diff options
Diffstat (limited to 'src/cuda-sim')
| -rw-r--r-- | src/cuda-sim/cuda-sim.cc | 4 | ||||
| -rw-r--r-- | src/cuda-sim/instructions.cc | 8 |
2 files changed, 11 insertions, 1 deletions
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index 57da23f..e194a2a 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -572,6 +572,9 @@ void ptx_instruction::set_bar_type() abort(); } } + else if(m_opcode==SST_OP) { + bar_type = SYNC; + } } @@ -635,6 +638,7 @@ void ptx_instruction::set_opcode_and_latency() case TEX_OP: op = LOAD_OP; mem_op=TEX; break; case ATOM_OP: op = LOAD_OP; break; case BAR_OP: op = BARRIER_OP; break; + case SST_OP: op = BARRIER_OP; break; case MEMBAR_OP: op = MEMORY_BARRIER_OP; break; case CALL_OP: { diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc index aeaf9e6..8bdb94f 100644 --- a/src/cuda-sim/instructions.cc +++ b/src/cuda-sim/instructions.cc @@ -3741,7 +3741,8 @@ void sqrt_impl( const ptx_instruction *pI, ptx_thread_info *thread ) void sst_impl( const ptx_instruction *pI, ptx_thread_info *thread ) { - const operand_info &dst = pI->dst(); + ptx_instruction * cpI = const_cast<ptx_instruction *>(pI); // constant + const operand_info &dst = cpI->dst(); const operand_info &src1 = pI->src1(); const operand_info &src2 = pI->src2(); const operand_info &src3 = pI->src3(); @@ -3763,8 +3764,13 @@ void sst_impl( const ptx_instruction *pI, ptx_thread_info *thread ) // store data in sstarr memory mem->write(addr,size/8,&src3_data.s64,thread,pI); + // sync threads + cpI->set_bar_id(dst_data.u32); + thread->m_last_effective_address = addr; thread->m_last_memory_space = space; + thread->m_last_dram_callback.function = bar_callback; + thread->m_last_dram_callback.instruction = cpI; int NUM_THREADS = 8; // (how do you get this dynamically?) if (src2_data.s64 == NUM_THREADS-1) { |