Adding gem5 source to svn.

git-svn-id: https://www4.informatik.uni-erlangen.de/i4svn/danceos/trunk/devel/fail@1819 8c4709b5-6ec9-48aa-a5cd-a96041d1645a
This commit is contained in:
friemel
2012-10-24 19:18:57 +00:00
parent f7ff71bd46
commit b41eec3f65
3222 changed files with 658579 additions and 1 deletions

View File

@ -0,0 +1,188 @@
# Copyright (c) 2006-2007 The Regents of The University of Michigan
# Copyright (c) 2009 Advanced Micro Devices, Inc.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Brad Beckmann
import math
import m5
from m5.objects import *
from m5.defines import buildEnv
#
# Note: the L1 Cache latency is only used by the sequencer on fast path hits
#
class L1Cache(RubyCache):
latency = 3
#
# Note: the L2 Cache latency is not currently used
#
class L2Cache(RubyCache):
latency = 15
def define_options(parser):
return
def create_system(options, system, piobus, dma_ports, ruby_system):
if buildEnv['PROTOCOL'] != 'MESI_CMP_directory':
panic("This script requires the MESI_CMP_directory protocol to be built.")
cpu_sequencers = []
#
# The ruby network creation expects the list of nodes in the system to be
# consistent with the NetDest list. Therefore the l1 controller nodes must be
# listed before the directory nodes and directory nodes before dma nodes, etc.
#
l1_cntrl_nodes = []
l2_cntrl_nodes = []
dir_cntrl_nodes = []
dma_cntrl_nodes = []
#
# Must create the individual controllers before the network to ensure the
# controller constructors are called before the network constructor
#
l2_bits = int(math.log(options.num_l2caches, 2))
block_size_bits = int(math.log(options.cacheline_size, 2))
cntrl_count = 0
for i in xrange(options.num_cpus):
#
# First create the Ruby objects associated with this cpu
#
l1i_cache = L1Cache(size = options.l1i_size,
assoc = options.l1i_assoc,
start_index_bit = block_size_bits)
l1d_cache = L1Cache(size = options.l1d_size,
assoc = options.l1d_assoc,
start_index_bit = block_size_bits)
l1_cntrl = L1Cache_Controller(version = i,
cntrl_id = cntrl_count,
L1IcacheMemory = l1i_cache,
L1DcacheMemory = l1d_cache,
l2_select_num_bits = l2_bits,
send_evictions = (
options.cpu_type == "detailed"),
ruby_system = ruby_system)
cpu_seq = RubySequencer(version = i,
icache = l1i_cache,
dcache = l1d_cache,
ruby_system = ruby_system)
l1_cntrl.sequencer = cpu_seq
if piobus != None:
cpu_seq.pio_port = piobus.slave
exec("system.l1_cntrl%d = l1_cntrl" % i)
#
# Add controllers and sequencers to the appropriate lists
#
cpu_sequencers.append(cpu_seq)
l1_cntrl_nodes.append(l1_cntrl)
cntrl_count += 1
l2_index_start = block_size_bits + l2_bits
for i in xrange(options.num_l2caches):
#
# First create the Ruby objects associated with this cpu
#
l2_cache = L2Cache(size = options.l2_size,
assoc = options.l2_assoc,
start_index_bit = l2_index_start)
l2_cntrl = L2Cache_Controller(version = i,
cntrl_id = cntrl_count,
L2cacheMemory = l2_cache,
ruby_system = ruby_system)
exec("system.l2_cntrl%d = l2_cntrl" % i)
l2_cntrl_nodes.append(l2_cntrl)
cntrl_count += 1
phys_mem_size = 0
for mem in system.memories.unproxy(system):
phys_mem_size += long(mem.range.second) - long(mem.range.first) + 1
mem_module_size = phys_mem_size / options.num_dirs
for i in xrange(options.num_dirs):
#
# Create the Ruby objects associated with the directory controller
#
mem_cntrl = RubyMemoryControl(version = i)
dir_size = MemorySize('0B')
dir_size.value = mem_module_size
dir_cntrl = Directory_Controller(version = i,
cntrl_id = cntrl_count,
directory = \
RubyDirectoryMemory(version = i,
size = dir_size,
use_map =
options.use_map),
memBuffer = mem_cntrl,
ruby_system = ruby_system)
exec("system.dir_cntrl%d = dir_cntrl" % i)
dir_cntrl_nodes.append(dir_cntrl)
cntrl_count += 1
for i, dma_port in enumerate(dma_ports):
#
# Create the Ruby objects associated with the dma controller
#
dma_seq = DMASequencer(version = i,
ruby_system = ruby_system)
dma_cntrl = DMA_Controller(version = i,
cntrl_id = cntrl_count,
dma_sequencer = dma_seq,
ruby_system = ruby_system)
exec("system.dma_cntrl%d = dma_cntrl" % i)
exec("system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
dma_cntrl_nodes.append(dma_cntrl)
cntrl_count += 1
all_cntrls = l1_cntrl_nodes + \
l2_cntrl_nodes + \
dir_cntrl_nodes + \
dma_cntrl_nodes
return (cpu_sequencers, dir_cntrl_nodes, all_cntrls)

