Updated for SPEC MPI2007 (new features are highlighted)
Last updated: $Date: 2009-12-01 09:38:49 -0500 (Tue, 01 Dec 2009) $ by $Author: keeper $
(To check for possible updates to this document, please
see http://www.spec.org/mpi2007/Docs/ )
Contents
1 Introduction
1.1 Who Needs runspec?
1.2 About Config Files
1.2.1 Finding a config file
1.2.2 Naming your config file
1.2.3 If you change only one thing...
1.3 About Defaults
1.4 About Disk Usage and Support for Multiple Users
1.4.1 Directory tree
1.4.2 Hey! Where did all my disk space go?
1.5 Multi-user support
1.5.1 Recommended sharing method: output_root
1.5.2 Alternative sharing methods
2 Before Using runspec
2.1 Install kit
2.2 Have a config file
2.3 Undefine SPEC
2.4 Set your path: Unix
2.5 Set your path: Windows
2.6 Check your disk space
3 Using runspec
3.1 Simplest usage
3.1.1 Reportable run
3.1.2 Running selected benchmarks
3.1.3 Output files
3.2 Syntax
3.2.1 Benchmark names in run lists
3.2.2 Run order for reportable runs
3.2.3 Run order when more than one tuning is present
3.3 Actions
3.4 Commonly used options
--action --check_version --config --flagsurl --help --ignore_errors --iterations --loose --output_format --ranks --rawformat --rebuild --reportable --tune
3.5 Less commonly used options
--basepeak --nobuild --comment --define --delay --deletework --extension --fake --fakereport --fakereportable --[no]feedback --[no]graph_auto --graph_max --graph_min --http_proxy --http_timeout --info_wrap_column --keeptmp --log_timestamp --machine --make_no_clobber --make_bundle --maxcompares --notes_wrap_column --preenv --reportonly --review --[no]setprocgroup --size --test --[no]table --undef --update --update_flags --unpack_bundle --use_bundle --username --verbose --version
4 Quick reference
Everyone who uses SPEC MPI2007 needs runspec. It is the primary tool in the suite. It is used to build the benchmarks, run them, and report on their results. All users of MPI2007 should read this document.
If you are a beginner, please start out by reading from the beginning through section 3.1 Simplest Usage. That will probably be enough to get you started.
In order to use runspec, you need a "config file", which contains detailed instructions about how to build and run the benchmarks. You may not have to learn much about config files in order to get started. Typically, you start off using a config file that someone else has previously written.
Where can you find such a config file? There are various sources:
Look in the directory $SPEC/config/ (Unix) or %SPEC%\config\ (Windows). You may find that there is a already a config file there with a name that indicates that it is appropriate for your system. You may even find that default.cfg already contains settings that would be a good starting place for your system.
Look at the SPEC web site (http://www.spec.org/mpi2007/) for a MPI2007 result submission that used your system, or a system similar to yours. You can download the config file from that submission.
Alternatively, you can write your own, using the instructions in config.html
Note: links to SPEC MPI2007 documents on this web page assume that you are reading the page from a directory that also contains the other SPEC MPI2007 documents. If by some chance you are reading this web page from a location where the links do not work, try accessing the referenced documents at one of the following locations:
Once you have found a config file that you would like to use as a starting point, you will probably find it convenient to copy it and modify it according to your needs. There are various options:
You can copy the config file to default.cfg. Doing so means that you won't even need to mention --config on your runspec command line.
You might find it useful to name config files after the date and the test attempt: jan07a.cfg, jan07b.cfg, and so forth. This is alleged to make it easier to trace the history of an experiment set.
If you are sharing a testbed with other users, it is probably wise to name the config file after yourself. For example, if Yusuf is trying out the new Solaris Fortran95 compiler, he might say:
and edit the new config file to add whatever options he wishes to try out in the new compiler.
At first, you may hesitate to change settings in config files, until you have a chance to read config.html. But there is one thing that you might want to change right away. Look for a line that says:
That line determines what extension will be added to your binaries. If there are comments next to that line giving instructions ("# set ext to A for this, or to B for that"), then set it accordingly. But if there are no such instructions, then usually you are free to set the extension to whatever you like, which can be very useful to ensure that your binaries are not accidentally over-written. You might add your name in the extension, if you are sharing a testbed with others. Or, you may find it convenient to keep binaries for a series of experiments, to facilitate later analysis; if you're naming your config files with names such as jan07a.cfg, you might choose to use "ext=jan07a" in the config file.
The SPEC tools have followed two principles regarding defaults:
This means (the good news) that something sensible will usually happen, even when you are not explicit about what you want. But it also means (the bad news) that if something unexpected happens, you may have to look in several places in order to figure out why it behaves differently than you expect.
The order of precedence for settings is:
| Highest precedence: | runspec command |
| Middle: | config file |
| Lowest: | the tools as shipped by SPEC |
Therefore, when this document tells you that something is the default, bear in mind that your config file may have changed that setting. With luck, the author of the config file will tell you so (perhaps in the comments to the config file).
The structure of the MPI2007 directory tree is:
$SPEC or %SPEC% - the root directory
benchspec - some suite-wide files
MPI2007 - the benchmarks
bin - tools to run and report on the suite
config - config files
result - log files and reports
tools - sources for the benchmark suite tools
Within each of the individual benchmarks, the structure is:
nnn.benchmark - root for this benchmark
Spec - SPEC metadata about the benchmark
build - all builds take place here
data
all - data used by all runs (if needed by the benchmark)
mref - the real data set, required for all medium result reporting
mtest - data for a simple test that an executable is functional
mtrain - another set of data to test that an executable is functional
lref - the real data set, required for all large result reporting
ltest - data for a simple test that an executable is functional
ltrain - another set of data to test that an executable is functional
commontest - shared data files used for ltest and mtest
commontrain - shared data files used for ltrain and mtrain
exe - compiled versions of the benchmark
run - all runs take place here
src - the sources for the benchmark
When you find yourself wondering "Where did all my disk space go?", the answer is "The run directories." Most (*) activity takes place in automatically created subdirectories of $SPEC/benchspec/MPI2007/*/run/ (Unix) or %SPEC%\benchspec\MPI2007\*\run\ (Windows).
