open-mpi · janjust · Oct 24, 2025 · bosilca · Oct 27, 2025 · janjust
diff --git a/.ci/mellanox/README.md b/.ci/mellanox/README.md
@@ -6,7 +6,7 @@
 CI is managed by [Azure Pipelines](https://docs.microsoft.com/en-us/azure/devops/pipelines/?view=azure-devops) service.
 
 Mellanox Open MPI CI includes:
-* Open MPI building with internal stable engineering versions of UCX and HCOLL. The building is run in Docker-based environment.
+* Open MPI building with internal stable engineering versions of UCX. The building is run in Docker-based environment.
 * Sanity functional testing.
 ### How to Run CI
 Mellanox Open MPI CI is triggered upon the following events:

diff --git a/config/ompi_check_libhcoll.m4 b/config/ompi_check_libhcoll.m4
diff --git a/contrib/amca-param-sets/ft-mpi b/contrib/amca-param-sets/ft-mpi
@@ -63,7 +63,7 @@ btl=^usnic
 # The following frameworks/components are UNTESTED, and probably won't work.
 # They should run without faults, and will probably crash/deadlock after a fault.
 # You may try at your own risk.
-#   coll    hcoll, portals4
+#   coll    portals4
 #   topo    (all)
 #   osc     (all)
 #   io      (all)
@@ -72,7 +72,7 @@ btl=^usnic
 # We will disable only the components for which good components are known to exist.
 # Other untested components are selectable but will issue a runtime warning at
 # initiation if FT is enabled.
-coll=^hcoll,portals4
+coll=^portals4
 
 #
 # The following frameworks/components are NOT WORKING. Do not enable these with FT.

diff --git a/contrib/platform/intel/bend/linux b/contrib/platform/intel/bend/linux
@@ -13,7 +13,7 @@ enable_ipv6=no
 enable_man_pages=no
 enable_mpi_fortran=no
 enable_memchecker=no
-enable_mca_no_build=memchecker,coll-adapt,coll-cuda,coll-demo,coll-ftagree,coll-han,coll-hcoll,coll-inter,coll-libnbc,coll-monitoring,coll-portals4,coll-tuned,common-monitoring,common-ompio,fbtl,fcoll,fs,io,mtl,osc,pml-cm,pml-monitoring,pml-ucx,pml-v,sharedfp,topo,vprotocol,btl-ofi,btl-portals4,btl-smcuda,btl-uct,btl-ugni,btl-usnic,common-cuda,common-ofi,common-ucx
+enable_mca_no_build=memchecker,coll-adapt,coll-cuda,coll-demo,coll-ftagree,coll-han,coll-inter,coll-libnbc,coll-monitoring,coll-portals4,coll-tuned,common-monitoring,common-ompio,fbtl,fcoll,fs,io,mtl,osc,pml-cm,pml-monitoring,pml-ucx,pml-v,sharedfp,topo,vprotocol,btl-ofi,btl-portals4,btl-smcuda,btl-uct,btl-ugni,btl-usnic,common-cuda,common-ofi,common-ucx
 enable_contrib_no_build=libompitrace
 with_memory_manager=no
 with_devel_headers=yes

diff --git a/contrib/platform/lanl/toss/README.md b/contrib/platform/lanl/toss/README.md
@@ -43,7 +43,6 @@ created.
       (change S to X; make sure numbers match those for the same entry in
       contrib/platform/lanl/toss/optimized-mlx.conf)
     - addition: pml = ob1 (disable MXM)
-    - addition: coll = ^hcoll (disable MXM)
 - toss3-hfi-optimized
   - copy of toss2-qib-optimized
 - toss3-hfi-optimized.conf

diff --git a/contrib/platform/lanl/toss/toss2-mlx-optimized.conf b/contrib/platform/lanl/toss/toss2-mlx-optimized.conf
@@ -106,4 +106,3 @@ ras_base_launch_orted_on_hn = true
 
 ## Disable MXM
 pml = ob1
-coll = ^hcoll
diff --git a/contrib/platform/mellanox/optimized b/contrib/platform/mellanox/optimized
@@ -22,11 +22,6 @@ if [ "$mellanox_autodetect" == "yes" ]; then
         with_ucx=$ucx_dir
     fi
 
