MOESI_CMP_token.py (9826:014ff1fbff6d) MOESI_CMP_token.py (9841:69c158420c51)
1# Copyright (c) 2006-2007 The Regents of The University of Michigan
2# Copyright (c) 2009 Advanced Micro Devices, Inc.
3# All rights reserved.
4#
5# Redistribution and use in source and binary forms, with or without
6# modification, are permitted provided that the following conditions are
7# met: redistributions of source code must retain the above copyright
8# notice, this list of conditions and the following disclaimer;
9# redistributions in binary form must reproduce the above copyright
10# notice, this list of conditions and the following disclaimer in the
11# documentation and/or other materials provided with the distribution;
12# neither the name of the copyright holders nor the names of its
13# contributors may be used to endorse or promote products derived from
14# this software without specific prior written permission.
15#
16# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27#
28# Authors: Brad Beckmann
29
30import math
31import m5
32from m5.objects import *
33from m5.defines import buildEnv
34from Ruby import create_topology
35
36#
37# Note: the L1 Cache latency is only used by the sequencer on fast path hits
38#
39class L1Cache(RubyCache):
40 latency = 2
41
42#
43# Note: the L2 Cache latency is not currently used
44#
45class L2Cache(RubyCache):
46 latency = 10
47
48def define_options(parser):
49 parser.add_option("--l1-retries", type="int", default=1,
50 help="Token_CMP: # of l1 retries before going persistent")
51 parser.add_option("--timeout-latency", type="int", default=300,
52 help="Token_CMP: cycles until issuing again");
53 parser.add_option("--disable-dyn-timeouts", action="store_true",
54 help="Token_CMP: disable dyanimc timeouts, use fixed latency instead")
55 parser.add_option("--allow-atomic-migration", action="store_true",
56 help="allow migratory sharing for atomic only accessed blocks")
57
58def create_system(options, system, piobus, dma_ports, ruby_system):
59
60 if buildEnv['PROTOCOL'] != 'MOESI_CMP_token':
61 panic("This script requires the MOESI_CMP_token protocol to be built.")
62
63 #
64 # number of tokens that the owner passes to requests so that shared blocks can
65 # respond to read requests
66 #
67 n_tokens = options.num_cpus + 1
68
69 cpu_sequencers = []
70
71 #
72 # The ruby network creation expects the list of nodes in the system to be
73 # consistent with the NetDest list. Therefore the l1 controller nodes must be
74 # listed before the directory nodes and directory nodes before dma nodes, etc.
75 #
76 l1_cntrl_nodes = []
77 l2_cntrl_nodes = []
78 dir_cntrl_nodes = []
79 dma_cntrl_nodes = []
80
81 #
82 # Must create the individual controllers before the network to ensure the
83 # controller constructors are called before the network constructor
84 #
85 l2_bits = int(math.log(options.num_l2caches, 2))
86 block_size_bits = int(math.log(options.cacheline_size, 2))
87
88 cntrl_count = 0
89
90 for i in xrange(options.num_cpus):
91 #
92 # First create the Ruby objects associated with this cpu
93 #
94 l1i_cache = L1Cache(size = options.l1i_size,
95 assoc = options.l1i_assoc,
96 start_index_bit = block_size_bits)
97 l1d_cache = L1Cache(size = options.l1d_size,
98 assoc = options.l1d_assoc,
99 start_index_bit = block_size_bits)
100
101 l1_cntrl = L1Cache_Controller(version = i,
102 cntrl_id = cntrl_count,
103 L1Icache = l1i_cache,
104 L1Dcache = l1d_cache,
105 l2_select_num_bits = l2_bits,
106 N_tokens = n_tokens,
107 retry_threshold = \
108 options.l1_retries,
109 fixed_timeout_latency = \
110 options.timeout_latency,
111 dynamic_timeout_enabled = \
112 not options.disable_dyn_timeouts,
113 no_mig_atomic = not \
114 options.allow_atomic_migration,
115 send_evictions = (
116 options.cpu_type == "detailed"),
1# Copyright (c) 2006-2007 The Regents of The University of Michigan
2# Copyright (c) 2009 Advanced Micro Devices, Inc.
