Exemple #1
0
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

    if options.checkpoint_dir:
        cptdir = options.checkpoint_dir
    elif m5.options.outdir:
        cptdir = m5.options.outdir
    else:
        cptdir = getcwd()

    if options.fast_forward and options.checkpoint_restore != None:
        fatal("Can't specify both --fast-forward and --checkpoint-restore")

    if options.standard_switch and not options.caches:
        fatal("Must specify --caches when using --standard-switch")

    np = options.num_cpus
    max_checkpoints = options.max_checkpoints
    switch_cpus = None

    if options.prog_interval:
        for i in xrange(np):
            testsys.cpu[i].progress_interval = options.prog_interval

    if options.maxinsts:
        for i in xrange(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)
        ]

        for i in xrange(np):
            if options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            switch_cpus[i].system = testsys
            switch_cpus[i].workload = testsys.cpu[i].workload
            switch_cpus[i].clock = testsys.cpu[0].clock
            # simulation period
            if options.maxinsts:
                switch_cpus[i].max_insts_any_thread = options.maxinsts
            # Add checker cpu if selected
            if options.checker:
                switch_cpus[i].addCheckerCpu()

        testsys.switch_cpus = switch_cpus
        switch_cpu_list = [(testsys.cpu[i], switch_cpus[i])
                           for i in xrange(np)]

    if options.standard_switch:
        if not options.caches:
            # O3 CPU must have a cache to work.
            print "O3 CPU must be used with caches"
            sys.exit(1)

        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)
        ]

        for i in xrange(np):
            switch_cpus[i].system = testsys
            switch_cpus_1[i].system = testsys
            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

            # if restoring, make atomic cpu simulate only a few instructions
            if options.checkpoint_restore != None:
                testsys.cpu[i].max_insts_any_thread = 1
            # Fast forward to specified location if we are not restoring
            elif options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            # Fast forward to a simpoint (warning: time consuming)
            elif options.simpoint:
                if testsys.cpu[i].workload[0].simpoint == 0:
                    fatal('simpoint not found')
                testsys.cpu[i].max_insts_any_thread = \
                    testsys.cpu[i].workload[0].simpoint
            # No distance specified, just switch
            else:
                testsys.cpu[i].max_insts_any_thread = 1

            # warmup period
            if options.warmup_insts:
                switch_cpus[i].max_insts_any_thread = options.warmup_insts

            # simulation period
            if options.maxinsts:
                switch_cpus_1[i].max_insts_any_thread = options.maxinsts

            # 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)]

    # set the checkpoint in the cpu before m5.instantiate is called
    if options.take_checkpoints != None and \
           (options.simpoint or options.at_instruction):
        offset = int(options.take_checkpoints)
        # Set an instruction break point
        if options.simpoint:
            for i in xrange(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
                testsys.cpu[i].max_insts_any_thread = checkpoint_inst
                # used for output below
                options.take_checkpoints = checkpoint_inst
        else:
            options.take_checkpoints = offset
            # Set all test cpus with the right number of instructions
            # for the upcoming simulation
            for i in xrange(np):
                testsys.cpu[i].max_insts_any_thread = offset

    checkpoint_dir = None
    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 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)
        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])
            checkpoint_dir = joinpath(cptdir, "cpt.%s" % cpts[cpt_num - 1])

    m5.instantiate(checkpoint_dir)

    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)
            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)
            exit_event = m5.simulate()
        else:
            print "Switch at curTick count:%s" % str(10000)
            exit_event = m5.simulate(10000)
        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)

        if options.standard_switch:
            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 @ tick %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 = ''

    # 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 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)

            # 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()
            print "exit cause = %s" % (exit_event.getCause())

            # 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(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(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(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
        if options.fast_forward:
            m5.stats.reset()
        print "**** REAL SIMULATION ****"
        exit_event = m5.simulate(maxtick)

        while exit_event.getCause() == "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()

    if exit_cause == '':
        exit_cause = exit_event.getCause()
    print 'Exiting @ tick %i because %s' % (m5.curTick(), exit_cause)

    if options.checkpoint_at_end:
        m5.checkpoint(joinpath(cptdir, "cpt.%d"))
Exemple #2
0
def run(options, root, testsys, cpu_class):
# NOTE: this function is called from example from configs/example/ruby_fs.py 
# like this: "Simulation.run(options, root, system, FutureClass)"
# so, "system" is "testsys" here;

