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
| 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):
| 58def create_system(options, system, 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 for i in xrange(options.num_cpus):
89 #
90 # First create the Ruby objects associated with this cpu
91 #
92 l1i_cache = L1Cache(size = options.l1i_size,
93 assoc = options.l1i_assoc,
94 start_index_bit = block_size_bits)
95 l1d_cache = L1Cache(size = options.l1d_size,
96 assoc = options.l1d_assoc,
97 start_index_bit = block_size_bits)
98
99 l1_cntrl = L1Cache_Controller(version = i,
100 L1Icache = l1i_cache,
101 L1Dcache = l1d_cache,
102 l2_select_num_bits = l2_bits,
103 N_tokens = n_tokens,
104 retry_threshold = \
105 options.l1_retries,
106 fixed_timeout_latency = \
107 options.timeout_latency,
108 dynamic_timeout_enabled = \
109 not options.disable_dyn_timeouts,
110 no_mig_atomic = not \
111 options.allow_atomic_migration,
112 send_evictions = (
113 options.cpu_type == "detailed"),
114 transitions_per_cycle = options.ports,
115 ruby_system = ruby_system)
116
117 cpu_seq = RubySequencer(version = i,
118 icache = l1i_cache,
119 dcache = l1d_cache,
120 ruby_system = ruby_system)
121
122 l1_cntrl.sequencer = cpu_seq
| 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 for i in xrange(options.num_cpus):
89 #
90 # First create the Ruby objects associated with this cpu
91 #
92 l1i_cache = L1Cache(size = options.l1i_size,
93 assoc = options.l1i_assoc,
94 start_index_bit = block_size_bits)
95 l1d_cache = L1Cache(size = options.l1d_size,
96 assoc = options.l1d_assoc,
97 start_index_bit = block_size_bits)
98
99 l1_cntrl = L1Cache_Controller(version = i,
100 L1Icache = l1i_cache,
101 L1Dcache = l1d_cache,
102 l2_select_num_bits = l2_bits,
103 N_tokens = n_tokens,
104 retry_threshold = \
105 options.l1_retries,
106 fixed_timeout_latency = \
107 options.timeout_latency,
108 dynamic_timeout_enabled = \
109 not options.disable_dyn_timeouts,
110 no_mig_atomic = not \
111 options.allow_atomic_migration,
112 send_evictions = (
113 options.cpu_type == "detailed"),
114 transitions_per_cycle = options.ports,
115 ruby_system = ruby_system)
116
117 cpu_seq = RubySequencer(version = i,
118 icache = l1i_cache,
119 dcache = l1d_cache,
120 ruby_system = ruby_system)
121
122 l1_cntrl.sequencer = cpu_seq
|
123
124 if piobus != None:
125 cpu_seq.pio_master_port = piobus.slave
126 cpu_seq.mem_master_port = piobus.slave
127 cpu_seq.pio_slave_port = piobus.master
128
| |
129 exec("ruby_system.l1_cntrl%d = l1_cntrl" % i)
| 123 exec("ruby_system.l1_cntrl%d = l1_cntrl" % i)
|
| 124
|
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 l2_index_start = block_size_bits + l2_bits
137
138 for i in xrange(options.num_l2caches):
139 #
140 # First create the Ruby objects associated with this cpu
141 #
142 l2_cache = L2Cache(size = options.l2_size,
143 assoc = options.l2_assoc,
144 start_index_bit = l2_index_start)
145
146 l2_cntrl = L2Cache_Controller(version = i,
147 L2cache = l2_cache,
148 N_tokens = n_tokens,
149 transitions_per_cycle = options.ports,
150 ruby_system = ruby_system)
151
152 exec("ruby_system.l2_cntrl%d = l2_cntrl" % i)
153 l2_cntrl_nodes.append(l2_cntrl)
154
155 phys_mem_size = sum(map(lambda r: r.size(), system.mem_ranges))
156 assert(phys_mem_size % options.num_dirs == 0)
157 mem_module_size = phys_mem_size / options.num_dirs
158
159 # Run each of the ruby memory controllers at a ratio of the frequency of
160 # the ruby system
161 # clk_divider value is a fix to pass regression.
