(To check for possible updates to this document, please see http://www.spec.org/accel/Docs/ )
Contents
1. Review Pre-requisites
2. Create destination. Have enough space; avoid space.
3. Mount the Benchmark ISO
4. Set your directory to the Benchmark ISO
5. Use install.sh
5.a. Destination selection
5.b. Toolset selection
5.c. The files are unpacked and tested
6. Source shrc or cshrc
7. Try to build one benchmark
8. Try running one benchmark with the test dataset
9. Try a real dataset
10. Try a full (reportable) run
Appendix 1: the DVD drive is on system A, but I want to install on system B. What do I do?
A1. Network mount
A2. Tar file
Appendix 2: Uninstalling SPEC ACCEL
Note: links to SPEC ACCEL documents on this web page assume that you are reading the page from a directory that also contains the other SPEC ACCEL 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:
The SPEC ACCEL suite has been tested under Unix and Linux. The benchmark environment should work with Mac OS X and Windows systems but has not been tested. Your benchmark can be installed under many operating systems.
Reminder: the SPEC license allows you to install on multiple systems as you may wish within your institution; but you may not share the software with the public.
The installation procedure for Unix, Linux, and Mac OS X is as follows:
Review the hardware and software requirements, in system-requirements.html
Create a directory on the destination disk. You should make sure that you have a disk that has at least 8GB free. (For more information on disk usage, see system-requirements.html.)
Don't put spaces in the path: even if you make it through the installation (doubtful), you are just asking for trouble, because there may be multiple programs from both SPEC and from your compiler that expect space to be an argument delimiter, not part of a path name. (This being the *Unix* install guide, you wouldn't have thought of using spaces in in the first place, would you?)
You can either burn a DVD of the benchmark ISO file, or you can just directly mount the benchmark ISO file you have downloaded. If you choose to mount the benchmark ISO file, the following examples may help you get it mounted. The examples assume the benchmark has been saved in the file accel-1.2.iso. The target location listed in these examples is /mnt but could be anything you have created.
After you are done installing, you may want to unmount the benchmark ISO. This can be done by making sure you are no longer in the install mount point and then issue the comman umount /mnt. This will unmount the filesystem. If you are on Solaris, you may also want to remove the lofi device that was created with lofiadm command. See the man page for further instructions.
| AIX: | loopmount -i accel-1.2.iso -o "-V cdrfs -o ro" -m /mnt |
| Linux: | mount -t iso9660 -o ro,loop accel-1.2.iso /mnt |
| Solaris: | mount -F hsfs -o ro `lofiadm -a accel-1.2.iso` /mnt |
If you have created a DVD, insert the the DVD, and, if necessary, issue a mount command for it. For many operating systems, the DVD will be automatically mounted. If not, you may have to enter an explicit mount command. If your operating system supports the Rock Ridge Interchange Protocol extensions to ISO 9660, be sure to select them, unless they are the default. The following examples are not intended to be comprehensive, but may get you started or at least give you clues which manpages to read:
| AIX: | mount -v cdrfs -r /dev/cd0 /cdrom |
| HP-UX: | mount -o cdcase /dev/disk/disk5 /mnt/cdrom/ |
| Linux: | mount -t iso9660 -o ro,exec /dev/cdrom /mnt |
| Solaris: | If Volume Management is running, you should find that the DVD is automatically mounted, as
/cdrom/label_of_volume/ If not, you should be able to mount it with commands similar to this:
mkdir /mnt1 mount -F hsfs -o ro /dev/dsk/c0t6d0s0 /mnt1 |
| Virtual Machines | If you are running in a virtual machine, you will need to convince the host operating system to allow your guest OS to have access to the DVD. The means of accomplishing this will vary. For reference, the following worked with a Linux guest running under Virtual Box V4.0.6, with Windows 7 as the host: (1) Shut down the virtual machine (don't just pause it; tell it to run its shutdown procedure). (2) The Settings dialog should now be visible (it's grayed out if the machine state is not shut down). (3) Use Settings to configure the DVD drive as both available to the guest OS and as "passthrough". (4) Boot the virtual machine. (5) Log in. (6) Insert the DVD. (7) At this point, the DVD was automatically mounted as /media/SPEC_ACCEL. |
Note that you may need root privileges to mount the DVD or benchmark ISO.
The following paragraphs assume that your benchmark mount point is on the same system as where you wish to install. If it is on a different system, please see Appendix 1.
