diff options
Diffstat (limited to 'src/gpgpu-sim/shader.cc')
| -rw-r--r-- | src/gpgpu-sim/shader.cc | 362 |
1 files changed, 250 insertions, 112 deletions
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index c6e7b8f..bcfda18 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -61,6 +61,20 @@ mem_fetch *shader_core_mem_fetch_allocator::alloc( m_core_id, m_cluster_id, m_memory_config, cycle); return mf; } + +mem_fetch *shader_core_mem_fetch_allocator::alloc( + new_addr_type addr, mem_access_type type, const active_mask_t &active_mask, + const mem_access_byte_mask_t &byte_mask, + const mem_access_sector_mask_t §or_mask, unsigned size, bool wr, + unsigned long long cycle, unsigned wid, unsigned sid, unsigned tpc, + mem_fetch *original_mf) const { + mem_access_t access(type, addr, size, wr, active_mask, byte_mask, sector_mask, + m_memory_config->gpgpu_ctx); + mem_fetch *mf = new mem_fetch( + access, NULL, wr ? WRITE_PACKET_SIZE : READ_PACKET_SIZE, wid, m_core_id, + m_cluster_id, m_memory_config, cycle, original_mf); + return mf; +} ///////////////////////////////////////////////////////////////////////////// std::list<unsigned> shader_core_ctx::get_regs_written(const inst_t &fvt) const { @@ -108,7 +122,7 @@ void shader_core_ctx::create_front_pipeline() { if (m_config->sub_core_model) { // in subcore model, each scheduler should has its own issue register, so - // num scheduler = reg width + // ensure num scheduler = reg width assert(m_config->gpgpu_num_sched_per_core == m_pipeline_reg[ID_OC_SP].get_size()); assert(m_config->gpgpu_num_sched_per_core == @@ -124,6 +138,11 @@ void shader_core_ctx::create_front_pipeline() { if (m_config->gpgpu_num_int_units > 0) assert(m_config->gpgpu_num_sched_per_core == m_pipeline_reg[ID_OC_INT].get_size()); + for (int j = 0; j < m_config->m_specialized_unit.size(); j++) { + if (m_config->m_specialized_unit[j].num_units > 0) + assert(m_config->gpgpu_num_sched_per_core == + m_config->m_specialized_unit[j].id_oc_spec_reg_width); + } } m_threadState = (thread_ctx_t *)calloc(sizeof(thread_ctx_t), @@ -172,6 +191,8 @@ void shader_core_ctx::create_schedulers() { ? CONCRETE_SCHEDULER_TWO_LEVEL_ACTIVE : sched_config.find("gto") != std::string::npos ? CONCRETE_SCHEDULER_GTO + : sched_config.find("rrr") != std::string::npos + ? CONCRETE_SCHEDULER_RRR : sched_config.find("old") != std::string::npos ? CONCRETE_SCHEDULER_OLDEST_FIRST : sched_config.find("warp_limiting") != @@ -206,6 +227,14 @@ void shader_core_ctx::create_schedulers() { &m_pipeline_reg[ID_OC_TENSOR_CORE], m_specilized_dispatch_reg, &m_pipeline_reg[ID_OC_MEM], i)); break; + case CONCRETE_SCHEDULER_RRR: + schedulers.push_back(new rrr_scheduler( + m_stats, this, m_scoreboard, m_simt_stack, &m_warp, + &m_pipeline_reg[ID_OC_SP], &m_pipeline_reg[ID_OC_DP], + &m_pipeline_reg[ID_OC_SFU], &m_pipeline_reg[ID_OC_INT], + &m_pipeline_reg[ID_OC_TENSOR_CORE], m_specilized_dispatch_reg, + &m_pipeline_reg[ID_OC_MEM], i)); + break; case CONCRETE_SCHEDULER_OLDEST_FIRST: schedulers.push_back(new oldest_scheduler( m_stats, this, m_scoreboard, m_simt_stack, &m_warp, @@ -377,41 +406,41 @@ void shader_core_ctx::create_exec_pipeline() { // m_fu = new simd_function_unit*[m_num_function_units]; - for (int k = 0; k < m_config->gpgpu_num_sp_units; k++) { - m_fu.push_back(new sp_unit(&m_pipeline_reg[EX_WB], m_config, this)); + for (unsigned k = 0; k < m_config->gpgpu_num_sp_units; k++) { + m_fu.push_back(new sp_unit(&m_pipeline_reg[EX_WB], m_config, this, k)); m_dispatch_port.push_back(ID_OC_SP); m_issue_port.push_back(OC_EX_SP); } - for (int k = 0; k < m_config->gpgpu_num_dp_units; k++) { - m_fu.push_back(new dp_unit(&m_pipeline_reg[EX_WB], m_config, this)); + for (unsigned k = 0; k < m_config->gpgpu_num_dp_units; k++) { + m_fu.push_back(new dp_unit(&m_pipeline_reg[EX_WB], m_config, this, k)); m_dispatch_port.push_back(ID_OC_DP); m_issue_port.push_back(OC_EX_DP); } - for (int k = 0; k < m_config->gpgpu_num_int_units; k++) { - m_fu.push_back(new int_unit(&m_pipeline_reg[EX_WB], m_config, this)); + for (unsigned k = 0; k < m_config->gpgpu_num_int_units; k++) { + m_fu.push_back(new int_unit(&m_pipeline_reg[EX_WB], m_config, this, k)); m_dispatch_port.push_back(ID_OC_INT); m_issue_port.push_back(OC_EX_INT); } - for (int k = 0; k < m_config->gpgpu_num_sfu_units; k++) { - m_fu.push_back(new sfu(&m_pipeline_reg[EX_WB], m_config, this)); + for (unsigned k = 0; k < m_config->gpgpu_num_sfu_units; k++) { + m_fu.push_back(new sfu(&m_pipeline_reg[EX_WB], m_config, this, k)); m_dispatch_port.push_back(ID_OC_SFU); m_issue_port.push_back(OC_EX_SFU); } - for (int k = 0; k < m_config->gpgpu_num_tensor_core_units; k++) { - m_fu.push_back(new tensor_core(&m_pipeline_reg[EX_WB], m_config, this)); + for (unsigned k = 0; k < m_config->gpgpu_num_tensor_core_units; k++) { + m_fu.push_back(new tensor_core(&m_pipeline_reg[EX_WB], m_config, this, k)); m_dispatch_port.push_back(ID_OC_TENSOR_CORE); m_issue_port.push_back(OC_EX_TENSOR_CORE); } - for (int j = 0; j < m_config->m_specialized_unit.size(); j++) { + for (unsigned j = 0; j < m_config->m_specialized_unit.size(); j++) { for (unsigned k = 0; k < m_config->m_specialized_unit[j].num_units; k++) { m_fu.push_back(new specialized_unit( &m_pipeline_reg[EX_WB], m_config, this, SPEC_UNIT_START_ID + j, m_config->m_specialized_unit[j].name, - m_config->m_specialized_unit[j].latency)); + m_config->m_specialized_unit[j].latency, k)); m_dispatch_port.push_back(m_config->m_specialized_unit[j].ID_OC_SPEC_ID); m_issue_port.push_back(m_config->m_specialized_unit[j].OC_EX_SPEC_ID); } @@ -1082,6 +1111,33 @@ void scheduler_unit::order_lrr( } } +template <class T> +void scheduler_unit::order_rrr( + std::vector<T> &result_list, const typename std::vector<T> &input_list, + const typename std::vector<T>::const_iterator &last_issued_from_input, + unsigned num_warps_to_add) { + result_list.clear(); + + if (m_num_issued_last_cycle > 0 || warp(m_current_turn_warp).done_exit() || + warp(m_current_turn_warp).waiting()) { + std::vector<shd_warp_t *>::const_iterator iter = + (last_issued_from_input == input_list.end()) ? + input_list.begin() : last_issued_from_input + 1; + for (unsigned count = 0; count < num_warps_to_add; ++iter, ++count) { + if (iter == input_list.end()) { + iter = input_list.begin(); + } + unsigned warp_id = (*iter)->get_warp_id(); + if (!(*iter)->done_exit() && !