For example, suppose Bob has a config file that he is using to test some new memory optimizations, and has set
in his config file. In that case, the tools would create directories such as these:
$ pwd /Users/bob/mpi2007/benchspec/MPI2007/121.pop2/run $ ls list run_base_mtest_BobMemoryOpt.0001 run_base_mtrain_BobMemoryOpt.0001 run_base_mref_BobMemoryOpt.0001 run_peak_mtest_BobMemoryOpt.0001 run_peak_mtrain_BobMemoryOpt.0001 run_peak_mref_BobMemoryOpt.0001 $
To get your disk space back, see the documentation of the various cleaning options, below; or issue a command such as the following (on Unix systems; Windows users can select the files with Explorer):
rm -Rf $SPEC/benchspec/MPI2007/*/run/run*BobMemory*
The effect of the above command would be to delete all the run directories associated with the benchmarks which used extension *BobMemory*. Note that the command did not delete the directories where the benchmarks were built (...MPI2007/*/build/*); sometimes it can come in handy to keep the build directories, perhaps to assist with debugging.
(*) Other space: In addition to the run directories, other consumers of disk space include: (1) temporary files; for a listing of these, see the documentation of keeptmp; and (2) the build directories. For the example above, underneath:
/Users/bob/mpi2007/benchspec/MPI2007/121.pop2/build/
will be found:
$ ls build_base_BobMemoryOpt.0001 build_peak_BobMemoryOpt.0001
For MPI2007 the directory names includes the string "build" or "run" followed by the extension. In addition, new for MPI2007 1.1 is the location of the build directories; if you prefer the old location, see build_in_build_dir.
(If you are not sharing a SPEC MPI2007 installation with other users, you can skip ahead to section 2.)
The SPEC MPI2007 toolset provides support for multiple users of a single installation, but the tools also rely upon users to make appropriate choices regarding setup of operating-system file protections. This section describes the multi-user features and describes ways of organizing protections. First, to the features that are always enabled:
The SPEC-distributed source directories and data directories are not changed during testing. Instead, working directories are created as needed for builds and runs.
Each user's build and run directories are tagged with the name of the user that they belong to (in the file nnn.benchmark/run/list). Directories created for one user are not re-used for a different user.
Multiple users can run tests at the same time. (Of course, if the jobs compete with each other for resources, it is likely that they will run more slowly.)
Multiple users can even run the "same" test at the same time, and they will automatically be given separate run directories.
If you have more than one user of SPEC MPI2007, you can use additional features and choose from several different ways to organize the on-disk layout to share usage of the product. The recommended way is described first.
The recommended method for sharing a SPEC MPI2007 installation among multiple users has 4 steps:
| Step | Example (Unix) |
| Protect most of the SPEC tree read-only | chmod -R ugo-w $SPEC |
| Allow shared access to the config directory | chmod 1777 $SPEC/config |
| Keep your own config files | cp config/assignment1.cfg config/alan1.cfg |
| Add an output_root to your config file | output_root=/home/${username}/spec |
More detail about the steps is below.
Most of the MPI2007 tree is shared, and can be protected read-only. For example, on a Unix system, you might set protections with:
chmod -R ugo-w $SPEC
The one exception is the config directory, $SPEC/config/ (Unix) or %SPEC%\config\ (Windows), which needs to be a read/write directory shared by all the users. It is written to by users when they create config files, and by the tools themselves: config files are updated after successful builds to associate them with their binaries.
On Unix, the above protection command needs to be supplemented with:
chmod 1777 $SPEC/config
which will have the effect (on most Unix systems) of allowing users to create config files which they can choose to protect to allow access only by themselves.
Config files usually would not be shared between users. For example, students might create their own copies of a config file.
Alan enters:
cd /cs403 . ./shrc cp config/assignment1.cfg config/alan1.cfg chmod u+w config/alan1.cfg runspec --config alan1 --action build 127.wrf2
Wendy enters:
cd /cs403 . ./shrc cp config/assignment1.cfg config/wendy1.cfg chmod u+w config/wendy1.cfg runspec --config wendy1 --action build 127.wrf2
Set output_root in the config files to change the destinations of the outputs.
To see the effect of output_root, consider an example with and without the feature. If $SPEC is set to /cs403 and if ext=feb27a, then normally the build directory for 127.wrf2 with base tuning would be:
But if the config files include (near the top, before any occurrence of a section marker):
output_root=/home/${username}/spec
ext=feb27a
then Alan's build directory for 127.wrf2 will be
and Wendy's will be
With the above setting of output_root, log files and reports that would normally go to /cs403/result instead will go to /home/alan/spec/result and /home/wendy/spec/result. Alan will find hmmer executables underneath /home/alan/spec/benchspec/MPI2007/127.wrf2/exe. And so forth.
Summary: output_root is the recommended way to separate users. Set the protection on the original tree to read-only, except for the config directory, which should be set to allow users to write, and protect, their own config files.
An alternative is to keep all the files in a single directory tree. In this case:
The directory tree must be writable by each of the users, which means that they have to trust each other not to modify or delete each others' files.
Directories such as result, nnn.benchmark/exe and nnn.benchmark/run are not segregated by user, so you can only have one version of (for example) benchspec/MPI2007/126.lammps/exe/lammps_base.jan07a
Note that user names do not appear in the directory names. For example, if Lizy, Aashish, and Ajay are sharing a directory tree on a Windows system, and each of them runs the ref workload for 115.fds4 with base tuning and a config file that sets ext=wwc9, there will be three directories created:
To discover which 115.fds4 run directories belong to Lizy:
F:\> cd %SPEC%\benchspec\MPI2007\115.fds4\run
F:\mpi2007\benchspec\MPI2007\115.fds4\run> findstr lizy list
To discover which result files belong to Aashish:
F:\mpi2007> cd %SPEC%\result
F:\mpi2007\result> findstr aashish *log
(Of course, on Unix, that would be grep instead of findstr).
Name convention: Users sharing a tree can adopt conventions to make their files more readily identifiable. As mentioned above, you can set your config file name to match your own name, and do the same for the extension.
Expid convention: Another alternative is to tag directories with labels that help to identify them based on an "experiment ID", with the config file feature expid, as described in config.html.
