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/alpha/linux/inorder-timing/stats.txt b/tests/quick/se/00.hello/ref/alpha/linux/inorder-timing/stats.txt
index 9728f1e09..dde2aed4b 100644
--- a/tests/quick/se/00.hello/ref/alpha/linux/inorder-timing/stats.txt
+++ b/tests/quick/se/00.hello/ref/alpha/linux/inorder-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000025                       # Nu
 sim_ticks                                    25046000                       # Number of ticks simulated
 final_tick                                   25046000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  25238                       # Simulator instruction rate (inst/s)
-host_op_rate                                    25236                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               98905790                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 225424                       # Number of bytes of host memory used
-host_seconds                                     0.25                       # Real time elapsed on the host
+host_inst_rate                                  22373                       # Simulator instruction rate (inst/s)
+host_op_rate                                    22372                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                               87684145                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 225308                       # Number of bytes of host memory used
+host_seconds                                     0.29                       # Real time elapsed on the host
 sim_insts                                        6390                       # Number of instructions simulated
 sim_ops                                          6390                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             19200                       # Number of bytes read from this memory
@@ -27,14 +27,15 @@ system.physmem.bw_inst_read::total          766589475                       # In
 system.physmem.bw_total::cpu.inst           766589475                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           429290106                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total             1195879582                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                           469                       # Total number of read requests seen
-system.physmem.writeReqs                            0                       # Total number of write requests seen
-system.physmem.cpureqs                            469                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                           469                       # 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                         469                       # 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                        29952                       # Total number of bytes read from memory
 system.physmem.bytesWritten                         0                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                  29952                       # 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                    65                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                    29                       # Track reads on a per bank basis
@@ -195,10 +196,10 @@ system.membus.trans_dist::ReadReq                 396                       # Tr
 system.membus.trans_dist::ReadResp                395                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                73                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               73                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          937                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           937                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        29952                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      29952                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          937                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    937                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        29952                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               29952                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  29952                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              559000                       # Layer occupancy (ticks)
@@ -307,15 +308,15 @@ system.cpu.stage3.utilization                2.663047                       # Pe
 system.cpu.stage4.idleCycles                    45635                       # Number of cycles 0 instructions are processed.
 system.cpu.stage4.runCycles                      4458                       # Number of cycles 1+ instructions are processed.
 system.cpu.stage4.utilization                8.899447                       # Percentage of cycles stage was utilized (processing insts).
-system.cpu.icache.tags.replacements                      0                       # number of replacements
-system.cpu.icache.tags.tagsinuse                141.294375                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                      560                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    301                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                   1.860465                       # 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     141.294375                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.068991                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.068991                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 0                       # number of replacements
+system.cpu.icache.tags.tagsinuse           141.294375                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                 560                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               301                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs              1.860465                       # 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   141.294375                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.068991                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.068991                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst          560                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total             560                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst           560                       # number of demand (read+write) hits
@@ -396,12 +397,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            397                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           396                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           73                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           73                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          603                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          336                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      939                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        19264                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side        10752                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 30016                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          603                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          336                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               939                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        19264                       # 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        10752                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          30016                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             30016                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         235000                       # Layer occupancy (ticks)
@@ -410,17 +411,17 @@ system.cpu.toL2Bus.respLayer0.occupancy        512750                       # La
 system.cpu.toL2Bus.respLayer0.utilization          2.0                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy        278750                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          1.1                       # Layer utilization (%)
-system.cpu.l2cache.tags.replacements                     0                       # number of replacements
-system.cpu.l2cache.tags.tagsinuse               197.784355                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                       1                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   395                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.002532                       # 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    141.343624                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     56.440731                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          197.784355                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                  1                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              395                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.002532                       # 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   141.343624                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    56.440731                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004313                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001722                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.006036                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.006036                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst            1                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total              1                       # number of ReadReq hits
 system.cpu.l2cache.demand_hits::cpu.inst            1                       # number of demand (read+write) hits
