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, 170 insertions, 172 deletions

diff --git a/tests/long/se/20.parser/ref/arm/linux/o3-timing/stats.txt b/tests/long/se/20.parser/ref/arm/linux/o3-timing/stats.txt
index d43c28cd4..d58d3f98b 100644
--- a/tests/long/se/20.parser/ref/arm/linux/o3-timing/stats.txt
+++ b/tests/long/se/20.parser/ref/arm/linux/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.202350                       # Nu
 sim_ticks                                202349747500                       # Number of ticks simulated
 final_tick                               202349747500                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 166059                       # Simulator instruction rate (inst/s)
-host_op_rate                                   187221                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               66507382                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 250660                       # Number of bytes of host memory used
-host_seconds                                  3042.52                       # Real time elapsed on the host
+host_inst_rate                                  95439                       # Simulator instruction rate (inst/s)
+host_op_rate                                   107602                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                               38223736                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 246676                       # Number of bytes of host memory used
+host_seconds                                  5293.82                       # Real time elapsed on the host
 sim_insts                                   505237723                       # Number of instructions simulated
 sim_ops                                     569624283                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            216896                       # Number of bytes read from this memory
@@ -34,14 +34,15 @@ system.physmem.bw_total::writebacks          30890515                       # To
 system.physmem.bw_total::cpu.inst             1071887                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data            45802993                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total               77765395                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                        148206                       # Total number of read requests seen
-system.physmem.writeReqs                        97667                       # Total number of write requests seen
-system.physmem.cpureqs                         245886                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                        148206                       # Total number of read requests accepted by DRAM controller
+system.physmem.writeReqs                        97667                       # Total number of write requests accepted by DRAM controller
+system.physmem.readBursts                      148206                       # Total number of DRAM read bursts. Each DRAM read request translates to either one or multiple DRAM read bursts
+system.physmem.writeBursts                      97667                       # Total number of DRAM write bursts. Each DRAM write request translates to either one or multiple DRAM write bursts
 system.physmem.bytesRead                      9485120                       # Total number of bytes read from memory
 system.physmem.bytesWritten                   6250688                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                9485120                       # bytesRead derated as per pkt->getSize()
 system.physmem.bytesConsumedWr                6250688                       # bytesWritten derated as per pkt->getSize()
-system.physmem.servicedByWrQ                       73                       # Number of read reqs serviced by write Q
+system.physmem.servicedByWrQ                       73                       # Number of DRAM read bursts serviced by write Q
 system.physmem.neitherReadNorWrite                  7                       # Reqs where no action is needed
 system.physmem.perBankRdReqs::0                  9580                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                  9220                       # Track reads on a per bank basis
@@ -297,10 +298,10 @@ system.membus.trans_dist::UpgradeReq                7                       # Tr
 system.membus.trans_dist::UpgradeResp               7                       # Transaction distribution
 system.membus.trans_dist::ReadExReq            101306                       # Transaction distribution
 system.membus.trans_dist::ReadExResp           101306                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side       394092                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                        394092                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     15735808                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   15735808                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port       394092                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                 394092                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     15735808                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            15735808                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               15735808                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1084180500                       # Layer occupancy (ticks)
@@ -628,12 +629,12 @@ system.cpu.toL2Bus.trans_dist::UpgradeReq           69                       # T
 system.cpu.toL2Bus.trans_dist::UpgradeResp           69                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq       348843                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp       348843                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side        33804                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side      3504826                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                  3538630                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side      1079232                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side    147703872                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size             148783104                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        33804                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side      3504826                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total           3538630                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side      1079232                       # 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    147703872                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total      148783104                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus         148783104                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus         4928                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy     2273504243                       # Layer occupancy (ticks)
@@ -642,15 +643,15 @@ system.cpu.toL2Bus.respLayer0.occupancy      26125731                       # La
 system.cpu.toL2Bus.respLayer0.utilization          0.0                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy    1828577727                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          0.9                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                  15008                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1099.436561                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                114505770                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  16868                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                6788.343016                       # 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    1099.436561                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.536834                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.536834                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements             15008                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1099.436561                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs           114505770                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             16868                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs           6788.343016                       # 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  1099.436561                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.536834                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.536834                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst    114505770                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total       114505770                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst     114505770                       # number of demand (read+write) hits
@@ -726,19 +727,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 25103.227023
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 25103.227023                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 25103.227023                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                115462                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             27105.054655                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                 1782175                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                146717                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                 12.147025                       # Average number of references to valid blocks.
