From adc311951d67b0685ebf2fab4ce6410f96f0039a Mon Sep 17 00:00:00 2001 From: sspenst Date: Mon, 11 Jul 2016 11:16:46 -0700 Subject: Reverted the previous commit to add a cleaner way of getting NUM_THREADS. Now, sst_impl doesn't functionally execute on the last indexed element of an array, but instead on the actual last thread that executes --- libcuda/cuda_runtime_api.cc | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-) (limited to 'libcuda') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 3eff4af..e8a0e91 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -180,9 +180,6 @@ cudaError_t g_last_cudaError = cudaSuccess; extern stream_manager *g_stream_manager; -dim3 gridDim = (dim3){1,1,1}; -dim3 blockDim = (dim3){1,1,1}; - void register_ptx_function( const char *name, function_info *impl ) { // no longer need this @@ -962,8 +959,8 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); kernel_info_t *grid = gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); std::string kname = grid->name(); - gridDim = config.grid_dim(); - blockDim = config.block_dim(); + dim3 gridDim = config.grid_dim(); + dim3 blockDim = config.block_dim(); 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,g_ptx_sim_mode,stream); -- cgit v1.3