# Copyright (c) 2015 Advanced Micro Devices, Inc. # All rights reserved. # # For use for simulation and test purposes only # # Redistribution and use in source and binary forms, with or without # modification, are permitted provided that the following conditions are met: # # 1. Redistributions of source code must retain the above copyright notice, # this list of conditions and the following disclaimer. # # 2. 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. # # 3. Neither the name of the copyright holder 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 HOLDER 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: Sooraj Puthoor from __future__ import print_function import optparse, os, re import math import glob import inspect import m5 from m5.objects import * from m5.util import addToPath addToPath('../') from ruby import Ruby from common import Options from common import Simulation from common import GPUTLBOptions, GPUTLBConfig ########################## Script Options ######################## def setOption(parser, opt_str, value = 1): # check to make sure the option actually exists if not parser.has_option(opt_str): raise Exception("cannot find %s in list of possible options" % opt_str) opt = parser.get_option(opt_str) # set the value exec("parser.values.%s = %s" % (opt.dest, value)) def getOption(parser, opt_str): # check to make sure the option actually exists if not parser.has_option(opt_str): raise Exception("cannot find %s in list of possible options" % opt_str) opt = parser.get_option(opt_str) # get the value exec("return_value = parser.values.%s" % opt.dest) return return_value # Adding script options parser = optparse.OptionParser() Options.addCommonOptions(parser) Options.addSEOptions(parser) parser.add_option("--cpu-only-mode", action="store_true", default=False, help="APU mode. Used to take care of problems in "\ "Ruby.py while running APU protocols") parser.add_option("-k", "--kernel-files", help="file(s) containing GPU kernel code (colon separated)") parser.add_option("-u", "--num-compute-units", type="int", default=1, help="number of GPU compute units"), parser.add_option("--num-cp", type="int", default=0, help="Number of GPU Command Processors (CP)") parser.add_option("--benchmark-root", help="Root of benchmark directory tree") # not super important now, but to avoid putting the number 4 everywhere, make # it an option/knob parser.add_option("--cu-per-sqc", type="int", default=4, help="number of CUs" \ "sharing an SQC (icache, and thus icache TLB)") parser.add_option("--simds-per-cu", type="int", default=4, help="SIMD units" \ "per CU") parser.add_option("--wf-size", type="int", default=64, help="Wavefront size(in workitems)") parser.add_option("--sp-bypass-path-length", type="int", default=4, \ help="Number of stages of bypass path in vector ALU for Single Precision ops") parser.add_option("--dp-bypass-path-length", type="int", default=4, \ help="Number of stages of bypass path in vector ALU for Double Precision ops") # issue period per SIMD unit: number of cycles before issuing another vector parser.add_option("--issue-period", type="int", default=4, \ help="Number of cycles per vector instruction issue period") parser.add_option("--glbmem-wr-bus-width", type="int", default=32, \ help="VGPR to Coalescer (Global Memory) data bus width in bytes") parser.add_option("--glbmem-rd-bus-width", type="int", default=32, \ help="Coalescer to VGPR (Global Memory) data bus width in bytes") # Currently we only support 1 local memory pipe parser.add_option("--shr-mem-pipes-per-cu", type="int", default=1, \ help="Number of Shared Memory pipelines per CU") # Currently we only support 1 global memory pipe parser.add_option("--glb-mem-pipes-per-cu", type="int", default=1, \ help="Number of Global Memory pipelines per CU") parser.add_option("--wfs-per-simd", type="int", default=10, help="Number of " \ "WF slots per SIMD") parser.add_option("--vreg-file-size", type="int", default=2048, help="number of physical vector registers per SIMD") parser.add_option("--bw-scalor", type="int", default=0, help="bandwidth scalor for