stats: Cumulative stats update

This patch updates the stats to reflect the: 1) addition of the
internal queue in SimpleMemory, 2) moving of the memory class outside
FSConfig, 3) fixing up of the 2D vector printing format, 4) specifying
burst size and interface width for the DRAM instead of relying on
cache-line size, 5) performing merging in the DRAM controller write
buffer, and 6) fixing how idle cycles are counted in the atomic and
timing CPU models.

The main reason for bundling them up is to minimise the changeset
size.

4 files changed, 178 insertions, 176 deletions

diff --git a/tests/long/se/40.perlbmk/ref/alpha/tru64/o3-timing/stats.txt b/tests/long/se/40.perlbmk/ref/alpha/tru64/o3-timing/stats.txt
index fb55fbe0e..669d3dfc7 100644
--- a/tests/long/se/40.perlbmk/ref/alpha/tru64/o3-timing/stats.txt
+++ b/tests/long/se/40.perlbmk/ref/alpha/tru64/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.631883                       # Nu
 sim_ticks                                631883288500                       # Number of ticks simulated
 final_tick                               631883288500                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 129491                       # Simulator instruction rate (inst/s)
-host_op_rate                                   129491                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               44882876                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 237188                       # Number of bytes of host memory used
-host_seconds                                 14078.49                       # Real time elapsed on the host
+host_inst_rate                                 177291                       # Simulator instruction rate (inst/s)
+host_op_rate                                   177291                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                               61450789                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 236780                       # Number of bytes of host memory used
+host_seconds                                 10282.75                       # Real time elapsed on the host
 sim_insts                                  1823043370                       # Number of instructions simulated
 sim_ops                                    1823043370                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            176064                       # Number of bytes read from this memory
@@ -34,14 +34,15 @@ system.physmem.bw_total::writebacks           6776745                       # To
 system.physmem.bw_total::cpu.inst              278634                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data            47944246                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total               54999625                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                        476114                       # Total number of read requests seen
-system.physmem.writeReqs                        66908                       # Total number of write requests seen
-system.physmem.cpureqs                         543022                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                        476114                       # Total number of read requests accepted by DRAM controller
+system.physmem.writeReqs                        66908                       # Total number of write requests accepted by DRAM controller
+system.physmem.readBursts                      476114                       # Total number of DRAM read bursts. Each DRAM read request translates to either one or multiple DRAM read bursts
+system.physmem.writeBursts                      66908                       # Total number of DRAM write bursts. Each DRAM write request translates to either one or multiple DRAM write bursts
 system.physmem.bytesRead                     30471232                       # Total number of bytes read from memory
 system.physmem.bytesWritten                   4282112                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd               30471232                       # bytesRead derated as per pkt->getSize()
 system.physmem.bytesConsumedWr                4282112                       # bytesWritten derated as per pkt->getSize()
-system.physmem.servicedByWrQ                       90                       # Number of read reqs serviced by write Q
+system.physmem.servicedByWrQ                       90                       # Number of DRAM read bursts serviced by write Q
 system.physmem.neitherReadNorWrite                  0                       # Reqs where no action is needed
 system.physmem.perBankRdReqs::0                 29447                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                 29799                       # Track reads on a per bank basis
@@ -262,10 +263,10 @@ system.membus.trans_dist::ReadResp             409257                       # Tr
 system.membus.trans_dist::Writeback             66908                       # Transaction distribution
 system.membus.trans_dist::ReadExReq             66856                       # Transaction distribution
 system.membus.trans_dist::ReadExResp            66856                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side      1019135                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                       1019135                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     34753344                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   34753344                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port      1019135                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                1019135                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     34753344                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            34753344                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               34753344                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1232718500                       # Layer occupancy (ticks)
@@ -581,12 +582,12 @@ system.cpu.toL2Bus.trans_dist::ReadResp       1470294                       # Tr
 system.cpu.toL2Bus.trans_dist::Writeback        95986                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq        71645                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp        71645                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side        20089                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side      3159776                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                  3179865                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side       642816                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side    104184384                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size             104827200                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        20089                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side      3159776                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total           3179865                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       642816                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side    104184384                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total      104827200                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus         104827200                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy      914949000                       # Layer occupancy (ticks)
@@ -595,15 +596,15 @@ system.cpu.toL2Bus.respLayer0.occupancy      15605000                       # La
 system.cpu.toL2Bus.respLayer0.utilization          0.0                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy    2398320750                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          0.4                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                   8334                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1655.074457                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                394735107                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  10044                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs               39300.588112                       # Average number of references to valid blocks.
