Convenience Scripts for Open|SpeedShop
Offline Mode of Operation
Version
2.0.2
November 9, 2012
Introduction
Open|SpeedShop versions 1.9.2 and beyond contain convenience scripts
that hide some of the underlying options for running the offline
version of Open|SpeedShop. These scripts hide the -f, -offline, and experiment specification options. The
new script names, which correspond to all the Open|SpeedShop
experiments, are:
- osspcsamp
- ossusertime
- osshwc
- osshwctime
- ossio
- ossiot
- ossmpi
- ossmpit
- ossmpiotf
- ossfpe
- osscompare
For example, the command:
osspcsamp
"executable"
is equivalent to the long hand version of executing program counter
sampling experiment on "executable":
openss
-offline
-f "executable" pcsamp
This convenience method is true for all the other Open|SpeedShop
experiments as well. In addition to the above mentioned
change, there are also additional options added to the convenience
commands. The additional options (default, high, low, and
input value(s)) are explained below in an experiment specific
manner.
"executable" is the general
form used for illustration. Real life examples are:
- "mpirun -np 64 sweep3d.mpi"
- "myExecutable < input"
- "/usr/bin/srun -N 64 -n 512
./smg2000 -n 5 5 5"
Use the same syntax that is used to run the executable normally
enclosed in quotes as the "exeutable" and pass it to the convenience
routines, or pass it to openss
-offline prefixed by the -f file/executable designator.
General Usage
Rules
GENERAL usage rules for using the oss<experiment_name>
scripts:
The script always will override the default experiment parameters
and any environment variable settings that are related to the
experiment parameters. The script will warn when it overrides an
experiment variable or default setting.
General and
Experiment Specific Environment Variables
There are several environment variables that affect the
OpenSpeedShop performance experiment execution. The experiment specific
environment variable section is suggested reading before
proceeding to the Experiment Specific Rules sections.
The experiment specific environment variables section describes both
general and experiment specific environment variables that are
references in the following sections.
Experiment
Specific Syntax and Rules
Experiment specific syntax and
rules for using the oss<experiment_name>
scripts follow below. Note that each numbered item
under the experiment section indicates the usage syntax for the oss<experiment_name>
script.
PCSAMP (osspcsamp)
The command options for the osspcsamp
command are as follows.
1) osspcsamp "executable"
If OPENSS_PCSAMP_RATE is not set, osspcsamp runs using the
default sampling rate which is currently 100 samples per second.
If OPENSS_PCSAMP_RATE is set, osspcsamp runs using the
sampling rate specified by the OPENSS_PCSAMP_RATE
environment variable.
2) osspcsamp "executable" default
If OPENSS_PCSAMP_RATE is not set, osspcsamp runs using the
default sampling rate which is currently 100 samples per second.
If OPENSS_PCSAMP_RATE is set, osspcsamp runs using the
default sampling rate which is currently 100 samples per second
and warns that it is not honoring the sampling rate specified by
the OPENSS_PCSAMP_RATE
environment variable.
3) osspcsamp "executable" high
If OPENSS_PCSAMP_RATE is not set, osspcsamp overrides the
default sampling rate by multiplying it by 4 (4 * 100), so 400 is
used as the sampling rate
If OPENSS_PCSAMP_RATE is set, osspcsamp overrides the
environment variable sampling rate by using the default sampling
rate multiplied by 4 (4 *100), so 400 is used as the sampling rate
4) osspcsamp "executable" low
If OPENSS_PCSAMP_RATE is not set, osspcsamp overrides the
default sampling rate by dividing by 2, so 50 is used as the
sampling rate
If OPENSS_PCSAMP_RATE is set, osspcsamp overrides the
environment variable sampling rate by using the default sampling
rate dividing by 2, so 50 is used as the sampling rate
5) osspcsamp "executable" <input_rate_value>
If OPENSS_PCSAMP_RATE is not set, osspcsamp overrides the
default sampling rate by using the <input_rate_value> value as the sampling
rate.
If OPENSS_PCSAMP_RATE is set, osspcsamp overrides the
environment variable sampling rate by using <input_rate_value> value
as the sampling rate.
Default Arguments:
The default sampling rate for the
PCSAMP experiment is currently 100 samples per second.
Example usage:
osspcsamp
"mpirun -np 64 nbody_exe"
USERTIME
(ossusertime)
The command options for the ossusertime
command are as follows.
