Move g_total_param_size

Signed-off-by: Mengchi Zhang <[email protected]>

4 files changed, 8 insertions, 6 deletions

diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc
index ef09051..0f4291d 100644
--- a/src/abstract_hardware_model.cc
+++ b/src/abstract_hardware_model.cc
@@ -771,8 +771,7 @@ bool kernel_info_t::children_all_finished() {
 
 void kernel_info_t::notify_parent_finished() {
    if(m_parent_kernel) {
-       extern unsigned long long g_total_param_size;
-       g_total_param_size -= ((m_kernel_entry->get_args_aligned_size() + 255)/256*256);
+       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());
    }
diff --git a/src/cuda-sim/cuda_device_runtime.cc b/src/cuda-sim/cuda_device_runtime.cc
index 354fa79..22304db 100644
--- a/src/cuda-sim/cuda_device_runtime.cc
+++ b/src/cuda-sim/cuda_device_runtime.cc
@@ -5,7 +5,6 @@
 #include <iostream>
 #include <map>
 
-unsigned long long g_total_param_size = 0;
 unsigned long long g_max_total_param_size = 0;
 
 
@@ -70,7 +69,7 @@ std::list<device_launch_operation_t> g_cuda_device_launch_op;
 
 //Handling device runtime api:
 //void * cudaGetParameterBufferV2(void *func, dim3 gridDimension, dim3 blockDimension, unsigned int sharedMemSize)
-void gpgpusim_cuda_getParameterBufferV2(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func)
+void cuda_device_runtime::gpgpusim_cuda_getParameterBufferV2(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func)
 {
     DEV_RUNTIME_REPORT("Calling cudaGetParameterBufferV2");
       
diff --git a/src/cuda-sim/cuda_device_runtime.h b/src/cuda-sim/cuda_device_runtime.h
index 851fed2..dd908ae 100644
--- a/src/cuda-sim/cuda_device_runtime.h
+++ b/src/cuda-sim/cuda_device_runtime.h
@@ -3,7 +3,6 @@
 #if (CUDART_VERSION >= 5000)
 #pragma once
 
-void gpgpusim_cuda_getParameterBufferV2(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func);
 void gpgpusim_cuda_launchDeviceV2(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func);
 void gpgpusim_cuda_streamCreateWithFlags(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func);
 #endif
@@ -14,10 +13,13 @@ class gpgpu_context;
 class cuda_device_runtime {
     public:
 	cuda_device_runtime( gpgpu_context* ctx ) {
+	    g_total_param_size = 0;
 	    gpgpu_ctx = ctx;
 	}
+	unsigned long long g_total_param_size;
 	// backward pointer
 	class gpgpu_context* gpgpu_ctx;
+	void gpgpusim_cuda_getParameterBufferV2(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func);
 	void launch_all_device_kernels();
 	void launch_one_device_kernel();
 };
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 46a262d..a72bc0c 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -56,6 +56,8 @@ class ptx_recognizer;
 #include "cuda_device_runtime.h"
 
 #include <stdarg.h>
+#include "../../libcuda/gpgpu_context.h"
+
 using half_float::half;
 
 unsigned ptx_instruction::g_num_ptx_inst_uid=0;
@@ -2095,7 +2097,7 @@ void call_impl( const ptx_instruction *pI, ptx_thread_info *thread )
 #if (CUDART_VERSION >= 5000)
    //Jin: handle device runtime apis for CDP
    else if(fname == "cudaGetParameterBufferV2") {
-      gpgpusim_cuda_getParameterBufferV2(pI, thread, target_func);
+      target_func->gpgpu_ctx->device_runtime->gpgpusim_cuda_getParameterBufferV2(pI, thread, target_func);
 	  return;
    }
    else if(fname == "cudaLaunchDeviceV2") {