misc: Merge branch 'release-staging-v20.0.0.0' into develop
[gem5.git] / configs / common / Simulation.py
index 1c9d4ff4ebbf311ee47309d7e02b936466c19541..e53c755c6c09e1e5528158eaa29966641cd1a8bc 100644 (file)
@@ -1,4 +1,17 @@
+# Copyright (c) 2012-2013 ARM Limited
+# All rights reserved
+#
+# The license below extends only to copyright in the software and shall
+# not be construed as granting a license to any other intellectual
+# property including but not limited to intellectual property relating
+# to a hardware implementation of the functionality of the software
+# licensed hereunder.  You may use the software subject to the license
+# terms below provided that you ensure that this notice is replicated
+# unmodified and in its entirety in all distributions of the software,
+# modified or unmodified, in source code or in binary form.
+#
 # Copyright (c) 2006-2008 The Regents of The University of Michigan
+# Copyright (c) 2010 Advanced Micro Devices, Inc.
 # All rights reserved.
 #
 # Redistribution and use in source and binary forms, with or without
 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-#
-# Authors: Lisa Hsu
 
+from __future__ import print_function
+from __future__ import absolute_import
+
+import six
+import sys
 from os import getcwd
 from os.path import join as joinpath
 
+from common import CpuConfig
+from common import ObjectList
+
 import m5
 from m5.defines import buildEnv
 from m5.objects import *
 from m5.util import *
 
+if six.PY3:
+    long = int
+
 addToPath('../common')
 
+def getCPUClass(cpu_type):
+    """Returns the required cpu class and the mode of operation."""
+    cls = ObjectList.cpu_list.get(cpu_type)
+    return cls, cls.memory_mode()
+
 def setCPUClass(options):
+    """Returns two cpu classes and the initial mode of operation.
 
-    atomic = False
-    if options.timing:
-        class TmpClass(TimingSimpleCPU): pass
-    elif options.detailed:
-        if not options.caches:
-            print "O3 CPU must be used with caches"
-            sys.exit(1)
-        class TmpClass(DerivO3CPU): pass
-    elif options.inorder:
-        if not options.caches:
-            print "InOrder CPU must be used with caches"
-            sys.exit(1)
-        class TmpClass(InOrderCPU): pass
-    else:
-        class TmpClass(AtomicSimpleCPU): pass
-        atomic = True
+       Restoring from a checkpoint or fast forwarding through a benchmark
+       can be done using one type of cpu, and then the actual
+       simulation can be carried out using another type. This function
+       returns these two types of cpus and the initial mode of operation
+       depending on the options provided.
+    """
 
+    TmpClass, test_mem_mode = getCPUClass(options.cpu_type)
     CPUClass = None
-    test_mem_mode = 'atomic'
+    if TmpClass.require_caches() and \
+            not options.caches and not options.ruby:
+        fatal("%s must be used with caches" % options.cpu_type)
 
-    if not atomic:
-        if options.checkpoint_restore != None or options.fast_forward:
+    if options.checkpoint_restore != None:
+        if options.restore_with_cpu != options.cpu_type:
             CPUClass = TmpClass
-            class TmpClass(AtomicSimpleCPU): pass
-        else:
-            test_mem_mode = 'timing'
+            TmpClass, test_mem_mode = getCPUClass(options.restore_with_cpu)
+    elif options.fast_forward:
+        CPUClass = TmpClass
+        TmpClass = AtomicSimpleCPU
+        test_mem_mode = 'atomic'
+
+    # Ruby only supports atomic accesses in noncaching mode
+    if test_mem_mode == 'atomic' and options.ruby:
+        warn("Memory mode will be changed to atomic_noncaching")
+        test_mem_mode = 'atomic_noncaching'
 
     return (TmpClass, test_mem_mode, CPUClass)
 
