def scriptCheckpoints(options, maxtick, cptdir):
if options.at_instruction or options.simpoint:
checkpoint_inst = 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 checkpoint inst number" or \
exit_cause == "sp simulation reached the interval size":
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_sync(options, maxreltick, cptdir):
import socket
quantum,port,host = options.sync.split(',')
sync_quantum = long(quantum)
TCP_IP = host
TCP_PORT = int(port)
BUFFER_SIZE = 1
pre_tick = m5.curTick()
try:
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.connect((TCP_IP, TCP_PORT))
exit_event = m5.simulate(sync_quantum)
exit_cause = exit_event.getCause()
total_tick_simulated = sync_quantum
num_checkpoints = 0
max_checkpoints = options.max_checkpoints
s.send("R")
data = s.recv(BUFFER_SIZE)
while total_tick_simulated <= maxreltick and\
(exit_cause == "simulate() limit reached" or\
exit_cause == "checkpoint"):
if exit_cause == "checkpoint":
m5.simulate(pre_tick + sync_quantum - m5.curTick())
# send "C" to barrier to notify that we should take a
# checkpoint at the begining of next quantum
s.send("C")
else:
s.send("R")
data = s.recv(BUFFER_SIZE)
pre_tick = m5.curTick()
exit_event = m5.simulate(sync_quantum)
total_tick_simulated += sync_quantum
# if we receive a "C" from barrier, start to dump a checkpoint
if data == "C" :
m5.checkpoint(joinpath(cptdir, "cpt.%d"))
num_checkpoints += 1
if num_checkpoints == max_checkpoints:
exit_cause = "maximum %d checkpoints dropped"\
% max_checkpoints
break
exit_cause = exit_event.getCause()
except KeyboardInterrupt:
sys.exit()
except Exception as ex:
print "exception %s occured." %(ex.args[1])
sys.exit()
return exit_event
def run(checkpoint_dir=m5.options.outdir):
# start simulation (and drop checkpoints when requested)
while True:
event = m5.simulate()
exit_msg = event.getCause()
if exit_msg == "checkpoint":
print("Dropping checkpoint at tick %d" % m5.curTick())
cpt_dir = os.path.join(checkpoint_dir, "cpt.%d" % m5.curTick())
m5.checkpoint(cpt_dir)
print("Checkpoint done.")
else:
print(exit_msg, " @ ", m5.curTick())
break
sys.exit(event.getCode())
def run_test(root):
"""Default run_test implementations. Scripts can override it."""
# instantiate configuration
m5.instantiate()
# simulate until program terminates
exit_event = m5.simulate(maxtick)
print('Exiting @ tick', m5.curTick(), 'because', exit_event.getCause())
def run(args):
cptdir = m5.options.outdir
if args.checkpoint:
print "Checkpoint directory: %s" % cptdir
while True:
event = m5.simulate()
exit_msg = event.getCause()
if exit_msg == "checkpoint":
print "Dropping checkpoint at tick %d" % m5.curTick()
cpt_dir = os.path.join(m5.options.outdir, "cpt.%d" % m5.curTick())
m5.checkpoint(os.path.join(cpt_dir))
print "Checkpoint done."
else:
print exit_msg, " @ ", m5.curTick()
break
sys.exit(event.getCode())
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
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_test(root):
"""gpu test requires a specialized run_test implementation to set up the
mmio space."""
# instantiate configuration
m5.instantiate()
# Now that the system has been constructed, setup the mmio space
root.system.cpu[0].workload[0].map(0x10000000, 0x200000000, 4096)
# simulate until program terminates
exit_event = m5.simulate(maxtick)
print 'Exiting @ tick', m5.curTick(), 'because', exit_event.getCause()
def handleEventPlayback(controller, options):
if not options.playback_file: return
from controller import playback
# Use all non-empty colon-separated words as event names.
events = filter(len, options.playback.split(':'))
offset = options.playback_offset or m5.curTick()
# The scheduleRecordedEvents() function uses m5.ticks.fromSeconds() which
# needs a fixed simulator frequency. Because this function is called early
# during M5 initialization the frequency is not yet fixed by M5 itself.
m5.ticks.fixGlobalFrequency()
playback.scheduleRecordedEvents(m5.event.mainq, controller, options.playback_file, events, 1.0/options.playback_speed, offset)
def main():
parser = argparse.ArgumentParser(epilog=__doc__)
parser.add_argument("commands_to_run", metavar="command(s)", nargs='*',
help="Command(s) to run")
parser.add_argument("--cpu", type=str, choices=cpu_types.keys(),
default="atomic",
help="CPU model to use")
parser.add_argument("--cpu-freq", type=str, default="4GHz")
parser.add_argument("--num-cores", type=int, default=1,
help="Number of CPU cores")
parser.add_argument("--mem-type", default="DDR3_1600_8x8",
choices=MemConfig.mem_names(),
help = "type of memory to use")
parser.add_argument("--mem-channels", type=int, default=2,
help = "number of memory channels")
parser.add_argument("--mem-ranks", type=int, default=None,
help = "number of memory ranks per channel")
parser.add_argument("--mem-size", action="store", type=str,
default="2GB",
help="Specify the physical memory size")
args = parser.parse_args()
# Create a single root node for gem5's object hierarchy. There can
# only exist one root node in the simulator at any given
# time. Tell gem5 that we want to use syscall emulation mode
# instead of full system mode.
root = Root(full_system=False)
# Populate the root node with a system. A system corresponds to a
# single node with shared memory.
root.system = create(args)
# Instantiate the C++ object hierarchy. After this point,
# SimObjects can't be instantiated anymore.
m5.instantiate()
# Start the simulator. This gives control to the C++ world and
# starts the simulator. The returned event tells the simulation
# script why the simulator exited.
event = m5.simulate()
# Print the reason for the simulation exit. Some exit codes are
# requests for service (e.g., checkpoints) from the simulation
# script. We'll just ignore them here and exit.
print(event.getCause(), " @ ", m5.curTick())
sys.exit(event.getCode())
def main():
parser = argparse.ArgumentParser(description="Simple system using HMC as\
main memory")
HMC.add_options(parser)
add_options(parser)
options = parser.parse_args()
# build the system
root = build_system(options)
# instantiate all of the objects we've created so far
m5.instantiate()
print("Beginning simulation!")
