diff options
| author | Jin Wang <[email protected]> | 2014-11-15 05:44:03 -0500 |
|---|---|---|
| committer | Jin Wang <[email protected]> | 2016-07-06 02:17:18 -0400 |
| commit | b70a5a69fa14cc01f707b910f8f021e36067922f (patch) | |
| tree | a2d3181a9671594b9691f38c0f591c5e95885260 /src/cuda-sim/cuda-sim.cc | |
| parent | a31392370bc7b23e115160b170e95b117597ebc4 (diff) | |
ADD: print kernel parameter size footprint. BUG: concurrent kernels on same shader, should use hw_cta_id to store shared mem info
Diffstat (limited to 'src/cuda-sim/cuda-sim.cc')
| -rw-r--r-- | src/cuda-sim/cuda-sim.cc | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index c87e3e4..3f5af7e 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -1499,7 +1499,8 @@ unsigned ptx_sim_init_thread( kernel_info_t &kernel, unsigned max_cta_per_sm = num_threads/cta_size; // e.g., 256 / 48 = 5 assert( max_cta_per_sm > 0 ); - unsigned sm_idx = (tid/cta_size)*gpgpu_param_num_shaders + sid; + //unsigned sm_idx = (tid/cta_size)*gpgpu_param_num_shaders + sid; + unsigned sm_idx = hw_cta_id*gpgpu_param_num_shaders + sid; if ( shared_memory_lookup.find(sm_idx) == shared_memory_lookup.end() ) { if ( g_debug_execution >= 1 ) { |