Merge pull request #17 from JRPan/rewrite-l2-breakdown

Rewrite l2 breakdown

7 files changed, 48 insertions, 55 deletions

diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index b33c50b..60d7328 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -876,7 +876,9 @@ class mem_fetch_allocator {
                             const mem_access_byte_mask_t &byte_mask,
                             const mem_access_sector_mask_t &sector_mask,
                             unsigned size, bool wr,
-                            unsigned long long cycle) const = 0;                    
+                            unsigned long long cycle,
+                            unsigned wid, unsigned sid,
+                            unsigned tpc, mem_fetch *original_mf) const = 0;                    
 };
 
 // the maximum number of destination, source, or address uarch operands in a
diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc
index 297a94c..7e7d2ad 100644
--- a/src/gpgpu-sim/gpu-cache.cc
+++ b/src/gpgpu-sim/gpu-cache.cc
@@ -282,7 +282,7 @@ enum cache_request_status tag_array::probe(new_addr_type addr, unsigned &idx,
       // percentage of dirty lines in the cache
       // number of dirty lines / total lines in the cache
       float dirty_line_percentage = 
-          (float) (m_dirty / (m_config.m_nset * m_config.m_assoc )) * 100;
+          ((float) m_dirty / (m_config.m_nset * m_config.m_assoc )) * 100;
       if (!line->is_modified_line() ||
           dirty_line_percentage >= m_config.m_wr_percent) {
         // if number of dirty lines in the cache is greater than
@@ -1338,7 +1338,7 @@ enum cache_request_status data_cache::wr_miss_wa_naive(
         evicted.m_block_addr,m_wrbk_type,
         mf->get_access_warp_mask(), evicted.m_byte_mask,
         evicted.m_sector_mask, evicted.m_modified_size,
-        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle);
+        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle,-1,-1,-1,NULL);
       // the evicted block may have wrong chip id when advanced L2 hashing  is
       // used, so set the right chip address from the original mf
       wb->set_chip(mf->get_tlx_addr().chip);
@@ -1391,7 +1391,7 @@ enum cache_request_status data_cache::wr_miss_wa_fetch_on_write(
         evicted.m_block_addr,m_wrbk_type,
         mf->get_access_warp_mask(), evicted.m_byte_mask,
         evicted.m_sector_mask, evicted.m_modified_size,
-        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle);
+        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle,-1,-1,-1,NULL);
         // the evicted block may have wrong chip id when advanced L2 hashing  is
         // used, so set the right chip address from the original mf
         wb->set_chip(mf->get_tlx_addr().chip);
@@ -1464,7 +1464,7 @@ enum cache_request_status data_cache::wr_miss_wa_fetch_on_write(
         evicted.m_block_addr,m_wrbk_type,
         mf->get_access_warp_mask(), evicted.m_byte_mask,
         evicted.m_sector_mask, evicted.m_modified_size,
-        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle);
+        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle,-1,-1,-1,NULL);
         // the evicted block may have wrong chip id when advanced L2 hashing  is
         // used, so set the right chip address from the original mf
         wb->set_chip(mf->get_tlx_addr().chip);
@@ -1549,7 +1549,7 @@ enum cache_request_status data_cache::wr_miss_wa_lazy_fetch_on_read(
         evicted.m_block_addr,m_wrbk_type,
         mf->get_access_warp_mask(), evicted.m_byte_mask,
         evicted.m_sector_mask, evicted.m_modified_size,
-        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle);
+        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle,-1,-1,-1,NULL);
       // the evicted block may have wrong chip id when advanced L2 hashing  is
       // used, so set the right chip address from the original mf
       wb->set_chip(mf->get_tlx_addr().chip);
@@ -1631,7 +1631,7 @@ enum cache_request_status data_cache::rd_miss_base(
         evicted.m_block_addr,m_wrbk_type,
         mf->get_access_warp_mask(), evicted.m_byte_mask,
         evicted.m_sector_mask, evicted.m_modified_size,
-        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle);
+        true, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle,-1,-1,-1,NULL);
       // the evicted block may have wrong chip id when advanced L2 hashing  is
       // used, so set the right chip address from the original mf
       wb->set_chip(mf->get_tlx_addr().chip);
diff --git a/src/gpgpu-sim/gpu-cache.h b/src/gpgpu-sim/gpu-cache.h
index 6698d92..007403f 100644
--- a/src/gpgpu-sim/gpu-cache.h
+++ b/src/gpgpu-sim/gpu-cache.h
@@ -563,10 +563,10 @@ class cache_config {
     char ct, rp, wp, ap, mshr_type, wap, sif;
 
