From 8db50a319911b803a386a7d7eca10afbea4d7e75 Mon Sep 17 00:00:00 2001 From: aamir Date: Fri, 2 Nov 2018 21:37:31 -0700 Subject: added incount and outcount, addressed all of the pull request comments --- src/cuda-sim/cuda-sim.cc | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-) (limited to 'src/cuda-sim/cuda-sim.cc') diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index 86a1d45..23c5ad5 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -602,7 +602,7 @@ void ptx_instruction::set_opcode_and_latency() * [3] MAD * [4] DIV */ - sscanf(opcode_latency_int, "%u,%u,%u,%u,%u,%u,%u", + sscanf(opcode_latency_int, "%u,%u,%u,%u,%u", &int_latency[0],&int_latency[1],&int_latency[2], &int_latency[3],&int_latency[4]); sscanf(opcode_latency_fp, "%u,%u,%u,%u,%u", @@ -850,12 +850,14 @@ void ptx_instruction::pre_decode() { pc = m_PC; isize = m_inst_size; - for(unsigned i=0; i<8; i++) { + for(unsigned i=0; i0) + outcount++; + + for(int i=0;i0) + incount++; // Get predicate if(has_pred()) { @@ -1316,7 +1327,10 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id) if(((inst_opcode==MMA_OP||inst_opcode==MMA_LD_OP||inst_opcode==MMA_ST_OP))){ if(inst.active_count()!=MAX_WARP_SIZE) - while(1); + { + printf("Tensor Core operation are warp synchronous operation. All the threads needs to be active."); + 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))){ -- cgit v1.3