{"payload":{"feedbackUrl":"https://github.com/orgs/community/discussions/53140","repo":{"id":37176333,"defaultBranch":"main","name":"ompi","ownerLogin":"bosilca","currentUserCanPush":false,"isFork":true,"isEmpty":false,"createdAt":"2015-06-10T04:57:20.000Z","ownerAvatar":"https://avatars.githubusercontent.com/u/642701?v=4","public":true,"private":false,"isOrgOwned":false},"refInfo":{"name":"","listCacheKey":"v0:1715969110.0","currentOid":""},"activityList":{"items":[{"before":"ee35f8458188c596bcb0c3ab0db171221bc219b5","after":"7a1a4e69e36fc175cfde540e911ab582e1afd9fb","ref":"refs/heads/fix/no_linear_allreduce_in_base_allreduce_scatter_allgather","pushedAt":"2024-05-23T16:24:22.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Dont redirect to another allreduce algorithm.\n\nThe decision to switch to a different algorithm based on count\ndoes not belong to the base algorithms, it should be done in the\nupper level decision functions.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Dont redirect to another allreduce algorithm."}},{"before":"5e22ccf25ac2e076e5f4f8a63b30fa6e1fe0dca0","after":"3e02132d1aec8791da556764cb4535a9ef55e92e","ref":"refs/heads/topic/add_SVE_op_support","pushedAt":"2024-05-23T02:45:12.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Add an AARCH64 OP component.\n\nIt provides support for NEON and SVE ISA.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Add an AARCH64 OP component."}},{"before":"c0c554130a803dd3697f2e202c4a166dbd7cc938","after":"5e22ccf25ac2e076e5f4f8a63b30fa6e1fe0dca0","ref":"refs/heads/topic/add_SVE_op_support","pushedAt":"2024-05-20T15:03:16.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Add an AARCH64 OP component.\n\nIt provides support for NEON and SVE ISA.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Add an AARCH64 OP component."}},{"before":"31b7e5cc0bc08cf4f13e6136d67c0a5a62607cc5","after":"c0c554130a803dd3697f2e202c4a166dbd7cc938","ref":"refs/heads/topic/add_SVE_op_support","pushedAt":"2024-05-17T18:08:45.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Add an AARCH64 OP component.\n\nIt provides support for NEON and SVE ISA.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Add an AARCH64 OP component."}},{"before":null,"after":"e6e5c9455309e19974dc2c15983ddb434b3a786b","ref":"refs/heads/topic/patterns_cleanup","pushedAt":"2024-05-17T18:05:10.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Remove the old patterns, they are not in use.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Remove the old patterns, they are not in use."}},{"before":null,"after":"ee35f8458188c596bcb0c3ab0db171221bc219b5","ref":"refs/heads/fix/no_linear_allreduce_in_base_allreduce_scatter_allgather","pushedAt":"2024-05-16T21:26:52.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Dont redirect to another allreduce algorithm.\n\nThe decision to switch to a different algorithm based on count\ndoes not belong to the base algorithms, it should be done in the\nupper level decision functions.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Dont redirect to another allreduce algorithm."}},{"before":null,"after":"cbaee8ca984b8edacbb5df26f133b1bb7879e328","ref":"refs/heads/fix/unused_tuned_alltoall_mca_params","pushedAt":"2024-05-16T21:16:19.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Mark unused MCA params as deprecated.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Mark unused MCA params as deprecated."}},{"before":"688da3089c7f30cf622902b2d6b2ad94904f6030","after":"31b7e5cc0bc08cf4f13e6136d67c0a5a62607cc5","ref":"refs/heads/topic/add_SVE_op_support","pushedAt":"2024-05-08T20:18:18.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Add an AARCH64 OP component.\n\nIt provides support for NEON and SVE ISA.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Add an AARCH64 OP component."}},{"before":"19bcc48a6325053cbabff6ee3d8541637400f6f9","after":"688da3089c7f30cf622902b2d6b2ad94904f6030","ref":"refs/heads/topic/add_SVE_op_support","pushedAt":"2024-05-04T14:08:18.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"flush\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"flush"}},{"before":"dfe43b8e557833552e1825e10c6bdd0d4fc58fa6","after":"19bcc48a6325053cbabff6ee3d8541637400f6f9","ref":"refs/heads/topic/add_SVE_op_support","pushedAt":"2024-05-02T19:34:59.000Z","pushType":"push","commitsCount":1,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"few updated to aarch64 support.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"few updated to aarch64 support."