3# All rights reserved.
4#
5# Redistribution and use in source and binary forms, with or without
6# modification, are permitted provided that the following conditions are
7# met: redistributions of source code must retain the above copyright
8# notice, this list of conditions and the following disclaimer;
9# redistributions in binary form must reproduce the above copyright
10# notice, this list of conditions and the following disclaimer in the
11# documentation and/or other materials provided with the distribution;
12# neither the name of the copyright holders nor the names of its
13# contributors may be used to endorse or promote products derived from
14# this software without specific prior written permission.
15#
16# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27#
28# Authors: Brad Beckmann
29
30import math
31import m5
32from m5.objects import *
33from m5.defines import buildEnv
34from Ruby import create_topology
35
36#
37# Note: the L1 Cache latency is only used by the sequencer on fast path hits
38#
39class L1Cache(RubyCache):
40 latency = 2
41
42#
43# Note: the L2 Cache latency is not currently used
44#
45class L2Cache(RubyCache):
46 latency = 10
47
48def define_options(parser):
49 parser.add_option("--l1-retries", type="int", default=1,
50 help="Token_CMP: # of l1 retries before going persistent")
51 parser.add_option("--timeout-latency", type="int", default=300,
52 help="Token_CMP: cycles until issuing again");
53 parser.add_option("--disable-dyn-timeouts", action="store_true",
54 help="Token_CMP: disable dyanimc timeouts, use fixed latency instead")
55 parser.add_option("--allow-atomic-migration", action="store_true",
56 help="allow migratory sharing for atomic only accessed blocks")
57
58def create_system(options, system, piobus, dma_ports, ruby_system):
59
60 if buildEnv['PROTOCOL'] != 'MOESI_CMP_token':
61 panic("This script requires the MOESI_CMP_token protocol to be built.")
62
63 #
64 # number of tokens that the owner passes to requests so that shared blocks can
65 # respond to read requests
66 #
67 n_tokens = options.num_cpus + 1
68
69 cpu_sequencers = []
70
71 #
72 # The ruby network creation expects the list of nodes in the system to be
73 # consistent with the NetDest list. Therefore the l1 controller nodes must be
74 # listed before the directory nodes and directory nodes before dma nodes, etc.
75 #
76 l1_cntrl_nodes = []
77 l2_cntrl_nodes = []
78 dir_cntrl_nodes = []
79 dma_cntrl_nodes = []
80
81 #
82 # Must create the individual controllers before the network to ensure the
83 # controller constructors are called before the network constructor
84 #
85 l2_bits = int(math.log(options.num_l2caches, 2))
86 block_size_bits = int(math.log(options.cacheline_size, 2))
87
88 cntrl_count = 0
89
90 for i in xrange(options.num_cpus):
91 #
92 # First create the Ruby objects associated with this cpu
93 #
94 l1i_cache = L1Cache(size = options.l1i_size,
95 assoc = options.l1i_assoc,
96 start_index_bit = block_size_bits)
97 l1d_cache = L1Cache(size = options.l1d_size,
98 assoc = options.l1d_assoc,
99 start_index_bit = block_size_bits)
100
101 l1_cntrl = L1Cache_Controller(version = i,
102 cntrl_id = cntrl_count,
103 L1Icache = l1i_cache,
104 L1Dcache = l1d_cache,
105 l2_select_num_bits = l2_bits,
106 N_tokens = n_tokens,
107 retry_threshold = \
108 options.l1_retries,
109 fixed_timeout_latency = \
110 options.timeout_latency,
111 dynamic_timeout_enabled = \
112 not options.disable_dyn_timeouts,
113 no_mig_atomic = not \
114 options.allow_atomic_migration,
115 send_evictions = (
