Rob Latham authored Oct 18, 2013 and Pavan Balaji committed Oct 31, 2013

MPI_Pack, MPI_Unpack, and MPI_Pack_size use integers to describe counts
of bytes and posistions in a bytestream.  In a world of 2 GiB+
datatypes, this is not correct.  It will take an army to change the
standard, but we can at least do the right thing internally.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Rob Latham authored Oct 28, 2013 and Pavan Balaji committed Oct 31, 2013

MPIR_Type_get_contig_blocks is buggy for certain datatypes (example:
HDF5 multi-dataset IO generated type results in a contig block count of
5 billion.  should be more like 200).  Nothing uses this function except
ROMIO's flattening code.  I'd like to see profiling suggesting the count
of contiguous blocks is a profitable short-circuit before spending a ton
of time debugging it.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Rob Latham authored Oct 18, 2013 and Pavan Balaji committed Oct 31, 2013

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Rob Latham authored Oct 18, 2013 and Pavan Balaji committed Oct 31, 2013

clang carped about too many parens around (( a == b))

Fixes #1929

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Rob Latham authored Oct 16, 2013 and Pavan Balaji committed Oct 31, 2013

formatting fixes and type promotions to chip away at the 1500 lines of
warnings that pop up with clang's -Wshorten-64-to-32 flag.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Jeff Hammond authored Oct 15, 2013 and Pavan Balaji committed Oct 31, 2013

goal: make MPICH count-safe
problem: casting to int when value exceeds INT_MAX
solution: remove all "(int)" in front of values assigned to the count
field

Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Jeff Hammond authored Oct 15, 2013 and Pavan Balaji committed Oct 31, 2013

strlen returns size_t.  since the return argument is localized, there is
no reason to not use the proper type to hold the result.

Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Rob Latham authored Oct 15, 2013 and Pavan Balaji committed Oct 31, 2013

this internal-to-romio code was using 'ints' to store things like
"datatype_size * count" calculation.  Promote that and all called paths
to ADIO_Offset

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Rob Latham authored Aug 27, 2013 and Pavan Balaji committed Oct 31, 2013

MPIO_CHECK_COUNT_SIZE was designed to ensure "count*datatype_size" did
not overflow integers.  Well, I've done my best to relax that need
throughout MPICH, so the cast to unsigned is not needed.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Rob Latham authored Jul 29, 2013 and Pavan Balaji committed Oct 31, 2013

The test suite has failures like this:

Assertion failed in file
/home/robl/work/mpich/src/mpid/ch3/src/ch3u_rma_ops.c at line 552
: ((dloop_vec[i].iov_len)/type_size) >= expr_inttype_min(count)

Even though dloop_vec[i].iov_len is 4, type_size is 4, and
expr_inttype_min(count) is -INT_MAX, this assertion is still failing.

while the _Generic-based test was giving strange assertion failures, the
cast-and-compare version still works and should be used.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Rob Latham authored Jul 19, 2013 and Pavan Balaji committed Oct 31, 2013

Not much in MPICH consults the max_contig_count field of the datatypes,
except ROMIO.  Resized types were reporting "huge negative number" for
the count of contiguous blocks because datatype code wasn't setting the
max_contig_count member.  Why 3?  The lower bound and upper bound parts
of the datatype, while not actually data or blocks, are effectively
blocks for the purposes of the one client of this code -- the ROMIO
datatype flattening code.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

When flattening a datatype, be mindful of overflowing integers

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Trying to find all places in the datatype/dataloop code that put size of
datatype into an integer.  It's easy, even under MPI-2, to describe more
than an integers worth of data with an MPI datatype.

since Dataloop code calls other dataloop code, just go ahead and make
everything a DLOOP_Count type

Fixes #1890
Fixes #1893

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

MPI-3 introduced MPI_Count and MPI_Type_size_x.  Use these new routines
throughout ROMIO to support large datatypes.

Still to do:
- update other ADIO drivers

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Dancing around with various schemes to detect/deal-with compiler padding
in structs so the 'payload' field ends up double-aligned.  The new
"large type , large count" work introduced a pointer into the nemisis
packet, so we have to be a bit more dynamic about how or if we pad out
the structure.

- put the members of MPID_nem_pkt_header inside a struct, so subsequent
  payload does not "fall into" compiler-introduced padding
- use this updated datatype to determine the size, and if we need to pad
  or not.
- use the result of the configure test to introduce padding, or omit the
  padding from the definition if the amount of padding is '0'

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Dave Goodell authored Apr 04, 2013 and Rob Latham committed Oct 31, 2013

Also includes random fixes to `-Wshorten-64-to-32` warnings which might
need to be teased out.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

No reviewer.

