ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
41a70a8f01
openmpi/ompi/attribute/attribute_predefined.c

457 строки
17 KiB
C
Исходник Обычный вид История

Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
/*
Update the copyright notices for IU and UTK. This commit was SVN r7999.
2005-11-05 22:57:48 +03:00
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
Add HLRS copyright This commit was SVN r3665.
2004-11-28 23:09:25 +03:00
* University of Stuttgart. All rights reserved.
Add UC copyright This commit was SVN r5009.
2005-03-24 15:43:37 +03:00
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
Rewriting much of the errorcode and errorclass code, since - we have to be able to attach a string to an error class, not just to an error code - according to MPI-2 the attribute MPI_LASTUSEDCODE has to be updated everytime you add a new code or a new class. Thus, you have to have single list for both. Thus, we got rid of the error_class structure. In the error-code structure, we can distinguish whether we are dealing with an error code or an error class by looking at the err->code element of the structure. In case its value is MPI_UNDEFINED, the according entry is a class, else it is an error code. All predefined error codes have the code and the class field set to the same value. The test MPI_Add_error_class1 passes now. Fixes trac:418 This commit was SVN r12764. The following Trac tickets were found above: Ticket 418 --> https://svn.open-mpi.org/trac/ompi/ticket/418
2006-12-05 22:07:02 +03:00
* Copyright (c) 2006 University of Houston. All rights reserved.
First cut at copyrights: IU, UTK, and some OSU. LANL and HLRS still pending. This commit was SVN r3655.
2004-11-22 04:38:40 +03:00
* $COPYRIGHT$
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
*
First cut at copyrights: IU, UTK, and some OSU. LANL and HLRS still pending. This commit was SVN r3655.
2004-11-22 04:38:40 +03:00
* Additional copyrights may follow
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
*
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
* $HEADER$
*/
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
/**
* @file
*
* Setup the predefined attributes in MPI.
*
* A number of pre-defined attributes are created here, most of which
* are exactly what one would expect, but there are a few exceptions
* -- so they're documented here.
*
* Predefined attributes are integer-valued or address-valued (per
* MPI-2; see section 4.12.7, keeping in mind that Example 4.13 is
* totally wrong -- see src/attribute/attribute.h for a lengthy
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
* explanation of this).
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
*
* The only address-valued attribute is MPI_WIN_BASE. We treat it as
* if it were set from C. All other attributes are integer-valued.
* We treat them as if they were set from Fortran MPI-1 (i.e.,
* MPI_ATTR_PUT) or Fortran MPI-2 (i.e., MPI_xxx_ATTR_SET). Most
* attributes are MPI-1 integer-valued, meaning that they are the size
* of MPI_Fint (INTEGER). But MPI_WIN_SIZE and MPI_WIN_DISP_UNIT are
* MPI-2 integer-valued, meaning that they are the size of MPI_Aint
* (INTEGER(KIND=MPI_ADDRESS_KIND)).
*
* MPI_TAG_UB is set to a fixed upper limit.
*
* MPI_HOST is set to MPI_PROC_NULL (per MPI-1, see 7.1.1, p192).
*
Didn't have a good preset value for MPI_IO (see MPI-1, 7.1.1, p192). We may need to revisit this per whacky platforms like Red Storm, Blue Gene, etc. This commit was SVN r6245.
2005-07-01 01:29:22 +04:00
* MPI_IO is set to MPI_ANY_SOURCE. We may need to revist this.
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
*
* MPI_WTIME_IS_GLOBAL is set to 0 (a conservative answer).
*
* MPI_APPNUM is set as the result of a GPR subscription.
*
* MPI_LASTUSEDCODE is set to an initial value and is reset every time
Rewriting much of the errorcode and errorclass code, since - we have to be able to attach a string to an error class, not just to an error code - according to MPI-2 the attribute MPI_LASTUSEDCODE has to be updated everytime you add a new code or a new class. Thus, you have to have single list for both. Thus, we got rid of the error_class structure. In the error-code structure, we can distinguish whether we are dealing with an error code or an error class by looking at the err->code element of the structure. In case its value is MPI_UNDEFINED, the according entry is a class, else it is an error code. All predefined error codes have the code and the class field set to the same value. The test MPI_Add_error_class1 passes now. Fixes trac:418 This commit was SVN r12764. The following Trac tickets were found above: Ticket 418 --> https://svn.open-mpi.org/trac/ompi/ticket/418
2006-12-05 22:07:02 +03:00
* MPI_ADD_ERROR_CLASS or MPI_ADD_ERROR_CODE is invoked.
* Its copy function is set to
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
* MPI_COMM_NULL_COPY_FN, meaning that *only* MPI_COMM_WORLD will have
* this attribute value. As such, we only have to update
* MPI_COMM_WORLD when this value changes (i.e., since this is an
* integer-valued attribute, we have to update this attribute on every
* communicator -- using NULL_COPY_FN ensures that only MPI_COMM_WORLD
* has this attribute value set).
*
* MPI_UNIVERSE_SIZE is set as the result of a GPR subscription.
*
* MPI_WIN_BASE is an address-valued attribute, and is set directly
* from MPI_WIN_CREATE. MPI_WIN_SIZE and MPI_WIN_DISP_UNIT are both
* integer-valued attributes, *BUT* at least the MPI_WIN_SIZE is an
* MPI_Aint, so in terms of consistency, both should be the same --
* hence, we treat them as MPI-2 Fortran integer-valued attributes.
