derek/gem5
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

configs: changed CHI --noc-config format

Browse Source 
Changed format from yaml to plain python. The new py configuration
file, when provided, must specialize the CHI node types defined in
configs/ruby/CHI_config.py (moved from configs/ruby/CHI.py). This
is required in order to setup the node->router bindings when the
CustomMesh topology is used.

See configs/example/noc_config/2x4.py (replaces
configs/example/noc_config/2x4.yaml) for an example.

--noc-config was also renamed to --chi-config, since the CHI node types
can be fully specialized in the configuration file.

Change-Id: Ic0c5407dba3d2483d5c30634c115b5410a5228fd
Signed-off-by: Tiago Mück <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/43123
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
This commit is contained in:
Tiago Mück
2021-03-16 17:00:33 -05:00
parent 377a96644e
commit 21dfaa49ca
4 changed files with 760 additions and 694 deletions
Download Patch File Download Diff File Expand all files Collapse all files

55
configs/example/noc_config/2x4.yaml → configs/example/noc_config/2x4.py
View File

@@ -33,38 +33,45 @@
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
# 2x4 mesh definition from ruby import CHI_config
# CustomMesh parameters for a 2x4 mesh. Routers will have the following layout:
# #
# 0 --- 1 --- 2 --- 3 # 0 --- 1 --- 2 --- 3
# | | | | # | | | |
# 4 --- 5 --- 6 --- 7 # 4 --- 5 --- 6 --- 7
# #
mesh: # Default parameter are configs/ruby/CHI_config.py
num_rows : 2 #
num_cols : 4 class NoC_Params(CHI_config.NoC_Params):
router_latency : 1 num_rows = 2
link_latency : 1 num_cols = 4
# Bindings for each CHI node type. # Specialization of nodes to define bindings for each CHI node type
# needed by CustomMesh.
# The default types are defined in CHI_Node and their derivatives in
# configs/ruby/CHI_config.py
CHI_RNF: class CHI_RNF(CHI_config.CHI_RNF):
# Uncomment to map num_nodes_per_router RNFs in each provided router, class NoC_Params(CHI_config.CHI_RNF.NoC_Params):
# assuming num. created CHI_RNFs == len(router_list)*num_nodes_per_router router_list = [1, 2, 5, 6]
# num_nodes_per_router: 1
router_list: [1, 2, 5, 6]
CHI_HNF: class CHI_HNF(CHI_config.CHI_HNF):
# num_nodes_per_router: 1 class NoC_Params(CHI_config.CHI_HNF.NoC_Params):
router_list: [1, 2, 5, 6] router_list = [1, 2, 5, 6]
CHI_SNF_MainMem: class CHI_SNF_MainMem(CHI_config.CHI_SNF_MainMem):
# num_nodes_per_router: 1 class NoC_Params(CHI_config.CHI_SNF_MainMem.NoC_Params):
router_list: [0, 4] router_list = [0, 4]
# Applies to CHI_SNF_BootMem and possibly other non-main memories class CHI_SNF_BootMem(CHI_config.CHI_SNF_BootMem):
CHI_SNF_IO: class NoC_Params(CHI_config.CHI_SNF_BootMem.NoC_Params):
router_list: [3] router_list = [3]
# Applies to CHI_RNI_DMA and CHI_RNI_IO class CHI_RNI_DMA(CHI_config.CHI_RNI_DMA):
CHI_RNI_IO: class NoC_Params(CHI_config.CHI_RNI_DMA.NoC_Params):
router_list: [7] router_list = [7]
class CHI_RNI_IO(CHI_config.CHI_RNI_IO):
class NoC_Params(CHI_config.CHI_RNI_IO.NoC_Params):
router_list = [7]

670
configs/ruby/CHI.py
View File

@@ -33,617 +33,25 @@
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
import math
import yaml
import m5 import m5
from m5.objects import * from m5.objects import *
from m5.defines import buildEnv from m5.defines import buildEnv
from .Ruby import create_topology, setup_memory_controllers from .Ruby import create_topology
def define_options(parser): def define_options(parser):
parser.add_option("--noc-config", action="store", type="string", parser.add_option("--chi-config", action="store", type="string",
default=None, default=None,
help="YAML NoC config. parameters and bindings. " help="NoC config. parameters and bindings. "
"required for CustomMesh topology") "Required for CustomMesh topology")
class Versions: def read_config_file(file):
''' ''' Read file as a module and return it '''
Helper class to obtain unique ids for a given controller class. import types
These are passed as the 'version' parameter when creating the controller. import importlib.machinery
''' loader = importlib.machinery.SourceFileLoader('chi_configs', file)
_seqs = 0 chi_configs = types.ModuleType(loader.name)
@classmethod loader.exec_module(chi_configs)
def getSeqId(cls): return chi_configs
val = cls._seqs
cls._seqs += 1
return val
_version = {}
@classmethod
def getVersion(cls, tp):
if tp not in cls._version:
cls._version[tp] = 0
val = cls._version[tp]
cls._version[tp] = val + 1
return val
class CHI_Node(SubSystem):
'''
Base class with common functions for setting up Cache or Memory
controllers that are part of a CHI RNF, RNFI, HNF, or SNF nodes.
Notice getNetworkSideControllers and getAllControllers must be implemented
in the derived classes.
'''
def __init__(self, ruby_system):
super(CHI_Node, self).__init__()
self._ruby_system = ruby_system
self._network = ruby_system.network
def getNetworkSideControllers(self):
'''
Returns all ruby controllers that need to be connected to the
network
'''
raise NotImplementedError()
def getAllControllers(self):
'''
Returns all ruby controllers associated with this node
'''
raise NotImplementedError()
def setDownstream(self, cntrls):
'''
Sets cntrls as the downstream list of all controllers in this node
'''
for c in self.getNetworkSideControllers():
c.downstream_destinations = cntrls
def connectController(self, cntrl):
'''
Creates and configures the messages buffers for the CHI input/output
ports that connect to the network
'''
cntrl.reqOut = MessageBuffer()
cntrl.rspOut = MessageBuffer()
cntrl.snpOut = MessageBuffer()
cntrl.datOut = MessageBuffer()
cntrl.reqIn = MessageBuffer()
cntrl.rspIn = MessageBuffer()
cntrl.snpIn = MessageBuffer()
cntrl.datIn = MessageBuffer()
# All CHI ports are always connected to the network.
# Controllers that are not part of the getNetworkSideControllers list
# still communicate using internal routers, thus we need to wire-up the
# ports
cntrl.reqOut.out_port = self._network.in_port
cntrl.rspOut.out_port = self._network.in_port
cntrl.snpOut.out_port = self._network.in_port
cntrl.datOut.out_port = self._network.in_port
cntrl.reqIn.in_port = self._network.out_port
cntrl.rspIn.in_port = self._network.out_port
cntrl.snpIn.in_port = self._network.out_port
cntrl.datIn.in_port = self._network.out_port
class TriggerMessageBuffer(MessageBuffer):
'''
MessageBuffer for triggering internal controller events.