If you haven't already done so by now, start a Terminal window (aka "command window", "shell", "console", "terminal emulator", "character cell window", "xterm", etc.) and issue a cd command to set your current working directory to the directory where the benchmark is mounted. The exact command will vary depending on the label on the media, the operating system, and the devices configured. It might look something like one of these:
$ cd /Volumes/SPEC_ACCEL $ cd /media/SPEC_ACCEL $ cd /dvdrom/spec_accel $ cd /mnt
Type:
./install.sh
Depending on your installation type, you may be prompted for a destination directory:
SPEC ACCEL Installation Top of the ACCEL tree is '/Volumes/SPEC_ACCEL' Enter the directory you wish to install to (e.g. /usr/accel) /Users/kgoel/accel
When answering the above question, note that you will have to use syntax acceptable to sh (so you might need to say something like "$HOME/mydir" instead of "~/mydir"). As mentioned above, don't use spaces.
Note: You can also specify the destination directory in the command
line, using the -d flag, for example, like this:
./install.sh -d
/Users/kgoel/accel
The installation procedure will show you the directories that will be used to install from and to. You will see a message such as this one:
Installing FROM /Volumes/SPEC_ACCEL Installing TO /Users/kgoel/accel Is this correct? (Please enter 'yes' or 'no') yes
Enter "yes" if the directories match your expectations. If there is an error, enter "no", and the procedure will exit, and you can try again, possibly using the -d flag mentioned in the note above.
The installation procedure will attempt to automatically determine your current platform type (hardware architecture, operating system, etc.) In some cases, the tools may identify several candidate matches for your architecture.
You typically do not have to worry about whether the toolset is an exact match to your current environment, because the toolset selection does not affect your benchmark scores, and because the installation procedure does a series of tests to ensure that the selected tools work on your system.
Examples: (1) the installation procedure may determine that SPEC tools built on version "N" of
your operating system are entirely functional on version "N+3". (2) Tools built on one Linux distribution often
work correctly on another: notably, certain versions of SuSE are compatible, from a tools point of view, with certain
versions of RedHat. (3) Tools built on AMD chips with 64-bit instructions ("amd64") are compatible with Intel chips
that implement the same instruction set under the names "EM64T" or "Intel 64" (but not compatibie with chips that
implement the Itanium instruction set, abbreviated "ia64"). (4) Often, though not always, 32-bit toolsets work correctly
on 64-bit operating systems.
Mostly, you don't need to worry about all this, because the installation
procedure does a comprehensive set of tests to verify compatibility.
If at least one candidate match is found, you will see a message such as:
The following toolset is expected to work on your platform. If the
automatically installed one does not work, please re-run install.sh and
exclude that toolset using the '-e' switch.
The toolset selected will not affect your benchmark scores.
macosx For MacOS X 10.4+ on Intel systems.
Built on MacOS X 10.6.6 with GCC 4.0.1, using
the 10.4u SDK.
If the installation procedure is unable to determine your system architecture, you will see a message such as:
We do not appear to have vendor supplied binaries for your
architecture. You will have to compile the tool binaries by
yourself. Please read the file
/Volumes/SPEC_ACCEL/Docs/tools_build.html
for instructions on how you might be able to build them.
If you see that message, please stop here, and examine the file tools-build.html.
Note: If the tools that are automatically installed on your system do not work, but you know that another set of tools that is in the list will work, you can exclude the ones that do not work. You may be instructed to do this during the first installation. Use the -e flag for install.sh, for example:
./install.sh -e linux-redhat72-ia32
The above will cause the tools for linux-redhat72-ia32 to be excluded from consideration.
Alternatively, you can explicitly direct which toolset is to be used with the -u flag for install.sh, for example:
./install.sh -u linux-suse10-amd64
The above will cause the tools for linux-suse10-amd64 to be installed, even if another toolset would have been chosen automatically. If you specify tools that do not work on your system, the installation procedure will stop without installing any tools.
Thousands of files will be unpacked from the distribution media, and quietly installed on your destination disk. (If you would prefer to see them all named you can set VERBOSE=1 in your environment before installing the kit.) Various tests will be performed to verify that the files have been correctly installed, and that the tools work correctly. You should see summary messages such as these:
================================================================= Attempting to install the the macosx toolset... <<-- or whatever toolset was selected Checking the integrity of your source tree... Checksums are all okay. Unpacking binary tools for macosx... <<-- your toolset Checking the integrity of your binary tools... Checksums are all okay. Testing the tools installation (this may take a minute) ........................................................................o....... ................................................................................ .......................................................... Installation successful. Source the shrc or cshrc in /Users/kgoel/accel <<-- your directory to set up your environment for the benchmark.
At this point, you will have consumed about 1.5GB of disk space on the destination drive.
Change your current directory to the top-level SPEC directory and source either shrc or cshrc:
For example, if you are using a Bourne-compatible shell (such as ash, bash, ksh, zsh), you could type:
If you are using a csh-compatible shell, you could type:
The effect of the above commands is to set up environment variables and paths for SPEC.