-    hcoll_dir=${hcoll_dir:="$(pkg-config --variable=prefix hcoll)"}
-    if [ -d $hcoll_dir ]; then
-        with_hcoll=$hcoll_dir
-    fi
-
     slurm_dir=${slurm_dir:="/usr"}
     if [ -f $slurm_dir/include/slurm/slurm.h ]; then
         with_slurm=$slurm_dir

diff --git a/docs/features/ulfm.rst b/docs/features/ulfm.rst
@@ -333,7 +333,7 @@ correctly after a failure.
   * ``cuda``, ``inter``, ``sync``, ``sm``: **untested** (they have not
     been modified to handle faults, but we expect correct post-fault
     behavior)
-  * ``hcoll``, ``portals4`` **disabled** (they have not been modified
+  * ``portals4`` **disabled** (it has not been modified
     to handle faults, and we expect unspecified post-fault behavior)
 
 * ``osc``: MPI one-sided communications

diff --git a/docs/installing-open-mpi/configure-cli-options/networking.rst b/docs/installing-open-mpi/configure-cli-options/networking.rst
@@ -14,15 +14,6 @@ can be used with ``configure``:
 
   FCA is the support library for Mellanox switches and HCAs.
 
-* ``--with-hcoll=DIR``:
-  Specify the directory where the Mellanox hcoll library and header
-  files are located.  This option is generally only necessary if the
-  hcoll headers and libraries are not in default compiler/linker
-  search paths.
-
-  hcoll is the support library for MPI collective operation offload on
-  Mellanox ConnectX-3 HCAs (and later).
-
 * ``--with-knem=DIR``:
   Specify the directory where the knem libraries and header files are
   located.  This option is generally only necessary if the knem headers

diff --git a/docs/tuning-apps/coll-tuned.rst b/docs/tuning-apps/coll-tuned.rst
@@ -3,7 +3,7 @@ Tuning Collectives
 
 Open MPI's ``coll`` framework provides a number of components implementing
 collective communication, including: ``han``, ``libnbc``, ``self``, ``ucc`` ``base``,
-``hcoll``, ``sync``, ``xhc``, ``accelerator``, ``basic``, ``ftagree``, ``inter``, ``portals4``,
+``sync``, ``xhc``, ``accelerator``, ``basic``, ``ftagree``, ``inter``, ``portals4``,
 and ``tuned``.  Some of these components may not be available depending on how
 Open MPI was compiled and what hardware is available on the system.  A run-time
 decision based on each component's self reported priority, selects which

diff --git a/docs/tuning-apps/networking/cuda.rst b/docs/tuning-apps/networking/cuda.rst
@@ -155,7 +155,6 @@ CUDA-aware support is available in:
 * The OFI (``ofi``) MTL with the CM (``cm``) PML.
 * Both CUDA-ized shared memory (``smcuda``) and TCP (``tcp``) BTLs
   with the OB1 (``ob1``) PML.
-* The HCOLL (``hcoll``) COLL
 
 /////////////////////////////////////////////////////////////////////////
 
@@ -702,30 +701,3 @@ to query rank information and utilize that to select a GPU.
 
 MPI internal CUDA resources are released during MPI_Finalize. Thus it is an
 application error to call cudaDeviceReset before MPI_Finalize is called.
-
-
-/////////////////////////////////////////////////////////////////////////
-
-How do I enable CUDA support in HCOLL collective component
-----------------------------------------------------------
-
-HCOLL component supports CUDA GPU buffers for the following
-collectives:
-
-MPI_Allreduce
-MPI_Bcast
-MPI_Allgather
-MPI_Ibarrier
-MPI_Ibcast
-MPI_Iallgather
-MPI_Iallreduce
-
-To enable CUDA GPU buffer support in these collectives pass the
-following environment variables via mpirun:
-
-.. code-block::
-
-   shell$ mpirun -x HCOLL_GPU_ENABLE=1 -x HCOLL_ENABLE_NBC=1 ..
-
-See `nVidia HCOLL documentation <https://docs.nvidia.com/networking/display/HPCXv29/HCOLL>`_
-for more information.
diff --git a/ompi/mca/coll/base/coll_base_allgather.c b/ompi/mca/coll/base/coll_base_allgather.c
@@ -291,8 +291,8 @@ int ompi_coll_base_allgather_intra_sparbit(const void *sbuf, size_t scount,
 