-system.cpu.icache.tags.warmup_cycle                      0                       # Cycle when the warmup percentage was hit.
-system.cpu.icache.tags.occ_blocks::cpu.inst    1655.074457                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.808142                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.808142                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements              8334                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1655.074457                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs           394735107                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             10044                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs          39300.588112                       # Average number of references to valid blocks.
+system.cpu.icache.tags.warmup_cycle                 0                       # Cycle when the warmup percentage was hit.
+system.cpu.icache.tags.occ_blocks::cpu.inst  1655.074457                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.808142                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.808142                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst    394735107                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total       394735107                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst     394735107                       # number of demand (read+write) hits
@@ -679,19 +680,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 27883.100946
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 27883.100946                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 27883.100946                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                443335                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             32690.569488                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                 1090072                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                476070                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  2.289731                       # Average number of references to valid blocks.
-system.cpu.l2cache.tags.warmup_cycle                     0                       # Cycle when the warmup percentage was hit.
+system.cpu.l2cache.tags.replacements           443335                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        32690.569488                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs            1090072                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           476070                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             2.289731                       # Average number of references to valid blocks.
+system.cpu.l2cache.tags.warmup_cycle                0                       # Cycle when the warmup percentage was hit.
 system.cpu.l2cache.tags.occ_blocks::writebacks  1328.456107                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst     35.162790                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data  31326.950592                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst    35.162790                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data 31326.950592                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.040541                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.001073                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.956023                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.997637                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.997637                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst         7293                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data      1053744                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total        1061037                       # number of ReadReq hits
@@ -817,15 +818,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 58975.018169
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 60811.242242                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 60800.628631                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                1527799                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4094.613876                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                667806397                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                1531895                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 435.934837                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle              399882250                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4094.613876                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.999662                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.999662                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements           1527799                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4094.613876                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs           667806397                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs           1531895                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            435.934837                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle         399882250                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4094.613876                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.999662                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.999662                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data    458073360                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total       458073360                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data    209733012                       # number of WriteReq hits
diff --git a/tests/long/se/40.perlbmk/ref/alpha/tru64/simple-timing/stats.txt b/tests/long/se/40.perlbmk/ref/alpha/tru64/simple-timing/stats.txt
index 3b39f56f2..421623453 100644
--- a/tests/long/se/40.perlbmk/ref/alpha/tru64/simple-timing/stats.txt
+++ b/tests/long/se/40.perlbmk/ref/alpha/tru64/simple-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  2.769740                       # Nu
 sim_ticks                                2769739533000                       # Number of ticks simulated
 final_tick                               2769739533000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                1559352                       # Simulator instruction rate (inst/s)
-host_op_rate                                  1559352                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                             2149839105                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 233980                       # Number of bytes of host memory used
-host_seconds                                  1288.35                       # Real time elapsed on the host
+host_inst_rate                                 892879                       # Simulator instruction rate (inst/s)
+host_op_rate                                   892879                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                             1230989339                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 233984                       # Number of bytes of host memory used
+host_seconds                                  2250.01                       # Real time elapsed on the host
 sim_insts                                  2008987605                       # Number of instructions simulated
 sim_ops                                    2008987605                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            137792                       # Number of bytes read from this memory
@@ -40,10 +40,10 @@ system.membus.trans_dist::ReadResp             408476                       # Tr
 system.membus.trans_dist::Writeback             66908                       # Transaction distribution
 system.membus.trans_dist::ReadExReq             66873                       # Transaction distribution
 system.membus.trans_dist::ReadExResp            66873                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side      1017606                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                       1017606                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     34704448                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   34704448                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port      1017606                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                1017606                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     34704448                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            34704448                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               34704448                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1077521000                       # Layer occupancy (ticks)
@@ -105,15 +105,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                 5539479066                       # Number of busy cycles
 system.cpu.not_idle_fraction                        1                       # Percentage of non-idle cycles
 system.cpu.idle_fraction                            0                       # Percentage of idle cycles
-system.cpu.icache.tags.replacements                   9046                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1478.418050                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs               2009410475                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  10596                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs               189638.587675                       # Average number of references to valid blocks.
-system.cpu.icache.tags.warmup_cycle                      0                       # Cycle when the warmup percentage was hit.
