summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/abstract_hardware_model.cc14
-rw-r--r--src/abstract_hardware_model.h2
-rw-r--r--src/gpgpu-sim/gpu-cache.cc2
-rw-r--r--src/gpgpu-sim/gpu-sim.cc12
-rw-r--r--src/gpgpu-sim/gpu-sim.h2
-rw-r--r--src/gpgpu-sim/shader.cc43
-rw-r--r--src/gpgpu-sim/shader.h4
7 files changed, 39 insertions, 40 deletions
diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc
index f75c525..64eb43c 100644
--- a/src/abstract_hardware_model.cc
+++ b/src/abstract_hardware_model.cc
@@ -315,11 +315,11 @@ void warp_inst_t::generate_mem_accesses()
case global_space: case local_space: case param_space_local:
if( m_config->gpgpu_coalesce_arch == 13 || m_config->gpgpu_coalesce_arch == 20) {
- if(isatomic())
- memory_coalescing_arch_atomic(is_write, access_type);
- else
- memory_coalescing_arch(is_write, access_type);
- } else abort();
+ if(isatomic())
+ memory_coalescing_arch_atomic(is_write, access_type);
+ else
+ memory_coalescing_arch(is_write, access_type);
+ } else abort();
break;
@@ -358,7 +358,7 @@ void warp_inst_t::memory_coalescing_arch( bool is_write, mem_access_type access_
unsigned segment_size = 0;
unsigned warp_parts;
- //TO DO: need to double check how double number are coalesced!
+ //TO DO: need to double check how doubles are coalesced!
if(data_size == 1)
{
//If it is byte data, then coalesce on the whole 32 threads, regardless the arch version
@@ -367,7 +367,7 @@ void warp_inst_t::memory_coalescing_arch( bool is_write, mem_access_type access_
else if(m_config->gpgpu_coalesce_arch == 13)
{
//mem_warp_parts should equal 2 for arch=13
- //we still use the parameter mem_warp_parts for arch=13 to ensure it is backward compatibility with older gpgpu config files
+ //use the parameter mem_warp_parts for arch=13 to ensure it is backward compatibility with older gpgpu config files
warp_parts = m_config->mem_warp_parts;
}
else if(m_config->gpgpu_coalesce_arch == 20)
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index cd185a1..910a9ed 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -685,7 +685,7 @@ public:
bool wr,
const active_mask_t &active_mask,
const mem_access_byte_mask_t &byte_mask,
- const mem_access_sector_mask_t &sector_mask)
+ const mem_access_sector_mask_t &sector_mask)
: m_warp_mask(active_mask), m_byte_mask(byte_mask), m_sector_mask(sector_mask)
{
init();
diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc
index cfd0dc8..7af7db0 100644
--- a/src/gpgpu-sim/gpu-cache.cc
+++ b/src/gpgpu-sim/gpu-cache.cc
@@ -868,7 +868,7 @@ data_cache::wr_miss_wa( new_addr_type addr,
false, // Now performing a read
mf->get_access_warp_mask(),
mf->get_access_byte_mask(),
- mf->get_access_sector_mask());
+ mf->get_access_sector_mask());
mem_fetch *n_mf = new mem_fetch( *ma,
NULL,
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 26432c4..483a7b6 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -302,7 +302,7 @@ void shader_core_config::reg_options(class OptionParser * opp)
"0");
option_parser_register(opp, "-gpgpu_local_mem_map", OPT_BOOL, &gpgpu_local_mem_map,
"Mapping from local memory space address to simulated GPU physical address space (default = enabled)",
- "1");
+ "1");
option_parser_register(opp, "-gpgpu_num_reg_banks", OPT_INT32, &gpgpu_num_reg_banks,
"Number of register banks (default = 8)",
"8");
@@ -346,17 +346,17 @@ void shader_core_config::reg_options(class OptionParser * opp)
"number of collector unit in ports (default = 0)",
"0");
option_parser_register(opp, "-gpgpu_coalesce_arch", OPT_INT32, &gpgpu_coalesce_arch,
- "Coalescing arch (GT200 = 13, Fermi = 20)",
+ "Coalescing arch (GT200 = 13, Fermi = 20)",
"13");
option_parser_register(opp, "-gpgpu_num_sched_per_core", OPT_INT32, &gpgpu_num_sched_per_core,
"Number of warp schedulers per core",
"1");
option_parser_register(opp, "-gpgpu_max_insn_issue_per_warp", OPT_INT32, &gpgpu_max_insn_issue_per_warp,
- "Max number of instructions that can be issued per warp in one cycle by scheduler (either 1 or 2)",
- "2");
+ "Max number of instructions that can be issued per warp in one cycle by scheduler (either 1 or 2)",
+ "2");
option_parser_register(opp, "-gpgpu_dual_issue_diff_exec_units", OPT_BOOL, &gpgpu_dual_issue_diff_exec_units,
- "should dual issue use two different execution unit resources",
- "1");
+ "should dual issue use two different execution unit resources",
+ "1");
option_parser_register(opp, "-gpgpu_simt_core_sim_order", OPT_INT32, &simt_core_sim_order,
"Select the simulation order of cores in a cluster (0=Fix, 1=Round-Robin)",
"1");
diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h
index 23e6144..7d92c66 100644
--- a/src/gpgpu-sim/gpu-sim.h
+++ b/src/gpgpu-sim/gpu-sim.h
@@ -314,8 +314,6 @@ public:
unsigned num_shader() const { return m_shader_config.num_shader(); }
unsigned num_cluster() const { return m_shader_config.n_simt_clusters; }
unsigned get_max_concurrent_kernel() const { return max_concurrent_kernel; }
- unsigned get_gpu_max_cycle_opt() const { return gpu_max_cycle_opt; }
- unsigned get_gpu_max_insn_opt() const { return gpu_max_insn_opt; }
private:
void init_clock_domains(void );
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 015995e..d826b69 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -132,8 +132,8 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu,
CONCRETE_SCHEDULER_TWO_LEVEL_ACTIVE :
sched_config.find("gto") != std::string::npos ?