Spend the disk space: A final alternative, of course, is to not share. You can simply give each user their own copy of the entire SPEC MPI2007 directory tree. This may be the easiest way to ensure that there are no surprises (at the expense of extra disk space.)
Before using runspec, you need to:
The runspec tool uses perl version 5.8.7, which is installed as specperl when you install MPI2007. If you haven't already installed the suite, then please see system-requirements.html followed by:
You won't get far unless you have a config file, but fortunately you can get started by using a pre-existing config file. See About Config Files, above.
If the environment variable SPEC is already defined (e.g. from a run of some other SPEC benchmark suite), it may be wise to undefine it first, e.g. by logging out and logging in, or by using whatever commands your system uses for removing definitions (such as unset).
To check whether the variable is already defined, type
echo $SPEC (Unix) or
echo %SPEC% (Windows)
On Unix systems, the desired output is nothing at all; on Windows systems, the desired output is %SPEC%.
Similarly, if your PATH includes tools from some other SPEC suite, it may be wise to remove them from your path.
Next, you need to
set your path appropriately for your system type.
See section 2.4 for Unix
or section 2.5 for Windows.
If you are using a Unix system, change your current directory to the top-level SPEC directory and source either shrc or cshrc:
Note that it is, in general, a good idea to ensure that you understand what is in your path, and that you have only what you truly need. If you have non-standard versions of commonly used utilities in your path, you may avoid unpleasant surprises by taking them out.
If you are using a Microsoft Windows system, start a Command Prompt Window (previously known as an "MSDOS window"). Change to the directory where you have installed MPI2007, then edit shrc.bat, following the instructions contained therein. For example:
C:\> f:
F:\> cd diego\mpi2007
F:\diego\mpi2007\> copy shrc.bat shrc.bat.orig
F:\diego\mpi2007\> notepad shrc.bat
and follow the instructions in shrc.bat to make the appropriate edits for your compiler paths.
Caution: you may find that the lines are not correctly formatted (the text appears to be all run together) when you edit this file. If so, see the section "Using Text Files on Windows" in the Windows installation guide.
You will have to uncomment one of two lines:
rem set SHRC_COMPILER_PATH_SET=yes
or
rem set SHRC_PRECOMPILED=yes
by removing "rem" from the beginning of the desired line.
If you uncomment the first line, you will have to follow the instructions a few lines further on, to set up the environment for your compiler.
If you uncomment the second line, you must have pre-compiled binaries for the benchmarks
Note that it is, in general, a good idea to ensure that you understand what is in your path, and that you have only what you truly need. If you have non-standard versions of commonly used utilities in your path, you may avoid unpleasant surprises by taking them out. In order to help you understand your path, shrc.bat will print it after it is done.
When you are done, set the path using your edited shrc.bat, for example:
F:\diego\mpi2007> shrc
Presumably, you checked to be sure you had enough space when you read system-requirements.html, but now might be a good time to double check that you still have enough. Typically, you will want to have at least 20 GB free disk space at the start of a run. Windows users can say "dir", and will find the free space at the bottom of the directory listing. Unix users can say "df -k ." to get a measure of freespace in KB.
If you have done some runs, and you are wondering where your space has gone, see section 1.4.2.
It is easiest to use runspec when:
Some kind person has already compiled the benchmarks.
That kind person provides both the compiled images and their corresponding config file (see About Config Files above).
The config file does not change the defaults in surprising or esoteric ways (see About Defaults above).
In this lucky circumstance, all that needs to be done is to name the config file, select which benchmark suite is to be run: medium for the SPECmpiM2007 benchmarks or large for the SPECmpiL2007 benchmarks, and say --reportable to attempt a full run.
For example, suppose that Wilfried wants to give Ryan a config file and compiled binaries with some new optimizations for a Unix system. Wilfried might type something like this:
[/usr/wilfried]$ cd $SPEC [/bigdisk/mpi2007]$ spectar -cvf - be*/C*/*/exe/*newint* config/newint.cfg | specbzip2 > newint.tar.bz2
and then Ryan might type something like this:
ryan% cd /usr/ryan/mpi2007 mpi2007% bash bash-2.05$ . ./shrc bash-2.05$ specbzip2 -dc newint.tar.bz2 | spectar -xf - bash-2.05$ runspec --config newint.cfg --nobuild --ranks=32 --reportable medium
In the example above, the --nobuild emphasizes that the tools should not attempt to build the binaries; instead, the prebuilt binaries should be used. If there is some reason why the tools don't like that idea (for example: the config file does not match the binaries), they will complain and refuse to run, but with --nobuild they won't go off and try to do a build.
As a another example, suppose that Reinhold has given Kaivalya a Windows config file with changes from 12 August, and Kaivalya wants to run the medium suite. He might say something like this:
F:\kaivalya\mpi2007\> shrc F:\kaivalya\mpi2007\> specbzip2 -dc reinhold_aug12a.tar.bz2 | spectar -xf - F:\kaivalya\mpi2007\> runspec --config reinhold_aug12a --reportable --ranks=64 medium
If you want to run a subset of the benchmarks, rather than running the whole suite, you can name them. Since a reportable run uses an entire suite, you will need to turn off reportable:
[/usr/mat/mpi2007]$ runspec --config mat_dec25j.cfg --noreportable 113.GemsFDTD
Look for the output of your runspec command in the directory $SPEC/result (Unix) or %SPEC%\result (Windows). There, you will find log files and result files. More information about log files can be found in config.html.
The format of the result files depends on what was selected in your config file, but will typically include at least .txt for ASCII text, and will always include .rsf, for raw (unformatted) run data. More information about result formats can be found below, under --output_format. Note that you can always re-generate the output, using the --rawformat option, also documented below.
This concludes the section on simplest usage.
If simple commands such as the above are not enough to meet your needs,
you can find out about commonly used options by continuing to read the next 3 sections (3.2, 3.3, and 3.4).