* All three of these atrributes have NULL_COPY_FN copy functions; it
* doesn't make sense to copy them to new windows (because they're
* values specific and unique to each window) -- especially when
* WIN_CREATE will explicitly set them on new windows anyway.
*
* These are not supported yet, but are included here for consistency:
*
* MPI_IMPI_CLIENT_SIZE, MPI_IMPI_CLIENT_COLOR, MPI_IMPI_HOST_SIZE,
* and MPI_IMPI_HOST_COLOR are integer-valued attributes.
*/
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
#include "ompi_config.h"
#include "mpi.h"
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
Next step in the project split, mainly source code re-arranging - move files out of toplevel include/ and etc/, moving it into the sub-projects - rather than including config headers with <project>/include, have them as <project> - require all headers to be included with a project prefix, with the exception of the config headers ({opal,orte,ompi}_config.h mpi.h, and mpif.h) This commit was SVN r8985.
2006-02-12 04:33:29 +03:00
#include "ompi/attribute/attribute.h"
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
Rewriting much of the errorcode and errorclass code, since - we have to be able to attach a string to an error class, not just to an error code - according to MPI-2 the attribute MPI_LASTUSEDCODE has to be updated everytime you add a new code or a new class. Thus, you have to have single list for both. Thus, we got rid of the error_class structure. In the error-code structure, we can distinguish whether we are dealing with an error code or an error class by looking at the err->code element of the structure. In case its value is MPI_UNDEFINED, the according entry is a class, else it is an error code. All predefined error codes have the code and the class field set to the same value. The test MPI_Add_error_class1 passes now. Fixes trac:418 This commit was SVN r12764. The following Trac tickets were found above: Ticket 418 --> https://svn.open-mpi.org/trac/ompi/ticket/418
2006-12-05 22:07:02 +03:00
#include "ompi/errhandler/errcode.h"
remove hard coded constants for value of MPI_TAG_UB and the max CID and add the values to the PML structure. This will allow PMLs that want to do hardware matching at the cost of a smaller range of valid tags and cids. Updated all the places that used the MPI_TAG_UB_VALUE constant to instead look at the pml struct. This commit was SVN r6778.
2005-08-09 18:56:04 +04:00
#include "ompi/communicator/communicator.h"
#include "orte/util/proc_info.h"
#include "ompi/mca/pml/pml.h"
Merge in the new data support subsystem for ORTE. MPI folks should not notice a difference. Longer explanation will be sent to developers mailing list. This commit was SVN r8912.
2006-02-07 06:32:36 +03:00
#include "orte/dss/dss.h"
remove hard coded constants for value of MPI_TAG_UB and the max CID and add the values to the PML structure. This will allow PMLs that want to do hardware matching at the cost of a smaller range of valid tags and cids. Updated all the places that used the MPI_TAG_UB_VALUE constant to instead look at the pml struct. This commit was SVN r6778.
2005-08-09 18:56:04 +04:00
#include "orte/mca/ns/ns.h"
#include "orte/mca/gpr/gpr.h"
#include "orte/mca/errmgr/errmgr.h"
#include "orte/mca/schema/schema.h"
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
/*
* Private functions
*/
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
static int create_comm(int target_keyval, bool want_inherit);
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
static int free_comm(int keyval);
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
/* JMS for when we implement windows */
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
static int create_win(int target_keyval);
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
static int free_win(int keyval);
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
static int set_f(int keyval, MPI_Fint value);
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
int ompi_attr_create_predefined(void)
The merged main trunk and gpr integration branch. Tested on Mac only so far - will check out and test on Linux. If that has a problem, will back all changes out (again), but I think we have this one correct. Will send out a more complete change notice once testing is complete. This commit was SVN r3644.
2004-11-20 22:12:43 +03:00
{
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
int rc, ret;
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
orte_gpr_subscription_t *subs, sub = ORTE_GPR_SUBSCRIPTION_EMPTY;
orte_gpr_trigger_t *trigs, trig = ORTE_GPR_TRIGGER_EMPTY;
char *jobseg;
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
/* Create all the keyvals */
/* DO NOT CHANGE THE ORDER OF CREATING THESE KEYVALS! This order
strictly adheres to the order in mpi.h. If you change the
order here, you must change the order in mpi.h as well! */
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
if (OMPI_SUCCESS != (ret = create_comm(MPI_TAG_UB, true)) ||
OMPI_SUCCESS != (ret = create_comm(MPI_HOST, true)) ||
OMPI_SUCCESS != (ret = create_comm(MPI_IO, true)) ||
OMPI_SUCCESS != (ret = create_comm(MPI_WTIME_IS_GLOBAL, true)) ||
OMPI_SUCCESS != (ret = create_comm(MPI_APPNUM, true)) ||
OMPI_SUCCESS != (ret = create_comm(MPI_LASTUSEDCODE, false)) ||
OMPI_SUCCESS != (ret = create_comm(MPI_UNIVERSE_SIZE, true)) ||
OMPI_SUCCESS != (ret = create_win(MPI_WIN_BASE)) ||
OMPI_SUCCESS != (ret = create_win(MPI_WIN_SIZE)) ||
OMPI_SUCCESS != (ret = create_win(MPI_WIN_DISP_UNIT)) ||
#if 0
/* JMS For when we implement IMPI */
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
OMPI_SUCCESS != (ret = create_comm(IMPI_CLIENT_SIZE, true)) ||
OMPI_SUCCESS != (ret = create_comm(IMPI_CLIENT_COLOR, true)) ||
OMPI_SUCCESS != (ret = create_comm(IMPI_HOST_SIZE, true)) ||
OMPI_SUCCESS != (ret = create_comm(IMPI_HOST_COLOR, true)) ||
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
#endif
0) {
return ret;
}
* set universe size to be size of comm_world in initial setup. Means that all required attributes are set even if the trigger never fires (the trigger can update the value, if needed. This is all in MPI_Init, so that does not violate the standard) This commit was SVN r6427.
