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.

3 files changed, 136 insertions, 134 deletions

diff --git a/tests/long/se/50.vortex/ref/alpha/tru64/inorder-timing/stats.txt b/tests/long/se/50.vortex/ref/alpha/tru64/inorder-timing/stats.txt
index fc992598c..c7e2525ee 100644
--- a/tests/long/se/50.vortex/ref/alpha/tru64/inorder-timing/stats.txt
+++ b/tests/long/se/50.vortex/ref/alpha/tru64/inorder-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.043769                       # Nu
 sim_ticks                                 43769191000                       # Number of ticks simulated
 final_tick                                43769191000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 112888                       # Simulator instruction rate (inst/s)
-host_op_rate                                   112888                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               55931443                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 233228                       # Number of bytes of host memory used
-host_seconds                                   782.55                       # Real time elapsed on the host
+host_inst_rate                                  69144                       # Simulator instruction rate (inst/s)
+host_op_rate                                    69144                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                               34257993                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 232832                       # Number of bytes of host memory used
+host_seconds                                  1277.63                       # Real time elapsed on the host
 sim_insts                                    88340673                       # Number of instructions simulated
 sim_ops                                      88340673                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            454592                       # Number of bytes read from this memory
@@ -34,14 +34,15 @@ system.physmem.bw_total::writebacks         166688208                       # To
 system.physmem.bw_total::cpu.inst            10386118                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           231632520                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total              408706846                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                        165515                       # Total number of read requests seen
-system.physmem.writeReqs                       113997                       # Total number of write requests seen
-system.physmem.cpureqs                         279512                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                        165515                       # Total number of read requests accepted by DRAM controller
+system.physmem.writeReqs                       113997                       # Total number of write requests accepted by DRAM controller
+system.physmem.readBursts                      165515                       # Total number of DRAM read bursts. Each DRAM read request translates to either one or multiple DRAM read bursts
+system.physmem.writeBursts                     113997                       # Total number of DRAM write bursts. Each DRAM write request translates to either one or multiple DRAM write bursts
 system.physmem.bytesRead                     10592960                       # Total number of bytes read from memory
 system.physmem.bytesWritten                   7295808                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd               10592960                       # bytesRead derated as per pkt->getSize()
 system.physmem.bytesConsumedWr                7295808                       # bytesWritten derated as per pkt->getSize()
-system.physmem.servicedByWrQ                        0                       # Number of read reqs serviced by write Q
+system.physmem.servicedByWrQ                        0                       # Number of DRAM read bursts serviced by write Q
 system.physmem.neitherReadNorWrite                  0                       # Reqs where no action is needed
 system.physmem.perBankRdReqs::0                 10379                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                 10437                       # Track reads on a per bank basis
@@ -311,10 +312,10 @@ system.membus.trans_dist::ReadResp              34625                       # Tr
 system.membus.trans_dist::Writeback            113997                       # Transaction distribution
 system.membus.trans_dist::ReadExReq            130890                       # Transaction distribution
 system.membus.trans_dist::ReadExResp           130890                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side       445027                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                        445027                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     17888768                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   17888768                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port       445027                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                 445027                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     17888768                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            17888768                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               17888768                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1218896000                       # Layer occupancy (ticks)
@@ -423,15 +424,15 @@ system.cpu.stage3.utilization               25.270124                       # Pe
 system.cpu.stage4.idleCycles                 41496378                       # Number of cycles 0 instructions are processed.
 system.cpu.stage4.runCycles                  46042005                       # Number of cycles 1+ instructions are processed.
 system.cpu.stage4.utilization               52.596362                       # Percentage of cycles stage was utilized (processing insts).