+def setMemClass(options):
+    """Returns a memory controller class."""
+
+    return ObjectList.mem_list.get(options.mem_type)
+
+def setWorkCountOptions(system, options):
+    if options.work_item_id != None:
+        system.work_item_id = options.work_item_id
+    if options.num_work_ids != None:
+        system.num_work_ids = options.num_work_ids
+    if options.work_begin_cpu_id_exit != None:
+        system.work_begin_cpu_id_exit = options.work_begin_cpu_id_exit
+    if options.work_end_exit_count != None:
+        system.work_end_exit_count = options.work_end_exit_count
+    if options.work_end_checkpoint_count != None:
+        system.work_end_ckpt_count = options.work_end_checkpoint_count
+    if options.work_begin_exit_count != None:
+        system.work_begin_exit_count = options.work_begin_exit_count
+    if options.work_begin_checkpoint_count != None:
+        system.work_begin_ckpt_count = options.work_begin_checkpoint_count
+    if options.work_cpus_checkpoint_count != None:
+        system.work_cpus_ckpt_count = options.work_cpus_checkpoint_count
+
+def findCptDir(options, cptdir, testsys):
+    """Figures out the directory from which the checkpointed state is read.
+
+    There are two different ways in which the directories holding checkpoints
+    can be named --
+    1. cpt.<benchmark name>.<instruction count when the checkpoint was taken>
+    2. cpt.<some number, usually the tick value when the checkpoint was taken>
+
+    This function parses through the options to figure out which one of the
+    above should be used for selecting the checkpoint, and then figures out
+    the appropriate directory.
+    """
+
+    from os.path import isdir, exists
+    from os import listdir
+    import re
+
+    if not isdir(cptdir):
+        fatal("checkpoint dir %s does not exist!", cptdir)
+
+    cpt_starttick = 0
+    if options.at_instruction or options.simpoint:
+        inst = options.checkpoint_restore
+        if options.simpoint:
+            # assume workload 0 has the simpoint
+            if testsys.cpu[0].workload[0].simpoint == 0:
+                fatal('Unable to find simpoint')
+            inst += int(testsys.cpu[0].workload[0].simpoint)
+
+        checkpoint_dir = joinpath(cptdir, "cpt.%s.%s" % (options.bench, inst))
+        if not exists(checkpoint_dir):
+            fatal("Unable to find checkpoint directory %s", checkpoint_dir)
+
+    elif options.restore_simpoint_checkpoint:
+        # Restore from SimPoint checkpoints
+        # Assumes that the checkpoint dir names are formatted as follows:
+        dirs = listdir(cptdir)
+        expr = re.compile('cpt\.simpoint_(\d+)_inst_(\d+)' +
+                    '_weight_([\d\.e\-]+)_interval_(\d+)_warmup_(\d+)')
+        cpts = []
+        for dir in dirs:
+            match = expr.match(dir)
+            if match:
+                cpts.append(dir)
+        cpts.sort()
+
+        cpt_num = options.checkpoint_restore
+        if cpt_num > len(cpts):
+            fatal('Checkpoint %d not found', cpt_num)
+        checkpoint_dir = joinpath(cptdir, cpts[cpt_num - 1])
+        match = expr.match(cpts[cpt_num - 1])
+        if match:
+            index = int(match.group(1))
+            start_inst = int(match.group(2))
+            weight_inst = float(match.group(3))
+            interval_length = int(match.group(4))
+            warmup_length = int(match.group(5))
+        print("Resuming from", checkpoint_dir)
+        simpoint_start_insts = []
+        simpoint_start_insts.append(warmup_length)
+        simpoint_start_insts.append(warmup_length + interval_length)
+        testsys.cpu[0].simpoint_start_insts = simpoint_start_insts
+        if testsys.switch_cpus != None:
+            testsys.switch_cpus[0].simpoint_start_insts = simpoint_start_insts
+
+        print("Resuming from SimPoint", end=' ')
+        print("#%d, start_inst:%d, weight:%f, interval:%d, warmup:%d" %
+            (index, start_inst, weight_inst, interval_length, warmup_length))
 
