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, 135 insertions, 133 deletions

diff --git a/tests/quick/se/00.hello/ref/arm/linux/o3-timing-checker/stats.txt b/tests/quick/se/00.hello/ref/arm/linux/o3-timing-checker/stats.txt
index 22dbcae6d..81f115ce9 100644
--- a/tests/quick/se/00.hello/ref/arm/linux/o3-timing-checker/stats.txt
+++ b/tests/quick/se/00.hello/ref/arm/linux/o3-timing-checker/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000016                       # Nu
 sim_ticks                                    16494000                       # Number of ticks simulated
 final_tick                                   16494000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  31208                       # Simulator instruction rate (inst/s)
-host_op_rate                                    38937                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                              112083077                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 244336                       # Number of bytes of host memory used
-host_seconds                                     0.15                       # Real time elapsed on the host
+host_inst_rate                                  32065                       # Simulator instruction rate (inst/s)
+host_op_rate                                    40006                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              115159682                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 240696                       # Number of bytes of host memory used
+host_seconds                                     0.14                       # Real time elapsed on the host
 sim_insts                                        4591                       # Number of instructions simulated
 sim_ops                                          5729                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             17344                       # Number of bytes read from this memory
@@ -27,14 +27,15 @@ system.physmem.bw_inst_read::total         1051533891                       # In
 system.physmem.bw_total::cpu.inst          1051533891                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           473384261                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total             1524918152                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                           393                       # Total number of read requests seen
-system.physmem.writeReqs                            0                       # Total number of write requests seen
-system.physmem.cpureqs                            393                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                           393                       # Total number of read requests accepted by DRAM controller
+system.physmem.writeReqs                            0                       # Total number of write requests accepted by DRAM controller
+system.physmem.readBursts                         393                       # Total number of DRAM read bursts. Each DRAM read request translates to either one or multiple DRAM read bursts
+system.physmem.writeBursts                          0                       # Total number of DRAM write bursts. Each DRAM write request translates to either one or multiple DRAM write bursts
 system.physmem.bytesRead                        25152                       # Total number of bytes read from memory
 system.physmem.bytesWritten                         0                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                  25152                       # bytesRead derated as per pkt->getSize()
 system.physmem.bytesConsumedWr                      0                       # 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                    86                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                    46                       # Track reads on a per bank basis
@@ -194,10 +195,10 @@ system.membus.trans_dist::ReadReq                 352                       # Tr
 system.membus.trans_dist::ReadResp                352                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                41                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               41                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          786                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           786                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        25152                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      25152                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          786                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    786                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        25152                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               25152                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  25152                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              487500                       # Layer occupancy (ticks)
@@ -563,12 +564,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            397                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           396                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           41                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           41                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          582                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          293                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      875                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        18624                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side         9344                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 27968                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          582                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          293                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               875                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        18624                       # 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         9344                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          27968                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             27968                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         219000                       # Layer occupancy (ticks)
@@ -577,15 +578,15 @@ system.cpu.toL2Bus.respLayer0.occupancy        485250                       # La
 system.cpu.toL2Bus.respLayer0.utilization          2.9                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy        231495                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          1.4                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                      4                       # number of replacements
-system.cpu.icache.tags.tagsinuse                145.483199                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     1583                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    291                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                   5.439863                       # 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     145.483199                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.071037                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.071037                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 4                       # number of replacements
+system.cpu.icache.tags.tagsinuse           145.483199                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                1583                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               291                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs              5.439863                       # 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   145.483199                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.071037                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.071037                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         1583                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            1583                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          1583                       # number of demand (read+write) hits
@@ -661,17 +662,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 64263.745704
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 64263.745704                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 64263.745704                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                     0                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse               183.328645                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                      40                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   352                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.113636                       # 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::cpu.inst    136.957008                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     46.371637                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          183.328645                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                 40                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              352                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.113636                       # 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::cpu.inst   136.957008                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    46.371637                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004180                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001415                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.005595                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.005595                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst           20                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data           20                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total             40                       # number of ReadReq hits
@@ -795,15 +796,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 54591.328413
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 60643.442623                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 56470.101781                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                      0                       # number of replacements
-system.cpu.dcache.tags.tagsinuse                 85.893510                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     2390                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    146                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  16.369863                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle                      0                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data      85.893510                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.020970                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.020970                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            85.893510                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                2390                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               146                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             16.369863                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle                 0                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data    85.893510                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.020970                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.020970                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1763                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1763                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          606                       # number of WriteReq hits
