Re: [OMPI devel] RFC: initial MPI_T support
As things stand, this branch cannot build. After spending some time trying to pinpoint the problem, I discovered that: 1. you are missing a file "ompi/runtime/ompi_info_support.c" 2. you have a duplicate definition in ompi/tools/ompi_info.c int ompi_info_register_framework_params(opal_pointer_array_t *component_map); That definition already exists in "ompi/runtime/ompi_info_support.h", and that function needs to be in the missing file as you refer to it from other places. Please don't commit this until we get it fixed and can verify we can at least build the result. Thanks Ralph On Apr 5, 2013, at 2:55 PM, Nathan Hjelm wrote: > Whoops, the command I gave you was wrong. Try: > > > git clone -b mpit-commit git://github.com/hjelmn/ompi-mca-var.git > > > -Nathan > > On Fri, Apr 05, 2013 at 01:14:41PM -0700, Ralph Castain wrote: >> Should clarify - I can build the trunk just fine. >> >> On Apr 5, 2013, at 1:14 PM, Ralph Castain wrote: >> >>> This is what I cloned: >>> >>> git clone git://github.com/hjelmn/ompi-mca-var.git mpit-commit >>> >>> >>> On Apr 5, 2013, at 1:08 PM, Nathan Hjelm wrote: >>> Should be. I rebased fro the trunk this morning. Are you sure you cloned the mpit-commit branch instead of the old mpit branch? I will remove the old one to reduce confusion. -Nathan On Fri, Apr 05, 2013 at 01:05:23PM -0700, Ralph Castain wrote: > Is this branch up-to-date? I hit this > > checking for pci/pci.h... no > configure: WARNING: Specified --enable-pci switch, but could not > configure: WARNING: find appropriate support > configure: error: Cannot continue > > > I assure you I didn't specify that switch - it looks old to me. > > > On Apr 5, 2013, at 12:53 PM, Nathan Hjelm wrote: > >> git clone git://github.com/hjelmn/ompi-mca-var.git mpit-commit >> >> -Nathan >> >> On Fri, Apr 05, 2013 at 12:50:55PM -0700, Ralph Castain wrote: >>> Nathan - I can't clone this repo - I get an error from git: >>> >>> [rhc@bend001 openmpi]$ git clone >>> https://github.com/hjelmn/ompi-mca-var/tree/mpit-commit >>> Initialized empty Git repository in >>> /home/common/openmpi/mpit-commit/.git/ >>> fatal: >>> https://github.com/hjelmn/ompi-mca-var/tree/mpit-commit/info/refs not >>> found: did you run git update-server-info on the server? >>> >>> Can you please let me know how I can get a copy of this code? >>> >>> >>> On Apr 5, 2013, at 12:44 PM, Ralph Castain wrote: >>> Let's bump it to Thurs - I trust what you say, but haven't had a chance to glance at the changes. Likewise, it would be nice to let Jeff return from vacation and look at it too. On Apr 5, 2013, at 12:11 PM, Nathan Hjelm wrote: > Yeah, I figured since the changes are self contained (doesn't tough > many outside pieces except info support and mpi.h.in) it would be ok > to do a short timeout. If not I can bump it to Thursday April 11. > > -Nathan > > On Fri, Apr 05, 2013 at 12:02:22PM -0700, Ralph Castain wrote: >> Udid you really mean a timeout of Mon?? Could you delay that >> a bit so we can actually have time to look at it? >> >> >> On Apr 5, 2013, at 11:56 AM, Nathan Hjelm wrote: >> >>> Also, please look at the thread level support. We had some >>> discussion at the forum about what level should be returned >>> depending on which function is called first (MPI_T_init_thread or >>> MPI_Init_thread). I don't think it was clarified what should be >>> done but since it will be in future errata what is in the current >>> implementation should be fine for now. >>> >>> -Nathan >>> >>> On Fri, Apr 05, 2013 at 12:52:12PM -0600, Nathan Hjelm wrote: What: Add initial support for the MPI 3.0 tools interface (MPI_T). Inital support includes full support for the MPI_T_cvar and MPI_T_category interfaces. No pvars are available at this time. Support for pvars will be added at a later time. Why: To be MPI 3.0 compliant the MPI_T interface must be implemented. This RFC implements the complete interface. When: Monday April 8, 2013. We can make adjustments to the implementation after it is committed to the trunk. The only changes that really need to be reviewed are: ompi/include/mpi.h.in, and structure of the incomming code. The changes can be found @ https://github.com/hjelmn/ompi-mca-var/tree/mpit-commit Jeff, I added the MPI_T error codes to mpif-values.pl even though >