2005-07-12 06:34:00 +04:00
/* Set default values for everything except APPNUM. Set UNIVERSE
size to comm_world size. It might grow later, it might not
(tiggers are not fired in all environments. In environments
where triggers aren't set, there won't be COMM_SPAWN, so APPNUM
probably isn't a big deal. */
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
* update ob1 to direct call * don't know what I was thinking, but can't use the MCA_PML_CALL macro on the two data values, as they don't have things that the macro can expand into This commit was SVN r6868.
2005-08-14 07:14:20 +04:00
if (OMPI_SUCCESS != (ret = set_f(MPI_TAG_UB, mca_pml.pml_max_tag)) ||
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
OMPI_SUCCESS != (ret = set_f(MPI_HOST, MPI_PROC_NULL)) ||
Didn't have a good preset value for MPI_IO (see MPI-1, 7.1.1, p192). We may need to revisit this per whacky platforms like Red Storm, Blue Gene, etc. This commit was SVN r6245.
2005-07-01 01:29:22 +04:00
OMPI_SUCCESS != (ret = set_f(MPI_IO, MPI_ANY_SOURCE)) ||
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
OMPI_SUCCESS != (ret = set_f(MPI_WTIME_IS_GLOBAL, 0)) ||
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
OMPI_SUCCESS != (ret = set_f(MPI_LASTUSEDCODE,
Rewriting much of the errorcode and errorclass code, since - we have to be able to attach a string to an error class, not just to an error code - according to MPI-2 the attribute MPI_LASTUSEDCODE has to be updated everytime you add a new code or a new class. Thus, you have to have single list for both. Thus, we got rid of the error_class structure. In the error-code structure, we can distinguish whether we are dealing with an error code or an error class by looking at the err->code element of the structure. In case its value is MPI_UNDEFINED, the according entry is a class, else it is an error code. All predefined error codes have the code and the class field set to the same value. The test MPI_Add_error_class1 passes now. Fixes trac:418 This commit was SVN r12764. The following Trac tickets were found above: Ticket 418 --> https://svn.open-mpi.org/trac/ompi/ticket/418
2006-12-05 22:07:02 +03:00
ompi_mpi_errcode_lastused)) ||
* set universe size to be size of comm_world in initial setup. Means that all required attributes are set even if the trigger never fires (the trigger can update the value, if needed. This is all in MPI_Init, so that does not violate the standard) This commit was SVN r6427.
2005-07-12 06:34:00 +04:00
OMPI_SUCCESS != (ret = set_f(MPI_UNIVERSE_SIZE,
ompi_comm_size(MPI_COMM_WORLD))) ||
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
#if 0
/* JMS For when we implement IMPI */
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
OMPI_SUCCESS != (ret = set(IMPI_CLIENT_SIZE,
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
&attr_impi_client_size)) ||
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
OMPI_SUCCESS != (ret = set(IMPI_CLIENT_COLOR,
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
&attr_impi_client_color)) ||
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
OMPI_SUCCESS != (ret = set(IMPI_HOST_SIZE,
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
&attr_impi_host_size)) ||
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
OMPI_SUCCESS != (ret = set(IMPI_HOST_COLOR,
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
&attr_impi_host_color)) ||
#endif
0) {
return ret;
}
/* Now that those are all created, setup the trigger to get the
UNIVERSE_SIZE and APPNUM values once everyone has passed
stg1. */
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
/* we have to create two subscriptions - one to retrieve the number of slots on
* each node so we can estimate the universe size, and the other to return our
* app_context index to properly set the appnum attribute.
*
* NOTE: when the 2.0 registry becomes available, this should be consolidated to
* a single subscription
*/
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
/* indicate that this is a standard subscription. This indicates
that the subscription will be common to all processes. Thus,
the resulting data can be consolidated into a
process-independent message and broadcast to all processes */
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
subs = &sub;
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
if (ORTE_SUCCESS !=
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
(rc = orte_schema.get_std_subscription_name(&sub.name,
OMPI_ATTRIBUTE_SUBSCRIPTION, ORTE_PROC_MY_NAME->jobid))) {
ORTE_ERROR_LOG(rc);
return rc;
}
sub.action = ORTE_GPR_NOTIFY_DELETE_AFTER_TRIG;
sub.values = (orte_gpr_value_t**)malloc(2 * sizeof(orte_gpr_value_t*));
sub.cnt = 2;
if (ORTE_SUCCESS != (rc = orte_gpr.create_value(&(sub.values[0]), ORTE_GPR_TOKENS_OR | ORTE_GPR_KEYS_OR,
ORTE_NODE_SEGMENT, 1, 0))) {
Okay, here is the massive checkin that restructures the registry trigger system for scalability. Actually, it isn't "quite" as large as it looks - it just touches a bunch of files. Also included is a fix to the attribute problem for singletons. Short explanation: The prior system placed triggers and subscriptions on the registry for each process - approximately eight/process. Each of these had to be checked every time there was a registry operation such as a "put" or "increment-value". For large numbers of processes, this repetitive checking consumed some significant time. The new system allows processes to "attach" to existing triggers and subscriptions, without creating a new one. Thus, there are now only eight triggers and five subscriptions on a job - *regardless of how many processes are being run*. This means that the registry now takes the same amount of time (which is pretty darn short) to process an operation regardless of how many processes are in a job. I'll provide some startup times from scalability tests shortly - need to complete the commit so I can move the system to an appropriate cluster. This commit was SVN r6164.