View File

@ -0,0 +1,158 @@
# Copyright (c) 2006-2007 The Regents of The University of Michigan
# Copyright (c) 2009 Advanced Micro Devices, Inc.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Brad Beckmann
import math
import m5
from m5.objects import *
from m5.defines import buildEnv
#
# Note: the cache latency is only used by the sequencer on fast path hits
#
class Cache(RubyCache):
latency = 3
def define_options(parser):
return
def create_system(options, system, piobus, dma_ports, ruby_system):
if buildEnv['PROTOCOL'] != 'MI_example':
panic("This script requires the MI_example protocol to be built.")
cpu_sequencers = []
#
# The ruby network creation expects the list of nodes in the system to be
# consistent with the NetDest list. Therefore the l1 controller nodes must be
# listed before the directory nodes and directory nodes before dma nodes, etc.
#
l1_cntrl_nodes = []
dir_cntrl_nodes = []
dma_cntrl_nodes = []
#
# Must create the individual controllers before the network to ensure the
# controller constructors are called before the network constructor
#
block_size_bits = int(math.log(options.cacheline_size, 2))
cntrl_count = 0
for i in xrange(options.num_cpus):
#
# First create the Ruby objects associated with this cpu
# Only one cache exists for this protocol, so by default use the L1D
# config parameters.
#
cache = Cache(size = options.l1d_size,
assoc = options.l1d_assoc,
start_index_bit = block_size_bits)
#
# Only one unified L1 cache exists. Can cache instructions and data.
#
l1_cntrl = L1Cache_Controller(version = i,
cntrl_id = cntrl_count,
cacheMemory = cache,
send_evictions = (
options.cpu_type == "detailed"),
ruby_system = ruby_system)
cpu_seq = RubySequencer(version = i,
icache = cache,
dcache = cache,
ruby_system = ruby_system)
l1_cntrl.sequencer = cpu_seq
if piobus != None:
cpu_seq.pio_port = piobus.slave
exec("system.l1_cntrl%d = l1_cntrl" % i)
#
# Add controllers and sequencers to the appropriate lists
#
cpu_sequencers.append(cpu_seq)
l1_cntrl_nodes.append(l1_cntrl)
cntrl_count += 1
phys_mem_size = 0
for mem in system.memories.unproxy(system):
phys_mem_size += long(mem.range.second) - long(mem.range.first) + 1
mem_module_size = phys_mem_size / options.num_dirs
for i in xrange(options.num_dirs):
#
# Create the Ruby objects associated with the directory controller
#
mem_cntrl = RubyMemoryControl(version = i)
dir_size = MemorySize('0B')
dir_size.value = mem_module_size
dir_cntrl = Directory_Controller(version = i,
cntrl_id = cntrl_count,
directory = \
RubyDirectoryMemory( \
version = i,
size = dir_size,
use_map = options.use_map,
map_levels = \
options.map_levels),
memBuffer = mem_cntrl,
ruby_system = ruby_system)
exec("system.dir_cntrl%d = dir_cntrl" % i)
dir_cntrl_nodes.append(dir_cntrl)
cntrl_count += 1
for i, dma_port in enumerate(dma_ports):
#
# Create the Ruby objects associated with the dma controller
#
dma_seq = DMASequencer(version = i,
ruby_system = ruby_system)
dma_cntrl = DMA_Controller(version = i,
cntrl_id = cntrl_count,
dma_sequencer = dma_seq,
ruby_system = ruby_system)
exec("system.dma_cntrl%d = dma_cntrl" % i)
exec("system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
dma_cntrl_nodes.append(dma_cntrl)
cntrl_count += 1
all_cntrls = l1_cntrl_nodes + dir_cntrl_nodes + dma_cntrl_nodes
return (cpu_sequencers, dir_cntrl_nodes, all_cntrls)