No reviewer.

We were downgrading void * content to int and then upgrading it back
to an MPIR_Pint, losing bits in the process.  This worked fine when
Fint was just 4 bytes, but not when Fint was 8 bytes.

Signed-off-by: William Gropp <wgropp@illinois.edu>

When freeing arrays, we were not checking for the case where the array
might be set to MPI_WEIGHTS_EMPTY.

Signed-off-by: William Gropp <wgropp@illinois.edu>

When MPI_Count was added to the status field, a bunch of bogus checks
got added into configure and MPI_Init.  This patch cleans up these
checks.

Signed-off-by: William Gropp <wgropp@illinois.edu>

An initial version of this patch was provided by Intel.

Signed-off-by: William Gropp <wgropp@illinois.edu>

Based on an Intel contributed patch.  The idea is to use the bits from
the cancelled field to extend the count, rather than increasing the
count datatype itself.

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Fixes to the bit manipulation based on feedback from Artem Yalozo @
Intel.

Fixes to the naming convention based on feedback from Bill Gropp.

Signed-off-by: William Gropp <wgropp@illinois.edu>

This reverts commit 81ad170b

.

Signed-off-by: William Gropp <wgropp@illinois.edu>

William Gropp authored Oct 23, 2013 and Pavan Balaji committed Oct 29, 2013

This selects the buildiface option that specifies a specific ordering
of the routines in mpicxx.h(.in) in order to provide ABI consistency.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

William Gropp authored Oct 23, 2013 and Pavan Balaji committed Oct 29, 2013

For MPICH builds, eliminate the C++ distgraphcomm test.  In other cases
test for presense of MPI::Distgraphcomm and only include the test if
available.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

William Gropp authored Oct 23, 2013 and Pavan Balaji committed Oct 29, 2013

Enhanced cxx/buildiface to output support for Distgraphcomm only if
requested through the feature interface.  Distgraphcomm was added
only in MPI 2.2 and for historical reasons is not part of the
MPICH ABI.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

William Gropp authored Oct 21, 2013 and Pavan Balaji committed Oct 29, 2013

Improve buildiface for C++ to detect an unknown feature request with
the -feature= option and abort.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

William Gropp authored Oct 21, 2013 and Pavan Balaji committed Oct 29, 2013

In support of the standard ABI, this change allows us to provide the
Distgraphcomm support defined in MPI 2.2, as well as support the ABI,
which only supports MPI 2.1.  This change is not complete in itself;
another commit is needed to select the ABI level in configure.

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

William Gropp authored Oct 08, 2013 and Pavan Balaji committed Oct 29, 2013

Add a way to output the order of function output in mpicxx.h.in file
and a way to read in a file specifying the order.  This is used to
meet ABI needs.  Also fixed some old MPI_xxx vs MPI::xxx cases, and
the almost duplicated Create_struct call.  All in a single commit
because the test was against the proposed ABI mpicxx.h.in file

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Junchao Zhang authored Oct 29, 2013 and Pavan Balaji committed Oct 29, 2013

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Junchao Zhang authored Oct 28, 2013 and Pavan Balaji committed Oct 29, 2013

Since in mpich_cvars.h, we used a type MPIR_T_cvar_range_value_t,
which is defined in mpitimpl.h.

Fix #1953

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Junchao Zhang authored Oct 28, 2013 and Pavan Balaji committed Oct 29, 2013

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

The new init. priority order of MPICH cvars is
MPICH_* < MPIR_PARAM_* < MPIR_CVAR_*

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Junchao Zhang authored Oct 29, 2013 and Kenneth Raffenetti committed Oct 29, 2013

The old code used (void*)(intptr_t)i to pass the thread id i.
However intptr_t is C99, so I changed it to (void*)(long)i.

Direct usage of (void*)i will issue a compilation warning
saying it is dangerous to convert integer to pointer.

Fixes #1956

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

We are calling the next release 3.1rc1.  The ABI string has already
been updated.

No reviewer.

No reviewer.

Some content was being printed even when not in verbose mode causing
the test to incorrectly fail.

Signed-off-by: Junchao Zhang <jczhang@mcs.anl.gov>

Junchao Zhang authored Oct 25, 2013 and Pavan Balaji committed Oct 26, 2013

Also change src/util/param/params.yml to src/util/cvar/cvars.yml,
maint/genparams to maint/gencvars

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>