-def run(options, root, testsys, cpu_class):
-    if options.maxtick:
-        maxtick = options.maxtick
-    elif options.maxtime:
-        simtime = m5.ticks.seconds(simtime)
-        print "simulating for: ", simtime
-        maxtick = simtime
     else:
-        maxtick = m5.MaxTick
+        dirs = listdir(cptdir)
+        expr = re.compile('cpt\.([0-9]+)')
+        cpts = []
+        for dir in dirs:
+            match = expr.match(dir)
+            if match:
+                cpts.append(match.group(1))
+
+        cpts.sort(lambda a,b: cmp(long(a), long(b)))
+
+        cpt_num = options.checkpoint_restore
+        if cpt_num > len(cpts):
+            fatal('Checkpoint %d not found', cpt_num)
+
+        cpt_starttick = int(cpts[cpt_num - 1])
+        checkpoint_dir = joinpath(cptdir, "cpt.%s" % cpts[cpt_num - 1])
+
+    return cpt_starttick, checkpoint_dir
+
+def scriptCheckpoints(options, maxtick, cptdir):
+    if options.at_instruction or options.simpoint:
+        checkpoint_inst = int(options.take_checkpoints)
+
+        # maintain correct offset if we restored from some instruction
+        if options.checkpoint_restore != None:
+            checkpoint_inst += options.checkpoint_restore
+
+        print("Creating checkpoint at inst:%d" % (checkpoint_inst))
+        exit_event = m5.simulate()
+        exit_cause = exit_event.getCause()
+        print("exit cause = %s" % exit_cause)
+
+        # skip checkpoint instructions should they exist
+        while exit_cause == "checkpoint":
+            exit_event = m5.simulate()
+            exit_cause = exit_event.getCause()
+
+        if exit_cause == "a thread reached the max instruction count":
+            m5.checkpoint(joinpath(cptdir, "cpt.%s.%d" % \
+                    (options.bench, checkpoint_inst)))
+            print("Checkpoint written.")
+
+    else:
+        when, period = options.take_checkpoints.split(",", 1)
+        when = int(when)
+        period = int(period)
+        num_checkpoints = 0
+
+        exit_event = m5.simulate(when - m5.curTick())
+        exit_cause = exit_event.getCause()
+        while exit_cause == "checkpoint":
+            exit_event = m5.simulate(when - m5.curTick())
+            exit_cause = exit_event.getCause()
+
+        if exit_cause == "simulate() limit reached":
+            m5.checkpoint(joinpath(cptdir, "cpt.%d"))
+            num_checkpoints += 1
+
+        sim_ticks = when
+        max_checkpoints = options.max_checkpoints
+
+        while num_checkpoints < max_checkpoints and \
+                exit_cause == "simulate() limit reached":
+            if (sim_ticks + period) > maxtick:
+                exit_event = m5.simulate(maxtick - sim_ticks)
+                exit_cause = exit_event.getCause()
+                break
+            else:
+                exit_event = m5.simulate(period)
+                exit_cause = exit_event.getCause()
+                sim_ticks += period
+                while exit_event.getCause() == "checkpoint":
+                    exit_event = m5.simulate(sim_ticks - m5.curTick())
+                if exit_event.getCause() == "simulate() limit reached":
+                    m5.checkpoint(joinpath(cptdir, "cpt.%d"))
+                    num_checkpoints += 1
+
+    return exit_event
+
+def benchCheckpoints(options, maxtick, cptdir):
+    exit_event = m5.simulate(maxtick - m5.curTick())
+    exit_cause = exit_event.getCause()
+
+    num_checkpoints = 0
+    max_checkpoints = options.max_checkpoints
+
+    while exit_cause == "checkpoint":
+        m5.checkpoint(joinpath(cptdir, "cpt.%d"))
+        num_checkpoints += 1
+        if num_checkpoints == max_checkpoints:
+            exit_cause = "maximum %d checkpoints dropped" % max_checkpoints
+            break
+
+        exit_event = m5.simulate(maxtick - m5.curTick())
+        exit_cause = exit_event.getCause()
+
+    return exit_event
+
+# Set up environment for taking SimPoint checkpoints
+# Expecting SimPoint files generated by SimPoint 3.2
+def parseSimpointAnalysisFile(options, testsys):
+    import re
+
+    simpoint_filename, weight_filename, interval_length, warmup_length = \
+        options.take_simpoint_checkpoints.split(",", 3)
+    print("simpoint analysis file:", simpoint_filename)
+    print("simpoint weight file:", weight_filename)
+    print("interval length:", interval_length)
+    print("warmup length:", warmup_length)
+
+    interval_length = int(interval_length)
+    warmup_length = int(warmup_length)
+
+    # Simpoint analysis output starts interval counts with 0.
+    simpoints = []
+    simpoint_start_insts = []
+
+    # Read in SimPoint analysis files
+    simpoint_file = open(simpoint_filename)
+    weight_file = open(weight_filename)
+    while True:
+        line = simpoint_file.readline()
+        if not line:
+            break
+        m = re.match("(\d+)\s+(\d+)", line)
+        if m:
+            interval = int(m.group(1))
+        else:
+            fatal('unrecognized line in simpoint file!')
+
+        line = weight_file.readline()
+        if not line:
+            fatal('not enough lines in simpoint weight file!')
+        m = re.match("([0-9\.e\-]+)\s+(\d+)", line)
+        if m:
+            weight = float(m.group(1))
+        else:
+            fatal('unrecognized line in simpoint weight file!')
+
+        if (interval * interval_length - warmup_length > 0):
+            starting_inst_count = \
+                interval * interval_length - warmup_length
+            actual_warmup_length = warmup_length
+        else:
+            # Not enough room for proper warmup
+            # Just starting from the beginning
+            starting_inst_count = 0
+            actual_warmup_length = interval * interval_length
+
+        simpoints.append((interval, weight, starting_inst_count,
+            actual_warmup_length))
+
+    # Sort SimPoints by starting inst count
+    simpoints.sort(key=lambda obj: obj[2])
+    for s in simpoints:
+        interval, weight, starting_inst_count, actual_warmup_length = s
+        print(str(interval), str(weight), starting_inst_count,
+            actual_warmup_length)
+        simpoint_start_insts.append(starting_inst_count)
+
+    print("Total # of simpoints:", len(simpoints))
+    testsys.cpu[0].simpoint_start_insts = simpoint_start_insts
+
+    return (simpoints, interval_length)
+
+def takeSimpointCheckpoints(simpoints, interval_length, cptdir):
+    num_checkpoints = 0
+    index = 0
+    last_chkpnt_inst_count = -1
+    for simpoint in simpoints:
+        interval, weight, starting_inst_count, actual_warmup_length = simpoint
+        if starting_inst_count == last_chkpnt_inst_count:
+            # checkpoint starting point same as last time
+            # (when warmup period longer than starting point)
+            exit_cause = "simpoint starting point found"
+            code = 0
+        else:
+            exit_event = m5.simulate()
+
+            # skip checkpoint instructions should they exist
+            while exit_event.getCause() == "checkpoint":
+                print("Found 'checkpoint' exit event...ignoring...")
+                exit_event = m5.simulate()
+
+            exit_cause = exit_event.getCause()
+            code = exit_event.getCode()
+
+        if exit_cause == "simpoint starting point found":
+            m5.checkpoint(joinpath(cptdir,
+                "cpt.simpoint_%02d_inst_%d_weight_%f_interval_%d_warmup_%d"
+                % (index, starting_inst_count, weight, interval_length,
+                actual_warmup_length)))
+            print("Checkpoint #%d written. start inst:%d weight:%f" %
+                (num_checkpoints, starting_inst_count, weight))
+            num_checkpoints += 1
+            last_chkpnt_inst_count = starting_inst_count
+        else:
+            break
+        index += 1
 
