gem5/configs/common/Simulation.py

517 lines
21 KiB
Python
Raw Normal View History

sim: Add the notion of clock domains to all ClockedObjects This patch adds the notion of source- and derived-clock domains to the ClockedObjects. As such, all clock information is moved to the clock domain, and the ClockedObjects are grouped into domains. The clock domains are either source domains, with a specific clock period, or derived domains that have a parent domain and a divider (potentially chained). For piece of logic that runs at a derived clock (a ratio of the clock its parent is running at) the necessary derived clock domain is created from its corresponding parent clock domain. For now, the derived clock domain only supports a divider, thus ensuring a lower speed compared to its parent. Multiplier functionality implies a PLL logic that has not been modelled yet (create a separate clock instead). The clock domains should be used as a mechanism to provide a controllable clock source that affects clock for every clocked object lying beneath it. The clock of the domain can (in a future patch) be controlled by a handler responsible for dynamic frequency scaling of the respective clock domains. All the config scripts have been retro-fitted with clock domains. For the System a default SrcClockDomain is created. For CPUs that run at a different speed than the system, there is a seperate clock domain created. This domain incorporates the CPU and the associated caches. As before, Ruby runs under its own clock domain. The clock period of all domains are pre-computed, such that no virtual functions or multiplications are needed when calling clockPeriod. Instead, the clock period is pre-computed when any changes occur. For this to be possible, each clock domain tracks its children.
2013-06-27 11:49:49 +02:00
# Copyright (c) 2012-2013 ARM Limited
# All rights reserved
#
# The license below extends only to copyright in the software and shall
# not be construed as granting a license to any other intellectual
# property including but not limited to intellectual property relating
# to a hardware implementation of the functionality of the software
# licensed hereunder. You may use the software subject to the license
# terms below provided that you ensure that this notice is replicated
# unmodified and in its entirety in all distributions of the software,
# modified or unmodified, in source code or in binary form.
#
# Copyright (c) 2006-2008 The Regents of The University of Michigan
# Copyright (c) 2010 Advanced Micro Devices, Inc.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Lisa Hsu
import sys
from os import getcwd
from os.path import join as joinpath
import CpuConfig
import MemConfig
import m5
from m5.defines import buildEnv
from m5.objects import *
from m5.util import *
addToPath('../common')
def getCPUClass(cpu_type):
"""Returns the required cpu class and the mode of operation."""
cls = CpuConfig.get(cpu_type)
return cls, cls.memory_mode()
def setCPUClass(options):
"""Returns two cpu classes and the initial mode of operation.
Restoring from a checkpoint or fast forwarding through a benchmark
can be done using one type of cpu, and then the actual
simulation can be carried out using another type. This function
returns these two types of cpus and the initial mode of operation
depending on the options provided.
"""
TmpClass, test_mem_mode = getCPUClass(options.cpu_type)
CPUClass = None
if TmpClass.require_caches() and \
not options.caches and not options.ruby:
fatal("%s must be used with caches" % options.cpu_type)
if options.checkpoint_restore != None:
if options.restore_with_cpu != options.cpu_type:
CPUClass = TmpClass
TmpClass, test_mem_mode = getCPUClass(options.restore_with_cpu)
elif options.fast_forward:
CPUClass = TmpClass
TmpClass = AtomicSimpleCPU
test_mem_mode = 'atomic'
return (TmpClass, test_mem_mode, CPUClass)
def setMemClass(options):
"""Returns a memory controller class."""
return MemConfig.get(options.mem_type)
def setWorkCountOptions(system, options):
if options.work_item_id != None:
system.work_item_id = options.work_item_id
if options.work_begin_cpu_id_exit != None:
system.work_begin_cpu_id_exit = options.work_begin_cpu_id_exit
if options.work_end_exit_count != None:
system.work_end_exit_count = options.work_end_exit_count
if options.work_end_checkpoint_count != None:
system.work_end_ckpt_count = options.work_end_checkpoint_count
if options.work_begin_exit_count != None:
system.work_begin_exit_count = options.work_begin_exit_count
if options.work_begin_checkpoint_count != None:
system.work_begin_ckpt_count = options.work_begin_checkpoint_count
if options.work_cpus_checkpoint_count != None:
system.work_cpus_ckpt_count = options.work_cpus_checkpoint_count
def findCptDir(options, cptdir, testsys):
"""Figures out the directory from which the checkpointed state is read.