    if options.maxtick:
        maxtick = options.maxtick
    elif options.maxtime:
        simtime = m5.ticks.seconds(simtime)
        print "simulating for: ", simtime
        maxtick = simtime
    else:
        maxtick = m5.MaxTick

    if options.checkpoint_dir:
        cptdir = options.checkpoint_dir
    elif m5.options.outdir:
        cptdir = m5.options.outdir
    else:
        cptdir = getcwd()

    if options.fast_forward and options.checkpoint_restore != None:
        fatal("Can't specify both --fast-forward and --checkpoint-restore")

    if options.standard_switch and not options.caches:
        fatal("Must specify --caches when using --standard-switch")

    np = options.num_cpus
    max_checkpoints = options.max_checkpoints
    switch_cpus = None

    if options.prog_interval:
        for i in xrange(np):
            testsys.cpu[i].progress_interval = options.prog_interval

    if options.maxinsts:
        for i in xrange(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)]

        for i in xrange(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
            # simulation period
            if options.maxinsts:
                switch_cpus[i].max_insts_any_thread = options.maxinsts

        testsys.switch_cpus = switch_cpus
        switch_cpu_list = [(testsys.cpu[i], switch_cpus[i]) for i in xrange(np)]

    if options.standard_switch:
        if not options.caches:
            # O3 CPU must have a cache to work.
            print "O3 CPU must be used with caches"
            sys.exit(1)

        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)]

        for i in xrange(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

            # if restoring, make atomic cpu simulate only a few instructions
            if options.checkpoint_restore != None:
                testsys.cpu[i].max_insts_any_thread = 1
            # Fast forward to specified location if we are not restoring
            elif options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            # Fast forward to a simpoint (warning: time consuming)
            elif options.simpoint:
                if testsys.cpu[i].workload[0].simpoint == 0:
                    fatal('simpoint not found')
                testsys.cpu[i].max_insts_any_thread = \
                    testsys.cpu[i].workload[0].simpoint
            # No distance specified, just switch
            else:
                testsys.cpu[i].max_insts_any_thread = 1

            # warmup period
            if options.warmup_insts:
                switch_cpus[i].max_insts_any_thread =  options.warmup_insts

            # simulation period
            if options.maxinsts:
                switch_cpus_1[i].max_insts_any_thread = options.maxinsts

        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)]

    # set the checkpoint in the cpu before m5.instantiate is called
    if options.take_checkpoints != None and \
           (options.simpoint or options.at_instruction):
        offset = int(options.take_checkpoints)
        # Set an instruction break point
        if options.simpoint:
            for i in xrange(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
                testsys.cpu[i].max_insts_any_thread = checkpoint_inst
                # used for output below
                options.take_checkpoints = checkpoint_inst
        else:
            options.take_checkpoints = offset
            # Set all test cpus with the right number of instructions
            # for the upcoming simulation
            for i in xrange(np):
                testsys.cpu[i].max_insts_any_thread = offset

    checkpoint_dir = None
    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 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)
        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])
            checkpoint_dir = joinpath(cptdir, "cpt.%s" % cpts[cpt_num - 1])

    m5.instantiate(checkpoint_dir)

    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)
            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)
            exit_event = m5.simulate()
        else:
            print "Switch at curTick count:%s" % str(10000)
            exit_event = m5.simulate(10000)
        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)

        if options.standard_switch:
            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 @ tick %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 = ''

    # 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 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)

            # 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()
            print "exit cause = %s" % (exit_event.getCause())

            # 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(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(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(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
        if options.fast_forward:
            m5.stats.reset()
        print "**** REAL SIMULATION ****"
        #exit_event = m5.simulate(maxtick)


        # --Note1: Ruby is created in ruby_fs.py by 
        # "Ruby.create_system(options, system, system.piobus, system._dma_devices)"
        # which assigned: "stats_filename = options.ruby_stats";
        # definition of "create_system" is in configs/ruby/Ruby.py, which
        # instantiate the ctor of RubySystem: "system.ruby = RubySystem(...)";
        #print testsys.ruby._cpu_ruby_ports 
        #print testsys.ruby.network.ni_flit_size
        #print testsys.ruby.profiler.ruby_system
        # the ctor of RubySystem is defined in src/mem/ruby/system/RubySystem.py;
        # which sets some defaults:
        #print testsys.ruby.stats_filename # i.e., ruby.stats
        #print testsys.ruby.type
        #print testsys.ruby.random_seed
        #print testsys.ruby.clock
        #print testsys.ruby.block_size_bytes
        #print testsys.ruby.mem_size
        #print testsys.ruby.no_mem_vec