These buffers should not be affected by the Ruby tester randomization
and allow poping messages enqueued in the same cycle.
'''
randomization = 'disabled'
allow_zero_latency = True
class OrderedTriggerMessageBuffer(TriggerMessageBuffer):
ordered = True
class CHI_Cache_Controller(Cache_Controller):
'''
Default parameters for a Cache controller
The Cache_Controller can also be used as a DMA requester or as
a pure directory if all cache allocation policies are disabled.
'''
def __init__(self, ruby_system):
super(CHI_Cache_Controller, self).__init__(
version = Versions.getVersion(Cache_Controller),
ruby_system = ruby_system,
mandatoryQueue = MessageBuffer(),
prefetchQueue = MessageBuffer(),
triggerQueue = TriggerMessageBuffer(),
retryTriggerQueue = OrderedTriggerMessageBuffer(),
replTriggerQueue = OrderedTriggerMessageBuffer(),
reqRdy = TriggerMessageBuffer(),
snpRdy = TriggerMessageBuffer())
# Set somewhat large number since we really a lot on internal
# triggers. To limit the controller performance, tweak other
# params such as: input port buffer size, cache banks, and output
# port latency
self.transitions_per_cycle = 128
# This should be set to true in the data cache controller to enable
# timeouts on unique lines when a store conditional fails
self.sc_lock_enabled = False
class CHI_L1Controller(CHI_Cache_Controller):
'''
Default parameters for a L1 Cache controller
'''
def __init__(self, ruby_system, sequencer, cache, prefetcher):
super(CHI_L1Controller, self).__init__(ruby_system)
self.sequencer = sequencer
self.cache = cache
self.use_prefetcher = False
self.send_evictions = True
self.is_HN = False
self.enable_DMT = False
self.enable_DCT = False
# Strict inclusive MOESI
self.allow_SD = True
self.alloc_on_seq_acc = True
self.alloc_on_seq_line_write = False
self.alloc_on_readshared = True
self.alloc_on_readunique = True
self.alloc_on_readonce = True
self.alloc_on_writeback = True
self.dealloc_on_unique = False
self.dealloc_on_shared = False
self.dealloc_backinv_unique = True
self.dealloc_backinv_shared = True
# Some reasonable default TBE params
self.number_of_TBEs = 16
self.number_of_repl_TBEs = 16
self.number_of_snoop_TBEs = 4
self.unify_repl_TBEs = False
class CHI_L2Controller(CHI_Cache_Controller):
'''
Default parameters for a L2 Cache controller
'''
def __init__(self, ruby_system, cache, prefetcher):
super(CHI_L2Controller, self).__init__(ruby_system)
self.sequencer = NULL
self.cache = cache
self.use_prefetcher = False
self.allow_SD = True
self.is_HN = False
self.enable_DMT = False
self.enable_DCT = False
self.send_evictions = False
# Strict inclusive MOESI
self.alloc_on_seq_acc = False
self.alloc_on_seq_line_write = False
self.alloc_on_readshared = True
self.alloc_on_readunique = True
self.alloc_on_readonce = True
self.alloc_on_writeback = True
self.dealloc_on_unique = False
self.dealloc_on_shared = False
self.dealloc_backinv_unique = True
self.dealloc_backinv_shared = True
# Some reasonable default TBE params
self.number_of_TBEs = 32
self.number_of_repl_TBEs = 32
self.number_of_snoop_TBEs = 16
self.unify_repl_TBEs = False
class CHI_HNFController(CHI_Cache_Controller):
'''
Default parameters for a coherent home node (HNF) cache controller
'''
def __init__(self, ruby_system, cache, prefetcher, addr_ranges):
super(CHI_HNFController, self).__init__(ruby_system)
self.sequencer = NULL
self.cache = cache
self.use_prefetcher = False
self.addr_ranges = addr_ranges
self.allow_SD = True
self.is_HN = True
self.enable_DMT = True
self.enable_DCT = True
self.send_evictions = False
# MOESI / Mostly inclusive for shared / Exclusive for unique
self.alloc_on_seq_acc = False
self.alloc_on_seq_line_write = False
self.alloc_on_readshared = True
self.alloc_on_readunique = False
self.alloc_on_readonce = True
self.alloc_on_writeback = True
self.dealloc_on_unique = True
self.dealloc_on_shared = False
self.dealloc_backinv_unique = False
self.dealloc_backinv_shared = False
# Some reasonable default TBE params
self.number_of_TBEs = 32
self.number_of_repl_TBEs = 32
self.number_of_snoop_TBEs = 1 # should not receive any snoop
self.unify_repl_TBEs = False
class CHI_DMAController(CHI_Cache_Controller):
'''
Default parameters for a DMA controller
'''
def __init__(self, ruby_system, sequencer):
super(CHI_DMAController, self).__init__(ruby_system)
self.sequencer = sequencer
class DummyCache(RubyCache):
dataAccessLatency = 0
tagAccessLatency = 1
size = "128"
assoc = 1
self.use_prefetcher = False
self.cache = DummyCache()
self.sequencer.dcache = NULL
# All allocations are false
# Deallocations are true (don't really matter)
self.allow_SD = False
self.is_HN = False
self.enable_DMT = False
self.enable_DCT = False
self.alloc_on_seq_acc = False
self.alloc_on_seq_line_write = False
self.alloc_on_readshared = False
self.alloc_on_readunique = False
self.alloc_on_readonce = False
self.alloc_on_writeback = False
self.dealloc_on_unique = False
self.dealloc_on_shared = False
self.dealloc_backinv_unique = False
self.dealloc_backinv_shared = False
self.send_evictions = False
self.number_of_TBEs = 16
self.number_of_repl_TBEs = 1
self.number_of_snoop_TBEs = 1 # should not receive any snoop
self.unify_repl_TBEs = False
class CPUSequencerWrapper:
'''
Other generic configuration scripts assume a matching number of sequencers
and cpus. This wraps the instruction and data sequencer so they are
compatible with the other scripts. This assumes all scripts are using
connectCpuPorts/connectIOPorts to bind ports
'''
def __init__(self, iseq, dseq):
# use this style due to __setattr__ override below
self.__dict__['inst_seq'] = iseq
self.__dict__['data_seq'] = dseq
self.__dict__['support_data_reqs'] = True
self.__dict__['support_inst_reqs'] = True
# Compatibility with certain scripts that wire up ports
# without connectCpuPorts
self.__dict__['slave'] = dseq.in_ports
self.__dict__['in_ports'] = dseq.in_ports
def connectCpuPorts(self, cpu):
assert(isinstance(cpu, BaseCPU))
cpu.icache_port = self.inst_seq.in_ports
for p in cpu._cached_ports:
if str(p) != 'icache_port':
exec('cpu.%s = self.data_seq.in_ports' % p)
cpu.connectUncachedPorts(self.data_seq)
def connectIOPorts(self, piobus):
self.data_seq.connectIOPorts(piobus)
def __setattr__(self, name, value):
setattr(self.inst_seq, name, value)
setattr(self.data_seq, name, value)
class CHI_RNF(CHI_Node):
'''
Defines a CHI request node.