Re: [OMPI devel] RFC: initial MPI_T support
After Nathan's latest commit, I can confirm this branch now builds and runs. I've looked at the code, and the added code is indeed completely separated from the existing code base and shouldn't create any operational issues. I haven't confirmed it actually works, but as Nathan saids, that can be resolved as we go. Still, let's give others a chance to verify as well. On Apr 6, 2013, at 8:37 AM, Ralph Castain wrote: > As things stand, this branch cannot build. After spending some time trying to > pinpoint the problem, I discovered that: > > 1. you are missing a file "ompi/runtime/ompi_info_support.c" > > 2. you have a duplicate definition in ompi/tools/ompi_info.c > int ompi_info_register_framework_params(opal_pointer_array_t *component_map); > > That definition already exists in "ompi/runtime/ompi_info_support.h", and > that function needs to be in the missing file as you refer to it from other > places. > > Please don't commit this until we get it fixed and can verify we can at least > build the result. > > Thanks > Ralph > > On Apr 5, 2013, at 2:55 PM, Nathan Hjelm wrote: > >> Whoops, the command I gave you was wrong. Try: >> >> >> git clone -b mpit-commit git://github.com/hjelmn/ompi-mca-var.git >> >> >> -Nathan >> >> On Fri, Apr 05, 2013 at 01:14:41PM -0700, Ralph Castain wrote: >>> Should clarify - I can build the trunk just fine. >>> >>> On Apr 5, 2013, at 1:14 PM, Ralph Castain wrote: >>> This is what I cloned: git clone git://github.com/hjelmn/ompi-mca-var.git mpit-commit On Apr 5, 2013, at 1:08 PM, Nathan Hjelm wrote: > Should be. I rebased fro the trunk this morning. Are you sure you cloned > the mpit-commit branch instead of the old mpit branch? I will remove the > old one to reduce confusion. > > -Nathan > > On Fri, Apr 05, 2013 at 01:05:23PM -0700, Ralph Castain wrote: >> Is this branch up-to-date? I hit this >> >> checking for pci/pci.h... no >> configure: WARNING: Specified --enable-pci switch, but could not >> configure: WARNING: find appropriate support >> configure: error: Cannot continue >> >> >> I assure you I didn't specify that switch - it looks old to me. >> >> >> On Apr 5, 2013, at 12:53 PM, Nathan Hjelm wrote: >> >>> git clone git://github.com/hjelmn/ompi-mca-var.git mpit-commit >>> >>> -Nathan >>> >>> On Fri, Apr 05, 2013 at 12:50:55PM -0700, Ralph Castain wrote: Nathan - I can't clone this repo - I get an error from git: [rhc@bend001 openmpi]$ git clone https://github.com/hjelmn/ompi-mca-var/tree/mpit-commit Initialized empty Git repository in /home/common/openmpi/mpit-commit/.git/ fatal: https://github.com/hjelmn/ompi-mca-var/tree/mpit-commit/info/refs not found: did you run git update-server-info on the server? Can you please let me know how I can get a copy of this code? On Apr 5, 2013, at 12:44 PM, Ralph Castain wrote: > Let's bump it to Thurs - I trust what you say, but haven't had a > chance to glance at the changes. Likewise, it would be nice to let > Jeff return from vacation and look at it too. > > > On Apr 5, 2013, at 12:11 PM, Nathan Hjelm wrote: > >> Yeah, I figured since the changes are self contained (doesn't tough >> many outside pieces except info support and mpi.h.in) it would be ok >> to do a short timeout. If not I can bump it to Thursday April 11. >> >> -Nathan >> >> On Fri, Apr 05, 2013 at 12:02:22PM -0700, Ralph Castain wrote: >>> Udid you really mean a timeout of Mon?? Could you delay >>> that a bit so we can actually have time to look at it? >>> >>> >>> On Apr 5, 2013, at 11:56 AM, Nathan Hjelm wrote: >>> Also, please look at the thread level support. We had some discussion at the forum about what level should be returned depending on which function is called first (MPI_T_init_thread or MPI_Init_thread). I don't think it was clarified what should be done but since it will be in future errata what is in the current implementation should be fine for now. -Nathan On Fri, Apr 05, 2013 at 12:52:12PM -0600, Nathan Hjelm wrote: > What: Add initial support for the MPI 3.0 tools interface > (MPI_T). Inital support includes full support for the MPI_T_cvar > and MPI_T_category interfaces. No pvars are available at this > time. Support for pvars will be added at a later time. > > Why: To be MPI 3.0 compliant the MPI_T interface must be > implement