scalability analysis") parser.add_option("--CPUClock", type="string", default="2GHz", help="CPU clock") parser.add_option("--GPUClock", type="string", default="1GHz", help="GPU clock") parser.add_option("--cpu-voltage", action="store", type="string", default='1.0V', help = """CPU voltage domain""") parser.add_option("--gpu-voltage", action="store", type="string", default='1.0V', help = """CPU voltage domain""") parser.add_option("--CUExecPolicy", type="string", default="OLDEST-FIRST", help="WF exec policy (OLDEST-FIRST, ROUND-ROBIN)") parser.add_option("--xact-cas-mode", action="store_true", help="enable load_compare mode (transactional CAS)") parser.add_option("--SegFaultDebug",action="store_true", help="checks for GPU seg fault before TLB access") parser.add_option("--FunctionalTLB",action="store_true", help="Assumes TLB has no latency") parser.add_option("--LocalMemBarrier",action="store_true", help="Barrier does not wait for writethroughs to complete") parser.add_option("--countPages", action="store_true", help="Count Page Accesses and output in per-CU output files") parser.add_option("--TLB-prefetch", type="int", help = "prefetch depth for"\ "TLBs") parser.add_option("--pf-type", type="string", help="type of prefetch: "\ "PF_CU, PF_WF, PF_PHASE, PF_STRIDE") parser.add_option("--pf-stride", type="int", help="set prefetch stride") parser.add_option("--numLdsBanks", type="int", default=32, help="number of physical banks per LDS module") parser.add_option("--ldsBankConflictPenalty", type="int", default=1, help="number of cycles per LDS bank conflict") parser.add_option('--fast-forward-pseudo-op', action='store_true', help = 'fast forward using kvm until the m5_switchcpu' ' pseudo-op is encountered, then switch cpus. subsequent' ' m5_switchcpu pseudo-ops will toggle back and forth') parser.add_option('--outOfOrderDataDelivery', action='store_true', default=False, help='enable OoO data delivery in the GM' ' pipeline') Ruby.define_options(parser) #add TLB options to the parser GPUTLBOptions.tlb_options(parser) (options, args) = parser.parse_args() # The GPU cache coherence protocols only work with the backing store setOption(parser, "--access-backing-store") # if benchmark root is specified explicitly, that overrides the search path if options.benchmark_root: benchmark_path = [options.benchmark_root] else: # Set default benchmark search path to current dir benchmark_path = ['.'] ########################## Sanity Check ######################## # Currently the gpu model requires ruby if buildEnv['PROTOCOL'] == 'None': fatal("GPU model requires ruby") # Currently the gpu model requires only timing or detailed CPU if not (options.cpu_type == "TimingSimpleCPU" or options.cpu_type == "DerivO3CPU"): fatal("GPU model requires TimingSimpleCPU or DerivO3CPU") # This file can support multiple compute units assert(options.num_compute_units >= 1) # Currently, the sqc (I-Cache of GPU) is shared by # multiple compute units(CUs). The protocol works just fine # even if sqc is not shared. Overriding this option here # so that the user need not explicitly set this (assuming # sharing sqc is the common usage) n_cu = options.num_compute_units num_sqc = int(math.ceil(float(n_cu) / options.cu_per_sqc)) options.num_sqc = num_sqc # pass this to Ruby ########################## Creating the GPU system ######################## # shader is the GPU shader = Shader(n_wf = options.wfs_per_simd, clk_domain = SrcClockDomain( clock = options.GPUClock, voltage_domain = VoltageDomain( voltage = options.gpu_voltage))) # GPU_RfO(Read For Ownership) implements SC/TSO memory model. # Other GPU protocols implement release consistency at GPU side. # So, all GPU protocols other than GPU_RfO should make their writes # visible to the global memory and should read from global memory # during kernal boundary. The pipeline initiates(or do not initiate) # the acquire/release operation depending on this impl_kern_boundary_sync # flag. This flag=true means pipeline initiates a acquire/release operation # at kernel boundary. if buildEnv['PROTOCOL'] == 'GPU_RfO': shader.impl_kern_boundary_sync = False else: shader.impl_kern_boundary_sync = True # Switching off per-lane TLB by default per_lane = False if options.TLB_config == "perLane": per_lane = True # List of compute units; one GPU can have multiple compute units compute_units = [] for i in xrange(n_cu): compute_units.append(ComputeUnit(cu_id = i, perLaneTLB = per_lane, num_SIMDs = options.simds_per_cu, wfSize = options.wf_size, spbypass_pipe_length = options.sp_bypass_path_length, dpbypass_pipe_length = options.dp_bypass_path_length, issue_period = options.issue_period, coalescer_to_vrf_bus_width = \ options.glbmem_rd_bus_width, vrf_to_coalescer_bus_width = \ options.glbmem_wr_bus_width, num_global_mem_pipes = \ options.glb_mem_pipes_per_cu, num_shared_mem_pipes = \ options.shr_mem_pipes_per_cu, n_wf = options.wfs_per_simd, execPolicy = options.CUExecPolicy, xactCasMode = options.xact_cas_mode, debugSegFault = options.SegFaultDebug, functionalTLB = options.FunctionalTLB, localMemBarrier = options.LocalMemBarrier, countPages = options.countPages, localDataStore = \ LdsState(banks = options.numLdsBanks, bankConflictPenalty = \ options.ldsBankConflictPenalty), out_of_order_data_delivery = options.outOfOrderDataDelivery)) wavefronts = [] vrfs = [] for j in xrange(options.simds_per_cu): for k in xrange(shader.n_wf): wavefronts.append(Wavefront(simdId = j, wf_slot_id = k, wfSize = options.wf_size)) vrfs.append(VectorRegisterFile(simd_id=j, num_regs_per_simd=options.vreg_file_size)) compute_units[-1].wavefronts = wavefronts compute_units[-1].vector_register_file = vrfs if options.TLB_prefetch: compute_units[-1].prefetch_depth = options.TLB_prefetch compute_units[-1].prefetch_prev_type = options.pf_type # attach the LDS and the CU to the bus (actually a Bridge) compute_units[-1].ldsPort = compute_units[-1].ldsBus.slave compute_units[-1].ldsBus.master = compute_units[-1].localDataStore.cuPort # Attach compute units to GPU shader.CUs = compute_units ########################## Creating the CPU system ######################## options.num_cpus = options.num_cpus # The shader core will be whatever is after the CPU cores are accounted for shader_idx = options.num_cpus # The command processor will be whatever is after the shader is accounted for cp_idx = shader_idx + 1 cp_list = [] # List of CPUs cpu_list = [] CpuClass, mem_mode = Simulation.getCPUClass(options.cpu_type) if CpuClass == AtomicSimpleCPU: fatal("AtomicSimpleCPU is not supported") if mem_mode != 'timing': fatal("Only the timing memory mode is supported") shader.timing = True if options.fast_forward and options.fast_forward_pseudo_op: fatal("Cannot fast-forward based both on the number of instructions and" " on pseudo-ops") fast_forward = options.fast_forward or options.fast_forward_pseudo_op if fast_forward: FutureCpuClass, future_mem_mode = CpuClass, mem_mode CpuClass = X86KvmCPU mem_mode = 'atomic_noncaching' # Leave shader.timing untouched, because its value only matters at the # start of the simulation and because we require switching cpus # *before* the first kernel launch. future_cpu_list = [] # Initial CPUs to be used during fast-forwarding. for i in xrange(options.num_cpus): cpu = CpuClass(cpu_id = i, clk_domain = SrcClockDomain( clock = options.CPUClock, voltage_domain = VoltageDomain( voltage = options.cpu_voltage))) cpu_list.append(cpu) if options.fast_forward: cpu.max_insts_any_thread = int(options.fast_forward) if fast_forward: MainCpuClass = FutureCpuClass else: MainCpuClass = CpuClass # CPs to be used throughout the simulation. for i in xrange(options.num_cp): cp = MainCpuClass(cpu_id = options.num_cpus + i, clk_domain = SrcClockDomain( clock = options.CPUClock, voltage_domain = VoltageDomain( voltage = options.cpu_voltage))) cp_list.append(cp) # Main CPUs (to be used after fast-forwarding if fast-forwarding is specified). for i in xrange(options.num_cpus): cpu = MainCpuClass(cpu_id = i, clk_domain = SrcClockDomain( clock = options.CPUClock, voltage_domain = VoltageDomain( voltage = options.cpu_voltage))) if fast_forward: cpu.switched_out = True future_cpu_list.append(cpu) else: cpu_list.append(cpu) ########################## Creating