diff --git a/tests/quick/se/00.hello/ref/arm/linux/o3-timing/stats.txt b/tests/quick/se/00.hello/ref/arm/linux/o3-timing/stats.txt
index 3ccfc050f..ace16d792 100644
--- a/tests/quick/se/00.hello/ref/arm/linux/o3-timing/stats.txt
+++ b/tests/quick/se/00.hello/ref/arm/linux/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000016                       # Nu
 sim_ticks                                    16494000                       # Number of ticks simulated
 final_tick                                   16494000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  66928                       # Simulator instruction rate (inst/s)
-host_op_rate                                    83502                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                              240363471                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 244336                       # Number of bytes of host memory used
-host_seconds                                     0.07                       # Real time elapsed on the host
+host_inst_rate                                  36590                       # Simulator instruction rate (inst/s)
+host_op_rate                                    45651                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              131406771                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 240696                       # Number of bytes of host memory used
+host_seconds                                     0.13                       # Real time elapsed on the host
 sim_insts                                        4591                       # Number of instructions simulated
 sim_ops                                          5729                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             17344                       # Number of bytes read from this memory
@@ -27,14 +27,15 @@ system.physmem.bw_inst_read::total         1051533891                       # In
 system.physmem.bw_total::cpu.inst          1051533891                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           473384261                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total             1524918152                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                           393                       # Total number of read requests seen
-system.physmem.writeReqs                            0                       # Total number of write requests seen
-system.physmem.cpureqs                            393                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                           393                       # Total number of read requests accepted by DRAM controller
+system.physmem.writeReqs                            0                       # Total number of write requests accepted by DRAM controller
+system.physmem.readBursts                         393                       # Total number of DRAM read bursts. Each DRAM read request translates to either one or multiple DRAM read bursts
+system.physmem.writeBursts                          0                       # Total number of DRAM write bursts. Each DRAM write request translates to either one or multiple DRAM write bursts
 system.physmem.bytesRead                        25152                       # Total number of bytes read from memory
 system.physmem.bytesWritten                         0                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                  25152                       # bytesRead derated as per pkt->getSize()
 system.physmem.bytesConsumedWr                      0                       # 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                    86                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                    46                       # Track reads on a per bank basis
@@ -194,10 +195,10 @@ system.membus.trans_dist::ReadReq                 352                       # Tr
 system.membus.trans_dist::ReadResp                352                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                41                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               41                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          786                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           786                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        25152                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      25152                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          786                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    786                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        25152                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               25152                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  25152                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              487500                       # Layer occupancy (ticks)
@@ -518,12 +519,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            397                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           396                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           41                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           41                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          582                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          293                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      875                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        18624                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side         9344                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 27968                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          582                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          293                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               875                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        18624                       # 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         9344                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          27968                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             27968                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         219000                       # Layer occupancy (ticks)
@@ -532,15 +533,15 @@ system.cpu.toL2Bus.respLayer0.occupancy        485250                       # La
 system.cpu.toL2Bus.respLayer0.utilization          2.9                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy        231495                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          1.4                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                      4                       # number of replacements
-system.cpu.icache.tags.tagsinuse                145.483199                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     1583                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    291                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                   5.439863                       # 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     145.483199                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.071037                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.071037                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 4                       # number of replacements
+system.cpu.icache.tags.tagsinuse           145.483199                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                1583                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               291                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs              5.439863                       # 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   145.483199                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.071037                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.071037                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         1583                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            1583                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          1583                       # number of demand (read+write) hits
@@ -616,17 +617,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 64263.745704
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 64263.745704                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 64263.745704                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                     0                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse               183.328645                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                      40                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   352                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.113636                       # 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::cpu.inst    136.957008                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     46.371637                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          183.328645                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                 40                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              352                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.113636                       # 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::cpu.inst   136.957008                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    46.371637                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004180                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001415                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.005595                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.005595                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst           20                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data           20                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total             40                       # number of ReadReq hits
@@ -750,15 +751,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 54591.328413
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 60643.442623                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 56470.101781                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                      0                       # number of replacements