event = m5.simulate(10000000000)
m5.stats.dump()
print('Exiting @ tick %i because %s (exit code is %i)' % (m5.curTick(),
event.getCause(),
event.getCode()))
print("Done")
def dump():
'''Dump all statistics data to the registered outputs'''
from m5 import options
curTick = m5.curTick()
global lastDump
assert lastDump <= curTick
if lastDump == curTick:
return
lastDump = curTick
internal.stats.processDumpQueue()
prepare()
# Prodromou: Based on a M5 flag, filter stats or not
if options.use_stat_flags:
stat_print_flags = [
"committedInsts",
"readReqs",
"writeReqs",
"system.mem_ctrls.avgQLat",
"microThreads",
"thinkingTime",
"readRowHitRate"
]
for output in outputList:
if output.valid():
output.begin()
for stat in stats_list:
for flag in stat_print_flags:
if flag in stat.name:
output.visit(stat)
output.end()
else:
for output in outputList:
if output.valid():
output.begin()
for stat in stats_list:
output.visit(stat)
output.end()
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 dump():
'''Dump all statistics data to the registered outputs'''
curTick = m5.curTick()
global lastDump
assert lastDump <= curTick
if lastDump == curTick:
return
lastDump = curTick
prepare()
for output in outputList:
if output.valid():
output.begin()
for stat in stats_list:
output.visit(stat)
output.end()
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 dump():
'''Dump all statistics data to the registered outputs'''
curTick = m5.curTick()
global lastDump
assert lastDump <= curTick
if lastDump == curTick:
return
lastDump = curTick
internal.stats.processDumpQueue()
prepare()
for output in outputList:
if output.valid():
output.begin()
for stat in stats_list:
output.visit(stat)
output.end()
for listener in dumpListeners:
listener()
def __init__(self, eventq, period):
super(ProgressEvent, self).__init__()
self.period = int(period)
self.eventq = eventq
self.eventq.schedule(self, m5.curTick() + self.period)
tmp_cpu_list = []
for old_cpu, new_cpu in switch_cpu_list:
tmp_cpu_list.append((new_cpu, old_cpu))
switch_cpu_list = tmp_cpu_list
m5.stats.reset()
exit_event = m5.simulate()
exit_cause = exit_event.getCause()
if switch_cpu_list[0][0].type == 'DerivO3CPU':
m5.stats.dump()
elif options.simpoint_mode == "checkpoint_gen":
exit_event = m5.simulate()
exit_cause = exit_event.getCause()
checkpoint_points = simple_end_points[:]
checkpoint_points.reverse()
while exit_cause == "simpoint starting point found":
point_inst = checkpoint_points.pop()
m5.checkpoint(os.path.join(options.checkpoint_dir, "simpoint.ckpt.inst.%d" % point_inst))
if not checkpoint_points:
exit_cause = "final simpoint checkpoint created"
break;
exit_event = m5.simulate()
exit_cause = exit_event.getCause()
print "Exiting @ tick %i because %s" % (m5.curTick(), exit_cause)
import m5
from m5.objects import *
root = Root(full_system=False)
root.hello = HelloObject(time_to_wait='2us', number_of_fires=10)
root.hello.goodbye_object = GoodbyeObject(buffer_size='100B')
m5.instantiate()
print("begin")
exit_event = m5.simulate()
print('exit @ tick {} because {}'.format(m5.curTick(), exit_event.getCause()))
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"))
def main():
parser = argparse.ArgumentParser(
description="Generic ARM big.LITTLE configuration")
parser.add_argument("--restore-from", type=str, default=None,
help="Restore from checkpoint")
parser.add_argument("--dtb", type=str, default=default_dtb,
help="DTB file to load")
parser.add_argument("--kernel", type=str, default=default_kernel,
help="Linux kernel")
parser.add_argument("--disk", action="append", type=str, default=[],
help="Disks to instantiate")
parser.add_argument("--bootscript", type=str, default=default_rcs,
help="Linux bootscript")
parser.add_argument("--atomic", action="store_true", default=False,
help="Use atomic CPUs")
parser.add_argument("--kernel-init", type=str, default="/sbin/init",
help="Override init")
parser.add_argument("--big-cpus", type=int, default=1,
help="Number of big CPUs to instantiate")
parser.add_argument("--little-cpus", type=int, default=1,
help="Number of little CPUs to instantiate")
parser.add_argument("--caches", action="store_true", default=False,
help="Instantiate caches")
parser.add_argument("--last-cache-level", type=int, default=2,
help="Last level of caches (e.g. 3 for L3)")
parser.add_argument("--big-cpu-clock", type=str, default="2GHz",
help="Big CPU clock frequency")
parser.add_argument("--little-cpu-clock", type=str, default="1GHz",
help="Little CPU clock frequency")
m5.ticks.fixGlobalFrequency()
options = parser.parse_args()
if options.atomic:
cpu_config = { 'cpu' : AtomicSimpleCPU }
big_cpu_config, little_cpu_config = cpu_config, cpu_config
else:
big_cpu_config = { 'cpu' : CpuConfig.get("arm_detailed"),
'l1i' : devices.L1I,
'l1d' : devices.L1D,
'wcache' : devices.WalkCache,
'l2' : devices.L2 }
little_cpu_config = { 'cpu' : MinorCPU,
'l1i' : devices.L1I,
'l1d' : devices.L1D,
'wcache' : devices.WalkCache,
'l2' : devices.L2 }
big_cpu_class = big_cpu_config['cpu']
little_cpu_class = little_cpu_config['cpu']
kernel_cmd = [
"earlyprintk=pl011,0x1c090000",
"console=ttyAMA0",
"lpj=19988480",
"norandmaps",
"loglevel=8",
"mem=%s" % default_mem_size,
"root=/dev/vda1",
"rw",
"init=%s" % options.kernel_init,
"vmalloc=768MB",
]
root = Root(full_system=True)
assert big_cpu_class.memory_mode() == little_cpu_class.memory_mode()
disks = default_disk if len(options.disk) == 0 else options.disk
system = createSystem(options.kernel, big_cpu_class.memory_mode(),
options.bootscript, disks=disks)
root.system = system
system.boot_osflags = " ".join(kernel_cmd)
# big cluster
if options.big_cpus > 0:
system.bigCluster = CpuCluster()
system.bigCluster.addCPUs(big_cpu_config, options.big_cpus,
options.big_cpu_clock)
# LITTLE cluster
if options.little_cpus > 0:
system.littleCluster = CpuCluster()
system.littleCluster.addCPUs(little_cpu_config, options.little_cpus,
options.little_cpu_clock)
# add caches
if options.caches:
cluster_mem_bus = addCaches(system, options.last_cache_level)
else:
if big_cpu_class.require_caches():
m5.util.panic("CPU model %s requires caches" % str(big_cpu_class))
if little_cpu_class.require_caches():
m5.util.panic("CPU model %s requires caches" %
str(little_cpu_class))
cluster_mem_bus = system.membus
# connect each cluster to the memory hierarchy
for cluster in system._clusters:
cluster.connectMemSide(cluster_mem_bus)
# Linux device tree
system.dtb_filename = SysPaths.binary(options.dtb)
# Get and load from the chkpt or simpoint checkpoint
if options.restore_from is not None:
m5.instantiate(options.restore_from)
else:
m5.instantiate()
# start simulation (and drop checkpoints when requested)
while True:
event = m5.simulate()
exit_msg = event.getCause()
if exit_msg == "checkpoint":
print "Dropping checkpoint at tick %d" % m5.curTick()
cpt_dir = os.path.join(m5.options.outdir, "cpt.%d" % m5.curTick())
m5.checkpoint(os.path.join(cpt_dir))
print "Checkpoint done."