View File

@ -0,0 +1,185 @@
# Copyright (c) 2006-2007 The Regents of The University of Michigan
# Copyright (c) 2009 Advanced Micro Devices, Inc.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Brad Beckmann
import math
import m5
from m5.objects import *
from m5.defines import buildEnv
#
# Note: the L1 Cache latency is only used by the sequencer on fast path hits
#
class L1Cache(RubyCache):
latency = 3
#
# Note: the L2 Cache latency is not currently used
#
class L2Cache(RubyCache):
latency = 15
def define_options(parser):
return
def create_system(options, system, piobus, dma_ports, ruby_system):
if buildEnv['PROTOCOL'] != 'MOESI_CMP_directory':
panic("This script requires the MOESI_CMP_directory protocol to be built.")
cpu_sequencers = []
#
# The ruby network creation expects the list of nodes in the system to be
# consistent with the NetDest list. Therefore the l1 controller nodes must be
# listed before the directory nodes and directory nodes before dma nodes, etc.
#
l1_cntrl_nodes = []
l2_cntrl_nodes = []
dir_cntrl_nodes = []
dma_cntrl_nodes = []
#
# Must create the individual controllers before the network to ensure the
# controller constructors are called before the network constructor
#
l2_bits = int(math.log(options.num_l2caches, 2))
block_size_bits = int(math.log(options.cacheline_size, 2))
cntrl_count = 0
for i in xrange(options.num_cpus):
#
# First create the Ruby objects associated with this cpu
#
l1i_cache = L1Cache(size = options.l1i_size,
assoc = options.l1i_assoc,
start_index_bit = block_size_bits)
l1d_cache = L1Cache(size = options.l1d_size,
assoc = options.l1d_assoc,
start_index_bit = block_size_bits)
l1_cntrl = L1Cache_Controller(version = i,
cntrl_id = cntrl_count,
L1IcacheMemory = l1i_cache,
L1DcacheMemory = l1d_cache,
l2_select_num_bits = l2_bits,
send_evictions = (
options.cpu_type == "detailed"),
ruby_system = ruby_system)
cpu_seq = RubySequencer(version = i,
icache = l1i_cache,
dcache = l1d_cache,
ruby_system = ruby_system)
l1_cntrl.sequencer = cpu_seq
if piobus != None:
cpu_seq.pio_port = piobus.slave
exec("system.l1_cntrl%d = l1_cntrl" % i)
#
# Add controllers and sequencers to the appropriate lists
#
cpu_sequencers.append(cpu_seq)
l1_cntrl_nodes.append(l1_cntrl)
cntrl_count += 1
l2_index_start = block_size_bits + l2_bits
for i in xrange(options.num_l2caches):
#
# First create the Ruby objects associated with this cpu
#
l2_cache = L2Cache(size = options.l2_size,
assoc = options.l2_assoc,
start_index_bit = l2_index_start)
l2_cntrl = L2Cache_Controller(version = i,
cntrl_id = cntrl_count,
L2cacheMemory = l2_cache,
ruby_system = ruby_system)
exec("system.l2_cntrl%d = l2_cntrl" % i)
l2_cntrl_nodes.append(l2_cntrl)
cntrl_count += 1
phys_mem_size = 0
for mem in system.memories.unproxy(system):
phys_mem_size += long(mem.range.second) - long(mem.range.first) + 1
mem_module_size = phys_mem_size / options.num_dirs
for i in xrange(options.num_dirs):
#
# Create the Ruby objects associated with the directory controller
#
mem_cntrl = RubyMemoryControl(version = i)
dir_size = MemorySize('0B')
dir_size.value = mem_module_size
dir_cntrl = Directory_Controller(version = i,
cntrl_id = cntrl_count,
directory = \
RubyDirectoryMemory(version = i,
size = dir_size),
memBuffer = mem_cntrl,
ruby_system = ruby_system)
exec("system.dir_cntrl%d = dir_cntrl" % i)
dir_cntrl_nodes.append(dir_cntrl)
cntrl_count += 1
for i, dma_port in enumerate(dma_ports):
#
# Create the Ruby objects associated with the dma controller
#
dma_seq = DMASequencer(version = i,
ruby_system = ruby_system)
dma_cntrl = DMA_Controller(version = i,
cntrl_id = cntrl_count,
dma_sequencer = dma_seq,
ruby_system = ruby_system)
exec("system.dma_cntrl%d = dma_cntrl" % i)
exec("system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
dma_cntrl_nodes.append(dma_cntrl)
cntrl_count += 1
all_cntrls = l1_cntrl_nodes + \
l2_cntrl_nodes + \
dir_cntrl_nodes + \
dma_cntrl_nodes
return (cpu_sequencers, dir_cntrl_nodes, all_cntrls)

