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/02.insttest/ref/sparc/linux/inorder-timing/stats.txt b/tests/quick/se/02.insttest/ref/sparc/linux/inorder-timing/stats.txt
index 60e6f3a9f..0c812fe4f 100644
--- a/tests/quick/se/02.insttest/ref/sparc/linux/inorder-timing/stats.txt
+++ b/tests/quick/se/02.insttest/ref/sparc/linux/inorder-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000027                       # Nu
 sim_ticks                                    27282000                       # Number of ticks simulated
 final_tick                                   27282000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  50184                       # Simulator instruction rate (inst/s)
-host_op_rate                                    50180                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               90285398                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 232468                       # Number of bytes of host memory used
-host_seconds                                     0.30                       # Real time elapsed on the host
+host_inst_rate                                  96636                       # Simulator instruction rate (inst/s)
+host_op_rate                                    96628                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              173854426                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 231852                       # Number of bytes of host memory used
+host_seconds                                     0.16                       # Real time elapsed on the host
 sim_insts                                       15162                       # Number of instructions simulated
 sim_ops                                         15162                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             19072                       # Number of bytes read from this memory
@@ -27,14 +27,15 @@ system.physmem.bw_inst_read::total          699068983                       # In
 system.physmem.bw_total::cpu.inst           699068983                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           323729932                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total             1022798915                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                           436                       # Total number of read requests seen
-system.physmem.writeReqs                            0                       # Total number of write requests seen
-system.physmem.cpureqs                            436                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                           436                       # 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                         436                       # 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                        27904                       # Total number of bytes read from memory
 system.physmem.bytesWritten                         0                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                  27904                       # 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                    97                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                    28                       # Track reads on a per bank basis
@@ -195,10 +196,10 @@ system.membus.trans_dist::ReadReq                 351                       # Tr
 system.membus.trans_dist::ReadResp                350                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                85                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               85                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          871                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           871                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        27840                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      27840                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          871                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    871                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        27840                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               27840                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  27840                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              519000                       # Layer occupancy (ticks)
@@ -275,15 +276,15 @@ system.cpu.stage3.utilization                5.274443                       # Pe
 system.cpu.stage4.idleCycles                    45256                       # Number of cycles 0 instructions are processed.
 system.cpu.stage4.runCycles                      9309                       # Number of cycles 1+ instructions are processed.
 system.cpu.stage4.utilization               17.060387                       # Percentage of cycles stage was utilized (processing insts).
-system.cpu.icache.tags.replacements                      0                       # number of replacements
-system.cpu.icache.tags.tagsinuse                168.400745                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     3004                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    299                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                  10.046823                       # 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     168.400745                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.082227                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.082227                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 0                       # number of replacements
+system.cpu.icache.tags.tagsinuse           168.400745                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                3004                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               299                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs             10.046823                       # 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   168.400745                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.082227                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.082227                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         3004                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            3004                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          3004                       # number of demand (read+write) hits
@@ -364,12 +365,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            354                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           352                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           85                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           85                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          600                       # 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                      876                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        19136                       # 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                 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          600                       # 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               876                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        19136                       # 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          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         219500                       # Layer occupancy (ticks)
@@ -378,17 +379,17 @@ system.cpu.toL2Bus.respLayer0.occupancy        507000                       # La
 system.cpu.toL2Bus.respLayer0.utilization          1.9                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy        223250                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          0.8                       # Layer utilization (%)
-system.cpu.l2cache.tags.replacements                     0                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse               199.371038                       # 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                   350                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.005714                       # 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    167.740493                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     31.630545                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          199.371038                       # 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              350                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.005714                       # 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   167.740493                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    31.630545                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.005119                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.000965                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.006084                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.006084                       # 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
@@ -503,15 +504,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 53302.675585
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 57367.753623                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 54586.384439                       # 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                 98.106033                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     3193                       # 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                  23.137681                       # 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      98.106033                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.023952                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.023952                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            98.106033                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                3193                       # 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             23.137681                       # 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    98.106033                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.023952                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.023952                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         2167                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            2167                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data         1020                       # number of WriteReq hits
diff --git a/tests/quick/se/02.insttest/ref/sparc/linux/o3-timing/stats.txt b/tests/quick/se/02.insttest/ref/sparc/linux/o3-timing/stats.txt
index 5128d5dc2..46cdc1496 100644
--- a/tests/quick/se/02.insttest/ref/sparc/linux/o3-timing/stats.txt
+++ b/tests/quick/se/02.insttest/ref/sparc/linux/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000027                       # Nu
 sim_ticks                                    26524500                       # Number of ticks simulated
 final_tick                                   26524500                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  52714                       # Simulator instruction rate (inst/s)
-host_op_rate                                    52709                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               96835127                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 234512                       # Number of bytes of host memory used
-host_seconds                                     0.27                       # Real time elapsed on the host
+host_inst_rate                                  95044                       # Simulator instruction rate (inst/s)
+host_op_rate                                    95035                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              174603061                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 232868                       # Number of bytes of host memory used
+host_seconds                                     0.15                       # Real time elapsed on the host
 sim_insts                                       14436                       # Number of instructions simulated
 sim_ops                                         14436                       # 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          808309299                       # In
 system.physmem.bw_total::cpu.inst           808309299                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           354690946                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total             1163000245                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                           482                       # Total number of read requests seen
-system.physmem.writeReqs                            0                       # Total number of write requests seen
-system.physmem.cpureqs                            482                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                           482                       # 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                         482                       # 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                        30848                       # Total number of bytes read from memory
 system.physmem.bytesWritten                         0                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                  30848                       # 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                   102                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                    29                       # Track reads on a per bank basis
@@ -194,10 +195,10 @@ system.membus.trans_dist::ReadReq                 399                       # Tr
 system.membus.trans_dist::ReadResp                399                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                83                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               83                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          964                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           964                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        30848                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      30848                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          964                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    964                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        30848                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               30848                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  30848                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              608000                       # Layer occupancy (ticks)
