cluster.py revision 11851 
1# Copyright (c) 2006-2007 The Regents of The University of Michigan
2# All rights reserved.
3#
4# Redistribution and use in source and binary forms, with or without
5# modification, are permitted provided that the following conditions are
6# met: redistributions of source code must retain the above copyright
7# notice, this list of conditions and the following disclaimer;
8# redistributions in binary form must reproduce the above copyright
9# notice, this list of conditions and the following disclaimer in the
10# documentation and/or other materials provided with the distribution;
11# neither the name of the copyright holders nor the names of its
12# contributors may be used to endorse or promote products derived from
13# this software without specific prior written permission.
14#
15# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
18# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
19# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
20# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
21# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
25# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26#
27# Authors: Ron Dreslinski
28
29# Simple test script
30#
31# "m5 test.py"
32
33import os
34import optparse
35import sys
36
37import m5
38from m5.objects import *
39
40# --------------------
41# Define Command Line Options
42# ====================
43
44parser = optparse.OptionParser()
45
46parser.add_option("-d", "--detailed", action="store_true")
47parser.add_option("-t", "--timing", action="store_true")
48parser.add_option("-m", "--maxtick", type="int")
49parser.add_option("-c", "--numclusters",
50                  help="Number of clusters", type="int")
51parser.add_option("-n", "--numcpus",
52                  help="Number of cpus in total", type="int")
53parser.add_option("-f", "--frequency",
54                  default = "1GHz",
55                  help="Frequency of each CPU")
56parser.add_option("--l1size",
57                  default = "32kB")
58parser.add_option("--l1latency",
59                  default = 1)
60parser.add_option("--l2size",
61                  default = "256kB")
62parser.add_option("--l2latency",
63                  default = 10)
64parser.add_option("--rootdir",
65                  help="ROot directory of Splash2",
66                  default="/dist/splash2/codes/")
67parser.add_option("-b", "--benchmark",
68                  help="Splash 2 benchmark to run")
69
70(options, args) = parser.parse_args()
71
72if args:
73    print "Error: script doesn't take any positional arguments"
74    sys.exit(1)
75
76# --------------------
77# Define Splash2 Benchmarks
78# ====================
79class Cholesky(Process):
80        executable = options.rootdir + '/kernels/cholesky/CHOLESKY'
81        cmd = 'CHOLESKY -p' + str(options.numcpus) + ' '\
82             + options.rootdir + '/kernels/cholesky/inputs/tk23.O'
83
84class FFT(Process):
85        executable = options.rootdir + 'kernels/fft/FFT'
86        cmd = 'FFT -p' + str(options.numcpus) + ' -m18'
87
88class LU_contig(Process):
89        executable = options.rootdir + 'kernels/lu/contiguous_blocks/LU'
90        cmd = 'LU -p' + str(options.numcpus)
91
92class LU_noncontig(Process):
93        executable = options.rootdir + 'kernels/lu/non_contiguous_blocks/LU'
94        cmd = 'LU -p' + str(options.numcpus)
95
96class Radix(Process):
97        executable = options.rootdir + 'kernels/radix/RADIX'
98        cmd = 'RADIX -n524288 -p' + str(options.numcpus)
99
100class Barnes(Process):
101        executable = options.rootdir + 'apps/barnes/BARNES'
102        cmd = 'BARNES'
103        input = options.rootdir + 'apps/barnes/input.p' + str(options.numcpus)
104
105class FMM(Process):
106        executable = options.rootdir + 'apps/fmm/FMM'
107        cmd = 'FMM'
108        input = options.rootdir + 'apps/fmm/inputs/input.2048.p' + str(options.numcpus)
109
110class Ocean_contig(Process):
111        executable = options.rootdir + 'apps/ocean/contiguous_partitions/OCEAN'
112        cmd = 'OCEAN -p' + str(options.numcpus)
113
114class Ocean_noncontig(Process):
115        executable = options.rootdir + 'apps/ocean/non_contiguous_partitions/OCEAN'
116        cmd = 'OCEAN -p' + str(options.numcpus)
117
118class Raytrace(Process):
119        executable = options.rootdir + 'apps/raytrace/RAYTRACE'
120        cmd = 'RAYTRACE -p' + str(options.numcpus) + ' ' \
121             + options.rootdir + 'apps/raytrace/inputs/teapot.env'
122
123class Water_nsquared(Process):
124        executable = options.rootdir + 'apps/water-nsquared/WATER-NSQUARED'
125        cmd = 'WATER-NSQUARED'
126        input = options.rootdir + 'apps/water-nsquared/input.p' + str(options.numcpus)
127
128class Water_spatial(Process):
129        executable = options.rootdir + 'apps/water-spatial/WATER-SPATIAL'
130        cmd = 'WATER-SPATIAL'
131        input = options.rootdir + 'apps/water-spatial/input.p' + str(options.numcpus)
132
133
134# --------------------
135# Base L1 Cache Definition
136# ====================
137
138class L1(Cache):
139    latency = options.l1latency
140    mshrs = 12
141    tgts_per_mshr = 8
142
143# ----------------------
144# Base L2 Cache Definition
145# ----------------------
146
147class L2(Cache):
148    latency = options.l2latency
149    mshrs = 92
150    tgts_per_mshr = 16
