diff options
| author | aamir <[email protected]> | 2018-11-02 23:17:24 -0700 |
|---|---|---|
| committer | aamir <[email protected]> | 2018-11-02 23:17:24 -0700 |
| commit | 9807c9393164bf01b14fcda0c35d75b248749c4f (patch) | |
| tree | 4503783f9bfe900ffcba35169504ffa9cad54fc1 /src/cuda-sim/cuda-sim.cc | |
| parent | 8db50a319911b803a386a7d7eca10afbea4d7e75 (diff) | |
some minor changes asked in the code review
Diffstat (limited to 'src/cuda-sim/cuda-sim.cc')
| -rw-r--r-- | src/cuda-sim/cuda-sim.cc | 12 |
1 files changed, 10 insertions, 2 deletions
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index 23c5ad5..b2b93e8 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -1271,6 +1271,13 @@ static unsigned get_tex_datasize( const ptx_instruction *pI, ptx_thread_info *th return data_size; } +int tensorcore_op(int inst_opcode){ + + if((inst_opcode==MMA_OP)&&(inst_opcode==MMA_LD_OP)&&(inst_opcode==MMA_ST_OP)) + return 1; + else + return 0; +} void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id) { @@ -1332,8 +1339,9 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id) assert(0); } } - - if(((inst_opcode!=MMA_OP)&&(inst_opcode!=MMA_LD_OP)&&(inst_opcode!=MMA_ST_OP))||((inst_opcode==MMA_OP||inst_opcode==MMA_LD_OP||inst_opcode==MMA_ST_OP)&&(lane_id==0))){ + + //Tensorcore is warp synchronous operation. So these instructions needs to be executed only once. To make the simulation faster removing the redundant tensorcore operation + if(!tensorcore_op(inst_opcode)||(tensorcore_op)&&(lane_id==0)){ switch ( inst_opcode ) { #define OP_DEF(OP,FUNC,STR,DST,CLASSIFICATION) case OP: FUNC(pI,this); op_classification = CLASSIFICATION; break; #define OP_W_DEF(OP,FUNC,STR,DST,CLASSIFICATION) case OP: FUNC(pI,get_core(),inst); op_classification = CLASSIFICATION; break; |