From this point forward, we are testing basic abilities of the SPEC ACCEL kit, including compiling benchmarks and running them. You may skip the remaining steps if all of the following are true:
Warning: even if someone else supplies binaries, you remain responsible for compliance with SPEC's Fair Use rule and the ACCEL run rules.
Change to the config directory, and test that you can build a benchmark using a config file supplied for your system. For example:
$ cd $SPEC/config
$ cp Example-macosx-gcc421.cfg Khushboo-macosx.cfg
$ runspec --config=Khushboo-macosx.cfg --action=build --platform NVIDIA --device GPU \
--tune=base fft
The above command assumes that you can identify a config file (in the directory $SPEC/config) that is appropriate for you. In this case, the user started with Example-macosx-gcc421.cfg. Your starting point will probably differ; here are some resources to help:
The "--tune=base" above indicates that we want to use only the simple tuning, if more than one kind of tuning is supplied in the config file. The "--platform NVIDIA" indicates that we want to use the NVIDIA platform. And the "--device GPU" indicates that we want to run the benchmark on a GPU. Optionally we can also run on a CPU.
Test that you can run a benchmark, using the minimal input set - the "test" workload. For example:
$ runspec --config=Khushboo-macosx.cfg --platform NVIDIA --device GPU --size=test \
--noreportable --tune=base --iterations=1 fft
The "\" above indicates that the command is continued on the next line. The "--noreportable" ensures that the tools will allow us to run just a single benchmark instead of the whole suite, "--iterations=1" says just run the benchmark once.
Check the results in $SPEC/result
Test that you can run a benchmark using the real input set - the "reference" workload. For example:
$ runspec --config=Khushboo-macosx.cfg --platform NVIDIA --device GPU --size=ref \
--noreportable --tune=base --iterations=1 fft
Check the results in $SPEC/result.
If everything has worked up to this point, you may wish to start a full run, perhaps leaving your computer to run overnight. The extended test will demand significant resources from your machine, including computational power and memory of several types. In order to avoid surprises, before starting the reportable run, you should review the section About Resources, in system-requirements.html.
Have a look at runspec.html to learn how to do a full run of the suite.
The command runspec -h will give you a brief summary of the many options for runspec.
To run a reportable run of the benchmark with simple (baseline) tuning:
C:\accel> runspec --tune=base --config=Khushboo-macosx.cfg --platform NVIDIA --device GPU opencl
Here is a complete Linux installation, with interspersed commentary. This example follows the steps listed above. We assume that Steps 1 through 3 are already complete (the pre-requisites are met, we have enough space, the benchmark is mounted).
Step 4: Set the current working directory to the benchmark mount point:
$ cd /media/SPEC_ACCEL
Step 5: Invoke install.sh. When prompted, we enter the destination directory:
$ ./install.sh
SPEC ACCEL Installation
Top of the ACCEL tree is '/media/SPEC_ACCEL'
Enter the directory you wish to install to (e.g. /usr/accel)
/accel
Installing FROM /media/SPEC_ACCEL
Installing TO /accel
Is this correct? (Please enter 'yes' or 'no')
yes
The following toolset is expected to work on your platform. If the
automatically installed one does not work, please re-run install.sh and
exclude that toolset using the '-e' switch.
The toolset selected will not affect your benchmark scores.
linux-suse10-amd64 For 64-bit AMD64/EM64T Linux systems running
SuSE Linux 10 or later, and other
compatible Linux distributions, including
some versions of RedHat Enterprise Linux
and Oracle Linux Server.
Built on SuSE Linux 10 with
GCC v4.1.0 (SUSE Linux)
linux-redhat72-ia32 For x86, IA-64, EM64T, and AMD64-based Linux
systems with GLIBC 2.2.4+.
Built on RedHat 7.2 (x86) with gcc 3.1.1
=================================================================
Attempting to install the linux-suse10-amd64 toolset...
Checking the integrity of your source tree...
Checksums are all okay.
Unpacking binary tools for linux-suse10-amd64...
Checking the integrity of your binary tools...
Checksums are all okay.
Testing the tools installation (this may take a minute)
........................................................................o.....................................
............................................................................................................
Installation successful. Source the shrc or cshrc in
/accel
to set up your environment for the benchmark.
Step 6: Now, we change the current working directory from the install media to the location of the new SPEC ACCEL tree. Since this user has a Bourne compatible shell, shrc is sourced (for csh compatible shells, use cshrc).
Next, the config file Example-linux64-amd64-gcc43+.cfg has been picked as a starting point for this system. The gcc compiler might not provide the best possible score for this particular SUT (System Under Test), a server with two Intel XEon X5690 processors, but this config file provides a reasonable start in order to demonstrate that the SPEC tree is functional.