-system.cpu.icache.tags.occ_blocks::cpu.inst    1478.418050                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.721884                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.721884                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements              9046                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1478.418050                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs          2009410475                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             10596                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs          189638.587675                       # Average number of references to valid blocks.
+system.cpu.icache.tags.warmup_cycle                 0                       # Cycle when the warmup percentage was hit.
+system.cpu.icache.tags.occ_blocks::cpu.inst  1478.418050                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.721884                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.721884                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst   2009410475                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total      2009410475                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst    2009410475                       # number of demand (read+write) hits
@@ -183,19 +183,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 19533.975085
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 19533.975085                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 19533.975085                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                442570                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             32706.854192                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                 1089464                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                475302                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  2.292151                       # Average number of references to valid blocks.
-system.cpu.l2cache.tags.warmup_cycle                     0                       # Cycle when the warmup percentage was hit.
+system.cpu.l2cache.tags.replacements           442570                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        32706.854192                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs            1089464                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           475302                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             2.292151                       # Average number of references to valid blocks.
+system.cpu.l2cache.tags.warmup_cycle                0                       # Cycle when the warmup percentage was hit.
 system.cpu.l2cache.tags.occ_blocks::writebacks  1300.510334                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst     26.518402                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data  31379.825456                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst    26.518402                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data 31379.825456                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.039688                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.000809                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.957636                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.998134                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.998134                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst         8443                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data      1051869                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total        1060312                       # number of ReadReq hits
@@ -321,15 +321,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst        40000
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 40000.006340                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 40000.006311                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                1526048                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4095.197836                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                720334778                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                1530144                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 470.762737                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle             1041395000                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4095.197836                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.999804                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.999804                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements           1526048                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4095.197836                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs           720334778                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs           1530144                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            470.762737                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle        1041395000                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4095.197836                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.999804                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.999804                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data    509611834                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total       509611834                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data    210722944                       # number of WriteReq hits
@@ -427,12 +427,12 @@ system.cpu.toL2Bus.trans_dist::ReadResp       1468788                       # Tr
 system.cpu.toL2Bus.trans_dist::Writeback        96129                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq        71952                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp        71952                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side        21192                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side      3156417                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                  3177609                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side       678144                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side    104081472                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size             104759616                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        21192                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side      3156417                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total           3177609                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       678144                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side    104081472                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total      104759616                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus         104759616                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy      914563500                       # Layer occupancy (ticks)
diff --git a/tests/long/se/40.perlbmk/ref/arm/linux/o3-timing/stats.txt b/tests/long/se/40.perlbmk/ref/arm/linux/o3-timing/stats.txt
index 68bfe2c31..5f69a496c 100644
--- a/tests/long/se/40.perlbmk/ref/arm/linux/o3-timing/stats.txt
+++ b/tests/long/se/40.perlbmk/ref/arm/linux/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.640648                       # Nu
 sim_ticks                                640648369500                       # Number of ticks simulated
 final_tick                               640648369500                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  99606                       # Simulator instruction rate (inst/s)
-host_op_rate                                   135651                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               46095119                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 254320                       # Number of bytes of host memory used
-host_seconds                                 13898.40                       # Real time elapsed on the host
+host_inst_rate                                 107808                       # Simulator instruction rate (inst/s)
+host_op_rate                                   146820                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                               49890421                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 251704                       # Number of bytes of host memory used
+host_seconds                                 12841.11                       # Real time elapsed on the host
 sim_insts                                  1384370590                       # Number of instructions simulated
