def __init__(self, ruby_system, ranges, mem_ctrls):
if len(mem_ctrls) > 1:
panic("the system can only have one memory control")
self.version = versionCount()
self.addr_ranges = ranges
self.ruby_system = ruby_system
self.directory = RubyDirectoryMemory()
self.memory = mem_ctrls[0].port
self.connectQueues(ruby_system)
def store_stat_value(stat, session, dumpCount):
""" Stores the value of a stat.
Args:
stat: The stat itself.
session: The session associated with the database.
dumpCount: The number of dumps that have occured. Used to store multiple
stats dumps in one database.
"""
if isinstance(stat, ScalarInfo):
temp = ScalarValueClass(id = stat.id, dump = dumpCount,
value = stat.value())
session.add(temp)
elif isinstance(stat, VectorInfo) or isinstance(stat, FormulaInfo):
temp = VectorValueClass(id = stat.id, dump = dumpCount,
value = stat.result())
session.add(temp)
elif isinstance(stat, Vector2dInfo):
temp = VectorValueClass(id = stat.id, dump = dumpCount,
value = stat.cvec)
session.add(temp)
elif isinstance(stat, DistInfo):
import array
data = stat.data
temp = DistValueClass(id = stat.id, dump = dumpCount)
temp.sum = data.sum
temp.squares = data.squares
temp.samples = data.samples
if data.type == Dist or data.type == Hist:
temp.min = data.min
temp.max = data.max
temp.bucket = data.bucket_size
a = array.array('f', data.cvec)
temp.vector = a.tostring()
if data.type == Dist:
temp.min_val = data.min_val
temp.max_val = data.max_val
temp.underflow = data.underflow
temp.overflow = data.overflow
session.add(temp)
else:
panic("Unable to output stat %s. Unsupported stat type!", stat)
def __init__(self, ruby_system, ranges, mem_ctrls):
"""ranges are the memory ranges assigned to this controller.
"""
if len(mem_ctrls) > 1:
panic("This cache system can only be connected to one mem ctrl")
super(DirController, self).__init__()
self.version = self.versionCount()
self.addr_ranges = ranges
self.ruby_system = ruby_system
self.directory = RubyDirectoryMemory()
# Connect this directory to the memory side.
self.memory_out_port = mem_ctrls[0].port
self.connectQueues(ruby_system)
def __init__(self, ruby_system, ranges, mem_ctrls):
"""ranges are the memory ranges assigned to this controller.
"""
if len(mem_ctrls) > 1:
panic("This cache system can only be connected to one mem ctrl")
super(DirController, self).__init__()
self.version = self.versionCount()
self.addr_ranges = ranges
self.ruby_system = ruby_system
self.directory = RubyDirectoryMemory()
# Connect this directory to the memory side.
self.memory = mem_ctrls[0].port
self.connectQueues(ruby_system)
def create_database(filename):
""" Create the database used to store the stats. If it exists, delete it.
Args:
filename: The filename usd to store the stats.
Return:
A handle to the database.
"""
if os.path.exists(filename):
os.remove(filename)
try:
db = create_engine('sqlite:///' + filename)
except:
panic("Failed to open database %s!", filename)
db.echo = False
return db
def getBlockSizeBits(self, system):
bits = int(math.log(system.cache_line_size, 2))
if 2**bits != system.cache_line_size.value:
panic("Cache line size not a power of 2!")
return bits
def getBlockSizeBits(self, system):
bits = int(math.log(system.cache_line_size, 2))
if 2**bits != system.cache_line_size.value:
panic("Cache line size not a power of 2!")
return bits
#
# Full system configuraiton for ruby
#
import os
import optparse
import sys
from os.path import join as joinpath
import m5
from m5.defines import buildEnv
from m5.objects import *
from m5.util import addToPath, panic
if not buildEnv['FULL_SYSTEM']:
panic("This script requires full-system mode (*_FS).")
