• Bug#1102068: libfabric: FTBFS on 32-bit arches: ofi_cma.h: error: passi

    From Drew Parsons@21:1/5 to All on Wed Apr 16 11:00:01 2025
    Source: libfabric
    Version: 2.1.0-1
    Followup-For: Bug #1102068
    Control: tags -1 ftbfs
    Control: reopen -1

    I think that bug closed by mpich 4.3.0-6 was meant to be one of the
    other mpich bugs (#1102612).

    32-bit arches are still failing to build libfabric 2.1.0-1 the same way,
    so reopening this bug.

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Alastair McKinstry@21:1/5 to Drew Parsons on Wed Apr 16 13:20:01 2025
    Apologies this was supposed to be #1102612

    Closing that as PMIX is now disabled; hydra in mpich (the mpiexec
    daemon) does not currently support pmix
    (which was added to mpich recently)

    Regards
    Alastair


    On 16/04/2025 09:54, Drew Parsons wrote:
    Source: libfabric
    Version: 2.1.0-1
    Followup-For: Bug #1102068
    Control: tags -1 ftbfs
    Control: reopen -1

    I think that bug closed by mpich 4.3.0-6 was meant to be one of the
    other mpich bugs (#1102612).

    32-bit arches are still failing to build libfabric 2.1.0-1 the same way,
    so reopening this bug.

    --
    Alastair McKinstry,
    GPG: 82383CE9165B347C787081A2CBE6BB4E5D9AD3A5
    e: alastair@mckinstry.ie, im: @alastair:mckinstry.ie @amckinstry@mastodon.ie

    Commander Vimes didn’t like the phrase “The innocent have nothing to fear,”
    believing the innocent had everything to fear, mostly from the guilty but in the longer term
    even more from those who say things like “The innocent have nothing to fear.”
    - T. Pratchett, Snuff

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Chris Hofstaedtler@21:1/5 to Drew Parsons on Wed May 7 20:10:02 2025
    On Fri, Apr 04, 2025 at 03:07:42PM +0200, Drew Parsons wrote:
    libfabric is failing to build on 32-bit architectures. The failure is blocking migration to testing, and will therefore block mpich migration.

    e.g. armhf https://buildd.debian.org/status/fetch.php?pkg=libfabric&arch=armhf&ver=2.0.0-4&stamp=1742062166&raw=0

    /bin/bash ./libtool --tag=CC --mode=compile gcc -DHAVE_CONFIG_H -I. -I./include -D_GNU_SOURCE -D__USE_XOPEN2K8 -DSYSCONFDIR=\"/etc\" -DRDMADIR=\"@rdmadir@\" -DPROVDLDIR=\"/usr/lib/arm-linux-gnueabihf/libfabric\" -I./prov/sockets/include -I./prov/
    sockets -I./prov/lpp/include -I./prov/lpp -I./prov/verbs/include -I./prov/verbs/include/linux -I./prov/verbs/src/ep_rdm -I./prov/verbs/src/ep_dgram -I./prov/lnx/include -I./prov/hook/include -I./prov/hook/perf/include -I./prov/hook/hook_debug/
    include -I./prov/hook/hook_hmem/include -I./prov/hook/dmabuf_peer_mem/include -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -D_TIME_BITS=64 -Wdate-time -D_FORTIFY_SOURCE=2 -Wall -O2 -DNDEBUG -g -O2 -Werror=implicit-function-declaration -ffile-prefix-map=/
    build/reproducible-path/libfabric-2.0.0=. -fstack-protector-strong -fstack-clash-protection -Wformat -Werror=format-security -c -o prov/hook/src/src_libfabric_la-hook.lo `test -f 'prov/hook/src/hook.c' || echo './'`prov/hook/src/hook.c
    In file included from ./include/ofi_util.h:59,
    from src/fabric.c:47:
    ./include/ofi_mr.h: In function 'ofi_mr_get_iov_from_dmabuf': ./include/ofi_mr.h:345:35: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
    345 | iov[i].iov_base = (void *) (
    | ^
    In file included from ./include/ofi_util.h:61:
    ./include/ofi_mem.h: In function 'smr_freestack_init': ./include/ofi_mem.h:281:53: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
    281 | next_aligned_addr = ofi_get_aligned_size((( (uint64_t) fs) +
    | ^ ./include/ofi_mem.h:283:54: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
    283 | fs->entry_base_offset = next_aligned_addr - ((uint64_t) fs);
    | ^
    In file included from ./include/ofi_shm_p2p.h:37,
    from src/fabric.c:54:
    ./include/ofi_cma.h: In function 'cma_copy':
    ./include/ofi_cma.h:67:40: error: passing argument 2 of 'ofi_consume_iov' from incompatible pointer type [-Wincompatible-pointer-types]
    67 | ofi_consume_iov(local, &local_cnt, (size_t) ret);
    | ^~~~~~~~~~
    | |
    | long unsigned int *
    In file included from ./include/ofi_cma.h:41:
    ./include/ofi_iov.h:204:51: note: expected 'size_t *' {aka 'unsigned int *'} but argument is of type 'long unsigned int *'


    This might be a gcc-15 error, which tightened type checking.

    Upstream has released v2.1.0 with gcc-15 fixes, though I don't know
    if it fixes this 32-bit build error.

    This is still a problem on armel, armhf, i386.

    Will something be done about it?

    Best,
    Chris

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Drew Parsons@21:1/5 to All on Mon May 12 15:40:02 2025
    Source: libfabric
    Followup-For: Bug #1102068
    Control: tags -1 ftbfs patch

    It looks like Gonzalo Silvalde Blanco's patch (salsa MR) at

    https://salsa.debian.org/hpc-team/libfabric/-/merge_requests/4

    should be able to fix the bug.

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)