Notice all contollers and sequencers are set as children of the cpus, so
this object acts more like a proxy for seting things up and has no topology
significance unless the cpus are set as its children at the top level
'''
def __init__(self, cpus, ruby_system,
l1Icache_type, l1Dcache_type,
cache_line_size,
l1Iprefetcher_type=None, l1Dprefetcher_type=None):
super(CHI_RNF, self).__init__(ruby_system)
self._block_size_bits = int(math.log(cache_line_size, 2))
# All sequencers and controllers
self._seqs = []
self._cntrls = []
# Last level controllers in this node, i.e., the ones that will send
# requests to the home nodes
self._ll_cntrls = []
self._cpus = cpus
# First creates L1 caches and sequencers
for cpu in self._cpus:
cpu.inst_sequencer = RubySequencer(version = Versions.getSeqId(),
ruby_system = ruby_system)
cpu.data_sequencer = RubySequencer(version = Versions.getSeqId(),
ruby_system = ruby_system)
self._seqs.append(CPUSequencerWrapper(cpu.inst_sequencer,
cpu.data_sequencer))
# caches
l1i_cache = l1Icache_type(start_index_bit = self._block_size_bits,
is_icache = True)
l1d_cache = l1Dcache_type(start_index_bit = self._block_size_bits,
is_icache = False)
# Placeholders for future prefetcher support
if l1Iprefetcher_type != None or l1Dprefetcher_type != None:
m5.fatal('Prefetching not supported yet')
l1i_pf = NULL
l1d_pf = NULL
# cache controllers
cpu.l1i = CHI_L1Controller(ruby_system, cpu.inst_sequencer,
l1i_cache, l1i_pf)
cpu.l1d = CHI_L1Controller(ruby_system, cpu.data_sequencer,
l1d_cache, l1d_pf)
cpu.inst_sequencer.dcache = NULL
cpu.data_sequencer.dcache = cpu.l1d.cache
cpu.l1d.sc_lock_enabled = True
cpu._ll_cntrls = [cpu.l1i, cpu.l1d]
for c in cpu._ll_cntrls:
self._cntrls.append(c)
self.connectController(c)
self._ll_cntrls.append(c)
def getSequencers(self):
return self._seqs
def getAllControllers(self):
return self._cntrls
def getNetworkSideControllers(self):
return self._cntrls
def setDownstream(self, cntrls):
for c in self._ll_cntrls:
c.downstream_destinations = cntrls
def getCpus(self):
return self._cpus
# Adds a private L2 for each cpu
def addPrivL2Cache(self, cache_type, pf_type=None):
self._ll_cntrls = []
for cpu in self._cpus:
l2_cache = cache_type(start_index_bit = self._block_size_bits,
is_icache = False)
if pf_type != None:
m5.fatal('Prefetching not supported yet')
l2_pf = NULL
cpu.l2 = CHI_L2Controller(self._ruby_system, l2_cache, l2_pf)
self._cntrls.append(cpu.l2)
self.connectController(cpu.l2)
self._ll_cntrls.append(cpu.l2)
for c in cpu._ll_cntrls:
c.downstream_destinations = [cpu.l2]
cpu._ll_cntrls = [cpu.l2]
class CHI_HNF(CHI_Node):
'''
Encapsulates an HNF cache/directory controller.
Before the first controller is created, the class method
CHI_HNF.createAddrRanges must be called before creating any CHI_HNF object
to set-up the interleaved address ranges used by the HNFs
'''
_addr_ranges = []
@classmethod
def createAddrRanges(cls, sys_mem_ranges, cache_line_size, num_hnfs):
# Create the HNFs interleaved addr ranges
block_size_bits = int(math.log(cache_line_size, 2))
cls._addr_ranges = []
llc_bits = int(math.log(num_hnfs, 2))
numa_bit = block_size_bits + llc_bits - 1
for i in range(num_hnfs):
ranges = []
for r in sys_mem_ranges:
addr_range = AddrRange(r.start, size = r.size(),
intlvHighBit = numa_bit,
intlvBits = llc_bits,
intlvMatch = i)
ranges.append(addr_range)
cls._addr_ranges.append((ranges, numa_bit, i))
@classmethod
def getAddrRanges(cls, hnf_idx):
assert(len(cls._addr_ranges) != 0)
return cls._addr_ranges[hnf_idx]
# The CHI controller can be a child of this object or another if
# 'parent' if specified
def __init__(self, hnf_idx, ruby_system, llcache_type, parent):
super(CHI_HNF, self).__init__(ruby_system)
addr_ranges,intlvHighBit,intlvMatch = CHI_HNF.getAddrRanges(hnf_idx)
# All ranges should have the same interleaving
assert(len(addr_ranges) >= 1)
assert(intlvMatch == hnf_idx)
ll_cache = llcache_type(start_index_bit = intlvHighBit + 1)
self._cntrl = CHI_HNFController(ruby_system, ll_cache, NULL,
addr_ranges)
if parent == None:
self.cntrl = self._cntrl
else:
parent.cntrl = self._cntrl
self.connectController(self._cntrl)
def getAllControllers(self):
return [self._cntrl]
def getNetworkSideControllers(self):
return [self._cntrl]
class CHI_SNF_Base(CHI_Node):
'''
Creates CHI node controllers for the memory controllers
'''
# The CHI controller can be a child of this object or another if
# 'parent' if specified
def __init__(self, ruby_system, parent):
super(CHI_SNF_Base, self).__init__(ruby_system)
self._cntrl = Memory_Controller(
version = Versions.getVersion(Memory_Controller),
ruby_system = ruby_system,
triggerQueue = TriggerMessageBuffer(),
responseFromMemory = MessageBuffer(),
requestToMemory = MessageBuffer(ordered = True),
reqRdy = TriggerMessageBuffer())
self.connectController(self._cntrl)
if parent:
parent.cntrl = self._cntrl
else:
self.cntrl = self._cntrl
def getAllControllers(self):
return [self._cntrl]
def getNetworkSideControllers(self):
return [self._cntrl]
def getMemRange(self, mem_ctrl):
# TODO need some kind of transparent API for
# MemCtrl+DRAM vs SimpleMemory
if hasattr(mem_ctrl, 'range'):
return mem_ctrl.range
else:
return mem_ctrl.dram.range
class CHI_SNF_BootMem(CHI_SNF_Base):
'''
Create the SNF for the boot memory
'''
def __init__(self, ruby_system, parent, bootmem):
super(CHI_SNF_BootMem, self).__init__(ruby_system, parent)
self._cntrl.memory_out_port = bootmem.port
self._cntrl.addr_ranges = self.getMemRange(bootmem)
class CHI_SNF_MainMem(CHI_SNF_Base):
'''
Create the SNF for a list main memory controllers
'''
def __init__(self, ruby_system, parent, mem_ctrl = None):
super(CHI_SNF_MainMem, self).__init__(ruby_system, parent)
if mem_ctrl:
self._cntrl.memory_out_port = mem_ctrl.port