116 options.cpu_type == "detailed"),
117 transitions_per_cycle = options.ports,
117 ruby_system = ruby_system)
118
119 cpu_seq = RubySequencer(version = i,
120 icache = l1i_cache,
121 dcache = l1d_cache,
122 ruby_system = ruby_system)
123
124 l1_cntrl.sequencer = cpu_seq
125
126 if piobus != None:
127 cpu_seq.pio_port = piobus.slave
128
129 exec("ruby_system.l1_cntrl%d = l1_cntrl" % i)
130 #
131 # Add controllers and sequencers to the appropriate lists
132 #
133 cpu_sequencers.append(cpu_seq)
134 l1_cntrl_nodes.append(l1_cntrl)
135
136 cntrl_count += 1
137
138 l2_index_start = block_size_bits + l2_bits
139
140 for i in xrange(options.num_l2caches):
141 #
142 # First create the Ruby objects associated with this cpu
143 #
144 l2_cache = L2Cache(size = options.l2_size,
145 assoc = options.l2_assoc,
146 start_index_bit = l2_index_start)
147
148 l2_cntrl = L2Cache_Controller(version = i,
149 cntrl_id = cntrl_count,
150 L2cache = l2_cache,
151 N_tokens = n_tokens,
118 ruby_system = ruby_system)
119
120 cpu_seq = RubySequencer(version = i,
121 icache = l1i_cache,
122 dcache = l1d_cache,
123 ruby_system = ruby_system)
124
125 l1_cntrl.sequencer = cpu_seq
126
127 if piobus != None:
128 cpu_seq.pio_port = piobus.slave
129
130 exec("ruby_system.l1_cntrl%d = l1_cntrl" % i)
131 #
132 # Add controllers and sequencers to the appropriate lists
133 #
134 cpu_sequencers.append(cpu_seq)
135 l1_cntrl_nodes.append(l1_cntrl)
136
137 cntrl_count += 1
138
139 l2_index_start = block_size_bits + l2_bits
140
141 for i in xrange(options.num_l2caches):
142 #
143 # First create the Ruby objects associated with this cpu
144 #
145 l2_cache = L2Cache(size = options.l2_size,
146 assoc = options.l2_assoc,
147 start_index_bit = l2_index_start)
148
149 l2_cntrl = L2Cache_Controller(version = i,
150 cntrl_id = cntrl_count,
151 L2cache = l2_cache,
152 N_tokens = n_tokens,
153 transitions_per_cycle = options.ports,
152 ruby_system = ruby_system)
153
154 exec("ruby_system.l2_cntrl%d = l2_cntrl" % i)
155 l2_cntrl_nodes.append(l2_cntrl)
156
157 cntrl_count += 1
158
159 phys_mem_size = sum(map(lambda r: r.size(), system.mem_ranges))
160 assert(phys_mem_size % options.num_dirs == 0)
161 mem_module_size = phys_mem_size / options.num_dirs
162
163 # Run each of the ruby memory controllers at a ratio of the frequency of
164 # the ruby system
165 # clk_divider value is a fix to pass regression.
166 ruby_system.memctrl_clk_domain = DerivedClockDomain(
167 clk_domain=ruby_system.clk_domain,
168 clk_divider=3)
169
170 for i in xrange(options.num_dirs):
171 #
172 # Create the Ruby objects associated with the directory controller
173 #
174
175 mem_cntrl = RubyMemoryControl(
176 clk_domain = ruby_system.memctrl_clk_domain,
177 version = i,
178 ruby_system = ruby_system)
179
180 dir_size = MemorySize('0B')
181 dir_size.value = mem_module_size
182
183 dir_cntrl = Directory_Controller(version = i,
184 cntrl_id = cntrl_count,
185 directory = \
186 RubyDirectoryMemory(version = i,
187 use_map = options.use_map,
188 size = dir_size),
189 memBuffer = mem_cntrl,
190 l2_select_num_bits = l2_bits,
154 ruby_system = ruby_system)
155
156 exec("ruby_system.l2_cntrl%d = l2_cntrl" % i)
157 l2_cntrl_nodes.append(l2_cntrl)
158
159 cntrl_count += 1
160
161 phys_mem_size = sum(map(lambda r: r.size(), system.mem_ranges))
162 assert(phys_mem_size % options.num_dirs == 0)
163 mem_module_size = phys_mem_size / options.num_dirs
164
165 # Run each of the ruby memory controllers at a ratio of the frequency of
166 # the ruby system
167 # clk_divider value is a fix to pass regression.