1) ossusertime "executable"
If OPENSS_USERTIME_RATE is
not set, ossusertime runs
using the default sampling rate which is currently 35 samples per
second.
If OPENSS_USERTIME_RATE is
set, ossusertime runs using
the sampling rate specified by the OPENSS_USERTIME_RATE
environment variable.
2) ossusertime "executable"
default
If OPENSS_USERTIME_RATE is
not set, ossusertime runs
using the default sampling rate which is currently 35 samples per
second.
If OPENSS_USERTIME_RATE is
set, ossusertime runs using
the default sampling rate which is currently 35 samples per second
and warns that it is not honoring the sampling rate specified by the
OPENSS_USERTIME_RATE
environment variable.
3) ossusertime "executable"
high
If OPENSS_USERTIME_RATE is
not set, ossusertime
overrides the default sampling rate by multiplying by 4, so 140 is
used as the sampling rate
If OPENSS_USERTIME_RATE is
set, ossusertime overrides
the environment variable sampling rate by using the default sampling
rate multiplied by 4, so 140 is used as the sampling rate
4) ossusertime "executable" low
If OPENSS_USERTIME_RATE is not set, ossusertime overrides the
default sampling rate by dividing by 2, so 17 is used as the
sampling rate
If OPENSS_USERTIME_RATE is set, ossusertime overrides the
environment variable sampling rate by using the default sampling
rate dividing by 2, so 17 is used as the sampling rate
5) ossusertime "executable" <input_rate_value>
If OPENSS_USERTIME_RATE is not set, ossusertime overrides the
default sampling rate by using the <input_rate_value> value as the sampling
rate.
If OPENSS_USERTIME_RATE is set, ossusertime overrides the
environment variable sampling rate by using <input_rate_value> value
as the sampling rate.
Default Arguments:
The default sampling rate for the
USERTIME experiment is currently 35 samples per second.
Example usage:
ossusertime
"mpirun -np 64 nbody_exe"
HWC (osshwc)
The command options for the osshwc
command are as follows.
0) osshwc
This command lists the available PAPI events that are available to be used as arguments for OPENSS_HWC_EVENT in the command variations below.
1) osshwc "executable"
If OPENSS_HWC_EVENT is not set, osshwc will run with default
papi event which is currently PAPI_TOT_CYC
If OPENSS_HWC_EVENT is set, osshwc warns the user that it
is overriding the environment variable OPENSS_HWC_EVENT setting.
If OPENSS_HWC_THRESHOLD is not set, osshwc will run using the
default papi event threshold which is currently 10000
If OPENSS_HWC_THRESHOLD is set, osshwc warns the user that it
is overriding the default and is using the environment variable OPENSS_HWC_THRESHOLD setting.
2) osshwc "executable" default
If OPENSS_HWC_EVENT is not set, osshwc will run with default
papi event which is currently PAPI_TOT_CYC
If OPENSS_HWC_EVENT is set, osshwc will run with default
papi event which is currently PAPI_TOT_CYC
and warns that it is not using the papi event specified by the OPENSS_HWC_EVENT environment
variable.
If OPENSS_HWC_THRESHOLD is not set, osshwc will run using the
default papi event threshold which is currently 10000
If OPENSS_HWC_THRESHOLD is set, osshwc will run using the
default papi event threshold which is currently 10000
3) osshwc "executable" <input_papi_event_name>
If OPENSS_HWC_EVENT is not set, osshwc will override the
default papi event value and use the <input_papi_event_name> event
If OPENSS_HWC_EVENT is set, osshwc warns the user that it
is overriding the environment variable OPENSS_HWC_EVENT setting.
If OPENSS_HWC_THRESHOLD is not set, osshwc will run using the
default papi event threshold which is currently 10000
If OPENSS_HWC_THRESHOLD is set, osshwc warns the user that it
is overriding the default and is using the environment variable OPENSS_HWC_THRESHOLD setting.
