
@@ -763,7 +755,7 @@ void cudaRegisterVarInternal(
     }
 	printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName);
 	printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size);
-	if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump())
+	if(GPGPUSim_Context(ctx)->get_device()->get_gpgpu()->get_config().use_cuobjdump())
 		ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle);
 	fflush(stdout);
 	if ( constant && !global && !ext ) {
@@ -872,7 +864,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx =
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	CUctx_st* context = GPGPUSim_Context();
+	CUctx_st* context = GPGPUSim_Context(ctx);
 	char *mode = getenv("PTX_SIM_MODE_FUNC");
 	if( mode )
 		sscanf(mode,"%u", &(ctx->func_sim->g_ptx_sim_mode));
@@ -944,7 +936,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx =
 	printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n",
 			kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z );
 	stream_operation op(grid,ctx->func_sim->g_ptx_sim_mode,stream);
-	g_stream_manager()->push(op);
+	ctx->the_gpgpusim->g_stream_manager->push(op);
 	ctx->api->g_cuda_launch_stack.pop_back();
 	return g_last_cudaError = cudaSuccess;
 }
@@ -960,7 +952,7 @@ cudaError_t cudaMallocInternal(void **devPtr, size_t size, gpgpu_context* gpgpu_
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	CUctx_st* context = GPGPUSim_Context();
+	CUctx_st* context = GPGPUSim_Context(ctx);
 	*devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size);
 	if(g_debug_execution >= 3){
 		printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *devPtr);
@@ -994,6 +986,29 @@ cudaError_t cudaMallocHostInternal(void **ptr, size_t size, gpgpu_context* gpgpu
 	}
 }
 
+__host__ cudaError_t CUDARTAPI cudaMallocPitchInternal(void **devPtr, size_t *pitch, size_t width, size_t height, gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	unsigned malloc_width_inbytes = width;
+	printf("GPGPU-Sim PTX: cudaMallocPitch (width = %d)\n", malloc_width_inbytes);
+	CUctx_st* context = GPGPUSim_Context(ctx);
+	*devPtr = context->get_device()->get_gpgpu()->gpu_malloc(malloc_width_inbytes*height);
+	pitch[0] = malloc_width_inbytes;
+	if ( *devPtr  ) {
+		return g_last_cudaError = cudaSuccess;
+	} else {
+		return g_last_cudaError = cudaErrorMemoryAllocation;
+	}
+}
+
 cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL)
 {
     gpgpu_context *ctx;
@@ -1011,7 +1026,7 @@ cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsign
 	//only cpu memory allocation happens in cudaHostAlloc. Linking with device pointer to pinned memory happens here.
 	//TODO: once kernel is executed, the contents in global pointer of GPU must be copied back to CPU host pointer!
 	flags=0;
-	CUctx_st* context = GPGPUSim_Context();
+	CUctx_st* context = GPGPUSim_Context(ctx);
 	gpgpu_t *gpu = context->get_device()->get_gpgpu();
 	std::map<void *, size_t>::const_iterator i = ctx->api->pinned_memory_size.find(pHost);
 	assert(i != ctx->api->pinned_memory_size.end());
@@ -1031,7 +1046,7 @@ cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsign
 	}
 }
 
-cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL)
+__host__ cudaError_t CUDARTAPI cudaMallocArrayInternal(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1), gpgpu_context* gpgpu_ctx = NULL)
 {
     gpgpu_context *ctx;
     if (gpgpu_ctx){
@@ -1042,226 +1057,8 @@ cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-#ifdef OPENGL_SUPPORT
-	GLint buffer_size=0;
-	CUctx_st* context = GPGPUSim_Context();
-
-	glbmap_entry_t *p = ctx->api->g_glbmap;
-	while ( p && p->m_bufferObj != bufferObj )
-		p = p->m_next;
-	if ( p == NULL ) {
-		glBindBuffer(GL_ARRAY_BUFFER,bufferObj);
-		glGetBufferParameteriv(GL_ARRAY_BUFFER,GL_BUFFER_SIZE,&buffer_size);
-		assert( buffer_size != 0 );
-		*devPtr = context->get_device()->get_gpgpu()->gpu_malloc(buffer_size);
-
-		// create entry and insert to front of list
-		glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t));
-		n->m_next = ctx->api->g_glbmap;
-		ctx->api->g_glbmap = n;
-
-		// initialize entry
-		n->m_bufferObj = bufferObj;
-		n->m_devPtr = *devPtr;
-		n->m_size = buffer_size;
-
-		p = n;
-	} else {
-		buffer_size = p->m_size;
-		*devPtr = p->m_devPtr;
-	}
-
-	if ( *devPtr  ) {
-		char *data = (char *) calloc(p->m_size,1);
-		glGetBufferSubData(GL_ARRAY_BUFFER,0,buffer_size,data);
-		memcpy_to_gpu( (size_t) *devPtr, data, buffer_size );
-		free(data);
-		printf("GPGPU-Sim PTX: cudaGLMapBufferObject %zu bytes starting at 0x%llx..\n", (size_t)buffer_size,
-				(unsigned long long) *devPtr);
-		return g_last_cudaError = cudaSuccess;
-	} else {
-		return g_last_cudaError = cudaErrorMemoryAllocation;
-	}
-
-	return g_last_cudaError = cudaSuccess;
-#else
-	fflush(stdout);
-	fflush(stderr);
-	printf("GPGPU-Sim PTX: GPGPU-Sim support for OpenGL integration disabled -- exiting\n");
-	fflush(stdout);
-	exit(50);
-#endif
-}
-
-#if CUDART_VERSION >= 6050
-CUresult
-cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path,
-    unsigned int numOptions, CUjit_option *options, void **optionValues, gpgpu_context* gpgpu_ctx = NULL)
-{
-    gpgpu_context *ctx;
-    if (gpgpu_ctx){
-	ctx = gpgpu_ctx;
-    } else {
-	ctx = GPGPU_Context();
-    }
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-    static bool addedFile = false;
-    if (addedFile){
-        printf("GPGPU-Sim PTX: ERROR: cuLinkAddFile does not support multiple files\n");
-        abort();
-    }
-
-    //blocking
-    assert(type==CU_JIT_INPUT_PTX);
-	CUctx_st *context = GPGPUSim_Context();
-    char *file = getenv("PTX_JIT_PATH");
-    if(file==NULL){
-        printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n");
-        abort();
-    }
-    strcat(file,"/");
-    strcat(file,path);
-	symbol_table *symtab = ctx->gpgpu_ptx_sim_load_ptx_from_filename( file );
-    std::string fname(path);
-    ctx->api->name_symtab[fname] = symtab;
-    context->add_binary(symtab, 1);
-    ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu());
-    ctx->api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu());
-    addedFile = true;
-	return CUDA_SUCCESS;
-}
-#endif
-
-#if (CUDART_VERSION >= 2010)
-
-cudaError_t cudaHostAllocInternal(void **pHost,  size_t bytes, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL)
-{
-    gpgpu_context *ctx;
-    if (gpgpu_ctx){
-	ctx = gpgpu_ctx;
-    } else {
-	ctx = GPGPU_Context();
-    }
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	*pHost = malloc(bytes);
-	//need to track the size allocated so that cudaHostGetDevicePointer() can function properly.
-	//TODO: vary this function behavior based on flags value (following nvidia documentation)
-	ctx->api->pinned_memory_size[*pHost]=bytes;
-	if( *pHost )
-		return g_last_cudaError = cudaSuccess;
-	else
-		return g_last_cudaError = cudaErrorMemoryAllocation;
-}
-
-#endif
-
-size_t getMaxThreadsPerBlock(struct cudaFuncAttributes *attr, gpgpu_context *ctx) {
-  _cuda_device_id *dev = ctx->GPGPUSim_Init();
-  struct cudaDeviceProp prop;
-
-  prop = *dev->get_prop();
-
-  size_t max = prop.maxThreadsPerBlock;
-
-  if (attr->numRegs && (prop.regsPerBlock / attr->numRegs) < max)
-    max = prop.regsPerBlock / attr->numRegs;
-
-  if (attr->sharedSizeBytes && (prop.sharedMemPerBlock / attr->sharedSizeBytes) < max)
-      max = prop.sharedMemPerBlock / attr->sharedSizeBytes;
-
-  return max;
-}
-
-cudaError_t CUDARTAPI cudaFuncGetAttributesInternal(struct cudaFuncAttributes *attr, const char *hostFun, gpgpu_context* gpgpu_ctx = NULL )
-{
-    gpgpu_context *ctx;
-    if (gpgpu_ctx){
-	ctx = gpgpu_ctx;
-    } else {
-	ctx = GPGPU_Context();
-    }
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	CUctx_st *context = GPGPUSim_Context();
-	function_info *entry = context->get_kernel(hostFun);
-	if( entry ) {
-		const struct gpgpu_ptx_sim_info *kinfo = entry->get_kernel_info();
-		attr->sharedSizeBytes = kinfo->smem;
-		attr->constSizeBytes  = kinfo->cmem;
-		attr->localSizeBytes  = kinfo->lmem;
-		attr->numRegs         = kinfo->regs;
-		if(kinfo->maxthreads > 0)
-		  attr->maxThreadsPerBlock = kinfo->maxthreads;
-		else
-		  attr->maxThreadsPerBlock = getMaxThreadsPerBlock(attr, ctx);
-#if CUDART_VERSION >= 3000
-		attr->ptxVersion      = kinfo->ptx_version;
-		attr->binaryVersion   = kinfo->sm_target;
-#endif
-	}
-	return g_last_cudaError = cudaSuccess;
-}
-
-
-/*******************************************************************************
- *                                                                              *
- *                                                                              *
- *                                                                              *
- *******************************************************************************/
-
-extern "C" {
-
-/*******************************************************************************
- *                                                                              *
- *                                                                              *
- *                                                                              *
- *******************************************************************************/
-cudaError_t cudaPeekAtLastError(void)
-{
-	return g_last_cudaError;
-}
-
-__host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) 
-{
-    return cudaMallocInternal(devPtr, size);
-}
-
-__host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size)
-{
-    return cudaMallocHostInternal(ptr, size);
-}
-__host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, size_t width, size_t height)
-{
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	unsigned malloc_width_inbytes = width;
-	printf("GPGPU-Sim PTX: cudaMallocPitch (width = %d)\n", malloc_width_inbytes);
-	CUctx_st* ctx = GPGPUSim_Context();
-	*devPtr = ctx->get_device()->get_gpgpu()->gpu_malloc(malloc_width_inbytes*height);
-	pitch[0] = malloc_width_inbytes;
-	if ( *devPtr  ) {
-		return g_last_cudaError = cudaSuccess;
-	} else {
-		return g_last_cudaError = cudaErrorMemoryAllocation;
-	}
-}
-
-__host__ cudaError_t CUDARTAPI cudaMallocArray(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1))
-{
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
 	unsigned size = width * height * ((desc->x + desc->y + desc->z + desc->w)/8);
-	CUctx_st* context = GPGPUSim_Context();
+	CUctx_st* context = GPGPUSim_Context(ctx);
 	(*array) = (struct cudaArray*) malloc(sizeof(struct cudaArray));
 	(*array)->desc = *desc;
 	(*array)->width = width;
@@ -1278,41 +1075,14 @@ __host__ cudaError_t CUDARTAPI cudaMallocArray(struct cudaArray **array, const s
 	}
 }
 
