diff options
| -rw-r--r-- | src/cuda-sim/cuda-sim.cc | 15 | ||||
| -rw-r--r-- | src/cuda-sim/cuda-sim.h | 10 |
2 files changed, 13 insertions, 12 deletions
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index 11ba4ec..832a223 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -112,8 +112,6 @@ cudaLaunchDeviceV2_init_perWarp, cudaLaunchDevicV2_perKernel>" "7200,8000,100,12000,1600"); } -static address_type get_converge_point(address_type pc); - void gpgpu_t::gpgpu_ptx_sim_bindNameToTexture(const char* name, const struct textureReference* texref, int dim, int readmode, int ext) { std::string texname(name); @@ -1096,7 +1094,7 @@ void ptx_instruction::pre_decode() } // get reconvergence pc - reconvergence_pc = get_converge_point(pc); + reconvergence_pc = gpgpu_ctx->func_sim->get_converge_point(pc); m_decoded=true; } @@ -2518,14 +2516,7 @@ void ptxinfo_opencl_addinfo( std::map<std::string,function_info*> &kernels ) clear_ptxinfo(); } -struct rec_pts { - gpgpu_recon_t *s_kernel_recon_points; - int s_num_recon; -}; - -class std::map<function_info*,rec_pts> g_rpts; - -struct rec_pts find_reconvergence_points( function_info *finfo ) +struct rec_pts cuda_sim::find_reconvergence_points( function_info *finfo ) { rec_pts tmp; std::map<function_info*,rec_pts>::iterator r=g_rpts.find(finfo); @@ -2564,7 +2555,7 @@ address_type get_return_pc( void *thd ) return the_thread->get_return_PC(); } -address_type get_converge_point( address_type pc ) +address_type cuda_sim::get_converge_point( address_type pc ) { // the branch could encode the reconvergence point and/or a bit that indicates the // reconvergence point is the return PC on the call stack in the case the branch has diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h index aa1fe40..0a1401a 100644 --- a/src/cuda-sim/cuda-sim.h +++ b/src/cuda-sim/cuda-sim.h @@ -120,6 +120,13 @@ void print_ptxinfo(); void clear_ptxinfo(); struct gpgpu_ptx_sim_info get_ptxinfo(); +class gpgpu_recon_t; +struct rec_pts { + gpgpu_recon_t *s_kernel_recon_points; + int s_num_recon; +}; + + class cuda_sim { public: cuda_sim( gpgpu_context* ctx ) { @@ -149,6 +156,7 @@ class cuda_sim { std::map<const void*,std::string> g_const_name_lookup; // indexed by hostVar int g_ptx_sim_mode; // if non-zero run functional simulation only (i.e., no notion of a clock cycle) unsigned gpgpu_param_num_shaders; + class std::map<function_info*,rec_pts> g_rpts; // backward pointer class gpgpu_context* gpgpu_ctx; //global functions @@ -165,6 +173,8 @@ class cuda_sim { void gpgpu_ptx_sim_register_const_variable(void*, const char *deviceName, size_t size ); void read_sim_environment_variables(); void set_param_gpgpu_num_shaders(int num_shaders); + struct rec_pts find_reconvergence_points( function_info *finfo ); + address_type get_converge_point( address_type pc ); }; #endif |