}},{"before":null,"after":"dfe43b8e557833552e1825e10c6bdd0d4fc58fa6","ref":"refs/heads/topic/add_SVE_op_support","pushedAt":"2024-05-01T15:08:04.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Add an AARCH64 OP component.\n\nIt provides support for NEON and SVE ISA.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Add an AARCH64 OP component."}},{"before":"f2dfbba1f26715bee67925d04d2392b9c335b1a7","after":null,"ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-05-01T13:05:30.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"}},{"before":"91ddbb0d1e9e00f9a51081fbb7217b736ea6787b","after":"f2dfbba1f26715bee67925d04d2392b9c335b1a7","ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-04-30T15:24:55.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix collective initialization and finalization\n\nInstead of allowing each collective module to present a list of\nfunctions it provide, let them register the functions they provide and\nsave the context of the previous collective if they choose to.\n\nThere are two major benefits to this approach:\n- tighter memory management in the collective module themselves. Each\n  collective enable and disable is called exactly once per communicator,\n  to chain or unchain themselves from the collective function pointers\n  struct. The disable is called in the reverse order of the enable,\n  allowing for proper chaining of collectives.\n- they only install the functions they want. So instead of checking in\n  the coll_select all the functions for all modules, each module can now\n  selectively iterate over only the functions it provides.\n\nWhat is still broken is the ability of a particular collective module to\nunchain itself in the middle of the execution. Instead, a properly\nimplemented module will have an enable/disable flag, and it should act\nas a passthrough if it chooses to desactivate.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix collective initialization and finalization"}},{"before":"2f63e230ecf95113f7f5339ce902aecd40e3ecbf","after":null,"ref":"refs/heads/topic/remove_sm_collective","pushedAt":"2024-04-30T15:11:40.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"}},{"before":"b3c22d9f556553192e59ac098303bcc5bc0b08c3","after":"91ddbb0d1e9e00f9a51081fbb7217b736ea6787b","ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-04-29T22:19:43.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix collective initialization and finalization\n\nInstead of allowing each collective module to present a list of\nfunctions it provide, let them register the functions they provide and\nsave the context of the previous collective if they choose to.\n\nThere are two major benefits to this approach:\n- tighter memory management in the collective module themselves. Each\n  collective enable and disable is called exactly once per communicator,\n  to chain or unchain themselves from the collective function pointers\n  struct. The disable is called in the reverse order of the enable,\n  allowing for proper chaining of collectives.\n- they only install the functions they want. So instead of checking in\n  the coll_select all the functions for all modules, each module can now\n  selectively iterate over only the functions it provides.\n\nWhat is still broken is the ability of a particular collective module to\nunchain itself in the middle of the execution. Instead, a properly\nimplemented module will have an enable/disable flag, and it should act\nas a passthrough if it chooses to desactivate.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix collective initialization and finalization"}},{"before":"31dc3b4cd34f39abc81988ffc2081ccd20d1f536","after":"b3c22d9f556553192e59ac098303bcc5bc0b08c3","ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-04-29T22:18:24.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix collective initialization and finalization\n\nInstead of allowing each collective module to present a list of\nfunctions it provide, let them register the functions they provide and\nsave the context of the previous collective if they choose to.\n\nThere are two major benefits to this approach:\n- tighter memory management in the collective module themselves. Each\n  collective enable and disable is called exactly once per communicator,\n  to chain or unchain themselves from the collective function pointers\n  struct. The disable is called in the reverse order of the enable,\n  allowing for proper chaining of collectives.