-__host__ cudaError_t CUDARTAPI cudaFree(void *devPtr)
-{
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	// TODO...  manage g_global_mem space?
-	return g_last_cudaError = cudaSuccess;
-}
-__host__ cudaError_t CUDARTAPI cudaFreeHost(void *ptr)
+__host__ cudaError_t CUDARTAPI cudaMemcpyInternal(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
     }
-	free (ptr);  // this will crash the system if called twice
-	return g_last_cudaError = cudaSuccess;
-}
-
-__host__ cudaError_t CUDARTAPI cudaFreeArray(struct cudaArray *array)
-{
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	// TODO...  manage g_global_mem space?
-	return g_last_cudaError = cudaSuccess;
-};
-
-
-/*******************************************************************************
- *                                                                              *
- *                                                                              *
- *                                                                              *
- *******************************************************************************/
-
-__host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind)
-{
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
@@ -1321,21 +1091,21 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t cou
 	if(g_debug_execution >= 3)
 		printf("GPGPU-Sim PTX: cudaMemcpy(): devPtr = %p\n", dst);
 	if( kind == cudaMemcpyHostToDevice )
-		g_stream_manager()->push( stream_operation(src,(size_t)dst,count,0) );
+		ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) );
 	else if( kind == cudaMemcpyDeviceToHost )
-		g_stream_manager()->push( stream_operation((size_t)src,dst,count,0) );
+		ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,0) );
 	else if( kind == cudaMemcpyDeviceToDevice )
-		g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,0) );
+		ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) );
 	else if ( kind == cudaMemcpyDefault ) {
 		if ((size_t)src >= GLOBAL_HEAP_START) {
 			if ((size_t)dst >= GLOBAL_HEAP_START)
-				g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device
+				ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device
 			else
-				g_stream_manager()->push( stream_operation((size_t)src,dst,count,0) ); // device to host
+				ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); // device to host
 		}
 		else {
 			if ((size_t)dst >= GLOBAL_HEAP_START)
-				g_stream_manager()->push( stream_operation(src,(size_t)dst,count,0) );
+				ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) );
 			else {
 				printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported transfer: host to host\n");
 				abort();
@@ -1349,12 +1119,18 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t cou
 	return g_last_cudaError = cudaSuccess;
 }
 
-__host__ cudaError_t CUDARTAPI cudaMemcpyToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind)
+__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayInternal(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL)
 { 
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	CUctx_st *context = GPGPUSim_Context();
+	CUctx_st *context = GPGPUSim_Context(ctx);
 	gpgpu_t *gpu = context->get_device()->get_gpgpu();
 	size_t size = count;
 	printf("GPGPU-Sim PTX: cudaMemcpyToArray\n");
@@ -1372,33 +1148,18 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyToArray(struct cudaArray *dst, size_t w
 	return g_last_cudaError = cudaSuccess;
 }
 
-
-__host__ cudaError_t CUDARTAPI cudaMemcpyFromArray(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind)
-{
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
-}
-
-
-__host__ cudaError_t CUDARTAPI cudaMemcpyArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t count, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice))
+__host__ cudaError_t CUDARTAPI cudaMemcpy2DInternal(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
     }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
-}
-
-
-__host__ cudaError_t CUDARTAPI cudaMemcpy2D(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind)
-{
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	CUctx_st *context = GPGPUSim_Context();
+	CUctx_st *context = GPGPUSim_Context(ctx);
 	gpgpu_t *gpu = context->get_device()->get_gpgpu();
 	size_t size = spitch*height;
 	gpgpusim_ptx_assert( (dpitch==spitch), "different src and dst pitch not supported yet" );
@@ -1415,13 +1176,18 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy2D(void *dst, size_t dpitch, const void
 	return g_last_cudaError = cudaSuccess;
 }
 
-
-__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind)
+__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayInternal(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL)
 {
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	CUctx_st *context = GPGPUSim_Context();
+	CUctx_st *context = GPGPUSim_Context(ctx);
 	gpgpu_t *gpu = context->get_device()->get_gpgpu();
 	size_t size = spitch*height;
 	size_t channel_size = dst->desc.w+dst->desc.x+dst->desc.y+dst->desc.z;
@@ -1447,29 +1213,14 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy2DToArray(struct cudaArray *dst, size_t
 	return g_last_cudaError = cudaSuccess;
 }
 
-
-__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArray(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind)
-{
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
-}
-
-
-__host__ cudaError_t CUDARTAPI cudaMemcpy2DArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t width, size_t height, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice))
+__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbolInternal(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice), gpgpu_context* gpgpu_ctx = NULL)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
     }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
-}
-
-
-__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice))
-{
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
@@ -1477,52 +1228,47 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void
 	assert(kind == cudaMemcpyHostToDevice);
 	printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol);
 	//stream_operation( const char *symbol, const void *src, size_t count, size_t offset )
-	g_stream_manager()->push( stream_operation(src,symbol,count,offset,0) );
+	ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,symbol,count,offset,0) );
 	//gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu());
 	return g_last_cudaError = cudaSuccess;
 }
 
 
-__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost))
+__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbolInternal(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost), gpgpu_context* gpgpu_ctx = NULL)
 {
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
 	//CUctx_st *context = GPGPUSim_Context();
 	assert(kind == cudaMemcpyDeviceToHost);
 	printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol);
-	g_stream_manager()->push( stream_operation(symbol,dst,count,offset,0) );
+	ctx->the_gpgpusim->g_stream_manager->push( stream_operation(symbol,dst,count,offset,0) );
 	//gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu());
 	return g_last_cudaError = cudaSuccess;
 }
 
-__host__ cudaError_t CUDARTAPI cudaMemGetInfo (size_t *free, size_t *total){
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	//placeholder; should interact with cudaMalloc and cudaFree?
-	*free = 10000000000;
-	*total = 10000000000;
-
-	return g_last_cudaError = cudaSuccess;
-}
-
-/*******************************************************************************
- *                                                                              *
- *                                                                              *
- *                                                                              *
- *******************************************************************************/
-
-__host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream)
+__host__ cudaError_t CUDARTAPI cudaMemcpyAsyncInternal(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL)
 {
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
 	struct CUstream_st *s = (struct CUstream_st *)stream;
 	switch( kind ) {
-	case cudaMemcpyHostToDevice: g_stream_manager()->push( stream_operation(src,(size_t)dst,count,s) ); break;
-	case cudaMemcpyDeviceToHost: g_stream_manager()->push( stream_operation((size_t)src,dst,count,s) ); break;
-	case cudaMemcpyDeviceToDevice: g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break;
+	case cudaMemcpyHostToDevice: ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,s) ); break;
+	case cudaMemcpyDeviceToHost: ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,s) ); break;
+	case cudaMemcpyDeviceToDevice: ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break;
 	default:
 		abort();
 	}
@@ -1530,98 +1276,249 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_
 }
 
 
-__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream)
+#if (CUDART_VERSION >= 8000)
+cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
     }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
+	printf("GPGPU-Sim PTX: cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags %p\n", hostFunc);
+	CUctx_st *context = GPGPUSim_Context(ctx);
+	function_info *entry = context->get_kernel(hostFunc);
+	printf("Calculate Maxium Active Block with function ptr=%p, blockSize=%d, SMemSize=%d\n", hostFunc, blockSize, dynamicSMemSize);
+	if (flags == cudaOccupancyDefault) {
+		//create kernel_info based on entry
+		dim3 gridDim(context->get_device()->get_gpgpu()->max_cta_per_core()
+                        * context->get_device()->get_gpgpu()->get_config().num_shader());
+		dim3 blockDim(blockSize);
+		kernel_info_t result(gridDim, blockDim, entry);
+		//if(entry == NULL){
+		//	*numBlocks = 1;
+		//	return g_last_cudaError = cudaErrorUnknown;
+		//}
+		*numBlocks = context->get_device()->get_gpgpu()->get_max_cta(result);
+		printf("Maximum size is %d with gridDim %d and blockDim %d\n", *numBlocks, gridDim.x, blockDim.x);
+		return g_last_cudaError = cudaSuccess;
+	} else {
+		cuda_not_implemented(__my_func__,__LINE__);
+		return g_last_cudaError = cudaErrorUnknown;
+	}
 }
 
+#endif
 
-__host__ cudaError_t CUDARTAPI cudaMemcpyFromArrayAsync(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream)
+__host__ cudaError_t CUDARTAPI cudaMemsetInternal(void *mem, int c, size_t count, gpgpu_context* gpgpu_ctx = NULL)
 {
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
+	CUctx_st *context = GPGPUSim_Context(ctx);
+	gpgpu_t *gpu = context->get_device()->get_gpgpu();
+	gpu->gpu_memset((size_t)mem, c, count);
+	return g_last_cudaError = cudaSuccess;
 }
 
-
-__host__ cudaError_t CUDARTAPI cudaMemcpy2DAsync(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream)
+//memset operation is done but i think its not async?
+__host__ cudaError_t CUDARTAPI cudaMemsetAsyncInternal(void *mem, int c, size_t count, 	cudaStream_t stream=0, gpgpu_context* gpgpu_ctx = NULL)
 {
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
+	printf("GPGPU-Sim PTX: WARNING: Asynchronous memset not supported (%s)\n", __my_func__);
+	CUctx_st *context = GPGPUSim_Context(ctx);
+	gpgpu_t *gpu = context->get_device()->get_gpgpu();
+	gpu->gpu_memset((size_t)mem, c, count);
+	return g_last_cudaError = cudaSuccess;
 }
 
-
-__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream)
+cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL)
 {
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
-}
+#ifdef OPENGL_SUPPORT
+	gpgpu_context *ctx;
+	if (gpgpu_ctx){
+	    ctx = gpgpu_ctx;
+	} else {
+	    ctx = GPGPU_Context();
+	}
+	GLint buffer_size=0;
+	CUctx_st* context = GPGPUSim_Context(ctx);
+
+	glbmap_entry_t *p = ctx->api->g_glbmap;
+	while ( p && p->m_bufferObj != bufferObj )
+		p = p->m_next;
+	if ( p == NULL ) {
+		glBindBuffer(GL_ARRAY_BUFFER,bufferObj);
+		glGetBufferParameteriv(GL_ARRAY_BUFFER,GL_BUFFER_SIZE,&buffer_size);
+		assert( buffer_size != 0 );
+		*devPtr = context->get_device()->get_gpgpu()->gpu_malloc(buffer_size);
 
+		// create entry and insert to front of list
+		glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t));
+		n->m_next = ctx->api->g_glbmap;
+		ctx->api->g_glbmap = n;
 
