MESI_Three_Level.py revision 10529:05b5a6cf3521
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 32import math 33import m5 34from m5.objects import * 35from m5.defines import buildEnv 36from Ruby import create_topology 37from Ruby import send_evicts 38 39# 40# Note: the L1 Cache latency is only used by the sequencer on fast path hits 41# 42class L0Cache(RubyCache): 43 latency = 1 44 45class L1Cache(RubyCache): 46 latency = 5 47 48# 49# Note: the L2 Cache latency is not currently used 50# 51class L2Cache(RubyCache): 52 latency = 15 53 54def define_options(parser): 55 parser.add_option("--num-clusters", type="int", default=1, 56 help="number of clusters in a design in which there are shared\ 57 caches private to clusters") 58 return 59 60def create_system(options, full_system, system, dma_ports, ruby_system): 61 62 if buildEnv['PROTOCOL'] != 'MESI_Three_Level': 63 fatal("This script requires the MESI_Three_Level protocol to be built.") 64 65 cpu_sequencers = [] 66 67 # 68 # The ruby network creation expects the list of nodes in the system to be 69 # consistent with the NetDest list. Therefore the l1 controller nodes must be 70 # listed before the directory nodes and directory nodes before dma nodes, etc. 71 # 72 l0_cntrl_nodes = [] 73 l1_cntrl_nodes = [] 74 l2_cntrl_nodes = [] 75 dir_cntrl_nodes = [] 76 dma_cntrl_nodes = [] 77 78 assert (options.num_cpus % options.num_clusters == 0) 79 num_cpus_per_cluster = options.num_cpus / options.num_clusters 80 81 assert (options.num_l2caches % options.num_clusters == 0) 82 num_l2caches_per_cluster = options.num_l2caches / options.num_clusters 83 84 l2_bits = int(math.log(num_l2caches_per_cluster, 2)) 85 block_size_bits = int(math.log(options.cacheline_size, 2)) 86 l2_index_start = block_size_bits + l2_bits 87 88 # 89 # Must create the individual controllers before the network to ensure the 90 # controller constructors are called before the network constructor 91 # 92 for i in xrange(options.num_clusters): 93 for j in xrange(num_cpus_per_cluster): 94 # 95 # First create the Ruby objects associated with this cpu 96 # 97 l0i_cache = L0Cache(size = '4096B', assoc = 1, is_icache = True, 98 start_index_bit = block_size_bits, replacement_policy="LRU") 99 100 l0d_cache = L0Cache(size = '4096B', assoc = 1, is_icache = False, 101 start_index_bit = block_size_bits, replacement_policy="LRU") 102 103 l0_cntrl = L0Cache_Controller(version = i*num_cpus_per_cluster + j, 104 Icache = l0i_cache, Dcache = l0d_cache, 105 send_evictions = send_evicts(options), 106 clk_domain=system.cpu[i].clk_domain, 107 ruby_system = ruby_system) 108 109 cpu_seq = RubySequencer(version = i, icache = l0i_cache, 110 clk_domain=system.cpu[i].clk_domain, 111 dcache = l0d_cache, ruby_system = ruby_system) 112 113 l0_cntrl.sequencer = cpu_seq 114 115 l1_cache = L1Cache(size = options.l1d_size, assoc = options.l1d_assoc, 116 start_index_bit = block_size_bits, is_icache = False) 117 118 l1_cntrl = L1Cache_Controller(version = i*num_cpus_per_cluster+j, 119 cache = l1_cache, l2_select_num_bits = l2_bits, 120 cluster_id = i, ruby_system = ruby_system) 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) 133 134 # Connect the L0 and L1 controllers 135 l0_cntrl.bufferToL1 = l1_cntrl.bufferFromL0 136 l0_cntrl.bufferFromL1 = l1_cntrl.bufferToL0 137 138 # Connect the L1 controllers and the network 139 l1_cntrl.requestToL2 = ruby_system.network.slave 140 l1_cntrl.responseToL2 = ruby_system.network.slave 141 l1_cntrl.unblockToL2 = ruby_system.network.slave 142 143 