blob: 58d762dc60adde5b90a13bfa8f72f6145e75e8b7 [file] [log] [blame]
# Copyright (c) 2015, 2018 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) 2006-2007 The Regents of The University of Michigan
# 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.
import argparse
import random
import sys
import m5
from m5.objects import *
# This example script stress tests the memory system by creating false
# sharing in a tree topology. At the bottom of the tree is a shared
# memory, and then at each level a number of testers are attached,
# along with a number of caches that them selves fan out to subtrees
# of testers and caches. Thus, it is possible to create a system with
# arbitrarily deep cache hierarchies, sharing or no sharing of caches,
# and testers not only at the L1s, but also at the L2s, L3s etc.
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter
)
parser.add_argument(
"-a", "--atomic", action="store_true", help="Use atomic (non-timing) mode"
)
parser.add_argument(
"-b", "--blocking", action="store_true", help="Use blocking caches"
)
parser.add_argument(
"-l", "--maxloads", metavar="N", default=0, help="Stop after N loads"
)
parser.add_argument(
"-m",
"--maxtick",
type=int,
default=m5.MaxTick,
metavar="T",
help="Stop after T ticks",
)
# The tree specification consists of two colon-separated lists of one
# or more integers, one for the caches, and one for the testers. The
# first integer is the number of caches/testers closest to main
# memory. Each cache then fans out to a subtree. The last integer in
# the list is the number of caches/testers associated with the
# uppermost level of memory. The other integers (if any) specify the
# number of caches/testers connected at each level of the crossbar
# hierarchy. The tester string should have one element more than the
# cache string as there should always be testers attached to the
# uppermost caches.
parser.add_argument(
"-c",
"--caches",
type=str,
default="2:2:1",
help="Colon-separated cache hierarchy specification, "
"see script comments for details ",
)
parser.add_argument(
"--noncoherent-cache",
action="store_true",
help="Adds a non-coherent, last-level cache",
)
parser.add_argument(
"-t",
"--testers",
type=str,
default="1:1:0:2",
help="Colon-separated tester hierarchy specification, "
"see script comments for details ",
)
parser.add_argument(
"-f",
"--functional",
type=int,
default=10,
metavar="PCT",
help="Target percentage of functional accesses ",
)
parser.add_argument(
"-u",
"--uncacheable",
type=int,
default=10,
metavar="PCT",
help="Target percentage of uncacheable accesses ",
)
parser.add_argument(
"-r",
"--random",
action="store_true",
help="Generate a random tree topology",
)
parser.add_argument(
"--progress",
type=int,
default=100000,
metavar="NLOADS",
help="Progress message interval ",
)
parser.add_argument(
"--sys-clock",
action="store",
type=str,
default="1GHz",
help="""Top-level clock for blocks running at system
speed""",
)
args = parser.parse_args()
# Get the total number of testers
def numtesters(cachespec, testerspec):
# Determine the tester multiplier for each level as the
# elements are per subsystem and it fans out
multiplier = [1]
for c in cachespec:
multiplier.append(multiplier[-1] * c)
total = 0
for t, m in zip(testerspec, multiplier):
total += t * m
return total
block_size = 64
# Start by parsing the command line args and do some basic sanity
# checking
if args.random:
# Generate a tree with a valid number of testers
while True:
tree_depth = random.randint(1, 4)
cachespec = [random.randint(1, 3) for i in range(tree_depth)]
testerspec = [random.randint(1, 3) for i in range(tree_depth + 1)]
if numtesters(cachespec, testerspec) < block_size:
break
print(
"Generated random tree -c",
":".join(map(str, cachespec)),
"-t",
":".join(map(str, testerspec)),
)
else:
try:
cachespec = [int(x) for x in args.caches.split(":")]
testerspec = [int(x) for x in args.testers.split(":")]
except:
print("Error: Unable to parse caches or testers option")
sys.exit(1)
if len(cachespec) < 1:
print("Error: Must have at least one level of caches")
sys.exit(1)
if len(cachespec) != len(testerspec) - 1:
print("Error: Testers must have one element more than caches")
sys.exit(1)
if testerspec[-1] == 0:
print("Error: Must have testers at the uppermost level")
sys.exit(1)
for t in testerspec:
if t < 0:
print("Error: Cannot have a negative number of testers")
sys.exit(1)
for c in cachespec:
if c < 1:
print("Error: Must have 1 or more caches at each level")
sys.exit(1)
if numtesters(cachespec, testerspec) > block_size:
print(
"Error: Limited to %s testers because of false sharing"
% (block_size)
)
sys.exit(1)
# Define a prototype L1 cache that we scale for all successive levels
proto_l1 = Cache(
size="32kB",
assoc=4,
tag_latency=1,
data_latency=1,
response_latency=1,
tgts_per_mshr=8,
clusivity="mostly_incl",
writeback_clean=True,
)
if args.blocking:
proto_l1.mshrs = 1
else:
proto_l1.mshrs = 4
cache_proto = [proto_l1]
# Now add additional cache levels (if any) by scaling L1 params, the
# first element is Ln, and the last element L1
for scale in cachespec[:-1]:
# Clone previous level and update params
prev = cache_proto[0]
next = prev()