-__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArrayAsync(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream)
+		// initialize entry
+		n->m_bufferObj = bufferObj;
+		n->m_devPtr = *devPtr;
+		n->m_size = buffer_size;
+
+		p = n;
+	} else {
+		buffer_size = p->m_size;
+		*devPtr = p->m_devPtr;
+	}
+
+	if ( *devPtr  ) {
+		char *data = (char *) calloc(p->m_size,1);
+		glGetBufferSubData(GL_ARRAY_BUFFER,0,buffer_size,data);
+		memcpy_to_gpu( (size_t) *devPtr, data, buffer_size );
+		free(data);
+		printf("GPGPU-Sim PTX: cudaGLMapBufferObject %zu bytes starting at 0x%llx..\n", (size_t)buffer_size,
+				(unsigned long long) *devPtr);
+		return g_last_cudaError = cudaSuccess;
+	} else {
+		return g_last_cudaError = cudaErrorMemoryAllocation;
+	}
+
+	return g_last_cudaError = cudaSuccess;
+#else
+	fflush(stdout);
+	fflush(stderr);
+	printf("GPGPU-Sim PTX: GPGPU-Sim support for OpenGL integration disabled -- exiting\n");
+	fflush(stdout);
+	exit(50);
+#endif
+}
+
+#if CUDART_VERSION >= 6050
+CUresult
+cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path,
+    unsigned int numOptions, CUjit_option *options, void **optionValues, gpgpu_context* gpgpu_ctx = NULL)
 {
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
+    static bool addedFile = false;
+    if (addedFile){
+        printf("GPGPU-Sim PTX: ERROR: cuLinkAddFile does not support multiple files\n");
+        abort();
+    }
+
+    //blocking
+    assert(type==CU_JIT_INPUT_PTX);
+	CUctx_st *context = GPGPUSim_Context(ctx);
+    char *file = getenv("PTX_JIT_PATH");
+    if(file==NULL){
+        printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n");
+        abort();
+    }
+    strcat(file,"/");
+    strcat(file,path);
+	symbol_table *symtab = ctx->gpgpu_ptx_sim_load_ptx_from_filename( file );
+    std::string fname(path);
+    ctx->api->name_symtab[fname] = symtab;
+    context->add_binary(symtab, 1);
+    ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu());
+    ctx->api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu());
+    addedFile = true;
+	return CUDA_SUCCESS;
 }
+#endif
 
-#if (CUDART_VERSION >= 8000)
-cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags)
+#if (CUDART_VERSION >= 2010)
+
+cudaError_t cudaHostAllocInternal(void **pHost,  size_t bytes, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL)
 {
-	printf("GPGPU-Sim PTX: cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags %p\n", hostFunc);
-	CUctx_st *context = GPGPUSim_Context();
-	function_info *entry = context->get_kernel(hostFunc);
-	printf("Calculate Maxium Active Block with function ptr=%p, blockSize=%d, SMemSize=%d\n", hostFunc, blockSize, dynamicSMemSize);
-	if (flags == cudaOccupancyDefault) {
-		//create kernel_info based on entry
-		dim3 gridDim(context->get_device()->get_gpgpu()->max_cta_per_core()
-                        * context->get_device()->get_gpgpu()->get_config().num_shader());
-		dim3 blockDim(blockSize);
-		kernel_info_t result(gridDim, blockDim, entry);
-		//if(entry == NULL){
-		//	*numBlocks = 1;
-		//	return g_last_cudaError = cudaErrorUnknown;
-		//}
-		*numBlocks = context->get_device()->get_gpgpu()->get_max_cta(result);
-		printf("Maximum size is %d with gridDim %d and blockDim %d\n", *numBlocks, gridDim.x, blockDim.x);
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	*pHost = malloc(bytes);
+	//need to track the size allocated so that cudaHostGetDevicePointer() can function properly.
+	//TODO: vary this function behavior based on flags value (following nvidia documentation)
+	ctx->api->pinned_memory_size[*pHost]=bytes;
+	if( *pHost )
 		return g_last_cudaError = cudaSuccess;
-	} else {
-		cuda_not_implemented(__my_func__,__LINE__);
-		return g_last_cudaError = cudaErrorUnknown;
-	}
+	else
+		return g_last_cudaError = cudaErrorMemoryAllocation;
 }
 
 #endif
 
+size_t getMaxThreadsPerBlock(struct cudaFuncAttributes *attr, gpgpu_context *ctx) {
+  _cuda_device_id *dev = ctx->GPGPUSim_Init();
+  struct cudaDeviceProp prop;
 
+  prop = *dev->get_prop();
 
-/*******************************************************************************
- *                                                                              *
- *                                                                              *
- *                                                                              *
- *******************************************************************************/
-__host__ cudaError_t CUDARTAPI cudaMemset(void *mem, int c, size_t count)
+  size_t max = prop.maxThreadsPerBlock;
+
+  if (attr->numRegs && (prop.regsPerBlock / attr->numRegs) < max)
+    max = prop.regsPerBlock / attr->numRegs;
+
+  if (attr->sharedSizeBytes && (prop.sharedMemPerBlock / attr->sharedSizeBytes) < max)
+      max = prop.sharedMemPerBlock / attr->sharedSizeBytes;
+
+  return max;
+}
+
+cudaError_t CUDARTAPI cudaFuncGetAttributesInternal(struct cudaFuncAttributes *attr, const char *hostFun, gpgpu_context* gpgpu_ctx = NULL )
 {
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	CUctx_st *context = GPGPUSim_Context();
-	gpgpu_t *gpu = context->get_device()->get_gpgpu();
-	gpu->gpu_memset((size_t)mem, c, count);
+	CUctx_st *context = GPGPUSim_Context(ctx);
+	function_info *entry = context->get_kernel(hostFun);
+	if( entry ) {
+		const struct gpgpu_ptx_sim_info *kinfo = entry->get_kernel_info();
+		attr->sharedSizeBytes = kinfo->smem;
+		attr->constSizeBytes  = kinfo->cmem;
+		attr->localSizeBytes  = kinfo->lmem;
+		attr->numRegs         = kinfo->regs;
+		if(kinfo->maxthreads > 0)
+		  attr->maxThreadsPerBlock = kinfo->maxthreads;
+		else
+		  attr->maxThreadsPerBlock = getMaxThreadsPerBlock(attr, ctx);
+#if CUDART_VERSION >= 3000
+		attr->ptxVersion      = kinfo->ptx_version;
+		attr->binaryVersion   = kinfo->sm_target;
+#endif
+	}
 	return g_last_cudaError = cudaSuccess;
 }
 
@@ -1826,19 +1723,652 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttributeInternal(int *value, enum c
 }
 #endif
 