2005-06-24 20:59:37 +04:00
ORTE_ERROR_LOG(rc);
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
free(sub.name);
return rc;
}
if (ORTE_SUCCESS != (rc = orte_gpr.create_keyval(&(sub.values[0]->keyvals[0]), ORTE_NODE_SLOTS_KEY, ORTE_UNDEF, NULL))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(sub.values[0]);
free(sub.name);
Okay, here is the massive checkin that restructures the registry trigger system for scalability. Actually, it isn't "quite" as large as it looks - it just touches a bunch of files. Also included is a fix to the attribute problem for singletons. Short explanation: The prior system placed triggers and subscriptions on the registry for each process - approximately eight/process. Each of these had to be checked every time there was a registry operation such as a "put" or "increment-value". For large numbers of processes, this repetitive checking consumed some significant time. The new system allows processes to "attach" to existing triggers and subscriptions, without creating a new one. Thus, there are now only eight triggers and five subscriptions on a job - *regardless of how many processes are being run*. This means that the registry now takes the same amount of time (which is pretty darn short) to process an operation regardless of how many processes are in a job. I'll provide some startup times from scalability tests shortly - need to complete the commit so I can move the system to an appropriate cluster. This commit was SVN r6164.
2005-06-24 20:59:37 +04:00
return rc;
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
}
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
if (ORTE_SUCCESS != (rc = orte_schema.get_job_segment_name(&jobseg, ORTE_PROC_MY_NAME->jobid))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(sub.values[0]);
free(sub.name);
return rc;
}
if (ORTE_SUCCESS != (rc = orte_gpr.create_value(&(sub.values[1]), ORTE_GPR_TOKENS_OR | ORTE_GPR_KEYS_OR,
jobseg, 2, 0))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(sub.values[0]);
free(sub.name);
free(jobseg);
return rc;
}
free(jobseg);
if (ORTE_SUCCESS != (rc = orte_gpr.create_keyval(&(sub.values[1]->keyvals[0]), ORTE_PROC_RANK_KEY, ORTE_UNDEF, NULL))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(sub.values[0]);
OBJ_RELEASE(sub.values[1]);
free(sub.name);
return rc;
}
if (ORTE_SUCCESS != (rc = orte_gpr.create_keyval(&(sub.values[1]->keyvals[1]), ORTE_PROC_APP_CONTEXT_KEY, ORTE_UNDEF, NULL))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(sub.values[0]);
OBJ_RELEASE(sub.values[1]);
free(sub.name);
return rc;
}
sub.cbfunc = ompi_attr_create_predefined_callback;
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
/* attach ourselves to the standard stage-1 trigger */
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
trigs = &trig;
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
if (ORTE_SUCCESS !=
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
(rc = orte_schema.get_std_trigger_name(&trig.name,
ORTE_STG1_TRIGGER, ORTE_PROC_MY_NAME->jobid))) {
Okay, here is the massive checkin that restructures the registry trigger system for scalability. Actually, it isn't "quite" as large as it looks - it just touches a bunch of files. Also included is a fix to the attribute problem for singletons. Short explanation: The prior system placed triggers and subscriptions on the registry for each process - approximately eight/process. Each of these had to be checked every time there was a registry operation such as a "put" or "increment-value". For large numbers of processes, this repetitive checking consumed some significant time. The new system allows processes to "attach" to existing triggers and subscriptions, without creating a new one. Thus, there are now only eight triggers and five subscriptions on a job - *regardless of how many processes are being run*. This means that the registry now takes the same amount of time (which is pretty darn short) to process an operation regardless of how many processes are in a job. I'll provide some startup times from scalability tests shortly - need to complete the commit so I can move the system to an appropriate cluster. This commit was SVN r6164.
2005-06-24 20:59:37 +04:00
ORTE_ERROR_LOG(rc);
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
OBJ_RELEASE(sub.values[0]);
OBJ_RELEASE(sub.values[1]);
free(sub.name);
Okay, here is the massive checkin that restructures the registry trigger system for scalability. Actually, it isn't "quite" as large as it looks - it just touches a bunch of files. Also included is a fix to the attribute problem for singletons. Short explanation: The prior system placed triggers and subscriptions on the registry for each process - approximately eight/process. Each of these had to be checked every time there was a registry operation such as a "put" or "increment-value". For large numbers of processes, this repetitive checking consumed some significant time. The new system allows processes to "attach" to existing triggers and subscriptions, without creating a new one. Thus, there are now only eight triggers and five subscriptions on a job - *regardless of how many processes are being run*. This means that the registry now takes the same amount of time (which is pretty darn short) to process an operation regardless of how many processes are in a job. I'll provide some startup times from scalability tests shortly - need to complete the commit so I can move the system to an appropriate cluster. This commit was SVN r6164.