next.size = prev.size * scale
next.tag_latency = prev.tag_latency * 10
next.data_latency = prev.data_latency * 10
next.response_latency = prev.response_latency * 10
next.assoc = prev.assoc * scale
next.mshrs = prev.mshrs * scale
# Swap the inclusivity/exclusivity at each level. L2 is mostly
# exclusive with respect to L1, L3 mostly inclusive, L4 mostly
# exclusive etc.
next.writeback_clean = not prev.writeback_clean
if prev.clusivity.value == "mostly_incl":
next.clusivity = "mostly_excl"
else:
next.clusivity = "mostly_incl"
cache_proto.insert(0, next)
# Make a prototype for the tester to be used throughout
proto_tester = MemTest(
max_loads=args.maxloads,
percent_functional=args.functional,
percent_uncacheable=args.uncacheable,
progress_interval=args.progress,
)
# Set up the system along with a simple memory and reference memory
system = System(physmem=SimpleMemory(), cache_line_size=block_size)
system.voltage_domain = VoltageDomain(voltage="1V")
system.clk_domain = SrcClockDomain(
clock=args.sys_clock, voltage_domain=system.voltage_domain
)
# For each level, track the next subsys index to use
next_subsys_index = [0] * (len(cachespec) + 1)
# Recursive function to create a sub-tree of the cache and tester
# hierarchy
def make_cache_level(ncaches, prototypes, level, next_cache):
global next_subsys_index, proto_l1, testerspec, proto_tester
index = next_subsys_index[level]
next_subsys_index[level] += 1
# Create a subsystem to contain the crossbar and caches, and
# any testers
subsys = SubSystem()
setattr(system, "l%dsubsys%d" % (level, index), subsys)
# The levels are indexing backwards through the list
ntesters = testerspec[len(cachespec) - level]
# Scale the progress threshold as testers higher up in the tree
# (smaller level) get a smaller portion of the overall bandwidth,
# and also make the interval of packet injection longer for the
# testers closer to the memory (larger level) to prevent them
# hogging all the bandwidth
limit = (len(cachespec) - level + 1) * 100000000
testers = [
proto_tester(interval=10 * (level * level + 1), progress_check=limit)
for i in range(ntesters)
]
if ntesters:
subsys.tester = testers
if level != 0:
# Create a crossbar and add it to the subsystem, note that
# we do this even with a single element on this level
xbar = L2XBar()
subsys.xbar = xbar
if next_cache:
xbar.mem_side_ports = next_cache.cpu_side
# Create and connect the caches, both the ones fanning out
# to create the tree, and the ones used to connect testers
# on this level
tree_caches = [prototypes[0]() for i in range(ncaches[0])]
tester_caches = [proto_l1() for i in range(ntesters)]
subsys.cache = tester_caches + tree_caches
for cache in tree_caches:
cache.mem_side = xbar.cpu_side_ports
make_cache_level(ncaches[1:], prototypes[1:], level - 1, cache)
for tester, cache in zip(testers, tester_caches):
tester.port = cache.cpu_side
cache.mem_side = xbar.cpu_side_ports
else:
if not next_cache:
print("Error: No next-level cache at top level")
sys.exit(1)
if ntesters > 1:
# Create a crossbar and add it to the subsystem
xbar = L2XBar()
subsys.xbar = xbar
xbar.mem_side_ports = next_cache.cpu_side
for tester in testers:
tester.port = xbar.cpu_side_ports
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 last-level cache and memory
# controller
last_subsys = getattr(system, "l%dsubsys0" % len(cachespec))
last_subsys.xbar.point_of_coherency = True
if args.noncoherent_cache:
system.llc = NoncoherentCache(
size="16MB",
assoc=16,
tag_latency=10,
data_latency=10,
sequential_access=True,
response_latency=20,
tgts_per_mshr=8,
mshrs=64,
)
last_subsys.xbar.mem_side_ports = system.llc.cpu_side
system.llc.mem_side = system.physmem.port
else:
last_subsys.xbar.mem_side_ports = system.physmem.port
root = Root(full_system=False, system=system)
if args.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.cpu_side_ports
# Instantiate configuration
m5.instantiate()
# Simulate until program terminates
exit_event = m5.simulate(args.maxtick)
print("Exiting @ tick", m5.curTick(), "because", exit_event.getCause())