@@ -535,15 +536,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 55479.235880
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 58120.535714                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 56425.373134                       # 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                103.103023                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     1601                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    168                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                   9.529762                       # 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     103.103023                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.025172                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.025172                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse           103.103023                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                1601                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               168                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs              9.529762                       # 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   103.103023                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.025172                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.025172                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1086                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1086                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          515                       # number of WriteReq hits
diff --git a/tests/quick/se/00.hello/ref/alpha/linux/o3-timing/stats.txt b/tests/quick/se/00.hello/ref/alpha/linux/o3-timing/stats.txt
index 38483afa5..60f469d0d 100644
--- a/tests/quick/se/00.hello/ref/alpha/linux/o3-timing/stats.txt
+++ b/tests/quick/se/00.hello/ref/alpha/linux/o3-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000021                       # Nu
 sim_ticks                                    20671000                       # Number of ticks simulated
 final_tick                                   20671000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  25591                       # Simulator instruction rate (inst/s)
-host_op_rate                                    25589                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                               83008053                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 227468                       # Number of bytes of host memory used
-host_seconds                                     0.25                       # Real time elapsed on the host
+host_inst_rate                                  24570                       # Simulator instruction rate (inst/s)
+host_op_rate                                    24568                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                               79697022                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 227340                       # Number of bytes of host memory used
+host_seconds                                     0.26                       # Real time elapsed on the host
 sim_insts                                        6372                       # Number of instructions simulated
 sim_ops                                          6372                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             20032                       # Number of bytes read from this memory
@@ -27,14 +27,15 @@ system.physmem.bw_inst_read::total          969087127                       # In
 system.physmem.bw_total::cpu.inst           969087127                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::cpu.data           538725751                       # Total bandwidth to/from this memory (bytes/s)
 system.physmem.bw_total::total             1507812878                       # Total bandwidth to/from this memory (bytes/s)
-system.physmem.readReqs                           488                       # Total number of read requests seen
-system.physmem.writeReqs                            0                       # Total number of write requests seen
-system.physmem.cpureqs                            488                       # Reqs generatd by CPU via cache - shady
+system.physmem.readReqs                           488                       # 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                         488                       # 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                        31168                       # Total number of bytes read from memory
 system.physmem.bytesWritten                         0                       # Total number of bytes written to memory
 system.physmem.bytesConsumedRd                  31168                       # 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                    69                       # Track reads on a per bank basis
 system.physmem.perBankRdReqs::1                    34                       # Track reads on a per bank basis
@@ -194,10 +195,10 @@ system.membus.trans_dist::ReadReq                 415                       # Tr
 system.membus.trans_dist::ReadResp                414                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                73                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               73                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          975                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           975                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        31168                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      31168                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          975                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    975                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        31168                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               31168                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  31168                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              619500                       # Layer occupancy (ticks)
@@ -511,12 +512,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            416                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           415                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           73                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           73                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          629                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          348                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      977                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        20096                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side        11136                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 31232                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          629                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          348                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               977                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        20096                       # 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        11136                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          31232                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             31232                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         244500                       # Layer occupancy (ticks)
@@ -525,15 +526,15 @@ system.cpu.toL2Bus.respLayer0.occupancy        531250                       # La
 system.cpu.toL2Bus.respLayer0.utilization          2.6                       # Layer utilization (%)
 system.cpu.toL2Bus.respLayer1.occupancy        281250                       # Layer occupancy (ticks)
 system.cpu.toL2Bus.respLayer1.utilization          1.4                       # Layer utilization (%)
-system.cpu.icache.tags.replacements                      0                       # number of replacements
-system.cpu.icache.tags.tagsinuse                159.268512                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     1898                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    314                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                   6.044586                       # 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     159.268512                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.077768                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.077768                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 0                       # number of replacements
+system.cpu.icache.tags.tagsinuse           159.268512                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                1898                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               314                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs              6.044586                       # 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   159.268512                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.077768                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.077768                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         1898                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            1898                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          1898                       # number of demand (read+write) hits
@@ -609,17 +610,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 67819.841270