The syntax for the runspec command is:
runspec [options] [list of benchmarks to run]
Options are described in the following sections. There, you will notice that many options have both long and short names. The long names are invoked using two dashes, and the short names use only a single dash. For long names that take a parameter, you can optionally use an equals sign. Long names can also be abbreviated, provided that you still enter enough letters for uniqueness. For example, the following commands all do the same thing:
runspec --config=dianne_july25a --debug=99 large runspec --config dianne_july25a --debug 99 large runspec --conf dianne_july25a --deb 99 large runspec -c dianne_july25a -v 99 large
The list of benchmarks to run can be:
Individual benchmarks can be named, numbered, or both; and they can be abbreviated, as long as you enter enough characters for uniqueness. For example, each of the following commands does the same thing:
runspec -c jason_july09d --noreportable 113.GemsFDTD 121.pop2 runspec -c jason_july09d --noreportable 113 121 runspec -c jason_july09d --noreportable GemsFDTD pop2 runspec -c jason_july09d --noreportable Gem po
It is also possible to exclude a benchmark, using a hat (^, also known as carat, typically found as shift-6). For example, suppose your system lacks a C++ compiler, and you therefore cannot run the benchmark 126.lammps. You could run all of the medium benchmarks except this one by entering a command such as this one:
runspec --noreportable -c kathy_sep14c medium ^lammps
Note that if hat has significance to your shell, you may need to protect it from interpretation by the shell, for example by putting it in single quotes. On Windows, you will need to use both a hat and double quotes for each benchmark you want to exclude, like this:
E:\mpi2007> runspec --noreportable -c cathy_apr21b medium "^lammps"
A reportable run runs all the benchmarks in a suite. The reference workloads are run at least two times. For odd number of runs, the medium runtime is used, and for even number of runs, the lesser of the two middle runtimes is used. For example, here are the runs for a reportable medium run of MPI2007 with 3 iterations:
$ grep runspec: *036.log runspec: runspec -c mar26a -T base -n 3 --reportable medium $ grep Running *036.log Running Benchmarks Running (#1) 104.milc mref (ref) base mar26a default Running (#1) 107.leslie3d mref (ref) base mar26a default Running (#1) 113.GemsFDTD mref (ref) base mar26a default Running (#1) 115.fds4 mref (ref) base mar26a default Running (#1) 121.pop2 mref (ref) base mar26a default Running (#1) 122.tachyon mref (ref) base mar26a default Running (#1) 126.lammps mref (ref) base mar26a default Running (#1) 127.wrf2 mref (ref) base mar26a default Running (#1) 128.GAPgeofem mref (ref) base mar26a default Running (#1) 129.tera_tf mref (ref) base mar26a default Running (#1) 130.socorro mref (ref) base mar26a default Running (#1) 132.zeusmp2 mref (ref) base mar26a default Running (#1) 137.lu mref (ref) base mar26a default Running (#2) 104.milc mref (ref) base mar26a default Running (#2) 107.leslie3d mref (ref) base mar26a default Running (#2) 113.GemsFDTD mref (ref) base mar26a default Running (#2) 115.fds4 mref (ref) base mar26a default Running (#2) 121.pop2 mref (ref) base mar26a default Running (#2) 122.tachyon mref (ref) base mar26a default Running (#2) 126.lammps mref (ref) base mar26a default Running (#2) 127.wrf2 mref (ref) base mar26a default Running (#2) 128.GAPgeofem mref (ref) base mar26a default Running (#2) 129.tera_tf mref (ref) base mar26a default Running (#2) 130.socorro mref (ref) base mar26a default Running (#2) 132.zeusmp2 mref (ref) base mar26a default Running (#2) 137.lu mref (ref) base mar26a default Running (#3) 104.milc mref (ref) base mar26a default Running (#3) 107.leslie3d mref (ref) base mar26a default Running (#3) 113.GemsFDTD mref (ref) base mar26a default Running (#3) 115.fds4 mref (ref) base mar26a default Running (#3) 121.pop2 mref (ref) base mar26a default Running (#3) 122.tachyon mref (ref) base mar26a default Running (#3) 126.lammps mref (ref) base mar26a default Running (#3) 127.wrf2 mref (ref) base mar26a default Running (#3) 128.GAPgeofem mref (ref) base mar26a default Running (#3) 129.tera_tf mref (ref) base mar26a default Running (#3) 130.socorro mref (ref) base mar26a default Running (#3) 132.zeusmp2 mref (ref) base mar26a default Running (#3) 137.lu mref (ref) base mar26a default
The above order can be summarized as:
mref1, mref2, mref3
Sometimes, it can be useful to understand when directory setup occurs. So, let's expand the list to include setup:
setup for mref mref1, mref2, mref3
If you run both base and peak tuning, base is always run first. If you do a reportable run with both base and peak, the order is:
setup for mref base mref1, base mref2, base mref3 peak mref1, peak mref2, peak mref3
When runspec is used, it normally (*) takes some kind of action for the set of benchmarks specified at the end of the command line (or defaulted from the config file). The default action is validate, which means that the benchmarks will be built if necessary, the run directories will be set up, the benchmarks will be run, and reports will be generated.
(*) Exception: if you use the --rawformat switch, then --action is ignored.
If you want to cause a different action, then you can enter one of the following runspec options:
| --action build | Compile the benchmarks. More information about compiling may be found in config.html, including information about additional files that are output during a build. |
| --action buildsetup | Set up build directories for the benchmarks, but do not attempt to compile them. |
| --action configpp | Preprocess the config file and dump it to stdout |
| --action onlyrun | Run the benchmarks but do not bother to verify that they got the correct answers. Reports are always marked "invalid", since the correctness checks are skipped. Therefore, this option is rarely useful, but it can be selected if, for example, you are generating a performance trace and wish to avoid tracing some of the tools overhead. |
| --action report | Synonym for --fakereport; see also --fakereportable. |
| --action run | Synonym for --action validate. in an attempt to better match what users expect "run" to do.) |
| --action runsetup | Synonym for --action setup |
| --action setup | Set up the run directories. Copy executables and data to work directories. |
| --action validate | Build (if needed), run, check for correct answers, and generate reports. |
In addition, the following cleanup actions are available (in order by level of vigor):
| --action clean | Empty all run and build directories for the specified benchmark set for the
current user. For example, if the current OS username is set to jeff and this
command is entered:
D:\mpi2007\> runspec --action clean --config may12a medium
then the tools will remove run directories with username jeff for medium
benchmarks generated by config file may12a.cfg (in nnn.benchmark\run and nnn.benchmark\build).
|
| --action clobber | Clean + remove all executables of the current type for the specified benchmark set. |
| --action trash | Same as clean, but do it for all users of this SPEC directory tree, and all types, regardless of what's in the config file. |
| --action realclean | A synonym for --action trash |
| --action scrub | Remove everybody's run and build directories and all executables for the specified benchmark set. |
Alternative cleaning method:
If you prefer, you can clean disk space by entering commands such as the following (on Unix systems):
rm -Rf $SPEC/benchspec/M*/*/run
rm -Rf $SPEC/benchspec/M*/*/exe
Notes
The above commands not only empty the contents of the run and exe directories; they also delete the directories themselves. That's fine; the tools will re-create the run and exe directories if they are needed again later on.