+    print('Exiting @ tick %i because %s' % (m5.curTick(), exit_cause))
+    print("%d checkpoints taken" % num_checkpoints)
+    sys.exit(code)
+
+def restoreSimpointCheckpoint():
+    exit_event = m5.simulate()
+    exit_cause = exit_event.getCause()
+
+    if exit_cause == "simpoint starting point found":
+        print("Warmed up! Dumping and resetting stats!")
+        m5.stats.dump()
+        m5.stats.reset()
+
+        exit_event = m5.simulate()
+        exit_cause = exit_event.getCause()
+
+        if exit_cause == "simpoint starting point found":
+            print("Done running SimPoint!")
+            sys.exit(exit_event.getCode())
+
+    print('Exiting @ tick %i because %s' % (m5.curTick(), exit_cause))
+    sys.exit(exit_event.getCode())
+
+def repeatSwitch(testsys, repeat_switch_cpu_list, maxtick, switch_freq):
+    print("starting switch loop")
+    while True:
+        exit_event = m5.simulate(switch_freq)
+        exit_cause = exit_event.getCause()
+
+        if exit_cause != "simulate() limit reached":
+            return exit_event
+
+        m5.switchCpus(testsys, repeat_switch_cpu_list)
+
+        tmp_cpu_list = []
+        for old_cpu, new_cpu in repeat_switch_cpu_list:
+            tmp_cpu_list.append((new_cpu, old_cpu))
+        repeat_switch_cpu_list = tmp_cpu_list
+
+        if (maxtick - m5.curTick()) <= switch_freq:
+            exit_event = m5.simulate(maxtick - m5.curTick())
+            return exit_event
+
+def run(options, root, testsys, cpu_class):
     if options.checkpoint_dir:
         cptdir = options.checkpoint_dir
     elif m5.options.outdir:
@@ -91,50 +445,108 @@ def run(options, root, testsys, cpu_class):
     if options.standard_switch and not options.caches:
         fatal("Must specify --caches when using --standard-switch")
 
+    if options.standard_switch and options.repeat_switch:
+        fatal("Can't specify both --standard-switch and --repeat-switch")
+
+    if options.repeat_switch and options.take_checkpoints:
+        fatal("Can't specify both --repeat-switch and --take-checkpoints")
+
     np = options.num_cpus
-    max_checkpoints = options.max_checkpoints
     switch_cpus = None
 
-    if options.prog_intvl:
-        for i in xrange(np):
-            testsys.cpu[i].progress_interval = options.prog_intvl
+    if options.prog_interval:
+        for i in range(np):
+            testsys.cpu[i].progress_interval = options.prog_interval
 
     if options.maxinsts:
-        for i in xrange(np):
+        for i in range(np):
             testsys.cpu[i].max_insts_any_thread = options.maxinsts
 
     if cpu_class:
-        switch_cpus = [cpu_class(defer_registration=True, cpu_id=(np+i))
-                       for i in xrange(np)]
+        switch_cpus = [cpu_class(switched_out=True, cpu_id=(i))
+                       for i in range(np)]
 