-system.cpu.l2cache.tags.warmup_cycle          102215583000                       # Cycle when the warmup percentage was hit.
+system.cpu.l2cache.tags.replacements           115462                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        27105.054655                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs            1782175                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           146717                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs            12.147025                       # Average number of references to valid blocks.
+system.cpu.l2cache.tags.warmup_cycle     102215583000                       # Cycle when the warmup percentage was hit.
 system.cpu.l2cache.tags.occ_blocks::writebacks 23019.815136                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst    365.213065                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data   3720.026454                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst   365.213065                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data  3720.026454                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.702509                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.011145                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.113526                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.827181                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.827181                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst        13469                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data       804438                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total         817907                       # number of ReadReq hits
@@ -889,15 +890,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 67880.014749
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 61549.679948                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 61694.476314                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                1192719                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4057.784175                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                190184088                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                1196815                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 158.908510                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle             4223544250                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4057.784175                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.990670                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.990670                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements           1192719                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4057.784175                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs           190184088                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs           1196815                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            158.908510                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle        4223544250                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4057.784175                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.990670                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.990670                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data    136217061                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total       136217061                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data     50989456                       # number of WriteReq hits
diff --git a/tests/long/se/20.parser/ref/arm/linux/simple-timing/stats.txt b/tests/long/se/20.parser/ref/arm/linux/simple-timing/stats.txt
index b28088e7d..3138d4062 100644
--- a/tests/long/se/20.parser/ref/arm/linux/simple-timing/stats.txt
+++ b/tests/long/se/20.parser/ref/arm/linux/simple-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.717366                       # Nu
 sim_ticks                                717366012000                       # Number of ticks simulated
 final_tick                               717366012000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 611042                       # Simulator instruction rate (inst/s)
-host_op_rate                                   688541                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                              868024183                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 246240                       # Number of bytes of host memory used
-host_seconds                                   826.44                       # Real time elapsed on the host
+host_inst_rate                                1130634                       # Simulator instruction rate (inst/s)
+host_op_rate                                  1274033                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                             1606137434                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 243872                       # Number of bytes of host memory used
+host_seconds                                   446.64                       # Real time elapsed on the host
 sim_insts                                   504986853                       # Number of instructions simulated
 sim_ops                                     569034839                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            177280                       # Number of bytes read from this memory
@@ -40,10 +40,10 @@ system.membus.trans_dist::ReadResp              41855                       # Tr
 system.membus.trans_dist::Writeback             95953                       # Transaction distribution
 system.membus.trans_dist::ReadExReq            100794                       # Transaction distribution
 system.membus.trans_dist::ReadExResp           100794                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side       381251                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                        381251                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     15270528                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   15270528                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port       381251                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                 381251                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     15270528                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            15270528                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               15270528                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1006226000                       # Layer occupancy (ticks)
@@ -115,15 +115,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                 1434732024                       # 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                   9788                       # number of replacements
-system.cpu.icache.tags.tagsinuse                982.663229                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                516599855                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  11521                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs               44839.845066                       # 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     982.663229                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.479816                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.479816                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements              9788                       # number of replacements
+system.cpu.icache.tags.tagsinuse           982.663229                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs           516599855                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             11521                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs          44839.845066                       # 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   982.663229                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.479816                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.479816                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst    516599855                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total       516599855                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst     516599855                       # number of demand (read+write) hits
@@ -193,19 +193,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 21105.199201
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 21105.199201                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 21105.199201                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                109895                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             27243.192324                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                 1668833                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                141072                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                 11.829654                       # Average number of references to valid blocks.