 sim_ops                                    1885325342                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            155648                       # Number of bytes read from this memory
@@ -34,14 +34,15 @@ system.physmem.bw_total::writebacks           6603111                       # To
 system.physmem.bw_total::cpu.inst              242954                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data            47208174                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total               54054239                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                        474992                       # Total number of read requests seen
-system.physmem.writeReqs                        66098                       # Total number of write requests seen
-system.physmem.cpureqs                         545451                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                        474992                       # Total number of read requests accepted by DRAM controller
+system.physmem.writeReqs                        66098                       # Total number of write requests accepted by DRAM controller
+system.physmem.readBursts                      474992                       # Total number of DRAM read bursts. Each DRAM read request translates to either one or multiple DRAM read bursts
+system.physmem.writeBursts                      66098                       # Total number of DRAM write bursts. Each DRAM write request translates to either one or multiple DRAM write bursts
 system.physmem.bytesRead                     30399488                       # Total number of bytes read from memory
 system.physmem.bytesWritten                   4230272                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd               30399488                       # bytesRead derated as per pkt->getSize()
 system.physmem.bytesConsumedWr                4230272                       # bytesWritten derated as per pkt->getSize()
-system.physmem.servicedByWrQ                      152                       # Number of read reqs serviced by write Q
+system.physmem.servicedByWrQ                      152                       # Number of DRAM read bursts serviced by write Q
 system.physmem.neitherReadNorWrite               4361                       # Reqs where no action is needed
 system.physmem.perBankRdReqs::0                 29873                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                 29675                       # Track reads on a per bank basis
@@ -244,10 +245,10 @@ system.membus.trans_dist::UpgradeReq             4361                       # Tr
 system.membus.trans_dist::UpgradeResp            4361                       # Transaction distribution
 system.membus.trans_dist::ReadExReq             66075                       # Transaction distribution
 system.membus.trans_dist::ReadExResp            66075                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side      1024803                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                       1024803                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     34629696                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   34629696                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port      1024803                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                1024803                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     34629696                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            34629696                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               34629696                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1215067500                       # Layer occupancy (ticks)
@@ -576,12 +577,12 @@ system.cpu.toL2Bus.trans_dist::UpgradeReq         4364                       # T
 system.cpu.toL2Bus.trans_dist::UpgradeResp         4364                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq        72519                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp        72519                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side        52387                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side      3178835                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                  3231222                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side      1536768                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side    104525120                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size             106061888                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        52387                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side      3178835                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total           3231222                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side      1536768                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side    104525120                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total      106061888                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus         106061888                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus       279232                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy      929276999                       # Layer occupancy (ticks)
@@ -590,15 +591,15 @@ system.cpu.toL2Bus.respLayer0.occupancy      43029998                       # La
 system.cpu.toL2Bus.respLayer0.utilization          0.0                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy    2407943085                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          0.4                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                  22329                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1638.931929                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                345969528                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  24011                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs               14408.792970                       # Average number of references to valid blocks.
-system.cpu.icache.tags.warmup_cycle                      0                       # Cycle when the warmup percentage was hit.
-system.cpu.icache.tags.occ_blocks::cpu.inst    1638.931929                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.800260                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.800260                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements             22329                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1638.931929                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs           345969528                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             24011                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs          14408.792970                       # Average number of references to valid blocks.
+system.cpu.icache.tags.warmup_cycle                 0                       # Cycle when the warmup percentage was hit.
+system.cpu.icache.tags.occ_blocks::cpu.inst  1638.931929                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.800260                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.800260                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst    345973619                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total       345973619                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst     345973619                       # number of demand (read+write) hits
@@ -674,19 +675,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 14882.555031
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 14882.555031                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 14882.555031                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                442208                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             32680.533022                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                 1109569                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                474957                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  2.336146                       # Average number of references to valid blocks.
-system.cpu.l2cache.tags.warmup_cycle                     0                       # Cycle when the warmup percentage was hit.
+system.cpu.l2cache.tags.replacements           442208                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        32680.533022                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs            1109569                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           474957                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             2.336146                       # Average number of references to valid blocks.
+system.cpu.l2cache.tags.warmup_cycle                0                       # Cycle when the warmup percentage was hit.
