diff options
| -rw-r--r-- | src/gpgpu-sim/shader.cc | 5 | ||||
| -rw-r--r-- | src/gpgpu-sim/shader.h | 3 |
2 files changed, 6 insertions, 2 deletions
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index 19c2291..50ad5ea 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -1089,14 +1089,15 @@ void ldst_unit::writeback() m_pending_writes[m_next_wb.warp_id()].erase(m_next_wb.out[r]); m_scoreboard->releaseRegister( m_next_wb.warp_id(), m_next_wb.out[r] ); m_stats->m_num_sim_insn[m_sid]++; + m_core->get_gpu()->gpu_sim_insn += m_next_wb.active_count(); } } else { // shared m_scoreboard->releaseRegister( m_next_wb.warp_id(), m_next_wb.out[r] ); m_stats->m_num_sim_insn[m_sid]++; + m_core->get_gpu()->gpu_sim_insn += m_next_wb.active_count(); } } } - m_core->get_gpu()->gpu_sim_insn += m_next_wb.active_count(); m_next_wb.clear(); m_last_inst_gpu_sim_cycle = gpu_sim_cycle; m_last_inst_gpu_tot_sim_cycle = gpu_tot_sim_cycle; @@ -1234,6 +1235,7 @@ void ldst_unit::cycle() } } if( !pending_requests ) { + m_core->get_gpu()->gpu_sim_insn += m_dispatch_reg->active_count(); m_scoreboard->releaseRegisters(m_dispatch_reg); m_stats->m_num_sim_insn[m_sid]++; } @@ -1243,6 +1245,7 @@ void ldst_unit::cycle() } else { // stores exit pipeline here m_core->dec_inst_in_pipeline(warp_id); + m_core->get_gpu()->gpu_sim_insn += m_dispatch_reg->active_count(); m_dispatch_reg->clear(); m_stats->m_num_sim_insn[m_sid]++; } diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h index dcb24be..e7d2f1b 100644 --- a/src/gpgpu-sim/shader.h +++ b/src/gpgpu-sim/shader.h @@ -1110,7 +1110,8 @@ public: mem_fetch *alloc( const warp_inst_t &inst, const mem_access_t &access ) const { warp_inst_t inst_copy = inst; - inst_copy.set_active(access.get_warp_mask()); + // tgrogers Oct 27, 2011 - I am commenting out this line of code, because it is screwing up our instruction counts on memory instructions. I am not sure what the point of this was, because no one seems to read the active mask on the memory fetches's warp instruction. + //inst_copy.set_active(access.get_warp_mask()); mem_fetch *mf = new mem_fetch(access, &inst_copy, access.is_write()?WRITE_PACKET_SIZE:READ_PACKET_SIZE, |