4) osshwc "executable" <input_papi_threshold>
If OPENSS_HWC_EVENT is not set, osshwc will run with default
papi event which is currently PAPI_TOT_CYC
If OPENSS_HWC_EVENT is set, osshwc will warn the user that
it is overriding the default papi event and is using the papi
event specified by OPENSS_HWC_EVENT
If OPENSS_HWC_THRESHOLD is not set, osshwc warns the user that it
is overriding the default papi threshold value and is using the <input_papi_threshold>
value
If OPENSS_HWC_THRESHOLD is set, osshwc warns the user that it
is overriding the environment variable OPENSS_HWC_THRESHOLD setting and is using the <input_papi_threshold>
value.
5) osshwc "executable" <input_papi_event_name> <input_papi_threshold>
If OPENSS_HWC_EVENT is not set, osshwc will override the
default papi event value and use the <input_papi_event_name> event
If OPENSS_HWC_EVENT is set, osshwc warns the user that it
is overriding the environment variable OPENSS_HWC_EVENT setting.
If OPENSS_HWC_THRESHOLD is not set, osshwc will override the
default papi threshold value and use the
<input_papi_threshold> value
If OPENSS_HWC_THRESHOLD is set, osshwc warns the user that it
is overriding the environment variable OPENSS_HWC_THRESHOLD setting and uses the
<input_papi_threshold> value.
Default Arguments:
The default papi event for the HWC
experiment is currently PAPI_TOT_CYC for the hardware counter event and the default papi event threshold is
currently 10000.
Example usage:
osshwc
"mpirun -np 64 nbody_exe" PAPI_L1_DCM 20000
if the user wanted to track the Level 1 data cache misses in their
application at double the default hardware counter threshold
value.
HWCSAMP
(osshwcsamp)
The command options for the osshwcsamp
command are as follows.
0) osshwcsamp
This command lists the available PAPI events that are available to
be used as arguments for OPENSS_HWCSAMP_EVENTS
in the command variations below.
1) osshwcsamp "executable"
If OPENSS_HWCSAMP_EVENTS is not set, osshwcsamp will run with
default papi event list which is currently PAPI_TOT_CYC
If OPENSS_HWCSAMP_EVENTS is set, osshwcsamp will use the papi
event specified by OPENSS_HWCSAMP_EVENTS
If OPENSS_HWCSAMP_RATE is not set, osshwcsamp will run using the
default papi event list sampling rate which is currently 100
samples per second
If OPENSS_HWCSAMP_RATE is set, osshwcsamp warns the user that
it is overriding the default and is using the environment variable
OPENSS_HWCSAMP_RATE setting.
2) osshwcsamp "executable" default
If OPENSS_HWCSAMP_EVENTS is not set, osshwcsamp will run with
default papi event list which is currently PAPI_TOT_CYC
If OPENSS_HWCSAMP_EVENTS is set, osshwcsamp will use the
default papi event list and warns that it is not honoring the papi
event specified by OPENSS_HWCSAMP_EVENTS
environment variable.
If OPENSS_HWCSAMP_RATE is not set, osshwcsamp will run using the
default papi event list sampling rate which is currently 100
samples per second
If OPENSS_HWCSAMP_RATE is set, osshwcsamp will run using the
default papi event list sampling rate and warns the user that it
not honoring the sampling rate specified by the environment
variable OPENSS_HWCSAMP_RATE
setting.
3) osshwcsamp "executable" <input_papi_event_list>
If OPENSS_HWCSAMP_EVENTS is not set, osshwcsamp will override the
default papi event list value and use the <input_papi_event_list>
event
If OPENSS_HWCSAMP_EVENTS is set, osshwcsamp warns the user that
it is overriding the environment variable OPENSS_HWCSAMP_EVENTS setting.
If OPENSS_HWCSAMP_RATE is not set, osshwcsamp will run using the
default papi event list sampling rate which is currently 100
samples per second
If OPENSS_HWCSAMP_RATE is set, osshwcsamp warns the user that
it is overriding the default and is using the environment variable
OPENSS_HWCSAMP_RATE
setting.
4) osshwcsamp "executable" <input_papi_rate>
If OPENSS_HWCSAMP_EVENTS is not set, osshwcsamp will run with
default papi event list which is currently PAPI_TOT_CYC
If OPENSS_HWCSAMP_EVENTS is set, osshwcsamp will warn the user
that it is overriding the default papi event list and is using the
papi event specified by OPENSS_HWCSAMP_EVENTS
If OPENSS_HWCSAMP_RATE is not set, osshwcsamp warns the user that
it is overriding the default papi sampling rate value and is using
the <input_papi_rate> value
If OPENSS_HWCSAMP_RATE is set, osshwcsamp warns the user that
it is overriding the environment variable OPENSS_HWCSAMP_RATE setting
and is using the <input_papi_rate>
value.