151    write_buffers = 8
152
153# ----------------------
154# Define the clusters with their cpus
155# ----------------------
156class Cluster:
157    pass
158
159cpusPerCluster = options.numcpus/options.numclusters
160
161busFrequency = Frequency(options.frequency)
162busFrequency *= cpusPerCluster
163
164all_cpus = []
165all_l1s = []
166all_l1buses = []
167if options.timing:
168    clusters = [ Cluster() for i in xrange(options.numclusters)]
169    for j in xrange(options.numclusters):
170        clusters[j].id = j
171    for cluster in clusters:
172        cluster.clusterbus = L2XBar(clock=busFrequency)
173        all_l1buses += [cluster.clusterbus]
174        cluster.cpus = [TimingSimpleCPU(cpu_id = i + cluster.id,
175                                        clock=options.frequency)
176                        for i in xrange(cpusPerCluster)]
177        all_cpus += cluster.cpus
178        cluster.l1 = L1(size=options.l1size, assoc = 4)
179        all_l1s += [cluster.l1]
180elif options.detailed:
181    clusters = [ Cluster() for i in xrange(options.numclusters)]
182    for j in xrange(options.numclusters):
183        clusters[j].id = j
184    for cluster in clusters:
185        cluster.clusterbus = L2XBar(clock=busFrequency)
186        all_l1buses += [cluster.clusterbus]
187        cluster.cpus = [DerivO3CPU(cpu_id = i + cluster.id,
188                                   clock=options.frequency)
189                        for i in xrange(cpusPerCluster)]
190        all_cpus += cluster.cpus
191        cluster.l1 = L1(size=options.l1size, assoc = 4)
192        all_l1s += [cluster.l1]
193else:
194    clusters = [ Cluster() for i in xrange(options.numclusters)]
195    for j in xrange(options.numclusters):
196        clusters[j].id = j
197    for cluster in clusters:
198        cluster.clusterbus = L2XBar(clock=busFrequency)
199        all_l1buses += [cluster.clusterbus]
200        cluster.cpus = [AtomicSimpleCPU(cpu_id = i + cluster.id,
201                                        clock=options.frequency)
202                        for i in xrange(cpusPerCluster)]
203        all_cpus += cluster.cpus
204        cluster.l1 = L1(size=options.l1size, assoc = 4)
205        all_l1s += [cluster.l1]
206
207# ----------------------
208# Create a system, and add system wide objects
209# ----------------------
210system = System(cpu = all_cpus, l1_ = all_l1s, l1bus_ = all_l1buses,
211                physmem = SimpleMemory(),
212                membus = SystemXBar(clock = busFrequency))
213system.clock = '1GHz'
214
215system.toL2bus = L2XBar(clock = busFrequency)
216system.l2 = L2(size = options.l2size, assoc = 8)
217
218# ----------------------
219# Connect the L2 cache and memory together
220# ----------------------
221
222system.physmem.port = system.membus.master
223system.l2.cpu_side = system.toL2bus.slave
224system.l2.mem_side = system.membus.master
225
226# ----------------------
227# Connect the L2 cache and clusters together
228# ----------------------
229for cluster in clusters:
230    cluster.l1.cpu_side = cluster.clusterbus.master
231    cluster.l1.mem_side = system.toL2bus.slave
232    for cpu in cluster.cpus:
233        cpu.icache_port = cluster.clusterbus.slave
234        cpu.dcache_port = cluster.clusterbus.slave
235
236# ----------------------
237# Define the root
238# ----------------------
239
240root = Root(full_system = False, system = system)
241
242# --------------------
243# Pick the correct Splash2 Benchmarks
244# ====================
245if options.benchmark == 'Cholesky':
246    root.workload = Cholesky()
247elif options.benchmark == 'FFT':
248    root.workload = FFT()
249elif options.benchmark == 'LUContig':
250    root.workload = LU_contig()
251elif options.benchmark == 'LUNoncontig':
252    root.workload = LU_noncontig()
253elif options.benchmark == 'Radix':
254    root.workload = Radix()
255elif options.benchmark == 'Barnes':
256    root.workload = Barnes()
257elif options.benchmark == 'FMM':
258    root.workload = FMM()
259elif options.benchmark == 'OceanContig':
260    root.workload = Ocean_contig()
261elif options.benchmark == 'OceanNoncontig':
262    root.workload = Ocean_noncontig()
263elif options.benchmark == 'Raytrace':
264    root.workload = Raytrace()
265elif options.benchmark == 'WaterNSquared':
266    root.workload = Water_nsquared()
267elif options.benchmark == 'WaterSpatial':
268    root.workload = Water_spatial()
269else:
270    m5.util.panic("""
271The --benchmark environment variable was set to something improper.
272Use Cholesky, FFT, LUContig, LUNoncontig, Radix, Barnes, FMM, OceanContig,
273OceanNoncontig, Raytrace, WaterNSquared, or WaterSpatial
274""")
275
276# --------------------
277# Assign the workload to the cpus
278# ====================
279
280for cluster in clusters:
281    for cpu in cluster.cpus:
282        cpu.workload = root.workload
283
284# ----------------------
285# Run the simulation
286# ----------------------
287
288if options.timing or options.detailed:
289    root.system.mem_mode = 'timing'
290
291# instantiate configuration
292m5.instantiate()
293
294# simulate until program terminates
295if options.maxtick:
296    exit_event = m5.simulate(options.maxtick)
297else:
298    exit_event = m5.simulate(m5.MaxTick)
299
300print 'Exiting @ tick', m5.curTick(), 'because', exit_event.getCause()
301
302