Just wanted to let everyone know we are making a minor change to some of
the tests in the intel test suite. There is a bug where the tests are
passing in a pointer rather then the address of a pointer to the
MPI_Buffer_detach call. Most of the time, this does not cause a problem
as we can see that these tests have been passing for years. However,
there are some cases on the 64-bit sparc platform where we get our
wonderful bus errors. Therefore, we plan to make this minor change to
several of the tests.
sr1-ubur-13 97 =>svn diff MPI_Testany_c.c
Index: MPI_Testany_c.c
===================================================================
--- MPI_Testany_c.c (revision 1220)
+++ MPI_Testany_c.c (working copy)
@@ -100,6 +100,7 @@
char
bsend_buff[NUMMESG * (8 * NUMELM + MPI_BSEND_OVERHEAD + 100)];
+ void * bsend_ptr;
/*----------------------- MPI Initialization
--------------------------*/
/*
@@ -297,7 +298,7 @@
fail++;
}
/* MPI_Test_cancelled Error Test */
- ierr = MPI_Buffer_detach(&bsend_buff, &size);
+ ierr = MPI_Buffer_detach(&bsend_ptr, &size);
if (ierr != MPI_SUCCESS) {
sprintf(info_buf,
"Non-zero return code (%d) from MPI_Buffer_detach",
sr1-ubur-13 98 =>
The tests being modified are:
M MPI_Test_c.c
M MPI_Testany_p_c.c
M MPI_Testall_p_c.c
M MPI_Testsome_p_c.c
M MPI_Testany_c.c
M MPI_Testall_c.c
M MPI_Test_p_c.c
M MPI_Waitsome_p_c.c
M MPI_Waitany_c.c
M MPI_Waitsome_c.c
M MPI_Waitall_c.c
M MPI_Testsome_c.c
M MPI_Waitany_p_c.c
M MPI_Waitall_p_c.c
sr1-ubur-13 99 =>
--
=========================
rolf.vandeva...@sun.com
781-442-3043
=========================