the GPU dispatcher ######################## # Dispatcher dispatches work from host CPU to GPU host_cpu = cpu_list[0] dispatcher = GpuDispatcher() ########################## Create and assign the workload ######################## # Check for rel_path in elements of base_list using test, returning # the first full path that satisfies test def find_path(base_list, rel_path, test): for base in base_list: if not base: # base could be None if environment var not set continue full_path = os.path.join(base, rel_path) if test(full_path): return full_path fatal("%s not found in %s" % (rel_path, base_list)) def find_file(base_list, rel_path): return find_path(base_list, rel_path, os.path.isfile) executable = find_path(benchmark_path, options.cmd, os.path.exists) # it's common for a benchmark to be in a directory with the same # name as the executable, so we handle that automatically if os.path.isdir(executable): benchmark_path = [executable] executable = find_file(benchmark_path, options.cmd) if options.kernel_files: kernel_files = [find_file(benchmark_path, f) for f in options.kernel_files.split(':')] else: # if kernel_files is not set, see if there's a unique .asm file # in the same directory as the executable kernel_path = os.path.dirname(executable) kernel_files = glob.glob(os.path.join(kernel_path, '*.asm')) if kernel_files: print("Using GPU kernel code file(s)", ",".join(kernel_files)) else: fatal("Can't locate kernel code (.asm) in " + kernel_path) # OpenCL driver driver = ClDriver(filename="hsa", codefile=kernel_files) for cpu in cpu_list: cpu.createThreads() cpu.workload = Process(executable = executable, cmd = [options.cmd] + options.options.split(), drivers = [driver]) for cp in cp_list: cp.workload = host_cpu.workload if fast_forward: for i in xrange(len(future_cpu_list)): future_cpu_list[i].workload = cpu_list[i].workload future_cpu_list[i].createThreads() ########################## Create the overall system ######################## # List of CPUs that must be switched when moving between KVM and simulation if fast_forward: switch_cpu_list = \ [(cpu_list[i], future_cpu_list[i]) for i in xrange(options.num_cpus)] # Full list of processing cores in the system. Note that # dispatcher is also added to cpu_list although it is # not a processing element cpu_list = cpu_list + [shader] + cp_list + [dispatcher] # creating the overall system # notice the cpu list is explicitly added as a parameter to System system = System(cpu = cpu_list, mem_ranges = [AddrRange(options.mem_size)], cache_line_size = options.cacheline_size, mem_mode = mem_mode) if fast_forward: system.future_cpu = future_cpu_list system.voltage_domain = VoltageDomain(voltage = options.sys_voltage) system.clk_domain = SrcClockDomain(clock = options.sys_clock, voltage_domain = system.voltage_domain) if fast_forward: have_kvm_support = 'BaseKvmCPU' in globals() if have_kvm_support and buildEnv['TARGET_ISA'] == "x86": system.vm = KvmVM() for i in xrange(len(host_cpu.workload)): host_cpu.workload[i].useArchPT = True host_cpu.workload[i].kvmInSE = True else: fatal("KvmCPU can only be used in SE mode with x86") # configure the TLB hierarchy GPUTLBConfig.config_tlb_hierarchy(options, system, shader_idx) # create Ruby system system.piobus = IOXBar(width=32, response_latency=0, frontend_latency=0, forward_latency=0) Ruby.create_system(options, None, system) system.ruby.clk_domain = SrcClockDomain(clock = options.ruby_clock, voltage_domain = system.voltage_domain) # attach the CPU ports to Ruby for i in range(options.num_cpus): ruby_port = system.ruby._cpu_ports[i] # Create interrupt controller system.cpu[i].createInterruptController() # Connect cache port's to ruby system.cpu[i].icache_port = ruby_port.slave system.cpu[i].dcache_port = ruby_port.slave ruby_port.mem_master_port = system.piobus.slave if buildEnv['TARGET_ISA'] == "x86": system.cpu[i].interrupts[0].pio = system.piobus.master system.cpu[i].interrupts[0].int_master = system.piobus.slave system.cpu[i].interrupts[0].int_slave = system.piobus.master if fast_forward: system.cpu[i].itb.walker.port = ruby_port.slave