Merge pull request #13 from echoedit/dev

Move some struct upward in cuda_runtime_api.cc

3 files changed, 143 insertions, 148 deletions

diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h
index 41337c6..73c077e 100644
--- a/libcuda/cuda_api_object.h
+++ b/libcuda/cuda_api_object.h
@@ -6,13 +6,10 @@
 #include <set>
 #include <string>
 
-class cuobjdumpSection;
-class cuobjdumpELFSection;
-class cuobjdumpPTXSection;
-class symbol_table;
-class gpgpu_ptx_sim_arg;
-class kernel_config;
-class kernel_info_t;
+#include "../src/gpgpu-sim/gpu-sim.h"
+#include "../src/cuda-sim/ptx_ir.h"
+#include "../src/abstract_hardware_model.h"
+#include "cuobjdump.h"
 
 typedef std::list<gpgpu_ptx_sim_arg> gpgpu_ptx_sim_arg_list_t;
 
@@ -29,6 +26,144 @@ struct glbmap_entry {
 
 typedef struct glbmap_entry glbmap_entry_t;
 
+struct _cuda_device_id {
+	_cuda_device_id(gpgpu_sim* gpu) {m_id = 0; m_next = NULL; m_gpgpu=gpu;}
+	struct _cuda_device_id *next() { return m_next; }
+	unsigned num_shader() const { return m_gpgpu->get_config().num_shader(); }
+	int num_devices() const {
+		if( m_next == NULL ) return 1;
+		else return 1 + m_next->num_devices();
+	}
+	struct _cuda_device_id *get_device( unsigned n )
+	{
+		assert( n < (unsigned)num_devices() );
+		struct _cuda_device_id *p=this;
+		for(unsigned i=0; i<n; i++)
+			p = p->m_next;
+		return p;
+	}
+	const struct cudaDeviceProp *get_prop() const
+	{
+		return m_gpgpu->get_prop();
+	}
+	unsigned get_id() const { return m_id; }
+
+	gpgpu_sim *get_gpgpu() { return m_gpgpu; }
+private:
+	unsigned m_id;
+	class gpgpu_sim *m_gpgpu;
+	struct _cuda_device_id *m_next;
+};
+
+struct CUctx_st {
+	CUctx_st( _cuda_device_id *gpu )
+	{
+		m_gpu = gpu;
+		m_binary_info.cmem = 0;
+		m_binary_info.gmem = 0;
+		no_of_ptx=0;
+	}
+
+	_cuda_device_id *get_device() { return m_gpu; }
+
+	void add_binary( symbol_table *symtab, unsigned fat_cubin_handle )
+	{
+		m_code[fat_cubin_handle] = symtab;
+		m_last_fat_cubin_handle = fat_cubin_handle;
+	}
+
+	void add_ptxinfo( const char *deviceFun, const struct gpgpu_ptx_sim_info &info )
+	{
+		symbol *s = m_code[m_last_fat_cubin_handle]->lookup(deviceFun);
+		assert( s != NULL );
+		function_info *f = s->get_pc();
+		assert( f != NULL );
+		f->set_kernel_info(info);
+	}
+
+	void add_ptxinfo( const struct gpgpu_ptx_sim_info &info )
+	{
+		m_binary_info = info;
+	}
+
+	void register_function( unsigned fat_cubin_handle, const char *hostFun, const char *deviceFun )
+	{
+		if( m_code.find(fat_cubin_handle) != m_code.end() ) {
+			symbol *s = m_code[fat_cubin_handle]->lookup(deviceFun);
+			if(s != NULL) {
+				function_info *f = s->get_pc();
+				assert( f != NULL );
+				m_kernel_lookup[hostFun] = f;
+			}
+			else {
+				printf("Warning: cannot find deviceFun %s\n", deviceFun);
+				m_kernel_lookup[hostFun] = NULL;
+			}
+	//		assert( s != NULL );
+	//		function_info *f = s->get_pc();
+	//		assert( f != NULL );
+	//		m_kernel_lookup[hostFun] = f;
+		} else {
+			m_kernel_lookup[hostFun] = NULL;
+		}
+	}
+
+    void register_hostFun_function( const char*hostFun, function_info* f){
+        m_kernel_lookup[hostFun] = f;
+    }
+
+	function_info *get_kernel(const char *hostFun)
+	{
+		std::map<const void*,function_info*>::iterator i=m_kernel_lookup.find(hostFun);
+		assert( i != m_kernel_lookup.end() );
+		return i->second;
+	}
+
+	int no_of_ptx;
+
+private:
+	_cuda_device_id *m_gpu; // selected gpu
+	std::map<unsigned,symbol_table*> m_code; // fat binary handle => global symbol table
+	unsigned m_last_fat_cubin_handle;
+	std::map<const void*,function_info*> m_kernel_lookup; // unique id (CUDA app function address) => kernel entry point
+	struct gpgpu_ptx_sim_info m_binary_info;
+
+};
+
+class kernel_config {
+public:
+	kernel_config( dim3 GridDim, dim3 BlockDim, size_t sharedMem, struct CUstream_st *stream )
+	{
+		m_GridDim=GridDim;
+		m_BlockDim=BlockDim;
+		m_sharedMem=sharedMem;
+		m_stream = stream;
+	}
+	kernel_config()
+	{
+		m_GridDim=dim3(-1,-1,-1);
+		m_BlockDim=dim3(-1,-1,-1);
+		m_sharedMem=0;
+		m_stream =NULL;
+	}
+	void set_arg( const void *arg, size_t size, size_t offset )
+	{
+		m_args.push_front( gpgpu_ptx_sim_arg(arg,size,offset) );
+	}
+	dim3 grid_dim() const { return m_GridDim; }
+	dim3 block_dim() const { return m_BlockDim; }
+	void set_grid_dim(dim3 *d) { m_GridDim = *d; }
+	void set_block_dim(dim3 *d) { m_BlockDim = *d; }
+	gpgpu_ptx_sim_arg_list_t get_args() { return m_args; }
+	struct CUstream_st *get_stream() { return m_stream; }
+
+private:
+	dim3 m_GridDim;
+	dim3 m_BlockDim;
+	size_t m_sharedMem;
+	struct CUstream_st *m_stream;
+	gpgpu_ptx_sim_arg_list_t m_args;
+};
 