(Note that the term "amd64" in the config file name does not designate a chip from a particular manufacturer; rather, it designates an instruction set, variously known as "amd64", "EM64T", and "x86_64". The config file is an OK starting point for this SUT.)
$ cd /accel/ $ . ./shrc $ cd config $ cp Example-linux64-amd64-gcc43+.cfg mytest.cfg $ runspec --config=mytest.cfg --platform NVIDIA --device GPU --action=build --tune=base fft runspec v2174 - Copyright 1999-2012 Standard Performance Evaluation Corporation Using 'linux-suse10-amd64' tools Reading MANIFEST... 13888 files Loading runspec modules................ Locating benchmarks...found 38 benchmarks in 12 benchsets. Reading config file '/accel/config/mytest.cfg' Running "specperl /accel/Docs/sysinfo" to gather system information. Benchmarks selected: 110.fft Compiling Binaries Building 110.fft base compsys default: (build_base_compsys.0000) [Mon Nov 4 11:16:52 2013] Build successes: 110.fft(base) Build Complete The log for this run is in /accel/result/ACCEL.001.log runspec finished at Mon Nov 4 11:18:27 2013; 5 total seconds elapsed
Just above, various compile and link commands may or may not be echoed to your screen, depending on the settings in your config file. At this point, we've accomplished a lot. The SPEC tree is installed, and we have verified that a benchmark can be compiled using the Fortran compiler. (The sharp-eyed reader may notice some warnings above about casts of pointers. These warnings from the compiler have been reviewed by SPEC's project leader for 110.fft, who has determined that they will not affect operation of the benchmark.)
Step 8: Now try running a benchmark, using the minimal test workload. The test workload runs in a tiny amount of time and does a minimal verification that the benchmark executable can at least start up:
$ runspec --config=mytest.cfg --platform NVIDIA --device GPU --size=test --noreportable --tune=base --iterations=1 fft
runspec v2174 - Copyright 1999-2012 Standard Performance Evaluation Corporation
Using 'linux-suse10-amd64' tools
Reading MANIFEST... 13888 files
Loading runspec modules................
Locating benchmarks...found 38 benchmarks in 12 benchsets.
Reading config file '/accel/config/raven.cfg'
Running "specperl /accel/Docs/sysinfo" to gather system information.
Benchmarks selected: 110.fft
Compiling Binaries
Up to date 110.fft base compsys default
Setting Up Run Directories
Setting up 110.fft test base compsys default: created (run_base_test_compsys.0000)
Running Benchmarks
Running 110.fft test base compsys default [Mon Nov 4 11:50:16 2013]
Success: 1x110.fft
Producing Raw Reports
mach: default
ext: compsys
size: test
set: opencl
format: raw -> /accel/result/ACCEL_OCL.002.test.rsf
Parsing flags for 110.fft base: done
Doing flag reduction: done
format: ASCII -> /accel/result/ACCEL_OCL.002.test.txt
set: openacc
The log for this run is in /accel/result/ACCEL.002.log
runspec finished at Mon Nov 4 11:50:37 2013; 23 total seconds elapsed
Notice about 20 lines up the notation "Success: 1x110.fft". That is what we want to see.
Step 9: let's try running fft with the real workload. This will take a while on the tested server running Linux.
$ runspec --config=mytest.cfg --platform NVIDIA --device GPU --size=ref --noreportable --tune=base --iterations=1 fft
runspec v2174 - Copyright 1999-2012 Standard Performance Evaluation Corporation
Using 'linux-suse10-amd64' tools
Reading MANIFEST... 13888 files
Loading runspec modules................
Locating benchmarks...found 38 benchmarks in 12 benchsets.
Reading config file '/N/dc2/scratch/huili/accelv1/config/raven.cfg'
Running "specperl /N/dc2/scratch/huili/accelv1/Docs/sysinfo" to gather system information.
