diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/abstract_hardware_model.cc | 1 | ||||
| -rw-r--r-- | src/abstract_hardware_model.h | 7 | ||||
| -rw-r--r-- | src/gpgpu-sim/shader.cc | 4 |
3 files changed, 8 insertions, 4 deletions
diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc index a7da2ff..19c3091 100644 --- a/src/abstract_hardware_model.cc +++ b/src/abstract_hardware_model.cc @@ -316,6 +316,7 @@ void warp_inst_t::memory_coalescing_arch_13( bool is_write, mem_access_type acce // local memory can only be accessed in 4B chunks by one thread unsigned data_size_coales = (space.get_type() == local_space || space.get_type() == param_space_local ) ? 4 : data_size; unsigned num_accesses = (space.get_type() == local_space || space.get_type() == param_space_local ) ? data_size/4 : 1; + assert(num_accesses <= MAX_ACCESSES_PER_INSN_PER_THREAD); for(unsigned access=0; access<num_accesses; access++) { new_addr_type addr = m_per_scalar_thread[thread].memreqaddr[access]; diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h index 8d4d009..97b243e 100644 --- a/src/abstract_hardware_model.h +++ b/src/abstract_hardware_model.h @@ -632,6 +632,8 @@ enum divergence_support_t { NUM_SIMD_MODEL }; +const unsigned MAX_ACCESSES_PER_INSN_PER_THREAD = 8; + class warp_inst_t: public inst_t { public: // constructors @@ -684,6 +686,7 @@ public: m_per_scalar_thread.resize(m_config->warp_size); m_per_scalar_thread_valid=true; } + assert(num_addrs <= MAX_ACCESSES_PER_INSN_PER_THREAD); for(unsigned i=0; i<num_addrs; i++) m_per_scalar_thread[n].memreqaddr[i] = addr[i]; } @@ -782,11 +785,11 @@ protected: struct per_thread_info { per_thread_info() { - for(unsigned i=0; i<8; i++) + for(unsigned i=0; i<MAX_ACCESSES_PER_INSN_PER_THREAD; i++) memreqaddr[i] = 0; } dram_callback_t callback; - new_addr_type memreqaddr[8]; // effective address, upto 8 different requests (to support 32B access in 8 chunks of 4B each) + new_addr_type memreqaddr[MAX_ACCESSES_PER_INSN_PER_THREAD]; // effective address, upto 8 different requests (to support 32B access in 8 chunks of 4B each) }; bool m_per_scalar_thread_valid; std::vector<per_thread_info> m_per_scalar_thread; diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index b37ceef..7ef5140 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -608,7 +608,7 @@ void shader_core_ctx::func_exec_inst( warp_inst_t &inst ) if( inst.has_callback(t) ) m_warp[inst.warp_id()].inc_n_atomic(); if (inst.space.is_local() && (inst.is_load() || inst.is_store())) { - new_addr_type localaddrs[8]; + new_addr_type localaddrs[MAX_ACCESSES_PER_INSN_PER_THREAD]; unsigned num_addrs; num_addrs = translate_local_memaddr(inst.get_addr(t), tid, m_config->n_simt_clusters*m_config->n_simt_cores_per_cluster, inst.data_size, (new_addr_type*) localaddrs ); @@ -797,7 +797,7 @@ unsigned shader_core_ctx::translate_local_memaddr( address_type localaddr, unsig assert(datasize%4 == 0); assert(datasize >= 4); - assert(datasize <= 32); // max 32B + assert(datasize/4 <= MAX_ACCESSES_PER_INSN_PER_THREAD); // max 32B assert(localaddr%4 == 0); // Required if accessing 4B per request, otherwise access will overflow into next thread's space for(unsigned i=0; i<datasize/4; i++) { address_type local_word = localaddr/4 + i; |