View File

@ -0,0 +1,209 @@
# Copyright (c) 2006-2007 The Regents of The University of Michigan
# Copyright (c) 2009 Advanced Micro Devices, Inc.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Brad Beckmann
import math
import m5
from m5.objects import *
from m5.defines import buildEnv
#
# Note: the L1 Cache latency is only used by the sequencer on fast path hits
#
class L1Cache(RubyCache):
latency = 2
#
# Note: the L2 Cache latency is not currently used
#
class L2Cache(RubyCache):
latency = 10
def define_options(parser):
parser.add_option("--l1-retries", type="int", default=1,
help="Token_CMP: # of l1 retries before going persistent")
parser.add_option("--timeout-latency", type="int", default=300,
help="Token_CMP: cycles until issuing again");
parser.add_option("--disable-dyn-timeouts", action="store_true",
help="Token_CMP: disable dyanimc timeouts, use fixed latency instead")
parser.add_option("--allow-atomic-migration", action="store_true",
help="allow migratory sharing for atomic only accessed blocks")
def create_system(options, system, piobus, dma_ports, ruby_system):
if buildEnv['PROTOCOL'] != 'MOESI_CMP_token':
panic("This script requires the MOESI_CMP_token protocol to be built.")
#
# number of tokens that the owner passes to requests so that shared blocks can
# respond to read requests
#
n_tokens = options.num_cpus + 1
cpu_sequencers = []
#
# The ruby network creation expects the list of nodes in the system to be
# consistent with the NetDest list. Therefore the l1 controller nodes must be
# listed before the directory nodes and directory nodes before dma nodes, etc.
#
l1_cntrl_nodes = []
l2_cntrl_nodes = []
dir_cntrl_nodes = []
dma_cntrl_nodes = []
#
# Must create the individual controllers before the network to ensure the
# controller constructors are called before the network constructor
#
l2_bits = int(math.log(options.num_l2caches, 2))
block_size_bits = int(math.log(options.cacheline_size, 2))
cntrl_count = 0
for i in xrange(options.num_cpus):
#
# First create the Ruby objects associated with this cpu
#
l1i_cache = L1Cache(size = options.l1i_size,
assoc = options.l1i_assoc,
start_index_bit = block_size_bits)
l1d_cache = L1Cache(size = options.l1d_size,
assoc = options.l1d_assoc,
start_index_bit = block_size_bits)
l1_cntrl = L1Cache_Controller(version = i,
cntrl_id = cntrl_count,
L1IcacheMemory = l1i_cache,
L1DcacheMemory = l1d_cache,
l2_select_num_bits = l2_bits,
N_tokens = n_tokens,
retry_threshold = \
options.l1_retries,
fixed_timeout_latency = \
options.timeout_latency,
dynamic_timeout_enabled = \
not options.disable_dyn_timeouts,
no_mig_atomic = not \
options.allow_atomic_migration,
send_evictions = (
options.cpu_type == "detailed"),
ruby_system = ruby_system)
cpu_seq = RubySequencer(version = i,
icache = l1i_cache,
dcache = l1d_cache,
ruby_system = ruby_system)
l1_cntrl.sequencer = cpu_seq
if piobus != None:
cpu_seq.pio_port = piobus.slave
exec("system.l1_cntrl%d = l1_cntrl" % i)
#
# Add controllers and sequencers to the appropriate lists
#
cpu_sequencers.append(cpu_seq)
l1_cntrl_nodes.append(l1_cntrl)
cntrl_count += 1
l2_index_start = block_size_bits + l2_bits
for i in xrange(options.num_l2caches):
#
# First create the Ruby objects associated with this cpu
#
l2_cache = L2Cache(size = options.l2_size,
assoc = options.l2_assoc,
start_index_bit = l2_index_start)
l2_cntrl = L2Cache_Controller(version = i,
cntrl_id = cntrl_count,
L2cacheMemory = l2_cache,
N_tokens = n_tokens,
ruby_system = ruby_system)
exec("system.l2_cntrl%d = l2_cntrl" % i)
l2_cntrl_nodes.append(l2_cntrl)
cntrl_count += 1
phys_mem_size = 0
for mem in system.memories.unproxy(system):
phys_mem_size += long(mem.range.second) - long(mem.range.first) + 1
mem_module_size = phys_mem_size / options.num_dirs
for i in xrange(options.num_dirs):
#
# Create the Ruby objects associated with the directory controller
#
mem_cntrl = RubyMemoryControl(version = i)
dir_size = MemorySize('0B')
dir_size.value = mem_module_size
dir_cntrl = Directory_Controller(version = i,
cntrl_id = cntrl_count,
directory = \
RubyDirectoryMemory(version = i,
size = dir_size),
memBuffer = mem_cntrl,
l2_select_num_bits = l2_bits,
ruby_system = ruby_system)
exec("system.dir_cntrl%d = dir_cntrl" % i)
dir_cntrl_nodes.append(dir_cntrl)
cntrl_count += 1
for i, dma_port in enumerate(dma_ports):
#
# Create the Ruby objects associated with the dma controller
#
dma_seq = DMASequencer(version = i,
ruby_system = ruby_system)
dma_cntrl = DMA_Controller(version = i,
cntrl_id = cntrl_count,
dma_sequencer = dma_seq,
ruby_system = ruby_system)
exec("system.dma_cntrl%d = dma_cntrl" % i)
exec("system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
dma_cntrl_nodes.append(dma_cntrl)
cntrl_count += 1
all_cntrls = l1_cntrl_nodes + \
l2_cntrl_nodes + \
dir_cntrl_nodes + \
dma_cntrl_nodes
return (cpu_sequencers, dir_cntrl_nodes, all_cntrls)