CONCRETE_SCHEDULER_GTO :
- sched_config.find("old") != std::string::npos ?
- CONCRETE_SCHEDULER_OLDEST_FIRST :
+ sched_config.find("old") != std::string::npos ?
+ CONCRETE_SCHEDULER_OLDEST_FIRST :
sched_config.find("warp_limiting") != std::string::npos ?
CONCRETE_SCHEDULER_WARP_LIMITING:
NUM_CONCRETE_SCHEDULERS;
@@ -187,16 +187,16 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu,
break;
case CONCRETE_SCHEDULER_OLDEST_FIRST:
schedulers.push_back(
- new oldest_scheduler( m_stats,
- this,
- m_scoreboard,
- m_simt_stack,
- &m_warp,
- &m_pipeline_reg[ID_OC_SP],
- &m_pipeline_reg[ID_OC_SFU],
- &m_pipeline_reg[ID_OC_MEM],
- i
- )
+ new oldest_scheduler( m_stats,
+ this,
+ m_scoreboard,
+ m_simt_stack,
+ &m_warp,
+ &m_pipeline_reg[ID_OC_SP],
+ &m_pipeline_reg[ID_OC_SFU],
+ &m_pipeline_reg[ID_OC_MEM],
+ i
+ )
);
break;
case CONCRETE_SCHEDULER_WARP_LIMITING:
@@ -750,12 +750,13 @@ void shader_core_ctx::issue_warp( register_set& pipe_reg_set, const warp_inst_t*
void shader_core_ctx::issue(){
- unsigned j;
- for (unsigned i = 0; i < schedulers.size(); i++) {
- j = (Issue_Prio + i) % schedulers.size();
- schedulers[j]->cycle();
- }
- Issue_Prio = (Issue_Prio+1)% schedulers.size();
+ //Ensure fair round robin issu between schedulers
+ unsigned j;
+ for (unsigned i = 0; i < schedulers.size(); i++) {
+ j = (Issue_Prio + i) % schedulers.size();
+ schedulers[j]->cycle();
+ }
+ Issue_Prio = (Issue_Prio+1)% schedulers.size();
//really is issue;
//for (unsigned i = 0; i < schedulers.size(); i++) {
@@ -876,8 +877,8 @@ void scheduler_unit::cycle()
unsigned checked=0;
unsigned issued=0;
exec_unit_type_t previous_issued_inst_exec_type = exec_unit_type_t::NONE;
- unsigned max_issue = m_shader->m_config->gpgpu_max_insn_issue_per_warp;
- bool diff_exec_units = m_shader->m_config->gpgpu_dual_issue_diff_exec_units;
+ unsigned max_issue = m_shader->m_config->gpgpu_max_insn_issue_per_warp;
+ bool diff_exec_units = m_shader->m_config->gpgpu_dual_issue_diff_exec_units;
while( !warp(warp_id).waiting() && !warp(warp_id).ibuffer_empty() && (checked < max_issue) && (checked <= issued) && (issued < max_issue) ) {
const warp_inst_t *pI = warp(warp_id).ibuffer_next_inst();
//Jin: handle cdp latency;
@@ -1062,7 +1063,7 @@ void oldest_scheduler::order_warps()
m_supervised_warps,
m_last_supervised_issued,
m_supervised_warps.size(),
- ORDERED_PRIORITY_FUNC_ONLY,
+ ORDERED_PRIORITY_FUNC_ONLY,
scheduler_unit::sort_warps_by_oldest_dynamic_id );
}
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 395c1ed..2fed420 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -316,7 +316,7 @@ enum concrete_scheduler
CONCRETE_SCHEDULER_GTO,
CONCRETE_SCHEDULER_TWO_LEVEL_ACTIVE,
CONCRETE_SCHEDULER_WARP_LIMITING,
- CONCRETE_SCHEDULER_OLDEST_FIRST,
+ CONCRETE_SCHEDULER_OLDEST_FIRST,
NUM_CONCRETE_SCHEDULERS
};
@@ -456,7 +456,7 @@ public:
: scheduler_unit ( stats, shader, scoreboard, simt, warp, sp_out, sfu_out, mem_out, id ){}
virtual ~oldest_scheduler () {}
virtual void order_warps ();
- virtual void done_adding_supervised_warps() {
+ virtual void done_adding_supervised_warps() {
m_last_supervised_issued = m_supervised_warps.begin();
}
generated by cgit v1.3 (git 2.53.0) at 2026-07-11 02:56:00 +0000