     int ntok =
-        sscanf(config, "%c:%u:%u:%u,%c:%c:%c:%c:%c,%c:%u:%u,%u:%u,%u,%u", &ct,
+        sscanf(config, "%c:%u:%u:%u,%c:%c:%c:%c:%c,%c:%u:%u,%u:%u,%u", &ct,
                &m_nset, &m_line_sz, &m_assoc, &rp, &wp, &ap, &wap, &sif,
                &mshr_type, &m_mshr_entries, &m_mshr_max_merge,
-               &m_miss_queue_size, &m_result_fifo_entries, &m_data_port_width, &m_wr_percent);
+               &m_miss_queue_size, &m_result_fifo_entries, &m_data_port_width);
 
     if (ntok < 12) {
       if (!strcmp(config, "none")) {
diff --git a/src/gpgpu-sim/l2cache.cc b/src/gpgpu-sim/l2cache.cc
index 00b14d7..0db6bd4 100644
--- a/src/gpgpu-sim/l2cache.cc
+++ b/src/gpgpu-sim/l2cache.cc
@@ -57,18 +57,19 @@ mem_fetch *partition_mf_allocator::alloc(new_addr_type addr,
   return mf;
 }
 
-mem_fetch *partition_mf_allocator::alloc(new_addr_type addr, 
-                            mem_access_type type,
+mem_fetch *partition_mf_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 &sector_mask,
                             unsigned size, bool wr,
-                            unsigned long long cycle) const {
+                            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, -1,
-                  -1, -1, m_memory_config, cycle);
+    new mem_fetch(access, NULL, wr ? WRITE_PACKET_SIZE : READ_PACKET_SIZE, wid,
+                  sid, tpc, m_memory_config, cycle,original_mf);
     return mf;
 }
 memory_partition_unit::memory_partition_unit(unsigned partition_id,
@@ -724,16 +725,11 @@ memory_sub_partition::breakdown_request_to_sector_requests(mem_fetch *mf) {
       for (unsigned k = i * SECTOR_SIZE; k < (i + 1) * SECTOR_SIZE; k++) {
         mask.set(k);
       }
-      const mem_access_t *ma = new mem_access_t(
-          mf->get_access_type(), mf->get_addr() + SECTOR_SIZE * i, SECTOR_SIZE,
-          mf->is_write(), mf->get_access_warp_mask(),
-          mf->get_access_byte_mask() & mask,
-          std::bitset<SECTOR_CHUNCK_SIZE>().set(i), m_gpu->gpgpu_ctx);
-
-      mem_fetch *n_mf =
-          new mem_fetch(*ma, NULL, mf->get_ctrl_size(), mf->get_wid(),
-                        mf->get_sid(), mf->get_tpc(), mf->get_mem_config(),
-                        m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle, mf);
+      mem_fetch *n_mf = m_mf_allocator->alloc(mf->get_addr() + SECTOR_SIZE * i, 
+        mf->get_access_type(),mf->get_access_warp_mask(), 
+        mf->get_access_byte_mask() & mask,std::bitset<SECTOR_CHUNCK_SIZE>().set(i), 
+        SECTOR_SIZE,mf->is_write(),m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle, 
+        mf->get_wid(),mf->get_sid(), mf->get_tpc(),mf);
 