-system.cpu.dcache.tags.tagsinuse                 85.893510                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     2390                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    146                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  16.369863                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle                      0                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data      85.893510                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.020970                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.020970                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            85.893510                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                2390                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               146                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             16.369863                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle                 0                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data    85.893510                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.020970                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.020970                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1763                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1763                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          606                       # number of WriteReq hits
diff --git a/tests/quick/se/00.hello/ref/arm/linux/simple-timing/stats.txt b/tests/quick/se/00.hello/ref/arm/linux/simple-timing/stats.txt
index 7a58b161f..13e2763d6 100644
--- a/tests/quick/se/00.hello/ref/arm/linux/simple-timing/stats.txt
+++ b/tests/quick/se/00.hello/ref/arm/linux/simple-timing/stats.txt
@@ -4,10 +4,10 @@ sim_seconds                                  0.000026                       # Nu
 sim_ticks                                    25969000                       # Number of ticks simulated
 final_tick                                   25969000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 220478                       # Simulator instruction rate (inst/s)
-host_op_rate                                   273604                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                             1251201624                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 241012                       # Number of bytes of host memory used
+host_inst_rate                                 229244                       # Simulator instruction rate (inst/s)
+host_op_rate                                   284503                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                             1301168988                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 238660                       # Number of bytes of host memory used
 host_seconds                                     0.02                       # Real time elapsed on the host
 sim_insts                                        4565                       # Number of instructions simulated
 sim_ops                                          5672                       # Number of ops (including micro ops) simulated
@@ -32,10 +32,10 @@ system.membus.trans_dist::ReadReq                 307                       # Tr
 system.membus.trans_dist::ReadResp                307                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                43                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               43                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          700                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           700                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        22400                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      22400                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          700                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    700                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        22400                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               22400                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  22400                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              350000                       # Layer occupancy (ticks)
@@ -107,15 +107,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                      51938                       # 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                      1                       # number of replacements
-system.cpu.icache.tags.tagsinuse                114.614391                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     4364                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    241                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                  18.107884                       # 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     114.614391                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.055964                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.055964                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 1                       # number of replacements
+system.cpu.icache.tags.tagsinuse           114.614391                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                4364                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               241                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs             18.107884                       # 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   114.614391                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.055964                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.055964                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         4364                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            4364                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          4364                       # number of demand (read+write) hits
@@ -185,17 +185,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 50211.618257
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 50211.618257                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 50211.618257                       # average overall mshr miss latency
 system.cpu.icache.no_allocate_misses                0                       # Number of misses that were no-allocate
-system.cpu.l2cache.tags.replacements                     0                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse               154.071129                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                      32                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   307                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.104235                       # 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::cpu.inst    105.889758                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     48.181371                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          154.071129                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                 32                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              307                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.104235                       # 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::cpu.inst   105.889758                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    48.181371                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.003231                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001470                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.004702                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.004702                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst           16                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::cpu.data           16                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total             32                       # number of ReadReq hits
@@ -313,15 +313,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst        40000
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data        40000                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total        40000                       # average overall mshr miss latency
 system.cpu.l2cache.no_allocate_misses               0                       # Number of misses that were no-allocate
-system.cpu.dcache.tags.replacements                      0                       # number of replacements
-system.cpu.dcache.tags.tagsinuse                 83.000387                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     1940                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    141                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  13.758865                       # Average number of references to valid blocks.
-system.cpu.dcache.tags.warmup_cycle                      0                       # Cycle when the warmup percentage was hit.
-system.cpu.dcache.tags.occ_blocks::cpu.data      83.000387                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.020264                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.020264                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            83.000387                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                1940                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               141                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             13.758865                       # Average number of references to valid blocks.
+system.cpu.dcache.tags.warmup_cycle                 0                       # Cycle when the warmup percentage was hit.
+system.cpu.dcache.tags.occ_blocks::cpu.data    83.000387                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.020264                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.020264                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1048                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1048                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          870                       # number of WriteReq hits
@@ -424,12 +424,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            339                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           339                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           43                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           43                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          482                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          282                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      764                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        15424                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side         9024                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 24448                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          482                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          282                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               764                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        15424                       # 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         9024                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          24448                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             24448                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         191000                       # Layer occupancy (ticks)