self._cntrl.addr_ranges = self.getMemRange(mem_ctrl)
# else bind ports and range later
class CHI_RNI_Base(CHI_Node):
'''
Request node without cache / DMA
'''
# The CHI controller can be a child of this object or another if
# 'parent' if specified
def __init__(self, ruby_system, parent):
super(CHI_RNI_Base, self).__init__(ruby_system)
self._sequencer = RubySequencer(version = Versions.getSeqId(),
ruby_system = ruby_system,
clk_domain = ruby_system.clk_domain)
self._cntrl = CHI_DMAController(ruby_system, self._sequencer)
if parent:
parent.cntrl = self._cntrl
else:
self.cntrl = self._cntrl
self.connectController(self._cntrl)
def getAllControllers(self):
return [self._cntrl]
def getNetworkSideControllers(self):
return [self._cntrl]
class CHI_RNI_DMA(CHI_RNI_Base):
'''
DMA controller wiredup to a given dma port
'''
def __init__(self, ruby_system, dma_port, parent):
super(CHI_RNI_DMA, self).__init__(ruby_system, parent)
assert(dma_port != None)
self._sequencer.in_ports = dma_port
class CHI_RNI_IO(CHI_RNI_Base):
'''
DMA controller wiredup to ruby_system IO port
'''
def __init__(self, ruby_system, parent):
super(CHI_RNI_IO, self).__init__(ruby_system, parent)
ruby_system._io_port = self._sequencer
def noc_params_from_config(config, noc_params):
# mesh options
noc_params.num_rows = config['mesh']['num_rows']
noc_params.num_cols = config['mesh']['num_cols']
if 'router_latency' in config['mesh']:
noc_params.router_latency = config['mesh']['router_latency']
if 'link_latency' in config['mesh']:
noc_params.router_link_latency = config['mesh']['link_latency']
noc_params.node_link_latency = config['mesh']['link_latency']
if 'router_link_latency' in config['mesh']:
noc_params.router_link_latency = config['mesh']['router_link_latency']
if 'node_link_latency' in config['mesh']:
noc_params.node_link_latency = config['mesh']['node_link_latency']
if 'cross_links' in config['mesh']:
noc_params.cross_link_latency = \
config['mesh']['cross_link_latency']
noc_params.cross_links = []
for x, y in config['mesh']['cross_links']:
noc_params.cross_links.append((x, y))
noc_params.cross_links.append((y, x))
else:
noc_params.cross_links = []
noc_params.cross_link_latency = 0
# CHI_RNF options
noc_params.CHI_RNF = config['CHI_RNF']
# CHI_RNI_IO
noc_params.CHI_RNI_IO = config['CHI_RNI_IO']
# CHI_HNF options
noc_params.CHI_HNF = config['CHI_HNF']
if 'pairing' in config['CHI_HNF']:
noc_params.pairing = config['CHI_HNF']['pairing']
# CHI_SNF_MainMem
noc_params.CHI_SNF_MainMem = config['CHI_SNF_MainMem']
# CHI_SNF_IO (applies to CHI_SNF_Bootmem)
noc_params.CHI_SNF_IO = config['CHI_SNF_IO']
def create_system(options, full_system, system, dma_ports, bootmem, def create_system(options, full_system, system, dma_ports, bootmem,
ruby_system): ruby_system):
@@ -657,25 +65,25 @@ def create_system(options, full_system, system, dma_ports, bootmem,
if options.num_l3caches < 1: if options.num_l3caches < 1:
m5.fatal('--num-l3caches must be at least 1') m5.fatal('--num-l3caches must be at least 1')
# Default parameters for the network # read specialized classes from config file if provided
class NoC_Params(object): if options.chi_config:
def __init__(self): chi_defs = read_config_file(options.chi_config)
self.topology = options.topology elif options.topology == 'CustomMesh':
self.network = options.network
self.router_link_latency = 1
self.node_link_latency = 1
self.router_latency = 1
self.router_buffer_size = 4
self.cntrl_msg_size = 8
self.data_width = 32
params = NoC_Params()
# read additional configurations from yaml file if provided
if options.noc_config:
with open(options.noc_config, 'r') as file:
noc_params_from_config(yaml.load(file), params)
elif params.topology == 'CustomMesh':
m5.fatal('--noc-config must be provided if topology is CustomMesh') m5.fatal('--noc-config must be provided if topology is CustomMesh')
else:
# Use the defaults from CHI_config
from . import CHI_config as chi_defs
# NoC params
params = chi_defs.NoC_Params
# Node types
CHI_RNF = chi_defs.CHI_RNF
CHI_HNF = chi_defs.CHI_HNF
CHI_SNF_MainMem = chi_defs.CHI_SNF_MainMem
CHI_SNF_BootMem = chi_defs.CHI_SNF_BootMem
CHI_RNI_DMA = chi_defs.CHI_RNI_DMA
CHI_RNI_IO = chi_defs.CHI_RNI_IO
# Declare caches and controller types used by the protocol # Declare caches and controller types used by the protocol
# Notice tag and data accesses are not concurrent, so the a cache hit # Notice tag and data accesses are not concurrent, so the a cache hit
@@ -824,17 +232,17 @@ def create_system(options, full_system, system, dma_ports, bootmem,
ruby_system.network.data_msg_size = params.data_width ruby_system.network.data_msg_size = params.data_width
ruby_system.network.buffer_size = params.router_buffer_size ruby_system.network.buffer_size = params.router_buffer_size
if params.topology == 'CustomMesh':
topology = create_topology(network_nodes, params)
elif params.topology in ['Crossbar', 'Pt2Pt']:
topology = create_topology(network_cntrls, params)
else:
m5.fatal("%s not supported!" % params.topology)
# Incorporate the params into options so it's propagated to # Incorporate the params into options so it's propagated to
# makeTopology by the parent script # makeTopology and create_topology the parent scripts
for k in dir(params): for k in dir(params):
if not k.startswith('__'): if not k.startswith('__'):
setattr(options, k, getattr(params, k)) setattr(options, k, getattr(params, k))
if options.topology == 'CustomMesh':
topology = create_topology(network_nodes, options)
elif options.topology in ['Crossbar', 'Pt2Pt']:
topology = create_topology(network_cntrls, options)
else:
m5.fatal("%s not supported!" % options.topology)
return (cpu_sequencers, mem_cntrls, topology) return (cpu_sequencers, mem_cntrls, topology)

646
configs/ruby/CHI_config.py Normal file
View File