       result.push_back(n_mf);
     }
@@ -750,16 +746,11 @@ memory_sub_partition::breakdown_request_to_sector_requests(mem_fetch *mf) {
       for (unsigned k = i * SECTOR_SIZE; k < (i + 1) * SECTOR_SIZE; k++) {
         mask.set(k);
       }
-      const mem_access_t *ma = new mem_access_t(
-          mf->get_access_type(), mf->get_addr(), SECTOR_SIZE,
-          mf->is_write(), mf->get_access_warp_mask(),
-          mf->get_access_byte_mask() & mask,
-          std::bitset<SECTOR_CHUNCK_SIZE>().set(i), m_gpu->gpgpu_ctx);
-
-      mem_fetch *n_mf =
-          new mem_fetch(*ma, NULL, mf->get_ctrl_size(), mf->get_wid(),
-                        mf->get_sid(), mf->get_tpc(), mf->get_mem_config(),
-                        m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle, mf);
+      mem_fetch *n_mf = m_mf_allocator->alloc(mf->get_addr(), 
+        mf->get_access_type(),mf->get_access_warp_mask(), 
+        mf->get_access_byte_mask() & mask,std::bitset<SECTOR_CHUNCK_SIZE>().set(i), 
+        SECTOR_SIZE,mf->is_write(),m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle, 
+        mf->get_wid(),mf->get_sid(), mf->get_tpc(),mf);
 
       result.push_back(n_mf);
     }
@@ -770,16 +761,11 @@ memory_sub_partition::breakdown_request_to_sector_requests(mem_fetch *mf) {
         for (unsigned k = i * SECTOR_SIZE; k < (i + 1) * SECTOR_SIZE; k++) {
           mask.set(k);
         }
-        const mem_access_t *ma = new mem_access_t(
-            mf->get_access_type(), mf->get_addr() + SECTOR_SIZE * i,
-            SECTOR_SIZE, mf->is_write(), mf->get_access_warp_mask(),
-            mf->get_access_byte_mask() & mask,
-            std::bitset<SECTOR_CHUNCK_SIZE>().set(i), m_gpu->gpgpu_ctx);
-
-        mem_fetch *n_mf =
-            new mem_fetch(*ma, NULL, mf->get_ctrl_size(), mf->get_wid(),
-                          mf->get_sid(), mf->get_tpc(), mf->get_mem_config(),
-                          m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle, mf);
+        mem_fetch *n_mf = m_mf_allocator->alloc(mf->get_addr() + SECTOR_SIZE * i, 
+          mf->get_access_type(),mf->get_access_warp_mask(), 
+          mf->get_access_byte_mask() & mask,std::bitset<SECTOR_CHUNCK_SIZE>().set(i), 
+          SECTOR_SIZE,mf->is_write(),m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle, 
+          mf->get_wid(),mf->get_sid(), mf->get_tpc(),mf);
 
         result.push_back(n_mf);
       }
diff --git a/src/gpgpu-sim/l2cache.h b/src/gpgpu-sim/l2cache.h
index 1f5d7c4..59432b8 100644
--- a/src/gpgpu-sim/l2cache.h
+++ b/src/gpgpu-sim/l2cache.h
@@ -56,7 +56,9 @@ class partition_mf_allocator : public mem_fetch_allocator {
                             const mem_access_byte_mask_t &byte_mask,
                             const mem_access_sector_mask_t &sector_mask,
                             unsigned size, bool wr,
-                            unsigned long long cycle) const;
+                            unsigned long long cycle,
+                            unsigned wid, unsigned sid,
+                            unsigned tpc, mem_fetch *original_mf) const;
 
  private:
   const memory_config *m_memory_config;
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 7f27b7b..51366de 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -62,18 +62,19 @@ mem_fetch *shader_core_mem_fetch_allocator::alloc(
   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 &sector_mask,
-  unsigned size, bool wr,
-  unsigned long long cycle) const {
+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 &sector_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, -1,
-                    m_core_id, m_cluster_id, m_memory_config, cycle);
+      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;
   }
 /////////////////////////////////////////////////////////////////////////////
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 42bbdcb..8662313 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -1903,7 +1903,9 @@ class shader_core_mem_fetch_allocator : public mem_fetch_allocator {
                             const mem_access_byte_mask_t &byte_mask,
                             const mem_access_sector_mask_t &sector_mask,
                             unsigned size, bool wr,
-                            unsigned long long cycle) const;
+                            unsigned long long cycle,
+                            unsigned wid, unsigned sid,
+                            unsigned tpc, mem_fetch *original_mf) const;
   mem_fetch *alloc(const warp_inst_t &inst, const mem_access_t &access,
                    unsigned long long cycle) const {
     warp_inst_t inst_copy = inst;