 class cuda_runtime_api {
     public:
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc
index 71340b3..2d9c4f7 100644
--- a/libcuda/cuda_runtime_api.cc
+++ b/libcuda/cuda_runtime_api.cc
@@ -141,8 +141,6 @@
 #include "../src/gpgpusim_entrypoint.h"
 #include "../src/stream_manager.h"
 #include "../src/abstract_hardware_model.h"
-typedef void * yyscan_t;
-#include "cuobjdump.h"
 
 #include <pthread.h>
 #include <semaphore.h>
@@ -208,145 +206,6 @@ void register_ptx_function( const char *name, function_info *impl )
 # endif
 #endif
 
-struct _cuda_device_id {
-	_cuda_device_id(gpgpu_sim* gpu) {m_id = 0; m_next = NULL; m_gpgpu=gpu;}
-	struct _cuda_device_id *next() { return m_next; }
-	unsigned num_shader() const { return m_gpgpu->get_config().num_shader(); }
-	int num_devices() const {
-		if( m_next == NULL ) return 1;
-		else return 1 + m_next->num_devices();
-	}
-	struct _cuda_device_id *get_device( unsigned n )
-	{
-		assert( n < (unsigned)num_devices() );
-		struct _cuda_device_id *p=this;
-		for(unsigned i=0; i<n; i++)
-			p = p->m_next;
-		return p;
-	}
-	const struct cudaDeviceProp *get_prop() const
-	{
-		return m_gpgpu->get_prop();
-	}
-	unsigned get_id() const { return m_id; }
-
-	gpgpu_sim *get_gpgpu() { return m_gpgpu; }
-private:
-	unsigned m_id;
-	class gpgpu_sim *m_gpgpu;
-	struct _cuda_device_id *m_next;
-};
-
-struct CUctx_st {
-	CUctx_st( _cuda_device_id *gpu )
-	{
-		m_gpu = gpu;
-		m_binary_info.cmem = 0;
-		m_binary_info.gmem = 0;
-		no_of_ptx=0;
-	}
-
-	_cuda_device_id *get_device() { return m_gpu; }
-
-	void add_binary( symbol_table *symtab, unsigned fat_cubin_handle )
-	{
-		m_code[fat_cubin_handle] = symtab;
-		m_last_fat_cubin_handle = fat_cubin_handle;
-	}
-
-	void add_ptxinfo( const char *deviceFun, const struct gpgpu_ptx_sim_info &info )
-	{
-		symbol *s = m_code[m_last_fat_cubin_handle]->lookup(deviceFun);
-		assert( s != NULL );
-		function_info *f = s->get_pc();
-		assert( f != NULL );
-		f->set_kernel_info(info);
-	}
-
-	void add_ptxinfo( const struct gpgpu_ptx_sim_info &info )
-	{
-		m_binary_info = info;
-	}
-
-	void register_function( unsigned fat_cubin_handle, const char *hostFun, const char *deviceFun )
-	{
-		if( m_code.find(fat_cubin_handle) != m_code.end() ) {
-			symbol *s = m_code[fat_cubin_handle]->lookup(deviceFun);
-			if(s != NULL) {
-				function_info *f = s->get_pc();
