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/quick/se/00.hello/ref/mips/linux/inorder-timing/stats.txt b/tests/quick/se/00.hello/ref/mips/linux/inorder-timing/stats.txt
index b2a150376..aeb0e2e25 100644
--- a/tests/quick/se/00.hello/ref/mips/linux/inorder-timing/stats.txt
+++ b/tests/quick/se/00.hello/ref/mips/linux/inorder-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000025                       # Nu
 sim_ticks                                    24587000                       # Number of ticks simulated
 final_tick                                   24587000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  41260                       # Simulator instruction rate (inst/s)
-host_op_rate                                    41253                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                              174426700                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 226212                       # Number of bytes of host memory used
-host_seconds                                     0.14                       # Real time elapsed on the host
+host_inst_rate                                  52979                       # Simulator instruction rate (inst/s)
+host_op_rate                                    52966                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              223940501                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 224928                       # Number of bytes of host memory used
+host_seconds                                     0.11                       # Real time elapsed on the host
 sim_insts                                        5814                       # Number of instructions simulated
 sim_ops                                          5814                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             20288                       # Number of bytes read from this memory
@@ -27,14 +27,15 @@ system.physmem.bw_inst_read::total          825151503                       # In
 system.physmem.bw_total::cpu.inst           825151503                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           359214219                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total             1184365722                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                           455                       # Total number of read requests seen
-system.physmem.writeReqs                            0                       # Total number of write requests seen
-system.physmem.cpureqs                            455                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                           455                       # 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                         455                       # 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                        29120                       # Total number of bytes read from memory
 system.physmem.bytesWritten                         0                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                  29120                       # 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                    28                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                     0                       # Track reads on a per bank basis
@@ -195,10 +196,10 @@ system.membus.trans_dist::ReadReq                 404                       # Tr
 system.membus.trans_dist::ReadResp                404                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                51                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               51                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          910                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           910                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        29120                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      29120                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          910                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    910                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        29120                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               29120                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  29120                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              551500                       # Layer occupancy (ticks)
@@ -293,15 +294,15 @@ system.cpu.stage3.utilization                2.517539                       # Pe
 system.cpu.stage4.idleCycles                    46285                       # Number of cycles 0 instructions are processed.
 system.cpu.stage4.runCycles                      2890                       # Number of cycles 1+ instructions are processed.
 system.cpu.stage4.utilization                5.876970                       # Percentage of cycles stage was utilized (processing insts).
-system.cpu.icache.tags.replacements                     13                       # number of replacements
-system.cpu.icache.tags.tagsinuse                150.350232                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                      428                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    319                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                   1.341693                       # 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     150.350232                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.073413                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.073413                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                13                       # number of replacements
+system.cpu.icache.tags.tagsinuse           150.350232                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                 428                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               319                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs              1.341693                       # 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   150.350232                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.073413                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.073413                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst          428                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total             428                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst           428                       # number of demand (read+write) hits
@@ -382,12 +383,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            406                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           406                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           51                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           51                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          638                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          276                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      914                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        20416                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side         8832                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 29248                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          638                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          276                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               914                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        20416                       # 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         8832                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          29248                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             29248                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         228500                       # Layer occupancy (ticks)
@@ -396,17 +397,17 @@ system.cpu.toL2Bus.respLayer0.occupancy        543000                       # La
 system.cpu.toL2Bus.respLayer0.utilization          2.2                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy        228000                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          0.9                       # Layer utilization (%)
-system.cpu.l2cache.tags.replacements                     0                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse               208.008874                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                       2                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   404                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.004950                       # 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    152.043119                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     55.965756                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          208.008874                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                  2                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              404                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.004950                       # 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   152.043119                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    55.965756                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004640                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001708                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.006348                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.006348                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst            2                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total              2                       # number of ReadReq hits
 system.cpu.l2cache.demand_hits::cpu.inst            2                       # number of demand (read+write) hits
@@ -521,15 +522,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 57861.198738
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 62804.347826                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 59360.439560                       # 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                 89.984709                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     1638                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    138                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  11.869565                       # 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      89.984709                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.021969                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.021969                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            89.984709                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                1638                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               138                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             11.869565                       # 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    89.984709                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.021969                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.021969                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1066                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1066                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          572                       # number of WriteReq hits
diff --git a/tests/quick/se/00.hello/ref/mips/linux/o3-timing/stats.txt b/tests/quick/se/00.hello/ref/mips/linux/o3-timing/stats.txt
index 6a930873f..f0a85d261 100644
--- a/tests/quick/se/00.hello/ref/mips/linux/o3-timing/stats.txt
+++ b/tests/quick/se/00.hello/ref/mips/linux/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000022                       # Nu
 sim_ticks                                    21805500                       # Number of ticks simulated
 final_tick                                   21805500                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  79844                       # Simulator instruction rate (inst/s)
-host_op_rate                                    79828                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                              337538221                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 228256                       # Number of bytes of host memory used
-host_seconds                                     0.06                       # Real time elapsed on the host
+host_inst_rate                                  44396                       # Simulator instruction rate (inst/s)
+host_op_rate                                    44386                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              187676879                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 228012                       # Number of bytes of host memory used
+host_seconds                                     0.12                       # Real time elapsed on the host
 sim_insts                                        5156                       # Number of instructions simulated
 sim_ops                                          5156                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             21440                       # Number of bytes read from this memory
@@ -27,14 +27,15 @@ system.physmem.bw_inst_read::total          983238174                       # In
 system.physmem.bw_total::cpu.inst           983238174                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           416775584                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total             1400013758                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                           477                       # Total number of read requests seen
-system.physmem.writeReqs                            0                       # Total number of write requests seen
-system.physmem.cpureqs                            477                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                           477                       # 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                         477                       # 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                        30528                       # Total number of bytes read from memory
 system.physmem.bytesWritten                         0                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                  30528                       # 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                    30                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                     0                       # Track reads on a per bank basis
@@ -194,10 +195,10 @@ system.membus.trans_dist::ReadReq                 426                       # Tr
 system.membus.trans_dist::ReadResp                426                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                51                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               51                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          954                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           954                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        30528                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      30528                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          954                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    954                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        30528                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               30528                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  30528                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              605000                       # Layer occupancy (ticks)