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 67819.841270                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 67819.841270                       # 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               218.982908                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                       1                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   414                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.002415                       # 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    159.353389                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     59.629519                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          218.982908                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                  1                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              414                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.002415                       # 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   159.353389                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    59.629519                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.004863                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001820                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.006683                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.006683                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst            1                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total              1                       # number of ReadReq hits
 system.cpu.l2cache.demand_hits::cpu.inst            1                       # number of demand (read+write) hits
@@ -734,15 +735,15 @@ system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.inst 54395.700637
 system.cpu.l2cache.overall_avg_mshr_miss_latency::cpu.data 62693.965517                       # average overall mshr miss latency
 system.cpu.l2cache.overall_avg_mshr_miss_latency::total 57354.508197                       # 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                106.762654                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     2236                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    174                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  12.850575                       # 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     106.762654                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.026065                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.026065                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse           106.762654                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                2236                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               174                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             12.850575                       # 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   106.762654                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.026065                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.026065                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1730                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1730                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          506                       # number of WriteReq hits
diff --git a/tests/quick/se/00.hello/ref/alpha/linux/simple-timing/stats.txt b/tests/quick/se/00.hello/ref/alpha/linux/simple-timing/stats.txt
index 2ce4c669d..6038d0a3c 100644
--- a/tests/quick/se/00.hello/ref/alpha/linux/simple-timing/stats.txt
+++ b/tests/quick/se/00.hello/ref/alpha/linux/simple-timing/stats.txt
@@ -4,11 +4,11 @@ sim_seconds                                  0.000033                       # Nu
 sim_ticks                                    32544000                       # Number of ticks simulated
 final_tick                                   32544000                       # Number of ticks from beginning of simulation (restored from checkpoints and never reset)
 sim_freq                                 1000000000000                       # Frequency of simulated ticks
-host_inst_rate                                  19861                       # Simulator instruction rate (inst/s)
-host_op_rate                                    19860                       # Simulator op (including micro ops) rate (op/s)
-host_tick_rate                              101141711                       # Simulator tick rate (ticks/s)
-host_mem_usage                                 224276                       # Number of bytes of host memory used
-host_seconds                                     0.32                       # Real time elapsed on the host
+host_inst_rate                                  27670                       # Simulator instruction rate (inst/s)
+host_op_rate                                    27667                       # Simulator op (including micro ops) rate (op/s)
+host_tick_rate                              140894748                       # Simulator tick rate (ticks/s)
+host_mem_usage                                 224272                       # Number of bytes of host memory used
+host_seconds                                     0.23                       # Real time elapsed on the host
 sim_insts                                        6390                       # Number of instructions simulated
 sim_ops                                          6390                       # Number of ops (including micro ops) simulated
 system.physmem.bytes_read::cpu.inst             17792                       # Number of bytes read from this memory
@@ -32,10 +32,10 @@ system.membus.trans_dist::ReadReq                 373                       # Tr
 system.membus.trans_dist::ReadResp                373                       # Transaction distribution
 system.membus.trans_dist::ReadExReq                73                       # Transaction distribution
 system.membus.trans_dist::ReadExResp               73                       # Transaction distribution
-system.membus.pkt_count_system.cpu.l2cache.mem_side          892                       # Packet count per connected master and slave (bytes)
-system.membus.pkt_count                           892                       # Packet count per connected master and slave (bytes)
-system.membus.tot_pkt_size_system.cpu.l2cache.mem_side        28544                       # Cumulative packet size per connected master and slave (bytes)
-system.membus.tot_pkt_size                      28544                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.pkt_count_system.cpu.l2cache.mem_side::system.physmem.port          892                       # Packet count per connected master and slave (bytes)
+system.membus.pkt_count::total                    892                       # Packet count per connected master and slave (bytes)
+system.membus.tot_pkt_size_system.cpu.l2cache.mem_side::system.physmem.port        28544                       # Cumulative packet size per connected master and slave (bytes)
+system.membus.tot_pkt_size::total               28544                       # Cumulative packet size per connected master and slave (bytes)
 system.membus.data_through_bus                  28544                       # Total data (bytes)
 system.membus.snoop_data_through_bus                0                       # Total snoop data (bytes)
 system.membus.reqLayer0.occupancy              446000                       # Layer occupancy (ticks)
@@ -97,15 +97,15 @@ system.cpu.num_idle_cycles                          0                       # Nu
 system.cpu.num_busy_cycles                      65088                       # 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                127.998991                       # Cycle average of tags in use
-system.cpu.icache.tags.total_refs                     6122                       # Total number of references to valid blocks.
-system.cpu.icache.tags.sampled_refs                    279                       # Sample count of references to valid blocks.
-system.cpu.icache.tags.avg_refs                  21.942652                       # 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     127.998991                       # Average occupied blocks per requestor
-system.cpu.icache.tags.occ_percent::cpu.inst      0.062500                       # Average percentage of cache occupancy
-system.cpu.icache.tags.occ_percent::total         0.062500                       # Average percentage of cache occupancy
+system.cpu.icache.tags.replacements                 0                       # number of replacements
+system.cpu.icache.tags.tagsinuse           127.998991                       # Cycle average of tags in use
+system.cpu.icache.tags.total_refs                6122                       # Total number of references to valid blocks.
+system.cpu.icache.tags.sampled_refs               279                       # Sample count of references to valid blocks.
+system.cpu.icache.tags.avg_refs             21.942652                       # 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   127.998991                       # Average occupied blocks per requestor
+system.cpu.icache.tags.occ_percent::cpu.inst     0.062500                       # Average percentage of cache occupancy
+system.cpu.icache.tags.occ_percent::total     0.062500                       # Average percentage of cache occupancy
 system.cpu.icache.ReadReq_hits::cpu.inst         6122                       # number of ReadReq hits
 system.cpu.icache.ReadReq_hits::total            6122                       # number of ReadReq hits
 system.cpu.icache.demand_hits::cpu.inst          6122                       # number of demand (read+write) hits
@@ -175,17 +175,17 @@ system.cpu.icache.demand_avg_mshr_miss_latency::total 52849.462366
 system.cpu.icache.overall_avg_mshr_miss_latency::cpu.inst 52849.462366                       # average overall mshr miss latency
 system.cpu.icache.overall_avg_mshr_miss_latency::total 52849.462366                       # 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               184.497210                       # Cycle average of tags in use
-system.cpu.l2cache.tags.total_refs                       1                       # Total number of references to valid blocks.