5) osshwcsamp "executable" <input_papi_event_list> <input_papi_rate>
If OPENSS_HWCSAMP_EVENTS is not set, osshwcsamp will override the
default papi event list value and use the <input_papi_event_list>
event
If OPENSS_HWCSAMP_EVENTS is set, osshwcsamp warns the user that
it is overriding the environment variable OPENSS_HWCSAMP_EVENTS setting.
If OPENSS_HWCSAMP_RATE is not set, osshwcsamp will override the
default papi sampling rate value and use the <input_papi_rate> value
If OPENSS_HWCSAMP_RATE is set, osshwcsamp warns the user that
it is overriding the environment variable OPENSS_HWCSAMP_RATE setting
and uses the <input_papi_rate>
value.
Default Arguments:
The default papi event list for
the HWCSAMP experiment is currently PAPI_TOT_CYC for the hardware counter samp event
and the default papi
event list sampling rate is currently 100 samples per second.
Example usage:
osshwcsamp
"mpirun -np 64 nbody_exe" PAPI_RES_STL
if the user wanted to monitor the Cycles stalled on any resource
with the default hardware counter samp ling rate value (100
samples per second).
HWCTIME
(osshwctime)
The command options for the osshwctime
command are as follows.
0) osshwctime
This command lists the available PAPI events that are available to
be used as arguments for OPENSS_HWCTIME_EVENT
in the command variations below.
1) osshwctime "executable"
If OPENSS_HWCTIME_EVENT is not set, osshwctime will run with
default papi event which is currently PAPI_TOT_CYC
If OPENSS_HWCTIME_EVENT is set, osshwctime will use the papi
event specified by OPENSS_HWCTIME_EVENT
If OPENSS_HWCTIME_THRESHOLD is not set, osshwctime will run using the
default papi event threshold which is currently 10000
If OPENSS_HWCTIME_THRESHOLD is set, osshwctime warns the user that
it is overriding the default and is using the environment variable
OPENSS_HWCTIME_THRESHOLD setting.
2) osshwctime "executable" default
If OPENSS_HWCTIME_EVENT is not set, osshwctime will run with
default papi event which is currently PAPI_TOT_CYC
If OPENSS_HWCTIME_EVENT is set, osshwctime will use the
default papi event and warns that it is not honoring the papi
event specified by OPENSS_HWCTIME_EVENT
environment variable.
If OPENSS_HWCTIME_THRESHOLD is not set, osshwctime will run using the
default papi event threshold which is currently 10000
If OPENSS_HWCTIME_THRESHOLD is set, osshwctime will run using the
default papi event threshold and warns the user that it not
honoring the threshold specified by the environment variable OPENSS_HWCTIME_THRESHOLD
setting.
3) osshwctime "executable" <input_papi_event_name>
If OPENSS_HWCTIME_EVENT is not set, osshwctime will override the
default papi event value and use the <input_papi_event_name> event
If OPENSS_HWCTIME_EVENT is set, osshwctime warns the user that
it is overriding the environment variable OPENSS_HWCTIME_EVENT setting.
If OPENSS_HWCTIME_THRESHOLD is not set, osshwctime will run using the
default papi event threshold which is currently 10000
If OPENSS_HWCTIME_THRESHOLD is set, osshwctime warns the user that
it is overriding the default and is using the environment variable
OPENSS_HWCTIME_THRESHOLD
setting.
4) osshwctime "executable" <input_papi_threshold>
If OPENSS_HWCTIME_EVENT is not set, osshwctime will run with
default papi event which is currently PAPI_TOT_CYC
If OPENSS_HWCTIME_EVENT is set, osshwctime will warn the user
that it is overriding the default papi event and is using the papi
event specified by OPENSS_HWCTIME_EVENT
If OPENSS_HWCTIME_THRESHOLD is not set, osshwctime warns the user that
it is overriding the default papi threshold value and is using the
<input_papi_threshold> value
If OPENSS_HWCTIME_THRESHOLD is set, osshwctime warns the user that
it is overriding the environment variable OPENSS_HWCTIME_THRESHOLD
setting and is using the <input_papi_threshold>
value.
