Difference between revisions of "Defining CPU Models stable tree v6230"

Revision as of 18:07, 11 June 2009

Overview

First, make sure you have basic understanding of how the CPU models function within the M5 framework. A good start is the CPU Models page.

This brief tutorial will show you how to create a custom CPU model called 'MyCPU', which will just be a renamed version of the AtomicSimpleCPU. After you learn how to compile and build 'MyCPU', then you have the liberty to edit the 'MyCPU' code at your heart's content without worrying about breaking any existing M5 CPU Models.

Port C++ Code for MyCPU

The easiest way is to derive a new C++ class of your CPU Model from M5 CPU Models that are already defined and the easiest model to start with is the 'AtomicSimpleCPU' located in the 'm5/src/cpu/simple' directory.

For this example, we'll just copy the files from the 'm5/src/cpu/simple' and place them in our own CPU directory: m5/src/cpu/mycpu.

me@mymachine:~/m5$ cd src/cpu 
me@mymachine:~/m5/src/cpu$ mkdir mycpu 
me@mymachine:~/m5/src/cpu$ cp -r simple/* mycpu

Now check the mycpu directory to make sure you've copied the files successfully:

me@mymachine:~/m5/src/cpu$ cd mycpu 
me@mymachine:~/m5/src/cpu/mycpu$ ls
AtomicSimpleCPU.py  SConscript  SConsopts  BaseSimpleCPU.py  TimingSimpleCPU.py  
atomic.cc  atomic.hh  base.cc  base.hh timing.cc  timing.hh

Let's move the AtomicSimpleCPU.py object file to MyCPU.py and remove the TimingSimpleCPU.py file. We'll edit MyCPU.py a little later:

me@mymachine:~/m5/src/cpu/mycpu$ mv AtomicSimpleCPU.py MyCPU.py
me@mymachine:~/m5/src/cpu/mycpu$ rm TimingSimpleCPU.py
me@mymachine:~/m5/src/cpu/mycpu$ ls
MyCPU.py  SConscript  SConsopts  atomic.cc  atomic.hh  base.cc  base.hh timing.cc  timing.hh

Since we want to change 'AtomicSimpleCPU' to 'MyCPU' we will just replace all the names in the atomic.* files and name them mycpu.* files:

me@mymachine:~/m5/src/cpu/mycpu$ perl -pe s/AtomicSimpleCPU/MyCPU/g atomic.hh > mycpu.hh
me@mymachine:~/m5/src/cpu/mycpu$ perl -pe s/AtomicSimpleCPU/MyCPU/g atomic.cc > mycpu.cc
me@mymachine:~/m5/src/cpu/mycpu$ ls
BaseSimpleCPU.py MyCPU.py SConscript SConsopts atomic.cc  atomic.hh  base.cc  base.hh  
mycpu.hh mycpu.cc timing.cc  timing.hh

The last thing you need to do is edit your mycpu.cc file to contain a reference to the 'cpu/mycpu/mycpu.hh" header file instead of the 'cpu/simple/atomic.hh' file:

#include "arch/locked_mem.hh"
...
#include "cpu/mycpu/mycpu.hh"
...

NOTE: The AtomicSimpleCPU is really just based off the BaseSimpleCPU (src/cpu/simple/base.hh) so your new CPU Model MyCPU is really a derivation off of this CPU model. Additionally, the BaseSimpleCPU model is derived from the BaseCPU (src/cpu/base.hh). As you can see, M5 is heavily object oriented.