There are two different ways in which the directories holding checkpoints
can be named --
1. cpt.<benchmark name>.<instruction count when the checkpoint was taken>
2. cpt.<some number, usually the tick value when the checkpoint was taken>
This function parses through the options to figure out which one of the
above should be used for selecting the checkpoint, and then figures out
the appropriate directory.
"""
from os.path import isdir, exists
from os import listdir
import re
if not isdir(cptdir):
fatal("checkpoint dir %s does not exist!", cptdir)
cpt_starttick = 0
if options.at_instruction or options.simpoint:
inst = options.checkpoint_restore
if options.simpoint:
# assume workload 0 has the simpoint
if testsys.cpu[0].workload[0].simpoint == 0:
fatal('Unable to find simpoint')
inst += int(testsys.cpu[0].workload[0].simpoint)
checkpoint_dir = joinpath(cptdir, "cpt.%s.%s" % (options.bench, inst))
if not exists(checkpoint_dir):
fatal("Unable to find checkpoint directory %s", checkpoint_dir)
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)
cpt_starttick = int(cpts[cpt_num - 1])
checkpoint_dir = joinpath(cptdir, "cpt.%s" % cpts[cpt_num - 1])
return cpt_starttick, checkpoint_dir
def scriptCheckpoints(options, maxtick, cptdir):
if options.at_instruction or options.simpoint:
checkpoint_inst = int(options.take_checkpoints)
# maintain correct offset if we restored from some instruction
if options.checkpoint_restore != None:
checkpoint_inst += options.checkpoint_restore
print "Creating checkpoint at inst:%d" % (checkpoint_inst)
exit_event = m5.simulate()
exit_cause = exit_event.getCause()
print "exit cause = %s" % exit_cause
# skip checkpoint instructions should they exist
while exit_cause == "checkpoint":
exit_event = m5.simulate()
exit_cause = exit_event.getCause()
if exit_cause == "a thread reached the max instruction count":
m5.checkpoint(joinpath(cptdir, "cpt.%s.%d" % \
(options.bench, checkpoint_inst)))
print "Checkpoint written."
else:
when, period = options.take_checkpoints.split(",", 1)
when = int(when)
period = int(period)
num_checkpoints = 0
exit_event = m5.simulate(when - m5.curTick())
exit_cause = exit_event.getCause()
while exit_cause == "checkpoint":
exit_event = m5.simulate(when - m5.curTick())
exit_cause = exit_event.getCause()
if exit_cause == "simulate() limit reached":
m5.checkpoint(joinpath(cptdir, "cpt.%d"))
num_checkpoints += 1
sim_ticks = when
max_checkpoints = options.max_checkpoints
while num_checkpoints < max_checkpoints and \
exit_cause == "simulate() limit reached":
if (sim_ticks + period) > maxtick:
exit_event = m5.simulate(maxtick - sim_ticks)
exit_cause = exit_event.getCause()
break
else:
exit_event = m5.simulate(period)
exit_cause = exit_event.getCause()
sim_ticks += period
while exit_event.getCause() == "checkpoint":
exit_event = m5.simulate(sim_ticks - m5.curTick())
if exit_event.getCause() == "simulate() limit reached":
m5.checkpoint(joinpath(cptdir, "cpt.%d"))
num_checkpoints += 1
return exit_event
def benchCheckpoints(options, maxtick, cptdir):
exit_event = m5.simulate(maxtick - m5.curTick())
exit_cause = exit_event.getCause()
num_checkpoints = 0
max_checkpoints = options.max_checkpoints
while exit_cause == "checkpoint":
m5.checkpoint(joinpath(cptdir, "cpt.%d"))
num_checkpoints += 1
if num_checkpoints == max_checkpoints:
exit_cause = "maximum %d checkpoints dropped" % max_checkpoints
break
exit_event = m5.simulate(maxtick - m5.curTick())
exit_cause = exit_event.getCause()
return exit_event
def repeatSwitch(testsys, repeat_switch_cpu_list, maxtick, switch_freq):
print "starting switch loop"
while True:
exit_event = m5.simulate(switch_freq)
exit_cause = exit_event.getCause()
if exit_cause != "simulate() limit reached":
return exit_event
m5.switchCpus(testsys, repeat_switch_cpu_list)
tmp_cpu_list = []
for old_cpu, new_cpu in repeat_switch_cpu_list:
tmp_cpu_list.append((new_cpu, old_cpu))
repeat_switch_cpu_list = tmp_cpu_list
if (maxtick - m5.curTick()) <= switch_freq:
exit_event = m5.simulate(maxtick - m5.curTick())
return exit_event
def run(options, root, testsys, cpu_class):
if options.checkpoint_dir:
cptdir = options.checkpoint_dir
elif m5.options.outdir:
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