5) osshwctime "executable" <input_papi_event_name> <input_papi_threshold>
If OPENSS_HWCTIME_EVENT is not set, osshwctime will override the
default papi event value and use the <input_papi_event_name> event
If OPENSS_HWCTIME_EVENT is set, osshwctime warns the user that
it is overriding the environment variable OPENSS_HWCTIME_EVENT setting.
If OPENSS_HWCTIME_THRESHOLD is not set, osshwctime will override the
default papi threshold value and use the <input_papi_threshold>
value
If OPENSS_HWCTIME_THRESHOLD is set, osshwctime warns the user that
it is overriding the environment variable OPENSS_HWCTIME_THRESHOLD
setting and uses the <input_papi_threshold>
value.
Default Arguments:
The default papi event for the
HWCTIME experiment is currently PAPI_TOT_CYC for the hardware counter time event
and the default papi
event threshold is currently 10000.
Example usage:
osshwctime
"mpirun -np 64 nbody_exe" PAPI_RES_STL
if the user wanted to monitor the Cycles stalled on any resource
with the default hardware counter time threshold value (10000).
IO (ossio)
The command options for the ossio
command are as follows.
1) ossio "executable"
If OPENSS_IO_TRACED is not set, ossio runs using the default
list of I/O functions to trace.
If OPENSS_IO_TRACED is set, ossio runs using the list of
I/O functions specified by the OPENSS_IO_TRACED
environment variable.
2) ossio "executable" default
If OPENSS_IO_TRACED is not set, ossio runs using the
default list of I/O functions to trace.
If OPENSS_IO_TRACED is set, ossio runs using the
default list of I/O functions to trace and warns that it is not
honoring the function list specified by the OPENSS_IO_TRACED environment
variable.
3) ossio "executable" <input_function_trace_list>
If OPENSS_IO_TRACED is not set, ossio overrides the default
list of I/O functions to trace by using the function list
specified by the <input_function_trace_list>
argument.
If OPENSS_IO_TRACED is set, ossio overrides the
environment variable list of I/O functions to trace by using the
function list specified by the <input_function_trace_list>
argument.
Default Arguments:
The default function list for the
IO experiment is currently the list of I/O functions consisting
of:
- close
- creat
- creat64
- dup
- dup2
- lseek
- lseek64
- open
- open64
- pipe
- pread
- pread64
- pwrite
- pwrite64
- read
- readv
- write
- writev
Example usage:
ossio
"mpirun -np 64 nbody_exe" read,write,pipe
if the user only wanted to trace read, write, and pipe.
IOT (ossiot)
The command options for the ossiot
command are as follows.
1) ossiot "executable"
If OPENSS_IOT_TRACED is not set, ossiot runs using the default
list of I/O functions to trace.
If OPENSS_IOT_TRACED is set, ossiot runs using the list of
I/O functions specified by the OPENSS_IOT_TRACED
environment variable.
2) ossiot "executable" default
If OPENSS_IOT_TRACED is not set, ossiot runs using the default
list of I/O functions to trace.
If OPENSS_IOT_TRACED is set, ossiot runs using the default
list of I/O functions to trace and warns that it is not honoring
the function list specified by the OPENSS_IOT_TRACED environment variable.
3) ossiot "executable" <input_function_trace_list>
If OPENSS_IOT_TRACED is not set, ossiot overrides the default
list of I/O functions to trace by using the function list
specified by the <input_function_trace_list>
argument.
If OPENSS_IOT_TRACED is set, ossiot overrides the
environment variable list of I/O functions to trace by using the
function list specified by the <input_function_trace_list>
argument.
Default Arguments:
The default parameter (I/O
function list) for the IOT experiment is currently the list of I/O
functions consisting of:
- close
- creat
- creat64
- dup
- dup2
- lseek
- lseek64
- open
- open64
- pipe
- pread
- pread64
- pwrite
- pwrite64
- read
- readv
- write
- writev
Example usage:
ossiot
"mpirun -np 64 nbody_exe" read,write,pipe
if the user only wanted to trace read, write, and pipe.
MPI (ossmpi)
The command options for the ossmpi
command are as follows.
1) ossmpi "executable"
If OPENSS_MPI_TRACED is not set, ossmpi runs using
the default list of MPI functions to trace.