Benchmarks selected: 110.fft
Compiling Binaries
Up to date 110.fft base compsys default
Setting Up Run Directories
Setting up 110.fft ref base compsys default: created (run_base_ref_compsys.0001)
Running Benchmarks
Running 110.fft ref base compsys default [Mon Nov 4 12:00:11 2013]
Success: 1x110.fft
Producing Raw Reports
mach: default
ext: compsys
size: ref
set: opencl
format: raw -> /accel/result/ACCEL_OCL.003.ref.rsf
Parsing flags for 110.fft base: done
Doing flag reduction: done
format: ASCII -> /accel/result/ACCEL_OCL.003.ref.txt
set: openacc
The log for this run is in /accel/result/ACCEL.003.log
runspec finished at Mon Nov 4 12:01:41 2013; 92 total seconds elapsed
Success with the real workload! So now let's look in the result directory and see what we find:
$ cd result $ ls ACCEL.001.log ACCEL.003.log ACCEL_OCL.002.test.txt ACCEL_OCL.003.ref.txt ACCEL.002.log ACCEL_OCL.002.test.rsf ACCEL_OCL.003.ref.rsf lock.ACCEL $ grep runspec: *log ACCEL.001.log:runspec: runspec --config=mytest.cfg --platform NVIDIA --device GPU --action=build --tune=base fft ACCEL.002.log:runspec: runspec --config=mytest.cfg --platform NVIDIA --device GPU --size=test --noreportable --tune=base --iterations=1 fft ACCEL.003.log:runspec: runspec --config=mytest.cfg --platform NVIDIA --device GPU --size=ref --noreportable --tune=base --iterations=1 fft $
Notice the three separate sets of files: .001, .002, and .003
ACCEL.001.log has the log from the compile.
ACCEL.002.log has the log from running 110.fft with the "test" input. The various outputs (.rsf, .txt) are all preceded by "ACCEL_OCL", because 110.fft is one of the OpenCL benchmarks. The tools also distinguish whether the input was a "test" input by putting that in the file name as well.
ACCEL.003.log has the log from running 110.fft with the "ref" input. Once again, the various outputs all start with "ACCEL_OCL".
Here is the complete .txt report from running 110.fft ref:
$ cat ACCEL_OCL.003.ref.txt
##############################################################################
# INVALID RUN -- INVALID RUN -- INVALID RUN -- INVALID RUN -- INVALID RUN #
# #
# 'reportable' flag not set during run #
# 123.nw (base) did not have enough runs! #
# 103.stencil (base) did not have enough runs! #
# 104.lbm (base) did not have enough runs! #
# 121.lavamd (base) did not have enough runs! #
# 101.tpacf (base) did not have enough runs! #
# 114.mriq (base) did not have enough runs! #
# 110.fft (base) did not have enough runs! #
# 124.hotspot (base) did not have enough runs! #
# 125.lud (base) did not have enough runs! #
# 128.heartwall (base) did not have enough runs! #
# 127.srad (base) did not have enough runs! #
# 140.bplustree (base) did not have enough runs! #
# 112.spmv (base) did not have enough runs! #
# 126.ge (base) did not have enough runs! #
# 120.kmeans (base) did not have enough runs! #
# 116.histo (base) did not have enough runs! #
# 117.bfs (base) did not have enough runs! #
# 118.cutcp (base) did not have enough runs! #
# 122.cfd (base) did not have enough runs! #
# Unknown flags were used! See #
# http://www.spec.org/accel/Docs/runspec.html#flagsurl #
# for information about how to get rid of this error. #
# #
# INVALID RUN -- INVALID RUN -- INVALID RUN -- INVALID RUN -- INVALID RUN #
##############################################################################
SPEC(R) ACCEL_OCL Summary
Computer System Incorporated Computer System XXX
Mon Nov 4 12:00:09 2013
ACCEL License: Test date: Nov-2013
Test sponsor: Computer System Incorporated Hardware availability: --
Tested by: Computer System Incorporated Software availability: --
Estimated Estimated
Base Base Base Peak Peak Peak
Benchmarks Ref. Run Time Ratio Ref. Run Time Ratio
-------------- ------ --------- --------- ------ --------- ---------
101.tpacf NR
103.stencil NR
104.lbm NR
110.fft 99 72.0 1.38 *
112.spmv NR
114.mriq NR
116.histo NR
117.bfs NR
118.cutcp NR
120.kmeans NR
121.lavamd NR
122.cfd NR
123.nw NR
124.hotspot NR
125.lud NR
126.ge NR
127.srad NR
128.heartwall NR
140.bplustree NR
==============================================================================
101.tpacf NR
103.stencil NR
104.lbm NR
110.fft 99 72.0 1.38 *
112.spmv NR
114.mriq NR
116.histo NR
117.bfs NR
118.cutcp NR
120.kmeans NR
121.lavamd NR
122.cfd NR
123.nw NR
124.hotspot NR
125.lud NR
126.ge NR
127.srad NR
128.heartwall NR
140.bplustree NR
Est. SPECaccel_ocl_base --
Est. SPECaccel_ocl_peak Not Run
HARDWARE
--------
CPU Name: AMD Opteron 6276
CPU Characteristics:
CPU MHz: 350
CPU MHz Maximum: --
FPU: --
CPU(s) enabled: -1 cores, 1 chip, -1 cores/chip, -1 threads/core
CPU(s) orderable: --
Primary Cache: --
Secondary Cache: --
L3 Cache: --
Other Cache: --
Memory: 4 GB
31.552 GB fixme: If using DDR3, format is:
'N GB (M x N GB nRxn PCn-nnnnnR-n, ECC)'
Disk Subsystem: 3.5P add more disk info here
Other Hardware: --
SOFTWARE
--------
Operating System: Computer System Unix Version YYY
SUSE Linux Enterprise Server 11 (x86_64)
2.6.32.59-0.7.1_1.0401.6845-cray_gem_c
Compiler: Computer System Compiler C and Fortran90
Auto Parallel: No
File System: lustre
System State: Run level N (add definition here)
Base Pointers: --
Peak Pointers: Not Applicable
Other Software: --
Platform Notes
--------------
Sysinfo program /accel/Docs/sysinfo
$Rev: 1623 $ $Date:: 2017-05-15 #$ 8f8c0fe9e19c658963a1e67685e50647
running on nid00922 Mon Nov 4 12:00:10 2013
This section contains SUT (System Under Test) info as seen by
some common utilities. To remove or add to this section, see:
http://www.spec.org/accel/Docs/config.html#sysinfo
From /proc/cpuinfo
model name : AMD Opteron(TM) Processor 6276
1 "physical id"s (chips)
16 "processors"
cores, siblings (Caution: counting these is hw and system dependent. The
following excerpts from /proc/cpuinfo might not be reliable. Use with
caution.)