2005-06-24 20:59:37 +04:00
return rc;
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
}
Okay, here is the massive checkin that restructures the registry trigger system for scalability. Actually, it isn't "quite" as large as it looks - it just touches a bunch of files. Also included is a fix to the attribute problem for singletons. Short explanation: The prior system placed triggers and subscriptions on the registry for each process - approximately eight/process. Each of these had to be checked every time there was a registry operation such as a "put" or "increment-value". For large numbers of processes, this repetitive checking consumed some significant time. The new system allows processes to "attach" to existing triggers and subscriptions, without creating a new one. Thus, there are now only eight triggers and five subscriptions on a job - *regardless of how many processes are being run*. This means that the registry now takes the same amount of time (which is pretty darn short) to process an operation regardless of how many processes are in a job. I'll provide some startup times from scalability tests shortly - need to complete the commit so I can move the system to an appropriate cluster. This commit was SVN r6164.
2005-06-24 20:59:37 +04:00
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
if (ORTE_SUCCESS != (rc = orte_gpr.subscribe(1, &subs, 1, &trigs))) {
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
ORTE_ERROR_LOG(rc);
Okay, here is the massive checkin that restructures the registry trigger system for scalability. Actually, it isn't "quite" as large as it looks - it just touches a bunch of files. Also included is a fix to the attribute problem for singletons. Short explanation: The prior system placed triggers and subscriptions on the registry for each process - approximately eight/process. Each of these had to be checked every time there was a registry operation such as a "put" or "increment-value". For large numbers of processes, this repetitive checking consumed some significant time. The new system allows processes to "attach" to existing triggers and subscriptions, without creating a new one. Thus, there are now only eight triggers and five subscriptions on a job - *regardless of how many processes are being run*. This means that the registry now takes the same amount of time (which is pretty darn short) to process an operation regardless of how many processes are in a job. I'll provide some startup times from scalability tests shortly - need to complete the commit so I can move the system to an appropriate cluster. This commit was SVN r6164.
2005-06-24 20:59:37 +04:00
}
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
OBJ_RELEASE(sub.values[0]);
OBJ_RELEASE(sub.values[1]);
free(sub.name);
free(trig.name);
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
return rc;
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
}
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
int ompi_attr_free_predefined(void)
{
int ret;
if (OMPI_SUCCESS != (ret = free_comm(MPI_TAG_UB)) ||
OMPI_SUCCESS != (ret = free_comm(MPI_HOST)) ||
OMPI_SUCCESS != (ret = free_comm(MPI_IO)) ||
OMPI_SUCCESS != (ret = free_comm(MPI_WTIME_IS_GLOBAL)) ||
OMPI_SUCCESS != (ret = free_comm(MPI_APPNUM)) ||
OMPI_SUCCESS != (ret = free_comm(MPI_LASTUSEDCODE)) ||
OMPI_SUCCESS != (ret = free_comm(MPI_UNIVERSE_SIZE)) ||
OMPI_SUCCESS != (ret = free_win(MPI_WIN_BASE)) ||
OMPI_SUCCESS != (ret = free_win(MPI_WIN_SIZE)) ||
OMPI_SUCCESS != (ret = free_win(MPI_WIN_DISP_UNIT)) ||
#if 0
/* JMS For when we implement IMPI */
OMPI_SUCCESS != (ret = free_comm(IMPI_CLIENT_SIZE)) ||
OMPI_SUCCESS != (ret = free_comm(IMPI_CLIENT_COLOR)) ||
OMPI_SUCCESS != (ret = free_comm(IMPI_HOST_SIZE)) ||
OMPI_SUCCESS != (ret = free_comm(IMPI_HOST_COLOR)) ||
#endif
0) {
return ret;
}
return OMPI_SUCCESS;
The merged main trunk and gpr integration branch. Tested on Mac only so far - will check out and test on Linux. If that has a problem, will back all changes out (again), but I think we have this one correct. Will send out a more complete change notice once testing is complete. This commit was SVN r3644.
2004-11-20 22:12:43 +03:00
}
void ompi_attr_create_predefined_callback(
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
orte_gpr_notify_data_t *data,
void *cbdata)
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
{
With the branch to 1.2 made.... Clean up the remainder of the size_t references in the runtime itself. Convert to orte_std_cntr_t wherever it makes sense (only avoid those places where the actual memory size is referenced). Remove the obsolete oob barrier function (we actually obsoleted it a long time ago - just never bothered to clean it up). I have done my best to go through all the components and catch everything, even if I couldn't test compile them since I wasn't on that type of system. Still, I cannot guarantee that problems won't show up when you test this on specific systems. Usually, these will just show as "warning: comparison between signed and unsigned" notes which are easily fixed (just change a size_t to orte_std_cntr_t). In some places, people didn't use size_t, but instead used some other variant (e.g., I found several places with uint32_t). I tried to catch all of them, but... Once we get all the instances caught and fixed, this should once and for all resolve many of the heterogeneity problems. This commit was SVN r11204.
2006-08-15 23:54:10 +04:00
orte_std_cntr_t i, j, k;
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
orte_gpr_keyval_t **keyval;
orte_gpr_value_t **value;
orte_jobid_t job;
Make the compilers happy. This commit was SVN r12729.
2006-12-03 03:19:11 +03:00
orte_std_cntr_t *cptr, rank, app_num = 0;
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
unsigned int universe_size = 0;
Merge in the new data support subsystem for ORTE. MPI folks should not notice a difference. Longer explanation will be sent to developers mailing list. This commit was SVN r8912.