-system.cpu.l2cache.tags.warmup_cycle          343698539000                       # Cycle when the warmup percentage was hit.
+system.cpu.l2cache.tags.replacements           109895                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        27243.192324                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs            1668833                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           141072                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs            11.829654                       # Average number of references to valid blocks.
+system.cpu.l2cache.tags.warmup_cycle     343698539000                       # Cycle when the warmup percentage was hit.
 system.cpu.l2cache.tags.occ_blocks::writebacks 23381.854289                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst    287.865470                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data   3573.472565                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst   287.865470                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data  3573.472565                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.713558                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.008785                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.109054                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.831396                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.831396                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst         8751                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data       743573                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total         752324                       # number of ReadReq hits
@@ -331,15 +331,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 40029.602888
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 40009.472473                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 40009.863371                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                1134822                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4065.297446                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                179817786                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                1138918                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 157.884752                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle            11885124000                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4065.297446                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.992504                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.992504                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements           1134822                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4065.297446                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs           179817786                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs           1138918                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            157.884752                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle       11885124000                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4065.297446                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.992504                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.992504                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data    122957658                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total       122957658                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data     53883046                       # number of WriteReq hits
@@ -445,12 +445,12 @@ system.cpu.toL2Bus.trans_dist::ReadResp        794179                       # Tr
 system.cpu.toL2Bus.trans_dist::Writeback      1064905                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq       356260                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp       356260                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side        23042                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side      3342741                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                  3365783                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side       737344                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side    141044672                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size             141782016                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        23042                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side      3342741                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total           3365783                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       737344                       # 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    141044672                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total      141782016                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus         141782016                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy     2172577000                       # Layer occupancy (ticks)
diff --git a/tests/long/se/20.parser/ref/x86/linux/o3-timing/stats.txt b/tests/long/se/20.parser/ref/x86/linux/o3-timing/stats.txt
index d91a5905c..4d8b3de9b 100644
--- a/tests/long/se/20.parser/ref/x86/linux/o3-timing/stats.txt
+++ b/tests/long/se/20.parser/ref/x86/linux/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.458202                       # Nu
 sim_ticks                                458201684000                       # Number of ticks simulated
 final_tick                               458201684000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 111882                       # Simulator instruction rate (inst/s)
-host_op_rate                                   206882                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               61997502                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 341328                       # Number of bytes of host memory used
-host_seconds                                  7390.65                       # Real time elapsed on the host
+host_inst_rate                                  77434                       # Simulator instruction rate (inst/s)
+host_op_rate                                   143185                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                               42909026                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 338808                       # Number of bytes of host memory used
+host_seconds                                 10678.45                       # Real time elapsed on the host
 sim_insts                                   826877109                       # Number of instructions simulated
 sim_ops                                    1528988701                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            201408                       # Number of bytes read from this memory
@@ -34,14 +34,15 @@ system.physmem.bw_total::writebacks          41005663                       # To
 system.physmem.bw_total::cpu.inst              439562                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data            53417735                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total               94862960                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                        385586                       # Total number of read requests seen
-system.physmem.writeReqs                       293576                       # Total number of write requests seen
-system.physmem.cpureqs                         810414                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                        385586                       # Total number of read requests accepted by DRAM controller