The above commands do NOT clean the build directories (unless you've set build_in_build_dir=0). Often, it's useful to preserve the build directories for debugging purposes, but if you'd like to get rid of them too, just add $SPEC/benchspec/M*/*/build to your list of directories.
Windows users:
Windows users can achieve a similar effect using Windows Explorer.
I have so much disk space, I'll never use all of it:
Run directories are automatically re-used for subsequent runs. If you prefer, you can ask the tools to never touch a used run directory. Do this by setting the environment variable:
SPEC_MPI2007_NO_RUNDIR_DEL
In this case, you should be prepared to do frequent cleaning, perhaps after reviewing the results of each run.
Most users of runspec will want to become familiar with the following options.
runspec --check_version --http_proxy http://webcache.tom.spokewrenchdad.com:8080or, equivalently, for those who prefer to abbreviate to the shortest possible amount of typing:
runspec --ch --http_p http://webcache.tom.spokewrenchdad.com:8080The command downloads a small file (~15 bytes) from www.spec.org which contains information about the most recent release, and compares that to your release. If your version is out of date, a warning will be printed.
Meaning: A "flags file" provides information about how to interpret and report on the flags (e.g. -O5, -fast, etc.) that are used in a config file.
The --flagsurl switch says that a flags file may be found at the specified URL (such as http://myflags.com/flags.xml). URL schemes supported are http, ftp, and file. A URL without a scheme is assumed to be a file or path name. If you need to specify an http proxy, you can do so in your config file, by using the --http_proxy command line switch, or via the environment variable http_proxy.
The special value noflags may be used to cause rawformat to remove a stored flags file when re-formatting a previously run result.
Flags files are required by run rule 4.2.5. If a run is marked "invalid" because some flags are "unknown", you may be able to resolve the invalid marking by finding, or creating, a flags file with proper descriptions and entering commands such as:
cp MPIM2007.559.mref.rsf retry rawformat --flagsurl myfixedflags.xml --output_format pdf,raw retry
The first command preserves the original raw file, which is always recommended before doing any operations that create a new raw file. The second command creates retry.rsf and retry.pdf, both of which will include descriptions of flags from myfixedflags.xml. If you are submitting a result to SPEC, the newly-generated rawfile is the one to submit.
Note that saying rawformat is equivalent to saying runspec --rawformat, as described below.
On Windows systems, the first command above would use copy instead of cp. Also, if Windows refuses to accept the syntax with a comma in it, you might have to generate just the rawfile as a first step, then generate other format(s).
You can find out more about how to write flag description files in flag-description.html. New with MPI2007 V1.1, included there is a complete example of a flags file update using rawformat --flagsurl.
New with MPI2007 V1.1, you can now format a single result using multiple flags files. This feature is intended to make it easier for multiple results to share what should be shared, while separating what should be separated. Common elements (such as a certain version of a compiler) can be placed into one flags file, while the elements that differ from one system to another (such as platform notes) can be maintained separately.
[/usr/mwong/mpi2007]$ runspec --config golden --iterations 1 128.GAPgeofemas the SPEC tools will inform you that you cannot change the number of iterations on a reportable run. But either of the following commands will override the config file and just run 128.GAPgeofem once:
[/usr/mwong/mpi2007]$ runspec --config golden --iterations 1 --loose 128.GAPgeofem [/usr/mwong/mpi2007]$ runspec --config golden --iterations 1 --noreportable 128.GAPgeofem
| all | implies all of the following except screen, check, and mail |
|---|---|
| cfg|config|conffile configfile|cfgfile |
config file used for this run (e.g. MPIM2007.030.mref.cfg) |
| check|chk|sub| subcheck|subtest|test |
Submission syntax check (automatically enabled for reportable runs). Causes many fields to be checked for acceptable
formatting - e.g. hardware available "Nov-2007", not "11/07";
memory size "4 GB", not "4Gb"; and so forth. SPEC enforces
consistent syntax for results submitted to its website as an aid to readers of results, and to increase the likelihood
that queries find the intended results. If you select --output_format subcheck on your local
system, you can find out about most formatting problems before you submit your results to SPEC. Even if you don't plan
to submit your results to SPEC, the Submission Check format can help you create reports that are more complete and readable.