        # () cris: description of changes
        # --Note2: initially writing into ruby.stats was done with overwriting;
        # so, for each dump point the file was re-written; to fix that I
        # changed function "OutputDirectory::create(...)" from src/base/output.cc
        # which is called by "RubyExitCallback::process()" from src/mem/ruby/system/System.cc
        # function that is the one called at the end of the gem5 run
        # as a calback to dump all ruby stats (callback is "registered" in the
        # ctor of RubySystem::RubySystem() inside the same file...); 
        # --Note3: using doExitCleanup inspired from src/python/m5/simulate.py 
        # (inside which ini and json files are created; you need to rebuild each
        # time you change that Python file):
        #m5.internal.core.doExitCleanup( False) #clear callback queue?
        # --Note4: python/m5/internal/core.py describes "m5.internal.core";

        # cris: here I want to dump stats every other delta ticks;
        # I need these to be able to generate reliability traces;
        
        NUM_OF_DUMPS = 100
        num_i = 0
        delta = maxtick/NUM_OF_DUMPS
        sim_ticks = m5.curTick()
        while (m5.curTick() < maxtick):
            sim_ticks += delta
            exit_event = m5.simulate(sim_ticks - m5.curTick())
            if exit_event.getCause() == "simulate() limit reached":
                #--Note5: "doExitCleanup()" is described in src/sim/core.cc;
                # I changed it to be able to call it multiple times;
                #--Note6: do not dump stats in ruby.stats for last iteration 
                # because it will be repeated once more via the exit callbacks
                # in src/python/m5/simulate.py...
                # Note6: next call of doExitCleanup does actually also reset/clear
                # the stats of ruby system via the RubyExitCallback::process() in
                # src/mem/ruby/system/System.cc
                if num_i < (NUM_OF_DUMPS-1):
                    print "Dumping also ruby stats at inst %d to file: ruby.stats" %(num_i)
                    m5.internal.core.doExitCleanup( False) #clear callback queue?
                print "Dumping gem5 stats at inst %d to file: stats.txt" %(num_i)
                # --Note7: dump() wites into stats.txt; dump() is defined in
                # src/python/m5/stats/__init__.py and does its 
                # thing via the functions described in base/stats/text.cc
                #atexit.register(stats.dump) <--- does not work (from simulate.py)
                m5.stats.dump()
                m5.stats.reset() # <--- what does it actually do?
                num_i += 1
                

        # alex's changes:
	#while exit_event.getCause() != "m5_exit instruction encountered":
	#    m5.stats.dump()
	#    m5.stats.reset()
	#    exit_event = m5.simulate(maxtick)

        #while exit_event.getCause() == "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()

    if exit_cause == '':
        exit_cause = exit_event.getCause()
    print 'Exiting @ tick %i because %s' % (m5.curTick(), exit_cause)

    if options.checkpoint_at_end:
        m5.checkpoint(joinpath(cptdir, "cpt.%d"))
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

    if options.checkpoint_dir:
        cptdir = options.checkpoint_dir
    elif m5.options.outdir:
        cptdir = m5.options.outdir
    else:
        cptdir = getcwd()

    if options.fast_forward and options.checkpoint_restore != None:
        fatal("Can't specify both --fast-forward and --checkpoint-restore")

    if options.standard_switch and not options.caches:
        fatal("Must specify --caches when using --standard-switch")

    np = options.num_cpus
    max_checkpoints = options.max_checkpoints
    new = None

    if options.prog_interval:
        for i in xrange(np):
            testsys.cpu[i].progress_interval = options.prog_interval

    if options.maxinsts:
        for i in xrange(np):
            testsys.cpu[i].max_insts_any_thread = options.maxinsts

    if cpu_class:
        new = [cpu_class(defer_registration=True, cpu_id=(np + i)) for i in xrange(np)]

        for i in xrange(np):
            if options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            new[i].system = testsys
            if not buildEnv["FULL_SYSTEM"]:
                new[i].workload = testsys.cpu[i].workload
            new[i].clock = testsys.cpu[0].clock
            # simulation period
            if options.maxinsts:
                new[i].max_insts_any_thread = options.maxinsts