@@ -0,0 +1,646 @@
# Copyright (c) 2021 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.
#
# 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.
'''
Definitions for CHI nodes and controller types. These are used by
create_system in configs/ruby/CHI.py or may be used in custom configuration
scripts. When used with create_system, the user may provide an additional
configuration file as the --chi-config parameter to specialize the classes
defined here.
When using the CustomMesh topology, --chi-config must be provided with
specialization of the NoC_Param classes defining the NoC dimensions and
node to router binding. See configs/example/noc_config/2x4.py for an example.
'''
import math
import m5
from m5.objects import *
class Versions:
'''
Helper class to obtain unique ids for a given controller class.
These are passed as the 'version' parameter when creating the controller.
'''
_seqs = 0
@classmethod
def getSeqId(cls):
val = cls._seqs
cls._seqs += 1
return val
_version = {}
@classmethod
def getVersion(cls, tp):
if tp not in cls._version:
cls._version[tp] = 0
val = cls._version[tp]
cls._version[tp] = val + 1
return val
class NoC_Params:
'''
Default parameters for the interconnect. The value of data_width is
also used to set the data_channel_size for all CHI controllers.
(see configs/ruby/CHI.py)
'''
router_link_latency = 1
node_link_latency = 1
router_latency = 1
router_buffer_size = 4
cntrl_msg_size = 8
data_width = 32
cross_links = []
cross_link_latency = 0
class CHI_Node(SubSystem):
'''
Base class with common functions for setting up Cache or Memory
controllers that are part of a CHI RNF, RNFI, HNF, or SNF nodes.
Notice getNetworkSideControllers and getAllControllers must be implemented
in the derived classes.
'''
class NoC_Params:
'''
NoC config. parameters and bindings required for CustomMesh topology.
Maps 'num_nodes_per_router' CHI nodes to each router provided in
'router_list'. This assumes len(router_list)*num_nodes_per_router
equals the number of nodes
If 'num_nodes_per_router' is left undefined, we circulate around
'router_list' until all nodes are mapped.
See 'distributeNodes' in configs/topologies/CustomMesh.py
'''
num_nodes_per_router = None
router_list = None
def __init__(self, ruby_system):
super(CHI_Node, self).__init__()
self._ruby_system = ruby_system
self._network = ruby_system.network
def getNetworkSideControllers(self):
'''
Returns all ruby controllers that need to be connected to the
network
'''
raise NotImplementedError()
def getAllControllers(self):
'''
Returns all ruby controllers associated with this node
'''
raise NotImplementedError()
def setDownstream(self, cntrls):
'''
Sets cntrls as the downstream list of all controllers in this node
'''
for c in self.getNetworkSideControllers():
c.downstream_destinations = cntrls
def connectController(self, cntrl):
'''
Creates and configures the messages buffers for the CHI input/output
ports that connect to the network
'''
cntrl.reqOut = MessageBuffer()
cntrl.rspOut = MessageBuffer()
cntrl.snpOut = MessageBuffer()
cntrl.datOut = MessageBuffer()
cntrl.reqIn = MessageBuffer()
cntrl.rspIn = MessageBuffer()
cntrl.snpIn = MessageBuffer()
cntrl.datIn = MessageBuffer()
# All CHI ports are always connected to the network.
# Controllers that are not part of the getNetworkSideControllers list
# still communicate using internal routers, thus we need to wire-up the
# ports
cntrl.reqOut.out_port = self._network.in_port
cntrl.rspOut.out_port = self._network.in_port
cntrl.snpOut.out_port = self._network.in_port
cntrl.datOut.out_port = self._network.in_port
cntrl.reqIn.in_port = self._network.out_port
cntrl.rspIn.in_port = self._network.out_port
cntrl.snpIn.in_port = self._network.out_port
cntrl.datIn.in_port = self._network.out_port
class TriggerMessageBuffer(MessageBuffer):
'''
MessageBuffer for triggering internal controller events.
These buffers should not be affected by the Ruby tester randomization
and allow poping messages enqueued in the same cycle.
'''
randomization = 'disabled'
allow_zero_latency = True
class OrderedTriggerMessageBuffer(TriggerMessageBuffer):
ordered = True
class CHI_Cache_Controller(Cache_Controller):
'''
Default parameters for a Cache controller
The Cache_Controller can also be used as a DMA requester or as
a pure directory if all cache allocation policies are disabled.