else:
print exit_msg, " @ ", m5.curTick()
break
sys.exit(event.getCode())
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(switched_out=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:
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 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(switched_out=True, cpu_id=(i))
for i in xrange(np)]
switch_cpus_1 = [DerivO3CPU(switched_out=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())
m5.switchCpus(testsys, switch_cpu_list)
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.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 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_event = 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_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(joinpath(cptdir, "cpt.%d"))
if not m5.options.interactive:
sys.exit(exit_event.getCode())
def run(options, root, testsys, cpu_class):
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 range(np):
testsys.cpu[i].progress_interval = options.prog_interval
if options.maxinsts:
for i in range(np):
testsys.cpu[i].max_insts_any_thread = options.maxinsts
if cpu_class:
switch_cpus = [
cpu_class(switched_out=True, cpu_id=(i)) for i in range(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
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.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 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(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 range(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].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:
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 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 \
(options.simpoint or options.at_instruction):
offset = int(options.take_checkpoints)
# Set an instruction break point
if options.simpoint:
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
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 range(np):
testsys.cpu[i].max_insts_any_thread = offset
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))
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()))
m5.switchCpus(testsys, switch_cpu_list)
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.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()
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_event = scriptCheckpoints(options, maxtick, cptdir)
# Take SimPoint checkpoints
elif options.take_simpoint_checkpoints != None:
takeSimpointCheckpoints(simpoints, interval_length, cptdir)
# Restore from SimPoint checkpoints
elif options.restore_simpoint_checkpoint != None:
restoreSimpointCheckpoint()
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_event = repeatSwitch(testsys, repeat_switch_cpu_list, maxtick,
options.repeat_switch)
elif options.advancedstats:
while 1:
exit_event = m5.simulate(
options.dumpinterval
) # The time scale is nano second (one tick)
m5.stats.dump()
# m5.stats.reset()
# print('MESSAGE', exit_event.getCause())
if (exit_event.getCause() ==
'exiting with last active thread context'):
break
print("all sample completed")
else:
exit_event = benchCheckpoints(options, maxtick, cptdir)
print('Exiting @ tick %i because %s' %
(m5.curTick(), exit_event.getCause()))
system.cpu.createInterruptController()
if isa == 'x86':
system.cpu.interrupts[0].pio = system.membus.master
system.cpu.interrupts[0].int_master = system.membus.slave
system.cpu.interrupts[0].int_slave = system.membus.master
system.system_port = system.membus.slave
system.mem_ctrl = DDR3_1600_8x8()
system.mem_ctrl.range = system.mem_ranges[0]
system.mem_ctrl.port = system.membus.master
process = Process()
apps_path = "/project/linuxlab/gem5/test_progs/"
if program == "daxpy" and isa == "x86":
process.cmd = [apps_path + '/daxpy/daxpy_x86']
elif program == "daxpy" and isa == "arm":
process.cmd = [apps_path + '/daxpy/daxpy_arm']
elif program == "queens" and isa == "x86":
process.cmd = [apps_path + '/queens/queens_x86']
process.cmd += ["10 -c"]
elif program == "queens" and isa == "arm":
process.cmd = [apps_path + 'queens/queens_arm']
process.cmd += ["10 -c"]
system.cpu.workload = process
system.cpu.createThreads()
root = Root(full_system=False, system=system)
m5.instantiate()
print("Beginning simulation!")
exit_event = m5.simulate()
print('Exiting @ tick because '.format(m5.curTick(), exit_event.getCause()))
# int controllers
system.cpu.createInterruptController()
system.cpu.interrupts[0].pio = system.membus.master
system.cpu.interrupts[0].int_master = system.membus.slave
system.cpu.interrupts[0].int_slave = system.membus.master
system.system_port = system.membus.slave
system.mem_ctrl = DDR3_1600_8x8()
system.mem_ctrl.range = system.mem_ranges[0]
system.mem_ctrl.port = system.membus.master
process = Process()
process.cmd = ['tests/test-progs/hello/bin/x86/linux/hello']
system.cpu.workload = process
system.cpu.createThreads()
# all the simulations have to have a root object, absolutely required,
# all others are optionsl
root = Root(full_system=False, system=system)
# walk through all the above python objects and create all C++ objects
m5.instantiate()
print "beginning simulation"
exit_event = m5.simulate()
print "done simulating at tick %d because %s!!" % (m5.curTick(),
exit_event.getCause())
checkpoint_dir = None
m5.instantiate(checkpoint_dir)
# Map workload to this address space
host_cpu.workload[0].map(0x10000000, 0x200000000, 4096)
if options.fast_forward:
print("Switch at instruction count: %d" % cpu_list[0].max_insts_any_thread)
exit_event = m5.simulate(maxtick)
if options.fast_forward:
if exit_event.getCause() == "a thread reached the max instruction count":
m5.switchCpus(system, switch_cpu_list)
print("Switched CPUS @ tick %s" % (m5.curTick()))
m5.stats.reset()
exit_event = m5.simulate(maxtick - m5.curTick())
elif options.fast_forward_pseudo_op:
while exit_event.getCause() == "switchcpu":
# If we are switching *to* kvm, then the current stats are meaningful
# Note that we don't do any warmup by default
if type(switch_cpu_list[0][0]) == FutureCpuClass:
print("Dumping stats...")