(*iter)->waiting()) { + result_list.push_back(*iter); + m_current_turn_warp = warp_id; + break; + } + } + } else { + result_list.push_back(&warp(m_current_turn_warp)); + } +} /** * A general function to order things in an priority-based way. * The core usage of the function is similar to order_lrr. @@ -1228,29 +1284,21 @@ void scheduler_unit::cycle() { previous_issued_inst_exec_type = exec_unit_type_t::MEM; } } else { - bool sp_pipe_avail = - (m_shader->m_config->gpgpu_num_sp_units > 0) && - m_sp_out->has_free(m_shader->m_config->sub_core_model, m_id); - bool sfu_pipe_avail = - (m_shader->m_config->gpgpu_num_sfu_units > 0) && - m_sfu_out->has_free(m_shader->m_config->sub_core_model, m_id); - bool tensor_core_pipe_avail = - (m_shader->m_config->gpgpu_num_tensor_core_units > 0) && - m_tensor_core_out->has_free( - m_shader->m_config->sub_core_model, m_id); - bool dp_pipe_avail = - (m_shader->m_config->gpgpu_num_dp_units > 0) && - m_dp_out->has_free(m_shader->m_config->sub_core_model, m_id); - bool int_pipe_avail = - (m_shader->m_config->gpgpu_num_int_units > 0) && - m_int_out->has_free(m_shader->m_config->sub_core_model, m_id); - // This code need to be refactored if (pI->op != TENSOR_CORE_OP && pI->op != SFU_OP && pI->op != DP_OP && !(pI->op >= SPEC_UNIT_START_ID)) { bool execute_on_SP = false; bool execute_on_INT = false; + bool sp_pipe_avail = + (m_shader->m_config->gpgpu_num_sp_units > 0) && + m_sp_out->has_free(m_shader->m_config->sub_core_model, + m_id); + bool int_pipe_avail = + (m_shader->m_config->gpgpu_num_int_units > 0) && + m_int_out->has_free(m_shader->m_config->sub_core_model, + m_id); + // if INT unit pipline exist, then execute ALU and INT // operations on INT unit and SP-FPU on SP unit (like in Volta) // if INT unit pipline does not exist, then execute all ALU, INT @@ -1311,6 +1359,11 @@ void scheduler_unit::cycle() { (pI->op == DP_OP) && !(diff_exec_units && previous_issued_inst_exec_type == exec_unit_type_t::DP)) { + bool dp_pipe_avail = + (m_shader->m_config->gpgpu_num_dp_units > 0) && + m_dp_out->has_free(m_shader->m_config->sub_core_model, + m_id); + if (dp_pipe_avail) { m_shader->issue_warp(*m_dp_out, pI, active_mask, warp_id, m_id); @@ -1326,6 +1379,11 @@ void scheduler_unit::cycle() { (pI->op == SFU_OP) || (pI->op == ALU_SFU_OP)) && !(diff_exec_units && previous_issued_inst_exec_type == exec_unit_type_t::SFU)) { + bool sfu_pipe_avail = + (m_shader->m_config->gpgpu_num_sfu_units > 0) && + m_sfu_out->has_free(m_shader->m_config->sub_core_model, + m_id); + if (sfu_pipe_avail) { m_shader->issue_warp(*m_sfu_out, pI, active_mask, warp_id, m_id); @@ -1337,6 +1395,11 @@ void scheduler_unit::cycle() { } else if ((pI->op == TENSOR_CORE_OP) && !(diff_exec_units && previous_issued_inst_exec_type == exec_unit_type_t::TENSOR)) { + bool tensor_core_pipe_avail = + (m_shader->m_config->gpgpu_num_tensor_core_units > 0) && + m_tensor_core_out->has_free( + m_shader->m_config->sub_core_model, m_id); + if (tensor_core_pipe_avail) { m_shader->issue_warp(*m_tensor_core_out, pI, active_mask, warp_id, m_id); @@ -1407,7 +1470,7 @@ void scheduler_unit::cycle() { m_last_supervised_issued = supervised_iter; } } - + m_num_issued_last_cycle = issued; if (issued == 1) m_stats->single_issue_nums[m_id]++; else if (issued > 1) @@ -1456,6 +1519,10 @@ void lrr_scheduler::order_warps() { order_lrr(m_next_cycle_prioritized_warps, m_supervised_warps, m_last_supervised_issued, m_supervised_warps.size()); } +void rrr_scheduler::order_warps() { + order_rrr(m_next_cycle_prioritized_warps, m_supervised_warps, + m_last_supervised_issued, m_supervised_warps.size()); +} void