+system.physmem.writeReqs                       293576                       # Total number of write requests accepted by DRAM controller
+system.physmem.readBursts                      385586                       # Total number of DRAM read bursts. Each DRAM read request translates to either one or multiple DRAM read bursts
+system.physmem.writeBursts                     293576                       # Total number of DRAM write bursts. Each DRAM write request translates to either one or multiple DRAM write bursts
 system.physmem.bytesRead                     24677504                       # Total number of bytes read from memory
 system.physmem.bytesWritten                  18788864                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd               24677504                       # bytesRead derated as per pkt->getSize()
 system.physmem.bytesConsumedWr               18788864                       # bytesWritten derated as per pkt->getSize()
-system.physmem.servicedByWrQ                      149                       # Number of read reqs serviced by write Q
+system.physmem.servicedByWrQ                      149                       # Number of DRAM read bursts serviced by write Q
 system.physmem.neitherReadNorWrite             131239                       # Reqs where no action is needed
 system.physmem.perBankRdReqs::0                 24063                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                 26436                       # Track reads on a per bank basis
@@ -316,11 +317,9 @@ system.membus.trans_dist::ReadExReq            206848                       # Tr
 system.membus.trans_dist::ReadExResp           206848                       # Transaction distribution
 system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port      1327226                       # Packet count per connected master and slave (bytes)
 system.membus.pkt_count_system.cpu.l2cache.mem_side::total      1327226                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count::system.physmem.port      1327226                       # Packet count per connected master and slave (bytes)
 system.membus.pkt_count::total                1327226                       # Packet count per connected master and slave (bytes)
 system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     43466368                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::total     43466368                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size::system.physmem.port     43466368                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.tot_pkt_size::total            43466368                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               43466368                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
@@ -606,12 +605,12 @@ system.cpu.toL2Bus.trans_dist::UpgradeReq       132628                       # T
 system.cpu.toL2Bus.trans_dist::UpgradeResp       132628                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq       771784                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp       771784                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side       146337                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side      7664164                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                  7810501                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side       435712                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side    311349248                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size             311784960                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       146337                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side      7664164                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total           7810501                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       435712                       # 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    311349248                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total      311784960                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus         311784960                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus      8494080                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy     4903151186                       # Layer occupancy (ticks)
@@ -620,15 +619,15 @@ system.cpu.toL2Bus.respLayer0.occupancy     209959241                       # La
 system.cpu.toL2Bus.respLayer0.utilization          0.0                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy    3959772656                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          0.9                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                   5293                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1036.459072                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                161843741                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                   6867                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs               23568.332751                       # 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    1036.459072                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.506084                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.506084                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements              5293                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1036.459072                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs           161843741                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs              6867                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs          23568.332751                       # 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  1036.459072                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.506084                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.506084                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst    161845824                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total       161845824                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst     161845824                       # number of demand (read+write) hits
@@ -704,19 +703,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total  3994.146443
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst  3994.146443                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total  3994.146443                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                352905                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             29673.331814                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                 3696859                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                385269                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  9.595527                       # Average number of references to valid blocks.
-system.cpu.l2cache.tags.warmup_cycle          199076310000                       # Cycle when the warmup percentage was hit.
+system.cpu.l2cache.tags.replacements           352905                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        29673.331814                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs            3696859                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           385269                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             9.595527                       # Average number of references to valid blocks.
+system.cpu.l2cache.tags.warmup_cycle     199076310000                       # Cycle when the warmup percentage was hit.