-system.cpu.l2cache.tags.sampled_refs                   373                       # Sample count of references to valid blocks.
-system.cpu.l2cache.tags.avg_refs                  0.002681                       # 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    128.017765                       # Average occupied blocks per requestor
-system.cpu.l2cache.tags.occ_blocks::cpu.data     56.479444                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.replacements                0                       # number of replacements
+system.cpu.l2cache.tags.tagsinuse          184.497210                       # Cycle average of tags in use
+system.cpu.l2cache.tags.total_refs                  1                       # Total number of references to valid blocks.
+system.cpu.l2cache.tags.sampled_refs              373                       # Sample count of references to valid blocks.
+system.cpu.l2cache.tags.avg_refs             0.002681                       # 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   128.017765                       # Average occupied blocks per requestor
+system.cpu.l2cache.tags.occ_blocks::cpu.data    56.479444                       # Average occupied blocks per requestor
 system.cpu.l2cache.tags.occ_percent::cpu.inst     0.003907                       # Average percentage of cache occupancy
 system.cpu.l2cache.tags.occ_percent::cpu.data     0.001724                       # Average percentage of cache occupancy
-system.cpu.l2cache.tags.occ_percent::total        0.005630                       # Average percentage of cache occupancy
+system.cpu.l2cache.tags.occ_percent::total     0.005630                       # Average percentage of cache occupancy
 system.cpu.l2cache.ReadReq_hits::cpu.inst            1                       # number of ReadReq hits
 system.cpu.l2cache.ReadReq_hits::total              1                       # number of ReadReq hits
 system.cpu.l2cache.demand_hits::cpu.inst            1                       # number of demand (read+write) hits
@@ -300,15 +300,15 @@ 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                103.762109                       # Cycle average of tags in use
-system.cpu.dcache.tags.total_refs                     1880                       # Total number of references to valid blocks.
-system.cpu.dcache.tags.sampled_refs                    168                       # Sample count of references to valid blocks.
-system.cpu.dcache.tags.avg_refs                  11.190476                       # 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     103.762109                       # Average occupied blocks per requestor
-system.cpu.dcache.tags.occ_percent::cpu.data      0.025333                       # Average percentage of cache occupancy
-system.cpu.dcache.tags.occ_percent::total         0.025333                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.replacements                 0                       # number of replacements
+system.cpu.dcache.tags.tagsinuse           103.762109                       # Cycle average of tags in use
+system.cpu.dcache.tags.total_refs                1880                       # Total number of references to valid blocks.
+system.cpu.dcache.tags.sampled_refs               168                       # Sample count of references to valid blocks.
+system.cpu.dcache.tags.avg_refs             11.190476                       # 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   103.762109                       # Average occupied blocks per requestor
+system.cpu.dcache.tags.occ_percent::cpu.data     0.025333                       # Average percentage of cache occupancy
+system.cpu.dcache.tags.occ_percent::total     0.025333                       # Average percentage of cache occupancy
 system.cpu.dcache.ReadReq_hits::cpu.data         1088                       # number of ReadReq hits
 system.cpu.dcache.ReadReq_hits::total            1088                       # number of ReadReq hits
 system.cpu.dcache.WriteReq_hits::cpu.data          792                       # number of WriteReq hits
@@ -403,12 +403,12 @@ system.cpu.toL2Bus.trans_dist::ReadReq            374                       # Tr
 system.cpu.toL2Bus.trans_dist::ReadResp           374                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExReq           73                       # Transaction distribution
 system.cpu.toL2Bus.trans_dist::ReadExResp           73                       # Transaction distribution
-system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side          558                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side          336                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.pkt_count                      894                       # Packet count per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side        17856                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size_system.cpu.dcache.mem_side        10752                       # Cumulative packet size per connected master and slave (bytes)
-system.cpu.toL2Bus.tot_pkt_size                 28608                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side          558                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count_system.cpu.dcache.mem_side::system.cpu.l2cache.cpu_side          336                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.pkt_count::total               894                       # Packet count per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size_system.cpu.icache.mem_side::system.cpu.l2cache.cpu_side        17856                       # 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        10752                       # Cumulative packet size per connected master and slave (bytes)
+system.cpu.toL2Bus.tot_pkt_size::total          28608                       # Cumulative packet size per connected master and slave (bytes)
 system.cpu.toL2Bus.data_through_bus             28608                       # Total data (bytes)
 system.cpu.toL2Bus.snoop_data_through_bus            0                       # Total snoop data (bytes)
 system.cpu.toL2Bus.reqLayer0.occupancy         223500                       # Layer occupancy (ticks)