]> xenbits.xensource.com Git - people/andrewcoop/hwloc.git/log
people/andrewcoop/hwloc.git
11 years agoplugins: cleanup hwloc_setup_pu_level() and export it to plugins x86-common
Brice Goglin [Thu, 13 Feb 2014 11:34:51 +0000 (12:34 +0100)]
plugins: cleanup hwloc_setup_pu_level() and export it to plugins

Xen will need it.

11 years agoplugins: export hwloc_alloc_root_sets()
Brice Goglin [Thu, 13 Feb 2014 11:29:44 +0000 (12:29 +0100)]
plugins: export hwloc_alloc_root_sets()

Renamed from the old hwloc_alloc_obj_cpusets(),
now officially specialized for root objects only,
and now exported to plugins (Xen will need it).

11 years agohwloc/plugins.h: protect cpuid functions with HWLOC_HAVE_X86_CPUID
Brice Goglin [Thu, 13 Feb 2014 10:28:32 +0000 (11:28 +0100)]
hwloc/plugins.h: protect cpuid functions with HWLOC_HAVE_X86_CPUID

But keep them enabled in the doc even when cpuid isn't available.

And make sure external plugins get HWLOC_HAVE_X86_CPUID from the public
autogen/config.h

11 years agox86: export hwloc_x86_cpuid_discovery() to plugins
Brice Goglin [Tue, 11 Feb 2014 12:00:39 +0000 (13:00 +0100)]
x86: export hwloc_x86_cpuid_discovery() to plugins

Make it accept a cpuid callback+data for Xen.

11 years agox86: make the x86 detection configurable with flags (internally)
Brice Goglin [Tue, 11 Feb 2014 09:46:54 +0000 (10:46 +0100)]
x86: make the x86 detection configurable with flags (internally)

Use flags to enable full discovery, caches only, or annotating cpuinfo only.
Enable them depending on whether caches or other objects being already present.

11 years agocontrib/update-my-copyright.pl: fix join() when rewriting the file
Brice Goglin [Thu, 13 Feb 2014 12:04:27 +0000 (13:04 +0100)]
contrib/update-my-copyright.pl: fix join() when rewriting the file

When working on NEWS, we were prefixing all lines (but the first) with a "1".
Join wants a "" for empty string, not //.

11 years agohwloc/plugins.h: move common PCI func to a dedicated section
Brice Goglin [Thu, 13 Feb 2014 09:38:45 +0000 (10:38 +0100)]
hwloc/plugins.h: move common PCI func to a dedicated section

11 years agocpuid: rename cpuid into x86_cpuid everywhere
Brice Goglin [Thu, 13 Feb 2014 08:26:55 +0000 (09:26 +0100)]
cpuid: rename cpuid into x86_cpuid everywhere

To avoid possible confusion if we add some sort of cpuid-related code for
other architectures such as ARM.

11 years agoRevert "x86: check hwloc_have_cpuid() in the backend instantiate instead of late...
Brice Goglin [Thu, 13 Feb 2014 08:46:35 +0000 (09:46 +0100)]
Revert "x86: check hwloc_have_cpuid() in the backend instantiate instead of late in the discovery"

This reverts commit 6a1cb59936460ff798970929f0bd52fa8da2cc00.

Don't require others caller of look_x86() (Xen soon) to manually check
for cpuid availability at runtime.

11 years agosynthetic: Remove duplicate check for the number of PU levels
Brice Goglin [Tue, 11 Feb 2014 21:27:51 +0000 (22:27 +0100)]
synthetic: Remove duplicate check for the number of PU levels

Found by Coverity Scan.

11 years agodiff: fix memory leak when appending a "too complex" diff entry
Brice Goglin [Tue, 11 Feb 2014 21:27:13 +0000 (22:27 +0100)]
diff: fix memory leak when appending a "too complex" diff entry

Found by Coverity Scan.

