memtest.py (10720:67b3e74de9ae) memtest.py (10750:30efc3828bb4)
1# Copyright (c) 2015 ARM Limited
2# All rights reserved.
3#
4# The license below extends only to copyright in the software and shall
5# not be construed as granting a license to any other intellectual
6# property including but not limited to intellectual property relating
7# to a hardware implementation of the functionality of the software
8# licensed hereunder. You may use the software subject to the license

--- 26 unchanged lines hidden (view full) ---

35# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
36# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
37# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38#
39# Authors: Ron Dreslinski
40# Andreas Hansson
41
42import optparse
1# Copyright (c) 2015 ARM Limited
2# All rights reserved.
3#
4# The license below extends only to copyright in the software and shall
5# not be construed as granting a license to any other intellectual
6# property including but not limited to intellectual property relating
7# to a hardware implementation of the functionality of the software
8# licensed hereunder. You may use the software subject to the license

--- 26 unchanged lines hidden (view full) ---

35# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
36# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
37# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38#
39# Authors: Ron Dreslinski
40# Andreas Hansson
41
42import optparse
43import random
43import sys
44
45import m5
46from m5.objects import *
47
44import sys
45
46import m5
47from m5.objects import *
48
49# This example script stress tests the memory system by creating false
50# sharing in a tree topology. At the bottom of the tree is a shared
51# memory, and then at each level a number of testers are attached,
52# along with a number of caches that them selves fan out to subtrees
53# of testers and caches. Thus, it is possible to create a system with
54# arbitrarily deep cache hierarchies, sharing or no sharing of caches,
55# and testers not only at the L1s, but also at the L2s, L3s etc.
56
48parser = optparse.OptionParser()
49
50parser.add_option("-a", "--atomic", action="store_true",
51 help="Use atomic (non-timing) mode")
52parser.add_option("-b", "--blocking", action="store_true",
53 help="Use blocking caches")
54parser.add_option("-l", "--maxloads", metavar="N", default=0,
55 help="Stop after N loads")
56parser.add_option("-m", "--maxtick", type="int", default=m5.MaxTick,
57 metavar="T",
58 help="Stop after T ticks")
59
57parser = optparse.OptionParser()
58
59parser.add_option("-a", "--atomic", action="store_true",
60 help="Use atomic (non-timing) mode")
61parser.add_option("-b", "--blocking", action="store_true",
62 help="Use blocking caches")
63parser.add_option("-l", "--maxloads", metavar="N", default=0,
64 help="Stop after N loads")
65parser.add_option("-m", "--maxtick", type="int", default=m5.MaxTick,
66 metavar="T",
67 help="Stop after T ticks")
68
60# This example script stress tests the memory system by creating false
61# sharing in a tree topology. At the bottom of the tree is a shared
62# memory, and then at each level a number of testers are attached,
63# along with a number of caches that them selves fan out to subtrees
64# of testers and caches. Thus, it is possible to create a system with
65# arbitrarily deep cache hierarchies, sharing or no sharing of caches,
66# and testers not only at the L1s, but also at the L2s, L3s etc.
67#
68# The tree specification consists of two colon-separated lists of one
69# or more integers, one for the caches, and one for the testers. The
70# first integer is the number of caches/testers closest to main
71# memory. Each cache then fans out to a subtree. The last integer in
72# the list is the number of caches/testers associated with the
73# uppermost level of memory. The other integers (if any) specify the
74# number of caches/testers connected at each level of the crossbar
75# hierarchy. The tester string should have one element more than the

--- 11 unchanged lines hidden (view full) ---

87parser.add_option("-f", "--functional", type="int", default=0,
88 metavar="PCT",
89 help="Target percentage of functional accesses "
90 "[default: %default]")
91parser.add_option("-u", "--uncacheable", type="int", default=0,
92 metavar="PCT",
93 help="Target percentage of uncacheable accesses "
94 "[default: %default]")
69# The tree specification consists of two colon-separated lists of one
70# or more integers, one for the caches, and one for the testers. The
71# first integer is the number of caches/testers closest to main
72# memory. Each cache then fans out to a subtree. The last integer in
73# the list is the number of caches/testers associated with the
74# uppermost level of memory. The other integers (if any) specify the
75# number of caches/testers connected at each level of the crossbar
76# hierarchy. The tester string should have one element more than the