'''
def __init__(self, ruby_system):
super(CHI_Cache_Controller, self).__init__(
version = Versions.getVersion(Cache_Controller),
ruby_system = ruby_system,
mandatoryQueue = MessageBuffer(),
prefetchQueue = MessageBuffer(),
triggerQueue = TriggerMessageBuffer(),
retryTriggerQueue = OrderedTriggerMessageBuffer(),
replTriggerQueue = OrderedTriggerMessageBuffer(),
reqRdy = TriggerMessageBuffer(),
snpRdy = TriggerMessageBuffer())
# Set somewhat large number since we really a lot on internal
# triggers. To limit the controller performance, tweak other
# params such as: input port buffer size, cache banks, and output
# port latency
self.transitions_per_cycle = 128
# This should be set to true in the data cache controller to enable
# timeouts on unique lines when a store conditional fails
self.sc_lock_enabled = False
class CHI_L1Controller(CHI_Cache_Controller):
'''
Default parameters for a L1 Cache controller
'''
def __init__(self, ruby_system, sequencer, cache, prefetcher):
super(CHI_L1Controller, self).__init__(ruby_system)
self.sequencer = sequencer
self.cache = cache
self.use_prefetcher = False
self.send_evictions = True
self.is_HN = False
self.enable_DMT = False
self.enable_DCT = False
# Strict inclusive MOESI
self.allow_SD = True
self.alloc_on_seq_acc = True
self.alloc_on_seq_line_write = False
self.alloc_on_readshared = True
self.alloc_on_readunique = True
self.alloc_on_readonce = True
self.alloc_on_writeback = True
self.dealloc_on_unique = False
self.dealloc_on_shared = False
self.dealloc_backinv_unique = True
self.dealloc_backinv_shared = True
# Some reasonable default TBE params
self.number_of_TBEs = 16
self.number_of_repl_TBEs = 16
self.number_of_snoop_TBEs = 4
self.unify_repl_TBEs = False
class CHI_L2Controller(CHI_Cache_Controller):
'''
Default parameters for a L2 Cache controller
'''
def __init__(self, ruby_system, cache, prefetcher):
super(CHI_L2Controller, self).__init__(ruby_system)
self.sequencer = NULL
self.cache = cache
self.use_prefetcher = False
self.allow_SD = True
self.is_HN = False
self.enable_DMT = False
self.enable_DCT = False
self.send_evictions = False
# Strict inclusive MOESI
self.alloc_on_seq_acc = False
self.alloc_on_seq_line_write = False
self.alloc_on_readshared = True
self.alloc_on_readunique = True
self.alloc_on_readonce = True
self.alloc_on_writeback = True
self.dealloc_on_unique = False
self.dealloc_on_shared = False
self.dealloc_backinv_unique = True
self.dealloc_backinv_shared = True
# Some reasonable default TBE params
self.number_of_TBEs = 32
self.number_of_repl_TBEs = 32
self.number_of_snoop_TBEs = 16
self.unify_repl_TBEs = False
class CHI_HNFController(CHI_Cache_Controller):
'''
Default parameters for a coherent home node (HNF) cache controller
'''
def __init__(self, ruby_system, cache, prefetcher, addr_ranges):
super(CHI_HNFController, self).__init__(ruby_system)
self.sequencer = NULL
self.cache = cache
self.use_prefetcher = False
self.addr_ranges = addr_ranges
self.allow_SD = True
self.is_HN = True
self.enable_DMT = True
self.enable_DCT = True
self.send_evictions = False
# MOESI / Mostly inclusive for shared / Exclusive for unique
self.alloc_on_seq_acc = False
self.alloc_on_seq_line_write = False
self.alloc_on_readshared = True
self.alloc_on_readunique = False
self.alloc_on_readonce = True
self.alloc_on_writeback = True
self.dealloc_on_unique = True
self.dealloc_on_shared = False
self.dealloc_backinv_unique = False
self.dealloc_backinv_shared = False
# Some reasonable default TBE params
self.number_of_TBEs = 32
self.number_of_repl_TBEs = 32
self.number_of_snoop_TBEs = 1 # should not receive any snoop
self.unify_repl_TBEs = False
class CHI_DMAController(CHI_Cache_Controller):
'''
Default parameters for a DMA controller
'''
def __init__(self, ruby_system, sequencer):
super(CHI_DMAController, self).__init__(ruby_system)
self.sequencer = sequencer
class DummyCache(RubyCache):
dataAccessLatency = 0
tagAccessLatency = 1
size = "128"
assoc = 1
self.use_prefetcher = False
self.cache = DummyCache()
self.sequencer.dcache = NULL
# All allocations are false
# Deallocations are true (don't really matter)
self.allow_SD = False
self.is_HN = False
self.enable_DMT = False
self.enable_DCT = False
self.alloc_on_seq_acc = False
self.alloc_on_seq_line_write = False
self.alloc_on_readshared = False
self.alloc_on_readunique = False
self.alloc_on_readonce = False
self.alloc_on_writeback = False
self.dealloc_on_unique = False
self.dealloc_on_shared = False
self.dealloc_backinv_unique = False
self.dealloc_backinv_shared = False
self.send_evictions = False
self.number_of_TBEs = 16
self.number_of_repl_TBEs = 1
self.number_of_snoop_TBEs = 1 # should not receive any snoop
self.unify_repl_TBEs = False
class CPUSequencerWrapper:
'''
Other generic configuration scripts assume a matching number of sequencers
and cpus. This wraps the instruction and data sequencer so they are
compatible with the other scripts. This assumes all scripts are using
connectCpuPorts/connectIOPorts to bind ports
'''
def __init__(self, iseq, dseq):
# use this style due to __setattr__ override below
self.__dict__['inst_seq'] = iseq
self.__dict__['data_seq'] = dseq
self.__dict__['support_data_reqs'] = True
self.__dict__['support_inst_reqs'] = True
# Compatibility with certain scripts that wire up ports
# without connectCpuPorts
self.__dict__['slave'] = dseq.in_ports
self.__dict__['in_ports'] = dseq.in_ports
def connectCpuPorts(self, cpu):
assert(isinstance(cpu, BaseCPU))
cpu.icache_port = self.inst_seq.in_ports
for p in cpu._cached_ports:
if str(p) != 'icache_port':
exec('cpu.%s = self.data_seq.in_ports' % p)
cpu.connectUncachedPorts(self.data_seq)
def connectIOPorts(self, piobus):
self.data_seq.connectIOPorts(piobus)
def __setattr__(self, name, value):
setattr(self.inst_seq, name, value)
setattr(self.data_seq, name, value)
class CHI_RNF(CHI_Node):
'''
Defines a CHI request node.