 system.cpu.l2cache.tags.occ_blocks::writebacks 21119.362848                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst    223.841801                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data   8330.127165                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst   223.841801                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data  8330.127165                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.644512                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.006831                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.254215                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.905558                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.905558                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst         3661                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data      1586701                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total        1590362                       # number of ReadReq hits
@@ -862,15 +861,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 65139.453621
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 59112.047244                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 59161.253496                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                2529980                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4088.352551                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                396070659                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                2534076                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 156.297861                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle             1764467250                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4088.352551                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.998133                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.998133                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements           2529980                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4088.352551                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs           396070659                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs           2534076                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            156.297861                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle        1764467250                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4088.352551                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.998133                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.998133                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data    247340077                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total       247340077                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data    148239061                       # number of WriteReq hits
diff --git a/tests/long/se/20.parser/ref/x86/linux/simple-timing/stats.txt b/tests/long/se/20.parser/ref/x86/linux/simple-timing/stats.txt
index 5255bf68c..7c10a36f9 100644
--- a/tests/long/se/20.parser/ref/x86/linux/simple-timing/stats.txt
+++ b/tests/long/se/20.parser/ref/x86/linux/simple-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  1.647873                       # Nu
 sim_ticks                                1647872849000                       # Number of ticks simulated
 final_tick                               1647872849000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 533286                       # Simulator instruction rate (inst/s)
-host_op_rate                                   986105                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                             1062778196                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 304632                       # Number of bytes of host memory used
-host_seconds                                  1550.53                       # Real time elapsed on the host
+host_inst_rate                                 418246                       # Simulator instruction rate (inst/s)
+host_op_rate                                   773383                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              833516309                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 254900                       # Number of bytes of host memory used
+host_seconds                                  1977.01                       # Real time elapsed on the host
 sim_insts                                   826877110                       # Number of instructions simulated
 sim_ops                                    1528988702                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            120704                       # Number of bytes read from this memory
@@ -42,11 +42,9 @@ system.membus.trans_dist::ReadExReq            206691                       # Tr
 system.membus.trans_dist::ReadExResp           206691                       # Transaction distribution
 system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port      1054572                       # Packet count per connected master and slave (bytes)
 system.membus.pkt_count_system.cpu.l2cache.mem_side::total      1054572                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count::system.physmem.port      1054572                       # Packet count per connected master and slave (bytes)
 system.membus.pkt_count::total                1054572                       # Packet count per connected master and slave (bytes)
 system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     43099456                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::total     43099456                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size::system.physmem.port     43099456                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.tot_pkt_size::total            43099456                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               43099456                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
@@ -77,15 +75,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                 3295745698                       # 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                   1253                       # number of replacements
-system.cpu.icache.tags.tagsinuse                881.356491                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs               1068344252                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                   2814                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs               379653.252310                       # 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     881.356491                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.430350                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.430350                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements              1253                       # number of replacements
+system.cpu.icache.tags.tagsinuse           881.356491                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs          1068344252                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs              2814                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs          379653.252310                       # 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   881.356491                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.430350                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.430350                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst   1068344252                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total      1068344252                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst    1068344252                       # number of demand (read+write) hits
@@ -155,19 +153,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 39153.518124
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 39153.518124                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 39153.518124                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                348459                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             29286.402664                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                 3655011                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                380814                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  9.597890                       # Average number of references to valid blocks.
-system.cpu.l2cache.tags.warmup_cycle          755936431000                       # Cycle when the warmup percentage was hit.
+system.cpu.l2cache.tags.replacements           348459                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        29286.402664                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs            3655011                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           380814                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             9.597890                       # Average number of references to valid blocks.
+system.cpu.l2cache.tags.warmup_cycle     755936431000                       # Cycle when the warmup percentage was hit.
 system.cpu.l2cache.tags.occ_blocks::writebacks 21041.299337                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst    139.758519                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data   8105.344807                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst   139.758519                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data  8105.344807                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.642129                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004265                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.247355                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.893750                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.893750                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst          928                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data      1554848                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total        1555776                       # number of ReadReq hits
@@ -293,15 +291,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 40006.362672
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 40000.324318                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 40000.354198                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                2514362                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4086.415783                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                530743930                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                2518458                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 210.741624                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle             8211724000                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4086.415783                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.997660                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.997660                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements           2514362                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4086.415783                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs           530743930                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs           2518458                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            210.741624                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle        8211724000                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4086.415783                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.997660                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.997660                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data    382374772                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total       382374772                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data    148369158                       # number of WriteReq hits
@@ -399,12 +397,12 @@ system.cpu.toL2Bus.trans_dist::ReadResp       1730228                       # Tr
 system.cpu.toL2Bus.trans_dist::Writeback      2323523                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq       791044                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp       791044                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side         5628                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side      7360439                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                  7366067                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side       180096                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side    309886784                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size             310066880                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side         5628                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side      7360439                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total           7366067                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       180096                       # 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    309886784                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total      310066880                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus         310066880                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy     4745920500                       # Layer occupancy (ticks)