--- 11 unchanged lines hidden (view full) ---

88parser.add_option("-f", "--functional", type="int", default=0,
89 metavar="PCT",
90 help="Target percentage of functional accesses "
91 "[default: %default]")
92parser.add_option("-u", "--uncacheable", type="int", default=0,
93 metavar="PCT",
94 help="Target percentage of uncacheable accesses "
95 "[default: %default]")
95
96parser.add_option("--progress", type="int", default=10000,
96parser.add_option("-r", "--random", action="store_true",
97 help="Generate a random tree topology")
98parser.add_option("--progress", type="int", default=100000,
97 metavar="NLOADS",
98 help="Progress message interval "
99 "[default: %default]")
100parser.add_option("--sys-clock", action="store", type="string",
101 default='1GHz',
102 help = """Top-level clock for blocks running at system
103 speed""")
104
105(options, args) = parser.parse_args()
106
107if args:
108 print "Error: script doesn't take any positional arguments"
109 sys.exit(1)
110
99 metavar="NLOADS",
100 help="Progress message interval "
101 "[default: %default]")
102parser.add_option("--sys-clock", action="store", type="string",
103 default='1GHz',
104 help = """Top-level clock for blocks running at system
105 speed""")
106
107(options, args) = parser.parse_args()
108
109if args:
110 print "Error: script doesn't take any positional arguments"
111 sys.exit(1)
112
113# Get the total number of testers
114def numtesters(cachespec, testerspec):
115 # Determine the tester multiplier for each level as the
116 # elements are per subsystem and it fans out
117 multiplier = [1]
118 for c in cachespec:
119 multiplier.append(multiplier[-1] * c)
120
121 total = 0
122 for t, m in zip(testerspec, multiplier):
123 total += t * m
124
125 return total
126
111block_size = 64
112
113# Start by parsing the command line options and do some basic sanity
114# checking
127block_size = 64
128
129# Start by parsing the command line options and do some basic sanity
130# checking
115try:
116 cachespec = [int(x) for x in options.caches.split(':')]
117 testerspec = [int(x) for x in options.testers.split(':')]
118except:
119 print "Error: Unable to parse caches or testers option"
120 sys.exit(1)
131if options.random:
132 # Generate a tree with a valid number of testers
133 while True:
134 tree_depth = random.randint(1, 4)
135 cachespec = [random.randint(1, 3) for i in range(tree_depth)]
136 testerspec = [random.randint(1, 3) for i in range(tree_depth + 1)]
137 if numtesters(cachespec, testerspec) < block_size:
138 break
121
139
122if len(cachespec) < 1:
123 print "Error: Must have at least one level of caches"
124 sys.exit(1)
140 print "Generated random tree -c", ':'.join(map(str, cachespec)), \
141 "-t", ':'.join(map(str, testerspec))
142else:
143 try:
144 cachespec = [int(x) for x in options.caches.split(':')]
145 testerspec = [int(x) for x in options.testers.split(':')]
146 except:
147 print "Error: Unable to parse caches or testers option"
148 sys.exit(1)
125
149
126if len(cachespec) != len(testerspec) - 1:
127 print "Error: Testers must have one element more than caches"
128 sys.exit(1)
150 if len(cachespec) < 1:
151 print "Error: Must have at least one level of caches"
152 sys.exit(1)
129
153
130if testerspec[-1] == 0:
131 print "Error: Must have testers at the uppermost level"
132 sys.exit(1)
133
134for t in testerspec:
135 if t < 0:
136 print "Error: Cannot have a negative number of testers"
154 if len(cachespec) != len(testerspec) - 1:
155 print "Error: Testers must have one element more than caches"
137 sys.exit(1)
138
156 sys.exit(1)
157
139for c in cachespec:
140 if c < 1:
141 print "Error: Must have 1 or more caches at each level"
158 if testerspec[-1] == 0:
159 print "Error: Must have testers at the uppermost level"
142 sys.exit(1)
143
160 sys.exit(1)
161
144# Determine the tester multiplier for each level as the string
145# elements are per subsystem and it fans out
146multiplier = [1]
147for c in cachespec:
148 if c < 1:
149 print "Error: Must have at least one cache per level"
150 multiplier.append(multiplier[-1] * c)
162 for t in testerspec:
163 if t < 0:
164 print "Error: Cannot have a negative number of testers"
165 sys.exit(1)
151
166
152numtesters = 0
153for t, m in zip(testerspec, multiplier):
154 numtesters += t * m
167 for c in cachespec:
168 if c < 1:
169 print "Error: Must have 1 or more caches at each level"
170 sys.exit(1)
155
171
156if numtesters > block_size:
157 print "Error: Number of testers limited to %s because of false sharing" \
158 % (block_size)
159 sys.exit(1)
172 if numtesters(cachespec, testerspec) > block_size:
173 print "Error: Limited to %s testers because of false sharing" \
174 % (block_size)
175 sys.exit(1)
160
161# Define a prototype L1 cache that we scale for all successive levels
162proto_l1 = BaseCache(size = '32kB', assoc = 4,
163 hit_latency = 1, response_latency = 1,
164 tgts_per_mshr = 8, is_top_level = True)
165
166if options.blocking:
167 proto_l1.mshrs = 1

--- 50 unchanged lines hidden (view full) ---

218 # The levels are indexing backwards through the list
219 ntesters = testerspec[len(cachespec) - level]
220
221 # Scale the progress threshold as testers higher up in the tree
222 # (smaller level) get a smaller portion of the overall bandwidth,
223 # and also make the interval of packet injection longer for the
224 # testers closer to the memory (larger level) to prevent them
225 # hogging all the bandwidth
176
177# Define a prototype L1 cache that we scale for all successive levels
178proto_l1 = BaseCache(size = '32kB', assoc = 4,
179 hit_latency = 1, response_latency = 1,
180 tgts_per_mshr = 8, is_top_level = True)
181
182if options.blocking:
183 proto_l1.mshrs = 1

--- 50 unchanged lines hidden (view full) ---

234 # The levels are indexing backwards through the list
235 ntesters = testerspec[len(cachespec) - level]
236
237 # Scale the progress threshold as testers higher up in the tree
238 # (smaller level) get a smaller portion of the overall bandwidth,
239 # and also make the interval of packet injection longer for the
240 # testers closer to the memory (larger level) to prevent them
241 # hogging all the bandwidth
226 limit = (len(cachespec) - level + 1) * 10000000
242 limit = (len(cachespec) - level + 1) * 100000000
227 testers = [proto_tester(interval = 10 * (level * level + 1),
228 progress_check = limit) \
229 for i in xrange(ntesters)]
230 if ntesters:
231 subsys.tester = testers
232
233 if level != 0:
234 # Create a crossbar and add it to the subsystem, note that

--- 59 unchanged lines hidden --- 		243 testers = [proto_tester(interval = 10 * (level * level + 1),
244 progress_check = limit) \
245 for i in xrange(ntesters)]
246 if ntesters:
247 subsys.tester = testers
248
249 if level != 0:
250 # Create a crossbar and add it to the subsystem, note that

--- 59 unchanged lines hidden ---