\n- they only install the functions they want. So instead of checking in\n  the coll_select all the functions for all modules, each module can now\n  selectively iterate over only the functions it provides.\n\nWhat is still broken is the ability of a particular collective module to\nunchain itself in the middle of the execution. Instead, a properly\nimplemented module will have an enable/disable flag, and it should act\nas a passthrough if it chooses to desactivate.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix collective initialization and finalization"}},{"before":"3333e1f798db0a68c1e30d73fe85a5ec2539f1fb","after":"31dc3b4cd34f39abc81988ffc2081ccd20d1f536","ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-04-29T20:51:51.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix collective initialization and finalization\n\nInstead of allowing each collective module to present a list of\nfunctions it provide, let them register the functions they provide and\nsave the context of the previous collective if they choose to.\n\nThere are two major benefits to this approach:\n- tighter memory management in the collective module themselves. Each\n  collective enable and disable is called exactly once per communicator,\n  to chain or unchain themselves from the collective function pointers\n  struct. The disable is called in the reverse order of the enable,\n  allowing for proper chaining of collectives.\n- they only install the functions they want. So instead of checking in\n  the coll_select all the functions for all modules, each module can now\n  selectively iterate over only the functions it provides.\n\nWhat is still broken is the ability of a particular collective module to\nunchain itself in the middle of the execution. Instead, a properly\nimplemented module will have an enable/disable flag, and it should act\nas a passthrough if it chooses to desactivate.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix collective initialization and finalization"}},{"before":"6d4070ae550665921f7cba3b04efc10023e4428c","after":"9fcfec2b9de0f23badc529633651db2cfbf34faa","ref":"refs/heads/main","pushedAt":"2024-04-29T16:01:09.000Z","pushType":"push","commitsCount":13,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Merge pull request #12500 from edgargabriel/topic/remove-pvfs2-support\n\nfs and fbtl/pvfs2: remove components","shortMessageHtmlLink":"Merge pull request <a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2267024620\" data-permission-text=\"Title is private\" data-url=\"https://github.com/open-mpi/ompi/issues/12500\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/open-mpi/ompi/pull/12500/hovercard\" href=\"https://github.com/open-mpi/ompi/pull/12500\">open-mpi#12500</a> from edgargabriel/topic/remove-pvfs…"}},{"before":null,"after":"2f63e230ecf95113f7f5339ce902aecd40e3ecbf","ref":"refs/heads/topic/remove_sm_collective","pushedAt":"2024-04-26T15:08:34.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Remove the SM collective component\n\nIt has been pointed out that the SM component was turned off for a very\nlong time, so it was not used and not tested. Moreover, when enabled the\nperformance are below tuned, so there is no reason to keep it around.\n\nThe good news is that there are two possible replacement that are up to\nreview and merge (ACOL and XHC), so at the developers meeting in Austin\n2024 we decided to remove the SM collective component.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Remove the SM collective component"}},{"before":"8514e717215a42ddb6db065c3baae3a2779a072a","after":"6d4070ae550665921f7cba3b04efc10023e4428c","ref":"refs/heads/main","pushedAt":"2024-04-25T14:35:54.000Z","pushType":"push","commitsCount":444,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Merge pull request #12451 from devreal/win_shared_query_ptrdiff\n\nFix incorrect disp_unit type passed to osc_win_shared_query","shortMessageHtmlLink":"Merge pull request <a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"2221401133\" data-permission-text=\"Title is private\" data-url=\"https://github.com/open-mpi/ompi/issues/12451\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/open-mpi/ompi/pull/12451/hovercard\" href=\"https://github.com/open-mpi/ompi/pull/12451\">open-mpi#12451</a> from devreal/win_shared_query_ptrdiff"}},{"before":"6f8ae384c6a76156a24f8c471f90c3ff90794eba","after":null,"ref":"refs/heads/fix/scatter_inter","pushedAt":"2024-04-22T23:49:19.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"}},{"before":null,"after":"6f8ae384c6a76156a24f8c471f90c3ff90794eba","ref":"refs/heads/fix/scatter_inter","pushedAt":"2024-04-22T18:02:08.