New with MPI2007 V1.1, Submission Check is automatically enabled when doing rawformat. |
| csv|spreadsheet | Comma-separated variable (e.g. MPIM2006.030.mref.csv). If you populate spreadsheets from your runs, you probably shouldn't be doing cut/paste of text files; you'll get more accurate data by using --output_format csv. In addition, the format has been updated with MPI2007 V1.1: CSV output now includes much more of the information in the other reports. All runs times are now included, and the selected run times are listed separately. The flags used are also included. Although details of the new features are not shown in the documentation, it is hoped that you will find the V1.1 format more complete and more useful. |
| default | implies HTML and text |
| flag|flags | Flag report (e.g. MPIM2007.030.flags.mref.html). Will also be produced when formats that use it are requested (PDF, HTML). |
| html|xhtml|www|web | web page (e.g. MPIM2007.030.mref.html) |
| mail|mailto|email | All generated reports will be sent to an address specified in the config file. |
| pdf|adobe | Portable Document Format (e.g. MPIM2007.030.mref.pdf). This format is the design center for SPEC MPI2007 reporting. Other formats contain less information: text lacks graphs, postscript lacks hyperlinks, and HTML is less structured. (It does not appear as part of "default" only because some systems may lack the ability to read PDF.) |
| postscript|ps| printer|print |
PostScript (e.g. MPIM2007.030.mref.ps) |
| raw|rsf | raw results, e.g. MPIM2007.030.mref.rsf. Note: you will automatically get an rsf file for commands that run a test or that update a result (such as rawformat --flagsurl). |
| screen|scr|disp| display|terminal|term |
ASCII text output to stdout. |
| text|txt|ASCII|asc | ASCII text, e.g. MPIM2007.030.mref.txt. |
Meaning: Provide the number of ranks to be used for the run. For example, the following command would do a 32-rank SPECmpiM2007 run:
/bigdisk/mpi2007$ runspec --config tony_may12a --rank 4 medium
Meaning: Do not attempt to do a run; instead, take an existing result file and just generate the reports. Using this option will cause any specified --actions to be ignored, and instead the result formatter will be invoked. This option is useful if (for example) you are just doing ASCII output during most of your runs, but now you would like to create additional reports for one or more especially interesting runs. To create the html and postscript files for experiment number 324, you could say:
runspec --rawformat --output_format html,ps $SPEC/result/MPIM2007.324.mref.rsf
You can achieve the same effect by invoking rawformat directly:
rawformat --output_format html,ps $SPEC/result/MPIM2007.324.mref.rsf
These two commands achieve the same effect because, in fact, saying runspec --rawformat just causes runspec to exit, invoking rawformat in its stead, and passing it whatever was on the command line - in this case, the --output_format html,ps string.
Note that when runnng rawformat, you will always get format "Submission Check", which encourages consistent formatting for various result fields when preparing final (submittable) reports. In addition, you will get the formats that you mention on the command line, or, if none are mentioned there, then you will get the defaults documented under output_format. The automatic addition of subcheck to the list of outputs is new with MPI2007 V1.1
For more information about rawformat, please see utility.html.
Meaning: Do not build binaries, even if they don't exist or MD5 sums don't match. This feature can be very handy if, for example, you have a long script with multiple invocations of runspec, and you would like to ensure that the build is only attempted once. (Perhaps your thought process might be, "If it fails the first time, fine, just forget about it until I come in Monday and look things over.") By adding --nobuild --ignore_errors to all runs after the first one, no attempt will be made to build the failed benchmarks after the first attempt.
The --nobuild feature also comes in handy when testing whether proposed config file options would potentially force an automatic rebuild.
Meaning: Defines a preprocessor macro named SYMBOL (for use in your config file) and optionally gives it the value VALUE. If no value is specified, the macro is defined with no value. SYMBOL may not contain equals signs ("=") or colons (":"). This option may be used multiple times. For example if a config file says:
%ifdef %{use_sparc_v9}
ext = darryl.native64
mach = native64
ARCH_SELECT = -xtarget=native64
%else
ext = darryl.native32
mach = default
ARCH_SELECT = -xtarget=native
%endif
default=base:
OPTIMIZE = -O ${ARCH_SELECT}
Then saying runspec --define use_sparc_v9=1 will cause base optimization to be -O -xtarget=native64
Meaning: In some cases, such as when doing version checks and loading flag description files, runspec will attempt to fetch a file, using http. If your web browser needs a proxy server in order to access the outside world, then runspec will probably want to use the same proxy server. The proxy server can be set by:
For example, a failure of this form:
$ runspec --rawformat --output_format txt \
--flagsurl http://portlandcyclers.net/evan.xml MPIM2007.007.mref.rsf
...
Retrieving flags file (http://portlandcyclers.net/evan.xml)...
ERROR: Specified flags URL (http://portlandcyclers.net/evan.xml) could not be retrieved.
The error returned was: 500 Can't connect to portlandcyclers.net:80
(Bad hostname 'portlandcyclers.net')
improves when a proxy is provided:
$ runspec --rawformat --output_format txt \ --flagsurl http://portlandcyclers.net/evan.xml \ --http_proxy=http://webcache.tom.spokewrenchdad.com:8080 MPIM2007.007.mref.rsf
Note that this setting will override the value of the http_proxy environment variable, as well as any setting in the config file.
By default, no proxy is used. The special value none may be used to unset any proxies set in the environment or via config file.
Meaning: The machines to build for or to run. Normally used only if the config file has been written to handle more than one machine type. The config file author should tell you what machines are supported by the config file.
Important note: If you specify multiple machine types, multiple runs will be performed. Because the benchmark binary names contain only the extension, it is quite possible, even easy, to cause binaries built in one run to be overwritten by subsequent runs. In practice, the ability to specify multiple machine types should rarely be used.
Note: The machine name may only consist of alphanumerics, underscores, hyphens, and periods.
Note to Windows users: Because of the command-line preprocessing performed by cmd.exe, it is not possible to run more than one machine type on Windows.
Meaning: Do not delete existing object files before attempting to build. This option should only be used for troubleshooting a problematic compile. It is against the run rules to use it when building binaries for an actual submission.
For a better way of troubleshooting a problematic compile, see the information about specmake in utility.html
Meaning: Package up the currently selected set of binaries, config files and other support files into a bundle that can be used to re-create the current run on a different system or installation.
When --make_bundle is present on the command line, most other switches have no immediate effect. The tools do not actually do the run at bundle creation time. Instead, a control file is written to the bundle to allow the run to occur on the destination system. The runspec command on the destination system will include all of your options other than those related to bundling.
Optional: additional files or directories may be specified on the command line for inclusion in the bundle.
Any such additional files must be underneath the $SPEC/ directory (or %SPEC%\ on Windows), but may not reside under any of the top-level subdirectories that ship with the suite (such as benchspec, bin, config, or result). Create a new subdirectory, such as %SPEC%\extras\ (on Windows) or $SPEC/extras/ (Unix). If your compiler license allows redistribution of run time libraries, you could place copies of them in that subdirectory, and use preenv variables to point $LD_LIBRARY_PATH at them.