-system.cpu.icache.tags.replacements                  84371                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1906.602529                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                 12250515                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  86417                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                 141.760475                       # 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    1906.602529                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.930958                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.930958                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements             84371                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1906.602529                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs            12250515                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             86417                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs            141.760475                       # 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  1906.602529                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.930958                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.930958                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst     12250515                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total        12250515                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst      12250515                       # number of demand (read+write) hits
@@ -513,12 +514,12 @@ system.cpu.toL2Bus.trans_dist::ReadResp        146995                       # Tr
 system.cpu.toL2Bus.trans_dist::Writeback       168352                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq       143769                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp       143769                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side       172834                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side       577046                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                   749880                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side      5530688                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side     23852736                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size              29383424                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       172834                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side       577046                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total            749880                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side      5530688                       # 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     23852736                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total       29383424                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus          29383424                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy      397910000                       # Layer occupancy (ticks)
@@ -527,19 +528,19 @@ system.cpu.toL2Bus.respLayer0.occupancy     131178984                       # La
 system.cpu.toL2Bus.respLayer0.utilization          0.3                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy     326782984                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          0.7                       # Layer utilization (%)
-system.cpu.l2cache.tags.replacements                131591                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             30902.226523                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                  151434                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                163651                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.925347                       # 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           131591                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        30902.226523                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs             151434                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           163651                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.925347                       # 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 27124.475533                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst   2007.439767                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data   1770.311223                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst  2007.439767                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data  1770.311223                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.827773                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.061262                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.054026                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.943061                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.943061                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst        79314                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data        33056                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total         112370                       # number of ReadReq hits
@@ -665,15 +666,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 69040.581444
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 87183.204240                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 86404.621938                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                 200251                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4076.642006                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                 33754840                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                 204347                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 165.183927                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle              293009000                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4076.642006                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.995274                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.995274                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements            200251                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4076.642006                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs            33754840                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs            204347                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            165.183927                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle         293009000                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4076.642006                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.995274                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.995274                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data     20180271                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total        20180271                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data     13574569                       # number of WriteReq hits
diff --git a/tests/long/se/50.vortex/ref/alpha/tru64/o3-timing/stats.txt b/tests/long/se/50.vortex/ref/alpha/tru64/o3-timing/stats.txt
index a1c1e25d4..5b9e2998f 100644
--- a/tests/long/se/50.vortex/ref/alpha/tru64/o3-timing/stats.txt
+++ b/tests/long/se/50.vortex/ref/alpha/tru64/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.024977                       # Nu
 sim_ticks                                 24977022500                       # Number of ticks simulated
 final_tick                                24977022500                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 130696                       # Simulator instruction rate (inst/s)
-host_op_rate                                   130696                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               41014411                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 236320                       # Number of bytes of host memory used
-host_seconds                                   608.98                       # Real time elapsed on the host
+host_inst_rate                                 124025                       # Simulator instruction rate (inst/s)
+host_op_rate                                   124025                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                               38920856                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 235900                       # Number of bytes of host memory used
+host_seconds                                   641.74                       # Real time elapsed on the host
 sim_insts                                    79591756                       # Number of instructions simulated
 sim_ops                                      79591756                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            489984                       # Number of bytes read from this memory
@@ -34,14 +34,15 @@ system.physmem.bw_total::writebacks         292149475                       # To
 system.physmem.bw_total::cpu.inst            19617390                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           406515068                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total              718281933                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                        166305                       # Total number of read requests seen
-system.physmem.writeReqs                       114016                       # Total number of write requests seen
-system.physmem.cpureqs                         280321                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                        166305                       # Total number of read requests accepted by DRAM controller
+system.physmem.writeReqs                       114016                       # Total number of write requests accepted by DRAM controller
+system.physmem.readBursts                      166305                       # Total number of DRAM read bursts. Each DRAM read request translates to either one or multiple DRAM read bursts
+system.physmem.writeBursts                     114016                       # Total number of DRAM write bursts. Each DRAM write request translates to either one or multiple DRAM write bursts
 system.physmem.bytesRead                     10643520                       # Total number of bytes read from memory
 system.physmem.bytesWritten                   7297024                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd               10643520                       # bytesRead derated as per pkt->getSize()
 system.physmem.bytesConsumedWr                7297024                       # bytesWritten derated as per pkt->getSize()
-system.physmem.servicedByWrQ                        3                       # Number of read reqs serviced by write Q
+system.physmem.servicedByWrQ                        3                       # Number of DRAM read bursts serviced by write Q
 system.physmem.neitherReadNorWrite                  0                       # Reqs where no action is needed
 system.physmem.perBankRdReqs::0                 10424                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                 10464                       # Track reads on a per bank basis
@@ -309,10 +310,10 @@ system.membus.trans_dist::ReadResp              35508                       # Tr
 system.membus.trans_dist::Writeback            114016                       # Transaction distribution
 system.membus.trans_dist::ReadExReq            130797                       # Transaction distribution
 system.membus.trans_dist::ReadExResp           130797                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side       446626                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                        446626                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     17940544                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   17940544                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port       446626                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                 446626                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     17940544                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            17940544                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               17940544                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1244155000                       # Layer occupancy (ticks)
@@ -628,12 +629,12 @@ system.cpu.toL2Bus.trans_dist::ReadResp        155431                       # Tr
 system.cpu.toL2Bus.trans_dist::Writeback       168929                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq       143410                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp       143410                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side       186551                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side       580061                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                   766612                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side      5969600                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side     23967680                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size              29937280                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       186551                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side       580061                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total            766612                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side      5969600                       # 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     23967680                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total       29937280                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus          29937280                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy      402814500                       # Layer occupancy (ticks)
@@ -642,15 +643,15 @@ system.cpu.toL2Bus.respLayer0.occupancy     141571734                       # La
 system.cpu.toL2Bus.respLayer0.utilization          0.6                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy     327076000                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          1.3                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                  91227                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1926.280031                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                 13799737                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  93275                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                 147.946792                       # Average number of references to valid blocks.
-system.cpu.icache.tags.warmup_cycle            20172265250                       # Cycle when the warmup percentage was hit.
-system.cpu.icache.tags.occ_blocks::cpu.inst    1926.280031                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.940566                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.940566                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements             91227                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1926.280031                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs            13799737                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             93275                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs            147.946792                       # Average number of references to valid blocks.
+system.cpu.icache.tags.warmup_cycle       20172265250                       # Cycle when the warmup percentage was hit.