View File

@ -0,0 +1,222 @@
# Copyright (c) 2006-2007 The Regents of The University of Michigan
# Copyright (c) 2009 Advanced Micro Devices, Inc.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Brad Beckmann
import math
import m5
from m5.objects import *
from m5.defines import buildEnv
#
# Note: the L1 Cache latency is only used by the sequencer on fast path hits
#
class L1Cache(RubyCache):
latency = 2
#
# Note: the L2 Cache latency is not currently used
#
class L2Cache(RubyCache):
latency = 10
#
# Probe filter is a cache, latency is not used
#
class ProbeFilter(RubyCache):
latency = 1
def define_options(parser):
parser.add_option("--allow-atomic-migration", action="store_true",
help="allow migratory sharing for atomic only accessed blocks")
parser.add_option("--pf-on", action="store_true",
help="Hammer: enable Probe Filter")
parser.add_option("--dir-on", action="store_true",
help="Hammer: enable Full-bit Directory")
def create_system(options, system, piobus, dma_ports, ruby_system):
if buildEnv['PROTOCOL'] != 'MOESI_hammer':
panic("This script requires the MOESI_hammer protocol to be built.")
cpu_sequencers = []
#
# The ruby network creation expects the list of nodes in the system to be
# consistent with the NetDest list. Therefore the l1 controller nodes must be
# listed before the directory nodes and directory nodes before dma nodes, etc.
#
l1_cntrl_nodes = []
dir_cntrl_nodes = []
dma_cntrl_nodes = []
#
# Must create the individual controllers before the network to ensure the
# controller constructors are called before the network constructor
#
block_size_bits = int(math.log(options.cacheline_size, 2))
cntrl_count = 0
for i in xrange(options.num_cpus):
#
# First create the Ruby objects associated with this cpu
#
l1i_cache = L1Cache(size = options.l1i_size,
assoc = options.l1i_assoc,
start_index_bit = block_size_bits,
is_icache = True)
l1d_cache = L1Cache(size = options.l1d_size,
assoc = options.l1d_assoc,
start_index_bit = block_size_bits)
l2_cache = L2Cache(size = options.l2_size,
assoc = options.l2_assoc,
start_index_bit = block_size_bits)
l1_cntrl = L1Cache_Controller(version = i,
cntrl_id = cntrl_count,
L1IcacheMemory = l1i_cache,
L1DcacheMemory = l1d_cache,
L2cacheMemory = l2_cache,
no_mig_atomic = not \
options.allow_atomic_migration,
send_evictions = (
options.cpu_type == "detailed"),
ruby_system = ruby_system)
cpu_seq = RubySequencer(version = i,
icache = l1i_cache,
dcache = l1d_cache,
ruby_system = ruby_system)
l1_cntrl.sequencer = cpu_seq
if piobus != None:
cpu_seq.pio_port = piobus.slave
if options.recycle_latency:
l1_cntrl.recycle_latency = options.recycle_latency
exec("system.l1_cntrl%d = l1_cntrl" % i)
#
# Add controllers and sequencers to the appropriate lists
#
cpu_sequencers.append(cpu_seq)
l1_cntrl_nodes.append(l1_cntrl)
cntrl_count += 1
phys_mem_size = 0
for mem in system.memories.unproxy(system):
phys_mem_size += long(mem.range.second) - long(mem.range.first) + 1
mem_module_size = phys_mem_size / options.num_dirs
#
# determine size and index bits for probe filter
# By default, the probe filter size is configured to be twice the
# size of the L2 cache.
#
pf_size = MemorySize(options.l2_size)
pf_size.value = pf_size.value * 2
dir_bits = int(math.log(options.num_dirs, 2))
pf_bits = int(math.log(pf_size.value, 2))
if options.numa_high_bit:
if options.numa_high_bit > 0:
# if numa high bit explicitly set, make sure it does not overlap
# with the probe filter index
assert(options.numa_high_bit - dir_bits > pf_bits)
# set the probe filter start bit to just above the block offset
pf_start_bit = 6
else:
if dir_bits > 0:
pf_start_bit = dir_bits + 5
else:
pf_start_bit = 6
for i in xrange(options.num_dirs):
#
# Create the Ruby objects associated with the directory controller
#
mem_cntrl = RubyMemoryControl(version = i)
dir_size = MemorySize('0B')
dir_size.value = mem_module_size
pf = ProbeFilter(size = pf_size, assoc = 4,
start_index_bit = pf_start_bit)
dir_cntrl = Directory_Controller(version = i,
cntrl_id = cntrl_count,
directory = \
RubyDirectoryMemory( \
version = i,
size = dir_size,
use_map = options.use_map,
map_levels = \
options.map_levels,
numa_high_bit = \
options.numa_high_bit),
probeFilter = pf,
memBuffer = mem_cntrl,
probe_filter_enabled = options.pf_on,
full_bit_dir_enabled = options.dir_on,
ruby_system = ruby_system)
if options.recycle_latency:
dir_cntrl.recycle_latency = options.recycle_latency
exec("system.dir_cntrl%d = dir_cntrl" % i)
dir_cntrl_nodes.append(dir_cntrl)
cntrl_count += 1
for i, dma_port in enumerate(dma_ports):
#
# Create the Ruby objects associated with the dma controller
#
dma_seq = DMASequencer(version = i,
ruby_system = ruby_system)
dma_cntrl = DMA_Controller(version = i,
cntrl_id = cntrl_count,
dma_sequencer = dma_seq,
ruby_system = ruby_system)
exec("system.dma_cntrl%d = dma_cntrl" % i)
exec("system.dma_cntrl%d.dma_sequencer.slave = dma_port" % i)
dma_cntrl_nodes.append(dma_cntrl)
if options.recycle_latency:
dma_cntrl.recycle_latency = options.recycle_latency
cntrl_count += 1
all_cntrls = l1_cntrl_nodes + dir_cntrl_nodes + dma_cntrl_nodes
return (cpu_sequencers, dir_cntrl_nodes, all_cntrls)