sim: Add the notion of clock domains to all ClockedObjects This patch adds the notion of source- and derived-clock domains to the ClockedObjects. As such, all clock information is moved to the clock domain, and the ClockedObjects are grouped into domains. The clock domains are either source domains, with a specific clock period, or derived domains that have a parent domain and a divider (potentially chained). For piece of logic that runs at a derived clock (a ratio of the clock its parent is running at) the necessary derived clock domain is created from its corresponding parent clock domain. For now, the derived clock domain only supports a divider, thus ensuring a lower speed compared to its parent. Multiplier functionality implies a PLL logic that has not been modelled yet (create a separate clock instead). The clock domains should be used as a mechanism to provide a controllable clock source that affects clock for every clocked object lying beneath it. The clock of the domain can (in a future patch) be controlled by a handler responsible for dynamic frequency scaling of the respective clock domains. All the config scripts have been retro-fitted with clock domains. For the System a default SrcClockDomain is created. For CPUs that run at a different speed than the system, there is a seperate clock domain created. This domain incorporates the CPU and the associated caches. As before, Ruby runs under its own clock domain. The clock period of all domains are pre-computed, such that no virtual functions or multiplications are needed when calling clockPeriod. Instead, the clock period is pre-computed when any changes occur. For this to be possible, each clock domain tracks its children.
2013-06-27 11:49:49 +02:00
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
sim: Add the notion of clock domains to all ClockedObjects This patch adds the notion of source- and derived-clock domains to the ClockedObjects. As such, all clock information is moved to the clock domain, and the ClockedObjects are grouped into domains. The clock domains are either source domains, with a specific clock period, or derived domains that have a parent domain and a divider (potentially chained). For piece of logic that runs at a derived clock (a ratio of the clock its parent is running at) the necessary derived clock domain is created from its corresponding parent clock domain. For now, the derived clock domain only supports a divider, thus ensuring a lower speed compared to its parent. Multiplier functionality implies a PLL logic that has not been modelled yet (create a separate clock instead). The clock domains should be used as a mechanism to provide a controllable clock source that affects clock for every clocked object lying beneath it. The clock of the domain can (in a future patch) be controlled by a handler responsible for dynamic frequency scaling of the respective clock domains. All the config scripts have been retro-fitted with clock domains. For the System a default SrcClockDomain is created. For CPUs that run at a different speed than the system, there is a seperate clock domain created. This domain incorporates the CPU and the associated caches. As before, Ruby runs under its own clock domain. The clock period of all domains are pre-computed, such that no virtual functions or multiplications are needed when calling clockPeriod. Instead, the clock period is pre-computed when any changes occur. For this to be possible, each clock domain tracks its children.
2013-06-27 11:49:49 +02:00
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
sim: Add the notion of clock domains to all ClockedObjects This patch adds the notion of source- and derived-clock domains to the ClockedObjects. As such, all clock information is moved to the clock domain, and the ClockedObjects are grouped into domains. The clock domains are either source domains, with a specific clock period, or derived domains that have a parent domain and a divider (potentially chained). For piece of logic that runs at a derived clock (a ratio of the clock its parent is running at) the necessary derived clock domain is created from its corresponding parent clock domain. For now, the derived clock domain only supports a divider, thus ensuring a lower speed compared to its parent. Multiplier functionality implies a PLL logic that has not been modelled yet (create a separate clock instead). The clock domains should be used as a mechanism to provide a controllable clock source that affects clock for every clocked object lying beneath it. The clock of the domain can (in a future patch) be controlled by a handler responsible for dynamic frequency scaling of the respective clock domains. All the config scripts have been retro-fitted with clock domains. For the System a default SrcClockDomain is created. For CPUs that run at a different speed than the system, there is a seperate clock domain created. This domain incorporates the CPU and the associated caches. As before, Ruby runs under its own clock domain. The clock period of all domains are pre-computed, such that no virtual functions or multiplications are needed when calling clockPeriod. Instead, the clock period is pre-computed when any changes occur. For this to be possible, each clock domain tracks its children.
2013-06-27 11:49:49 +02:00
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())