        testsys.new = new
        switch_cpu_list = [(testsys.cpu[i], new[i]) for i in xrange(np)]

    if options.standard_switch:
        if not options.caches:
            # O3 CPU must have a cache to work.
            print "O3 CPU must be used with caches"
            sys.exit(1)

        new = [TimingSimpleCPU(defer_registration=True, cpu_id=(np + i)) for i in xrange(np)]
        new_1 = [DerivO3CPU(defer_registration=True, cpu_id=(2 * np + i)) for i in xrange(np)]

        for i in xrange(np):
            new[i].system = testsys
            new_1[i].system = testsys
            if not buildEnv["FULL_SYSTEM"]:
                new[i].workload = testsys.cpu[i].workload
                new_1[i].workload = testsys.cpu[i].workload
            new[i].clock = testsys.cpu[0].clock
            new_1[i].clock = testsys.cpu[0].clock

            # if restoring, make atomic cpu simulate only a few instructions
            if options.checkpoint_restore != None:
                testsys.cpu[i].max_insts_any_thread = 1
            # Fast forward to specified location if we are not restoring
            elif options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            # Fast forward to a simpoint (warning: time consuming)
            elif options.simpoint:
                if testsys.cpu[i].workload[0].simpoint == 0:
                    fatal("simpoint not found")
                testsys.cpu[i].max_insts_any_thread = testsys.cpu[i].workload[0].simpoint
            # No distance specified, just switch
            else:
                testsys.cpu[i].max_insts_any_thread = 1

            # warmup period
            if options.warmup_insts:
                new[i].max_insts_any_thread = options.warmup_insts

            # simulation period
            if options.maxinsts:
                new_1[i].max_insts_any_thread = options.maxinsts

        testsys.new = new
        testsys.new_1 = new_1
        switch_cpu_list = [(testsys.cpu[i], new[i]) for i in xrange(np)]
        switch_cpu_list1 = [(new[i], new_1[i]) for i in xrange(np)]

    # set the checkpoint in the cpu before m5.instantiate is called
    if options.take_checkpoints != None and (options.simpoint or options.at_instruction):
        offset = int(options.take_checkpoints)
        # Set an instruction break point
        if options.simpoint:
            for i in xrange(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
                testsys.cpu[i].max_insts_any_thread = checkpoint_inst
                # used for output below
                options.take_checkpoints = checkpoint_inst
        else:
            options.take_checkpoints = offset
            # Set all test cpus with the right number of instructions
            # for the upcoming simulation
            for i in xrange(np):
                testsys.cpu[i].max_insts_any_thread = offset

    checkpoint_dir = None
    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 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)
        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])
            checkpoint_dir = joinpath(cptdir, "cpt.%s" % cpts[cpt_num - 1])

    m5.instantiate(checkpoint_dir)

    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)
            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)
            exit_event = m5.simulate()
        else:
            print "Switch at curTick count:%s" % str(10000)
            exit_event = m5.simulate(10000)
        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)

        if options.standard_switch:
            print "Switch at instruction count:%d" % (testsys.new[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 @ tick %s" % (m5.curTick())
            print "Simulation ends instruction count:%d" % (testsys.new_1[0].max_insts_any_thread)
            m5.drain(testsys)
            m5.switchCpus(switch_cpu_list1)
            m5.resume(testsys)

    num_checkpoints = 0
    exit_cause = ""

    # 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 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)

            # 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()
            print "exit cause = %s" % (exit_event.getCause())

            # 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(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(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(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
        if options.fast_forward:
            m5.stats.reset()
        print "**** REAL SIMULATION ****"
        exit_event = m5.simulate(maxtick)

        while exit_event.getCause() == "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()

    if exit_cause == "":
        exit_cause = exit_event.getCause()
    print "Exiting @ tick %i because %s" % (m5.curTick(), exit_cause)

    if options.checkpoint_at_end:
        m5.checkpoint(joinpath(cptdir, "cpt.%d"))
Exemple #4
0
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

    if options.checkpoint_dir:
        cptdir = options.checkpoint_dir
    elif m5.options.outdir:
        cptdir = m5.options.outdir
    else:
        cptdir = getcwd()

    if options.fast_forward and options.checkpoint_restore != None:
        fatal("Can't specify both --fast-forward and --checkpoint-restore")