View File

@ -0,0 +1,138 @@
# Copyright (c) 2006-2007 The Regents of The University of Michigan
# Copyright (c) 2009 Advanced Micro Devices, Inc.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Brad Beckmann
import m5
from m5.objects import *
from m5.defines import buildEnv
from m5.util import addToPath
#
# Note: the cache latency is only used by the sequencer on fast path hits
#
class Cache(RubyCache):
latency = 3
def define_options(parser):
return
def create_system(options, system, piobus, dma_ports, ruby_system):
if buildEnv['PROTOCOL'] != 'Network_test':
panic("This script requires the Network_test protocol to be built.")
cpu_sequencers = []
#
# The Garnet tester protocol does not support fs nor dma
#
assert(piobus == None)
assert(dma_ports == [])
#
# The ruby network creation expects the list of nodes in the system to be
# consistent with the NetDest list. Therefore the l1 controller nodes must be
# listed before the directory nodes and directory nodes before dma nodes, etc.
#
l1_cntrl_nodes = []
dir_cntrl_nodes = []
#
# Must create the individual controllers before the network to ensure the
# controller constructors are called before the network constructor
#
cntrl_count = 0
for i in xrange(options.num_cpus):
#
# First create the Ruby objects associated with this cpu
# Only one cache exists for this protocol, so by default use the L1D
# config parameters.
#
cache = Cache(size = options.l1d_size,
assoc = options.l1d_assoc)
#
# Only one unified L1 cache exists. Can cache instructions and data.
#
l1_cntrl = L1Cache_Controller(version = i,
cntrl_id = cntrl_count,
cacheMemory = cache,
ruby_system = ruby_system)
cpu_seq = RubySequencer(icache = cache,
dcache = cache,
using_network_tester = True,
ruby_system = ruby_system)
l1_cntrl.sequencer = cpu_seq
if piobus != None:
cpu_seq.pio_port = piobus.slave
exec("system.l1_cntrl%d = l1_cntrl" % i)
#
# Add controllers and sequencers to the appropriate lists
#
cpu_sequencers.append(cpu_seq)
l1_cntrl_nodes.append(l1_cntrl)
cntrl_count += 1
phys_mem_size = 0
for mem in system.memories.unproxy(system):
phys_mem_size += long(mem.range.second) - long(mem.range.first) + 1
mem_module_size = phys_mem_size / options.num_dirs
for i in xrange(options.num_dirs):
#
# Create the Ruby objects associated with the directory controller
#
mem_cntrl = RubyMemoryControl(version = i)
dir_size = MemorySize('0B')
dir_size.value = mem_module_size
dir_cntrl = Directory_Controller(version = i,
cntrl_id = cntrl_count,
directory = \
RubyDirectoryMemory(version = i,
size = dir_size),
memBuffer = mem_cntrl,
ruby_system = ruby_system)
exec("system.dir_cntrl%d = dir_cntrl" % i)
dir_cntrl_nodes.append(dir_cntrl)
cntrl_count += 1
all_cntrls = l1_cntrl_nodes + dir_cntrl_nodes
return (cpu_sequencers, dir_cntrl_nodes, all_cntrls)