Making M5 Recognize MyCPU

Now that you've created a separate directory and files for your MyCPU code (i.e. m5/src/cpu/mycpu), there are a couple files that need to be updated so that M5 can recognize M5 as a build option:

• m5/src/cpu/mycpu/MyCPU.py: Edit your MyCPU python script file (e.g. MyCPU.py) so that your CPU can be recognized as a simulation object. For this example, we will just use the same code tha was previously in AtomicSimpleCPU.py file but replace the name 'AtomicSimpleCPU' with 'MyCPU'.

from m5.params import *
from m5 import build_env
from BaseCPU import BaseCPU

class MyCPU(BaseCPU):
    type = 'MyCPU'
    width = Param.Int(1, "CPU width")
    simulate_stalls = Param.Bool(False, "Simulate cache stall cycles")
    function_trace = Param.Bool(False, "Enable function trace")
    function_trace_start = Param.Tick(0, "Cycle to start function trace")
    if build_env['FULL_SYSTEM']:
        profile = Param.Latency('0ns', "trace the kernel stack")
    icache_port = Port("Instruction Port")
    dcache_port = Port("Data Port")
    _mem_ports = ['icache_port', 'dcache_port']

• m5/src/cpu/mycpu/SConscript: Edit the SConscript for your CPU model and add the relevant files that need to be built in here. Your file should only contain this code:

Import('*')

if 'MyCPU' in env['CPU_MODELS']:
    Source('mycpu.cc')
    Source('base.cc')

    SimObject('MyCPU.py')

    TraceFlag('MyCPU')   

• m5/src/cpu/mycpu/SConsopts: Edit the SConscript for your CPU model and add the relevant files that need to be built in here. Your file should only contain this code:

Import('*')

all_cpu_list.append('MyCPU')
default_cpus.append('MyCPU')

• m5/src/cpu/static_inst.hh: Put a forward class declaration of your model in here

...
class CheckerCPU;
class FastCPU;
class AtomicSimpleCPU;
class TimingSimpleCPU;
class InorderCPU;
class MyCPU;
...

• m5/src/cpu/cpu_models.py: Add in CPU Model-specific information for the ISA Parser. The ISA Parser will use this when referring to the "Execution Context" for executing instructions. For instance, the AtomicSimpleCPU's instructions get all of their information from the actual CPU (since it's a 1 CPI machine). Thus, instructions only need to know the current state or "Execution Context" of the 'AtomicSimpleCPU' object. However, the instructions in a O3CPU needed to know the register values (& other state) only known to that current instruction so it's "Execution Context" is the O3DynInst object. (check out the ISA Description Language documentation page for more details)

...
CpuModel('AtomicSimpleCPU', 'atomic_simple_cpu_exec.cc',
         '#include "cpu/simple/atomic.hh"',
         { 'CPU_exec_context': 'AtomicSimpleCPU' })
CpuModel('MyCPU', 'mycpu_exec.cc',
         '#include "cpu/mycpu/mycpu.hh"',
         { 'CPU_exec_context': 'MyCPU' })
...

Building MyCPU

Navigate to the M5 top-level directory and build your model:

me@mymachine:~/m5/src/python/objects$cd ~/m5
me@mymachine:~/m5$scons build/MIPS_SE/m5.debug CPU_MODELS=MyCPU
scons: Reading SConscript files ...
Checking for C header file Python.h... (cached) yes
...
Building in /y/ksewell/research/m5-sim/newmem-clean/build/MIPS_SE
Options file /y/ksewell/research/m5-sim/newmem-clean/build/options/MIPS_SE not found,
  using defaults in build_opts/MIPS_SE
Compiling in MIPS_SE with MySQL support.
scons: done reading SConscript files.
scons: Building targets ...
make_hh(["build/MIPS_SE/base/traceflags.hh"], ["build/MIPS_SE/base/traceflags.py"])
Generating switch header build/MIPS_SE/arch/interrupts.hh
Generating switch header build/MIPS_SE/arch/isa_traits.hh
Defining FULL_SYSTEM as 0 in build/MIPS_SE/config/full_system.hh.
Generating switch header build/MIPS_SE/arch/regfile.hh
Generating switch header build/MIPS_SE/arch/types.hh
Defining NO_FAST_ALLOC as 0 in build/MIPS_SE/config/no_fast_alloc.hh.
...
g++ -o build/MIPS_SE/arch/mips/faults.do -c -pipe -fno-strict-aliasing -Wall -Wno-sign-compare -Werror -Wundef -ggdb3 -DTHE_ISA=MIPS_ISA -DDEBUG -DTRACING_ON=1 -Iext/dnet 
-I/usr/include/python2.4 -Ibuild/libelf/include -I/usr/include/mysql -Ibuild/MIPS_SE build/MIPS_SE/arch/mips/faults.cc
g++ -o build/MIPS_SE/arch/mips/isa_traits.do -c -pipe -fno-strict-aliasing -Wall -Wno-sign-compare -Werror -Wundef -ggdb3 -DTHE_ISA=MIPS_ISA -DDEBUG -DTRACING_ON=1 -Iext/dnet
 -I/usr/include/python2.4 -Ibuild/libelf/include -I/usr/include/mysql -Ibuild/MIPS_SE build/MIPS_SE/arch/mips/isa_traits.cc
g++ -o build/MIPS_SE/arch/mips/utility.do -c -pipe -fno-strict-aliasing -Wall -Wno-sign-compare -Werror -Wundef -ggdb3 -DTHE_ISA=MIPS_ISA -DDEBUG -DTRACING_ON=1 -Iext/dnet 
-I/usr/include/python2.4 -Ibuild/libelf/include -I/usr/include/mysql -Ibuild/MIPS_SE build/MIPS_SE/arch/mips/utility.cc
...
cat build/MIPS_SE/m5.debug.bin build/MIPS_SE/m5py.zip > build/MIPS_SE/m5.debug
chmod +x build/MIPS_SE/m5.debug
scons: done building targets.

If you are compiling on a dual-core CPU, use this command-line to speed-up the compilation:

scons -j2 build/MIPS_SE/m5.debug CPU_MODELS=MyCPU

Creating Configuration Scripts for MyCPU

Now that you have a M5 binary built for use with the MIPS Architecture in a M5, MyCPU Model, you are almost ready to simulate. Note that the standard M5 command line requires taht your provide at least a configuration script for the M5 binary to use.

The easiest way to get up and running is to use the sample Syscall-Emulation script: configs/example/se.py.

You'll note that line 9 of the se.py Python script imports the details of what type of Simulation will be run (e.g. what CPU Model?) from the Simulation.py file found here: m5/configs/common/Simulation.py. And then, the Simulation.py file imports it's CPU Model options from the Options.py file in the same directory. Edit those two files and your M5 binary will be ready to simulate.

• m5/configs/common/Options.py: Add a option for your model in this file...

...
parser.add_option("--my_cpu", action="store_true", help="Use MyCPU Model")
...

• m5/configs/common/Simulation.py: Edit the Simulation script to recognize your model as a CPU class. After your edits, the setCPUClass function should look like this:

...
def setCPUClass(options):
    
    atomic = False
    if options.timing:
        TmpClass = TimingSimpleCPU
    elif options.my_cpu:
	TmpClass = MyCPU
        atomic = True
    elif options.detailed:
        if not options.caches:
            print "O3 CPU must be used with caches"
            sys.exit(1)
        TmpClass = DerivO3CPU
    else:
        TmpClass = AtomicSimpleCPU
        atomic = True
        
    CPUClass = None
    test_mem_mode = 'atomic'

    if not atomic:    
        if options.checkpoint_restore:
            CPUClass = TmpClass
            TmpClass = AtomicSimpleCPU
        else:
            test_mem_mode = 'timing'

    return (TmpClass, test_mem_mode, CPUClass)
...

Testing MyCPU

Once you've edited the configuration scripts, you can run a M5 simulation from the top level directory with this command line:

me@mymachine:~/m5$build/MIPS_SE/m5.debug configs/example/se.py --my_cpu --cmd=tests/test-progs/hello/bin/mips/linux/hello

NOTE: This binary path refers to the m5/tests directory.

Summary

So the above "tutorial" showed you how to build your own CPU model in M5. Now, it's up to you to customize the CPU Model as you like and do your experiments! Good luck!