diff options
| -rw-r--r-- | src/gpgpu-sim/gpu-sim.cc | 9 | ||||
| -rw-r--r-- | src/gpgpu-sim/local_interconnect.h | 30 |
2 files changed, 27 insertions, 12 deletions
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 92d5366..4f071c7 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -257,9 +257,12 @@ void shader_core_config::reg_options(class OptionParser * opp) "per-shader L1 data cache config " " {<nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>,<mshr>:<N>:<merge>,<mq> | none}", "none" ); + option_parser_register(opp, "-l1_banks", OPT_UINT32, &m_L1D_config.l1_banks, + "The number of L1 cache banks", + "1"); option_parser_register(opp, "-l1_latency", OPT_UINT32, &m_L1D_config.l1_latency, "L1 Hit Latency", - "0"); + "1"); option_parser_register(opp, "-smem_latency", OPT_UINT32, &smem_latency, "smem Latency", "3"); @@ -320,8 +323,8 @@ void shader_core_config::reg_options(class OptionParser * opp) option_parser_register(opp, "-gpgpu_shmem_size", OPT_UINT32, &gpgpu_shmem_size, "Size of shared memory per shader core (default 16kB)", "16384"); - option_parser_register(opp, "-adaptive_volta_cache_config", OPT_BOOL, &adaptive_volta_cache_config, - "adaptive_volta_cache_config", + option_parser_register(opp, "-adaptive_cache_config", OPT_BOOL, &adaptive_volta_cache_config, + "adaptive_cache_config", "0"); option_parser_register(opp, "-gpgpu_shmem_sizeDefault", OPT_UINT32, &gpgpu_shmem_sizeDefault, "Size of shared memory per shader core (default 16kB)", diff --git a/src/gpgpu-sim/local_interconnect.h b/src/gpgpu-sim/local_interconnect.h index f4a2af1..a784da8 100644 --- a/src/gpgpu-sim/local_interconnect.h +++ b/src/gpgpu-sim/local_interconnect.h @@ -35,27 +35,35 @@ using namespace std; +enum Interconnect_type { + REQ_NET=0, + REPLY_NET=1 +}; + +enum Arbiteration_type { + NAIVE_RR=0, + iSLIP=1 +}; + struct inct_config { - //config for local interconnect unsigned in_buffer_limit; unsigned out_buffer_limit; unsigned subnets; + Arbiteration_type arbiter_algo; }; -enum Interconnect_type { - REQ_NET=0, - REPLY_NET=1 -}; class xbar_router { public: - xbar_router(unsigned router_id, enum Interconnect_type m_type, unsigned n_shader, unsigned n_mem, unsigned m_in_buffer_limit, unsigned m_out_buffer_limit); + xbar_router(unsigned router_id, enum Interconnect_type m_type, unsigned n_shader, unsigned n_mem, unsigned m_in_buffer_limit, unsigned m_out_buffer_limit, enum Arbiteration_type m_arbit_type); ~xbar_router(); void Push(unsigned input_deviceID, unsigned output_deviceID, void* data, unsigned int size); void* Pop(unsigned ouput_deviceID); - void Advance(); + void Advance( ); + + bool Busy() const; bool Has_Buffer_In(unsigned input_deviceID, unsigned size, bool update_counter=false); bool Has_Buffer_Out(unsigned output_deviceID, unsigned size); @@ -70,6 +78,9 @@ public: unsigned long long packets_num; private: + void iSLIP_Advance(); + void RR_Advance(); + struct Packet{ Packet(void* m_data, unsigned m_output_deviceID) { data = m_data; @@ -82,11 +93,12 @@ private: vector<queue<Packet> > out_buffers; unsigned _n_shader, _n_mem, total_nodes; unsigned in_buffer_limit, out_buffer_limit; - vector<unsigned> next_node; -// unsigned next_node; + vector<unsigned> next_node; //used for iSLIP arbit + unsigned next_node_id; //used for RR arbit unsigned m_id; enum Interconnect_type router_type; unsigned active_in_buffers,active_out_buffers; + Arbiteration_type arbit_type; friend class LocalInterconnect; |