[mpich2-dev] Coverage annotations and mpid_rma.c

[Rajeev Thakur]

Yes, there should be some annotation for the coverage test. I am ok with
option 3 where all this gets moved to the upper layer.
 
Rajeev


  _____  

From: owner-mpich2-dev at mcs.anl.gov [mailto:owner-mpich2-dev at mcs.anl.gov] On
Behalf Of William Gropp
Sent: Monday, April 28, 2008 12:31 PM
To: mpich2-dev at mcs.anl.gov
Subject: [mpich2-dev] Coverage annotations and mpid_rma.c


I was looking at some of the files that have substantial uncovered sections,
and one such file is ch3/src/mpid_rma.c . All of the uncovered code in that
file is of the form 

if (! initialized) { initialize(); }

This raises a number of questions:

1) Should there be a coverage annotation for this case (e.g.,
BEGIN_OPTIONALINIT ... END_OPTIONALINIT)?


2) In almost all cases, these initializes cannot happen in correct code
(e.g., some other routine must have been called first, such as
MPI_Win_create). Should these be marked as BEGIN ERROR HANDLING .. END ERROR
HANDLING instead? If so, should it be in the #if error handling defintions?

3) This entire file is used to provide for and make available a function
table for the RMA functions (except MPID_Win_test). This adds an extra
function call to all RMA operations, including latency-sensitive operations
such as lock or put. Would it make sense to promote something like this
interface into the top-level routines (e.g., MPI_Win_lock), perhaps with a
macro that would basically say "either convert to a specific routine or
convert to a reference to a standardized function table"? That would then
eliminate this file from ch3 entirely.

Bill



William Gropp
Paul and Cynthia Saylor Professor of Computer Science
University of Illinois Urbana-Champaign




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mcs.anl.gov/mailman/private/mpich2-dev/attachments/20080509/d786a161/attachment.htm>