-        for i in xrange(np):
+        for i in range(np):
             if options.fast_forward:
                 testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
-            switch_cpus[i].system =  testsys
-            if not buildEnv['FULL_SYSTEM']:
-                switch_cpus[i].workload = testsys.cpu[i].workload
-            switch_cpus[i].clock = testsys.cpu[0].clock
+            switch_cpus[i].system = testsys
+            switch_cpus[i].workload = testsys.cpu[i].workload
+            switch_cpus[i].clk_domain = testsys.cpu[i].clk_domain
+            switch_cpus[i].progress_interval = \
+                testsys.cpu[i].progress_interval
+            switch_cpus[i].isa = testsys.cpu[i].isa
             # simulation period
-            if options.max_inst:
-                switch_cpus[i].max_insts_any_thread = options.max_inst
+            if options.maxinsts:
+                switch_cpus[i].max_insts_any_thread = options.maxinsts
+            # Add checker cpu if selected
+            if options.checker:
+                switch_cpus[i].addCheckerCpu()
+            if options.bp_type:
+                bpClass = ObjectList.bp_list.get(options.bp_type)
+                switch_cpus[i].branchPred = bpClass()
+            if options.indirect_bp_type:
+                IndirectBPClass = ObjectList.indirect_bp_list.get(
+                    options.indirect_bp_type)
+                switch_cpus[i].branchPred.indirectBranchPred = \
+                    IndirectBPClass()
+
+        # If elastic tracing is enabled attach the elastic trace probe
+        # to the switch CPUs
+        if options.elastic_trace_en:
+            CpuConfig.config_etrace(cpu_class, switch_cpus, options)
 
         testsys.switch_cpus = switch_cpus
-        switch_cpu_list = [(testsys.cpu[i], switch_cpus[i]) for i in xrange(np)]
+        switch_cpu_list = [(testsys.cpu[i], switch_cpus[i]) for i in range(np)]
+
+    if options.repeat_switch:
+        switch_class = getCPUClass(options.cpu_type)[0]
+        if switch_class.require_caches() and \
+                not options.caches:
+            print("%s: Must be used with caches" % str(switch_class))
+            sys.exit(1)
+        if not switch_class.support_take_over():
+            print("%s: CPU switching not supported" % str(switch_class))
+            sys.exit(1)
+
+        repeat_switch_cpus = [switch_class(switched_out=True, \
+                                               cpu_id=(i)) for i in range(np)]
+
+        for i in range(np):
+            repeat_switch_cpus[i].system = testsys
+            repeat_switch_cpus[i].workload = testsys.cpu[i].workload
+            repeat_switch_cpus[i].clk_domain = testsys.cpu[i].clk_domain
+            repeat_switch_cpus[i].isa = testsys.cpu[i].isa
+
+            if options.maxinsts:
+                repeat_switch_cpus[i].max_insts_any_thread = options.maxinsts
+
+            if options.checker:
+                repeat_switch_cpus[i].addCheckerCpu()
+
+        testsys.repeat_switch_cpus = repeat_switch_cpus
+
+        if cpu_class:
+            repeat_switch_cpu_list = [(switch_cpus[i], repeat_switch_cpus[i])
+                                      for i in range(np)]
+        else:
+            repeat_switch_cpu_list = [(testsys.cpu[i], repeat_switch_cpus[i])
+                                      for i in range(np)]
 
     if options.standard_switch:
-        switch_cpus = [TimingSimpleCPU(defer_registration=True, cpu_id=(np+i))
-                       for i in xrange(np)]
-        switch_cpus_1 = [DerivO3CPU(defer_registration=True, cpu_id=(2*np+i))
-                        for i in xrange(np)]
+        switch_cpus = [TimingSimpleCPU(switched_out=True, cpu_id=(i))
+                       for i in range(np)]
+        switch_cpus_1 = [DerivO3CPU(switched_out=True, cpu_id=(i))
+                        for i in range(np)]
 
-        for i in xrange(np):
+        for i in range(np):
             switch_cpus[i].system =  testsys
             switch_cpus_1[i].system =  testsys
-            if not buildEnv['FULL_SYSTEM']:
-                switch_cpus[i].workload = testsys.cpu[i].workload
-                switch_cpus_1[i].workload = testsys.cpu[i].workload
-            switch_cpus[i].clock = testsys.cpu[0].clock
-            switch_cpus_1[i].clock = testsys.cpu[0].clock
+            switch_cpus[i].workload = testsys.cpu[i].workload
+            switch_cpus_1[i].workload = testsys.cpu[i].workload
+            switch_cpus[i].clk_domain = testsys.cpu[i].clk_domain
+            switch_cpus_1[i].clk_domain = testsys.cpu[i].clk_domain
+            switch_cpus[i].isa = testsys.cpu[i].isa
+            switch_cpus_1[i].isa = testsys.cpu[i].isa
 
             # if restoring, make atomic cpu simulate only a few instructions
             if options.checkpoint_restore != None:
@@ -157,18 +569,22 @@ def run(options, root, testsys, cpu_class):
                 switch_cpus[i].max_insts_any_thread =  options.warmup_insts
 
             # simulation period
-            if options.max_inst:
-                switch_cpus_1[i].max_insts_any_thread = options.max_inst
+            if options.maxinsts:
+                switch_cpus_1[i].max_insts_any_thread = options.maxinsts
 