2006-02-07 06:32:36 +03:00
int rc;
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
/* Set some default values */
Bring over the update to terminate orteds that are generated by a dynamic spawn such as comm_spawn. This introduces the concept of a job "family" - i.e., jobs that have a parent/child relationship. Comm_spawn'ed jobs have a parent (the one that spawned them). We track that relationship throughout the lineage - i.e., if a comm_spawned job in turn calls comm_spawn, then it has a parent (the one that spawned it) and a "root" job (the original job that started things). Accordingly, there are new APIs to the name service to support the ability to get a job's parent, root, immediate children, and all its descendants. In addition, the terminate_job, terminate_orted, and signal_job APIs for the PLS have been modified to accept attributes that define the extent of their actions. For example, doing a "terminate_job" with an attribute of ORTE_NS_INCLUDE_DESCENDANTS will terminate the given jobid AND all jobs that descended from it. I have tested this capability on a MacBook under rsh, Odin under SLURM, and LANL's Flash (bproc). It worked successfully on non-MPI jobs (both simple and including a spawn), and MPI jobs (again, both simple and with a spawn). This commit was SVN r12597.
2006-11-14 22:34:59 +03:00
job = ORTE_PROC_MY_NAME->jobid;
- query the daemon in a persistent universe to find out how many app slots we have to fill in MPI_UNIVERSE_SIZE - in a non-persistent universe, just use the size of COMM_WORLD for MPI_UNIVERSE_SIZE. Note the comment in the file, however, I don't think that this is sufficient -- e.g., reserve 8 CPUs in a batch environment and then run "./master" -- UNIVERSE_SIZE should still be 8, even though we don't have a persistent universe. So more integration is necessary here. - removed some debugging printfs and added a forgotten error condition This commit was SVN r2751.
2004-09-18 14:04:31 +04:00
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
/* Query the gpr to find out how many CPUs there will be.
- query the daemon in a persistent universe to find out how many app slots we have to fill in MPI_UNIVERSE_SIZE - in a non-persistent universe, just use the size of COMM_WORLD for MPI_UNIVERSE_SIZE. Note the comment in the file, however, I don't think that this is sufficient -- e.g., reserve 8 CPUs in a batch environment and then run "./master" -- UNIVERSE_SIZE should still be 8, even though we don't have a persistent universe. So more integration is necessary here. - removed some debugging printfs and added a forgotten error condition This commit was SVN r2751.
2004-09-18 14:04:31 +04:00
This will only return a non-empty list in a persistent
universe. If we don't have a persistent universe, then just
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
default to the size of MPI_COMM_WORLD.
- query the daemon in a persistent universe to find out how many app slots we have to fill in MPI_UNIVERSE_SIZE - in a non-persistent universe, just use the size of COMM_WORLD for MPI_UNIVERSE_SIZE. Note the comment in the file, however, I don't think that this is sufficient -- e.g., reserve 8 CPUs in a batch environment and then run "./master" -- UNIVERSE_SIZE should still be 8, even though we don't have a persistent universe. So more integration is necessary here. - removed some debugging printfs and added a forgotten error condition This commit was SVN r2751.
2004-09-18 14:04:31 +04:00
JMS: I think we need more here -- there are cases where you
wouldn't have a persistent universe but still may have a
comm_size(COMM_WORLD) != UNIVERSE_SIZE. For example, say you
reserve 8 CPUs in a batch environment and then run ./master,
where the master is supposed to SPAWN the other processes.
Perhaps need some integration with the LLM here...? [shrug] */
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
/* RHC: Needed to change this code so it wouldn't issue a gpr.get
The merged main trunk and gpr integration branch. Tested on Mac only so far - will check out and test on Linux. If that has a problem, will back all changes out (again), but I think we have this one correct. Will send out a more complete change notice once testing is complete. This commit was SVN r3644.
2004-11-20 22:12:43 +03:00
* during the compound command phase of mpi_init. Since all you need
* is to have the data prior to dtypes etc., and since this function
* is called right before we send the compound command, I've changed
* it to a subscription and a callback function. This allows you to
* get the data AFTER the compound command executes. Nothing else
* happens in-between anyway, so this shouldn't cause a problem.
*/
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
if (0 == data->cnt) { /* no data returned */
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
universe_size = ompi_comm_size(MPI_COMM_WORLD);
- query the daemon in a persistent universe to find out how many app slots we have to fill in MPI_UNIVERSE_SIZE - in a non-persistent universe, just use the size of COMM_WORLD for MPI_UNIVERSE_SIZE. Note the comment in the file, however, I don't think that this is sufficient -- e.g., reserve 8 CPUs in a batch environment and then run "./master" -- UNIVERSE_SIZE should still be 8, even though we don't have a persistent universe. So more integration is necessary here. - removed some debugging printfs and added a forgotten error condition This commit was SVN r2751.
2004-09-18 14:04:31 +04:00
} else {
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
value = (orte_gpr_value_t**)(data->values)->addr;
for (i=0, k=0; k < data->cnt &&
i < (data->values)->size; i++) {
if (NULL != value[i]) {
k++;
Initialize the rank variable before it is used. This commit was SVN r12565.