162 ruby_system.memctrl_clk_domain = DerivedClockDomain(
163 clk_domain=ruby_system.clk_domain,
164 clk_divider=3)
165
166 for i in xrange(options.num_dirs):
167 #
168 # Create the Ruby objects associated with the directory controller
169 #
170
171 mem_cntrl = RubyMemoryControl(
172 clk_domain = ruby_system.memctrl_clk_domain,
173 version = i,
174 ruby_system = ruby_system)
175
176 dir_size = MemorySize('0B')
177 dir_size.value = mem_module_size
178
179 dir_cntrl = Directory_Controller(version = i,
180 directory = \
181 RubyDirectoryMemory(version = i,
182 use_map = options.use_map,
183 size = dir_size),
184 memBuffer = mem_cntrl,
185 l2_select_num_bits = l2_bits,
186 transitions_per_cycle = options.ports,
187 ruby_system = ruby_system)
188
189 exec("ruby_system.dir_cntrl%d = dir_cntrl" % i)
190 dir_cntrl_nodes.append(dir_cntrl)
191
192 for i, dma_port in enumerate(dma_ports):
193 #
194 # Create the Ruby objects associated with the dma controller
195 #
196 dma_seq = DMASequencer(version = i,
197 ruby_system = ruby_system)
198
199 dma_cntrl = DMA_Controller(version = i,
200 dma_sequencer = dma_seq,
201 transitions_per_cycle = options.ports,
202 ruby_system = ruby_system)
203
204 exec("ruby_system.dma_cntrl%d = dma_cntrl" % i)
205 exec("ruby_system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
206 dma_cntrl_nodes.append(dma_cntrl)
207
208 all_cntrls = l1_cntrl_nodes + \
209 l2_cntrl_nodes + \
210 dir_cntrl_nodes + \
211 dma_cntrl_nodes
212
213 topology = create_topology(all_cntrls, options)
214
215 return (cpu_sequencers, dir_cntrl_nodes, topology)
| 125 #
126 # Add controllers and sequencers to the appropriate lists
127 #
128 cpu_sequencers.append(cpu_seq)
129 l1_cntrl_nodes.append(l1_cntrl)
130
131 l2_index_start = block_size_bits + l2_bits
132
133 for i in xrange(options.num_l2caches):
134 #
135 # First create the Ruby objects associated with this cpu
136 #
137 l2_cache = L2Cache(size = options.l2_size,
138 assoc = options.l2_assoc,
139 start_index_bit = l2_index_start)
140
141 l2_cntrl = L2Cache_Controller(version = i,
142 L2cache = l2_cache,
143 N_tokens = n_tokens,
144 transitions_per_cycle = options.ports,
145 ruby_system = ruby_system)
146
147 exec("ruby_system.l2_cntrl%d = l2_cntrl" % i)
148 l2_cntrl_nodes.append(l2_cntrl)
149
150 phys_mem_size = sum(map(lambda r: r.size(), system.mem_ranges))
151 assert(phys_mem_size % options.num_dirs == 0)
152 mem_module_size = phys_mem_size / options.num_dirs
153
154 # Run each of the ruby memory controllers at a ratio of the frequency of
155 # the ruby system
156 # clk_divider value is a fix to pass regression.
157 ruby_system.memctrl_clk_domain = DerivedClockDomain(
158 clk_domain=ruby_system.clk_domain,
159 clk_divider=3)
160
161 for i in xrange(options.num_dirs):
162 #
163 # Create the Ruby objects associated with the directory controller
164 #
165
166 mem_cntrl = RubyMemoryControl(
167 clk_domain = ruby_system.memctrl_clk_domain,
168 version = i,
169 ruby_system = ruby_system)
170
171 dir_size = MemorySize('0B')
172 dir_size.value = mem_module_size
173
174 dir_cntrl = Directory_Controller(version = i,
175 directory = \
176 RubyDirectoryMemory(version = i,
177 use_map = options.use_map,
178 size = dir_size),
179 memBuffer = mem_cntrl,
180 l2_select_num_bits = l2_bits,
181 transitions_per_cycle = options.ports,
182 ruby_system = ruby_system)
183
184 exec("ruby_system.dir_cntrl%d = dir_cntrl" % i)
185 dir_cntrl_nodes.append(dir_cntrl)
186
187 for i, dma_port in enumerate(dma_ports):
188 #
189 # Create the Ruby objects associated with the dma controller
190 #
191 dma_seq = DMASequencer(version = i,
192 ruby_system = ruby_system)
193
194 dma_cntrl = DMA_Controller(version = i,
195 dma_sequencer = dma_seq,
196 transitions_per_cycle = options.ports,
197 ruby_system = ruby_system)
198
199 exec("ruby_system.dma_cntrl%d = dma_cntrl" % i)
200 exec("ruby_system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
201 dma_cntrl_nodes.append(dma_cntrl)
202
203 all_cntrls = l1_cntrl_nodes + \
204 l2_cntrl_nodes + \
205 dir_cntrl_nodes + \
206 dma_cntrl_nodes
207
208 topology = create_topology(all_cntrls, options)
209
210 return (cpu_sequencers, dir_cntrl_nodes, topology)
|