config, mem, hmc: fix HMC test script

This patch keeps the logic behind the HMC model implementation untouched.

Additional changes:
- simple hello world script using HMC (SE simulation)

Usage examples:

./build/ARM/gem5.opt configs/example/hmctest.py
./build/ARM/gem5.opt configs/example/hmctest.py --enable-global-monitor --enable-link-monitor --arch=same
./build/ARM/gem5.opt configs/example/hmctest.py --enable-global-monitor --enable-link-monitor --arch=mixed
./build/ARM/gem5.opt configs/example/hmc_hello.py
./build/ARM/gem5.opt configs/example/hmc_hello.py --enable-global-monitor --enable-link-monitor

Change-Id: I64eb6c9abb45376b6ed72722926acddd50765394
Reviewed-on: https://gem5-review.googlesource.com/6061
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>

1 files changed, 95 insertions, 153 deletions

diff --git a/configs/example/hmctest.py b/configs/example/hmctest.py
index 3e0fa124a..9ed3ef45d 100644
--- a/configs/example/hmctest.py
+++ b/configs/example/hmctest.py
@@ -1,171 +1,113 @@
-import optparse
 import sys
+import argparse
 import subprocess
+from pprint import pprint
 
 import m5
 from m5.objects import *
-from m5.util import addToPath
+from m5.util import *
 
 addToPath('../')
 
 from common import MemConfig
 from common import HMC
 
-parser = optparse.OptionParser()
 