l1_cntrl.requestFromL2 = ruby_system.network.master 144 l1_cntrl.responseFromL2 = ruby_system.network.master 145 146 147 for j in xrange(num_l2caches_per_cluster): 148 l2_cache = L2Cache(size = options.l2_size, 149 assoc = options.l2_assoc, 150 start_index_bit = l2_index_start) 151 152 l2_cntrl = L2Cache_Controller( 153 version = i * num_l2caches_per_cluster + j, 154 L2cache = l2_cache, cluster_id = i, 155 transitions_per_cycle=options.ports, 156 ruby_system = ruby_system) 157 158 exec("ruby_system.l2_cntrl%d = l2_cntrl" % ( 159 i * num_l2caches_per_cluster + j)) 160 l2_cntrl_nodes.append(l2_cntrl) 161 162 # Connect the L2 controllers and the network 163 l2_cntrl.DirRequestFromL2Cache = ruby_system.network.slave 164 l2_cntrl.L1RequestFromL2Cache = ruby_system.network.slave 165 l2_cntrl.responseFromL2Cache = ruby_system.network.slave 166 167 l2_cntrl.unblockToL2Cache = ruby_system.network.master 168 l2_cntrl.L1RequestToL2Cache = ruby_system.network.master 169 l2_cntrl.responseToL2Cache = ruby_system.network.master 170 171 phys_mem_size = sum(map(lambda r: r.size(), system.mem_ranges)) 172 assert(phys_mem_size % options.num_dirs == 0) 173 mem_module_size = phys_mem_size / options.num_dirs 174 175 # Run each of the ruby memory controllers at a ratio of the frequency of 176 # the ruby system 177 # clk_divider value is a fix to pass regression. 178 ruby_system.memctrl_clk_domain = DerivedClockDomain( 179 clk_domain=ruby_system.clk_domain, 180 clk_divider=3) 181 182 for i in xrange(options.num_dirs): 183 # 184 # Create the Ruby objects associated with the directory controller 185 # 186 dir_size = MemorySize('0B') 187 dir_size.value = mem_module_size 188 189 dir_cntrl = Directory_Controller(version = i, 190 directory = RubyDirectoryMemory( 191 version = i, size = dir_size), 192 transitions_per_cycle = options.ports, 193 ruby_system = ruby_system) 194 195 exec("ruby_system.dir_cntrl%d = dir_cntrl" % i) 196 dir_cntrl_nodes.append(dir_cntrl) 197 198 # Connect the directory controllers and the network 199 dir_cntrl.requestToDir = ruby_system.network.master 200 dir_cntrl.responseToDir = ruby_system.network.master 201 dir_cntrl.responseFromDir = ruby_system.network.slave 202 203 for i, dma_port in enumerate(dma_ports): 204 # 205 # Create the Ruby objects associated with the dma controller 206 # 207 dma_seq = DMASequencer(version = i, 208 ruby_system = ruby_system) 209 210 dma_cntrl = DMA_Controller(version = i, 211 dma_sequencer = dma_seq, 212 transitions_per_cycle = options.ports, 213 ruby_system = ruby_system) 214 215 exec("ruby_system.dma_cntrl%d = dma_cntrl" % i) 216 exec("ruby_system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i) 217 dma_cntrl_nodes.append(dma_cntrl) 218 219 all_cntrls = l0_cntrl_nodes + \ 220 l1_cntrl_nodes + \ 221 l2_cntrl_nodes + \ 222 dir_cntrl_nodes + \ 223 dma_cntrl_nodes 224 225 # Create the io controller and the sequencer 226 if full_system: 227 io_seq = DMASequencer(version=len(dma_ports), ruby_system=ruby_system) 228 ruby_system._io_port = io_seq 229 io_controller = DMA_Controller(version = len(dma_ports), 230 dma_sequencer = io_seq, 231 ruby_system = ruby_system) 232 ruby_system.io_controller = io_controller 233 234 # Connect the dma controller to the network 235 io_controller.responseFromDir = ruby_system.network.master 236 io_controller.requestToDir = ruby_system.network.slave 237 238 all_cntrls = all_cntrls + [io_controller] 239 240 topology = create_topology(all_cntrls, options) 241 return (cpu_sequencers, dir_cntrl_nodes, topology) 242