-//memset operation is done but i think its not async?
-__host__ cudaError_t CUDARTAPI cudaMemsetAsync(void *mem, int c, size_t count, 	cudaStream_t stream=0)
+__host__ cudaError_t CUDARTAPI cudaBindTextureInternal(size_t *offset,
+		const struct textureReference *texref,
+		const void *devPtr,
+		const struct cudaChannelFormatDesc *desc,
+		size_t size __dv(UINT_MAX),
+		gpgpu_context* gpgpu_ctx = NULL)
 {
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
 	if(g_debug_execution >= 3){
 	    announce_call(__my_func__);
     }
-	printf("GPGPU-Sim PTX: WARNING: Asynchronous memset not supported (%s)\n", __my_func__);
-	CUctx_st *context = GPGPUSim_Context();
+	CUctx_st *context = GPGPUSim_Context(ctx);
 	gpgpu_t *gpu = context->get_device()->get_gpgpu();
-	gpu->gpu_memset((size_t)mem, c, count);
+	printf("GPGPU-Sim PTX: in cudaBindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference));
+	struct cudaArray *array;
+	array = (struct cudaArray*) malloc(sizeof(struct cudaArray));
+	array->desc = *desc;
+	array->size = size;
+	array->width = size;
+	array->height = 1;
+	array->dimensions = 1;
+	array->devPtr = (void*)devPtr;
+	array->devPtr32 = (int)(long long)devPtr;
+	offset = 0;
+	printf("GPGPU-Sim PTX:   size = %zu\n", size);
+	printf("GPGPU-Sim PTX:   texref = %p, array = %p\n", texref, array);
+	printf("GPGPU-Sim PTX:   devPtr32 = %x\n", array->devPtr32);
+	printf("GPGPU-Sim PTX:   Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref));
+	printf("GPGPU-Sim PTX:   ChannelFormatDesc: x=%d, y=%d, z=%d, w=%d\n", desc->x, desc->y, desc->z, desc->w);
+	printf("GPGPU-Sim PTX:   Texture Normalized? = %d\n", texref->normalized);
+	gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array);
+	devPtr = (void*)(long long)array->devPtr32;
+	printf("GPGPU-Sim PTX: devPtr = %p\n", devPtr);
+	return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaBindTextureToArrayInternal(const struct textureReference *texref, const struct cudaArray *array, const struct cudaChannelFormatDesc *desc, gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	CUctx_st *context = GPGPUSim_Context(ctx);
+	gpgpu_t *gpu = context->get_device()->get_gpgpu();
+	printf("GPGPU-Sim PTX: in cudaBindTextureToArray: %p %p\n", texref, array);
+	printf("GPGPU-Sim PTX:   devPtr32 = %x\n", array->devPtr32);
+	printf("GPGPU-Sim PTX:   Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref));
+	printf("GPGPU-Sim PTX:   Texture Normalized? = %d\n", texref->normalized);
+	gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array);
+	return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaUnbindTextureInternal(const struct textureReference *texref, gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+   CUctx_st *context = GPGPUSim_Context(ctx);
+   gpgpu_t *gpu = context->get_device()->get_gpgpu();
+   printf("GPGPU-Sim PTX: in cudaUnbindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference));
+   printf("GPGPU-Sim PTX:   Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref));
+
+   gpu->gpgpu_ptx_sim_unbindTexture(texref);
+   return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaLaunchKernelInternal( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL )
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+
+    if(g_debug_execution >= 3){
+        announce_call(__my_func__);
+    }
+    CUctx_st *context = GPGPUSim_Context(ctx);
+    function_info *entry = context->get_kernel(hostFun);
+#if CUDART_VERSION < 10000
+    cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream, ctx);
+#endif
+    for(unsigned i = 0; i < entry->num_args(); i++){
+        std::pair<size_t, unsigned> p = entry->get_param_config(i);
+        cudaSetupArgumentInternal(args[i], p.first, p.second);
+    }
+
+    cudaLaunchInternal(hostFun);
+    return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaStreamCreateInternal(cudaStream_t *stream, gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	printf("GPGPU-Sim PTX: cudaStreamCreate\n");
+#if (CUDART_VERSION >= 3000)
+	*stream = new struct CUstream_st();
+	ctx->the_gpgpusim->g_stream_manager->add_stream(*stream);
+#else
+	*stream = 0;
+	printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__);
+#endif
+	return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaStreamDestroyInternal(cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+#if (CUDART_VERSION >= 3000)
+	//per-stream synchronization required for application using external libraries without explicit synchronization in the code to 
+	//avoid the stream_manager from spinning forever to destroy non-empty streams without making any forward progress. 
+	stream->synchronize();
+	ctx->the_gpgpusim->g_stream_manager->destroy_stream(stream);
+#endif
+	return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaStreamSynchronizeInternal(cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+#if (CUDART_VERSION >= 3000)
+	if( stream == NULL )
+		ctx->synchronize();
+		return g_last_cudaError = cudaSuccess;
+	stream->synchronize();
+#else
+	printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__);
+#endif
+	return g_last_cudaError = cudaSuccess;
+}
+
+void __cudaRegisterTextureInternal(
+		void **fatCubinHandle,
+		const struct textureReference *hostVar,
+		const void **deviceAddress,
+		const char *deviceName,
+		int dim,
+		int norm,
+		int ext,
+		gpgpu_context* gpgpu_ctx = NULL
+) //passes in a newly created textureReference
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	std::string devStr (deviceName);
+	#if (CUDART_VERSION > 4020)
+	if (devStr.size() > 2 && devStr.data()[0] == ':' && devStr.data()[1] == ':')
+		devStr = devStr.replace(0, 2, "");
+	#endif
+	CUctx_st *context = GPGPUSim_Context(ctx);
+	gpgpu_t *gpu = context->get_device()->get_gpgpu();
+	printf("GPGPU-Sim PTX: in __cudaRegisterTexture:\n");
+	gpu->gpgpu_ptx_sim_bindNameToTexture(devStr.data(), hostVar, dim, norm, ext);
+	printf("GPGPU-Sim PTX:   int dim = %d\n", dim);
+	printf("GPGPU-Sim PTX:   int norm = %d\n", norm);
+	printf("GPGPU-Sim PTX:   int ext = %d\n", ext);
+	printf("GPGPU-Sim PTX:   Execution warning: Not finished implementing \"%s\"\n", __my_func__ );
+}
+
+cudaError_t cudaGLUnmapBufferObjectInternal(GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+#ifdef OPENGL_SUPPORT
+	gpgpu_context *ctx;
+	if (gpgpu_ctx){
+	    ctx = gpgpu_ctx;
+	} else {
+	    ctx = GPGPU_Context();
+	}
+	CUctx_st* ctx = GPGPUSim_Context(ctx);
+	glbmap_entry_t *p = ctx->api->g_glbmap;
+	while ( p && p->m_bufferObj != bufferObj )
+		p = p->m_next;
+	if ( p == NULL )
+		return g_last_cudaError = cudaErrorUnknown;
+
+	char *data = (char *) calloc(p->m_size,1);
+	memcpy_from_gpu( data,(size_t)p->m_devPtr,p->m_size );
+	glBufferSubData(GL_ARRAY_BUFFER,0,p->m_size,data);
+	free(data);
+
+	return g_last_cudaError = cudaSuccess;
+#else
+	fflush(stdout);
+	fflush(stderr);
+	printf("GPGPU-Sim PTX: support for OpenGL integration disabled -- exiting\n");
+	fflush(stdout);
+	exit(50);
+#endif
+}
+
+#if CUDART_VERSION >= 3000
+
+__host__ cudaError_t CUDARTAPI cudaFuncSetCacheConfigInternal(const char *func, enum cudaFuncCache  cacheConfig, gpgpu_context* gpgpu_ctx = NULL )
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	CUctx_st *context = GPGPUSim_Context(ctx);
+	context->get_device()->get_gpgpu()->set_cache_config(context->get_kernel(func)->get_name(), (FuncCache)cacheConfig);
+	return g_last_cudaError = cudaSuccess;
+}
+
+#endif
+
+#if CUDART_VERSION >= 4000
+CUresult CUDAAPI cuLaunchKernelInternal(CUfunction f,
+                                unsigned int gridDimX,
+                                unsigned int gridDimY,
+                                unsigned int gridDimZ,
+                                unsigned int blockDimX,
+                                unsigned int blockDimY,
+                                unsigned int blockDimZ,
+                                unsigned int sharedMemBytes,
+                                CUstream hStream,
+                                void **kernelParams,
+                                void **extra,
+				gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+    if (extra!=NULL){
+	    printf("GPGPU-Sim CUDA DRIVER API: ERROR: Currently do not support void** extra.\n");
+        abort();
+    }
+    const char *hostFun = (const char*) f;
+	CUctx_st *context = GPGPUSim_Context(ctx);
+	function_info *entry = context->get_kernel(hostFun);
+    cudaConfigureCallInternal(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream, ctx);
+    for(unsigned i = 0; i < entry->num_args(); i++){
+        std::pair<size_t, unsigned> p = entry->get_param_config(i);
+        cudaSetupArgumentInternal(kernelParams[i], p.first, p.second, ctx);
+    }
+    cudaLaunchInternal(hostFun, ctx);
+	return CUDA_SUCCESS;
+}
+#endif /* CUDART_VERSION >= 4000 */
+
+CUevent_st *get_event(cudaEvent_t event)
+{
+	unsigned event_uid;
+#if CUDART_VERSION >= 3000
+	event_uid = event->get_uid();
+#else
+	event_uid = event;
+#endif
+	event_tracker_t::iterator e = g_timer_events.find(event_uid);
+	if( e == g_timer_events.end() )
+		return NULL;
+	return e->second;
+}
+
+__host__ cudaError_t CUDARTAPI cudaEventRecordInternal(cudaEvent_t event, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	CUevent_st *e = get_event(event);
+	if( !e ) return g_last_cudaError = cudaErrorUnknown;
+	struct CUstream_st *s = (struct CUstream_st *)stream;
+	stream_operation op(e,s);
+	ctx->the_gpgpusim->g_stream_manager->push(op);
+	return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaStreamWaitEventInternal(cudaStream_t stream, cudaEvent_t event, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+   //reference: https://www.cs.cmu.edu/afs/cs/academic/class/15668-s11/www/cuda-doc/html/group__CUDART__STREAM_gfe68d207dc965685d92d3f03d77b0876.html 
+	CUevent_st *e = get_event(event);
+	if( !e ){
+	   printf("GPGPU-Sim API: Warning: cudaEventRecord has not been called on event before calling cudaStreamWaitEvent.\nNothing to be done.\n");
+      return g_last_cudaError = cudaSuccess;
+   }
+   if (!stream){
+     ctx->the_gpgpusim->g_stream_manager->pushCudaStreamWaitEventToAllStreams(e, flags);
+   } else {
+	   struct CUstream_st *s = (struct CUstream_st *)stream;
+	   stream_operation op(s,e,flags);
+	   ctx->the_gpgpusim->g_stream_manager->push(op);
+   }
+	return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaThreadExitInternal(gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	ctx->exit_simulation();
 	return g_last_cudaError = cudaSuccess;
 }
 
+__host__ cudaError_t CUDARTAPI cudaThreadSynchronizeInternal(gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	//Called on host side
+	ctx->synchronize();
+	return g_last_cudaError = cudaSuccess;
+}
+
+cudaError_t CUDARTAPI cudaDeviceSynchronizeInternal(gpgpu_context* gpgpu_ctx = NULL)
+{
+    gpgpu_context *ctx;
+    if (gpgpu_ctx){
+	ctx = gpgpu_ctx;
+    } else {
+	ctx = GPGPU_Context();
+    }
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	//Blocks until the device has completed all preceding requested tasks
+	ctx->synchronize();
+	return g_last_cudaError = cudaSuccess;
+}
+
+/*******************************************************************************
+ *                                                                              *
+ *                                                                              *
+ *                                                                              *
+ *******************************************************************************/
+
+extern "C" {
+
+/*******************************************************************************
+ *                                                                              *
+ *                                                                              *
+ *                                                                              *
+ *******************************************************************************/
+cudaError_t cudaPeekAtLastError(void)
+{
+	return g_last_cudaError;
+}
+
+__host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) 
+{
+    return cudaMallocInternal(devPtr, size);
+}
+
+__host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size)
+{
+    return cudaMallocHostInternal(ptr, size);
+}
+__host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, size_t width, size_t height)
+{
+    return cudaMallocPitchInternal(devPtr, pitch, width, height);
+}
+
+__host__ cudaError_t CUDARTAPI cudaMallocArray(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1))
+{
+    return cudaMallocArrayInternal(array, desc, width, height);
+}
+
+__host__ cudaError_t CUDARTAPI cudaFree(void *devPtr)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	// TODO...  manage g_global_mem space?
+	return g_last_cudaError = cudaSuccess;
+}
+__host__ cudaError_t CUDARTAPI cudaFreeHost(void *ptr)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	free (ptr);  // this will crash the system if called twice
+	return g_last_cudaError = cudaSuccess;
+}
+
+__host__ cudaError_t CUDARTAPI cudaFreeArray(struct cudaArray *array)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	// TODO...  manage g_global_mem space?
+	return g_last_cudaError = cudaSuccess;
+};
+
+
+/*******************************************************************************
+ *                                                                              *
+ *                                                                              *
+ *                                                                              *
+ *******************************************************************************/
+
+__host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind)
+{
+    return cudaMemcpyInternal(dst, src, count, kind);
+}
+
+__host__ cudaError_t CUDARTAPI cudaMemcpyToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind)
+{ 
+    return cudaMemcpyToArrayInternal(dst, wOffset, hOffset, src, count, kind);
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpyFromArray(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpyArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t count, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice))
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpy2D(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind)
+{
+    return cudaMemcpy2DInternal(dst, dpitch, src, spitch, width, height, kind);
+}
+
+__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind)
+{
+    return cudaMemcpy2DToArrayInternal(dst, wOffset, hOffset, src, spitch, width, height, kind);
+}
+
+__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArray(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+__host__ cudaError_t CUDARTAPI cudaMemcpy2DArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t width, size_t height, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice))
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice))
+{
+    return cudaMemcpyToSymbolInternal(symbol, src, count, offset, kind);
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost))
+{
+    return cudaMemcpyFromSymbolInternal(dst, symbol, count, offset, kind);
+}
+
+__host__ cudaError_t CUDARTAPI cudaMemGetInfo (size_t *free, size_t *total){
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	//placeholder; should interact with cudaMalloc and cudaFree?
+	*free = 10000000000;
+	*total = 10000000000;
+
+	return g_last_cudaError = cudaSuccess;
+}
+
+/*******************************************************************************
+ *                                                                              *
+ *                                                                              *
+ *                                                                              *
+ *******************************************************************************/
+
+__host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream)
+{
+    return cudaMemcpyAsyncInternal(dst, src, count, kind, stream);
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpyFromArrayAsync(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpy2DAsync(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArrayAsync(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream)
+{
+	if(g_debug_execution >= 3){
+	    announce_call(__my_func__);
+    }
+	cuda_not_implemented(__my_func__,__LINE__);
+	return g_last_cudaError = cudaErrorUnknown;
+}
+
+#if (CUDART_VERSION >= 8000)
+cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags)
+{
+    return cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal(numBlocks, hostFunc, blockSize, dynamicSMemSize, flags);
+}
+
+#endif
+
+
+
+/*******************************************************************************
+ *                                                                              *
+ *                                                                              *
+ *                                                                              *
+ *******************************************************************************/
+__host__ cudaError_t CUDARTAPI cudaMemset(void *mem, int c, size_t count)
+{
+    return cudaMemsetInternal(mem, c, count);
+}
+
+//memset operation is done but i think its not async?
+__host__ cudaError_t CUDARTAPI cudaMemsetAsync(void *mem, int c, size_t count, 	cudaStream_t stream=0)
+{
+    return cudaMemsetAsyncInternal(mem, c, count, stream=0);
+}
+
 __host__ cudaError_t CUDARTAPI cudaMemset2D(void *mem, size_t pitch, int c, size_t width, size_t height)
 {
 	if(g_debug_execution >= 3){
@@ -1986,61 +2516,18 @@ __host__ cudaError_t CUDARTAPI cudaBindTexture(size_t *offset,
 		const struct cudaChannelFormatDesc *desc,
 		size_t size __dv(UINT_MAX))
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	CUctx_st *context = GPGPUSim_Context();
-	gpgpu_t *gpu = context->get_device()->get_gpgpu();
-	printf("GPGPU-Sim PTX: in cudaBindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference));
-	struct cudaArray *array;
-	array = (struct cudaArray*) malloc(sizeof(struct cudaArray));
-	array->desc = *desc;
-	array->size = size;
-	array->width = size;
-	array->height = 1;
-	array->dimensions = 1;
-	array->devPtr = (void*)devPtr;
-	array->devPtr32 = (int)(long long)devPtr;
-	offset = 0;
-	printf("GPGPU-Sim PTX:   size = %zu\n", size);
-	printf("GPGPU-Sim PTX:   texref = %p, array = %p\n", texref, array);
-	printf("GPGPU-Sim PTX:   devPtr32 = %x\n", array->devPtr32);
-	printf("GPGPU-Sim PTX:   Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref));
-	printf("GPGPU-Sim PTX:   ChannelFormatDesc: x=%d, y=%d, z=%d, w=%d\n", desc->x, desc->y, desc->z, desc->w);
-	printf("GPGPU-Sim PTX:   Texture Normalized? = %d\n", texref->normalized);
-	gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array);
-	devPtr = (void*)(long long)array->devPtr32;
-	printf("GPGPU-Sim PTX: devPtr = %p\n", devPtr);
-	return g_last_cudaError = cudaSuccess;
+    return cudaBindTextureInternal(offset, texref, devPtr, desc, size __dv(UINT_MAX));
 }
 