@@ -495,12 +496,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            429                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           429                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           51                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           51                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          676                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          284                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      960                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        21632                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side         9088                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 30720                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          676                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          284                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               960                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        21632                       # 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         9088                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          30720                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             30720                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         240000                       # Layer occupancy (ticks)
@@ -509,15 +510,15 @@ system.cpu.toL2Bus.respLayer0.occupancy        573500                       # La
 system.cpu.toL2Bus.respLayer0.utilization          2.6                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy        230000                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          1.1                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                     17                       # number of replacements
-system.cpu.icache.tags.tagsinuse                160.845390                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     1531                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    338                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                   4.529586                       # 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     160.845390                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.078538                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.078538                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                17                       # number of replacements
+system.cpu.icache.tags.tagsinuse           160.845390                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                1531                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               338                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs              4.529586                       # 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   160.845390                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.078538                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.078538                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         1531                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            1531                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          1531                       # number of demand (read+write) hits
@@ -593,17 +594,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 70585.798817
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 70585.798817                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 70585.798817                       # 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               220.792115                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                       3                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   426                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.007042                       # 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    163.133804                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     57.658310                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          220.792115                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                  3                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              426                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.007042                       # 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   163.133804                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    57.658310                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004978                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001760                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.006738                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.006738                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst            3                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total              3                       # number of ReadReq hits
 system.cpu.l2cache.demand_hits::cpu.inst            3                       # number of demand (read+write) hits
@@ -718,15 +719,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 57459.701493
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 64862.676056                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 59663.522013                       # 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                 91.308892                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     2395                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    142                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  16.866197                       # 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      91.308892                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.022292                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.022292                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            91.308892                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                2395                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               142                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             16.866197                       # 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    91.308892                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.022292                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.022292                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1832                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1832                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          563                       # number of WriteReq hits
diff --git a/tests/quick/se/00.hello/ref/mips/linux/simple-timing/stats.txt b/tests/quick/se/00.hello/ref/mips/linux/simple-timing/stats.txt
index bfb8470a6..d3256ea4d 100644
--- a/tests/quick/se/00.hello/ref/mips/linux/simple-timing/stats.txt
+++ b/tests/quick/se/00.hello/ref/mips/linux/simple-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000032                       # Nu
 sim_ticks                                    31633000                       # Number of ticks simulated
 final_tick                                   31633000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 482351                       # Simulator instruction rate (inst/s)
-host_op_rate                                   481309                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                             2613274672                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 225064                       # Number of bytes of host memory used
-host_seconds                                     0.01                       # Real time elapsed on the host
+host_inst_rate                                 304637                       # Simulator instruction rate (inst/s)
+host_op_rate                                   304230                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                             1653175117                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 224940                       # Number of bytes of host memory used
+host_seconds                                     0.02                       # Real time elapsed on the host
 sim_insts                                        5814                       # Number of instructions simulated
 sim_ops                                          5814                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             19264                       # Number of bytes read from this memory
@@ -32,10 +32,10 @@ system.membus.trans_dist::ReadReq                 388                       # Tr
 system.membus.trans_dist::ReadResp                388                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                51                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               51                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          878                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           878                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        28096                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      28096                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          878                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    878                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        28096                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               28096                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  28096                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              439000                       # Layer occupancy (ticks)
@@ -83,15 +83,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                      63266                       # 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                     13                       # number of replacements
-system.cpu.icache.tags.tagsinuse                132.545353                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     5513                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    303                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                  18.194719                       # 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     132.545353                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.064719                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.064719                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                13                       # number of replacements
+system.cpu.icache.tags.tagsinuse           132.545353                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                5513                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               303                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs             18.194719                       # 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   132.545353                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.064719                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.064719                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         5513                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            5513                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          5513                       # number of demand (read+write) hits
@@ -161,17 +161,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 52722.772277
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 52722.772277                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 52722.772277                       # 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               188.114191                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                       2                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   388                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.005155                       # 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    133.890657                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     54.223533                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          188.114191                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                  2                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              388                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.005155                       # 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   133.890657                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    54.223533                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004086                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001655                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.005741                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.005741                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst            2                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total              2                       # number of ReadReq hits
 system.cpu.l2cache.demand_hits::cpu.inst            2                       # number of demand (read+write) hits
@@ -286,15 +286,15 @@ 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                 87.492114                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     1950                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    138                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  14.130435                       # 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      87.492114                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.021360                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.021360                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            87.492114                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                1950                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               138                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             14.130435                       # 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    87.492114                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.021360                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.021360                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1076                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1076                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          874                       # number of WriteReq hits
@@ -389,12 +389,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            390                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           390                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           51                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           51                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          606                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          276                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      882                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        19392                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side         8832                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 28224                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          606                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          276                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               882                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        19392                       # 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         8832                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          28224                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             28224                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         220500                       # Layer occupancy (ticks)