Bump soname to 21:0:6 (vs 20:3:5 for 2.7.1),

Reuse AM 1.16.5 and AC 2.71, but bump LT to 2.4.7.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Still too experimental, might get removed if we don't find some manpower to revive it

And also disable netloc in sonarscanner job

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 2082c24
                       and 15a927b)
(cherry picked from commit eddcbcf)
(cherry picked from commit 30075ab)
(cherry picked from commit 29af441)
(cherry picked from commit dfab43e)
(cherry picked from commit 4fcdc93)
(cherry picked from commit b2eff2b)
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Reuse the timeslot of v2.3 which is unlikely to ever run again.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

It's already defined in Cygwin's /usr/include/w32api/_mingw.h

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit c91712e)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit ac18e13)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 4732ec4)

Thanks to Jonathan Peyton for the fix.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 1f0283e)

This issue seems to happen more frequently, don't report it unless
all errors are enabled.

This warning means that some PCI devices might be missing.
That's not critical enough to warn users (most of them know nothing about hwloc).
For the record, CUDA init warnings were demoted the same a couple releases ago.

Refs open-mpi#354

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 69fbc20)

…ors too

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 64e01e8)

Forgotten in commit 69fbc20

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit c100a52)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 3329842)

Don't fallthough to xml filenames after setting COMPREPLY for --refname.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit c57d0fe)

We were still using index instead of id.

Thanks to Guillaume Mercier.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 61083f1)

Closes open-mpi#542

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 8d81da3)

Escape double-quotes in comments (they break the parser and the remaining
enum item doc is ignored in manpages).

Hide initializer values.

Uniformize the source formatting to keep it easy to read even if the
manpage formatting isn't perfect (looks like doxygen/manpage wants
only two paragraph per enum item, but we'd have to change lots of
other places, and it would make the source harder to read).

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit d316fa7)

MinGW doesn't get uint64_t on our CI.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 601b58d)

Possible version values aren't documented in NVML API doc,
but wikipedia and other sources say 50GB/s per link
for v3 (Ampere) and v4 (Hopper).

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 352348a)

hwloc_backends.c needs some open/read/close/mktemp redirection
to _foo. Copy what we did in pci-common.c and add mktemp.

Thanks to Mario Emmenlauer for the report.

Fixes open-mpi#546

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 0de22a5)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 3b5d5d8)

128 FP32 cores again.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit a908085)

Looks like we always need free the locally allocated infos.

(cherry picked from commit b86f63f)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 4630002)

And 8.7 is officially documented now.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 604fdd5)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit e84553d)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit b5c07dd)

And use "CXLMem" as a string for PCI_CLASS_MEMORY_CXL = 0x0502

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit e0f56e2)

We'll need to double-check whether they are useful in the end.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit a172c84)

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit c938d81)

…onditional

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit 471e43d)

HWLOC_VERSION* macros were X.Yrc1 instead of X.Y in the MSVC builds.

Windows CMAKE builds don't include suffixes such as rc1, but we still
update the hwloc version there for consistency (and in case X.Y was
changed too).

Android isn't updated since it isn't distributed in tarballs.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit accb702)

Forgotten in f7c9aa8

Thanks to Michel Lesoinne for the report.

Closes open-mpi#564

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
(cherry picked from commit aa0ef16)

Currently only effective for MacOS. Hoping it can be improved to become general for more systems.