168 ruby_system.memctrl_clk_domain = DerivedClockDomain(
169 clk_domain=ruby_system.clk_domain,
170 clk_divider=3)
171
172 for i in xrange(options.num_dirs):
173 #
174 # Create the Ruby objects associated with the directory controller
175 #
176
177 mem_cntrl = RubyMemoryControl(
178 clk_domain = ruby_system.memctrl_clk_domain,
179 version = i,
180 ruby_system = ruby_system)
181
182 dir_size = MemorySize('0B')
183 dir_size.value = mem_module_size
184
185 dir_cntrl = Directory_Controller(version = i,
186 cntrl_id = cntrl_count,
187 directory = \
188 RubyDirectoryMemory(version = i,
189 use_map = options.use_map,
190 size = dir_size),
191 memBuffer = mem_cntrl,
192 l2_select_num_bits = l2_bits,
193 transitions_per_cycle = options.ports,
191 ruby_system = ruby_system)
192
193 exec("ruby_system.dir_cntrl%d = dir_cntrl" % i)
194 dir_cntrl_nodes.append(dir_cntrl)
195
196 cntrl_count += 1
197
198 for i, dma_port in enumerate(dma_ports):
199 #
200 # Create the Ruby objects associated with the dma controller
201 #
202 dma_seq = DMASequencer(version = i,
203 ruby_system = ruby_system)
204
205 dma_cntrl = DMA_Controller(version = i,
206 cntrl_id = cntrl_count,
207 dma_sequencer = dma_seq,
194 ruby_system = ruby_system)
195
196 exec("ruby_system.dir_cntrl%d = dir_cntrl" % i)
197 dir_cntrl_nodes.append(dir_cntrl)
198
199 cntrl_count += 1
200
201 for i, dma_port in enumerate(dma_ports):
202 #
203 # Create the Ruby objects associated with the dma controller
204 #
205 dma_seq = DMASequencer(version = i,
206 ruby_system = ruby_system)
207
208 dma_cntrl = DMA_Controller(version = i,
209 cntrl_id = cntrl_count,
210 dma_sequencer = dma_seq,
211 transitions_per_cycle = options.ports,
208 ruby_system = ruby_system)
209
210 exec("ruby_system.dma_cntrl%d = dma_cntrl" % i)
211 exec("ruby_system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
212 dma_cntrl_nodes.append(dma_cntrl)
213 cntrl_count += 1
214
215 all_cntrls = l1_cntrl_nodes + \
216 l2_cntrl_nodes + \
217 dir_cntrl_nodes + \
218 dma_cntrl_nodes
219
220 topology = create_topology(all_cntrls, options)
221
222 return (cpu_sequencers, dir_cntrl_nodes, topology)
212 ruby_system = ruby_system)
213
214 exec("ruby_system.dma_cntrl%d = dma_cntrl" % i)
215 exec("ruby_system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
216 dma_cntrl_nodes.append(dma_cntrl)
217 cntrl_count += 1
218
219 all_cntrls = l1_cntrl_nodes + \
220 l2_cntrl_nodes + \
221 dir_cntrl_nodes + \
222 dma_cntrl_nodes
223
224 topology = create_topology(all_cntrls, options)
225
226 return (cpu_sequencers, dir_cntrl_nodes, topology)