If OPENSS_MPI_TRACED is set, ossmpi runs using the
list of MPI functions specified by the OPENSS_MPI_TRACED environment variable.
2) ossmpi "executable" default
If OPENSS_MPI_TRACED is not set, ossmpi runs using the default
list of MPI functions to trace.
If OPENSS_MPI_TRACED is set, ossmpi runs using the default
list of MPI functions to trace and warns that it is not honoring
the function list specified by the OPENSS_MPI_TRACED environment variable.
3) ossmpi "executable" <input_function_trace_list>
If OPENSS_MPI_TRACED is not set, ossmpi overrides the default
list of MPI functions to trace by using the function list
specified by the <input_function_trace_list>
argument.
If OPENSS_MPI_TRACED is set, ossmpi overrides the
environment variable list of MPI functions to trace by using the
function list specified by the <input_function_trace_list>
argument.
Default Arguments:
The default parameters for the MPI
experiment is currently the list of MPI functions consisting of:
- MPI_Allgather
- MPI_Allgatherv
- MPI_Allreduce
- MPI_Alltoall
- MPI_Alltoallv
- MPI_Barrier
- MPI_Bcast
- MPI_Bsend
- MPI_Bsend_init
- MPI_Cancel
- MPI_Cart_create
- MPI_Cart_sub
- MPI_Comm_create
- MPI_Comm_dup
- MPI_Comm_free
- MPI_Comm_split
- MPI_Finalize
- MPI_Gather
- MPI_Gatherv
- MPI_Get_count
- MPI_Graph_create
- MPI_Ibsend
- MPI_Init
- MPI_Intercomm_create
- MPI_Intercomm_merge
- MPI_Iprobe
- MPI_Irecv
- MPI_Irsend
- MPI_Isend
- MPI_Issend
- MPI_Pack
- MPI_Probe
- MPI_Recv
- MPI_Recv_init
- MPI_Reduce
- MPI_Reduce_scatter
- MPI_Request_free
- MPI_Rsend
- MPI_Rsend_init
- MPI_Scan
- MPI_Scatter
- MPI_Scatterv
- MPI_Send
- MPI_Sendrecv
- MPI_Sendrecv_replace
- MPI_Ssend
- MPI_Ssend_init
- MPI_Start
- MPI_Startall
- MPI_Test
- MPI_Testall
- MPI_Testany
- MPI_Testsome
- MPI_Unpack
- MPI_Wait
- MPI_Waitall
- MPI_Waitany
- MPI_Waitsome
And additionally, one of the MPI groupings can be specified by using
the "Acceptable Argument" values listed in the table below.
MPI Category
|
Acceptable Argument
|
All the MPI Functions |
all |
Collective Communicators |
collective_com |
Persistent_Communicators |
persistent_com |
Synchronous Point to Point |
synchronous_p2p |
Asynchronous Point to Point |
asynchronous_p2p |
Process Topologies |
process_topologies |
Groups Contexts Communicators |
graphs_contexts_comms |
Environment |
environment |
Datatypes |
datatypes |
Example usage:
ossmpi
"mpirun -np 64 nbody_exe" MPI_Scatter, MPI_Scatterv
if the user only wanted to trace the MPI scatter functions.
MPIT (ossmpit)
The command options for the ossmpit
command are as follows.
1) ossmpit "executable"
If OPENSS_MPIT_TRACED is not set, ossmpit runs using the default
list of MPI functions to trace.
If OPENSS_MPIT_TRACED is set, ossmpit runs using the list of
MPI functions specified by the OPENSS_MPIT_TRACED
environment variable.
2) ossmpit "executable" default
If OPENSS_MPIT_TRACED is not set, ossmpit runs using the default
list of MPI functions to trace.
If OPENSS_MPIT_TRACED is set, ossmpit runs using the default
list of MPI functions to trace and warns that it is not honoring
the function list specified by the OPENSS_MPIT_TRACED environment variable.
3) ossmpit "executable" <input_function_trace_list>
If OPENSS_MPIT_TRACED is not set, ossmpit overrides
the default list of MPI functions to trace by using the function
list specified by the <input_function_trace_list>
argument.
If OPENSS_MPIT_TRACED is set, ossmpit overrides the
environment variable list of MPI functions to trace by using the
function list specified by the <input_function_trace_list>
argument.