m5.stats.dump()
m5.switchCpus(system, switch_cpu_list)
print("Switched CPUS @ tick %s" % (m5.curTick()))
m5.stats.reset()
# This lets us switch back and forth without keeping a counter
switch_cpu_list = [(x[1], x[0]) for x in switch_cpu_list]
exit_event = m5.simulate(maxtick - m5.curTick())
def main():
parser = argparse.ArgumentParser(epilog=__doc__)
parser.add_argument("commands_to_run",
metavar="command(s)",
nargs='*',
help="Command(s) to run")
parser.add_argument("--cpu",
type=str,
choices=cpu_types.keys(),
default="atomic",
help="CPU model to use")
parser.add_argument("--cpu-freq", type=str, default="4GHz")
parser.add_argument("--num-cores",
type=int,
default=1,
help="Number of CPU cores")
parser.add_argument("--mem-type",
default="DDR3_1600_8x8",
choices=ObjectList.mem_list.get_names(),
help="type of memory to use")
parser.add_argument("--mem-channels",
type=int,
default=2,
help="number of memory channels")
parser.add_argument("--mem-ranks",
type=int,
default=None,
help="number of memory ranks per channel")
parser.add_argument("--mem-size",
action="store",
type=str,
default="2GB",
help="Specify the physical memory size")
args = parser.parse_args()
# Create a single root node for gem5's object hierarchy. There can
# only exist one root node in the simulator at any given
# time. Tell gem5 that we want to use syscall emulation mode
# instead of full system mode.
root = Root(full_system=False)
# Populate the root node with a system. A system corresponds to a
# single node with shared memory.
root.system = create(args)
# Instantiate the C++ object hierarchy. After this point,
# SimObjects can't be instantiated anymore.
m5.instantiate()
# Start the simulator. This gives control to the C++ world and
# starts the simulator. The returned event tells the simulation
# script why the simulator exited.
event = m5.simulate()
# Print the reason for the simulation exit. Some exit codes are
# requests for service (e.g., checkpoints) from the simulation
# script. We'll just ignore them here and exit.
print(event.getCause(), " @ ", m5.curTick())
sys.exit(event.getCode())
def __call__(self):
print "Progress! Time now %fs" % (m5.curTick() / 1e12)
self.eventq.schedule(self, m5.curTick() + self.period)
def __init__(self, eventq, period):
super(ProgressEvent, self).__init__()
self.period = int(period)
self.eventq = eventq
self.eventq.schedule(self, m5.curTick() + self.period)
def get_current_tick(self) -> int:
"""
Returns the current tick.
"""
return m5.curTick()
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":
# cuiwl: dump and reset stats.
print "Checkpoint %d." % (num_checkpoints)
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":
# dump and reset stats.
print "Checkpoint %d." % (num_checkpoints)
m5.stats.dump()
m5.stats.reset()
m5.checkpoint(joinpath(cptdir, "cpt.%d"))
num_checkpoints += 1
if options.stat_prefix:
shutil.copyfile("/usr/local/google/home/cuiwl/gem5/m5out/stats.txt", "/usr/local/google/home/cuiwl/gem5/m5out/stats/" + options.stat_prefix + "_" + time.strftime("%Y%m%d") + "_" + time.strftime("%H:%M:%S") + ".txt")
else:
shutil.copyfile("/usr/local/google/home/cuiwl/gem5/m5out/stats.txt", "/usr/local/google/home/cuiwl/gem5/m5out/stats/none" + "_" + time.strftime("%Y%m%d") + "_" + time.strftime("%H:%M:%S") + ".txt")
return exit_event
numa_cache_upward.append(domain.numa_cache_upward)
system.mem_ranges.append(domain.mem_ranges)
mem_ctrl.append(domain.mem_ctrl)
membus.append(domain.membus)
l2bus.append(domain.l2bus)
l2cache.append(domain.l2cache)
cpu += domain.cpu
system.numa_cache_downward = numa_cache_downward
system.numa_cache_upward = numa_cache_upward
system.mem_ctrl = mem_ctrl
system.membus = membus
system.l2bus = l2bus
system.l2cache = l2cache
system.cpu = cpu
system.system_port = system.systembus.slave
root = Root(full_system = False, system = system)
m5.instantiate()
exit_event = m5.simulate()
print 'Exiting @ tick ' + str(m5.curTick()) + ' because ' + exit_event.getCause()
make_level(spec[1:], prototypes[1:], cache, "cpu_side")
else:
# we just built the MemTest objects
parent.cpu = objs
for t in objs:
t.test = getattr(attach_obj, attach_port)
t.functional = system.funcmem.port
make_level(treespec, prototypes, system.physmem, "port")
# -----------------------
# run simulation
# -----------------------
root = Root( system = system )
if options.atomic:
root.system.mem_mode = 'atomic'
else:
root.system.mem_mode = 'timing'
# Not much point in this being higher than the L1 latency
m5.ticks.setGlobalFrequency('1ns')
# instantiate configuration
m5.instantiate(root)
# simulate until program terminates
exit_event = m5.simulate(options.maxtick)
print 'Exiting @ tick', m5.curTick(), 'because', exit_event.getCause()
system.cpu.createInterruptController()
system.cpu.interrupts[0].pio = system.membus.master
system.cpu.interrupts[0].int_master = system.membus.slave
system.cpu.interrupts[0].int_slave = system.membus.master
system.system_port = system.membus.slave
system.mem_ctrl = DDR3_1600_8x8()
system.mem_ctrl.range = system.mem_ranges[0]
system.mem_ctrl.port = system.membus.master
process = Process()
process.cmd = ['tests/test-progs/hello/bin/x86/linux/hello']
system.cpu.workload = process
system.cpu.createThreads()
root = Root(full_system = False, system = system)
m5.instantiate()
print("Beginning simulation!")
