[Mpi-forum] First reading: MPI_T / Tools Chapter
Adam T. Moody
moody20 at llnl.gov
Wed Mar 16 16:10:20 CDT 2011
Hi Martin,
Here is round two of my review (the second half of the proposal).
First, in yesterday's email, I suggested the following change:
p 12, line 13
Change
"This routine returns, if datatpye represents a valid enumeration
datatype, N representing the range of the enumeration 0 to N-1 as well
as a string with a name for it."
"The arguments name and ..."
to
"If datatype is a valid enumeration datatype, this routine returns
the number of items in the enumeration range and the name of the
datatype. If the range runs from 0 to N-1, the value N is returned in
num. The arguments name and ..."
But now I think a better way to write this is:
"If the argument datatype is a valid enumeration datatype, this
routine returns the number of items in the enumeration range and the
name of the datatype. The number of items, N, is returned in the
argument num. If N > 0, then the enumeration range runs from 0 to N-1.
If N = 0, then the enumeration range is empty. The arguments name and ..."
Maybe we don't need to worry about N=0?
Alright, now here are comments for the second half of the document.
p 16, line 43
"local to a single MPI process" --> "local to an MPI process"
p 16, line 46
"the interface has a cleaner semantics" --> "the interface provides
cleaner semantics"
p 17, line 2
Extra spaces?
"Extra performance ... with a class describing ..."
p 17, line 2
"describing its the" --> "describing its"
p 17, lines 2-5
Could reword these sentences a bit
"Each performance variable is associated with a class describing its
the basic semantics.
The class of a variable also de nes its basic behavior, when and how an
MPI implementation
can change its value, and what the initial value of this variable is at
the time it is either used
for the first time or reset. In the following this referred to as the
starting value."
How about?
"Each performance variable is associated with a class that describes
its basic semantics, basic behavior, its starting value, and when and
how an MPI implementation can changes its value. The starting value is
the value the variable assumes when it is used for the first time or
whenever it is reset."
p 17, line 6
Change
"Further, it ... to represent it."
to
"Further, the class of a variable ... to represent the variable."
p 17, line 11
"Variables of thie class are expected to be ..."
Can they be something other than an enumeration datatype? If not, I
would drop the "expected to be" phrase in this sentence. If so, I would
list what other types they may be.
p 17, lines 10-45
Change the pattern of
"The default starting value is ... (at the time the starting value
is set) of the resource."
to
"The default starting value is ... of the resource at the time the
starting value is set."
p 17, line 10 - p 18, line 27
Change the pattern of
"The default starting value"
to just
"The starting value"
The term "starting value" is well defined in the intro, and adding
"default" can be confusing.
p 17, line 26
Change
"It will be returned as an MPI_T_DOUBLE_DATATYPE"
to
"It is represented by an MPI_T_DOUBLE datatype."
p 18, line 1
Drop "during the execution time of an application". Perhaps, I'm
wrong, but I don't think this phrase is necessary. It can be
misinterpretted to mean over the duration of the application, which can
be wrong since the variable may count events only through some window of
execution.
p 18, line 3
Do you really need this clause?
"by one for each specific event that is observed".
If not, I'd suggest dropping it.
p 18, line 24
"variables if this class" --> "variables of this class"
p 18, line 26
Why only specify the units for double?
How does one determine the units for MPI_T_INT and MPI_T_LONG_LONG?
p 18 , line 35
Change
"of an MPI application when new variables become available through
dynamic loading."
to
"of an MPI application when new variables become available (e.g.,
through dynamic loading)."
A similar statement is made for control variables on p 13, line 11.
p 18, line 35
Extra space?
"an MPI application when"
p 19, line 36
Change
"The argument verbosity returns the verbosity level (see Section
1.3.1) assigned by the MPI implementation to the variable."
to
"The argument verbosity returns the verbosity level of the variable
(see Section 1.3.1)."
p 19, line 38
Can it be anything other that what's in the list?
"in the parameter var_class and can be"
If not, I'd make this statement stronger, also break into two sentences.
"in the parameter var_class. The class value must be"
p 14, line 10
Same as above
p 19, line 40
Change
"The argument datatype returns the MPI_T datatype in which the value
for this performance variable is returned."
to
"The argument datatype returns the MPI_T datatype that is used to
represent the performance variable."
p 14, line 12
Same as above
p 19, line 42
"control" --> "performance"
p 20, line 13
Change
"Upon return, the argument readonly will be set to zero if the
variable can be written or reset by the user, or one if the variable can
only be read."
to
"Upon return, the argument readonly is set to zero if the variable
can be written or reset by the user. It is set to one if the variable
can only be read."