    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
    switch_cpus = None

    if options.prog_interval:
        for i in xrange(np):
            testsys.cpu[i].progress_interval = options.prog_interval

    if options.maxinsts:
        for i in xrange(np):
            testsys.cpu[i].max_insts_any_thread = options.maxinsts

    if options.pred_type:
        for i in xrange(np):
            testsys.cpu[i].predType = options.pred_type

    if options.global_hist_size:
        for i in xrange(np):
            testsys.cpu[i].globalHistoryBits = options.global_hist_size

    if options.global_pred_size:
        for i in xrange(np):
            testsys.cpu[i].globalPredictorSize = options.global_pred_size

    if options.local_pred_size:
        for i in xrange(np):
            testsys.cpu[i].localPredictorSize = options.local_pred_size

    if cpu_class:
        switch_cpus = [
            cpu_class(defer_registration=True, cpu_id=(i)) for i in xrange(np)
        ]

        for i in xrange(np):
            if options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            switch_cpus[i].system = testsys
            switch_cpus[i].workload = testsys.cpu[i].workload
            switch_cpus[i].clock = testsys.cpu[i].clock
            # simulation period
            if options.maxinsts:
                switch_cpus[i].max_insts_any_thread = options.maxinsts
            # Add checker cpu if selected
            if options.checker:
                switch_cpus[i].addCheckerCpu()

        testsys.switch_cpus = switch_cpus
        switch_cpu_list = [(testsys.cpu[i], switch_cpus[i])
                           for i in xrange(np)]

    if options.repeat_switch:
        if options.cpu_type == "arm_detailed":
            if not options.caches:
                print "O3 CPU must be used with caches"
                sys.exit(1)

            repeat_switch_cpus = [O3_ARM_v7a_3(defer_registration=True, \
                                  cpu_id=(i)) for i in xrange(np)]
        elif options.cpu_type == "detailed":
            if not options.caches:
                print "O3 CPU must be used with caches"
                sys.exit(1)

            repeat_switch_cpus = [DerivO3CPU(defer_registration=True, \
                                  cpu_id=(i)) for i in xrange(np)]
        elif options.cpu_type == "inorder":
            print "inorder CPU switching not supported"
            sys.exit(1)
        elif options.cpu_type == "timing":
            repeat_switch_cpus = [TimingSimpleCPU(defer_registration=True, \
                                  cpu_id=(i)) for i in xrange(np)]
        else:
            repeat_switch_cpus = [AtomicSimpleCPU(defer_registration=True, \
                                  cpu_id=(i)) for i in xrange(np)]

        for i in xrange(np):
            repeat_switch_cpus[i].system = testsys
            repeat_switch_cpus[i].workload = testsys.cpu[i].workload
            repeat_switch_cpus[i].clock = testsys.cpu[i].clock

            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 xrange(np)]
        else:
            repeat_switch_cpu_list = [(testsys.cpu[i], repeat_switch_cpus[i])
                                      for i in xrange(np)]

    if options.standard_switch:
        switch_cpus = [
            TimingSimpleCPU(defer_registration=True, cpu_id=(i))
            for i in xrange(np)
        ]
        switch_cpus_1 = [
            DerivO3CPU(defer_registration=True, cpu_id=(i)) for i in xrange(np)
        ]

        for i in xrange(np):
            switch_cpus[i].system = testsys
            switch_cpus_1[i].system = testsys
            switch_cpus[i].workload = testsys.cpu[i].workload
            switch_cpus_1[i].workload = testsys.cpu[i].workload
            switch_cpus[i].clock = testsys.cpu[i].clock
            switch_cpus_1[i].clock = testsys.cpu[i].clock

            # if restoring, make atomic cpu simulate only a few instructions
            if options.checkpoint_restore != None:
                testsys.cpu[i].max_insts_any_thread = 1
            # Fast forward to specified location if we are not restoring
            elif options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            # Fast forward to a simpoint (warning: time consuming)
            elif options.simpoint:
                if testsys.cpu[i].workload[0].simpoint == 0:
                    fatal('simpoint not found')
                testsys.cpu[i].max_insts_any_thread = \
                    testsys.cpu[i].workload[0].simpoint
            # No distance specified, just switch
            else:
                testsys.cpu[i].max_insts_any_thread = 1