11 years agox86: use fake binding hooks to let look_procs work on unsupported OSes
Brice Goglin [Tue, 11 Feb 2014 14:30:38 +0000 (15:30 +0100)]
x86: use fake binding hooks to let look_procs work on unsupported OSes

11 years agox86: simplify the management of binding hooks
Brice Goglin [Tue, 11 Feb 2014 10:44:10 +0000 (11:44 +0100)]
x86: simplify the management of binding hooks

We check early whether binding is available, record the thread or proc
support there, and just use that one later, instead of trying both.

11 years agox86: check hwloc_have_cpuid() in the backend instantiate instead of late in the discovery
Brice Goglin [Tue, 11 Feb 2014 10:30:53 +0000 (11:30 +0100)]
x86: check hwloc_have_cpuid() in the backend instantiate instead of late in the discovery

11 years agosynthetic: rework the storage of synthetic info in the backend private data
Brice Goglin [Tue, 4 Feb 2014 19:23:53 +0000 (20:23 +0100)]
synthetic: rework the storage of synthetic info in the backend private data

11 years agosynthetic: add support for L2cache, L1i, group3, etc for for input and export
Brice Goglin [Mon, 3 Feb 2014 21:13:09 +0000 (22:13 +0100)]
synthetic: add support for L2cache, L1i, group3, etc for for input and export

11 years agoAdd hwloc_obj_type_sscanf() to extend hwloc_obj_type_of_string()
Brice Goglin [Mon, 3 Feb 2014 19:21:59 +0000 (20:21 +0100)]
Add hwloc_obj_type_sscanf() to extend hwloc_obj_type_of_string()

The former was already used in hwloc-calc.h to parse things like
"L2iCache" or "Group3". Make it officially part of the API and
deprecate the old hwloc_obj_type_of_string().

11 years agotests: don't abuse synthetic topology strings
Brice Goglin [Tue, 4 Feb 2014 19:14:35 +0000 (20:14 +0100)]
tests: don't abuse synthetic topology strings

Use what's officially supported, no deprecated or possibly ambiguous syntax.

11 years agohwloc_obj_attr_snprintf: don't mix local and total memory when printing verbose attri...
Brice Goglin [Tue, 4 Feb 2014 21:04:07 +0000 (22:04 +0100)]
hwloc_obj_attr_snprintf: don't mix local and total memory when printing verbose attributes

11 years agoNEWS: final v1.8.1 bullets
Brice Goglin [Tue, 4 Feb 2014 14:04:20 +0000 (15:04 +0100)]
NEWS: final v1.8.1 bullets

11 years agoMerge branch 'master' of github.com:open-mpi/hwloc
Samuel Thibault [Tue, 4 Feb 2014 10:02:20 +0000 (11:02 +0100)]
Merge branch 'master' of github.com:open-mpi/hwloc

11 years agoGive a name to the lstopo window
Samuel Thibault [Tue, 4 Feb 2014 10:01:51 +0000 (11:01 +0100)]
Give a name to the lstopo window

11 years agodoc: add hwloc_linux_get_tid_last_cpu_location manpage
Brice Goglin [Mon, 3 Feb 2014 20:52:41 +0000 (21:52 +0100)]
doc: add hwloc_linux_get_tid_last_cpu_location manpage

11 years agohwloc-distrib(1): clarify --from
Brice Goglin [Mon, 3 Feb 2014 18:33:30 +0000 (19:33 +0100)]
hwloc-distrib(1): clarify --from

11 years agoNEWS: minor fixes
Brice Goglin [Mon, 3 Feb 2014 18:33:07 +0000 (19:33 +0100)]
NEWS: minor fixes

11 years agoBring 1.8.1 NEWS in master and update 1.9 NEWS bullets
Brice Goglin [Thu, 30 Jan 2014 14:41:36 +0000 (15:41 +0100)]
Bring 1.8.1 NEWS in master and update 1.9 NEWS bullets