 
 __host__ cudaError_t CUDARTAPI cudaBindTextureToArray(const struct textureReference *texref, const struct cudaArray *array, const struct cudaChannelFormatDesc *desc)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	CUctx_st *context = GPGPUSim_Context();
-	gpgpu_t *gpu = context->get_device()->get_gpgpu();
-	printf("GPGPU-Sim PTX: in cudaBindTextureToArray: %p %p\n", texref, array);
-	printf("GPGPU-Sim PTX:   devPtr32 = %x\n", array->devPtr32);
-	printf("GPGPU-Sim PTX:   Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref));
-	printf("GPGPU-Sim PTX:   Texture Normalized? = %d\n", texref->normalized);
-	gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array);
-	return g_last_cudaError = cudaSuccess;
+    return cudaBindTextureToArrayInternal(texref, array, desc);
 }
 
-__host__ cudaError_t CUDARTAPI cudaUnbindTexture(const struct textureReference *texref){
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-   CUctx_st *context = GPGPUSim_Context();
-   gpgpu_t *gpu = context->get_device()->get_gpgpu();
-   printf("GPGPU-Sim PTX: in cudaUnbindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference));
-   printf("GPGPU-Sim PTX:   Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref));
-
-   gpu->gpgpu_ptx_sim_unbindTexture(texref);
-   return g_last_cudaError = cudaSuccess;
+__host__ cudaError_t CUDARTAPI cudaUnbindTexture(const struct textureReference *texref)
+{
+    return cudaUnbindTextureInternal(texref);
 }
 
 __host__ cudaError_t CUDARTAPI cudaGetTextureAlignmentOffset(size_t *offset, const struct textureReference *texref)
@@ -2125,24 +2612,9 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun )
     return cudaLaunchInternal( hostFun );
 }
 
-__host__ cudaError_t CUDARTAPI cudaLaunchKernel ( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream )
+__host__ cudaError_t CUDARTAPI cudaLaunchKernel( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream )
 {
-
-    if(g_debug_execution >= 3){
-        announce_call(__my_func__);
-    }
-    CUctx_st *context = GPGPUSim_Context();
-    function_info *entry = context->get_kernel(hostFun);
-#if CUDART_VERSION < 10000
-    cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream);
-#endif
-    for(unsigned i = 0; i < entry->num_args(); i++){
-        std::pair<size_t, unsigned> p = entry->get_param_config(i);
-        cudaSetupArgumentInternal(args[i], p.first, p.second);
-    }
-
-    cudaLaunchInternal(hostFun);
-    return g_last_cudaError = cudaSuccess;
+    return cudaLaunchKernelInternal(hostFun, gridDim, blockDim, args, sharedMem, stream);
 }
 
 
@@ -2154,18 +2626,7 @@ __host__ cudaError_t CUDARTAPI cudaLaunchKernel ( const char* hostFun, dim3 grid
 
 __host__ cudaError_t CUDARTAPI cudaStreamCreate(cudaStream_t *stream)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	printf("GPGPU-Sim PTX: cudaStreamCreate\n");
-#if (CUDART_VERSION >= 3000)
-	*stream = new struct CUstream_st();
-	g_stream_manager()->add_stream(*stream);
-#else
-	*stream = 0;
-	printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__);
-#endif
-	return g_last_cudaError = cudaSuccess;
+    return cudaStreamCreateInternal(stream);
 }
 
 //TODO: introduce priorities
@@ -2192,32 +2653,12 @@ __host__ __device__ cudaError_t CUDARTAPI cudaStreamCreateWithFlags(cudaStream_t
 
 __host__ cudaError_t CUDARTAPI cudaStreamDestroy(cudaStream_t stream)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-#if (CUDART_VERSION >= 3000)
-	//per-stream synchronization required for application using external libraries without explicit synchronization in the code to 
-	//avoid the stream_manager from spinning forever to destroy non-empty streams without making any forward progress. 
-	stream->synchronize();
-	g_stream_manager()->destroy_stream(stream);
-#endif
-	return g_last_cudaError = cudaSuccess;
+    return cudaStreamDestroyInternal(stream);
 }
 
 __host__ cudaError_t CUDARTAPI cudaStreamSynchronize(cudaStream_t stream)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-#if (CUDART_VERSION >= 3000)
-	if( stream == NULL )
-		synchronize();
-		return g_last_cudaError = cudaSuccess;
-	stream->synchronize();
-#else
-	printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__);
-#endif
-	return g_last_cudaError = cudaSuccess;
+    return cudaStreamSynchronizeInternal(stream);
 }
 
 __host__ cudaError_t CUDARTAPI cudaStreamQuery(cudaStream_t stream)
@@ -2256,52 +2697,14 @@ __host__ cudaError_t CUDARTAPI cudaEventCreate(cudaEvent_t *event)
 	return g_last_cudaError = cudaSuccess;
 }
 
-CUevent_st *get_event(cudaEvent_t event)
-{
-	unsigned event_uid;
-#if CUDART_VERSION >= 3000
-	event_uid = event->get_uid();
-#else
-	event_uid = event;
-#endif
-	event_tracker_t::iterator e = g_timer_events.find(event_uid);
-	if( e == g_timer_events.end() )
-		return NULL;
-	return e->second;
-}
-
 __host__ cudaError_t CUDARTAPI cudaEventRecord(cudaEvent_t event, cudaStream_t stream)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	CUevent_st *e = get_event(event);
-	if( !e ) return g_last_cudaError = cudaErrorUnknown;
-	struct CUstream_st *s = (struct CUstream_st *)stream;
-	stream_operation op(e,s);
-	g_stream_manager()->push(op);
-	return g_last_cudaError = cudaSuccess;
+    return cudaEventRecordInternal(event, stream);
 }
 
 __host__ cudaError_t CUDARTAPI cudaStreamWaitEvent(cudaStream_t stream, cudaEvent_t event, unsigned int flags)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-   //reference: https://www.cs.cmu.edu/afs/cs/academic/class/15668-s11/www/cuda-doc/html/group__CUDART__STREAM_gfe68d207dc965685d92d3f03d77b0876.html 
-	CUevent_st *e = get_event(event);
-	if( !e ){
-	   printf("GPGPU-Sim API: Warning: cudaEventRecord has not been called on event before calling cudaStreamWaitEvent.\nNothing to be done.\n");
-      return g_last_cudaError = cudaSuccess;
-   }
-   if (!stream){
-      g_stream_manager()->pushCudaStreamWaitEventToAllStreams(e, flags);
-   } else {
-	   struct CUstream_st *s = (struct CUstream_st *)stream;
-	   stream_operation op(s,e,flags);
-	   g_stream_manager()->push(op);
-   }
-	return g_last_cudaError = cudaSuccess;
+    return cudaStreamWaitEventInternal(stream, event, flags);
 }
 
 __host__ cudaError_t CUDARTAPI cudaEventQuery(cudaEvent_t event)
@@ -2374,22 +2777,13 @@ __host__ cudaError_t CUDARTAPI cudaEventElapsedTime(float *ms, cudaEvent_t start
 
 __host__ cudaError_t CUDARTAPI cudaThreadExit(void)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	exit_simulation();
-	return g_last_cudaError = cudaSuccess;
+    return cudaThreadExitInternal();
 }
 
 __host__ cudaError_t CUDARTAPI cudaThreadSynchronize(void)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	//Called on host side
-	synchronize();
-	return g_last_cudaError = cudaSuccess;
-};
+    return cudaThreadSynchronizeInternal();
+}
 
 int CUDARTAPI __cudaSynchronizeThreads(void**, void*)
 {
@@ -2528,7 +2922,7 @@ void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){
  *	enabled
  * */
 void cuda_runtime_api::extract_code_using_cuobjdump(){
-    CUctx_st *context = GPGPUSim_Context();
+    CUctx_st *context = GPGPUSim_Context(gpgpu_ctx);
 
     //prevent the dumping by cuobjdump everytime we execute the code!
     const char *override_cuobjdump = getenv("CUOBJDUMP_SIM_FILE"); 
@@ -2886,7 +3280,7 @@ cuobjdumpPTXSection* cuda_runtime_api::findPTXSection(const std::string identifi
 
 //! Extract the code using cuobjdump and remove unnecessary sections
 void cuda_runtime_api::cuobjdumpInit(){
-	CUctx_st *context = GPGPUSim_Context();
+	CUctx_st *context = GPGPUSim_Context(gpgpu_ctx);
 	extract_code_using_cuobjdump(); //extract all the output of cuobjdump to _cuobjdump_*.*
 	const char* pre_load = getenv("CUOBJDUMP_SIM_FILE");
 	if (pre_load ==NULL || strlen(pre_load)==0){
@@ -2899,7 +3293,7 @@ void cuda_runtime_api::cuobjdumpInit(){
 //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it
 void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){
 
