MOESI_hammer.py revision 10116:d61a59beb670
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 48# 49# Probe filter is a cache, latency is not used 50# 51class ProbeFilter(RubyCache): 52 latency = 1 53 54def define_options(parser): 55 parser.add_option("--allow-atomic-migration", action="store_true", 56 help="allow migratory sharing for atomic only accessed blocks") 57 parser.add_option("--pf-on", action="store_true", 58 help="Hammer: enable Probe Filter") 59 parser.add_option("--dir-on", action="store_true", 60 help="Hammer: enable Full-bit Directory") 61 62def create_system(options, system, dma_ports, ruby_system): 63 64 if buildEnv['PROTOCOL'] != 'MOESI_hammer': 65 panic("This script requires the MOESI_hammer protocol to be built.") 66 67 cpu_sequencers = [] 68 69 # 70 # The ruby network creation expects the list of nodes in the system to be 71 # consistent with the NetDest list. Therefore the l1 controller nodes must be 72 # listed before the directory nodes and directory nodes before dma nodes, etc. 73 # 74 l1_cntrl_nodes = [] 75 dir_cntrl_nodes = [] 76 dma_cntrl_nodes = [] 77 78 # 79 # Must create the individual controllers before the network to ensure the 80 # controller constructors are called before the network constructor 81 # 82 block_size_bits = int(math.log(options.cacheline_size, 2)) 83 84 for i in xrange(options.num_cpus): 85 # 86 # First create the Ruby objects associated with this cpu 87 # 88 l1i_cache = L1Cache(size = options.l1i_size, 89 assoc = options.l1i_assoc, 90 start_index_bit = block_size_bits, 91 is_icache = True) 92 l1d_cache = L1Cache(size = options.l1d_size, 93 assoc = options.l1d_assoc, 94 start_index_bit = block_size_bits) 95 l2_cache = L2Cache(size = options.l2_size, 96 assoc = options.l2_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 L2cache = l2_cache, 103 no_mig_atomic = not \ 104 options.allow_atomic_migration, 105 send_evictions = ( 106 options.cpu_type == "detailed"), 107 transitions_per_cycle = options.ports, 108 ruby_system = ruby_system) 109 110 cpu_seq = RubySequencer(version = i, 111 icache = l1i_cache, 112 dcache = l1d_cache, 113 ruby_system = ruby_system) 114 115 l1_cntrl.sequencer = cpu_seq 116 if options.recycle_latency: 117 l1_cntrl.recycle_latency = options.recycle_latency 118 119 exec("ruby_system.l1_cntrl%d = l1_cntrl" % i) 120 # 121 # Add controllers and sequencers to the appropriate lists 122 # 123 cpu_sequencers.append(cpu_seq) 124 l1_cntrl_nodes.append(l1_cntrl) 125 126 phys_mem_size = sum(map(lambda r: r.size(), system.mem_ranges)) 127 assert(phys_mem_size % options.num_dirs == 0) 128 mem_module_size = phys_mem_size / options.num_dirs 129 130 # 131 # determine size and index bits for probe filter 132 # By default, the probe filter size is configured to be twice the 133 # size of the L2 cache. 134 # 135 pf_size = MemorySize(options.l2_size) 136 pf_size.value = pf_size.value * 2 137 dir_bits = int(math.log(options.num_dirs, 2)) 138 pf_bits = int(math.log(pf_size.value, 2)) 139 if options.numa_high_bit: 140 if options.pf_on or options.dir_on: 141 # if numa high bit explicitly set, make sure it does not overlap 142 # with the probe filter index 143 assert(options.numa_high_bit - dir_bits > pf_bits) 144 145 # set the probe filter start bit to just above the block offset 146 pf_start_bit = block_size_bits 147 else: 148 if dir_bits > 0: 149 pf_start_bit = dir_bits + block_size_bits - 1 150 else: 151 pf_start_bit = block_size_bits 152 153 # Run each of the ruby memory controllers at a ratio of the frequency of 154 # the ruby system 155 # clk_divider value is a fix to pass regression. 156 ruby_system.memctrl_clk_domain = DerivedClockDomain( 157 clk_domain=ruby_system.clk_domain, 158 clk_divider=3) 159 160 for i in xrange(options.num_dirs): 161 # 162 # Create the Ruby objects associated with the directory controller 163 # 164 165 mem_cntrl = RubyMemoryControl( 166 clk_domain = ruby_system.memctrl_clk_domain, 167 version = i, 168 ruby_system = ruby_system) 169 170 dir_size = MemorySize('0B') 171 dir_size.value = mem_module_size 172 173 pf = ProbeFilter(size = pf_size, assoc = 4, 174 start_index_bit = pf_start_bit) 175 176 dir_cntrl = Directory_Controller(version = i, 177 directory = \ 178 RubyDirectoryMemory( \ 179 version = i, 180 size = dir_size, 181 use_map = options.use_map, 182 map_levels = \ 183 options.map_levels, 184 numa_high_bit = \ 185 options.numa_high_bit), 186 probeFilter = pf, 187 memBuffer = mem_cntrl, 188 probe_filter_enabled = options.pf_on, 189 full_bit_dir_enabled = options.dir_on, 190 transitions_per_cycle = options.ports, 191 ruby_system = ruby_system) 192 193 if options.recycle_latency: 194 dir_cntrl.recycle_latency = options.recycle_latency 195 196 exec("ruby_system.dir_cntrl%d = dir_cntrl" % i) 197 dir_cntrl_nodes.append(dir_cntrl) 198 199 for i, dma_port in enumerate(dma_ports): 200 # 201 # Create the Ruby objects associated with the dma controller 202 # 203 dma_seq = DMASequencer(version = i, 204 ruby_system = ruby_system) 205 206 dma_cntrl = DMA_Controller(version = i, 207 dma_sequencer = dma_seq, 208 transitions_per_cycle = options.ports, 209 ruby_system = ruby_system) 210 211 exec("ruby_system.dma_cntrl%d = dma_cntrl" % i) 212 exec("ruby_system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i) 213 dma_cntrl_nodes.append(dma_cntrl) 214 215 if options.recycle_latency: 216 dma_cntrl.recycle_latency = options.recycle_latency 217 218 all_cntrls = l1_cntrl_nodes + dir_cntrl_nodes + dma_cntrl_nodes 219 topology = create_topology(all_cntrls, options) 220 221 return (cpu_sequencers, dir_cntrl_nodes, topology) 222