system.cpu[i].dtb.walker.port = ruby_port.slave # attach CU ports to Ruby # Because of the peculiarities of the CP core, you may have 1 CPU but 2 # sequencers and thus 2 _cpu_ports created. Your GPUs shouldn't be # hooked up until after the CP. To make this script generic, figure out # the index as below, but note that this assumes there is one sequencer # per compute unit and one sequencer per SQC for the math to work out # correctly. gpu_port_idx = len(system.ruby._cpu_ports) \ - options.num_compute_units - options.num_sqc gpu_port_idx = gpu_port_idx - options.num_cp * 2 wavefront_size = options.wf_size for i in xrange(n_cu): # The pipeline issues wavefront_size number of uncoalesced requests # in one GPU issue cycle. Hence wavefront_size mem ports. for j in xrange(wavefront_size): system.cpu[shader_idx].CUs[i].memory_port[j] = \ system.ruby._cpu_ports[gpu_port_idx].slave[j] gpu_port_idx += 1 for i in xrange(n_cu): if i > 0 and not i % options.cu_per_sqc: print("incrementing idx on ", i) gpu_port_idx += 1 system.cpu[shader_idx].CUs[i].sqc_port = \ system.ruby._cpu_ports[gpu_port_idx].slave gpu_port_idx = gpu_port_idx + 1 # attach CP ports to Ruby for i in xrange(options.num_cp): system.cpu[cp_idx].createInterruptController() system.cpu[cp_idx].dcache_port = \ system.ruby._cpu_ports[gpu_port_idx + i * 2].slave system.cpu[cp_idx].icache_port = \ system.ruby._cpu_ports[gpu_port_idx + i * 2 + 1].slave system.cpu[cp_idx].interrupts[0].pio = system.piobus.master system.cpu[cp_idx].interrupts[0].int_master = system.piobus.slave system.cpu[cp_idx].interrupts[0].int_slave = system.piobus.master cp_idx = cp_idx + 1 # connect dispatcher to the system.piobus dispatcher.pio = system.piobus.master dispatcher.dma = system.piobus.slave ################# Connect the CPU and GPU via GPU Dispatcher ################### # CPU rings the GPU doorbell to notify a pending task # using this interface. # And GPU uses this interface to notify the CPU of task completion # The communcation happens through emulated driver. # Note this implicit setting of the cpu_pointer, shader_pointer and tlb array # parameters must be after the explicit setting of the System cpu list if fast_forward: shader.cpu_pointer = future_cpu_list[0] dispatcher.cpu = future_cpu_list[0] else: shader.cpu_pointer = host_cpu dispatcher.cpu = host_cpu dispatcher.shader_pointer = shader dispatcher.cl_driver = driver ########################## Start simulation ######################## root = Root(system=system, full_system=False) m5.ticks.setGlobalFrequency('1THz') if options.abs_max_tick: maxtick = options.abs_max_tick else: maxtick = m5.MaxTick # Benchmarks support work item annotations Simulation.setWorkCountOptions(system, options) # Checkpointing is not supported by APU model if (options.checkpoint_dir != None or options.checkpoint_restore != None): fatal("Checkpointing not supported by apu model") checkpoint_dir = None m5.instantiate(checkpoint_dir) # Map workload to this address space host_cpu.workload[0].map(0x10000000, 0x200000000, 4096) if options.fast_forward: print("Switch at instruction count: %d" % cpu_list[0].max_insts_any_thread) exit_event = m5.simulate(maxtick) if options.fast_forward: if exit_event.getCause() == "a thread reached the max instruction count": m5.switchCpus(system, switch_cpu_list) print("Switched CPUS @ tick %s" % (m5.curTick())) m5.stats.reset() exit_event = m5.simulate(maxtick - m5.curTick()) elif options.fast_forward_pseudo_op: while exit_event.getCause() == "switchcpu": # If we are switching *to* kvm, then the current stats are meaningful # Note that we don't do any warmup by default if type(switch_cpu_list[0][0]) == FutureCpuClass: print("Dumping stats...") m5.stats.dump() m5.switchCpus(system, switch_cpu_list) print("Switched CPUS @ tick %s" % (m5.curTick())) m5.stats.reset() # This lets us switch back and forth without keeping a counter switch_cpu_list = [(x[1], x[0]) for x in switch_cpu_list] exit_event = m5.simulate(maxtick - m5.curTick()) print("Ticks:", m5.curTick()) print('Exiting because ', exit_event.getCause()) sys.exit(exit_event.getCode())