            /* Since each process sends several non-contiguos blocks of data, each block sent (and therefore each send and recv call) needs a different tag. */
            /* As base OpenMPI only provides one tag for allgather, we are forced to use a tag space from other components in the send and recv calls */
-           MCA_PML_CALL(isend(tmpsend + (ptrdiff_t) send_disp * scount * rext, scount, rdtype, sendto, MCA_COLL_BASE_TAG_HCOLL_BASE - send_disp, MCA_PML_BASE_SEND_STANDARD, comm, requests + transfer_count));
-           MCA_PML_CALL(irecv(tmprecv + (ptrdiff_t) recv_disp * rcount * rext, rcount, rdtype, recvfrom, MCA_COLL_BASE_TAG_HCOLL_BASE - recv_disp, comm, requests + data_expected - exclusion + transfer_count));
+           MCA_PML_CALL(isend(tmpsend + (ptrdiff_t) send_disp * scount * rext, scount, rdtype, sendto, MCA_COLL_BASE_TAG_NEIGHBOR_BASE - send_disp, MCA_PML_BASE_SEND_STANDARD, comm, requests + transfer_count));
+           MCA_PML_CALL(irecv(tmprecv + (ptrdiff_t) recv_disp * rcount * rext, rcount, rdtype, recvfrom, MCA_COLL_BASE_TAG_NEIGHBOR_BASE - recv_disp, comm, requests + data_expected - exclusion + transfer_count));
        }
        ompi_request_wait_all(transfer_count * 2, requests, MPI_STATUSES_IGNORE);
 

diff --git a/ompi/mca/coll/base/coll_base_allgatherv.c b/ompi/mca/coll/base/coll_base_allgatherv.c
@@ -332,12 +332,12 @@ int ompi_coll_base_allgatherv_intra_sparbit(const void *sbuf, size_t scount,
            if(ompi_count_array_get(rcounts, send_disp) > 0)
                MCA_PML_CALL(isend(tmpsend + ompi_disp_array_get(rdispls, send_disp) * rext,
                                   ompi_count_array_get(rcounts, send_disp), rdtype, sendto,
-                                  MCA_COLL_BASE_TAG_HCOLL_BASE - send_disp,
+                                  MCA_COLL_BASE_TAG_NEIGHBOR_BASE - send_disp,
                                   MCA_PML_BASE_SEND_STANDARD, comm, requests + step_requests++));
            if(ompi_count_array_get(rcounts, recv_disp) > 0)
                MCA_PML_CALL(irecv(tmprecv + ompi_disp_array_get(rdispls, recv_disp) * rext,
                                   ompi_count_array_get(rcounts, recv_disp), rdtype, recvfrom,
-                                  MCA_COLL_BASE_TAG_HCOLL_BASE - recv_disp, comm,
+                                  MCA_COLL_BASE_TAG_NEIGHBOR_BASE - recv_disp, comm,
                                   requests + step_requests++));
        }
        ompi_request_wait_all(step_requests, requests, MPI_STATUSES_IGNORE);

diff --git a/ompi/mca/coll/base/coll_tags.h b/ompi/mca/coll/base/coll_tags.h
@@ -69,10 +69,8 @@
 #define MCA_COLL_BASE_TAG_NONBLOCKING_END ((-1 * INT_MAX/2) + 1)
 #define MCA_COLL_BASE_TAG_NEIGHBOR_BASE  (MCA_COLL_BASE_TAG_NONBLOCKING_END - 1)
 #define MCA_COLL_BASE_TAG_NEIGHBOR_END   (MCA_COLL_BASE_TAG_NEIGHBOR_BASE - 1024)
-#define MCA_COLL_BASE_TAG_HCOLL_BASE (-1 * INT_MAX/2)
-#define MCA_COLL_BASE_TAG_HCOLL_END (-1 * INT_MAX)
 
 #define MCA_COLL_BASE_TAG_BASE MCA_COLL_BASE_TAG_BLOCKING_BASE
-#define MCA_COLL_BASE_TAG_END  MCA_COLL_BASE_TAG_HCOLL_END
+#define MCA_COLL_BASE_TAG_END  MCA_COLL_BASE_TAG_NEIGHBOR_END
 
 #endif /* MCA_COLL_BASE_TAGS_H */
diff --git a/ompi/mca/coll/hcoll/Makefile.am b/ompi/mca/coll/hcoll/Makefile.am
Original file line number	Diff line number	Diff line change
Expand Up		@@ -106,4 +106,3 @@ ras_base_launch_orted_on_hn = true

		## Disable MXM
		pml = ob1
		coll = ^hcoll