-				assert( f != NULL );
-				m_kernel_lookup[hostFun] = f;
-			}
-			else {
-				printf("Warning: cannot find deviceFun %s\n", deviceFun);
-				m_kernel_lookup[hostFun] = NULL;
-			}
-	//		assert( s != NULL );
-	//		function_info *f = s->get_pc();
-	//		assert( f != NULL );
-	//		m_kernel_lookup[hostFun] = f;
-		} else {
-			m_kernel_lookup[hostFun] = NULL;
-		}
-	}
-
-    void register_hostFun_function( const char*hostFun, function_info* f){
-        m_kernel_lookup[hostFun] = f;
-    }
-
-	function_info *get_kernel(const char *hostFun)
-	{
-		std::map<const void*,function_info*>::iterator i=m_kernel_lookup.find(hostFun);
-		assert( i != m_kernel_lookup.end() );
-		return i->second;
-	}
-
-	int no_of_ptx;
-
-private:
-	_cuda_device_id *m_gpu; // selected gpu
-	std::map<unsigned,symbol_table*> m_code; // fat binary handle => global symbol table
-	unsigned m_last_fat_cubin_handle;
-	std::map<const void*,function_info*> m_kernel_lookup; // unique id (CUDA app function address) => kernel entry point
-	struct gpgpu_ptx_sim_info m_binary_info;
-
-};
-
-class kernel_config {
-public:
-	kernel_config( dim3 GridDim, dim3 BlockDim, size_t sharedMem, struct CUstream_st *stream )
-	{
-		m_GridDim=GridDim;
-		m_BlockDim=BlockDim;
-		m_sharedMem=sharedMem;
-		m_stream = stream;
-	}
-	kernel_config()
-	{
-		m_GridDim=dim3(-1,-1,-1);
-		m_BlockDim=dim3(-1,-1,-1);
-		m_sharedMem=0;
-		m_stream =NULL;
-	}
-	void set_arg( const void *arg, size_t size, size_t offset )
-	{
-		m_args.push_front( gpgpu_ptx_sim_arg(arg,size,offset) );
-	}
-	dim3 grid_dim() const { return m_GridDim; }
-	dim3 block_dim() const { return m_BlockDim; }
-	void set_grid_dim(dim3 *d) { m_GridDim = *d; }
-	void set_block_dim(dim3 *d) { m_BlockDim = *d; }
-	gpgpu_ptx_sim_arg_list_t get_args() { return m_args; }
-	struct CUstream_st *get_stream() { return m_stream; }
-
-private:
-	dim3 m_GridDim;
-	dim3 m_BlockDim;
-	size_t m_sharedMem;
-	struct CUstream_st *m_stream;
-	gpgpu_ptx_sim_arg_list_t m_args;
-};
-
 struct _cuda_device_id *GPGPUSim_Init()
 {
 	//static _cuda_device_id *the_device = NULL;
diff --git a/libcuda/cuobjdump.h b/libcuda/cuobjdump.h
index 49af3e2..6ab6778 100644
--- a/libcuda/cuobjdump.h
+++ b/libcuda/cuobjdump.h
@@ -4,6 +4,7 @@
 #include <list>
 #include <iostream>
 
+typedef void * yyscan_t;
 struct cuobjdump_parser {
     yyscan_t scanner;
     int elfserial;