-	CUctx_st *context = GPGPUSim_Context();
+	CUctx_st *context = GPGPUSim_Context(this);
 	if(api->fatbin_registered[handle]) return;
 	api->fatbin_registered[handle] = true;
 	std::string fname = api->fatbinmap[handle];
@@ -3093,15 +3487,11 @@ cudaError_t cudaDeviceReset ( void ) {
     }
 	return g_last_cudaError = cudaSuccess;
 }
-cudaError_t CUDARTAPI cudaDeviceSynchronize(void){
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	//Blocks until the device has completed all preceding requested tasks
-	synchronize();
-	return g_last_cudaError = cudaSuccess;
-}
 
+cudaError_t CUDARTAPI cudaDeviceSynchronize(void)
+{
+    return cudaDeviceSynchronizeInternal();
+}
 
 void __cudaRegisterShared(
 		void **fatCubinHandle,
@@ -3140,22 +3530,7 @@ void __cudaRegisterTexture(
 		int ext
 ) //passes in a newly created textureReference
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	std::string devStr (deviceName);
-	#if (CUDART_VERSION > 4020)
-	if (devStr.size() > 2 && devStr.data()[0] == ':' && devStr.data()[1] == ':')
-		devStr = devStr.replace(0, 2, "");
-	#endif
-	CUctx_st *context = GPGPUSim_Context();
-	gpgpu_t *gpu = context->get_device()->get_gpgpu();
-	printf("GPGPU-Sim PTX: in __cudaRegisterTexture:\n");
-	gpu->gpgpu_ptx_sim_bindNameToTexture(devStr.data(), hostVar, dim, norm, ext);
-	printf("GPGPU-Sim PTX:   int dim = %d\n", dim);
-	printf("GPGPU-Sim PTX:   int norm = %d\n", norm);
-	printf("GPGPU-Sim PTX:   int ext = %d\n", ext);
-	printf("GPGPU-Sim PTX:   Execution warning: Not finished implementing \"%s\"\n", __my_func__ );
+    __cudaRegisterTextureInternal(fatCubinHandle, hostVar, deviceAddress, deviceName, dim, norm, ext);
 }
 
 
@@ -3187,30 +3562,7 @@ cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj)
 
 cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-#ifdef OPENGL_SUPPORT
-	CUctx_st* ctx = GPGPUSim_Context();
-	glbmap_entry_t *p = ctx->api->g_glbmap;
-	while ( p && p->m_bufferObj != bufferObj )
-		p = p->m_next;
-	if ( p == NULL )
-		return g_last_cudaError = cudaErrorUnknown;
-
-	char *data = (char *) calloc(p->m_size,1);
-	memcpy_from_gpu( data,(size_t)p->m_devPtr,p->m_size );
-	glBufferSubData(GL_ARRAY_BUFFER,0,p->m_size,data);
-	free(data);
-
-	return g_last_cudaError = cudaSuccess;
-#else
-	fflush(stdout);
-	fflush(stderr);
-	printf("GPGPU-Sim PTX: support for OpenGL integration disabled -- exiting\n");
-	fflush(stdout);
-	exit(50);
-#endif
+    return cudaGLUnmapBufferObjectInternal(bufferObj);
 }
 
 cudaError_t cudaGLUnregisterBufferObject(GLuint bufferObj) 
@@ -3322,12 +3674,7 @@ cudaError_t CUDARTAPI cudaRuntimeGetVersion(int *runtimeVersion)
 #if CUDART_VERSION >= 3000
 __host__ cudaError_t CUDARTAPI cudaFuncSetCacheConfig(const char *func, enum cudaFuncCache  cacheConfig )
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-	CUctx_st *context = GPGPUSim_Context();
-	context->get_device()->get_gpgpu()->set_cache_config(context->get_kernel(func)->get_name(), (FuncCache)cacheConfig);
-	return g_last_cudaError = cudaSuccess;
+    return cudaFuncSetCacheConfigInternal(func, cacheConfig);
 }
 
 //Jin: hack for cdp
@@ -5033,23 +5380,7 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f,
                                 void **kernelParams,
                                 void **extra)
 {
-	if(g_debug_execution >= 3){
-	    announce_call(__my_func__);
-    }
-    if (extra!=NULL){
-	    printf("GPGPU-Sim CUDA DRIVER API: ERROR: Currently do not support void** extra.\n");
-        abort();
-    }
-    const char *hostFun = (const char*) f;
-	CUctx_st *context = GPGPUSim_Context();
-	function_info *entry = context->get_kernel(hostFun);
-    cudaConfigureCallInternal(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream);
-    for(unsigned i = 0; i < entry->num_args(); i++){
-        std::pair<size_t, unsigned> p = entry->get_param_config(i);
-        cudaSetupArgument(kernelParams[i], p.first, p.second);
-    }
-    cudaLaunchInternal(hostFun);
-	return CUDA_SUCCESS;
+    return cuLaunchKernelInternal(f, gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY, blockDimZ, sharedMemBytes, hStream, kernelParams, extra);
 }
 #endif /* CUDART_VERSION >= 4000 */
 
diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h
index 45c5cdd..61d7507 100644
--- a/libcuda/gpgpu_context.h
+++ b/libcuda/gpgpu_context.h
@@ -52,6 +52,10 @@ class gpgpu_context {
 	cuda_device_runtime* device_runtime;
 	ptx_stats* stats;
 	// member function list
+	void synchronize();
+	void exit_simulation();
+	void print_simulation_time();
+	int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid );
 	void cuobjdumpParseBinary(unsigned int handle);
 	class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num );
 	class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename );
@@ -60,6 +64,7 @@ class gpgpu_context {
 	void print_ptx_file( const char *p, unsigned source_num, const char *filename );
 	class symbol_table* init_parser(const char*);
 	class gpgpu_sim *gpgpu_ptx_sim_init_perf();
+	void start_sim_thread(int api);
 	struct _cuda_device_id *GPGPUSim_Init();
 	void ptx_reg_options(option_parser_t opp);
 	const ptx_instruction* pc_to_instruction(unsigned pc); 
diff --git a/libopencl/opencl_runtime_api.cc b/libopencl/opencl_runtime_api.cc
index 7f029c7..b032c05 100644
--- a/libopencl/opencl_runtime_api.cc
+++ b/libopencl/opencl_runtime_api.cc
@@ -647,13 +647,13 @@ unsigned _cl_kernel::sm_context_uid = 0;
 class _cl_device_id *GPGPUSim_Init()
 {
    static _cl_device_id *the_device = NULL;
+   gpgpu_context *ctx;
+   ctx = GPGPU_Context();
    if( !the_device ) { 
-       gpgpu_context *ctx;
-       ctx = GPGPU_Context();
       gpgpu_sim *the_gpu = ctx->gpgpu_ptx_sim_init_perf(); 
       the_device = new _cl_device_id(the_gpu);
    } 
-   start_sim_thread(2);
+   ctx->start_sim_thread(2);
    return the_device;
 }
 
@@ -960,7 +960,7 @@ clEnqueueNDRangeKernel(cl_command_queue command_queue,
    if ( ctx->func_sim->g_ptx_sim_mode )
       ctx->func_sim->gpgpu_opencl_ptx_sim_main_func( grid );
    else
-      gpgpu_opencl_ptx_sim_main_perf( grid );
+      ctx->gpgpu_opencl_ptx_sim_main_perf( grid );
    return CL_SUCCESS;
 }
 
diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc
index 9a91818..07232ee 100644
--- a/src/abstract_hardware_model.cc
+++ b/src/abstract_hardware_model.cc
@@ -808,14 +808,14 @@ void kernel_info_t::notify_parent_finished() {
    if(m_parent_kernel) {
        m_kernel_entry->gpgpu_ctx->device_runtime->g_total_param_size -= ((m_kernel_entry->get_args_aligned_size() + 255)/256*256);
        m_parent_kernel->remove_child(this);
-       g_stream_manager()->register_finished_kernel(m_parent_kernel->get_uid());
+       m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->register_finished_kernel(m_parent_kernel->get_uid());
    }
 }
 
 CUstream_st * kernel_info_t::create_stream_cta(dim3 ctaid) {
     assert(get_default_stream_cta(ctaid));
     CUstream_st * stream = new CUstream_st();
-    g_stream_manager()->add_stream(stream);
+    m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->add_stream(stream);
     assert(m_cta_streams.find(ctaid) != m_cta_streams.end());
     assert(m_cta_streams[ctaid].size() >= 1); //must have default stream
     m_cta_streams[ctaid].push_back(stream);
@@ -831,7 +831,7 @@ CUstream_st * kernel_info_t::get_default_stream_cta(dim3 ctaid) {
     else {
       m_cta_streams[ctaid] = std::list<CUstream_st *>();
       CUstream_st * stream = new CUstream_st();
-      g_stream_manager()->add_stream(stream);
+      m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->add_stream(stream);
       m_cta_streams[ctaid].push_back(stream);
       return stream;
     }
@@ -863,7 +863,7 @@ void kernel_info_t::destroy_cta_streams() {
      for(auto s = m_cta_streams.begin(); s != m_cta_streams.end(); s++) {
         stream_size += s->second.size();
         for(auto ss = s->second.begin(); ss != s->second.end(); ss++)
-        g_stream_manager()->destroy_stream(*ss);
+        m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->destroy_stream(*ss);
         s->second.clear();
      }
      printf("size %lu\n", stream_size);
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 28b4bf9..7a130ea 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -429,7 +429,7 @@ void gpgpu_t::memcpy_to_gpu( size_t dst_start_addr, const void *src, size_t coun
 
    // Copy into the performance model.
    //extern gpgpu_sim* g_the_gpu;
-   g_the_gpu()->perf_memcpy_to_gpu(dst_start_addr, count);
+   gpgpu_ctx->the_gpgpusim->g_the_gpu->perf_memcpy_to_gpu(dst_start_addr, count);
    if(g_debug_execution >= 3) {
       printf( " done.\n");
       fflush(stdout);
@@ -448,7 +448,7 @@ void gpgpu_t::memcpy_from_gpu( void *dst, size_t src_start_addr, size_t count )
 
    // Copy into the performance model.
    //extern gpgpu_sim* g_the_gpu;
-   g_the_gpu()->perf_memcpy_to_gpu(src_start_addr, count);
+   gpgpu_ctx->the_gpgpusim->g_the_gpu->perf_memcpy_to_gpu(src_start_addr, count);
    if(g_debug_execution >= 3) {
       printf( " done.\n");
       fflush(stdout);
@@ -1254,7 +1254,7 @@ void function_info::param_to_shared( memory_space *shared_mem, symbol_table *sym
 {
    // TODO: call this only for PTXPlus with GT200 models 
    //extern gpgpu_sim* g_the_gpu;
-   if (not g_the_gpu()->get_config().convert_to_ptxplus()) return;
+   if (not gpgpu_ctx->the_gpgpusim->g_the_gpu->get_config().convert_to_ptxplus()) return;
 
    // copies parameters into simulated shared memory
    for( std::map<unsigned,param_info>::iterator i=m_ptx_kernel_param_info.begin(); i!=m_ptx_kernel_param_info.end(); i++ ) {
@@ -2137,13 +2137,13 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
 	 kernel_func_info->set_pdom();
     }
 