Maybe change "zero" and "one" to "true" and "false", which is how flags
in other MPI functions are described?
p 20, line 25-28
Break this into two senences like above. And again, change "will be
set" to "is set".
p 20, line 27
"automatically active" --> "always active" or "continuously active"
p 20, line 27
"can not by" --> "cannot be"
p 21, line 6
"existing section, i.e., calls to ... of the freed session" -->
"existing session. Calls to ... of a session after it is freed."
p 21, line 14
Change
"The type of MPI object is returned by a previous call to
MPI_T_PVAR_GET_INFO in the bind argument."
to
"The type of MPI object a variable may be bound to is determined by
calling MPI_T_PVAR_GET_INFO."
p 21, line 29-36
Apply changes from earlier email about control vars.
p 22, line 5
Change
"Performance variables that have the continuous flag set during the
query operation are continuously operating once a handle has been
allocated and can be queried any time. They cannot be stopped or paused
by the user."
to
"Performance variables that have the continuous flag set during the
query operation are continuously operating once a handle has been
allocated. Such variables may be queried at any time, but they may not
be stopped or paused by the user."
p 22, line 6
Change
"All other variables are in a stopped state after their handle has
been allocated; their values are not updated as the program executes,
and must be started by the user."
to
"All other variables are in a stopped state after their handle has
been allocated; their values are not updated until they have been
started by the user."
p 22, line 25
Change
"are ignored when used with MPI_T_PVAR_ALL_HANDLES."
to
"are ignored when MPI_T_PVAR_ALL_HANDLES is specified."
p 22, line 40
Same as above
p 23, line 13
"size and fits" --> "size to hold" as described in previous email
p 23, line 15 and line 35
These statements feel more like rationale. You could simply say:
"The constant MPI_T_PVAR_ALL_HANDLES cannot be used as an argument
for MPI_T_PVAR_WRITE."
p 23, line 46
"sets of the" --> "sets the"
p 23, line 47
Extra space?
" to its starting"
p 24, line 4
Extra space?
"Readonly variables are"
p 24, line 4
Change
"are ignored when used with MPI_T_PVAR_ALL_HANDLES."
to
"are ignored when MPI_T_PVAR_ALL_HANDLES is specified."
p 24, line 18
Same rationale feeling text here.
p 24, line 18
Since a user must just call READ and RESET, what is the purpose of this
function? Atomicity, convenience, or both? I think some explanation
for the user would be useful here.
p 24, line 26
"distinguishing" --> "distinguish"
p 25, line 4
Dynamic loading again (see comment for p 18, line 35)
p 27, line 9-13
Change
"The functions will only write up to len elements into the
respective array. If the category contains more than len variables or
other categories respectively , the function returns an arbitrary
subset; if it contains less than len variables
or other categories respectively, all will be returned and the remaining
array entries will not be modified."
to
"Starting from array index 0, each function writes up to len
elements into the array. If the category contains more than len
elements, the function returns an arbitrary subset. If the category
contains less than len elements, the set of elements is returned in the
first len array entries, and the remaining array entries are not modified."
p 27, line 17
Table 1.3.9??
p 27, line 17-19
Change
"None of the error codes returned by an MPI_T routine are fatal to
the MPI process or invoke an MPI error handler."
to
"Errors returned by MPI_T routines are not fatal and do not invoke
MPI error handlers."
p 27, line 21
This seems to contradict the previous statements somewhat. If it's
undefined, then an implementation *may* be fatal. So perhaps we should
say that errors are not required to be fatal, or they are not fatal so
long as the input parameters are valid, or we recommend that
implementors avoid making them fatal ...
p 27, line 26
Switch "complying" and extra "s" on "implementation"
"a complying MPI implemetations" --> "a compliant MPI_T implementation"
-Adam
Martin Schulz wrote:
>In preparation for the first reading at the March MPI forum,
>I have updated ticket #266, the proposal for the MPI Tool
>Information Interface and the restructuring of the profiling
>or tools chapter.
>
>https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/266
>
>Changes compared to the document at the February
>meeting are marked in green, changes caused by the
>name change requested by the forum are marked in red.
>
>Martin
>
>
><https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/269>________________________________________________________________________
>Martin Schulz, schulzm at llnl.gov <mailto:schulzm at llnl.gov>,
>http://people.llnl.gov/schulzm
>CASC @ Lawrence Livermore National Laboratory, Livermore, USA
>
>
>
>
>
>------------------------------------------------------------------------
>
>_______________________________________________
>mpi-forum mailing list
>mpi-forum at lists.mpi-forum.org
>http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum
>
>
More information about the mpi-forum
mailing list