addToPath('../common')
addToPath('../ruby')
import Ruby
from FSConfig import *
from SysPaths import *
from Benchmarks import *
import Simulation
from Caches import *
# Get paths we might need. It's expected this file is in m5/configs/example.
config_path = os.path.dirname(os.path.abspath(__file__))
config_root = os.path.dirname(config_path)
#
# Full system configuraiton for ruby
#
import os
import optparse
import sys
from os.path import join as joinpath
import m5
from m5.defines import buildEnv
from m5.objects import *
from m5.util import addToPath, panic
if not buildEnv['FULL_SYSTEM']:
panic("This script requires full-system mode (*_FS).")
addToPath('../common')
addToPath('../ruby')
import Ruby
from FSConfig import *
from SysPaths import *
from Benchmarks import *
import Simulation
from Caches import *
# Get paths we might need. It's expected this file is in m5/configs/example.
config_path = os.path.dirname(os.path.abspath(__file__))
config_root = os.path.dirname(config_path)
def __init__(self):
if buildEnv['PROTOCOL'] != 'MSI':
panic("This system assumes MSI to be the protocol")
super(MyCacheSystem, self).__init__()
# Simple test script
#
# "m5 test.py"
import os
import optparse
import sys
from os.path import join as joinpath
import m5
from m5.defines import buildEnv
from m5.objects import *
from m5.util import addToPath, panic
if buildEnv['FULL_SYSTEM']:
panic("This script requires syscall emulation mode (*_SE).")
addToPath('../common')
addToPath('../ruby')
import Ruby
import Simulation
from cpu2000 import *
# Get paths we might need. It's expected this file is in m5/configs/example.
config_path = os.path.dirname(os.path.abspath(__file__))
config_root = os.path.dirname(config_path)
m5_root = os.path.dirname(config_root)
parser = optparse.OptionParser()
# Simple test script
#
# "m5 test.py"
import os
import optparse
import sys
from os.path import join as joinpath
import m5
from m5.defines import buildEnv
from m5.objects import *
from m5.util import addToPath, panic
if buildEnv['FULL_SYSTEM']:
panic("This script requires syscall emulation mode (*_SE).")
addToPath('../common')
addToPath('../ruby')
import Ruby
import Simulation
from cpu2000 import *
# Get paths we might need. It's expected this file is in m5/configs/example.
config_path = os.path.dirname(os.path.abspath(__file__))
config_root = os.path.dirname(config_path)
m5_root = os.path.dirname(config_root)
parser = optparse.OptionParser()
def makeGpuFSSystem(args):
# Boot options are standard gem5 options plus:
# - Framebuffer device emulation 0 to reduce driver code paths.
# - Blacklist amdgpu as it cannot (currently) load in KVM CPU.
# - Blacklist psmouse as amdgpu driver adds proprietary commands that
# cause gem5 to panic.
boot_options = ['earlyprintk=ttyS0', 'console=ttyS0,9600',
'lpj=7999923', 'root=/dev/sda1',
'drm_kms_helper.fbdev_emulation=0',
'modprobe.blacklist=amdgpu',
'modprobe.blacklist=psmouse']
cmdline = ' '.join(boot_options)
if MemorySize(args.mem_size) < MemorySize('2GB'):
panic("Need at least 2GB of system memory to load amdgpu module")
# Use the common FSConfig to setup a Linux X86 System
(TestCPUClass, test_mem_mode, FutureClass) = Simulation.setCPUClass(args)
bm = SysConfig(disks=[args.disk_image], mem=args.mem_size)
system = makeLinuxX86System(test_mem_mode, args.num_cpus, bm, True,
cmdline=cmdline)
system.workload.object_file = binary(args.kernel)
# Set the cache line size for the entire system.
system.cache_line_size = args.cacheline_size
# Create a top-level voltage and clock domain.
system.voltage_domain = VoltageDomain(voltage = args.sys_voltage)
system.clk_domain = SrcClockDomain(clock = args.sys_clock,
voltage_domain = system.voltage_domain)
# Create a CPU voltage and clock domain.
system.cpu_voltage_domain = VoltageDomain()
system.cpu_clk_domain = SrcClockDomain(clock = args.cpu_clock,
voltage_domain =
system.cpu_voltage_domain)
# Setup VGA ROM region
system.shadow_rom_ranges = [AddrRange(0xc0000, size = Addr('128kB'))]