-    unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), g_the_gpu()->getShaderCoreConfig()->warp_size, g_the_gpu()->getShaderCoreConfig()->n_thread_per_shader, g_the_gpu()->getShaderCoreConfig()->gpgpu_shmem_size, g_the_gpu()->getShaderCoreConfig()->gpgpu_shader_registers, g_the_gpu()->getShaderCoreConfig()->max_cta_per_core);
+    unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->warp_size, gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->n_thread_per_shader, gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->gpgpu_shmem_size, gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->gpgpu_shader_registers, gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->max_cta_per_core);
     printf("Max CTA : %d\n",max_cta_tot);
 
-    int cp_op= g_the_gpu()->checkpoint_option;
-    int cp_kernel= g_the_gpu()->checkpoint_kernel;
-    cp_count= g_the_gpu()->checkpoint_insn_Y;
-    cp_cta_resume= g_the_gpu()->checkpoint_CTA_t;
+    int cp_op= gpgpu_ctx->the_gpgpusim->g_the_gpu->checkpoint_option;
+    int cp_kernel= gpgpu_ctx->the_gpgpusim->g_the_gpu->checkpoint_kernel;
+    cp_count= gpgpu_ctx->the_gpgpusim->g_the_gpu->checkpoint_insn_Y;
+    cp_cta_resume= gpgpu_ctx->the_gpgpusim->g_the_gpu->checkpoint_CTA_t;
     int cta_launched =0;
 
     //we excute the kernel one CTA (Block) at the time, as synchronization functions work block wise
@@ -2155,8 +2155,8 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
         {
            functionalCoreSim cta(
                &kernel,
-               g_the_gpu(),
-               g_the_gpu()->getShaderCoreConfig()->warp_size
+               gpgpu_ctx->the_gpgpusim->g_the_gpu,
+               gpgpu_ctx->the_gpgpusim->g_the_gpu->getShaderCoreConfig()->warp_size
            );
            cta.execute(cp_count,temp);
 
@@ -2177,7 +2177,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
 	{
       char f1name[2048];
       snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", kernel.get_uid() );
-      g_checkpoint->store_global_mem(g_the_gpu()->get_global_memory(), f1name , (char *)"%08x");
+      g_checkpoint->store_global_mem(gpgpu_ctx->the_gpgpusim->g_the_gpu->get_global_memory(), f1name , (char *)"%08x");
 	}
 
 
@@ -2188,7 +2188,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
    //openCL kernel simulation calls don't register the kernel so we don't register its exit
    if(!openCL) {
       //extern stream_manager *g_stream_manager;
-      g_stream_manager()->register_finished_kernel(kernel.get_uid());
+      gpgpu_ctx->the_gpgpusim->g_stream_manager->register_finished_kernel(kernel.get_uid());
    }
 
    //******PRINTING*******
@@ -2203,7 +2203,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL
    //g_simulation_starttime is initilized by gpgpu_ptx_sim_init_perf() in gpgpusim_entrypoint.cc upon starting gpgpu-sim
    time_t end_time, elapsed_time, days, hrs, minutes, sec;
    end_time = time((time_t *)NULL);
-   elapsed_time = MAX(end_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1);
+   elapsed_time = MAX(end_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1);
 	
 
    //calculating and printing simulation time in terms of days, hours, minutes and seconds
diff --git a/src/cuda-sim/cuda_device_runtime.cc b/src/cuda-sim/cuda_device_runtime.cc
index dc3adc3..4baced5 100644
--- a/src/cuda-sim/cuda_device_runtime.cc
+++ b/src/cuda-sim/cuda_device_runtime.cc
@@ -27,7 +27,6 @@
    }
 
 
-//extern stream_manager *g_stream_manager();
 
 //Handling device runtime api:
 //void * cudaGetParameterBufferV2(void *func, dim3 gridDimension, dim3 blockDimension, unsigned int sharedMemSize)
@@ -285,7 +284,7 @@ void cuda_device_runtime::launch_one_device_kernel() {
         device_launch_operation_t &op = g_cuda_device_launch_op.front();
 
         stream_operation stream_op = stream_operation(op.grid, gpgpu_ctx->func_sim->g_ptx_sim_mode, op.stream);
-        g_stream_manager()->push(stream_op);
+        gpgpu_ctx->the_gpgpusim->g_stream_manager->push(stream_op);
         g_cuda_device_launch_op.pop_front();
     }
 }
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index d236c74..56ea8c4 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -1144,7 +1144,7 @@ void gpgpu_sim::gpu_print_stat()
 
    time_t curr_time;
    time(&curr_time);
-   unsigned long long elapsed_time = MAX( curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1 );
+   unsigned long long elapsed_time = MAX( curr_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1 );
    printf( "gpu_total_sim_rate=%u\n", (unsigned)( ( gpu_tot_sim_insn + gpu_sim_insn ) / elapsed_time ) );
 
    //shader_print_l1_miss_stat( stdout );
@@ -1717,7 +1717,7 @@ void gpgpu_sim::cycle()
          time_t days, hrs, minutes, sec;
          time_t curr_time;
          time(&curr_time);
-         unsigned long long  elapsed_time = MAX(curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1);
+         unsigned long long  elapsed_time = MAX(curr_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1);
          if ( (elapsed_time - last_liveness_message_time) >= m_config.liveness_message_freq && DTRACE(LIVENESS) ) {
             days    = elapsed_time/(3600*24);
             hrs     = elapsed_time/3600 - 24*days;
diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc
index d3deb24..846773d 100644
--- a/src/gpgpusim_entrypoint.cc
+++ b/src/gpgpusim_entrypoint.cc
@@ -43,46 +43,28 @@ static int sg_argc = 3;
 static const char *sg_argv[] = {"", "-config","gpgpusim.config"};
 
 
-GPGPUsim_ctx* the_gpgpusim =  NULL;
-
-GPGPUsim_ctx* GPGPUsim_ctx_ptr(){
-	if(the_gpgpusim == NULL)
-		the_gpgpusim = GPGPU_Context()->the_gpgpusim;
-
-	return the_gpgpusim;
-}
-
-class gpgpu_sim* g_the_gpu() {
-	return GPGPUsim_ctx_ptr()->g_the_gpu;
-}
-
-class stream_manager* g_stream_manager()  {
-	return GPGPUsim_ctx_ptr()->g_stream_manager;
-}
-
-static void print_simulation_time();
-
-void *gpgpu_sim_thread_sequential(void*)
+void * gpgpu_sim_thread_sequential(void * ctx_ptr)
 {
+    gpgpu_context * ctx = (gpgpu_context *)ctx_ptr;
    // at most one kernel running at a time
    bool done;
    do {
-      sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_start));
+      sem_wait(&(ctx->the_gpgpusim->g_sim_signal_start));
       done = true;
-      if( GPGPUsim_ctx_ptr()->g_the_gpu->get_more_cta_left() ) {
+      if( ctx->the_gpgpusim->g_the_gpu->get_more_cta_left() ) {
           done = false;
-          GPGPUsim_ctx_ptr()->g_the_gpu->init();
-          while( GPGPUsim_ctx_ptr()->g_the_gpu->active() ) {
-        	  GPGPUsim_ctx_ptr()->g_the_gpu->cycle();
-        	  GPGPUsim_ctx_ptr()->g_the_gpu->deadlock_check();
+          ctx->the_gpgpusim->g_the_gpu->init();
+          while( ctx->the_gpgpusim->g_the_gpu->active() ) {
+        	  ctx->the_gpgpusim->g_the_gpu->cycle();
+        	  ctx->the_gpgpusim->g_the_gpu->deadlock_check();
           }
-          GPGPUsim_ctx_ptr()->g_the_gpu->print_stats();
-          GPGPUsim_ctx_ptr()->g_the_gpu->update_stats();
-          print_simulation_time();
+          ctx->the_gpgpusim->g_the_gpu->print_stats();
+          ctx->the_gpgpusim->g_the_gpu->update_stats();
+          ctx->print_simulation_time();
       }
-      sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish));
+      sem_post(&(ctx->the_gpgpusim->g_sim_signal_finish));
    } while(!done);
-   sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit));
+   sem_post(&(ctx->the_gpgpusim->g_sim_signal_exit));
    return NULL;
 }
 
@@ -94,8 +76,9 @@ static void termination_callback()
     fflush(stdout);
 }
 
-void *gpgpu_sim_thread_concurrent(void*)
+void *gpgpu_sim_thread_concurrent(void * ctx_ptr)
 {
+    gpgpu_context * ctx = (gpgpu_context *)ctx_ptr;
     atexit(termination_callback);
     // concurrent kernel execution simulation thread
     do {
@@ -103,19 +86,19 @@ void *gpgpu_sim_thread_concurrent(void*)
            printf("GPGPU-Sim: *** simulation thread starting and spinning waiting for work ***\n");
            fflush(stdout);
         }
-        while( GPGPUsim_ctx_ptr()->g_stream_manager->empty_protected() && !GPGPUsim_ctx_ptr()->g_sim_done )
+        while( ctx->the_gpgpusim->g_stream_manager->empty_protected() && !ctx->the_gpgpusim->g_sim_done )
             ;
         if(g_debug_execution >= 3) {
            printf("GPGPU-Sim: ** START simulation thread (detected work) **\n");
-           GPGPUsim_ctx_ptr()->g_stream_manager->print(stdout);
+           ctx->the_gpgpusim->g_stream_manager->print(stdout);
            fflush(stdout);
         }
-        pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
-        GPGPUsim_ctx_ptr()->g_sim_active = true;
-        pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
+        pthread_mutex_lock(&(ctx->the_gpgpusim->g_sim_lock));
+        ctx->the_gpgpusim->g_sim_active = true;
+        pthread_mutex_unlock(&(ctx->the_gpgpusim->g_sim_lock));
         bool active = false;
         bool sim_cycles = false;
-        GPGPUsim_ctx_ptr()->g_the_gpu->init();
+        ctx->the_gpgpusim->g_the_gpu->init();
         do {
             // check if a kernel has completed
             // launch operation on device if one is pending and can be run
@@ -127,82 +110,82 @@ void *gpgpu_sim_thread_concurrent(void*)
             // another kernel, the gpu is not re-initialized and the inter-kernel
             // behaviour may be incorrect. Check that a kernel has finished and
             // no other kernel is currently running.
-            if(GPGPUsim_ctx_ptr()->g_stream_manager->operation(&sim_cycles) && !GPGPUsim_ctx_ptr()->g_the_gpu->active())
+            if(ctx->the_gpgpusim->g_stream_manager->operation(&sim_cycles) && !ctx->the_gpgpusim->g_the_gpu->active())
                 break;
 
             //functional simulation
-            if( GPGPUsim_ctx_ptr()->g_the_gpu->is_functional_sim()) {
-                kernel_info_t * kernel = GPGPUsim_ctx_ptr()->g_the_gpu->get_functional_kernel();
+            if( ctx->the_gpgpusim->g_the_gpu->is_functional_sim()) {
+                kernel_info_t * kernel = ctx->the_gpgpusim->g_the_gpu->get_functional_kernel();
                 assert(kernel);
-                GPGPUsim_ctx_ptr()->gpgpu_ctx->func_sim->gpgpu_cuda_ptx_sim_main_func(*kernel);
-                GPGPUsim_ctx_ptr()->g_the_gpu->finish_functional_sim(kernel);
+                ctx->the_gpgpusim->gpgpu_ctx->func_sim->gpgpu_cuda_ptx_sim_main_func(*kernel);
+                ctx->the_gpgpusim->g_the_gpu->finish_functional_sim(kernel);
             }
 