+system.cpu.icache.tags.occ_blocks::cpu.inst  1926.280031                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.940566                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.940566                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst     13799737                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total        13799737                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst      13799737                       # number of demand (read+write) hits
@@ -726,19 +727,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 17000.812278
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 17000.812278                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 17000.812278                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                132400                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             30717.176709                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                  159637                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                164461                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.970668                       # 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           132400                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        30717.176709                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs             159637                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           164461                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.970668                       # 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 26388.752281                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst   2106.212865                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data   2222.211563                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst  2106.212865                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data  2222.211563                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.805321                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.064277                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.067817                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.937414                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.937414                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst        85619                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data        34304                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total         119923                       # number of ReadReq hits
@@ -864,15 +865,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 70298.942144
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 89605.561649                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 88716.653338                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                 201470                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4074.474898                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                 34190075                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                 205566                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 166.321644                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle              215349000                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4074.474898                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.994745                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.994745                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements            201470                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4074.474898                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs            34190075                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs            205566                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            166.321644                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle         215349000                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4074.474898                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.994745                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.994745                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data     20615905                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total        20615905                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data     13574108                       # number of WriteReq hits
diff --git a/tests/long/se/50.vortex/ref/alpha/tru64/simple-timing/stats.txt b/tests/long/se/50.vortex/ref/alpha/tru64/simple-timing/stats.txt
index 060f66d07..bac018361 100644
--- a/tests/long/se/50.vortex/ref/alpha/tru64/simple-timing/stats.txt
+++ b/tests/long/se/50.vortex/ref/alpha/tru64/simple-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.133635                       # Nu
 sim_ticks                                133634727000                       # Number of ticks simulated
 final_tick                               133634727000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 671194                       # Simulator instruction rate (inst/s)
-host_op_rate                                   671194                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                             1015328507                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 233108                       # Number of bytes of host memory used
-host_seconds                                   131.62                       # Real time elapsed on the host
+host_inst_rate                                 775893                       # Simulator instruction rate (inst/s)
+host_op_rate                                   775893                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                             1173708198                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 233104                       # Number of bytes of host memory used
+host_seconds                                   113.86                       # Real time elapsed on the host
 sim_insts                                    88340673                       # Number of instructions simulated
 sim_ops                                      88340673                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst            432896                       # Number of bytes read from this memory
@@ -40,10 +40,10 @@ system.membus.trans_dist::ReadResp              34272                       # Tr
 system.membus.trans_dist::Writeback            113982                       # Transaction distribution
 system.membus.trans_dist::ReadExReq            130881                       # Transaction distribution
 system.membus.trans_dist::ReadExResp           130881                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side       444288                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                        444288                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side     17864640                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                   17864640                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port       444288                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                 444288                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port     17864640                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total            17864640                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus               17864640                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy          1190991000                       # Layer occupancy (ticks)
@@ -105,15 +105,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                  267269454                       # 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                  74391                       # number of replacements
-system.cpu.icache.tags.tagsinuse               1871.686406                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                 88361638                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                  76436                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                1156.021220                       # 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    1871.686406                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.913909                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.913909                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements             74391                       # number of replacements
+system.cpu.icache.tags.tagsinuse          1871.686406                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs            88361638                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs             76436                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs           1156.021220                       # 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  1871.686406                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.913909                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.913909                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst     88361638                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total        88361638                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst      88361638                       # number of demand (read+write) hits
@@ -183,19 +183,19 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 14721.335496
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 14721.335496                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 14721.335496                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                131235                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse             30728.810101                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                  142024                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                163291                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.869760                       # 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           131235                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse        30728.810101                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs             142024                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs           163291                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.869760                       # 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 27298.448351                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.inst   1874.507766                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data   1555.853984                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.inst  1874.507766                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data  1555.853984                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::writebacks     0.833083                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.057205                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.047481                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.937769                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.937769                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst        69672                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data        33258                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total         102930                       # number of ReadReq hits
@@ -321,15 +321,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 40052.631579
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 40003.137844                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 40005.164908                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                 200248                       # number of replacements
-system.cpu.dcache.tags.tagsinuse               4078.863631                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                 34685671                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                 204344                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                 169.741568                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle              936463000                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data    4078.863631                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.995816                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.995816                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements            200248                       # number of replacements
+system.cpu.dcache.tags.tagsinuse          4078.863631                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs            34685671                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs            204344                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs            169.741568                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle         936463000                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data  4078.863631                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.995816                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.995816                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data     20215872                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total        20215872                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data     14469799                       # number of WriteReq hits
@@ -427,12 +427,12 @@ system.cpu.toL2Bus.trans_dist::ReadResp        137202                       # Tr
 system.cpu.toL2Bus.trans_dist::Writeback       168375                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq       143578                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp       143578                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side       152872                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side       577063                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                   729935                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side      4891904                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side     23854016                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size              28745920                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side       152872                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side       577063                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total            729935                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side      4891904                       # 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     23854016                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total       28745920                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus          28745920                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy      392952500                       # Layer occupancy (ticks)