Notice all contollers and sequencers are set as children of the cpus, so
this object acts more like a proxy for seting things up and has no topology
significance unless the cpus are set as its children at the top level
'''
def __init__(self, cpus, ruby_system,
l1Icache_type, l1Dcache_type,
cache_line_size,
l1Iprefetcher_type=None, l1Dprefetcher_type=None):
super(CHI_RNF, self).__init__(ruby_system)
self._block_size_bits = int(math.log(cache_line_size, 2))
# All sequencers and controllers
self._seqs = []
self._cntrls = []
# Last level controllers in this node, i.e., the ones that will send
# requests to the home nodes
self._ll_cntrls = []
self._cpus = cpus
# First creates L1 caches and sequencers
for cpu in self._cpus:
cpu.inst_sequencer = RubySequencer(version = Versions.getSeqId(),
ruby_system = ruby_system)
cpu.data_sequencer = RubySequencer(version = Versions.getSeqId(),
ruby_system = ruby_system)
self._seqs.append(CPUSequencerWrapper(cpu.inst_sequencer,
cpu.data_sequencer))
# caches
l1i_cache = l1Icache_type(start_index_bit = self._block_size_bits,
is_icache = True)
l1d_cache = l1Dcache_type(start_index_bit = self._block_size_bits,
is_icache = False)
# Placeholders for future prefetcher support
if l1Iprefetcher_type != None or l1Dprefetcher_type != None:
m5.fatal('Prefetching not supported yet')
l1i_pf = NULL
l1d_pf = NULL
# cache controllers
cpu.l1i = CHI_L1Controller(ruby_system, cpu.inst_sequencer,
l1i_cache, l1i_pf)
cpu.l1d = CHI_L1Controller(ruby_system, cpu.data_sequencer,
l1d_cache, l1d_pf)
cpu.inst_sequencer.dcache = NULL
cpu.data_sequencer.dcache = cpu.l1d.cache
cpu.l1d.sc_lock_enabled = True
cpu._ll_cntrls = [cpu.l1i, cpu.l1d]
for c in cpu._ll_cntrls:
self._cntrls.append(c)
self.connectController(c)
self._ll_cntrls.append(c)
def getSequencers(self):
return self._seqs
def getAllControllers(self):
return self._cntrls
def getNetworkSideControllers(self):
return self._cntrls
def setDownstream(self, cntrls):
for c in self._ll_cntrls:
c.downstream_destinations = cntrls
def getCpus(self):
return self._cpus
# Adds a private L2 for each cpu
def addPrivL2Cache(self, cache_type, pf_type=None):
self._ll_cntrls = []
for cpu in self._cpus:
l2_cache = cache_type(start_index_bit = self._block_size_bits,
is_icache = False)
if pf_type != None:
m5.fatal('Prefetching not supported yet')
l2_pf = NULL
cpu.l2 = CHI_L2Controller(self._ruby_system, l2_cache, l2_pf)
self._cntrls.append(cpu.l2)
self.connectController(cpu.l2)
self._ll_cntrls.append(cpu.l2)
for c in cpu._ll_cntrls:
c.downstream_destinations = [cpu.l2]
cpu._ll_cntrls = [cpu.l2]
class CHI_HNF(CHI_Node):
'''
Encapsulates an HNF cache/directory controller.
Before the first controller is created, the class method
CHI_HNF.createAddrRanges must be called before creating any CHI_HNF object
to set-up the interleaved address ranges used by the HNFs
'''
class NoC_Params(CHI_Node.NoC_Params):
'''HNFs may also define the 'pairing' parameter to allow pairing'''
pairing = None
_addr_ranges = []
@classmethod
def createAddrRanges(cls, sys_mem_ranges, cache_line_size, num_hnfs):
# Create the HNFs interleaved addr ranges
block_size_bits = int(math.log(cache_line_size, 2))
cls._addr_ranges = []
llc_bits = int(math.log(num_hnfs, 2))
numa_bit = block_size_bits + llc_bits - 1
for i in range(num_hnfs):
ranges = []
for r in sys_mem_ranges:
addr_range = AddrRange(r.start, size = r.size(),
intlvHighBit = numa_bit,
intlvBits = llc_bits,
intlvMatch = i)
ranges.append(addr_range)
cls._addr_ranges.append((ranges, numa_bit, i))
@classmethod
def getAddrRanges(cls, hnf_idx):
assert(len(cls._addr_ranges) != 0)
return cls._addr_ranges[hnf_idx]
# The CHI controller can be a child of this object or another if
# 'parent' if specified
def __init__(self, hnf_idx, ruby_system, llcache_type, parent):
super(CHI_HNF, self).__init__(ruby_system)
addr_ranges,intlvHighBit,intlvMatch = self.getAddrRanges(hnf_idx)
# All ranges should have the same interleaving
assert(len(addr_ranges) >= 1)
assert(intlvMatch == hnf_idx)
ll_cache = llcache_type(start_index_bit = intlvHighBit + 1)
self._cntrl = CHI_HNFController(ruby_system, ll_cache, NULL,
addr_ranges)
if parent == None:
self.cntrl = self._cntrl
else:
parent.cntrl = self._cntrl
self.connectController(self._cntrl)
def getAllControllers(self):
return [self._cntrl]
def getNetworkSideControllers(self):
return [self._cntrl]
class CHI_SNF_Base(CHI_Node):
'''
Creates CHI node controllers for the memory controllers
'''
# The CHI controller can be a child of this object or another if
# 'parent' if specified
def __init__(self, ruby_system, parent):
super(CHI_SNF_Base, self).__init__(ruby_system)
self._cntrl = Memory_Controller(
version = Versions.getVersion(Memory_Controller),
ruby_system = ruby_system,
triggerQueue = TriggerMessageBuffer(),
responseFromMemory = MessageBuffer(),
requestToMemory = MessageBuffer(ordered = True),
reqRdy = TriggerMessageBuffer())
self.connectController(self._cntrl)
if parent:
parent.cntrl = self._cntrl
else:
self.cntrl = self._cntrl
def getAllControllers(self):
return [self._cntrl]
def getNetworkSideControllers(self):
return [self._cntrl]
def getMemRange(self, mem_ctrl):
# TODO need some kind of transparent API for
# MemCtrl+DRAM vs SimpleMemory
if hasattr(mem_ctrl, 'range'):
return mem_ctrl.range
else:
return mem_ctrl.dram.range
class CHI_SNF_BootMem(CHI_SNF_Base):
'''
Create the SNF for the boot memory
'''
def __init__(self, ruby_system, parent, bootmem):
super(CHI_SNF_BootMem, self).__init__(ruby_system, parent)
self._cntrl.memory_out_port = bootmem.port
self._cntrl.addr_ranges = self.getMemRange(bootmem)
class CHI_SNF_MainMem(CHI_SNF_Base):
'''
Create the SNF for a list main memory controllers
'''
def __init__(self, ruby_system, parent, mem_ctrl = None):
super(CHI_SNF_MainMem, self).__init__(ruby_system, parent)
if mem_ctrl:
self._cntrl.memory_out_port = mem_ctrl.port
self._cntrl.addr_ranges = self.getMemRange(mem_ctrl)
