diff options
| -rw-r--r-- | src/gpgpu-sim/shader.cc | 15 | ||||
| -rw-r--r-- | src/gpgpu-sim/shader.h | 8 |
2 files changed, 20 insertions, 3 deletions
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index e7204c8..db6a532 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -441,6 +441,12 @@ void shader_core_stats::print( FILE* fout ) const fprintf(fout,"gpgpu_mem_divergence_hist "); gpgpu_mem_divergence_hist->fprint(fout); fprintf(fout,"\n"); + fprintf(fout,"gpgpu_gmem_ld_divergence_hist "); + gpgpu_gmem_ld_divergence_hist->fprint(fout); + fprintf(fout,"\n"); + fprintf(fout,"gpgpu_gmem_st_divergence_hist "); + gpgpu_gmem_st_divergence_hist->fprint(fout); + fprintf(fout,"\n"); fprintf(fout, "gpgpu_n_load_insn = %d\n", gpgpu_n_load_insn); fprintf(fout, "gpgpu_n_store_insn = %d\n", gpgpu_n_store_insn); @@ -746,12 +752,17 @@ void shader_core_ctx::func_exec_inst( warp_inst_t &inst ) { unsigned starting_queue_size; execute_warp_inst_t(inst); - if( inst.is_load() || inst.is_store() ) + if( inst.is_load() || inst.is_store() ) { starting_queue_size = inst.accessq_count(); inst.generate_mem_accesses(); if ( inst.space.get_type() == global_space ) { - m_stats->gpgpu_mem_divergence_hist->add2bin(inst.accessq_count() - starting_queue_size); + if (inst.is_load()) + m_stats->gpgpu_gmem_ld_divergence_hist->add2bin(inst.accessq_count() - starting_queue_size); + else if (inst.is_store()) + m_stats->gpgpu_gmem_st_divergence_hist->add2bin(inst.accessq_count() - starting_queue_size); } + m_stats->gpgpu_mem_divergence_hist->add2bin(inst.accessq_count() - starting_queue_size); + } } void shader_core_ctx::issue_warp( register_set& pipe_reg_set, const warp_inst_t* next_inst, const active_mask_t &active_mask, unsigned warp_id ) diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h index 45658ac..da7eb32 100644 --- a/src/gpgpu-sim/shader.h +++ b/src/gpgpu-sim/shader.h @@ -1475,7 +1475,9 @@ struct shader_core_stats_pod { //warps combined memory divergence histogram linear_histogram* gpgpu_mem_divergence_hist; - + linear_histogram* gpgpu_gmem_ld_divergence_hist; + linear_histogram* gpgpu_gmem_st_divergence_hist; + int gpgpu_n_mem_l2_writeback; int gpgpu_n_mem_l1_write_allocate; int gpgpu_n_mem_l2_write_allocate; @@ -1545,6 +1547,8 @@ public: gpgpu_n_shmem_bank_access = (unsigned *)calloc(config->num_shader(), sizeof(unsigned)); gpgpu_mem_divergence_hist = new linear_histogram(1, "", config->warp_size+1); + gpgpu_gmem_ld_divergence_hist = new linear_histogram(1, "", config->warp_size+1); + gpgpu_gmem_st_divergence_hist = new linear_histogram(1, "", config->warp_size+1); m_shader_dynamic_warp_issue_distro.resize( config->num_shader() ); m_shader_warp_slot_issue_distro.resize( config->num_shader() ); @@ -1560,6 +1564,8 @@ public: free(shader_cycle_distro); free(last_shader_cycle_distro); free(gpgpu_mem_divergence_hist); + free(gpgpu_gmem_ld_divergence_hist); + free(gpgpu_gmem_st_divergence_hist); } void new_grid() |