            # warmup period
            if options.warmup_insts:
                switch_cpus[i].max_insts_any_thread = options.warmup_insts

            # simulation period
            if options.maxinsts:
                switch_cpus_1[i].max_insts_any_thread = options.maxinsts

            # 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)]

    # set the checkpoint in the cpu before m5.instantiate is called
    if options.take_checkpoints != None and \
           (options.simpoint or options.at_instruction):
        offset = int(options.take_checkpoints)
        # Set an instruction break point
        if options.simpoint:
            for i in xrange(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
                testsys.cpu[i].max_insts_any_thread = checkpoint_inst
                # used for output below
                options.take_checkpoints = checkpoint_inst
        else:
            options.take_checkpoints = offset
            # Set all test cpus with the right number of instructions
            # for the upcoming simulation
            for i in xrange(np):
                testsys.cpu[i].max_insts_any_thread = offset

    checkpoint_dir = None
    if options.checkpoint_restore != None:
        maxtick, checkpoint_dir = findCptDir(options, maxtick, cptdir, testsys)
    m5.instantiate(checkpoint_dir)

    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)
            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)
            exit_event = m5.simulate()
        else:
            print "Switch at curTick count:%s" % str(10000)
            exit_event = m5.simulate(10000)
        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.doDrain(testsys)
        m5.changeToTiming(testsys)
        m5.switchCpus(switch_cpu_list)
        m5.resume(testsys)

        if options.standard_switch:
            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.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.doDrain(testsys)
            m5.switchCpus(switch_cpu_list1)
            m5.resume(testsys)

    # 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 and options.checkpoint_restore:
        if m5.options.outdir:
            cptdir = m5.options.outdir
        else:
            cptdir = getcwd()

    if options.take_checkpoints != None:
        # 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_cause = scriptCheckpoints(options, maxtick, cptdir)
    else:
        if options.fast_forward:
            m5.stats.reset()
        print "**** REAL SIMULATION ****"

        # 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_cause = repeatSwitch(testsys, repeat_switch_cpu_list, maxtick,
                                      options.repeat_switch)
        else:
            exit_cause = benchCheckpoints(options, maxtick, cptdir)

    print 'Exiting @ tick %i because %s' % (m5.curTick(), exit_cause)
    if options.checkpoint_at_end:
        m5.checkpoint(joinpath(cptdir, "cpt.%d"))
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

    if options.checkpoint_dir:
        cptdir = options.checkpoint_dir
    elif m5.options.outdir:
        cptdir = m5.options.outdir
    else:
        cptdir = getcwd()

    if options.fast_forward and options.checkpoint_restore != None:
        fatal("Can't specify both --fast-forward and --checkpoint-restore")

    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
    switch_cpus = None

    if options.prog_interval:
        for i in xrange(np):
            testsys.cpu[i].progress_interval = options.prog_interval

    if options.maxinsts:
        for i in xrange(np):
            testsys.cpu[i].max_insts_any_thread = options.maxinsts

    if cpu_class:
        switch_cpus = [cpu_class(defer_registration=True, cpu_id=(i))
                       for i in xrange(np)]

        for i in xrange(np):
            if options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            switch_cpus[i].system =  testsys
            switch_cpus[i].workload = testsys.cpu[i].workload
            switch_cpus[i].clock = testsys.cpu[i].clock
            # simulation period
            if options.maxinsts:
                switch_cpus[i].max_insts_any_thread = options.maxinsts
            # Add checker cpu if selected
            if options.checker:
                switch_cpus[i].addCheckerCpu()

        testsys.switch_cpus = switch_cpus
        switch_cpu_list = [(testsys.cpu[i], switch_cpus[i]) for i in xrange(np)]

    if options.repeat_switch:
        if options.cpu_type == "arm_detailed":
            if not options.caches:
                print "O3 CPU must be used with caches"
                sys.exit(1)

            repeat_switch_cpus = [O3_ARM_v7a_3(defer_registration=True, \
                                  cpu_id=(i)) for i in xrange(np)]
        elif options.cpu_type == "detailed":
            if not options.caches:
                print "O3 CPU must be used with caches"
                sys.exit(1)

            repeat_switch_cpus = [DerivO3CPU(defer_registration=True, \
                                  cpu_id=(i)) for i in xrange(np)]
        elif options.cpu_type == "inorder":
            print "inorder CPU switching not supported"
            sys.exit(1)
        elif options.cpu_type == "timing":
            repeat_switch_cpus = [TimingSimpleCPU(defer_registration=True, \
                                  cpu_id=(i)) for i in xrange(np)]
        else:
            repeat_switch_cpus = [AtomicSimpleCPU(defer_registration=True, \
                                  cpu_id=(i)) for i in xrange(np)]

        for i in xrange(np):
            repeat_switch_cpus[i].system = testsys
            repeat_switch_cpus[i].workload = testsys.cpu[i].workload
            repeat_switch_cpus[i].clock = testsys.cpu[i].clock