exit_event = m5.simulate()
print('Exiting @ tick {} because {}'
.format(m5.curTick(), exit_event.getCause()))
process = Process()
system.cpu.process = process
system.cpu.max_insts_any_thread = opts.maxinsts or default_max_insts
# Set the command
# cmd is a list which begins with the executable (like argv)
process.cmd = binary + input_file
process.output = output_file
# Set the cpu to use the process as its workload and create thread contexts
system.cpu.workload = process
system.cpu.createThreads()
# set up the root SimObject and start the simulation
root = Root(full_system = False, system = system)
# instantiate all of the objects we've created above
m5.instantiate()
# this code allows us to dump stats in intermediate states, every n instructions
dump_n_instructions = 100000
i = 0
while True:
i += dump_n_instructions
system.cpu.scheduleInstStop(0, dump_n_instructions, 'dump statistics')
event = m5.simulate()
if event.getCause() == 'dump statistics':
m5.stats.dump()
print(str(datetime.datetime.now()), 'Instruction Count: {}, Dumping stats'.format(i))
else:
print('Exiting @ tick %i because %s', m5.curTick(), event.getCause())
break
def main():
parser = argparse.ArgumentParser(
description="Generic ARM big.LITTLE configuration")
parser.add_argument("--restore-from",
type=str,
default=None,
help="Restore from checkpoint")
parser.add_argument("--dtb",
type=str,
default=default_dtb,
help="DTB file to load")
parser.add_argument("--kernel",
type=str,
default=default_kernel,
help="Linux kernel")
parser.add_argument("--disk",
action="append",
type=str,
default=[],
help="Disks to instantiate")
parser.add_argument("--bootscript",
type=str,
default=default_rcs,
help="Linux bootscript")
parser.add_argument("--atomic",
action="store_true",
default=False,
help="Use atomic CPUs")
parser.add_argument("--kernel-init",
type=str,
default="/sbin/init",
help="Override init")
parser.add_argument("--big-cpus",
type=int,
default=1,
help="Number of big CPUs to instantiate")
parser.add_argument("--little-cpus",
type=int,
default=1,
help="Number of little CPUs to instantiate")
parser.add_argument("--caches",
action="store_true",
default=False,
help="Instantiate caches")
parser.add_argument("--last-cache-level",
type=int,
default=2,
help="Last level of caches (e.g. 3 for L3)")
parser.add_argument("--big-cpu-clock",
type=str,
default="2GHz",
help="Big CPU clock frequency")
parser.add_argument("--little-cpu-clock",
type=str,
default="1GHz",
help="Little CPU clock frequency")
m5.ticks.fixGlobalFrequency()
options = parser.parse_args()
kernel_cmd = [
"earlyprintk=pl011,0x1c090000",
"console=ttyAMA0",
"lpj=19988480",
"norandmaps",
"loglevel=8",
"mem=%s" % default_mem_size,
"root=/dev/vda1",
"rw",
"init=%s" % options.kernel_init,
"vmalloc=768MB",
]
root = Root(full_system=True)
disks = default_disk if len(options.disk) == 0 else options.disk
system = createSystem(options.kernel, options.bootscript, disks=disks)
root.system = system
system.boot_osflags = " ".join(kernel_cmd)
AtomicCluster = devices.AtomicCluster
if options.big_cpus + options.little_cpus == 0:
m5.util.panic("Empty CPU clusters")
# big cluster
if options.big_cpus > 0:
if options.atomic:
system.bigCluster = AtomicCluster(system, options.big_cpus,
options.big_cpu_clock)
else:
system.bigCluster = BigCluster(system, options.big_cpus,
options.big_cpu_clock)
mem_mode = system.bigCluster.memoryMode()
# little cluster
if options.little_cpus > 0:
if options.atomic:
system.littleCluster = AtomicCluster(system, options.little_cpus,
options.little_cpu_clock)
else:
system.littleCluster = LittleCluster(system, options.little_cpus,
options.little_cpu_clock)
mem_mode = system.littleCluster.memoryMode()
if options.big_cpus > 0 and options.little_cpus > 0:
if system.bigCluster.memoryMode() != system.littleCluster.memoryMode():
m5.util.panic("Memory mode missmatch among CPU clusters")
system.mem_mode = mem_mode
# create caches
system.addCaches(options.caches, options.last_cache_level)
if not options.caches:
if options.big_cpus > 0 and system.bigCluster.requireCaches():
m5.util.panic("Big CPU model requires caches")
if options.little_cpus > 0 and system.littleCluster.requireCaches():
m5.util.panic("Little CPU model requires caches")
# Linux device tree
system.dtb_filename = SysPaths.binary(options.dtb)
# Get and load from the chkpt or simpoint checkpoint
if options.restore_from is not None:
m5.instantiate(options.restore_from)
else:
m5.instantiate()
# start simulation (and drop checkpoints when requested)
while True:
event = m5.simulate()
exit_msg = event.getCause()
if exit_msg == "checkpoint":
print "Dropping checkpoint at tick %d" % m5.curTick()
cpt_dir = os.path.join(m5.options.outdir, "cpt.%d" % m5.curTick())
m5.checkpoint(os.path.join(cpt_dir))
print "Checkpoint done."
else:
print exit_msg, " @ ", m5.curTick()
break
sys.exit(event.getCode())
def main():
parser = argparse.ArgumentParser(
description="Generic ARM big.LITTLE configuration")
parser.add_argument("--restore-from", type=str, default=None,
help="Restore from checkpoint")
parser.add_argument("--dtb", type=str, default=default_dtb,
help="DTB file to load")
parser.add_argument("--kernel", type=str, default=default_kernel,
help="Linux kernel")
parser.add_argument("--disk", action="append", type=str, default=[],
help="Disks to instantiate")
parser.add_argument("--bootscript", type=str, default=default_rcs,
help="Linux bootscript")
parser.add_argument("--atomic", action="store_true", default=False,
help="Use atomic CPUs")
parser.add_argument("--kernel-init", type=str, default="/sbin/init",
help="Override init")
parser.add_argument("--big-cpus", type=int, default=1,
help="Number of big CPUs to instantiate")
parser.add_argument("--little-cpus", type=int, default=1,
help="Number of little CPUs to instantiate")
parser.add_argument("--caches", action="store_true", default=False,
help="Instantiate caches")
parser.add_argument("--last-cache-level", type=int, default=2,
help="Last level of caches (e.g. 3 for L3)")
parser.add_argument("--big-cpu-clock", type=str, default="2GHz",
help="Big CPU clock frequency")
parser.add_argument("--little-cpu-clock", type=str, default="1GHz",
help="Little CPU clock frequency")
m5.ticks.fixGlobalFrequency()
options = parser.parse_args()
kernel_cmd = [
"earlyprintk=pl011,0x1c090000",
"console=ttyAMA0",
"lpj=19988480",
"norandmaps",
"loglevel=8",
"mem=%s" % default_mem_size,
"root=/dev/vda1",
"rw",
"init=%s" % options.kernel_init,
"vmalloc=768MB",
]
root = Root(full_system=True)
disks = [default_disk] if len(options.disk) == 0 else options.disk
system = createSystem(options.caches,
options.kernel,
options.bootscript,
disks=disks)
root.system = system
system.boot_osflags = " ".join(kernel_cmd)
AtomicCluster = devices.AtomicCluster
if options.big_cpus + options.little_cpus == 0:
m5.util.panic("Empty CPU clusters")
# big cluster
if options.big_cpus > 0:
if options.atomic:
system.bigCluster = AtomicCluster(system, options.big_cpus,
options.big_cpu_clock)
else:
system.bigCluster = BigCluster(system, options.big_cpus,
options.big_cpu_clock)
mem_mode = system.bigCluster.memoryMode()
# little cluster
if options.little_cpus > 0:
if options.atomic:
system.littleCluster = AtomicCluster(system, options.little_cpus,
options.little_cpu_clock)
else:
system.littleCluster = LittleCluster(system, options.little_cpus,
options.little_cpu_clock)
mem_mode = system.littleCluster.memoryMode()
if options.big_cpus > 0 and options.little_cpus > 0:
if system.bigCluster.memoryMode() != system.littleCluster.memoryMode():
m5.util.panic("Memory mode missmatch among CPU clusters")
system.mem_mode = mem_mode
# create caches
system.addCaches(options.caches, options.last_cache_level)
if not options.caches:
if options.big_cpus > 0 and system.bigCluster.requireCaches():
m5.util.panic("Big CPU model requires caches")
if options.little_cpus > 0 and system.littleCluster.requireCaches():
m5.util.panic("Little CPU model requires caches")
# Linux device tree
system.dtb_filename = SysPaths.binary(options.dtb)
# Get and load from the chkpt or simpoint checkpoint
if options.restore_from is not None:
m5.instantiate(options.restore_from)
else:
m5.instantiate()
# start simulation (and drop checkpoints when requested)
while True:
event = m5.simulate()
exit_msg = event.getCause()
if exit_msg == "checkpoint":
print "Dropping checkpoint at tick %d" % m5.curTick()
cpt_dir = os.path.join(m5.options.outdir, "cpt.%d" % m5.curTick())
m5.checkpoint(os.path.join(cpt_dir))
print "Checkpoint done."