             //performance simulation
-            if( GPGPUsim_ctx_ptr()->g_the_gpu->active() ) {
-            	GPGPUsim_ctx_ptr()->g_the_gpu->cycle();
+            if( ctx->the_gpgpusim->g_the_gpu->active() ) {
+            	ctx->the_gpgpusim->g_the_gpu->cycle();
             	sim_cycles = true;
-            	GPGPUsim_ctx_ptr()->g_the_gpu->deadlock_check();
+            	ctx->the_gpgpusim->g_the_gpu->deadlock_check();
             }else {
-                if(GPGPUsim_ctx_ptr()->g_the_gpu->cycle_insn_cta_max_hit()){
-                	GPGPUsim_ctx_ptr()->g_stream_manager->stop_all_running_kernels();
-                	GPGPUsim_ctx_ptr()->g_sim_done = true;
-                	GPGPUsim_ctx_ptr()->break_limit = true;
+                if(ctx->the_gpgpusim->g_the_gpu->cycle_insn_cta_max_hit()){
+                	ctx->the_gpgpusim->g_stream_manager->stop_all_running_kernels();
+                	ctx->the_gpgpusim->g_sim_done = true;
+                	ctx->the_gpgpusim->break_limit = true;
                 }
             }
 
-            active=GPGPUsim_ctx_ptr()->g_the_gpu->active() || !(GPGPUsim_ctx_ptr()->g_stream_manager->empty_protected());
+            active=ctx->the_gpgpusim->g_the_gpu->active() || !(ctx->the_gpgpusim->g_stream_manager->empty_protected());
 
-        } while( active && !GPGPUsim_ctx_ptr()->g_sim_done);
+        } while( active && !ctx->the_gpgpusim->g_sim_done);
         if(g_debug_execution >= 3) {
            printf("GPGPU-Sim: ** STOP simulation thread (no work) **\n");
            fflush(stdout);
         }
         if(sim_cycles) {
-        	GPGPUsim_ctx_ptr()->g_the_gpu->print_stats();
-        	GPGPUsim_ctx_ptr()->g_the_gpu->update_stats();
-            print_simulation_time();
+        	ctx->the_gpgpusim->g_the_gpu->print_stats();
+        	ctx->the_gpgpusim->g_the_gpu->update_stats();
+            ctx->print_simulation_time();
         }
-        pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
-        GPGPUsim_ctx_ptr()->g_sim_active = false;
-        pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
-    } while( !GPGPUsim_ctx_ptr()->g_sim_done );
+        pthread_mutex_lock(&(ctx->the_gpgpusim->g_sim_lock));
+        ctx->the_gpgpusim->g_sim_active = false;
+        pthread_mutex_unlock(&(ctx->the_gpgpusim->g_sim_lock));
+    } while( !ctx->the_gpgpusim->g_sim_done );
 
     printf("GPGPU-Sim: *** simulation thread exiting ***\n");
     fflush(stdout);
 
-    if(GPGPUsim_ctx_ptr()->break_limit) {
+    if(ctx->the_gpgpusim->break_limit) {
     	printf("GPGPU-Sim: ** break due to reaching the maximum cycles (or instructions) **\n");
     	exit(1);
     }
 
-    sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit));
+    sem_post(&(ctx->the_gpgpusim->g_sim_signal_exit));
     return NULL;
 }
 
-void synchronize()
+void gpgpu_context::synchronize()
 {
     printf("GPGPU-Sim: synchronize waiting for inactive GPU simulation\n");
-    GPGPUsim_ctx_ptr()->g_stream_manager->print(stdout);
+    the_gpgpusim->g_stream_manager->print(stdout);
     fflush(stdout);
 //    sem_wait(&g_sim_signal_finish);
     bool done = false;
     do {
-        pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
-        done = ( GPGPUsim_ctx_ptr()->g_stream_manager->empty() && !GPGPUsim_ctx_ptr()->g_sim_active ) || GPGPUsim_ctx_ptr()->g_sim_done;
-        pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock));
+        pthread_mutex_lock(&(the_gpgpusim->g_sim_lock));
+        done = ( the_gpgpusim->g_stream_manager->empty() && !the_gpgpusim->g_sim_active ) || the_gpgpusim->g_sim_done;
+        pthread_mutex_unlock(&(the_gpgpusim->g_sim_lock));
     } while (!done);
     printf("GPGPU-Sim: detected inactive GPU simulation thread\n");
     fflush(stdout);
 //    sem_post(&g_sim_signal_start);
 }
 
-void exit_simulation()
+void gpgpu_context::exit_simulation()
 {
-	GPGPUsim_ctx_ptr()->g_sim_done=true;
+	the_gpgpusim->g_sim_done=true;
     printf("GPGPU-Sim: exit_simulation called\n");
     fflush(stdout);
-    sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit));
+    sem_wait(&(the_gpgpusim->g_sim_signal_exit));
     printf("GPGPU-Sim: simulation thread signaled exit\n");
     fflush(stdout);
 }
@@ -219,8 +202,8 @@ gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf()
    func_sim->ptx_opcocde_latency_options(opp);
 
    icnt_reg_options(opp);
-   GPGPUsim_ctx_ptr()->g_the_gpu_config = new gpgpu_sim_config(this);
-   GPGPUsim_ctx_ptr()->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options
+   the_gpgpusim->g_the_gpu_config = new gpgpu_sim_config(this);
+   the_gpgpusim->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options
 
    option_parser_cmdline(opp, sg_argc, sg_argv); // parse configuration options
    fprintf(stdout, "GPGPU-Sim: Configuration options:\n\n");
@@ -228,37 +211,37 @@ gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf()
    // Set the Numeric locale to a standard locale where a decimal point is a "dot" not a "comma"
    // so it does the parsing correctly independent of the system environment variables
    assert(setlocale(LC_NUMERIC,"C"));
-   GPGPUsim_ctx_ptr()->g_the_gpu_config->init();
+   the_gpgpusim->g_the_gpu_config->init();
 
-   GPGPUsim_ctx_ptr()->g_the_gpu = new gpgpu_sim(*(GPGPUsim_ctx_ptr()->g_the_gpu_config), this);
-   GPGPUsim_ctx_ptr()->g_stream_manager = new stream_manager((GPGPUsim_ctx_ptr()->g_the_gpu), func_sim->g_cuda_launch_blocking);
+   the_gpgpusim->g_the_gpu = new gpgpu_sim(*(the_gpgpusim->g_the_gpu_config), this);
+   the_gpgpusim->g_stream_manager = new stream_manager((the_gpgpusim->g_the_gpu), func_sim->g_cuda_launch_blocking);
 
-   GPGPUsim_ctx_ptr()->g_simulation_starttime = time((time_t *)NULL);
+   the_gpgpusim->g_simulation_starttime = time((time_t *)NULL);
 
-   sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_start),0,0);
-   sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish),0,0);
-   sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit),0,0);
+   sem_init(&(the_gpgpusim->g_sim_signal_start),0,0);
+   sem_init(&(the_gpgpusim->g_sim_signal_finish),0,0);
+   sem_init(&(the_gpgpusim->g_sim_signal_exit),0,0);
 
-   return GPGPUsim_ctx_ptr()->g_the_gpu;
+   return the_gpgpusim->g_the_gpu;
 }
 
-void start_sim_thread(int api)
+void gpgpu_context::start_sim_thread(int api)
 {
-    if( GPGPUsim_ctx_ptr()->g_sim_done ) {
-    	GPGPUsim_ctx_ptr()->g_sim_done = false;
+    if( the_gpgpusim->g_sim_done ) {
+    	the_gpgpusim->g_sim_done = false;
         if( api == 1 ) {
-           pthread_create(&(GPGPUsim_ctx_ptr()->g_simulation_thread),NULL,gpgpu_sim_thread_concurrent,NULL);
+           pthread_create(&(the_gpgpusim->g_simulation_thread),NULL,gpgpu_sim_thread_concurrent,(void *)this);
         } else {
-           pthread_create(&(GPGPUsim_ctx_ptr()->g_simulation_thread),NULL,gpgpu_sim_thread_sequential,NULL);
+           pthread_create(&(the_gpgpusim->g_simulation_thread),NULL,gpgpu_sim_thread_sequential,(void *)this);
         }
     }
 }
 
-void print_simulation_time()
+void gpgpu_context::print_simulation_time()
 {
    time_t current_time, difference, d, h, m, s;
    current_time = time((time_t *)NULL);
-   difference = MAX(current_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1);
+   difference = MAX(current_time - the_gpgpusim->g_simulation_starttime, 1);
 
    d = difference/(3600*24);
    h = difference/3600 - 24*d;
@@ -268,18 +251,18 @@ void print_simulation_time()
    fflush(stderr);
    printf("\n\ngpgpu_simulation_time = %u days, %u hrs, %u min, %u sec (%u sec)\n",
           (unsigned)d, (unsigned)h, (unsigned)m, (unsigned)s, (unsigned)difference );
-   printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(GPGPUsim_ctx_ptr()->g_the_gpu->gpu_tot_sim_insn / difference) );
-   const unsigned cycles_per_sec = (unsigned)(GPGPUsim_ctx_ptr()->g_the_gpu->gpu_tot_sim_cycle / difference);
+   printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(the_gpgpusim->g_the_gpu->gpu_tot_sim_insn / difference) );
+   const unsigned cycles_per_sec = (unsigned)(the_gpgpusim->g_the_gpu->gpu_tot_sim_cycle / difference);
    printf("gpgpu_simulation_rate = %u (cycle/sec)\n", cycles_per_sec );
-   printf("gpgpu_silicon_slowdown = %ux\n", GPGPUsim_ctx_ptr()->g_the_gpu->shader_clock() * 1000 / cycles_per_sec);
+   printf("gpgpu_silicon_slowdown = %ux\n", the_gpgpusim->g_the_gpu->shader_clock() * 1000 / cycles_per_sec);
    fflush(stdout);
 }
 
-int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid )
+int gpgpu_context::gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid )
 {
-   GPGPUsim_ctx_ptr()->g_the_gpu->launch(grid);
-   sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_start));
-   sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish));
+   the_gpgpusim->g_the_gpu->launch(grid);
+   sem_post(&(the_gpgpusim->g_sim_signal_start));
+   sem_wait(&(the_gpgpusim->g_sim_signal_finish));
    return 0;
 }
 
diff --git a/src/gpgpusim_entrypoint.h b/src/gpgpusim_entrypoint.h
index 887b3c8..9f408df 100644
--- a/src/gpgpusim_entrypoint.h
+++ b/src/gpgpusim_entrypoint.h
@@ -76,12 +76,4 @@ class GPGPUsim_ctx {
 
 };
 
-void start_sim_thread(int api);
-
-class gpgpu_sim* g_the_gpu();
-struct GPGPUsim_ctx* GPGPUsim_ctx_ptr();
-class stream_manager* g_stream_manager();
-
-int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid );
-
 #endif