            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 xrange(np)]
        else:
            repeat_switch_cpu_list = [(testsys.cpu[i], repeat_switch_cpus[i])
                                      for i in xrange(np)]

    if options.standard_switch:
        switch_cpus = [TimingSimpleCPU(defer_registration=True, cpu_id=(i))
                       for i in xrange(np)]
        switch_cpus_1 = [DerivO3CPU(defer_registration=True, cpu_id=(i))
                        for i in xrange(np)]

        for i in xrange(np):
            switch_cpus[i].system =  testsys
            switch_cpus_1[i].system =  testsys
            switch_cpus[i].workload = testsys.cpu[i].workload
            switch_cpus_1[i].workload = testsys.cpu[i].workload
            switch_cpus[i].clock = testsys.cpu[i].clock
            switch_cpus_1[i].clock = testsys.cpu[i].clock

            # if restoring, make atomic cpu simulate only a few instructions
            if options.checkpoint_restore != None:
                testsys.cpu[i].max_insts_any_thread = 1
            # Fast forward to specified location if we are not restoring
            elif options.fast_forward:
                testsys.cpu[i].max_insts_any_thread = int(options.fast_forward)
            # Fast forward to a simpoint (warning: time consuming)
            elif options.simpoint:
                if testsys.cpu[i].workload[0].simpoint == 0:
                    fatal('simpoint not found')
                testsys.cpu[i].max_insts_any_thread = \
                    testsys.cpu[i].workload[0].simpoint
            # No distance specified, just switch
            else:
                testsys.cpu[i].max_insts_any_thread = 1

            # warmup period
            if options.warmup_insts:
                switch_cpus[i].max_insts_any_thread =  options.warmup_insts

            # simulation period
            if options.maxinsts:
                switch_cpus_1[i].max_insts_any_thread = options.maxinsts

            # 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)]

    # set the checkpoint in the cpu before m5.instantiate is called
    if options.take_checkpoints != None and \
           (options.simpoint or options.at_instruction):
        offset = int(options.take_checkpoints)
        # Set an instruction break point
        if options.simpoint:
            for i in xrange(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
                testsys.cpu[i].max_insts_any_thread = checkpoint_inst
                # used for output below
                options.take_checkpoints = checkpoint_inst
        else:
            options.take_checkpoints = offset
            # Set all test cpus with the right number of instructions
            # for the upcoming simulation
            for i in xrange(np):
                testsys.cpu[i].max_insts_any_thread = offset

    checkpoint_dir = None
    if options.checkpoint_restore != None:
        maxtick, checkpoint_dir = findCptDir(options, maxtick, cptdir, testsys)
    m5.instantiate(checkpoint_dir)

    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)
            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)
            exit_event = m5.simulate()
        else:
            print "Switch at curTick count:%s" % str(10000)
            exit_event = m5.simulate(10000)
        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)

        if options.standard_switch:
            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.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.drain(testsys)
            m5.switchCpus(switch_cpu_list1)
            m5.resume(testsys)

    # 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 and options.checkpoint_restore:
        if m5.options.outdir:
            cptdir = m5.options.outdir
        else:
            cptdir = getcwd()

    if options.take_checkpoints != None :
        # 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_cause = scriptCheckpoints(options)
    else:
        if options.fast_forward:
            m5.stats.reset()
        print "**** REAL SIMULATION ****"

        # 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_cause = repeatSwitch(testsys, repeat_switch_cpu_list,
                                      maxtick, options.repeat_switch)
        else:
            exit_cause = benchCheckpoints(options, maxtick, cptdir)

    print 'Exiting @ tick %i because %s' % (m5.curTick(), exit_cause)
    if options.checkpoint_at_end:
        m5.checkpoint(joinpath(cptdir, "cpt.%d"))