else:
print exit_msg, " @ ", m5.curTick()
break
sys.exit(event.getCode())
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"))
dispatcher.cpu = host_cpu
dispatcher.shader_pointer = shader
dispatcher.cl_driver = driver
########################## Start simulation ########################
root = Root(system=system, full_system=False)
m5.ticks.setGlobalFrequency('1THz')
if options.abs_max_tick:
maxtick = options.abs_max_tick
else:
maxtick = m5.MaxTick
# Benchmarks support work item annotations
Simulation.setWorkCountOptions(system, options)
# Checkpointing is not supported by APU model
if (options.checkpoint_dir != None or options.checkpoint_restore != None):
fatal("Checkpointing not supported by apu model")
checkpoint_dir = None
m5.instantiate(checkpoint_dir)
# Map workload to this address space
host_cpu.workload[0].map(0x10000000, 0x200000000, 4096)
exit_event = m5.simulate(maxtick)
print "Ticks:", m5.curTick()
print 'Exiting because ', exit_event.getCause()
sys.exit(exit_event.getCode())
# obj1.master = obj2.slave <==> obj2.slave = obj1.master # But the direction of data stream between cache and the membus should be double direction ? system.cpu.createInterruptController() print type(system.cpu.interrupts.pio) system.cpu.interrupts.pio = system.membus.master system.cpu.interrupts.int_master = system.membus.slave system.cpu.interrupts.int_slave = system.membus.master system.system_port = system.membus.slave # x86 specific, allow system to r/w memory system.mem_ctrl = DDR3_1600_x64() # memory controller system.mem_ctrl.range = system.mem_ranges[0] system.mem_ctrl.port = system.membus.master process = LiveProcess() process.cmd = ['tests/test-progs/hello/bin/x86/linux/hello'] system.cpu.workload = process system.cpu.createThreads() root = Root(full_system=False, system=system) m5.instantiate() print 'Beginning simulation!' exit_event = m5.simulate() print 'Exiting @ tick %i because %s' % (m5.curTick(), exit_event.getCause())
root.sim_quantum = int(1e9) # 1 ms
# instantiate all of the objects we've created above
m5.instantiate()
# Keep running until we are done.
print("Running the simulation")
curCpus = 'system.cpu'
exit_event = m5.simulate()
while exit_event.getCause() != "m5_exit instruction encountered":
if exit_event.getCause() == "user interrupt received":
print("User interrupt. Exiting")
break
print('Pause @ tick %i because %s' %
(m5.curTick(), exit_event.getCause()))
if exit_event.getCause() == "switchcpu":
if 'system.cpu' == curCpus:
system.switchCpus(system.cpu, system.timingCpu)
curCpus = 'system.timingCpu'
else:
system.switchCpus(system.timingCpu, system.cpu)
curCpus = 'system.cpu'
m5.stats.reset()
exit_event = m5.simulate(100000000)
elif exit_event.getCause() == "simulate() limit reached":
if 'system.timingCpu' == curCpus:
m5.stats.dump()
m5.stats.reset()
print("Continuing...")
system.mem_ctrl.range = system.mem_ranges[0]
system.mem_ctrl.port = system.membus.master
'''
Process setup
'''
# Process
process = Process()
process.cmd = [binary]
# Set CPU to execute process
system.cpu.workload = process
# Create the execution threads
system.cpu.createThreads()
'''
Simulation environment setup
'''
# Root of the simulation
root = Root(full_system=False, system=system)
'''
Run the setup
'''
# Instantiating the simulator
m5.instantiate()
# Run the simulator, kick off discrete event simulation
exit_event = m5.simulate()
m5.stats.dump()
print("{} at tick {}.".format(exit_event.getCause(), m5.curTick()))
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"))
system.cpu = TimingSimpleCPU() system.membus = SystemXBar() system.cpu.icache_port = system.membus.slave system.cpu.dcache_port = system.membus.slave system.cpu.createInterruptController() system.cpu.interrupts[0].pio = system.membus.master system.cpu.interrupts[0].int_master = system.membus.slave system.cpu.interrupts[0].int_slave = system.membus.master system.system_port = system.membus.slave system.mem_ctrl = DDR3_1600_8x8() system.mem_ctrl.range = system.mem_ranges[0] system.mem_ctrl.port = system.membus.master process = Process() #process.cmd = ['tests/test-progs/hello/bin/x86/linux/hello'] process.cmd = ['tests/test-progs/hello/bin/x86/linux/loop'] system.cpu.workload = process system.cpu.createThreads() root = Root(full_system=False, system=system) m5.instantiate() print "Beginning Simulation" exit_event = m5.simulate() print 'Exiting @ tick %i' % (m5.curTick())
system.system_port = system.membus.slave
#Caches
if options.caches or options.l2cache:
system.iocache = IOCache(addr_ranges=[system.physmem.range])
system.iocache.cpu_side = system.iobus.master
system.iocache.mem_side = system.membus.slave
else:
system.iobridge = Bridge(delay='50ns',
ranges = [system.physmem.range])
system.iobridge.slave = system.iobus.master
system.iobridge.master = system.membus.slave
CacheConfig.config_cache(options, system)
#######################################################################
#
# Start simulation
root = Root(full_system=True, system=system)
m5.instantiate()
print '..... STARTING SIMULATION'
exit_event = m5.simulate()
exit_cause = exit_event.getCause()
print 'Exiting @ tick %i because %s' % (m5.curTick(), exit_cause)
# instantiate all of the objects we've created above
m5.instantiate()
globalStart = time.time()
print("Running the simulation")
print("Using cpu: {}".format(cpu))
exit_event = m5.simulate()
if exit_event.getCause() == "workbegin":
# Reached the start of ROI
# start of ROI is marked by an
# m5_work_begin() call
print("Resetting stats at the start of ROI!")
m5.stats.reset()
start_tick = m5.curTick()
start_insts = system.totalInsts()
# switching cpu if argument cpu == atomic or timing
if cpu == 'atomic':
system.switchCpus(system.cpu, system.atomicCpu)
if cpu == 'timing':
system.switchCpus(system.cpu, system.timingCpu)
else:
print("Unexpected termination of simulation !")