View File

@ -0,0 +1,189 @@
# Copyright (c) 2012 ARM Limited
# All rights reserved.
#
# The license below extends only to copyright in the software and shall
# not be construed as granting a license to any other intellectual
# property including but not limited to intellectual property relating
# to a hardware implementation of the functionality of the software
# licensed hereunder. You may use the software subject to the license
# terms below provided that you ensure that this notice is replicated
# unmodified and in its entirety in all distributions of the software,
# modified or unmodified, in source code or in binary form.
#
# Copyright (c) 2006-2007 The Regents of The University of Michigan
# Copyright (c) 2009 Advanced Micro Devices, Inc.
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Brad Beckmann
import math
import m5
from m5.objects import *
from m5.defines import buildEnv
def define_options(parser):
# By default, ruby uses the simple timing cpu
parser.set_defaults(cpu_type="timing")
# ruby network options
parser.add_option("--topology", type="string", default="Crossbar",
help="check src/mem/ruby/network/topologies for complete set")
parser.add_option("--mesh-rows", type="int", default=1,
help="the number of rows in the mesh topology")
parser.add_option("--garnet-network", type="string", default=None,
help="'fixed'|'flexible'")
parser.add_option("--network-fault-model", action="store_true", default=False,
help="enable network fault model: see src/mem/ruby/network/fault_model/")
# ruby mapping options
parser.add_option("--numa-high-bit", type="int", default=0,
help="high order address bit to use for numa mapping. " \
"0 = highest bit, not specified = lowest bit")
# ruby sparse memory options
parser.add_option("--use-map", action="store_true", default=False)
parser.add_option("--map-levels", type="int", default=4)
parser.add_option("--recycle-latency", type="int", default=10,
help="Recycle latency for ruby controller input buffers")
parser.add_option("--random_seed", type="int", default=1234,
help="Used for seeding the random number generator")
parser.add_option("--ruby_stats", type="string", default="ruby.stats")
protocol = buildEnv['PROTOCOL']
exec "import %s" % protocol
eval("%s.define_options(parser)" % protocol)
def create_system(options, system, piobus = None, dma_ports = []):
system.ruby = RubySystem(clock = options.clock,
stats_filename = options.ruby_stats,
no_mem_vec = options.use_map)
ruby = system.ruby
protocol = buildEnv['PROTOCOL']
exec "import %s" % protocol
try:
(cpu_sequencers, dir_cntrls, all_cntrls) = \
eval("%s.create_system(options, system, piobus, dma_ports, ruby)"
% protocol)
except:
print "Error: could not create sytem for ruby protocol %s" % protocol
raise
# Create a port proxy for connecting the system port. This is
# independent of the protocol and kept in the protocol-agnostic
# part (i.e. here).
sys_port_proxy = RubyPortProxy(ruby_system = ruby)
# Give the system port proxy a SimObject parent without creating a
# full-fledged controller
system.sys_port_proxy = sys_port_proxy
# Connect the system port for loading of binaries etc
system.system_port = system.sys_port_proxy.slave
#
# Set the network classes based on the command line options
#
if options.garnet_network == "fixed":
class NetworkClass(GarnetNetwork_d): pass
class IntLinkClass(GarnetIntLink_d): pass
class ExtLinkClass(GarnetExtLink_d): pass
class RouterClass(GarnetRouter_d): pass
elif options.garnet_network == "flexible":
class NetworkClass(GarnetNetwork): pass
class IntLinkClass(GarnetIntLink): pass
class ExtLinkClass(GarnetExtLink): pass
class RouterClass(GarnetRouter): pass
else:
class NetworkClass(SimpleNetwork): pass
class IntLinkClass(SimpleIntLink): pass
class ExtLinkClass(SimpleExtLink): pass
class RouterClass(BasicRouter): pass
#
# Important: the topology must be created before the network and after the
# controllers.
#
exec "import %s" % options.topology
try:
net_topology = eval("%s.makeTopology(all_cntrls, options, \
IntLinkClass, ExtLinkClass, \
RouterClass)" \
% options.topology)
except:
print "Error: could not create topology %s" % options.topology
raise
if options.network_fault_model:
assert(options.garnet_network == "fixed")
fault_model = FaultModel()
network = NetworkClass(ruby_system = ruby, topology = net_topology,\
enable_fault_model=True, fault_model = fault_model)
else:
network = NetworkClass(ruby_system = ruby, topology = net_topology)
#
# Loop through the directory controlers.
# Determine the total memory size of the ruby system and verify it is equal
# to physmem. However, if Ruby memory is using sparse memory in SE
# mode, then the system should not back-up the memory state with
# the Memory Vector and thus the memory size bytes should stay at 0.
# Also set the numa bits to the appropriate values.
#
total_mem_size = MemorySize('0B')
dir_bits = int(math.log(options.num_dirs, 2))
if options.numa_high_bit:
numa_bit = options.numa_high_bit
else:
# if not specified, use the lowest bits above the block offest
if dir_bits > 0:
# add 5 because bits 0-5 are the block offset
numa_bit = dir_bits + 5
else:
numa_bit = 6
for dir_cntrl in dir_cntrls:
total_mem_size.value += dir_cntrl.directory.size.value
dir_cntrl.directory.numa_high_bit = numa_bit
phys_mem_size = 0
for mem in system.memories.unproxy(system):
phys_mem_size += long(mem.range.second) - long(mem.range.first) + 1
assert(total_mem_size.value == phys_mem_size)
ruby_profiler = RubyProfiler(ruby_system = ruby,
num_of_sequencers = len(cpu_sequencers))
ruby.network = network
ruby.profiler = ruby_profiler
ruby.mem_size = total_mem_size
ruby._cpu_ruby_ports = cpu_sequencers
ruby.random_seed = options.random_seed