ruby: add a three level MESI protocol.
[gem5.git] / configs / ruby / MESI_Three_Level.py
1 # Copyright (c) 2006-2007 The Regents of The University of Michigan
2 # Copyright (c) 2009 Advanced Micro Devices, Inc.
3 # Copyright (c) 2013 Mark D. Hill and David A. Wood
4 # All rights reserved.
5 #
6 # Redistribution and use in source and binary forms, with or without
7 # modification, are permitted provided that the following conditions are
8 # met: redistributions of source code must retain the above copyright
9 # notice, this list of conditions and the following disclaimer;
10 # redistributions in binary form must reproduce the above copyright
11 # notice, this list of conditions and the following disclaimer in the
12 # documentation and/or other materials provided with the distribution;
13 # neither the name of the copyright holders nor the names of its
14 # contributors may be used to endorse or promote products derived from
15 # this software without specific prior written permission.
16 #
17 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20 # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
21 # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
23 # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 #
29 # Authors: Brad Beckmann
30 # Nilay Vaish
31
32 import math
33 import m5
34 from m5.objects import *
35 from m5.defines import buildEnv
36 from Ruby import create_topology
37
38 #
39 # Note: the L1 Cache latency is only used by the sequencer on fast path hits
40 #
41 class L0Cache(RubyCache):
42 latency = 1
43
44 class L1Cache(RubyCache):
45 latency = 5
46
47 #
48 # Note: the L2 Cache latency is not currently used
49 #
50 class L2Cache(RubyCache):
51 latency = 15
52
53 def define_options(parser):
54 parser.add_option("--num-clusters", type="int", default=1,
55 help="number of clusters in a design in which there are shared\
56 caches private to clusters")
57 return
58
59 def create_system(options, system, piobus, dma_ports, ruby_system):
60
61 if buildEnv['PROTOCOL'] != 'MESI_Three_Level':
62 fatal("This script requires the MESI_Three_Level protocol to be built.")
63
64 cpu_sequencers = []
65
66 #
67 # The ruby network creation expects the list of nodes in the system to be
68 # consistent with the NetDest list. Therefore the l1 controller nodes must be
69 # listed before the directory nodes and directory nodes before dma nodes, etc.
70 #
71 l0_cntrl_nodes = []
72 l1_cntrl_nodes = []
73 l2_cntrl_nodes = []
74 dir_cntrl_nodes = []
75 dma_cntrl_nodes = []
76
77 assert (options.num_cpus % options.num_clusters == 0)
78 num_cpus_per_cluster = options.num_cpus / options.num_clusters
79
80 assert (options.num_l2caches % options.num_clusters == 0)
81 num_l2caches_per_cluster = options.num_l2caches / options.num_clusters
82
83 l2_bits = int(math.log(num_l2caches_per_cluster, 2))
84 block_size_bits = int(math.log(options.cacheline_size, 2))
85 l2_index_start = block_size_bits + l2_bits
86
87 #
88 # Must create the individual controllers before the network to ensure the
89 # controller constructors are called before the network constructor
90 #
91 for i in xrange(options.num_clusters):
92 for j in xrange(num_cpus_per_cluster):
93 #
94 # First create the Ruby objects associated with this cpu
95 #
96 l0i_cache = L0Cache(size = '4096B', assoc = 1, is_icache = True,
97 start_index_bit = block_size_bits, replacement_policy="LRU")
98
99 l0d_cache = L0Cache(size = '4096B', assoc = 1, is_icache = False,
100 start_index_bit = block_size_bits, replacement_policy="LRU")
101
102 l0_cntrl = L0Cache_Controller(version = i*num_cpus_per_cluster + j,
103 Icache = l0i_cache, Dcache = l0d_cache,
104 send_evictions = (options.cpu_type == "detailed"),
105 ruby_system = ruby_system)
106
107 cpu_seq = RubySequencer(version = i, icache = l0i_cache,
108 dcache = l0d_cache, ruby_system = ruby_system)
109
110 l0_cntrl.sequencer = cpu_seq
111
112 l1_cache = L1Cache(size = options.l1d_size, assoc = options.l1d_assoc,
113 start_index_bit = block_size_bits, is_icache = False)
114
115 l1_cntrl = L1Cache_Controller(version = i*num_cpus_per_cluster+j,
116 cache = l1_cache, l2_select_num_bits = l2_bits,
117 cluster_id = i, ruby_system = ruby_system)
118
119 if piobus != None:
120 cpu_seq.pio_port = piobus.slave
121
122 exec("ruby_system.l0_cntrl%d = l0_cntrl" % (
123 i*num_cpus_per_cluster+j))
124 exec("ruby_system.l1_cntrl%d = l1_cntrl" % (
125 i*num_cpus_per_cluster+j))
126
127 #
128 # Add controllers and sequencers to the appropriate lists
129 #
130 cpu_sequencers.append(cpu_seq)
131 l0_cntrl_nodes.append(l0_cntrl)
132 l1_cntrl_nodes.append(l1_cntrl)
133 l0_cntrl.peer = l1_cntrl
134
135 for j in xrange(num_l2caches_per_cluster):
136 l2_cache = L2Cache(size = options.l2_size,
137 assoc = options.l2_assoc,
138 start_index_bit = l2_index_start)
139
140 l2_cntrl = L2Cache_Controller(
141 version = i * num_l2caches_per_cluster + j,
142 L2cache = l2_cache, cluster_id = i,
143 transitions_per_cycle=options.ports,
144 ruby_system = ruby_system)
145
146 exec("ruby_system.l2_cntrl%d = l2_cntrl" % (
147 i * num_l2caches_per_cluster + j))
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 size = dir_size,
178 use_map =
179 options.use_map),
180 memBuffer = mem_cntrl,
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 = l0_cntrl_nodes + \
204 l1_cntrl_nodes + \
205 l2_cntrl_nodes + \
206 dir_cntrl_nodes + \
207 dma_cntrl_nodes
208
209 topology = create_topology(all_cntrls, options)
210 return (cpu_sequencers, dir_cntrl_nodes, topology)