exit()
# Simulate the ROI
exit_event = m5.simulate()
# Reached the end of ROI
# Finish executing the benchmark
def __call__(self):
print("Progress! Time now %fs" % (m5.curTick()/1e12))
self.eventq.schedule(self, m5.curTick() + self.period)
isa = str(m5.defines.buildEnv['TARGET_ISA']).lower()
# Run application and use the compiled ISA to find the binary
# grab the specific path to the binary
thispath = os.path.dirname(os.path.realpath(__file__))
binary = os.path.join(thispath, '../../../', 'tests/test-progs/threads/bin/',
isa, 'linux/threads')
# Create a process for a simple "multi-threaded" application
process = Process()
# Set the command
# cmd is a list which begins with the executable (like argv)
process.cmd = [binary]
# Set the cpu to use the process as its workload and create thread contexts
for cpu in system.cpu:
cpu.workload = process
cpu.createThreads()
# Set up the pseudo file system for the threads function above
config_filesystem(system)
# set up the root SimObject and start the simulation
root = Root(full_system=False, system=system)
# instantiate all of the objects we've created above
m5.instantiate()
print("Beginning simulation!")
exit_event = m5.simulate()
print('Exiting @ tick {} because {}'.format(m5.curTick(),
exit_event.getCause()))
ctrl.port = system.membus.master
if args.dump_period:
m5.internal.stats.periodicStatDump(int(args.dump_period * 1e9))
root = m5.objects.Root(full_system=False, system=system)
if args.fast_forward:
cpu_types[args.cpu_type].numThreads = args.num_threads
system.switch_cpus = [
cpu_types[args.cpu_type](switched_out=True, cpu_id=i)
for i in xrange(args.num_cpus)
]
for i in xrange(args.num_cpus):
system.cpu[i].max_insts_any_thread = args.fast_forward
system.switch_cpus[i].system = system
system.switch_cpus[i].workload = system.cpu[i].workload
system.switch_cpus[i].clk_domain = system.cpu[i].clk_domain
system.switch_cpus[i].progress_interval = system.cpu[
i].progress_interval
system.switch_cpus[i].createThreads()
if args.max_instructions:
system.switch_cpus[i].max_insts_all_threads = args.max_instructions
m5.instantiate(None)
if args.fast_forward:
exit_event = m5.simulate(args.stop_at_tick)
m5.switchCpus(system, [(system.cpu[i], system.switch_cpus[i])
for i in xrange(args.num_cpus)])
exit_event = m5.simulate(args.stop_at_tick - m5.curTick())
print("Exiting at tick %i because %s" % (m5.curTick(), exit_event.getCause()))
for (i, cpu) in enumerate(cpus):
#
# Tie the cpu memtester ports to the correct system ports
#
cpu.port = system.ruby._cpu_ports[i].slave
#
# Since the memtester is incredibly bursty, increase the deadlock
# threshold to 5 million cycles
#
system.ruby._cpu_ports[i].deadlock_threshold = 5000000
# -----------------------
# run simulation
# -----------------------
root = Root(full_system=False, system=system)
root.system.mem_mode = 'timing'
# Not much point in this being higher than the L1 latency
m5.ticks.setGlobalFrequency('1ns')
# instantiate configuration
m5.instantiate()
# simulate until program terminates
exit_event = m5.simulate(options.abs_max_tick)
print('Exiting @ tick', m5.curTick(), 'because', exit_event.getCause())
for core in system.cpu:
core.workload = process
core.createThreads()
print(bcol.OK("Done"))
print("")
print(bcol.I("Instantiating system ..."))
root = Root(full_system=False, system=system)
m5.instantiate()
print(bcol.OK("Done"))
print("")
print(bcol.H( \
"*********************** Beginning gem5 simulation ************************"))
simulation = m5.simulate()
if simulation.getCause() == "exiting with last active thread context":
print(bcol.H( \
"**************************************************************************"))
print(bcol.OK("Done."))
print(bcol.OK("Finished at tick %i." % (m5.curTick())))
print(bcol.OK("Status: %s." % (simulation.getCause())))
print(bcol.H( \
"**************************************************************************"))
else:
print(bcol.H( \
"**************************************************************************"))
print(bcol.E("Status: %s." % (simulation.getCause())))
print(bcol.H( \
"**************************************************************************"))
def run_system_with_cpu(
process, options, output_dir,
warmup_cpu_class=None,
warmup_instructions=0,
real_cpu_create_function=lambda cpu_id: DerivO3CPU(cpu_id=cpu_id),
):
# Override the -d outdir --outdir option to gem5
m5.options.outdir = output_dir
m5.core.setOutputDir(m5.options.outdir)
m5.stats.reset()
max_tick = options.abs_max_tick
if options.rel_max_tick:
max_tick = options.rel_max_tick
elif options.maxtime:
max_tick = int(options.maxtime * 1000 * 1000 * 1000 * 1000)
eprint("Simulating until tick=%s" % (max_tick))
real_cpus = [real_cpu_create_function(0)]
mem_mode = real_cpus[0].memory_mode()
if warmup_cpu_class:
warmup_cpus = [warmup_cpu_class(cpu_id=0)]
warmup_cpus[0].max_insts_any_thread = warmup_instructions
else:
warmup_cpus = real_cpus
system = System(cpu = warmup_cpus,
mem_mode = mem_mode,
mem_ranges = [AddrRange(options.mem_size)],
cache_line_size = options.cacheline_size)
system.multi_thread = False
system.voltage_domain = VoltageDomain(voltage = options.sys_voltage)
system.clk_domain = SrcClockDomain(clock = options.sys_clock,
voltage_domain = system.voltage_domain)
system.cpu_voltage_domain = VoltageDomain()
system.cpu_clk_domain = SrcClockDomain(clock = options.cpu_clock,
voltage_domain =
system.cpu_voltage_domain)
system.cache_line_size = options.cacheline_size
if warmup_cpu_class:
for cpu in real_cpus:
cpu.clk_domain = system.cpu_clk_domain
cpu.workload = process
cpu.system = system
cpu.switched_out = True
cpu.createThreads()
system.switch_cpus = real_cpus
for cpu in system.cpu:
cpu.clk_domain = system.cpu_clk_domain
cpu.workload = process
if options.prog_interval:
cpu.progress_interval = options.prog_interval
cpu.createThreads();
MemClass = Simulation.setMemClass(options)
system.membus = SystemXBar()
system.system_port = system.membus.slave
system.cpu[0].connectAllPorts(system.membus)
MemConfig.config_mem(options, system)
root = Root(full_system = False, system = system)
m5.options.outdir = output_dir
m5.instantiate(None) # None == no checkpoint
if warmup_cpu_class:
eprint("Running warmup with warmup CPU class (%d instrs.)" % (warmup_instructions))
eprint("Starting simulation")
exit_event = m5.simulate(max_tick)
if warmup_cpu_class:
max_tick -= m5.curTick()
m5.stats.reset()
debug_print("Finished warmup; running real simulation")
m5.switchCpus(system, real_cpus)
exit_event = m5.simulate(max_tick)
eprint("Done simulation @ tick = %s: %s" % (m5.curTick(), exit_event.getCause()))
m5.stats.dump()
system.readfile = writeBenchScript(m5.options.outdir, benchmark_name,
benchmark_size, synthetic)
m5.disableAllListeners()
# set up the root SimObject and start the simulation
root = Root(full_system = True, system = system)
root.sim_quantum = int(1e9) # 1 ms
# instantiate all of the objects we've created above
m5.instantiate()
print("Running the simulation with: ", system.cpu)
exit_event = m5.simulate()
print('Exiting @ tick {} because {}'
.format(m5.curTick(), exit_event.getCause()))
if exit_event.getCause() == "work started count reach":
m5.stats.reset()
start_tick = m5.curTick()
start_insts = system.totalInsts()
if cpu_type != 'kvm':
system.switchToTiming()
print("Switch to detailed cpu model")
else:
print("ROI is not annotated!")