Default Arguments:
The default parameters for the
MPIT experiment is currently the list of MPI functions consisting
of:
- MPI_Allgather
- MPI_Allgatherv
- MPI_Allreduce
- MPI_Alltoall
- MPI_Alltoallv
- MPI_Barrier
- MPI_Bcast
- MPI_Bsend
- MPI_Bsend_init
- MPI_Cancel
- MPI_Cart_create
- MPI_Cart_sub
- MPI_Comm_create
- MPI_Comm_dup
- MPI_Comm_free
- MPI_Comm_split
- MPI_Finalize
- MPI_Gather
- MPI_Gatherv
- MPI_Get_count
- MPI_Graph_create
- MPI_Ibsend
- MPI_Init
- MPI_Intercomm_create
- MPI_Intercomm_merge
- MPI_Iprobe
- MPI_Irecv
- MPI_Irsend
- MPI_Isend
- MPI_Issend
- MPI_Pack
- MPI_Probe
- MPI_Recv
- MPI_Recv_init
- MPI_Reduce
- MPI_Reduce_scatter
- MPI_Request_free
- MPI_Rsend
- MPI_Rsend_init
- MPI_Scan
- MPI_Scatter
- MPI_Scatterv
- MPI_Send
- MPI_Sendrecv
- MPI_Sendrecv_replace
- MPI_Ssend
- MPI_Ssend_init
- MPI_Start
- MPI_Startall
- MPI_Test
- MPI_Testall
- MPI_Testany
- MPI_Testsome
- MPI_Unpack
- MPI_Wait
- MPI_Waitall
- MPI_Waitany
- MPI_Waitsome
And additionally, one of the MPI groupings can be specified by using
the "Acceptable Argument" values listed in the table below.
MPI Category
|
Acceptable Argument
|
All the MPI Functions |
all |
Collective Communicators |
collective_com |
Persistent_Communicators |
persistent_com |
Synchronous Point to Point |
synchronous_p2p |
Asynchronous Point to Point |
asynchronous_p2p |
Process Topologies |
process_topologies |
Groups Contexts Communicators |
graphs_contexts_comms |
Environment |
environment |
Datatypes |
datatypes |
Example usage:
ossmpit
"mpirun -np 64 nbody_exe" MPI_Scatter, MPI_Scatterv
if the user only wanted to trace the MPI scatter functions.
MPIOTF (ossmpiotf)
The command options for the ossmpiotf
command are as follows.
1) ossmpiotf "executable"
If OPENSS_MPIOTF_TRACED is not set, ossmpiotf runs using the
default list of MPI functions to trace.
If OPENSS_MPIOTF_TRACED is set, ossmpiotf runs using the list
of MPI functions specified by the OPENSS_MPIOTF_TRACED environment variable.
2) ossmpiotf "executable" default
If OPENSS_MPIOTF_TRACED is not set, ossmpiotf runs using the
default list of MPI functions to trace.
If OPENSS_MPIOTF_TRACED is set, ossmpiotf runs using the
default list of MPI functions to trace and warns that it is not
honoring the function list specified by the OPENSS_MPIOTF_TRACED
environment variable.
3) ossmpiotf "executable" <input_function_trace_list>
If OPENSS_MPIOTF_TRACED is not set, ossmpiotf overrides the
default list of MPI functions to trace by using the function list
specified by the <input_function_trace_list>
argument.
If OPENSS_MPIOTF_TRACED is set, ossmpiotf overrides the
environment variable list of MPI functions to trace by using the
function list specified by the <input_function_trace_list>
argument.
