MOESI_hammer.py revision 12976:125099a94768
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, create_directories 35from Ruby import send_evicts 36 37# 38# Declare caches used by the protocol 39# 40class L1Cache(RubyCache): pass 41class L2Cache(RubyCache): pass 42# 43# Probe filter is a cache 44# 45class ProbeFilter(RubyCache): pass 46 47def define_options(parser): 48 parser.add_option("--allow-atomic-migration", action="store_true", 49 help="allow migratory sharing for atomic only accessed blocks") 50 parser.add_option("--pf-on", action="store_true", 51 help="Hammer: enable Probe Filter") 52 parser.add_option("--dir-on", action="store_true", 53 help="Hammer: enable Full-bit Directory") 54 55def create_system(options, full_system, system, dma_ports, bootmem, 56 ruby_system): 57 58 if buildEnv['PROTOCOL'] != 'MOESI_hammer': 59 panic("This script requires the MOESI_hammer protocol to be built.") 60 61 cpu_sequencers = [] 62 63 # 64 # The ruby network creation expects the list of nodes in the system to be 65 # consistent with the NetDest list. Therefore the l1 controller nodes must be 66 # listed before the directory nodes and directory nodes before dma nodes, etc. 67 # 68 l1_cntrl_nodes = [] 69 dma_cntrl_nodes = [] 70 71 # 72 # Must create the individual controllers before the network to ensure the 73 # controller constructors are called before the network constructor 74 # 75 block_size_bits = int(math.log(options.cacheline_size, 2)) 76 77 for i in xrange(options.num_cpus): 78 # 79 # First create the Ruby objects associated with this cpu 80 # 81 l1i_cache = L1Cache(size = options.l1i_size, 82 assoc = options.l1i_assoc, 83 start_index_bit = block_size_bits, 84 is_icache = True) 85 l1d_cache = L1Cache(size = options.l1d_size, 86 assoc = options.l1d_assoc, 87 start_index_bit = block_size_bits) 88 l2_cache = L2Cache(size = options.l2_size, 89 assoc = options.l2_assoc, 90 start_index_bit = block_size_bits) 91 92 # the ruby random tester reuses num_cpus to specify the 93 # number of cpu ports connected to the tester object, which 94 # is stored in system.cpu. because there is only ever one 95 # tester object, num_cpus is not necessarily equal to the 96 # size of system.cpu; therefore if len(system.cpu) == 1 97 # we use system.cpu[0] to set the clk_domain, thereby ensuring 98 # we don't index off the end of the cpu list. 99 if len(system.cpu) == 1: 100 clk_domain = system.cpu[0].clk_domain 101 else: 102 clk_domain = system.cpu[i].clk_domain 103 104 l1_cntrl = L1Cache_Controller(version=i, L1Icache=l1i_cache, 105 L1Dcache=l1d_cache, L2cache=l2_cache, 106 no_mig_atomic=not \ 107 options.allow_atomic_migration, 108 send_evictions=send_evicts(options), 109 transitions_per_cycle=options.ports, 110 clk_domain=clk_domain, 111 ruby_system=ruby_system) 112 113 cpu_seq = RubySequencer(version=i, icache=l1i_cache, 114 dcache=l1d_cache,clk_domain=clk_domain, 115 ruby_system=ruby_system) 116 117 l1_cntrl.sequencer = cpu_seq 118 if options.recycle_latency: 119 l1_cntrl.recycle_latency = options.recycle_latency 120 121 exec("ruby_system.l1_cntrl%d = l1_cntrl" % i) 122 123 # Add controllers and sequencers to the appropriate lists 124 cpu_sequencers.append(cpu_seq) 125 l1_cntrl_nodes.append(l1_cntrl) 126 127 # Connect the L1 controller and the network 128 # Connect the buffers from the controller to network 129 l1_cntrl.requestFromCache = MessageBuffer() 130 l1_cntrl.requestFromCache.master = ruby_system.network.slave 131 l1_cntrl.responseFromCache = MessageBuffer() 132 l1_cntrl.responseFromCache.master = ruby_system.network.slave 133 l1_cntrl.unblockFromCache = MessageBuffer() 134 l1_cntrl.unblockFromCache.master = ruby_system.network.slave 135 136 l1_cntrl.triggerQueue = MessageBuffer() 137 138 # Connect the buffers from the network to the controller 139 l1_cntrl.mandatoryQueue = MessageBuffer() 140 l1_cntrl.forwardToCache = MessageBuffer() 141 l1_cntrl.forwardToCache.slave = ruby_system.network.master 142 l1_cntrl.responseToCache = MessageBuffer() 143 l1_cntrl.responseToCache.slave = ruby_system.network.master 144 145 146 # 147 # determine size and index bits for probe filter 148 # By default, the probe filter size is configured to be twice the 149 # size of the L2 cache. 