-            if not options.caches:
-                # O3 CPU must have a cache to work.
-                print "O3 CPU must be used with caches"
-                sys.exit(1)
+            # attach the checker cpu if selected
+            if options.checker:
+                switch_cpus[i].addCheckerCpu()
+                switch_cpus_1[i].addCheckerCpu()
 
-            testsys.switch_cpus = switch_cpus
-            testsys.switch_cpus_1 = switch_cpus_1
-            switch_cpu_list = [(testsys.cpu[i], switch_cpus[i]) for i in xrange(np)]
-            switch_cpu_list1 = [(switch_cpus[i], switch_cpus_1[i]) for i in xrange(np)]
+        testsys.switch_cpus = switch_cpus
+        testsys.switch_cpus_1 = switch_cpus_1
+        switch_cpu_list = [
+            (testsys.cpu[i], switch_cpus[i]) for i in range(np)
+        ]
+        switch_cpu_list1 = [
+            (switch_cpus[i], switch_cpus_1[i]) for i in range(np)
+        ]
 
     # set the checkpoint in the cpu before m5.instantiate is called
     if options.take_checkpoints != None and \
@@ -176,7 +592,7 @@ def run(options, root, testsys, cpu_class):
         offset = int(options.take_checkpoints)
         # Set an instruction break point
         if options.simpoint:
-            for i in xrange(np):
+            for i in range(np):
                 if testsys.cpu[i].workload[0].simpoint == 0:
                     fatal('no simpoint for testsys.cpu[%d].workload[0]', i)
                 checkpoint_inst = int(testsys.cpu[i].workload[0].simpoint) + offset
@@ -187,193 +603,134 @@ def run(options, root, testsys, cpu_class):
             options.take_checkpoints = offset
             # Set all test cpus with the right number of instructions
             # for the upcoming simulation
-            for i in xrange(np):
+            for i in range(np):
                 testsys.cpu[i].max_insts_any_thread = offset
 
-    m5.instantiate(root)
-
-    if options.checkpoint_restore != None:
-        from os.path import isdir, exists
-        from os import listdir
-        import re
-
-        if not isdir(cptdir):
-            fatal("checkpoint dir %s does not exist!", cptdir)
-
-        if options.at_instruction:
-            checkpoint_dir = joinpath(cptdir, "cpt.%s.%s" % \
-                    (options.bench, options.checkpoint_restore))
-            if not exists(checkpoint_dir):
-                fatal("Unable to find checkpoint directory %s", checkpoint_dir)
-
-            print "Restoring checkpoint ..."
-            m5.restoreCheckpoint(root, checkpoint_dir)
-            print "Done."
-        elif options.simpoint:
-            # assume workload 0 has the simpoint
-            if testsys.cpu[0].workload[0].simpoint == 0:
-                fatal('Unable to find simpoint')
-
-            options.checkpoint_restore += \
-                int(testsys.cpu[0].workload[0].simpoint)
-
-            checkpoint_dir = joinpath(cptdir, "cpt.%s.%d" % \
-                    (options.bench, options.checkpoint_restore))
-            if not exists(checkpoint_dir):
-                fatal("Unable to find checkpoint directory %s.%s",
-                      options.bench, options.checkpoint_restore)
-
-            print "Restoring checkpoint ..."
-            m5.restoreCheckpoint(root,checkpoint_dir)
-            print "Done."
-        else:
-            dirs = listdir(cptdir)
-            expr = re.compile('cpt\.([0-9]*)')
-            cpts = []
-            for dir in dirs:
-                match = expr.match(dir)
-                if match:
-                    cpts.append(match.group(1))
-
-            cpts.sort(lambda a,b: cmp(long(a), long(b)))
-
-            cpt_num = options.checkpoint_restore
-
-            if cpt_num > len(cpts):
-                fatal('Checkpoint %d not found', cpt_num)
-
-            ## Adjust max tick based on our starting tick
-            maxtick = maxtick - int(cpts[cpt_num - 1])
-
-            ## Restore the checkpoint
-            m5.restoreCheckpoint(root,
-                    joinpath(cptdir, "cpt.%s" % cpts[cpt_num - 1]))
+    if options.take_simpoint_checkpoints != None:
+        simpoints, interval_length = parseSimpointAnalysisFile(options, testsys)
+
+    checkpoint_dir = None
+    if options.checkpoint_restore:
+        cpt_starttick, checkpoint_dir = findCptDir(options, cptdir, testsys)
+    root.apply_config(options.param)
+    m5.instantiate(checkpoint_dir)
+
+    # Initialization is complete.  If we're not in control of simulation
+    # (that is, if we're a slave simulator acting as a component in another
+    #  'master' simulator) then we're done here.  The other simulator will
+    # call simulate() directly. --initialize-only is used to indicate this.
+    if options.initialize_only:
+        return
+
+    # Handle the max tick settings now that tick frequency was resolved
+    # during system instantiation
+    # NOTE: the maxtick variable here is in absolute ticks, so it must
+    # include any simulated ticks before a checkpoint
+    explicit_maxticks = 0
+    maxtick_from_abs = m5.MaxTick
+    maxtick_from_rel = m5.MaxTick
+    maxtick_from_maxtime = m5.MaxTick
+    if options.abs_max_tick:
+        maxtick_from_abs = options.abs_max_tick
+        explicit_maxticks += 1
+    if options.rel_max_tick:
+        maxtick_from_rel = options.rel_max_tick
+        if options.checkpoint_restore:
+            # NOTE: this may need to be updated if checkpoints ever store
+            # the ticks per simulated second
+            maxtick_from_rel += cpt_starttick
+            if options.at_instruction or options.simpoint:
+                warn("Relative max tick specified with --at-instruction or" \
+                     " --simpoint\n      These options don't specify the " \
+                     "checkpoint start tick, so assuming\n      you mean " \
+                     "absolute max tick")
+        explicit_maxticks += 1
+    if options.maxtime:
+        maxtick_from_maxtime = m5.ticks.fromSeconds(options.maxtime)
+        explicit_maxticks += 1
+    if explicit_maxticks > 1:
+        warn("Specified multiple of --abs-max-tick, --rel-max-tick, --maxtime."\
+             " Using least")
+    maxtick = min([maxtick_from_abs, maxtick_from_rel, maxtick_from_maxtime])
+
+    if options.checkpoint_restore != None and maxtick < cpt_starttick:
+        fatal("Bad maxtick (%d) specified: " \
+              "Checkpoint starts starts from tick: %d", maxtick, cpt_starttick)
 