cpu cores : 16
siblings : 16
physical 0: cores 0 1 2 3 4 5 6 7
cache size : 2048 KB
From /proc/meminfo
MemTotal: 33084660 kB
HugePages_Total: 0
Hugepagesize: 2048 kB
/usr/bin/lsb_release -d
SUSE Linux Enterprise Server 11 (x86_64)
From /etc/*release* /etc/*version*
SuSE-release:
SUSE Linux Enterprise Server 11 (x86_64)
VERSION = 11
PATCHLEVEL = 1
mazama-release:
Mazama Wed Oct 31 02:36:27 CDT 2012 on hssbld0 by bwdev
lsb-cray-mazama-7.0.0
uname -a:
Linux nid00922 2.6.32.59-0.7.1_1.0401.6845-cray_gem_c #1 SMP Thu Nov 15
00:24:59 UTC 2012 x86_64 x86_64 x86_64 GNU/Linux
SPEC is set to: /accel
Filesystem Type Size Used Avail Use% Mounted on
10.10.0.172@o2ib:/dev/sda3 lustre 3.5P 812T 2.7P 23% /
Cannot run dmidecode; consider saying 'chmod +s /usr/sbin/dmidecode'
(End of data from sysinfo program)
General Notes
-------------
Baseline C: gcc
Fortran: f90 -64 -mp -O2
Base Unknown Flags
------------------
110.fft: "cc" (in CC) "cc" (in LD) "-O2" (in OPTIMIZE)
"-I/opt/nvidia/cudatoolkit/default/include" (in EXTRA_CFLAGS)
"-L/opt/cray/nvidia/default/lib64 -lcuda -lOpenCL" (in LIBS)
Base Compiler Invocation
------------------------
C benchmarks:
110.fft: No flags used
Base Optimization Flags
-----------------------
C benchmarks:
110.fft: No flags used
Base Other Flags
----------------
C benchmarks:
110.fft: No flags used
SPEC is a registered trademark of the Standard Performance Evaluation
Corporation. All other brand and product names appearing in this
result are trademarks or registered trademarks of their respective
holders.
##############################################################################
# INVALID RUN -- INVALID RUN -- INVALID RUN -- INVALID RUN -- INVALID RUN #
# #
# 'reportable' flag not set during run #
# 123.nw (base) did not have enough runs! #
# 103.stencil (base) did not have enough runs! #
# 104.lbm (base) did not have enough runs! #
# 121.lavamd (base) did not have enough runs! #
# 101.tpacf (base) did not have enough runs! #
# 114.mriq (base) did not have enough runs! #
# 110.fft (base) did not have enough runs! #
# 124.hotspot (base) did not have enough runs! #
# 125.lud (base) did not have enough runs! #
# 128.heartwall (base) did not have enough runs! #
# 127.srad (base) did not have enough runs! #
# 140.bplustree (base) did not have enough runs! #
# 112.spmv (base) did not have enough runs! #
# 126.ge (base) did not have enough runs! #
# 120.kmeans (base) did not have enough runs! #
# 116.histo (base) did not have enough runs! #
# 117.bfs (base) did not have enough runs! #
# 118.cutcp (base) did not have enough runs! #
# 122.cfd (base) did not have enough runs! #
# Unknown flags were used! See #
# http://www.spec.org/accel/Docs/runspec.html#flagsurl #
# for information about how to get rid of this error. #
# #
# INVALID RUN -- INVALID RUN -- INVALID RUN -- INVALID RUN -- INVALID RUN #
##############################################################################
---------------------------------------------------------------------------------
For questions about this result, please contact the tester.