@@ -472,12 +473,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            401                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           401                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           83                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           83                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          674                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          294                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      968                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        21568                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side         9408                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 30976                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          674                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          294                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               968                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        21568                       # 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         9408                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          30976                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             30976                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         242000                       # Layer occupancy (ticks)
@@ -486,15 +487,15 @@ system.cpu.toL2Bus.respLayer0.occupancy        570000                       # La
 system.cpu.toL2Bus.respLayer0.utilization          2.1                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy        235750                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          0.9                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                      0                       # number of replacements
-system.cpu.icache.tags.tagsinuse                187.665560                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     4873                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    337                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                  14.459941                       # 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     187.665560                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.091634                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.091634                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 0                       # number of replacements
+system.cpu.icache.tags.tagsinuse           187.665560                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                4873                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               337                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs             14.459941                       # 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   187.665560                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.091634                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.091634                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         4873                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            4873                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          4873                       # number of demand (read+write) hits
@@ -570,17 +571,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 66274.480712
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 66274.480712                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 66274.480712                       # 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               221.542392                       # 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                   399                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.005013                       # 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    187.054257                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     34.488135                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          221.542392                       # 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              399                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.005013                       # 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   187.054257                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    34.488135                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.005708                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001052                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.006761                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.006761                       # 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
@@ -695,15 +696,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 52991.044776
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 57909.863946                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 54491.182573                       # 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                 98.809715                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     4001                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    147                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  27.217687                       # 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      98.809715                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.024123                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.024123                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            98.809715                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                4001                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               147                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             27.217687                       # 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    98.809715                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.024123                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.024123                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         2962                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            2962                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data         1033                       # number of WriteReq hits
diff --git a/tests/quick/se/02.insttest/ref/sparc/linux/simple-timing/stats.txt b/tests/quick/se/02.insttest/ref/sparc/linux/simple-timing/stats.txt
index ac8c29d55..45bd7d946 100644
--- a/tests/quick/se/02.insttest/ref/sparc/linux/simple-timing/stats.txt
+++ b/tests/quick/se/02.insttest/ref/sparc/linux/simple-timing/stats.txt
@@ -4,10 +4,10 @@ sim_seconds                                  0.000041                       # Nu
 sim_ticks                                    41368000                       # Number of ticks simulated
 final_tick                                   41368000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                 479032                       # Simulator instruction rate (inst/s)
-host_op_rate                                   478642                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                             1304958787                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 231320                       # Number of bytes of host memory used
+host_inst_rate                                 554996                       # Simulator instruction rate (inst/s)
+host_op_rate                                   554737                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                             1512828488                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 230824                       # Number of bytes of host memory used
 host_seconds                                     0.03                       # Real time elapsed on the host
 sim_insts                                       15162                       # Number of instructions simulated
 sim_ops                                         15162                       # Number of ops (including micro ops) simulated
@@ -32,10 +32,10 @@ system.membus.trans_dist::ReadReq                 331                       # Tr
 system.membus.trans_dist::ReadResp                331                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                85                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               85                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          832                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           832                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        26624                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      26624                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          832                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    832                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        26624                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               26624                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  26624                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              416000                       # Layer occupancy (ticks)
@@ -65,15 +65,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                      82736                       # 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                      0                       # number of replacements
-system.cpu.icache.tags.tagsinuse                153.782734                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                    14928                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    280                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                  53.314286                       # 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     153.782734                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.075089                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.075089                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 0                       # number of replacements
+system.cpu.icache.tags.tagsinuse           153.782734                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs               14928                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               280                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs             53.314286                       # 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   153.782734                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.075089                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.075089                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst        14928                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total           14928                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst         14928                       # number of demand (read+write) hits
@@ -143,17 +143,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total        52700
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst        52700                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total        52700                       # 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               184.632038                       # 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                   331                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.006042                       # 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    153.110886                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     31.521152                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          184.632038                       # 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              331                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.006042                       # 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   153.110886                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    31.521152                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004673                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.000962                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.005635                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.005635                       # 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
@@ -268,15 +268,15 @@ 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                 97.994344                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     3535                       # 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                  25.615942                       # 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      97.994344                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.023924                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.023924                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse            97.994344                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                3535                       # 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             25.615942                       # 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    97.994344                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.023924                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.023924                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         2172                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            2172                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data         1357                       # number of WriteReq hits
@@ -375,12 +375,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            333                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           333                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           85                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           85                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          560                       # 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                      836                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        17920                       # 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                 26752                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          560                       # 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               836                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        17920                       # 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          26752                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             26752                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         209000                       # Layer occupancy (ticks)