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix the intercom scatter.\n\nThe original implementation overwrote the requests array, leading to\nsegfaults. While fixing this I also fix few other typos and removed all\nbasic functions that were just an indirection to the base functions.\n\nFixes #12482.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix the intercom scatter."}},{"before":"3153988420bfe14964ac5b7110de49437b8d0c97","after":"3333e1f798db0a68c1e30d73fe85a5ec2539f1fb","ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-04-20T18:16:42.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix collective initialization and finalization\n\nInstead of allowing each collective module to present a list of\nfunctions it provide, let them register the functions they provide and\nsave the context of the previous collective if they choose to.\n\nThere are two major benefits to this approach:\n- tighter memory management in the collective module themselves. Each\n  collective enable and disable is called exactly once per communicator,\n  to chain or unchain themselves from the collective function pointers\n  struct. The disable is called in the reverse order of the enable,\n  allowing for proper chaining of collectives.\n- they only install the functions they want. So instead of checking in\n  the coll_select all the functions for all modules, each module can now\n  selectively iterate over only the functions it provides.\n\nWhat is still broken is the ability of a particular collective module to\nunchain itself in the middle of the execution. Instead, a properly\nimplemented module will have an enable/disable flag, and it should act\nas a passthrough if it chooses to desactivate.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix collective initialization and finalization"}},{"before":"cc2ebadfc4d87293a96d54962556e59d96a53ee4","after":"3153988420bfe14964ac5b7110de49437b8d0c97","ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-04-20T17:24:24.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix collective initialization and finalization\n\nInstead of allowing each collective module to present a list of\nfunctions it provide, let them register the functions they provide and\nsave the context of the previous collective if they choose to.\n\nThere are two major benefits to this approach:\n- tighter memory management in the collective module themselves. Each\n  collective enable and disable is called exactly once per communicator,\n  to chain or unchain themselves from the collective function pointers\n  struct. The disable is called in the reverse order of the enable,\n  allowing for proper chaining of collectives.\n- they only install the functions they want. So instead of checking in\n  the coll_select all the functions for all modules, each module can now\n  selectively iterate over only the functions it provides.\n\nWhat is still broken is the ability of a particular collective module to\nunchain itself in the middle of the execution. Instead, a properly\nimplemented module will have an enable/disable flag, and it should act\nas a passthrough if it chooses to desactivate.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix collective initialization and finalization"}},{"before":"ee1610752d90e844499fbe02deb7452fc681b633","after":"8514e717215a42ddb6db065c3baae3a2779a072a","ref":"refs/heads/main","pushedAt":"2024-04-20T17:24:24.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Merge pull request #11848 from jsquyres/pr/rtd-update\n\ndocs: update readthedocs.org config","shortMessageHtmlLink":"Merge pull request <a class=\"issue-link js-issue-link\" data-error-text=\"Failed to load title\" data-id=\"1839462793\" data-permission-text=\"Title is private\" data-url=\"https://github.com/open-mpi/ompi/issues/11848\" data-hovercard-type=\"pull_request\" data-hovercard-url=\"/open-mpi/ompi/pull/11848/hovercard\" href=\"https://github.com/open-mpi/ompi/pull/11848\">open-mpi#11848</a> from jsquyres/pr/rtd-update"}},{"before":"41256e376de3db4fb014fbe641bbbec9e5addfb0","after":"cc2ebadfc4d87293a96d54962556e59d96a53ee4","ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-04-20T03:32:56.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix collective initialization and finalization\n\nInstead of allowing each collective module to present a list of\nfunctions it provide, let them register the functions they provide and\nsave the context of the previous collective if they choose to.