In the following example, we begin by building a binary, check its runtime requirements, and populate a directory with the needed run time libraries:
$ cat may08a.cfg ext = may08a int=default: CXX = /bench/gcc/bin/g++ OPTIMIZE = -O $ runspec --config may08a --action build 126.lammps runspec v6152 - Copyright 1999-2008 Standard Performance Evaluation Corporation ... Compiling Binaries Building 126.lammps base may08a default: (build_base_may08a.0000) Build successes: 126.lammps(base) Build Complete The log for this run is in /bench/mpi2007/result/MPI2007.006.log $ go lammps exe /bench/mpi2007/benchspec/MPI2007/126.lammps/exe $ ldd lammps_base.may08a libstdc++.so.6 => /bench/gcc/lib/gcc/sparc-sun-solaris2.10/4.2.0/libstdc++.so.6 libm.so.2 => /lib/libm.so.2 libgcc_s.so.1 => /bench/gcc/lib/gcc/sparc-sun-solaris2.10/4.2.0/libgcc_s.so.1 libc.so.1 => /lib/libc.so.1 /platform/SUNW,Sun-Blade-1000/lib/libc_psr.so.1 $ mkdir $SPEC/extras $ cp /bench/gcc/lib/gcc/sparc-sun-solaris2.10/4.2.0/libstdc++.so.6 $SPEC/extras $ cp /bench/gcc/lib/gcc/sparc-sun-solaris2.10/4.2.0/libgcc_s.so.1 $SPEC/extras
Next, add the needed preENV line for the run time library to the config file. Then, bundle it all up:
$ go config /bench/mpi2007/config $ cat > tmp preENV_LD_LIBRARY_PATH = $[top]/extras/:\$LD_LIBRARY_PATH $ cat may08a.cfg >> tmp $ mv tmp may08a.cfg $ runspec --config may08a --size mtest --iterations 1 \ 126.lammps --make_bundle mumble $SPEC/extras ... Bundling finished. The completed bundle is in /bench/mpi2007/mumble.mpi2007bundle.bz2
The above command causes the entire contents of $SPEC/extras/ to be added to the bundle, along with the binary for 126.lammps and the config file may08.cfg.
Bundle verification: If you would like to verify the contents of a bundle, you can do so with "specbzip2 -dc" and "spectar -tf -", like so:
$ specbzip2 -dc mumble.mpi2007bundle.bz2 | spectar -tf - config/mumble.control extras/libgcc_s.so.1 extras/libstdc++.so.6 config/may08a.cfg benchspec/MPI2007/126.lammps/exe/lammps_base.may08a config/MD5.mumble.control.d359135ba4f5a5762428dc9cadfb94ae extras/MD5.libgcc_s.so.1.6b056cb37dd2c88d1ed736000202af5d extras/MD5.libstdc++.so.6.2c0e98f4d4d21db7ecdd2b247e9b39de config/MD5.may08a.cfg.4fbadb2ecf67dea8c2f9be5346bc871c benchspec/MPI2007/126.lammps/exe/MD5.lammps_base.may08a.9c1fae1704a21ec3af874d88496a6564 $
Don't worry about the odd looking extra files in the bundle; these are md5 checksums, which are used to help ensure bundle integrity.
Using a bundle: See the descriptions of --use_bundle or --unpack_bundle for information on what to do with a bundle when you've got one.
Note: --make_bundle can't bundle files up that aren't underneath the top-level $SPEC directory. If you use the output_root config file option with --make_bundle, please make sure that it points to somewhere under $SPEC.
WARNING: Although the features to create and use bundles are intended to make it easier to run SPEC MPI2007, the tester remains responsible for compliance with the run rules. And, of course, both the creators and the users of bundles are responsible for compliance with any applicable software licenses.
The --make-bundle feature is new with MPI2007 V1.1
$ runspec --newflags --verbose 7 runspec v5603 - Copyright 1999-2007 Standard Performance Evaluation Corporation Using 'solaris-sparc' tools Reading MANIFEST... 18342 files Loading runspec modules................ Locating benchmarks...found 13 benchmarks in 15 benchsets. Checking for flag updates for 104.milc Checking for flag updates for 107.leslie3d Checking for flag updates for 113.GemsFDTD . . . Checking for updates to Docs/flags/flags-advanced.xml Checking for updates to Docs/flags/flags-simple.xml Flag and config file update successful! There is no log file for this run. runspec finished at Thu Jan 10 06:22:13 2008; 9 total seconds elapsed $
Meaning: Unpack a previously-created bundle of binaries and config file, but do not attempt to start a run using the settings in the bundle. For your reference, the command that would have been used is printed out. See --make_bundle for more information about bundles.
Meaning: Use a previously-created bundle of binaries and config file for the current run. Unless overridden, the run will use the set of extension, machine name, tuning levels, and benchmarks that were in effect when the bundle was created. If you specify a run that would use binaries that the current bundle doesn't contain, they will attempt to be built as usual before the run. See --make_bundle for more information about bundles.
The following is an excerpt from the output that is printed when we use the bundle that was created in the example at --make_bundle:
$ runspec --use_bundle /net/remotesys/bench/mpi2007/mumble.mpi2007bundle.bz2
runspec v6152 - Copyright 1999-2008 Standard Performance Evaluation Corporation
Using 'solaris-sparc' tools
Reading MANIFEST... 18359 files
Loading runspec modules................
Locating benchmarks...found 13 benchmarks in 15 benchsets.
Use Bundle: /net/remotesys/bench/mpi2007/mumble.mpi2007bundle.bz2
Uncompressing bundle file "/net/remotesys/bench/mpi2007/mumble.mpi2007bundle.bz2"...done!
Reading bundle table of contents...5 files
Unpacking bundle file...done
Bundle unpacking complete.
About to run:
/export/home/kit111/bin/specperl /export/home/kit111/bin/runspec --config=may08a.cfg
--ext=may08a --mach=default --size=mtest --iterations=1 --tune=base 126.lammps
runspec v6152 - Copyright 1999-2008 Standard Performance Evaluation Corporation
Using 'solaris-sparc' tools
Reading MANIFEST... 18359 files
Loading runspec modules................
Locating benchmarks...found 13 benchmarks in 15 benchsets.
Reading config file '/export/home/kit111/config/may08a.cfg'
Setting up environment for runspec...
About to re-exec runspec...
