diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/abstract_hardware_model.cc | 3 | ||||
| -rw-r--r-- | src/cuda-sim/instructions.cc | 10 |
2 files changed, 10 insertions, 3 deletions
diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc index b24a77e..83e76fe 100644 --- a/src/abstract_hardware_model.cc +++ b/src/abstract_hardware_model.cc @@ -389,7 +389,8 @@ void warp_inst_t::memory_coalescing_arch_13( bool is_write, mem_access_type acce assert(num_accesses <= MAX_ACCESSES_PER_INSN_PER_THREAD); - for(unsigned access=0; access<num_accesses; access++) { +// for(unsigned access=0; access<num_accesses; access++) { + for(unsigned access=0; (access<MAX_ACCESSES_PER_INSN_PER_THREAD)&&(m_per_scalar_thread[thread].memreqaddr[access]!=0); access++) { new_addr_type addr = m_per_scalar_thread[thread].memreqaddr[access]; unsigned block_address = line_size_based_tag_func(addr,segment_size); unsigned chunk = (addr&127)/32; // which 32-byte chunk within in a 128-byte chunk does this thread access? diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc index 6f1b5d3..42c63ca 100644 --- a/src/cuda-sim/instructions.cc +++ b/src/cuda-sim/instructions.cc @@ -3363,7 +3363,7 @@ void vp_ld_impl(const ptx_instruction *pI, core_t *core, warp_inst_t &inst) else if(wmma_layout==COL){ //mem->read(new_addr+4*stride*i,size/8,&data[i].s64); mem->read(new_addr+4*stride*i,size/8,&data[i].s64); - mem_txn_addr[num_mem_txn++]=new_addr+4*i; + mem_txn_addr[num_mem_txn++]=new_addr+4*stride*i; } } } @@ -3410,7 +3410,13 @@ void vp_ld_impl(const ptx_instruction *pI, core_t *core, warp_inst_t &inst) inst.set_addr(thrd, (new_addr_type *)mem_txn_addr , num_mem_txn); inst.data_size = 4; // 4 byte transaction assert( inst.memory_op == insn_memory_op ); - + printf("\nVPLOAD_THREAD%d:",thrd); + for(int lll=0;lll<num_mem_txn;lll++) + { + printf("%x ",mem_txn_addr[lll]); + } + printf("\n"); + int num_reg; if(g_debug_instruction){ printf("\nvp_ld:thread%d= ",thrd); |