# Create specified number of CPUs. GPUFS really only needs one.
system.cpu = [TestCPUClass(clk_domain=system.cpu_clk_domain, cpu_id=i)
for i in range(args.num_cpus)]
if ObjectList.is_kvm_cpu(TestCPUClass) or \
ObjectList.is_kvm_cpu(FutureClass):
system.kvm_vm = KvmVM()
# Create AMDGPU and attach to southbridge
shader = createGPU(system, args)
connectGPU(system, args)
# This arbitrary address is something in the X86 I/O hole
hsapp_gpu_map_paddr = 0xe00000000
gpu_hsapp = HSAPacketProcessor(pioAddr=hsapp_gpu_map_paddr,
numHWQueues=args.num_hw_queues)
dispatcher = GPUDispatcher()
gpu_cmd_proc = GPUCommandProcessor(hsapp=gpu_hsapp,
dispatcher=dispatcher)
shader.dispatcher = dispatcher
shader.gpu_cmd_proc = gpu_cmd_proc
# GPU, HSAPP, and GPUCommandProc are DMA devices
system._dma_ports.append(gpu_hsapp)
system._dma_ports.append(gpu_cmd_proc)
system._dma_ports.append(system.pc.south_bridge.gpu)
gpu_hsapp.pio = system.iobus.mem_side_ports
gpu_cmd_proc.pio = system.iobus.mem_side_ports
system.pc.south_bridge.gpu.pio = system.iobus.mem_side_ports
# Create Ruby system using Ruby.py for now
Ruby.create_system(args, True, system, system.iobus,
system._dma_ports)
# Create a seperate clock domain for Ruby
system.ruby.clk_domain = SrcClockDomain(clock = args.ruby_clock,
voltage_domain = system.voltage_domain)
for (i, cpu) in enumerate(system.cpu):
#
# Tie the cpu ports to the correct ruby system ports
#
cpu.clk_domain = system.cpu_clk_domain
cpu.createThreads()
cpu.createInterruptController()
system.ruby._cpu_ports[i].connectCpuPorts(cpu)
# The shader core will be whatever is after the CPU cores are accounted for
shader_idx = args.num_cpus
system.cpu.append(shader)
gpu_port_idx = len(system.ruby._cpu_ports) \
- args.num_compute_units - args.num_sqc \
- args.num_scalar_cache
gpu_port_idx = gpu_port_idx - args.num_cp * 2
# Connect token ports. For this we need to search through the list of all
# sequencers, since the TCP coalescers will not necessarily be first. Only
# TCP coalescers use a token port for back pressure.
token_port_idx = 0
for i in range(len(system.ruby._cpu_ports)):
if isinstance(system.ruby._cpu_ports[i], VIPERCoalescer):
system.cpu[shader_idx].CUs[token_port_idx].gmTokenPort = \
system.ruby._cpu_ports[i].gmTokenPort
token_port_idx += 1
wavefront_size = args.wf_size
for i in range(args.num_compute_units):
# The pipeline issues wavefront_size number of uncoalesced requests
# in one GPU issue cycle. Hence wavefront_size mem ports.
for j in range(wavefront_size):
system.cpu[shader_idx].CUs[i].memory_port[j] = \
system.ruby._cpu_ports[gpu_port_idx].in_ports[j]
gpu_port_idx += 1
for i in range(args.num_compute_units):
if i > 0 and not i % args.cu_per_sqc:
gpu_port_idx += 1
system.cpu[shader_idx].CUs[i].sqc_port = \
system.ruby._cpu_ports[gpu_port_idx].in_ports
gpu_port_idx = gpu_port_idx + 1
for i in range(args.num_compute_units):
if i > 0 and not i % args.cu_per_scalar_cache:
gpu_port_idx += 1
system.cpu[shader_idx].CUs[i].scalar_port = \
system.ruby._cpu_ports[gpu_port_idx].in_ports
gpu_port_idx = gpu_port_idx + 1
return system