     if options.standard_switch or cpu_class:
         if options.standard_switch:
-            print "Switch at instruction count:%s" % \
-                    str(testsys.cpu[0].max_insts_any_thread)
+            print("Switch at instruction count:%s" %
+                    str(testsys.cpu[0].max_insts_any_thread))
             exit_event = m5.simulate()
         elif cpu_class and options.fast_forward:
-            print "Switch at instruction count:%s" % \
-                    str(testsys.cpu[0].max_insts_any_thread)
+            print("Switch at instruction count:%s" %
+                    str(testsys.cpu[0].max_insts_any_thread))
             exit_event = m5.simulate()
         else:
-            print "Switch at curTick count:%s" % str(10000)
+            print("Switch at curTick count:%s" % str(10000))
             exit_event = m5.simulate(10000)
-        print "Switched CPUS @ cycle = %s" % (m5.curTick())
+        print("Switched CPUS @ tick %s" % (m5.curTick()))
 
-        # when you change to Timing (or Atomic), you halt the system
-        # given as argument.  When you are finished with the system
-        # changes (including switchCpus), you must resume the system
-        # manually.  You DON'T need to resume after just switching
-        # CPUs if you haven't changed anything on the system level.
-
-        m5.changeToTiming(testsys)
-        m5.switchCpus(switch_cpu_list)
-        m5.resume(testsys)
+        m5.switchCpus(testsys, switch_cpu_list)
 
         if options.standard_switch:
-            print "Switch at instruction count:%d" % \
-                    (testsys.switch_cpus[0].max_insts_any_thread)
+            print("Switch at instruction count:%d" %
+                    (testsys.switch_cpus[0].max_insts_any_thread))
 
             #warmup instruction count may have already been set
             if options.warmup_insts:
                 exit_event = m5.simulate()
             else:
-                exit_event = m5.simulate(options.warmup)
-            print "Switching CPUS @ cycle = %s" % (m5.curTick())
-            print "Simulation ends instruction count:%d" % \
-                    (testsys.switch_cpus_1[0].max_insts_any_thread)
-            m5.drain(testsys)
-            m5.switchCpus(switch_cpu_list1)
-            m5.resume(testsys)
-
-    num_checkpoints = 0
-    exit_cause = ''
+                exit_event = m5.simulate(options.standard_switch)
+            print("Switching CPUS @ tick %s" % (m5.curTick()))
+            print("Simulation ends instruction count:%d" %
+                    (testsys.switch_cpus_1[0].max_insts_any_thread))
+            m5.switchCpus(testsys, switch_cpu_list1)
+
+    # If we're taking and restoring checkpoints, use checkpoint_dir
+    # option only for finding the checkpoints to restore from.  This
+    # lets us test checkpointing by restoring from one set of
+    # checkpoints, generating a second set, and then comparing them.
+    if (options.take_checkpoints or options.take_simpoint_checkpoints) \
+        and options.checkpoint_restore:
+
+        if m5.options.outdir:
+            cptdir = m5.options.outdir
+        else:
+            cptdir = getcwd()
 