2006-11-13 05:37:12 +03:00
rank = ORTE_VPID_INVALID;
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
if (0 < value[i]->cnt) { /* make sure some data was returned here */
keyval = value[i]->keyvals;
for (j=0; j < value[i]->cnt; j++) {
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
if (0 == strcmp(ORTE_NODE_SLOTS_KEY, keyval[j]->key)) {
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
/* Process slot count */
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&cptr, keyval[j]->value, ORTE_STD_CNTR))) {
ORTE_ERROR_LOG(rc);
return;
}
universe_size += (unsigned int)(*cptr);
} else if (0 == strcmp(ORTE_PROC_RANK_KEY, keyval[j]->key)) {
/* Process rank */
if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&cptr, keyval[j]->value, ORTE_STD_CNTR))) {
ORTE_ERROR_LOG(rc);
return;
}
rank = *cptr;
} else if (0 == strcmp(ORTE_PROC_APP_CONTEXT_KEY, keyval[j]->key)) {
/* App context number */
if (ORTE_SUCCESS != (rc = orte_dss.get((void**)&cptr, keyval[j]->value, ORTE_STD_CNTR))) {
Merge in the new data support subsystem for ORTE. MPI folks should not notice a difference. Longer explanation will be sent to developers mailing list. This commit was SVN r8912.
2006-02-07 06:32:36 +03:00
ORTE_ERROR_LOG(rc);
return;
}
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
app_num = *cptr;
First phase of the scalable RTE changes: 1. Modify the registry to eliminate redundant data copying for startup messages. 2. Revise the subscription/trigger system to avoid redundant storage of triggers and subscriptions. This dramatically reduces the search time when a registry action occurs - to illustrate the point, there are now only a handful of triggers on the system for each job. Before, there were a handful of triggers for each PROCESS in the job, all of which had to be checked every time something happened on the registry. This is much, much faster now. 3. Update all subscriptions to the new format. There are now "named" subscriptions - this allows you to "name" a subscription that all the processes will be using. The first one to hit the registry actually defines the subscription. From then on, any subsequent "subscribes" to the same name just cause that process to "attach" to the existing subscription. This keeps the number of subscriptions being tracked by the registry to a minimum, while ensuring that each process still gets notified. 4. Do the same for triggers. Also fixed a duplicate subscription problem that was causing people to receive data equal to the number of processes times the data they should have received from a trigger/subscription. Sorry about that... :-( ...but it's all better now! Uncovered a situation where the modex data seems to be getting entered on the registry a second time - the latter time coming after the compound command has been "fired", thereby causing all the subscriptions to fire. Asked Tim and Jeff to look into this. Second phase of the changes will involve modifying the xcast system so that the same message gets sent to all processes. This will further reduce the message traffic, and - once we have a true "broadcast" version of xcast - really speed things up and improve scalability. This commit was SVN r6542.
2005-07-18 22:49:00 +04:00
}
Fix ticket #545 Setup subscriptions to correctly return the MPI_APPNUM attribute. Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated. Update the hello_nodename test to check for MPI_APPNUM. Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there. This commit was SVN r12377.
2006-11-01 00:29:07 +03:00
}
/* see if this value is for this process. We need to
* perform the check since this subscription is associated
* with a trigger - hence, the data from all the procs
* is included in the message
*/
if (rank == ORTE_PROC_MY_NAME->vpid) {
set_f(MPI_APPNUM, (MPI_Fint) app_num);
}
* results from initial merge of the tim branch into the trunk. Compiles and ompi_info works, but that's all that has been tested. This commit was SVN r4827.
2005-03-14 23:57:21 +03:00
}
}
- query the daemon in a persistent universe to find out how many app slots we have to fill in MPI_UNIVERSE_SIZE - in a non-persistent universe, just use the size of COMM_WORLD for MPI_UNIVERSE_SIZE. Note the comment in the file, however, I don't think that this is sufficient -- e.g., reserve 8 CPUs in a batch environment and then run "./master" -- UNIVERSE_SIZE should still be 8, even though we don't have a persistent universe. So more integration is necessary here. - removed some debugging printfs and added a forgotten error condition This commit was SVN r2751.
2004-09-18 14:04:31 +04:00
}
}
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
/* Same as above -- ignore errors here because there's nothing we
can do if there's any error anyway */
- query the daemon in a persistent universe to find out how many app slots we have to fill in MPI_UNIVERSE_SIZE - in a non-persistent universe, just use the size of COMM_WORLD for MPI_UNIVERSE_SIZE. Note the comment in the file, however, I don't think that this is sufficient -- e.g., reserve 8 CPUs in a batch environment and then run "./master" -- UNIVERSE_SIZE should still be 8, even though we don't have a persistent universe. So more integration is necessary here. - removed some debugging printfs and added a forgotten error condition This commit was SVN r2751.
2004-09-18 14:04:31 +04:00
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
set_f(MPI_UNIVERSE_SIZE, universe_size);
The merged main trunk and gpr integration branch. Tested on Mac only so far - will check out and test on Linux. If that has a problem, will back all changes out (again), but I think we have this one correct. Will send out a more complete change notice once testing is complete. This commit was SVN r3644.
2004-11-20 22:12:43 +03:00
return;
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
}
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
static int create_comm(int target_keyval, bool want_inherit)
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
{
int err;
int keyval;
ompi_attribute_fn_ptr_union_t copy;
ompi_attribute_fn_ptr_union_t del;
keyval = -1;
Add one more argument to the copy functions for the MPI objects. As this argument is the last one on the list and as on C the caller "make it right" this addition will not affect the way we handle the user defined copy functions. Only the C version of the function has this additional parameter. As it represent the pointer to the newly created MPI object It hold the key to allow us to modify the new object (communicator, window or type) depending on some key stored on the initial communicator. This commit was SVN r9371.