print('Exiting @ tick {} because {}'
.format(m5.curTick(), exit_event.getCause()))
exit()
exit_event = m5.simulate()
if m5.defines.buildEnv['TARGET_ISA'] == "x86":
system.cpu.interrupts[0].pio = system.membus.master
system.cpu.interrupts[0].int_master = system.membus.slave
system.cpu.interrupts[0].int_slave = system.membus.master
# Connect the system up to the membus
system.system_port = system.membus.slave
# Create a DDR3 memory controller
system.mem_ctrl = DDR3_1600_x64()
system.mem_ctrl.range = system.mem_ranges[0]
system.mem_ctrl.port = system.membus.master
# Create a process for a simple "Hello World" application
process = LiveProcess()
# Set the command
# cmd is a list which begins with the executable (like argv)
process.cmd = [binary]
# Set the cpu to use the process as its workload and create thread contexts
system.cpu.workload = process
system.cpu.createThreads()
# set up the root SimObject and start the simulation
root = Root(full_system=False, system=system)
# instantiate all of the objects we've created above
m5.instantiate()
print "Beginning simulation!"
exit_event = m5.simulate()
print 'Exiting @ tick %i because %s' % (m5.curTick(), exit_event.getCause())
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, 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"))
root = Root(full_system=True, system=board)
# sim_quantum must be set if KVM cores are used.
root.sim_quantum = int(1e9)
m5.instantiate()
# We maintain the wall clock time.
globalStart = time.time()
print("Running the simulation")
print("Using KVM cpu")
start_tick = m5.curTick()
end_tick = m5.curTick()
m5.stats.reset()
# We start the simulation
exit_event = m5.simulate()
# The first exit_event ends with a `workbegin` cause. This means that the
# system booted successfully and the execution on the program started.
if exit_event.getCause() == "workbegin":
print("Done booting Linux")
print("Resetting stats at the start of ROI!")
def run(options, root, testsys, cpu_class):
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(switched_out=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].clk_domain = testsys.cpu[i].clk_domain
switch_cpus[i].progress_interval = testsys.cpu[i].progress_interval
# 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:
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 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].clk_domain = testsys.cpu[i].clk_domain
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(switched_out=True, cpu_id=(i))
for i in xrange(np)]
switch_cpus_1 = [DerivO3CPU(switched_out=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].clk_domain = testsys.cpu[i].clk_domain
switch_cpus_1[i].clk_domain = testsys.cpu[i].clk_domain
# 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
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)
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)
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())
m5.switchCpus(testsys, switch_cpu_list)
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.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()
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_event = scriptCheckpoints(options, maxtick, cptdir)
# Take SimPoint checkpoints
elif options.take_simpoint_checkpoints != None:
takeSimpointCheckpoints(simpoints, interval_length, cptdir)
# Restore from SimPoint checkpoints
elif options.restore_simpoint_checkpoint != None:
restoreSimpointCheckpoint()
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_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(joinpath(cptdir, "cpt.%d"))
if not m5.options.interactive:
sys.exit(exit_event.getCode())
import SimpleOpts
from system import MySystem
SimpleOpts.add_option("--script",
default='',
help="Script to execute in the simulated system")
if __name__ == "__m5_main__":
(opts, args) = SimpleOpts.parse_args()
# create the system we are going to simulate
system = MySystem(opts)
# Read in the script file passed in via an option.
# This file gets read and executed by the simulated system after boot.
# Note: The disk image needs to be configured to do this.
system.readfile = opts.script
# set up the root SimObject and start the simulation
root = Root(full_system=True, system=system)
# instantiate all of the objects we've created above
m5.instantiate()
# Keep running until we are done.
print "Running the simulation"
exit_event = m5.simulate()
print 'Exiting @ tick %i because %s' % (m5.curTick(),
exit_event.getCause())
tester.port = xbar.slave
else:
# Single tester
testers[0].port = next_cache.cpu_side
# Top level call to create the cache hierarchy, bottom up
make_cache_level(cachespec, cache_proto, len(cachespec), None)
# Connect the lowest level crossbar to the memory
last_subsys = getattr(system, "l%dsubsys0" % len(cachespec))
last_subsys.xbar.master = system.physmem.port
root = Root(full_system=False, system=system)
if options.atomic:
root.system.mem_mode = "atomic"
else:
root.system.mem_mode = "timing"
# The system port is never used in the tester so merely connect it
# to avoid problems
root.system.system_port = last_subsys.xbar.slave
# Instantiate configuration
m5.instantiate()
# Simulate until program terminates
exit_event = m5.simulate(options.maxtick)
print "Exiting @ tick", m5.curTick(), "because", exit_event.getCause()
def run(options, root, testsys, cpu_class):
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(switched_out=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].clk_domain = testsys.cpu[i].clk_domain
# 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:
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 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].clk_domain = testsys.cpu[i].clk_domain
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(switched_out=True, cpu_id=(i)) for i in xrange(np)
]
switch_cpus_1 = [
DerivO3CPU(switched_out=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].clk_domain = testsys.cpu[i].clk_domain
switch_cpus_1[i].clk_domain = testsys.cpu[i].clk_domain
# 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:
cpt_starttick, checkpoint_dir = findCptDir(options, cptdir, testsys)
m5.instantiate(checkpoint_dir)
# 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)
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())
m5.switchCpus(testsys, switch_cpu_list)
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.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 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_event = 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_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(joinpath(cptdir, "cpt.%d"))
if not m5.options.interactive:
sys.exit(exit_event.getCode())