Default Arguments:
The default parameters for the
MPIOTF experiment is currently the list of MPI functions
consisting of:
- MPI_Allgather
- MPI_Allgatherv
- MPI_Allreduce
- MPI_Alltoall
- MPI_Alltoallv
- MPI_Barrier
- MPI_Bcast
- MPI_Bsend
- MPI_Bsend_init
- MPI_Cancel
- MPI_Cart_create
- MPI_Cart_sub
- MPI_Comm_create
- MPI_Comm_dup
- MPI_Comm_free
- MPI_Comm_split
- MPI_Finalize
- MPI_Gather
- MPI_Gatherv
- MPI_Get_count
- MPI_Graph_create
- MPI_Ibsend
- MPI_Init
- MPI_Intercomm_create
- MPI_Intercomm_merge
- MPI_Iprobe
- MPI_Irecv
- MPI_Irsend
- MPI_Isend
- MPI_Issend
- MPI_Pack
- MPI_Probe
- MPI_Recv
- MPI_Recv_init
- MPI_Reduce
- MPI_Reduce_scatter
- MPI_Request_free
- MPI_Rsend
- MPI_Rsend_init
- MPI_Scan
- MPI_Scatter
- MPI_Scatterv
- MPI_Send
- MPI_Sendrecv
- MPI_Sendrecv_replace
- MPI_Ssend
- MPI_Ssend_init
- MPI_Start
- MPI_Startall
- MPI_Test
- MPI_Testall
- MPI_Testany
- MPI_Testsome
- MPI_Unpack
- MPI_Wait
- MPI_Waitall
- MPI_Waitany
- MPI_Waitsome
And additionally, one of the MPI groupings can be specified by using
the "Acceptable Argument" values listed in the table below.
MPI Category
|
Acceptable Argument
|
All the MPI Functions |
all |
Collective Communicators |
collective_com |
Persistent_Communicators |
persistent_com |
Synchronous Point to Point |
synchronous_p2p |
Asynchronous Point to Point |
asynchronous_p2p |
Process Topologies |
process_topologies |
Groups Contexts Communicators |
graphs_contexts_comms |
Environment |
environment |
Datatypes |
datatypes |
Example usage:
ossmpiotf
"mpirun -np 64 nbody_exe" MPI_Scatter, MPI_Scatterv
if the user only wanted to trace the MPI scatter functions.
FPE (ossfpe)
The command options for the ossfpe
command are as follows.
1) ossfpe "executable"
If OPENSS_FPE_EVENT is not set, ossfpe runs using the default list of floating point events to monitor.
If OPENSS_FPE_EVENT is set, ossfpe runs using the list of floating point events to monitor specified by the OPENSS_FPE_EVENT environment variable.
2) ossfpe "executable" default
If OPENSS_FPE_EVENT is not set, ossfpe runs using the default list of floating point events to monitor.
If OPENSS_FPE_EVENT is set, ossfpe runs using the default list of floating point events to monitor and warns that it is not honoring the events specified by the OPENSS_FPE_EVENT environment variable.
3) ossfpe "executable" <input_function_trace_list>
If OPENSS_FPE_EVENT is not set, ossfpe overrides the default list of floating point events to monitor by using the function list specified by the <input_function_trace_list> argument.
If OPENSS_FPE_EVENT is set, ossfpe overrides the environment variable list of floating point events monitor by using the function list specified by the <input_function_trace_list> argument.
Default Arguments:
The default parameters for the FPE
experiment is currently the list of floating point exception
events consisting of:
- inexact_result
- division_by_zero
- underflow
- overflow
- invalid_operation
Example usage:
ossfpe
"mpirun -np 64 nbody_exe" division_by_zero
if the user only wanted to trace the division by zero floating point
exception.
COMPARE
(osscompare)
The command options for the osscompare
command are as follows.
1) osscompare "database_file, database_file [, database_file] ..."
osscompare runs using the default comparison metric, exclusive CPU time, comparing the performance data from each of the experiments represented by the datbase files specified.
2) osscompare "database_file, database_file [, database_file] ..." [time | percent]
osscompare runs using the either, the time (exclusive CPU time) or percentage of CPU time, comparison metric: exclusive CPU time, comparing the performance data from each of the experiments represented by the datbase files specified.
3) osscompare " as above in 1) or 2) "
If OPENSS_CREATE_CSV is not set, osscompare generates the comparison output to stdout.
If OPENSS_CREATE_CSV is set, osscompare generates the comparison output to stdout and creates two additional files in the directory the command was run in. These two files contain:
- File 1: contains the same stdout comparison output that
contains a side-by-side comparison
- File 2: contains a comma separated list that corresponds to
the side-by-side comparison output, but can be input into
a spreadsheet program.
Default Arguments:
The default parameters for the
osscompare:
- The comparison metric is:
exclusive CPU time
- The view type is: Function
level
- Output filename is
internally generated
- Row count is unlimited.
These are changable by using the osscompare arguments for metric,
viewtype, oname and rows.
Example usage:
osscompare
"smg2000-pcsamp-run1.openss, smg2000-pcsamp-run2.openss"