2006-03-23 07:47:14 +03:00
copy.attr_communicator_copy_fn = (MPI_Comm_internal_copy_attr_function*)
(want_inherit ? MPI_COMM_DUP_FN : MPI_COMM_NULL_COPY_FN);
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
del.attr_communicator_delete_fn = MPI_COMM_NULL_DELETE_FN;
err = ompi_attr_create_keyval(COMM_ATTR, copy, del,
&keyval, NULL, OMPI_KEYVAL_PREDEFINED);
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
if (MPI_SUCCESS != err) {
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
return err;
}
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
if (target_keyval != keyval) {
return OMPI_ERR_BAD_PARAM;
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
}
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
return OMPI_SUCCESS;
}
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
static int free_comm(int keyval)
{
int key = keyval;
return ompi_attr_free_keyval (COMM_ATTR, &key, true);
}
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
static int create_win(int target_keyval)
{
int err;
int keyval;
ompi_attribute_fn_ptr_union_t copy;
ompi_attribute_fn_ptr_union_t del;
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
keyval = -1;
Add one more argument to the copy functions for the MPI objects. As this argument is the last one on the list and as on C the caller "make it right" this addition will not affect the way we handle the user defined copy functions. Only the C version of the function has this additional parameter. As it represent the pointer to the newly created MPI object It hold the key to allow us to modify the new object (communicator, window or type) depending on some key stored on the initial communicator. This commit was SVN r9371.
2006-03-23 07:47:14 +03:00
copy.attr_win_copy_fn = (MPI_Win_internal_copy_attr_function*)MPI_WIN_NULL_COPY_FN;
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
del.attr_win_delete_fn = MPI_WIN_NULL_DELETE_FN;
err = ompi_attr_create_keyval(WIN_ATTR, copy, del,
&keyval, NULL, OMPI_KEYVAL_PREDEFINED);
if (MPI_SUCCESS != err) {
return err;
}
if (target_keyval != keyval) {
return OMPI_ERR_BAD_PARAM;
}
Some fixes for the attribute code: - move the attribute init section in ompi_mpi_init() down below where communicators are setup (we need MPI_COMM_WORLD to be setup before attributes and keyvals are setup) - removed confusing extra wrapper class around ompi_hash_table_t; looks like it was a victim of slow eroding of members so I put took it out back and put it out of its misery - added preliminary definitions for all the pre-defined keyvals. Still need more work here to assign their initial values, but I think Edgar was running into an atrribute problem and it may have been that the pre-defined attrs didn't yet exist. - removed some LAM-specific predefined keyvals from mpi.h This commit was SVN r2695.
2004-09-16 04:00:09 +04:00
return OMPI_SUCCESS;
}
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
- Upon Finalize, also release the memory of predefined attributes. This commit was SVN r9169.
2006-02-27 18:15:48 +03:00
static int free_win(int keyval)
{
int key = keyval;
return ompi_attr_free_keyval (WIN_ATTR, &key, true);
}
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
static int set_f(int keyval, MPI_Fint value)
{
return ompi_attr_set_fortran_mpi1(COMM_ATTR, MPI_COMM_WORLD,
Hey, sports fans!! Guess what?? Here's the huge registry check-in you've all been waiting for with baited breath. The revised version sends a single message to all processes at the various stage gates, thus making the startup much more scalable. I could provide you with all the tawdry details, but won't for now - you are welcome to ask, though, and I'll merrily bore your ears to tears. In addition, the commit contains the following: 1. set the ignore properties on ompi/debuggers and orte/mca/pls/poe 2. Added simplified subscribe and put functions to the registry's API. I have also converted all of the ompi functions that registered subscriptions to the new API, and caught their associated put's as well. In a follow-on commit, I'll be adding support for George's hetero arch registry subscription (wanted to get this one in first). This commit was SVN r7118.
2005-09-01 05:07:30 +04:00
&MPI_COMM_WORLD->c_keyhash,
keyval, value,
Submitted by: Jeff "I love MPI attributes" Squyres Reviewed by: Brian "MPI attributes ROCK" Barrett Bunches of changes to the attribute engine: - After many hours of discussion about MPI attributes, we came to the conclusion that MPI-2 Example 4.13 (the C->Fortran example) is just wrong. If you accept that, the rest of the text makes much more sense. - There are 9 inter-language cases: all combinations of (read, write) with C, Fortran MPI-1, and Fortran MPI-2 for each value. Each of the 9 cases have specific code for what is supposed to happen (and is labeled in the code with comments). There is a *lengthy* comment at the top of src/attribute/attribute.c that describes all of this. - All predefined attributes are now treated as if they were put from MPI-1 Fortran calls, with the exception of the window predefined attributes (which are irrelevant on the beta, because there is no one-sided support; preliminary fixes included in this patch, but will be fully addressed on the trunk) - MPI API calls (particularly the Fortran wrappers) are now fundamentally simpler -- they do *not* call the back-end MPI C API calls; instead, they call directly back into the attribute engine. - The MPI_LASTUSEDCODE attribute only exists on MPI_COMM_WORLD and is updated appropriately when user error classes are added. --> Note: Edgar made a suggestion that for communicator attributes, we ignore the communicator argument when retrieving attributes and simply return the value. This will likely only happen on the trunk, and will alleviate (from the user's perspective) the restriction that LASTUSEDCODE is only on MPI_COMM_WORLD. - The predefined attributes are now "better". We create keyvals separately than assigning values, and correctly distinguish between comm, type, and win attributes. Initial values are now set as if they were called from MPI-1 fortran. - Added a comment to the top of src/attribute/attribute_predefined.c explaining what each of the predefined attributes were and what OMPI sets them to be. This commit was SVN r6193.
2005-06-27 23:17:11 +04:00
true, true);
}
Копировать постоянную ссылку