For other inquiries, please contact webmaster@spec.org.
Copyright 2013 Standard Performance Evaluation Corporation
Tested with SPEC ACCEL v31.
Report generated on Mon Nov 4 12:01:41 2013 by ACCEL ASCII formatter v2174.
Done. The suite is installed, and we can run at least one benchmark for real (see the report of the time spent in 110.fft above).
If the title of this section describes your situation, you basically have two choices.
You might be able to mount the DVD on one system and use network services to make it available on other systems.
Please note that the SPEC ACCEL license agreement does not allow you to post the DVD on any public server. If your institution has a SPEC ACCEL license, then it's fine to post it on an internal server that is accessible only to members of your institution.
Whether you attempt a network mount will probably depend on:
If your network environment allows easy cross-system mounting, or if you feel brave about reading manpages, you can use a network mount for the installation. Otherwise, you can fall back on the tar file.
Network mount, easy:
for example, System A Solaris/Opteron + System B Solaris/SPARC
Your operating system may be configured to automatically mount the drive and automatically make it visible to other network systems, or may make it visible with minimal user intervention. During one set of testing, system A (with the DVD drive) was an Opteron-based system running Solaris 10. The SPEC ACCEL DVD was inserted. The operating system mounted it automatically, and from a terminal window, a (non-privileged) user entered the Solaris
command to make it visible to other hosts.
On System B, a Solaris SPARC system, a non-privileged user typed:
cd /net/systemA/cdrom/spec_accel ./install.sh
and the installation proceeded normally, picking up from step 5, above.
Network mount, medium difficulty:
for example, System A Solaris/Opteron + System B Tru64 Unix/Alpha
Subsequent to the tests of the previous paragraphs, the DVD drive on System A (Solaris/Opteron) was also visible to a system running Compaq Tru64 UNIX V5.1A. But in this case, a little assistance was needed from the privileged (root) account on system B:
echo "systemA.domain.com:/cdrom/spec_accel /systemA nfs ro,bg,soft 0 0" >> /etc/fstab mkdir /systemA /usr/sbin/mount /systemA
Then, the non-privileged user was able to say:
cd /systemA ./install.sh
and once again the installation proceeded normally, picking up from step 5, above.
Network mount, a bit harder:
for example, System A SuSE/x86 + System B Mac OS X/PowerPC
The SPEC ACCEL DVD was also inserted into a system running SuSE Linux 9.0, and used from a Mac OS X PowerBook. On both these systems, there are probably automatic tools that would have accomplished the following more quickly, but the tester happened to read the manpages in the particular order that he happened to read them in. The following succeeded:
On System A, root added
/dev/cdrom /cd iso9660 ro,user,noauto,unhide
to /etc/fstab as suggested by man mount; the DVD was inserted; and the user typed mount /cd. On System A, root also added:
/cd 192.168.0.0/24(ro,insecure,no_root_squash,sync)
to /etc/exports, and then typed:
exportfs -r rpc.nfsd -p 8 rpc.mountd cat /var/lib/nfs/etab
On System B, root typed:
mkdir /remote mount -t nfs 192.168.0.106:/cd /remote
Finally, the user typed
cd /remote ./install.sh
and installation continued as normal, with step 5.
If the DVD drive is on a system other than the one where you wish to do the installation, and if you do not wish to try to get a network mount working, then the final fallback is to use the compressed tarfile. If you choose this option, please carefully observe the warnings.
Go to the system with the DVD drive ("System A"). Insert the SPEC ACCEL DVD, and, if required, issue a mount command.
From a terminal window (aka command window), cd to the top level directory on the DVD.
You are going to retrieve five things from the DVD. First, find the large tarfile and its corresponding md5 file:
cd install_archives ls -l accel.tar.xz*
or, if System A is a Windows system, then:
cd install_archives dir accel.tar.xz*
In either case, you should see one moderately large file > 500MB, accel.tar.xz, and a small file associated with it that contains a checksum, accel.tar.xz.md5.
If you don't see the above files, try looking for cpu*tar*. The name might change if, for example, a maintenance update of SPEC ACCEL changes the name slightly to indicate an updated version.
Do whatever is required in order to transfer both files intact to the system where you wish to do the installation ("System B"). If you use ftp, do not forget to use image (binary) mode. For example:
$ ftp
ftp> op systemB
Name: imauser
Password:
ftp> cd /kits
ftp> bin <-------- important
200 Type set to I.
ftp> put accel.tar.xz
ftp> put accel.tar.xz.md5
Please note that the SPEC ACCEL license agreement does not allow you to post the above file on any public ftp server. If your institution has a SPEC ACCEL license, then it's fine to post it on an internal server that is accessible only to members of your institution.