-    # Checkpoints being taken via the command line at <when> and at
-    # subsequent periods of <period>.  Checkpoint instructions
-    # received from the benchmark running are ignored and skipped in
-    # favor of command line checkpoint instructions.
     if options.take_checkpoints != None :
-        if options.at_instruction or options.simpoint:
-            checkpoint_inst = int(options.take_checkpoints)
+        # Checkpoints being taken via the command line at <when> and at
+        # subsequent periods of <period>.  Checkpoint instructions
+        # received from the benchmark running are ignored and skipped in
+        # favor of command line checkpoint instructions.
+        exit_event = scriptCheckpoints(options, maxtick, cptdir)
 
-            # maintain correct offset if we restored from some instruction
-            if options.checkpoint_restore != None:
-                checkpoint_inst += options.checkpoint_restore
+    # Take SimPoint checkpoints
+    elif options.take_simpoint_checkpoints != None:
+        takeSimpointCheckpoints(simpoints, interval_length, cptdir)
 
-            print "Creating checkpoint at inst:%d" % (checkpoint_inst)
-            exit_event = m5.simulate()
-            print "exit cause = %s" % (exit_event.getCause())
+    # Restore from SimPoint checkpoints
+    elif options.restore_simpoint_checkpoint != None:
+        restoreSimpointCheckpoint()
 
-            # skip checkpoint instructions should they exist
-            while exit_event.getCause() == "checkpoint":
-                exit_event = m5.simulate()
-
-            if exit_event.getCause() == \
-                   "a thread reached the max instruction count":
-                m5.checkpoint(root, joinpath(cptdir, "cpt.%s.%d" % \
-                        (options.bench, checkpoint_inst)))
-                print "Checkpoint written."
-                num_checkpoints += 1
-
-            if exit_event.getCause() == "user interrupt received":
-                exit_cause = exit_event.getCause();
-        else:
-            when, period = options.take_checkpoints.split(",", 1)
-            when = int(when)
-            period = int(period)
-
-            exit_event = m5.simulate(when)
-            while exit_event.getCause() == "checkpoint":
-                exit_event = m5.simulate(when - m5.curTick())
-
-            if exit_event.getCause() == "simulate() limit reached":
-                m5.checkpoint(root, joinpath(cptdir, "cpt.%d"))
-                num_checkpoints += 1
-
-            sim_ticks = when
-            exit_cause = "maximum %d checkpoints dropped" % max_checkpoints
-            while num_checkpoints < max_checkpoints and \
-                    exit_event.getCause() == "simulate() limit reached":
-                if (sim_ticks + period) > maxtick:
-                    exit_event = m5.simulate(maxtick - sim_ticks)
-                    exit_cause = exit_event.getCause()
-                    break
-                else:
-                    exit_event = m5.simulate(period)
-                    sim_ticks += period
-                    while exit_event.getCause() == "checkpoint":
-                        exit_event = m5.simulate(sim_ticks - m5.curTick())
-                    if exit_event.getCause() == "simulate() limit reached":
-                        m5.checkpoint(root, joinpath(cptdir, "cpt.%d"))
-                        num_checkpoints += 1
-
-            if exit_event.getCause() != "simulate() limit reached":
-                exit_cause = exit_event.getCause();
-
-    else: # no checkpoints being taken via this script
+    else:
         if options.fast_forward:
             m5.stats.reset()
-        print "**** REAL SIMULATION ****"
-        exit_event = m5.simulate(maxtick)
+        print("**** REAL SIMULATION ****")
 
-        while exit_event.getCause() == "checkpoint":
-            m5.checkpoint(root, joinpath(cptdir, "cpt.%d"))
-            num_checkpoints += 1
-            if num_checkpoints == max_checkpoints:
-                exit_cause = "maximum %d checkpoints dropped" % max_checkpoints
-                break
-
-            exit_event = m5.simulate(maxtick - m5.curTick())
-            exit_cause = exit_event.getCause()
-
-    if exit_cause == '':
-        exit_cause = exit_event.getCause()
-    print 'Exiting @ cycle %i because %s' % (m5.curTick(), exit_cause)
+        # If checkpoints are being taken, then the checkpoint instruction
+        # will occur in the benchmark code it self.
+        if options.repeat_switch and maxtick > options.repeat_switch:
+            exit_event = repeatSwitch(testsys, repeat_switch_cpu_list,
+                                      maxtick, options.repeat_switch)
+        else:
+            exit_event = benchCheckpoints(options, maxtick, cptdir)
 
+    print('Exiting @ tick %i because %s' %
+          (m5.curTick(), exit_event.getCause()))
     if options.checkpoint_at_end:
-        m5.checkpoint(root, joinpath(cptdir, "cpt.%d"))
+        m5.checkpoint(joinpath(cptdir, "cpt.%d"))
 
+    if exit_event.getCode() != 0:
+        print("Simulated exit code not 0! Exit code is", exit_event.getCode())