150 # 151 pf_size = MemorySize(options.l2_size) 152 pf_size.value = pf_size.value * 2 153 dir_bits = int(math.log(options.num_dirs, 2)) 154 pf_bits = int(math.log(pf_size.value, 2)) 155 if options.numa_high_bit: 156 if options.pf_on or options.dir_on: 157 # if numa high bit explicitly set, make sure it does not overlap 158 # with the probe filter index 159 assert(options.numa_high_bit - dir_bits > pf_bits) 160 161 # set the probe filter start bit to just above the block offset 162 pf_start_bit = block_size_bits 163 else: 164 if dir_bits > 0: 165 pf_start_bit = dir_bits + block_size_bits - 1 166 else: 167 pf_start_bit = block_size_bits 168 169 # Run each of the ruby memory controllers at a ratio of the frequency of 170 # the ruby system 171 # clk_divider value is a fix to pass regression. 172 ruby_system.memctrl_clk_domain = DerivedClockDomain( 173 clk_domain=ruby_system.clk_domain, 174 clk_divider=3) 175 176 mem_dir_cntrl_nodes, rom_dir_cntrl_node = create_directories( 177 options, bootmem, ruby_system, system) 178 dir_cntrl_nodes = mem_dir_cntrl_nodes[:] 179 if rom_dir_cntrl_node is not None: 180 dir_cntrl_nodes.append(rom_dir_cntrl_node) 181 for dir_cntrl in dir_cntrl_nodes: 182 pf = ProbeFilter(size = pf_size, assoc = 4, 183 start_index_bit = pf_start_bit) 184 185 dir_cntrl.probeFilter = pf 186 dir_cntrl.probe_filter_enabled = options.pf_on 187 dir_cntrl.full_bit_dir_enabled = options.dir_on 188 189 if options.recycle_latency: 190 dir_cntrl.recycle_latency = options.recycle_latency 191 192 # Connect the directory controller to the network 193 dir_cntrl.forwardFromDir = MessageBuffer() 194 dir_cntrl.forwardFromDir.master = ruby_system.network.slave 195 dir_cntrl.responseFromDir = MessageBuffer() 196 dir_cntrl.responseFromDir.master = ruby_system.network.slave 197 dir_cntrl.dmaResponseFromDir = MessageBuffer(ordered = True) 198 dir_cntrl.dmaResponseFromDir.master = ruby_system.network.slave 199 200 dir_cntrl.triggerQueue = MessageBuffer(ordered = True) 201 202 dir_cntrl.unblockToDir = MessageBuffer() 203 dir_cntrl.unblockToDir.slave = ruby_system.network.master 204 dir_cntrl.responseToDir = MessageBuffer() 205 dir_cntrl.responseToDir.slave = ruby_system.network.master 206 dir_cntrl.requestToDir = MessageBuffer() 207 dir_cntrl.requestToDir.slave = ruby_system.network.master 208 dir_cntrl.dmaRequestToDir = MessageBuffer(ordered = True) 209 dir_cntrl.dmaRequestToDir.slave = ruby_system.network.master 210 dir_cntrl.responseFromMemory = MessageBuffer() 211 212 213 for i, dma_port in enumerate(dma_ports): 214 # 215 # Create the Ruby objects associated with the dma controller 216 # 217 dma_seq = DMASequencer(version = i, 218 ruby_system = ruby_system, 219 slave = dma_port) 220 221 dma_cntrl = DMA_Controller(version = i, 222 dma_sequencer = dma_seq, 223 transitions_per_cycle = options.ports, 224 ruby_system = ruby_system) 225 226 exec("ruby_system.dma_cntrl%d = dma_cntrl" % i) 227 dma_cntrl_nodes.append(dma_cntrl) 228 229 if options.recycle_latency: 230 dma_cntrl.recycle_latency = options.recycle_latency 231 232 # Connect the dma controller to the network 233 dma_cntrl.responseFromDir = MessageBuffer(ordered = True) 234 dma_cntrl.responseFromDir.slave = ruby_system.network.master 235 dma_cntrl.requestToDir = MessageBuffer() 236 dma_cntrl.requestToDir.master = ruby_system.network.slave 237 dma_cntrl.mandatoryQueue = MessageBuffer() 238 239 all_cntrls = l1_cntrl_nodes + dir_cntrl_nodes + dma_cntrl_nodes 240 241 # Create the io controller and the sequencer 242 if full_system: 243 io_seq = DMASequencer(version=len(dma_ports), ruby_system=ruby_system) 244 ruby_system._io_port = io_seq 245 io_controller = DMA_Controller(version = len(dma_ports), 246 dma_sequencer = io_seq, 247 ruby_system = ruby_system) 248 ruby_system.io_controller = io_controller 249 250 # Connect the dma controller to the network 251 io_controller.responseFromDir = MessageBuffer(ordered = True) 252 io_controller.responseFromDir.slave = ruby_system.network.master 253 io_controller.requestToDir = MessageBuffer() 254 io_controller.requestToDir.master = ruby_system.network.slave 255 io_controller.mandatoryQueue = MessageBuffer() 256 257 all_cntrls = all_cntrls + [io_controller] 258 259 ruby_system.network.number_of_virtual_networks = 6 260 topology = create_topology(all_cntrls, options) 261 return (cpu_sequencers, mem_dir_cntrl_nodes, topology) 262