# else bind ports and range later
class CHI_RNI_Base(CHI_Node):
'''
Request node without cache / DMA
'''
# The CHI controller can be a child of this object or another if
# 'parent' if specified
def __init__(self, ruby_system, parent):
super(CHI_RNI_Base, self).__init__(ruby_system)
self._sequencer = RubySequencer(version = Versions.getSeqId(),
ruby_system = ruby_system,
clk_domain = ruby_system.clk_domain)
self._cntrl = CHI_DMAController(ruby_system, self._sequencer)
if parent:
parent.cntrl = self._cntrl
else:
self.cntrl = self._cntrl
self.connectController(self._cntrl)
def getAllControllers(self):
return [self._cntrl]
def getNetworkSideControllers(self):
return [self._cntrl]
class CHI_RNI_DMA(CHI_RNI_Base):
'''
DMA controller wiredup to a given dma port
'''
def __init__(self, ruby_system, dma_port, parent):
super(CHI_RNI_DMA, self).__init__(ruby_system, parent)
assert(dma_port != None)
self._sequencer.in_ports = dma_port
class CHI_RNI_IO(CHI_RNI_Base):
'''
DMA controller wiredup to ruby_system IO port
'''
def __init__(self, ruby_system, parent):
super(CHI_RNI_IO, self).__init__(ruby_system, parent)
ruby_system._io_port = self._sequencer

83
configs/topologies/CustomMesh.py
View File

@@ -42,7 +42,7 @@ from m5.objects import *
from m5.defines import buildEnv from m5.defines import buildEnv
if buildEnv['PROTOCOL'] == 'CHI': if buildEnv['PROTOCOL'] == 'CHI':
import ruby.CHI as CHI import ruby.CHI_config as CHI
from topologies.BaseTopology import SimpleTopology from topologies.BaseTopology import SimpleTopology
@@ -163,8 +163,12 @@ class CustomMesh(SimpleTopology):
return node_router return node_router
def distributeNodes(self, num_nodes_per_router, router_idx_list, def distributeNodes(self, node_placement_config, node_list):
node_list): if len(node_list) == 0:
return
num_nodes_per_router = node_placement_config.num_nodes_per_router
router_idx_list = node_placement_config.router_list
if num_nodes_per_router: if num_nodes_per_router:
# evenly distribute nodes to all listed routers # evenly distribute nodes to all listed routers
@@ -233,25 +237,45 @@ class CustomMesh(SimpleTopology):
self._node_link_latency = options.link_latency self._node_link_latency = options.link_latency
# classify nodes into different types # classify nodes into different types
rnf_list = [] rnf_nodes = []
hnf_list = [] hnf_nodes = []
mem_ctrls = [] mem_nodes = []
io_mem_ctrls = [] io_mem_nodes = []
io_rni_ctrls = [] rni_dma_nodes = []
rni_io_nodes = []
# Notice below that all the type must be the same for all nodes with
# the same base type.
rnf_params = None
hnf_params = None
mem_params = None
io_mem_params = None
rni_dma_params = None
rni_io_params = None
def check_same(val, curr):
assert(curr == None or curr == val)
return val
for n in self.nodes: for n in self.nodes:
if isinstance(n, CHI.CHI_RNF): if isinstance(n, CHI.CHI_RNF):
rnf_list.append(n) rnf_nodes.append(n)
rnf_params = check_same(type(n).NoC_Params, rnf_params)
elif isinstance(n, CHI.CHI_HNF): elif isinstance(n, CHI.CHI_HNF):
hnf_list.append(n) hnf_nodes.append(n)
hnf_params = check_same(type(n).NoC_Params, hnf_params)
elif isinstance(n, CHI.CHI_SNF_MainMem): elif isinstance(n, CHI.CHI_SNF_MainMem):
mem_ctrls.append(n) mem_nodes.append(n)
mem_params = check_same(type(n).NoC_Params, mem_params)
elif isinstance(n, CHI.CHI_SNF_BootMem): elif isinstance(n, CHI.CHI_SNF_BootMem):
io_mem_ctrls.append(n) io_mem_nodes.append(n)
io_mem_params = check_same(type(n).NoC_Params, io_mem_params)
elif isinstance(n, CHI.CHI_RNI_DMA): elif isinstance(n, CHI.CHI_RNI_DMA):
io_rni_ctrls.append(n) rni_dma_nodes.append(n)
rni_dma_params = check_same(type(n).NoC_Params, rni_dma_params)
elif isinstance(n, CHI.CHI_RNI_IO): elif isinstance(n, CHI.CHI_RNI_IO):
io_rni_ctrls.append(n) rni_io_nodes.append(n)
rni_io_params = check_same(type(n).NoC_Params, rni_io_params)
else: else:
fatal('topologies.CustomMesh: {} not supported' fatal('topologies.CustomMesh: {} not supported'
.format(n.__class__.__name__)) .format(n.__class__.__name__))
@@ -269,39 +293,20 @@ class CustomMesh(SimpleTopology):
options.cross_links, options.cross_link_latency) options.cross_links, options.cross_link_latency)
# Place CHI_RNF on the mesh # Place CHI_RNF on the mesh
num_nodes_per_router = options.CHI_RNF['num_nodes_per_router'] \ self.distributeNodes(rnf_params, rnf_nodes)
if 'num_nodes_per_router' in options.CHI_RNF else None
self.distributeNodes(num_nodes_per_router,
options.CHI_RNF['router_list'],
rnf_list)
# Place CHI_HNF on the mesh # Place CHI_HNF on the mesh
num_nodes_per_router = options.CHI_HNF['num_nodes_per_router'] \ self.distributeNodes(hnf_params, hnf_nodes)
if 'num_nodes_per_router' in options.CHI_HNF else None
self.distributeNodes(num_nodes_per_router,
options.CHI_HNF['router_list'],
hnf_list)
# Place CHI_SNF_MainMem on the mesh # Place CHI_SNF_MainMem on the mesh
num_nodes_per_router = options.CHI_SNF_MainMem['num_nodes_per_router']\ self.distributeNodes(mem_params, mem_nodes)
if 'num_nodes_per_router' in options.CHI_SNF_MainMem else None
self.distributeNodes(num_nodes_per_router,
options.CHI_SNF_MainMem['router_list'],
mem_ctrls)
# Place all IO mem nodes on the mesh # Place all IO mem nodes on the mesh
num_nodes_per_router = options.CHI_SNF_IO['num_nodes_per_router'] \ self.distributeNodes(io_mem_params, io_mem_nodes)
if 'num_nodes_per_router' in options.CHI_SNF_IO else None
self.distributeNodes(num_nodes_per_router,
options.CHI_SNF_IO['router_list'],
io_mem_ctrls)
# Place all IO request nodes on the mesh # Place all IO request nodes on the mesh
num_nodes_per_router = options.CHI_RNI_IO['num_nodes_per_router'] \ self.distributeNodes(rni_dma_params, rni_dma_nodes)
if 'num_nodes_per_router' in options.CHI_RNI_IO else None self.distributeNodes(rni_io_params, rni_io_nodes)
self.distributeNodes(num_nodes_per_router,
options.CHI_RNI_IO['router_list'],
io_rni_ctrls)
# Set up # Set up
network.int_links = self._int_links network.int_links = self._int_links