Merge pull request #94 from deval281shah/dev

changes for ubuntu 18.04 support

3 files changed, 53 insertions, 3 deletions

diff --git a/README b/README
index 29d6366..998cb51 100644
--- a/README
+++ b/README
@@ -20,7 +20,7 @@ April 19-21, 2009.
 
 If you use cuDNN and Pytorch support, the Checkpoint function or the Debigging tool for functional simulation error in GPGPU-Sim for your research,
 please cite:
-Jonathan Lew, Deval Shah, Suchita Pati, Shaylin Cattell, Mengchi Zhang, Amruth Sandhupatla, Christopher Ng, Negar Goli, Matthew D. Sinclair, Timothy G. Rogers, Tor Aamodt
+Jonathan Lew, Deval Shah, Suchita Pati, Shaylin Cattell, Mengchi Zhang, Amruth Sandhupatla, Christopher Ng, Negar Goli, Matthew D. Sinclair, Timothy G. Rogers, Tor M. Aamodt
 Analyzing Machine Learning Workloads Using a Detailed GPU Simulator, arXiv:1811.08933,
 https://arxiv.org/abs/1811.08933
 
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc
index 95a3c24..e6b53c3 100644
--- a/libcuda/cuda_runtime_api.cc
+++ b/libcuda/cuda_runtime_api.cc
@@ -1199,6 +1199,26 @@ __host__ cudaError_t CUDARTAPI cudaGetDevice(int *device)
 	*device = g_active_device;
 	return g_last_cudaError = cudaSuccess;
 }
+__host__ cudaError_t CUDARTAPI cudaDeviceGetLimit ( size_t* pValue, cudaLimit limit )
+{
+	if(g_debug_execution >= 3){
+            announce_call(__my_func__);
+   	 }
+        cuda_not_implemented(__my_func__,__LINE__);
+        return g_last_cudaError = cudaSuccess;
+
+}
+
+
+__host__ cudaError_t CUDARTAPI cudaStreamGetPriority ( cudaStream_t hStream, int* priority )
+{
+        if(g_debug_execution >= 3){
+            announce_call(__my_func__);
+   	 }
+        cuda_not_implemented(__my_func__,__LINE__);
+        return g_last_cudaError = cudaSuccess;
+
+}
 
 __host__ cudaError_t CUDARTAPI cudaDeviceGetPCIBusId (
 		char *pciBusId,
@@ -1235,6 +1255,16 @@ __host__ cudaError_t cudaIpcOpenMemHandle(
 	return g_last_cudaError = cudaErrorUnknown;
 }
 
+__host__ cudaError_t CUDARTAPI cudaDestroyTextureObject(cudaTextureObject_t texObject)
+{
+        if(g_debug_execution >= 3){
+            announce_call(__my_func__);
+   	 }
+        cuda_not_implemented(__my_func__,__LINE__);
+        return g_last_cudaError = cudaErrorUnknown;
+}
+
+
 /*******************************************************************************
  *                                                                              *
  *                                                                              *
@@ -2552,8 +2582,17 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin )
         // Making this a runtime variable based on the app, enables GPGPU-Sim compiled
         // with a newer version of CUDA to run apps compiled with older versions of
         // CUDA. This is especially useful for PTXPLUS execution.
-        int app_cuda_version = get_app_cuda_version();
-        assert( app_cuda_version == CUDART_VERSION / 1000  && "The app must be compiled with same major version as the simulator." );
+        //Skip cuda version check for pytorch application
+	std::string app_binary_path =  get_app_binary();
+        int pos = app_binary_path.find("python");
+        if (pos==std::string::npos){
+        	// Not pytorch app : checking cuda version
+		int app_cuda_version = get_app_cuda_version();
+        	assert( app_cuda_version == CUDART_VERSION / 1000  && "The app must be compiled with same major version as the simulator." );
+        }
+
+	//int app_cuda_version = get_app_cuda_version();
+        //assert( app_cuda_version == CUDART_VERSION / 1000  && "The app must be compiled with same major version as the simulator." );
         const char* filename;
 #if CUDART_VERSION < 6000
             // FatBin handle from the .fatbin.c file (one of the intermediate files generated by NVCC)
@@ -4447,6 +4486,16 @@ CUresult CUDAAPI cuPointerGetAttribute(void *data, CUpointer_attribute attribute
 #endif /* CUDART_VERSION >= 4000 */
 
 #if CUDART_VERSION >= 8000
+__host__ cudaError_t CUDARTAPI cudaCreateTextureObject ( cudaTextureObject_t* pTexObject, const cudaResourceDesc* pResDesc, const cudaTextureDesc* pTexDesc, const cudaResourceViewDesc* pResViewDesc )
+{
+         if(g_debug_execution >= 3){
+            announce_call(__my_func__);
+        }
+        cuda_not_implemented(__my_func__,__LINE__);
+        return g_last_cudaError = cudaSuccess;
+
+}
+
 CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream)
 {
 	if(g_debug_execution >= 3){
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index 08aa88c..28a22e0 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -73,6 +73,7 @@ enum FuncCache
 #include <set>
 
 typedef unsigned long long new_addr_type;
+typedef unsigned long long cudaTextureObject_t;
 typedef unsigned address_type;
 typedef unsigned addr_t;