\n\nThere are two major benefits to this approach:\n- tighter memory management in the collective module themselves. Each\n  collective enable and disable is called exactly once per communicator,\n  to chain or unchain themselves from the collective function pointers\n  struct. The disable is called in the reverse order of the enable,\n  allowing for proper chaining of collectives.\n- they only install the functions they want. So instead of checking in\n  the coll_select all the functions for all modules, each module can now\n  selectively iterate over only the functions it provides.\n\nWhat is still broken is the ability of a particular collective module to\nunchain itself in the middle of the execution. Instead, a properly\nimplemented module will have an enable/disable flag, and it should act\nas a passthrough if it chooses to desactivate.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix collective initialization and finalization"}},{"before":"ebcc4e018124b7d3bc012ad19bf55428c79e687c","after":"41256e376de3db4fb014fbe641bbbec9e5addfb0","ref":"refs/heads/topic/fix_collective_init_fini","pushedAt":"2024-04-19T23:27:19.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Fix collective initialization and finalization\n\nInstead of allowing each collective module to present a list of\nfunctions it provide, let them register the functions they provide and\nsave the context of the previous collective if they choose to.\n\nThere are two major benefits to this approach:\n- tighter memory management in the collective module themselves. Each\n  collective enable and disable is called exactly once per communicator,\n  to chain or unchain themselves from the collective function pointers\n  struct. The disable is called in the reverse order of the enable,\n  allowing for proper chaining of collectives.\n- they only install the functions they want. So instead of checking in\n  the coll_select all the functions for all modules, each module can now\n  selectively iterate over only the functions it provides.\n\nWhat is still broken is the ability of a particular collective module to\nunchain itself in the middle of the execution. Instead, a properly\nimplemented module will have an enable/disable flag, and it should act\nas a passthrough if it chooses to desactivate.\n\nSigned-off-by: George Bosilca <bosilca@icl.utk.edu>","shortMessageHtmlLink":"Fix collective initialization and finalization"}},{"before":"1edc1b5b05b203d949a2d64430247fb6a2cfea56","after":null,"ref":"refs/heads/topic/null_as_unused_buffers_for_collectives","pushedAt":"2024-04-01T15:45:19.000Z","pushType":"branch_deletion","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"}},{"before":"c0c77b702998704c61fee40b4867895fc7ab681e","after":"1f0fb1f5721918460af101a49dde5bb9efbd9c04","ref":"refs/heads/topic/dpu_offload_hcoll","pushedAt":"2024-03-26T19:37:00.000Z","pushType":"force_push","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Correctly link to the ocoms dependency.\n\nWhen correctly installed hcoll will pull the ocoms library\nautomatically. However, as this library is installed with an rpath if\nthe installation directory moved to a different location the ocoms\nlibrary will not be automatically pulled during link time. By adding it\nexplicitly we force the linker to look for it (will be found in the same\ndirectory as hcoll) and use it during link.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Correctly link to the ocoms dependency."}},{"before":null,"after":"c0c77b702998704c61fee40b4867895fc7ab681e","ref":"refs/heads/topic/dpu_offload_hcoll","pushedAt":"2024-03-26T19:35:56.000Z","pushType":"branch_creation","commitsCount":0,"pusher":{"login":"bosilca","name":null,"path":"/bosilca","primaryAvatarUrl":"https://avatars.githubusercontent.com/u/642701?s=80&v=4"},"commit":{"message":"Correctly link to the ocoms depedency.\n\nWhen correctly installed hcoll will pull the ocoms library\nautomatically. However, as this library is installed with an rpath if\nthe installation directory moved to a different location the ocoms\nlibrary will not be automatically pulled during link time. By adding it\nexplicitly we force the linker to look for it (will be found in the same\ndirectory as hcoll) and use it during link.\n\nSigned-off-by: George Bosilca <gbosilca@nvidia.com>","shortMessageHtmlLink":"Correctly link to the ocoms depedency."}}],"hasNextPage":true,"hasPreviousPage":false,"activityType":"all","actor":null,"timePeriod":"all","sort":"DESC","perPage":30,"cursor":"djE6ks8AAAAEUjxH8AA","startCursor":null,"endCursor":null}},"title":"Activity · bosilca/ompi"}