diff options
Diffstat (limited to 'src/gpgpu-sim/shader.cc')
| -rw-r--r-- | src/gpgpu-sim/shader.cc | 21 |
1 files changed, 16 insertions, 5 deletions
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index c0161dd..6f68283 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -1,6 +1,7 @@ // Copyright (c) 2009-2021, Tor M. Aamodt, Wilson W.L. Fung, Ali Bakhoda, -// George L. Yuan, Andrew Turner, Inderpreet Singh, Vijay Kandiah, Nikos Hardavellas -// The University of British Columbia, Northwestern University +// George L. Yuan, Andrew Turner, Inderpreet Singh, Vijay Kandiah, Nikos Hardavellas, +// Mahmoud Khairy, Junrui Pan, Timothy G. Rogers +// The University of British Columbia, Northwestern University, Purdue University // All rights reserved. // // Redistribution and use in source and binary forms, with or without @@ -950,7 +951,7 @@ void shader_core_ctx::fetch() { m_threadState[tid].m_active = false; unsigned cta_id = m_warp[warp_id]->get_cta_id(); if (m_thread[tid] == NULL) { - register_cta_thread_exit(cta_id, m_kernel); + register_cta_thread_exit(cta_id, m_warp[warp_id]->get_kernel_info()); } else { register_cta_thread_exit(cta_id, &(m_thread[tid]->get_kernel())); @@ -2091,10 +2092,11 @@ bool ldst_unit::constant_cycle(warp_inst_t &inst, mem_stage_stall_type &rc_fail, mem_stage_stall_type fail; if (m_config->perfect_inst_const_cache) { fail = NO_RC_FAIL; + unsigned access_count = inst.accessq_count(); while (inst.accessq_count() > 0) inst.accessq_pop_back(); if (inst.is_load()) { for (unsigned r = 0; r < MAX_OUTPUT_VALUES; r++) - if (inst.out[r] > 0) m_pending_writes[inst.warp_id()][inst.out[r]]--; + if (inst.out[r] > 0) m_pending_writes[inst.warp_id()][inst.out[r]] -= access_count; } } else { fail = process_memory_access_queue(m_L1C, inst); @@ -3897,6 +3899,8 @@ void shader_core_ctx::get_icnt_power_stats(long &n_simt_to_mem, n_mem_to_simt += m_stats->n_mem_to_simt[m_sid]; } +kernel_info_t* shd_warp_t::get_kernel_info() const { return m_shader->get_kernel_info(); } + bool shd_warp_t::functional_done() const { return get_n_completed() == m_warp_size; } @@ -4238,12 +4242,19 @@ bool opndcoll_rfu_t::collector_unit_t::allocate(register_set *pipeline_reg_set, warp_inst_t **pipeline_reg = pipeline_reg_set->get_ready(); if ((pipeline_reg) and !((*pipeline_reg)->empty())) { m_warp_id = (*pipeline_reg)->warp_id(); + std::vector<int> prev_regs; // remove duplicate regs within same instr for (unsigned op = 0; op < MAX_REG_OPERANDS; op++) { int reg_num = (*pipeline_reg) ->arch_reg.src[op]; // this math needs to match that used in // function_info::ptx_decode_inst - if (reg_num >= 0) { // valid register + bool new_reg = true; + for (auto r : prev_regs) { + if (r == reg_num) + new_reg = false; + } + if (reg_num >= 0 && new_reg) { // valid register + prev_regs.push_back(reg_num); m_src_op[op] = op_t(this, op, reg_num, m_num_banks, m_bank_warp_shift, m_sub_core_model, m_num_banks_per_sched, (*pipeline_reg)->get_schd_id()); |