gto_scheduler::order_warps() { order_by_priority(m_next_cycle_prioritized_warps, m_supervised_warps, @@ -1569,7 +1636,10 @@ void swl_scheduler::order_warps() { } } -void shader_core_ctx::read_operands() {} +void shader_core_ctx::read_operands() { + for (int i = 0; i < m_config->reg_file_port_throughput; ++i) + m_operand_collector.step(); +} address_type coalesced_segment(address_type addr, unsigned segment_size_lg2bytes) { @@ -1669,8 +1739,15 @@ void shader_core_ctx::execute() { m_fu[n]->active_lanes_in_pipeline(); unsigned issue_port = m_issue_port[n]; register_set &issue_inst = m_pipeline_reg[issue_port]; - warp_inst_t **ready_reg = issue_inst.get_ready(); - if (issue_inst.has_ready() && m_fu[n]->can_issue(**ready_reg)) { + unsigned reg_id; + bool partition_issue = + m_config->sub_core_model && m_fu[n]->is_issue_partitioned(); + if (partition_issue) { + reg_id = m_fu[n]->get_issue_reg_id(); + } + warp_inst_t **ready_reg = issue_inst.get_ready(partition_issue, reg_id); + if (issue_inst.has_ready(partition_issue, reg_id) && + m_fu[n]->can_issue(**ready_reg)) { bool schedule_wb_now = !m_fu[n]->stallable(); int resbus = -1; if (schedule_wb_now && @@ -1970,6 +2047,21 @@ void ldst_unit::L1_latency_queue_cycle() { } else { assert(status == MISS || status == HIT_RESERVED); l1_latency_queue[j][0] = NULL; + if (m_config->m_L1D_config.get_write_policy() != WRITE_THROUGH && + mf_next->get_inst().is_store() && + (m_config->m_L1D_config.get_write_allocate_policy() == + FETCH_ON_WRITE || + m_config->m_L1D_config.get_write_allocate_policy() == + LAZY_FETCH_ON_READ) && + !was_writeallocate_sent(events)) { + unsigned dec_ack = + (m_config->m_L1D_config.get_mshr_type() == SECTOR_ASSOC) + ? (mf_next->get_data_size() / SECTOR_SIZE) + : 1; + mf_next->set_reply(); + for (unsigned i = 0; i < dec_ack; ++i) m_core->store_ack(mf_next); + if (!write_sent && !read_sent) delete mf_next; + } } } @@ -2112,22 +2204,32 @@ simd_function_unit::simd_function_unit(const shader_core_config *config) { m_dispatch_reg = new warp_inst_t(config); } +void simd_function_unit::issue(register_set &source_reg) { + bool partition_issue = + m_config->sub_core_model && this->is_issue_partitioned(); + source_reg.move_out_to(partition_issue, this->get_issue_reg_id(), + m_dispatch_reg); + occupied.set(m_dispatch_reg->latency); +} + sfu::sfu(register_set *result_port, const shader_core_config *config, - shader_core_ctx *core) - : pipelined_simd_unit(result_port, config, config->max_sfu_latency, core) { + shader_core_ctx *core, unsigned issue_reg_id) + : pipelined_simd_unit(result_port, config, config->max_sfu_latency, core, + issue_reg_id) { m_name = "SFU"; } tensor_core::tensor_core(register_set *result_port, const shader_core_config *config, - shader_core_ctx *core) + shader_core_ctx *core, unsigned issue_reg_id) : pipelined_simd_unit(result_port, config, config->max_tensor_core_latency, - core) { + core, issue_reg_id) { m_name = "TENSOR_CORE"; } void sfu::issue(register_set &source_reg) { - warp_inst_t **ready_reg = source_reg.get_ready(); + warp_inst_t **ready_reg = + source_reg.get_ready(m_config->sub_core_model, m_issue_reg_id); // m_core->incexecstat((*ready_reg)); (*ready_reg)->op_pipe = SFU__OP; @@ -2136,7 +2238,8 @@ void sfu::issue(register_set &source_reg) { } void tensor_core::issue(register_set &source_reg) { - warp_inst_t **ready_reg = source_reg.get_ready(); + warp_inst_t **ready_reg = + source_reg.get_ready(m_config->sub_core_model, m_issue_reg_id); // m_core->incexecstat((*ready_reg)); (*ready_reg)->op_pipe = TENSOR_CORE__OP; @@ -2208,34 +2311,39 @@ void tensor_core::active_lanes_in_pipeline() { } sp_unit::sp_unit(register_set *result_port, const shader_core_config *config, - shader_core_ctx *core) - : pipelined_simd_unit(result_port, config, config->max_sp_latency, core) { + shader_core_ctx *core, unsigned issue_reg_id) + : pipelined_simd_unit(result_port, config, config->max_sp_latency, core, + issue_reg_id) { m_name = "SP "; } specialized_unit::specialized_unit(register_set *result_port, const shader_core_config *config, shader_core_ctx *core, unsigned supported_op, - char *unit_name, unsigned latency) - : pipelined_simd_unit(result_port, config, latency, core) { + char *unit_name, unsigned latency, + unsigned issue_reg_id) + : pipelined_simd_unit(result_port, config, latency, core, issue_reg_id) { m_name = unit_name; m_supported_op = supported_op; } dp_unit::dp_unit(register_set *result_port, const shader_core_config *config, - shader_core_ctx *core) - : pipelined_simd_unit(result_port, config, config->max_dp_latency, core) { + shader_core_ctx *core, unsigned issue_reg_id) + : pipelined_simd_unit(result_port, config, config->max_dp_latency, core, + issue_reg_id) { m_name = "DP "; } int_unit::int_unit(register_set *result_port, const shader_core_config *config, - shader_core_ctx *core) - : pipelined_simd_unit(result_port, config, config->max_int_latency, core) { + shader_core_ctx *core, unsigned issue_reg_id) + : pipelined_simd_unit(result_port, config, config->max_int_latency, core, + issue_reg_id) { m_name = "INT "; } void sp_unit ::issue(register_set &source_reg) { - warp_inst_t **ready_reg = source_reg.get_ready(); + warp_inst_t **ready_reg = + source_reg.get_ready(m_config->sub_core_model, m_issue_reg_id); // m_core->incexecstat((*ready_reg)); (*ready_reg)->op_pipe = SP__OP; m_core->incsp_stat(m_core->get_config()->warp_size, (*ready_reg)->latency); @@ -2243,7 +2351,8 @@ void sp_unit ::issue(register_set &source_reg) { } void dp_unit ::issue(register_set &source_reg) { - warp_inst_t **ready_reg = source_reg.get_ready(); + warp_inst_t **ready_reg = + source_reg.get_ready(m_config->sub_core_model, m_issue_reg_id); // m_core->incexecstat((*ready_reg)); (*ready_reg)->op_pipe = DP__OP; m_core->incsp_stat(m_core->get_config()->warp_size, (*ready_reg)->latency); @@ -2251,7 +2360,8 @@ void dp_unit ::issue(register_set &source_reg) { } void specialized_unit ::issue(register_set &source_reg) { - warp_inst_t **ready_reg = source_reg.get_ready(); + warp_inst_t **ready_reg = + source_reg.get_ready(m_config->sub_core_model, m_issue_reg_id); // m_core->incexecstat((*ready_reg)); (*ready_reg)->op_pipe = SPECIALIZED__OP; m_core->incsp_stat(m_core->get_config()->warp_size, (*ready_reg)->latency); @@ -2259,7 +2369,8 @@ void specialized_unit ::issue(register_set &source_reg) { } void int_unit ::issue(register_set &source_reg) { - warp_inst_t **ready_reg = source_reg.get_ready(); + warp_inst_t **ready_reg = + source_reg.get_ready(m_config->sub_core_model, m_issue_reg_id); // m_core->incexecstat((*ready_reg)); (*ready_reg)->op_pipe = INTP__OP; m_core->incsp_stat(m_core->get_config()->warp_size, (*ready_reg)->latency); @@ -2269,7 +2380,8 @@ void