------------------------------------------------------------------------------
------------------------------------------------------------------------------
runspec v6152 - Copyright 1999-2008 Standard Performance Evaluation Corporation
Using 'solaris-sparc' tools
Reading MANIFEST... 18359 files
Loading runspec modules................
Locating benchmarks...found 13 benchmarks in 15 benchsets.
Reading config file '/export/home/kit111/config/may08a.cfg'
Benchmarks selected: 126.lammps
Compiling Binaries
Up to date 126.lammps base may08a default
Setting Up Run Directories
Setting up 126.lammps mtest base may08a default: existing (run_base_mtest_may08a.0000)
Running Benchmarks
Running 126.lammps mtest base may08a default
Success: 1x126.lammps
Note in the example that runspec restarted itself twice. It unpacked the bundle, then restarted itself to run the command that had been entered at the time that the bundle was created. Upon doing so, it discovered the preENV line for $LD_LIBRARY_PATH in the config file. It applied the environment setting, then began all over again.
WARNING: Although the features to create and use bundles are intended to make it easier to run SPEC MPI2007, the tester remains responsible for compliance with the run rules. And, of course, both the creators and the users of bundles are responsible for compliance with any applicable software licenses.
The --use_bundle option is new with SPEC MPI2007 V1.1.
(This table is organized alphabetically, without regard to upper/lower case, and without regard to the presence of a leading "no").
| -a | Same as --action |
|---|---|
| --action action | Do: build|buildsetup|clean|clobber|configpp| onlyrun|realclean|report|run|runsetup|scrub| setup|trash|validate |
| --basepeak | Copy base results to peak (use with --rawformat) |
| --nobuild | Do not attempt to build binaries |
| -c | Same as --config |
| --check_version | Check whether an updated version of MPI2007 is available |
| --comment "text" | Add a comment to the log and the stored configfile. |
| --config file | Set config file for runspec to use |
| -D | Same as --rebuild |
| -d | Same as --deletework |
| --debug | Same as --verbose |
| --define SYMBOL[=VALUE] | Define a config preprocessor macro |
| --delay secs | Add delay before and after benchmark invocation |
| --deletework | Force work directories to be rebuilt |
| --dryrun | Same as --fake |
| --dry-run | Same as --fake |
| -e | Same as --extension |
| --ext | Same as --extension |
| --extension ext[,ext...] | Set the extensions |
| -F | Same as --flagsurl |
| --fake | Show what commands would be executed. |
| --fakereport | Generate a report without compiling codes or doing a run. |
| --fakereportable | Generate a fake report as if "--reportable" were set. |
| --[no]feedback | Control whether builds use feedback directed optimization |
| --flagupdate | Same as --update |
| --flagsupdate | Same as --update |
| --flagsurl url | Location (url or filespec) where to find your flags file |
| --getflags | Same as --update |
| --graph_auto | Let the tools pick minimum and maximum for the graph |
| --graph_min N | Set the minimum for the graph |
| --graph_max N | Set the maximum for the graph |
| -h | Same as --help |
| --help | Print usage message |
| --http_proxy | Specify the proxy for internet access |
| --http_timeout | Timeout when attempting http access |
| -I | Same as --ignore_errors |
| -i | Same as --size |
| --ignore_errors | Continue with benchmark runs even if some fail |
| --ignoreerror | Same as --ignore_errors |
| --info_wrap_column N | Set wrap width for non-notes informational items |
| --infowrap | Same as --info_wrap_column |
| --input | Same as --size |
| --iterations N | Run each benchmark N times |
| --keeptmp | Keep temporary files |
| -l | Same as --loose |
| --loose | Do not produce a reportable result |
| --noloose | Same as --reportable |
| -m | Same as --machine |
| -M | Same as --make_no_clobber |
| --mach | Same as --machine |
| --machine name[,name...] | Set the machine types |
| --make_bundle | Create a package of binaries and config file |
| --make_no_clobber | Do not delete existing object files before building. |
| --max_active_compares | Same as --maxcompares |
| --maxcompares N | Set the number of concurrent compares to N |
| --mockup | Same as --fakereportable |
| -n | Same as --iterations |
| -N | Same as --nobuild |
| --newflags | Same as --update |
| --notes_wrap_column N | Set wrap width for notes lines |
| -noteswrap | Same as --notes_wrap_column |
| -o | Same as --output_format |
| --output_format format[,format...] | Generate: all|cfg|check|csv|flags|html|mail|pdf|ps|raw|screen|text |
| --preenv | Allow environment settings in config file to be applied |
| -R | Same as --rawformat |
| -r | Same as --ranks N |
| --ranks [N] | Set the number of MPI ranks to run |
| --rawformat | Format raw file |
| --rebuild | Force a rebuild of binaries |
| --reportable | Produce a reportable result |
| --noreportable | Same as --loose |
| --reportonly | Same as --fakereport |
| --[no]review | Format results for review |
| -s | Same as --reportable |
| -S SYMBOL[=VALUE] | Same as --define |
| -S SYMBOL:VALUE | Same as --define |
| --[no]setprocgroup | [Don't] try to create all processes in one group. |
| --size size[,size...] | Select data set(s): mtest|mtrain|mref|ltest|ltrain|lref |
| --strict | Same as --reportable |
| --nostrict | Same as --loose |
| -T | Same as --tune |
| --[no]table | Do [not] include a detailed table of results |
| --test | Run various perl validation tests on specperl |
| --train_with | Change the training workload |
| --tune | Set the tuning levels to one of: base|peak|all |
| --tuning | Same as --tune |
| --undef SYMBOL | Remove any definition of this config preprocessor macro |
| -U | Same as --username |
| --unpack_bundle | Unpack a package of binaries and config file |
| --use_bundle | Use a package of binaries and config file |
| --update | Check www.spec.org for updates to benchmark and example flag files, and config files |
| --update_flags | Same as --update |
| --username | Name of user to tag as owner for run directories |
| -v | Same as --verbose |
| --verbose | Set verbosity level for messages to N |
| -V | Same as --version |
| --version | Output lots of version information |
| -? | Same as --help |
Copyright © 1999-2010 Standard Performance Evaluation Corporation
All Rights Reserved