 system.cpu.l2cache.tags.occ_blocks::writebacks  1291.826262                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst     50.114345                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data  31338.592416                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst    50.114345                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data 31338.592416                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.039423                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.001529                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.956378                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.997331                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.997331                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst        21578                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data      1057872                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total        1079450                       # number of ReadReq hits
@@ -837,15 +838,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 58784.539474
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 62076.363848                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 62059.509423                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                1532805                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4094.435174                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                972917364                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                1536901                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 633.038409                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle              392115250                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4094.435174                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.999618                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.999618                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements           1532805                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4094.435174                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs           972917364                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs           1536901                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            633.038409                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle         392115250                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4094.435174                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.999618                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.999618                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data    696790485                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total       696790485                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data    276093216                       # number of WriteReq hits
diff --git a/tests/long/se/40.perlbmk/ref/arm/linux/simple-timing/stats.txt b/tests/long/se/40.perlbmk/ref/arm/linux/simple-timing/stats.txt
index b119e8929..65ea4c6ca 100644
--- a/tests/long/se/40.perlbmk/ref/arm/linux/simple-timing/stats.txt
+++ b/tests/long/se/40.perlbmk/ref/arm/linux/simple-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  2.326119                       # Nu
 sim_ticks                                2326118592000                       # Number of ticks simulated
 final_tick                               2326118592000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 575384                       # Simulator instruction rate (inst/s)
-host_op_rate                                   780549                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                              968736790                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 250996                       # Number of bytes of host memory used
-host_seconds                                  2401.19                       # Real time elapsed on the host
+host_inst_rate                                 481372                       # Simulator instruction rate (inst/s)
+host_op_rate                                   653016                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              810455855                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 248632                       # Number of bytes of host memory used
+host_seconds                                  2870.14                       # Real time elapsed on the host
 sim_insts                                  1381604339                       # Number of instructions simulated
 sim_ops                                    1874244941                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            113472                       # Number of bytes read from this memory
@@ -40,10 +40,10 @@ system.membus.trans_dist::ReadResp             408063                       # Tr
 system.membus.trans_dist::Writeback             66099                       # Transaction distribution
 system.membus.trans_dist::ReadExReq             66093                       # Transaction distribution
 system.membus.trans_dist::ReadExResp            66093                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side      1014411                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                       1014411                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     34576320                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   34576320                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port      1014411                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                1014411                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     34576320                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            34576320                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               34576320                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1069047000                       # Layer occupancy (ticks)
@@ -115,15 +115,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                 4652237184                       # Number of busy cycles
 system.cpu.not_idle_fraction                        1                       # Percentage of non-idle cycles
 system.cpu.idle_fraction                            0                       # Percentage of idle cycles
-system.cpu.icache.tags.replacements                  18364                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1392.317060                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs               1390251699                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  19803                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs               70204.095289                       # Average number of references to valid blocks.
-system.cpu.icache.tags.warmup_cycle                      0                       # Cycle when the warmup percentage was hit.
-system.cpu.icache.tags.occ_blocks::cpu.inst    1392.317060                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.679842                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.679842                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements             18364                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1392.317060                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs          1390251699                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             19803                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs          70204.095289                       # Average number of references to valid blocks.
+system.cpu.icache.tags.warmup_cycle                 0                       # Cycle when the warmup percentage was hit.
+system.cpu.icache.tags.occ_blocks::cpu.inst  1392.317060                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.679842                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.679842                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst   1390251699                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total      1390251699                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst    1390251699                       # number of demand (read+write) hits
@@ -193,19 +193,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 14760.642327
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 14760.642327                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 14760.642327                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                441378                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             32692.891822                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                 1102614                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                474121                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  2.325596                       # Average number of references to valid blocks.
-system.cpu.l2cache.tags.warmup_cycle                     0                       # Cycle when the warmup percentage was hit.
+system.cpu.l2cache.tags.replacements           441378                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        32692.891822                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs            1102614                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           474121                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             2.325596                       # Average number of references to valid blocks.
+system.cpu.l2cache.tags.warmup_cycle                0                       # Cycle when the warmup percentage was hit.
 system.cpu.l2cache.tags.occ_blocks::writebacks  1298.141733                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst     30.233408                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data  31364.516681                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst    30.233408                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data 31364.516681                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.039616                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.000923                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.957169                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.997708                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.997708                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst        18030                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data      1054583                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total        1072613                       # number of ReadReq hits
@@ -331,15 +331,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 40003.384095
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data        40000                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 40000.012654                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                1529557                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4094.947189                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                895757408                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                1533653                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 584.067848                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle              991199000                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4094.947189                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.999743                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.999743                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements           1529557                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4094.947189                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs           895757408                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs           1533653                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            584.067848                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle         991199000                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4094.947189                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.999743                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.999743                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data    618874540                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total       618874540                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data    276862898                       # number of WriteReq hits
@@ -445,12 +445,12 @@ system.cpu.toL2Bus.trans_dist::ReadResp       1480676                       # Tr
 system.cpu.toL2Bus.trans_dist::Writeback        96257                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq        72780                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp        72780                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side        39606                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side      3163563                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                  3203169                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side      1267392                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side    104314240                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size             105581632                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        39606                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side      3163563                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total           3203169                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side      1267392                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side    104314240                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total      105581632                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus         105581632                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy      921113500                       # Layer occupancy (ticks)