11 years agotests/linux: add /proc/hwloc-nofile-info with Architecture field to some tarballs...
Brice Goglin [Mon, 27 Jan 2014 22:33:53 +0000 (23:33 +0100)]
tests/linux: add /proc/hwloc-nofile-info with Architecture field to some tarballs to test the new CPUVendor/... code

11 years agoLinux: add CPUVendor/CPUModelNumber/CPUFamilyNumber on Linux x86, x86_64, ia64 and...
Brice Goglin [Mon, 27 Jan 2014 22:33:19 +0000 (23:33 +0100)]
Linux: add CPUVendor/CPUModelNumber/CPUFamilyNumber on Linux x86, x86_64, ia64 and Xeon Phi

Uses the uname architecture to properly match /proc/cpuinfo fields.

Requested by Ralph Castain.

11 years agoLinux: use the uname architecture to better match cpumodel information
Brice Goglin [Mon, 27 Jan 2014 17:45:34 +0000 (18:45 +0100)]
Linux: use the uname architecture to better match cpumodel information

Look at the uname architecture to select which cpuinfo parsing function
will be used.
For now, we have one for x86/x86_64/mic, one for ia64 and one for arm.
Others (or when the architecture is unknown, when fsroot was changed
and there's no /proc/hwloc-nofile-info) fallback to the default "dumb"
parsing function.

11 years agoLinux: prepare for gathering more than just the main cpu model name
Brice Goglin [Mon, 27 Jan 2014 16:01:44 +0000 (17:01 +0100)]
Linux: prepare for gathering more than just the main cpu model name

Pass some structures between functions instead of a single cpumodel pointer.

11 years agoLinux: allow to read uname from fake fs root
Brice Goglin [Mon, 27 Jan 2014 15:19:27 +0000 (16:19 +0100)]
Linux: allow to read uname from fake fs root

Read a new "fake" /proc/hwloc-nofile-info where we can store things
that cannot be saved as a file. Put the output of uname there when
gathering the topology and use it on input to overwrite local or
missing data.

This improves testing under tests/linux by allowing architecture
specific code to run from fake fsroot. Will be useful for advanced
cpuinfo parsing.

We manually remove /proc/hwloc-nofile-info in test-gather-topology.sh
because it only partially brings fake fsroot information.
Removing it creates the same !is_thissystem() output from the fake
fsroot and from the real machine.

Once we gather other information in /proc/hwloc-nofile-info:
* sysconf(_SC_LARGE_PAGESIZE);
* hwloc_getpagesize();
* hwloc_fallback_nbprocessors(topology);
we may be able to really use /proc/hwloc-nofile-info in test-gather-topology.sh
and really discover the exact same thing from the saved topology and
from the original machine. We'll see that later.

11 years agocore: allow backends to reuse a cached uname info when adding uname string infos
Brice Goglin [Mon, 27 Jan 2014 14:43:57 +0000 (15:43 +0100)]
core: allow backends to reuse a cached uname info when adding uname string infos

And ignore any empty uname field in case the backend gives incomplete info.

11 years agoREADME: Remove the doxygen footer so that the README doesn't get modified too often
Brice Goglin [Wed, 29 Jan 2014 21:05:34 +0000 (22:05 +0100)]
README: Remove the doxygen footer so that the README doesn't get modified too often

By the way, update the README.

11 years agorename: fix diff renaming
Brice Goglin [Wed, 29 Jan 2014 20:57:50 +0000 (21:57 +0100)]
rename: fix diff renaming

11 years agodoc: document new CPU info attributes
Brice Goglin [Wed, 29 Jan 2014 20:48:53 +0000 (21:48 +0100)]
doc: document new CPU info attributes

11 years agoUpdate hwloc/rename.h
Brice Goglin [Wed, 29 Jan 2014 20:12:52 +0000 (21:12 +0100)]
Update hwloc/rename.h

11 years agox86: Add CPUVendor/CPUModelNumber/CPUFamilyNumber info
Brice Goglin [Tue, 28 Jan 2014 09:27:49 +0000 (10:27 +0100)]
x86: Add CPUVendor/CPUModelNumber/CPUFamilyNumber info

Requested by Ralph Castaing.