int_unit ::issue(register_set &source_reg) { pipelined_simd_unit::pipelined_simd_unit(register_set *result_port, const shader_core_config *config, unsigned max_latency, - shader_core_ctx *core) + shader_core_ctx *core, + unsigned issue_reg_id) : simd_function_unit(config) { m_result_port = result_port; m_pipeline_depth = max_latency; @@ -2277,6 +2389,7 @@ pipelined_simd_unit::pipelined_simd_unit(register_set *result_port, for (unsigned i = 0; i < m_pipeline_depth; i++) m_pipeline_reg[i] = new warp_inst_t(config); m_core = core; + m_issue_reg_id = issue_reg_id; active_insts_in_pipeline = 0; } @@ -2303,7 +2416,10 @@ void pipelined_simd_unit::cycle() { void pipelined_simd_unit::issue(register_set &source_reg) { // move_warp(m_dispatch_reg,source_reg); - warp_inst_t **ready_reg = source_reg.get_ready(); + bool partition_issue = + m_config->sub_core_model && this->is_issue_partitioned(); + warp_inst_t **ready_reg = + source_reg.get_ready(partition_issue, m_issue_reg_id); m_core->incexecstat((*ready_reg)); // source_reg.move_out_to(m_dispatch_reg); simd_function_unit::issue(source_reg); @@ -2360,7 +2476,7 @@ ldst_unit::ldst_unit(mem_fetch_interface *icnt, Scoreboard *scoreboard, const shader_core_config *config, const memory_config *mem_config, shader_core_stats *stats, unsigned sid, unsigned tpc) - : pipelined_simd_unit(NULL, config, config->smem_latency, core), + : pipelined_simd_unit(NULL, config, config->smem_latency, core, 0), m_next_wb(config) { assert(config->smem_latency > 1); init(icnt, mf_allocator, core, operand_collector, scoreboard, config, @@ -2388,7 +2504,7 @@ ldst_unit::ldst_unit(mem_fetch_interface *icnt, Scoreboard *scoreboard, const shader_core_config *config, const memory_config *mem_config, shader_core_stats *stats, unsigned sid, unsigned tpc, l1_cache *new_l1d_cache) - : pipelined_simd_unit(NULL, config, 3, core), + : pipelined_simd_unit(NULL, config, 3, core, 0), m_L1D(new_l1d_cache), m_next_wb(config) { init(icnt, mf_allocator, core, operand_collector, scoreboard, config, @@ -2550,8 +2666,7 @@ inst->space.get_type() != shared_space) { unsigned warp_id = inst->warp_id(); */ void ldst_unit::cycle() { writeback(); - for (int i = 0; i < m_config->reg_file_port_throughput; ++i) - m_operand_collector->step(); + for (unsigned stage = 0; (stage + 1) < m_pipeline_depth; stage++) if (m_pipeline_reg[stage]->empty() && !m_pipeline_reg[stage + 1]->empty()) move_warp(m_pipeline_reg[stage], m_pipeline_reg[stage + 1]); @@ -3264,49 +3379,46 @@ unsigned int shader_core_config::max_cta(const kernel_info_t &k) const { if (adaptive_cache_config && !k.cache_config_set) { // For more info about adaptive cache, see // https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory-7-x - unsigned total_shmed = kernel_info->smem * result; - assert(total_shmed >= 0 && total_shmed <= gpgpu_shmem_size); - // assert(gpgpu_shmem_size == 98304); //Volta has 96 KB shared - // assert(m_L1D_config.get_nset() == 4); //Volta L1 has four sets - if (total_shmed < gpgpu_shmem_size) { - switch (adaptive_cache_config) { - case FIXED: - break; - case ADAPTIVE_VOLTA: { - // For Volta, we assign the remaining shared memory to L1 cache - // For more info about adaptive cache, see - // https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory-7-x - // assert(gpgpu_shmem_size == 98304); //Volta has 96 KB