Next, you are going to look on the DVD for versions of specxz, specmd5sum, and spectar that are compatible with system B. Please do not use the tar supplied by your operating system unless you are sure that it can handle long path names. Many commonly-supplied tar utilities cannnot.
Please do not use Windows Zip utilities, as these will not preserve line endings.
If you have GNU tar and the genuine xz, then you can use those; otherwise, please hunt around on the DVD to find prebuilt versions that are compatible with your environment, like so:
$ cd /media/SPEC_ACCEL/
$ cd tools/bin
$ ls
aix5L-ppc64 linux-redhat72-ia32 macosx solaris10-sparc
hpux11iv3-ipf linux-rhas4r4-ia64 solaris-sparc solaris10-x86
linux-debian6-armv6 linux-suse10-amd64 solaris-x86 windows-i386
$ cd aix5L-ppc64
$ cat description
For PowerPC systems running AIX 5L V5.3 or later
Built on AIX 5L 5300-02 with the
IBM XL C/C++ for AIX V9.0.0.25 compiler
$ ls -g spec*
-r-xr-xr-x. 1 imauser 52635 Aug 19 2011 specmd5sum
-r-xr-xr-x. 1 imauser 594483 Aug 19 2011 spectar
-r-xr-xr-x. 1 imauser 250543 Aug 19 2011 specxz
$
Once you've found the right versions of specxz, specmd5sum, and spectar for the system where you intend to install (system B), transfer them to system B using the same methods that you used for the big tarfile.
On system B, use specmd5sum to check that the file transfer worked correctly. In this example, we assume that you have placed all 5 of the files mentioned above in the /kits directory:
$ cd /kits $ chmod +x spec* $ specmd5sum -c accel.tar.xz.md5 accel.tar.xz: OK
Unpack the tarfile, like so:
$ cd /mybigdisk $ mkdir accel $ cd accel $ /kits/specxz -dc /kits/accel.tar.xz | /kits/spectar -xf -
Be patient: it will take a bit of time to unpack! It might take 15 minutes, depending on the speed of your processor and disks. Go for a coffee break.
Now, at last, type ./install.sh and pick up with step 5, above. Your output will be similar, but not identical, to the output shown in step 5 above: you won't see the "Unpacking xxxx" messages, because you already did the unpacking.
Note that the directory where you unpack the tarfile will be the directory you install FROM and also the directory you install TO. This is normal, and expected, for a tarfile installation.
You will see a question similar to this:
Installing FROM /mybigdisk/accel Installing TO /mybigdisk/accel Is this correct? (Please enter 'yes' or 'no') yes
If you enter "no", installation will stop. If you try to install TO another directory, using the -d flag, the installation will not succeed when using the tar file method.
At this time, SPEC does not provide an uninstall utility for SPEC ACCEL. Confusingly, there is a file named uninstall.sh in the top directory, but it does not remove the whole product; it only removes the SPEC tool set, and does not affect the benchmarks (which consume the bulk of the disk space).
To remove SPEC ACCEL on Windows systems, select the top directory in Windows Explorer and delete it.
To remove SPEC ACCEL on Unix systems, use rm -Rf on the directory where you installed the suite, for example:
rm -Rf /home/cs3000/saturos/spec/accel
If you have been using the output_root feature, you will have to track those down separately. Therefore, prior to removing the tree,, you might want to look for mentions of output root, for example:
Windows:
cd %SPEC%\config
findstr output_root *cfg
Unix:
cd $SPEC/config
grep output_root *cfg
Note: instead of deleting the entire directory tree, some users find it useful to keep the config and result subdirectories, while deleting everything else.
Copyright 2014-2017 Standard Performance Evaluation Corporation
All Rights Reserved
q. Do you have to be root? Occasionally, users of Unix systems have asked whether it is necessary to elevate privileges, or to become 'root', prior to entering the above command. SPEC recommends (*) that you do not become root, because: (1) To the best of SPEC's knowledge, no component of SPEC ACCEL needs to modify system directories, nor does any component need to call privileged system interfaces. (2) Therefore, if you find that it appears that there is some reason why you need to be root, the cause is likely to be outside the SPEC toolset - for example, disk protections, or quota limits. (3) For safe benchmarking, it is better to avoid being root, for the same reason that it is a good idea to wear seat belts in a car: accidents happen, humans make mistakes. For example, if you accidentally type:
when you meant to say:
then you will very grateful if you are not privileged at that moment.
(*) This is only a recommendation, not a requirement nor a rule.