-# Use a HMC_2500_1x32 (1 channel, 32-bits wide) by default
-parser.add_option("--mem-type", type = "choice", default = "HMC_2500_1x32",
-                  choices = MemConfig.mem_names(),
-                  help = "type of memory to use")
-
-parser.add_option("--ranks", "-r", type = "int", default = 1,
-                  help = "Number of ranks to iterate across")
-
-parser.add_option("--rd_perc", type ="int", default=100,
-                  help = "Percentage of read commands")
-
-parser.add_option("--mode", type ="choice", default ="DRAM",
-                  choices = ["DRAM", "DRAM_ROTATE", "RANDOM"],
-                  help = "DRAM: Random traffic; \
-                          DRAM_ROTATE: Traffic rotating across banks and ranks"
-                          )
-
-parser.add_option("--addr_map", type ="int", default = 1,
-                  help = "0: RoCoRaBaCh; 1: RoRaBaCoCh/RoRaBaChCo")
-
-parser.add_option("--arch", type = "choice", default = "distributed",
-                  choices = ["same", "distributed", "mixed"],
-                  help = "same: HMC-4 links with same range\
-                  distributed: HMC-4 links with distributed range\
-                  mixed: mixed with same & distributed range")
-
-parser.add_option("--linkaggr", type = "int", default = 0,
-                  help = "1: enable link crossbar, 0: disable link crossbar")
-
-parser.add_option("--num_cross", type = "int", default = 4,
-                  help = "1: number of crossbar in HMC=1;\
-                  4: number of crossbar = 4")
-
-parser.add_option("--tlm-memory", type = "string",
-                  help="use external port for SystemC TLM cosimulation")
-
-parser.add_option("--elastic-trace-en", action ="store_true",
-                  help = """Enable capture of data dependency and instruction
-                  fetch traces using elastic trace probe.""")
-
-(options, args) = parser.parse_args()
-
-if args:
-    print "Error: script doesn't take any positional arguments"
-    sys.exit(1)
-
-system = System()
-system.clk_domain = SrcClockDomain(clock='100GHz',
-                                   voltage_domain=
-                                   VoltageDomain(voltage = '1V'))
-# Create additional crossbar for arch1
-if options.arch == "distributed" or options.arch == "mixed" :
-    system.membus = NoncoherentXBar( width=8 )
-    system.membus.badaddr_responder = BadAddr()
-    system.membus.default = Self.badaddr_responder.pio
-    system.membus.width = 8
-    system.membus.frontend_latency = 3
-    system.membus.forward_latency = 4
-    system.membus.response_latency = 2
-
-    system.membus.clk_domain = SrcClockDomain(clock='100GHz', voltage_domain=
-            VoltageDomain(voltage = '1V'))
-
-# we are considering 4GB HMC device with following parameters
+def add_options(parser):
+    parser.add_argument("--external-memory-system", default=0, action="store",
+                        type=int, help="External memory system")
+    # TLM related options, currently optional in configs/common/MemConfig.py
+    parser.add_argument("--tlm-memory", action="store_true", help="use\
+                        external port for SystemC TLM co-simulation. Default:\
+                        no")
+    # Elastic traces related options, currently optional in
+    # configs/common/MemConfig.py
+    parser.add_argument("--elastic-trace-en", action="store_true",
+                        help="enable capture of data dependency and\
+                        instruction fetch traces using elastic trace\
+                        probe.\nDefault: no")
+    # Options related to traffic generation
+    parser.add_argument("--num-tgen", default=4, action="store", type=int,
+                        choices=[4], help="number of traffic generators.\
+                        Right now this script supports only 4.\nDefault: 4")
+    parser.add_argument("--tgen-cfg-file",
+                        default="./configs/example/hmc_tgen.cfg",
+                        type=str, help="Traffic generator(s) configuration\
+                        file. Note: this script uses the same configuration\
+                        file for all traffic generators")
+
+
+# considering 4GB HMC device with following parameters
 # hmc_device_size = '4GB'
-# hmc_num_vaults = 16
 # hmc_vault_size = '256MB'
 # hmc_stack_size = 8
 # hmc_bank_in_stack = 2
 # hmc_bank_size = '16MB'
 # hmc_bank_in_vault = 16
-
-# determine the burst length in bytes
-burst_size = 256
-num_serial_links = 4
-num_vault_ctrl = 16
-options.mem_channels = 1
-options.external_memory_system = 0
-options.mem_ranks=1
-stride_size = burst_size
-system.cache_line_size = burst_size
-
-# Enable performance monitoring
-options.enable_global_monitor = True
-options.enable_link_monitor = False
-
-# Bytes used for calculations
-oneGBytes = 1024 * 1024 * 1024
-oneMBytes = 1024 * 1024
-
-# Memory ranges of 16 vault controller - Total_HMC_size / 16
-mem_range_vault = [ AddrRange(i * 256 * oneMBytes, ((i + 1) * 256 * oneMBytes)
-    - 1)
-        for i in range(num_vault_ctrl)]
-
-# Memmory ranges of serial link for arch-0
-# Same as the ranges of vault controllers - 4 vault - to - 1 serial link
-if options.arch == "same":
-    ser_range  = [ AddrRange(0, (4 * oneGBytes) - 1)
-            for i in range(num_serial_links)]
-    options.ser_ranges = ser_range
-
-# Memmory ranges of serial link for arch-1
-# Distributed range accross links
-if options.arch == "distributed":
-    ser_range  = [ AddrRange(i * oneGBytes, ((i + 1) * oneGBytes) - 1)
-            for i in range(num_serial_links)]
-    options.ser_ranges = ser_range
-
-# Memmory ranges of serial link for arch-2
-# "Mixed" address distribution over links
-if options.arch == "mixed":
-    ser_range0  = AddrRange(0              , (1 * oneGBytes) - 1)
-    ser_range1  = AddrRange(1 * oneGBytes  , (2 * oneGBytes) - 1)
-    ser_range2  = AddrRange(0              , (4 * oneGBytes) - 1)
-    ser_range3  = AddrRange(0              , (4 * oneGBytes) - 1)
-    options.ser_ranges = [ser_range0, ser_range1, ser_range2, ser_range3]
-
-# Assign ranges of vault controller to system ranges
-system.mem_ranges = mem_range_vault
-
-# open traffic generator
-cfg_file_name = "./tests/quick/se/70.tgen/traffic.cfg"
-cfg_file = open(cfg_file_name, 'r')
-
-# number of traffic generator
-np = 4
-# create a traffic generator, and point it to the file we just created
-system.tgen = [ TrafficGen(config_file = cfg_file_name) for i in xrange(np)]
-
-# Config memory system with given HMC arch
-MemConfig.config_mem(options, system)
-
-if options.arch == "distributed":
-    for i in xrange(np):
-        system.tgen[i].port = system.membus.slave
-    # connect the system port even if it is not used in this example
-    system.system_port = system.membus.slave
-
-if options.arch == "mixed":
-    for i in xrange(int(np/2)):
-        system.tgen[i].port = system.membus.slave
-    # connect the system port even if it is not used in this example
-    system.system_port = system.membus.slave
-
-
-# run Forrest, run!
-root = Root(full_system = False, system = system)
-root.system.mem_mode = 'timing'
-
-m5.instantiate()
-m5.simulate(10000000000)
-
-m5.stats.dump()
-
-print "Done!"
+def build_system(options):
+    # create the system we are going to simulate
+    system = System()
+    # use timing mode for the interaction between master-slave ports
+    system.mem_mode = 'timing'
+    # set the clock fequency of the system
+    clk = '100GHz'
+    vd = VoltageDomain(voltage='1V')
+    system.clk_domain = SrcClockDomain(clock=clk, voltage_domain=vd)
+    # add traffic generators to the system
+    system.tgen = [TrafficGen(config_file=options.tgen_cfg_file) for i in
+                   xrange(options.num_tgen)]
+    # Config memory system with given HMC arch
+    MemConfig.config_mem(options, system)
+    # Connect the traffic generatiors
+    if options.arch == "distributed":
+        for i in xrange(options.num_tgen):
+            system.tgen[i].port = system.membus.slave
+        # connect the system port even if it is not used in this example
+        system.system_port = system.membus.slave
+    if options.arch == "mixed":
+        for i in xrange(int(options.num_tgen/2)):
+            system.tgen[i].port = system.membus.slave
+        hh = system.hmc_host
+        if options.enable_global_monitor:
+            system.tgen[2].port = hh.lmonitor[2].slave
+            hh.lmonitor[2].master = hh.seriallink[2].slave
+            system.tgen[3].port = hh.lmonitor[3].slave
+            hh.lmonitor[3].master = hh.seriallink[3].slave
+        else:
+            system.tgen[2].port = hh.seriallink[2].slave
+            system.tgen[3].port = hh.seriallink[3].slave
+        # connect the system port even if it is not used in this example
+        system.system_port = system.membus.slave
+    if options.arch == "same":
+        hh = system.hmc_host
+        for i in xrange(options.num_links_controllers):
+            if options.enable_global_monitor:
+                system.tgen[i].port = hh.lmonitor[i].slave
+            else:
+                system.tgen[i].port = hh.seriallink[i].slave
+    # set up the root SimObject
+    root = Root(full_system=False, system=system)
+    return root
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Simple system using HMC as\
+                                     main memory")
+    HMC.add_options(parser)
+    add_options(parser)
+    options = parser.parse_args()
+    # build the system
+    root = build_system(options)
+    # instantiate all of the objects we've created so far
+    m5.instantiate()
+    print "Beginning simulation!"
+    event = m5.simulate(10000000000)
+    m5.stats.dump()
+    print 'Exiting @ tick %i because %s (exit code is %i)' % (m5.curTick(),
+                                                              event.getCause(),
+                                                              event.getCode())
+    print "Done"
+
+
+if __name__ == "__m5_main__":
+    main()