shared + unsigned total_shmem = kernel_info->smem * result; + assert(total_shmem >= 0 && total_shmem <= shmem_opt_list.back()); - // To Do: make it flexible and not tuned to 9KB share memory - unsigned max_assoc = m_L1D_config.get_max_assoc(); - if (total_shmed == 0) - m_L1D_config.set_assoc(max_assoc); // L1 is 128KB and shd=0 - else if (total_shmed > 0 && total_shmed <= 8192) - m_L1D_config.set_assoc(0.9375 * - max_assoc); // L1 is 120KB and shd=8KB - else if (total_shmed > 8192 && total_shmed <= 16384) - m_L1D_config.set_assoc(0.875 * - max_assoc); // L1 is 112KB and shd=16KB - else if (total_shmed > 16384 && total_shmed <= 32768) - m_L1D_config.set_assoc(0.75 * max_assoc); // L1 is 96KB and - // shd=32KB - else if (total_shmed > 32768 && total_shmed <= 65536) - m_L1D_config.set_assoc(0.5 * max_assoc); // L1 is 64KB and shd=64KB - else if (total_shmed > 65536 && total_shmed <= gpgpu_shmem_size) - m_L1D_config.set_assoc(0.25 * max_assoc); // L1 is 32KB and - // shd=96KB - else - assert(0); - break; - } - default: - assert(0); + // Unified cache config is in KB. Converting to B + unsigned total_unified = m_L1D_config.m_unified_cache_size * 1024; + + bool l1d_configured = false; + unsigned max_assoc = m_L1D_config.get_max_assoc(); + + for (std::vector<unsigned>::const_iterator it = shmem_opt_list.begin(); + it < shmem_opt_list.end(); it++) { + if (total_shmem <= *it) { + float l1_ratio = 1 - ((float)*(it) / total_unified); + // make sure the ratio is between 0 and 1 + assert(0 <= l1_ratio && l1_ratio <= 1); + // round to nearest instead of round down + m_L1D_config.set_assoc(max_assoc * l1_ratio + 0.5f); + l1d_configured = true; + break; } + } - printf("GPGPU-Sim: Reconfigure L1 cache to %uKB\n", - m_L1D_config.get_total_size_inKB()); + assert(l1d_configured && "no shared memory option found"); + + if (m_L1D_config.is_streaming()) { + // for streaming cache, if the whole memory is allocated + // to the L1 cache, then make the allocation to be on_MISS + // otherwise, make it ON_FILL to eliminate line allocation fails + // i.e. MSHR throughput is the same, independent on the L1 cache + // size/associativity + if (total_shmem == 0) { + m_L1D_config.set_allocation_policy(ON_MISS); + printf("GPGPU-Sim: Reconfigure L1 allocation to ON_MISS\n"); + } else { + m_L1D_config.set_allocation_policy(ON_FILL); + printf("GPGPU-Sim: Reconfigure L1 allocation to ON_FILL\n"); + } } + printf("GPGPU-Sim: Reconfigure L1 cache to %uKB\n", + m_L1D_config.get_total_size_inKB()); k.cache_config_set = true; } @@ -3867,15 +3979,26 @@ void opndcoll_rfu_t::init(unsigned num_banks, shader_core_ctx *shader) { assert((m_bank_warp_shift == 5) || (m_warp_size != 32)); sub_core_model = shader->get_config()->sub_core_model; - m_num_warp_sceds = shader->get_config()->gpgpu_num_sched_per_core; - if (sub_core_model) + m_num_warp_scheds = shader->get_config()->gpgpu_num_sched_per_core; + unsigned reg_id; + if (sub_core_model) { assert(num_banks % shader->get_config()->gpgpu_num_sched_per_core == 0); + assert(m_num_warp_scheds <= m_cu.size() && + m_cu.size() % m_num_warp_scheds == 0); + } m_num_banks_per_sched = num_banks / shader->get_config()->gpgpu_num_sched_per_core; for (unsigned j = 0; j < m_cu.size(); j++) { + if (sub_core_model) { + unsigned cusPerSched = m_cu.size() / m_num_warp_scheds; + reg_id = j / cusPerSched; + } m_cu[j]->init(j, num_banks, m_bank_warp_shift, shader->get_config(), this, - sub_core_model, m_num_banks_per_sched); + sub_core_model, reg_id, m_num_banks_per_sched); + } + for (unsigned j = 0; j < m_dispatch_units.size(); j++) { + m_dispatch_units[j].init(sub_core_model,m_num_warp_scheds); } m_initialized = true; } @@ -3974,7 +4097,22 @@ void opndcoll_rfu_t::allocate_cu(unsigned port_num) { for (unsigned j = 0; j < inp.m_cu_sets.size(); j++) { std::vector<collector_unit_t> &cu_set = m_cus[inp.m_cu_sets[j]]; bool allocated = false; - for (unsigned k = 0; k < cu_set.size(); k++) { + unsigned cuLowerBound = 0; + unsigned cuUpperBound = cu_set.size(); + unsigned schd_id; + if (sub_core_model) { + // Sub core model only allocates on the subset of CUs assigned to the + // scheduler that issued + unsigned reg_id = (*inp.m_in[i]).get_ready_reg_id(); + schd_id = (*inp.m_in[i]).get_schd_id(reg_id); + assert(cu_set.size() % m_num_warp_scheds == 0 && + cu_set.size() >= m_num_warp_scheds); + unsigned cusPerSched = cu_set.size() / m_num_warp_scheds; + cuLowerBound = schd_id * cusPerSched; + cuUpperBound = cuLowerBound + cusPerSched; + assert(0 <= cuLowerBound && cuUpperBound <= cu_set.size()); + } + for (unsigned k = cuLowerBound; k < cuUpperBound; k++) { if (cu_set[k].is_free()) { collector_unit_t *cu = &cu_set[k]; allocated = cu->allocate(inp.m_in[i], inp.m_out[i]); @@ -3984,8 +4122,9 @@ void opndcoll_rfu_t::allocate_cu(unsigned port_num) { } if (allocated) break; // cu has been allocated, no need to search more. } - break; // can only service a single input, if it failed it will fail for - // others. + // break; // can only service a single input, if it failed it will fail + // for + // others. } } } @@ -4032,7 +4171,8 @@ void opndcoll_rfu_t::allocate_reads() { } bool opndcoll_rfu_t::collector_unit_t::ready() const { - return (!m_free) && m_not_ready.none() && (*m_output_register).has_free(); + return (!m_free) && m_not_ready.none() && + (*m_output_register).has_free(m_sub_core_model, m_reg_id); } void opndcoll_rfu_t::collector_unit_t::dump( @@ -4050,12 +4190,10 @@ void opndcoll_rfu_t::collector_unit_t::dump( } } -void opndcoll_rfu_t::collector_unit_t::init(unsigned n, unsigned num_banks, - unsigned log2_warp_size, - const core_config *config, - opndcoll_rfu_t *rfu, - bool sub_core_model, - unsigned banks_per_sched) { +void opndcoll_rfu_t::collector_unit_t::init( + unsigned n, unsigned num_banks, unsigned log2_warp_size, + const core_config *config, opndcoll_rfu_t *rfu, bool sub_core_model, + unsigned reg_id, unsigned banks_per_sched) { m_rfu = rfu; m_cuid = n; m_num_banks = num_banks; @@ -4063,6 +4201,7 @@ void opndcoll_rfu_t::collector_unit_t::init(unsigned n, unsigned num_banks, m_warp = new warp_inst_t(config); m_bank_warp_shift = log2_warp_size; m_sub_core_model = sub_core_model; + m_reg_id = reg_id; m_num_banks_per_sched = banks_per_sched; } @@ -4097,8 +4236,7 @@ bool opndcoll_rfu_t::collector_unit_t::allocate(register_set *pipeline_reg_set, void opndcoll_rfu_t::collector_unit_t::dispatch() { assert(m_not_ready.none()); - // move_warp(*m_output_register,m_warp); - m_output_register->move_in(m_warp); + m_output_register->move_in(m_sub_core_model, m_reg_id, m_warp); m_free = true; m_output_register = NULL; for (unsigned i = 0; i < MAX_REG_OPERANDS * 2; i++) m_src_op[i].reset(); |