11 years agocore: Add add_info_nodup() to optionally check for duplicate info keypairs
Brice Goglin [Wed, 29 Jan 2014 19:51:29 +0000 (20:51 +0100)]
core: Add add_info_nodup() to optionally check for duplicate info keypairs

11 years agox86: Add uname info when not done yet, or at least set Architecture=x86/x86_64
Brice Goglin [Wed, 29 Jan 2014 19:07:09 +0000 (20:07 +0100)]
x86: Add uname info when not done yet, or at least set Architecture=x86/x86_64

11 years agox86: factorize the code that uses thread or process-binding for cpuid lookups
Brice Goglin [Wed, 29 Jan 2014 15:39:03 +0000 (16:39 +0100)]
x86: factorize the code that uses thread or process-binding for cpuid lookups

11 years agoSave the windows build script for ci.inria.fr under contrib
Brice Goglin [Wed, 29 Jan 2014 10:03:37 +0000 (11:03 +0100)]
Save the windows build script for ci.inria.fr under contrib

11 years agoFix cpuid on windows 64bits so that the x86 backend is enabled as expected
Brice Goglin [Tue, 28 Jan 2014 14:41:57 +0000 (15:41 +0100)]
Fix cpuid on windows 64bits so that the x86 backend is enabled as expected

unsigned long isn't 64bits on win64, so the cpuid assembly code fails to
build during configure (%ebx and %rbx gets mixed in the assembly),
and the entire x86 backend gets disabled.

Use hwloc_uint64_t instead. It needs hwloc/autogen/config.h which isn't
ready during configure yet, so manually add #include/#define during
the cpuid check.

Thanks to Robin Scher for reporting the problem.

11 years agoRun "make doc" before "make distcheck" in the nightly tarball script.
Jeff Squyres [Mon, 27 Jan 2014 14:50:58 +0000 (09:50 -0500)]
Run "make doc" before "make distcheck" in the nightly tarball script.

It is now necessary to have the docs built before running "make dist",
so update the nightly script to do that.

11 years agodoc: only try to copy doxygen.css when the doc was rebuilt
Brice Goglin [Fri, 24 Jan 2014 09:53:31 +0000 (10:53 +0100)]
doc: only try to copy doxygen.css when the doc was rebuilt

11 years agodoc: build the top-level README as a single rule
Brice Goglin [Fri, 24 Jan 2014 09:47:40 +0000 (10:47 +0100)]
doc: build the top-level README as a single rule

2 rules with intermediate file being removed means
that those rules always rerun.

11 years agodoc: further simplify doc/Makefile.am
Brice Goglin [Fri, 24 Jan 2014 09:40:31 +0000 (10:40 +0100)]
doc: further simplify doc/Makefile.am

When doxygen and readme are buildable, we always build them.
When not, we just do nothing and let config/distscript.csh fail later.

11 years agodoc/Makefile.am: factorize AM if/endif and remove obsolete stuff
Brice Goglin [Fri, 24 Jan 2014 09:21:22 +0000 (10:21 +0100)]
doc/Makefile.am: factorize AM if/endif and remove obsolete stuff

11 years agodist: no need to specify anything from doxygen-doc/ under doc/Makefile.am
Brice Goglin [Thu, 23 Jan 2014 18:17:18 +0000 (19:17 +0100)]
dist: no need to specify anything from doxygen-doc/ under doc/Makefile.am

doxygen-doc/ is manually copied by config/distscript.csh

Therefore, let config/distscript.csh handle failure to create the doc
when doxygen is missing.

By the way, update obsolete comments about adding entire directories
to EXTRA_DIST.

11 years agodist: don't automatically rebuild doxygen on each make dist
Brice Goglin [Thu, 23 Jan 2014 14:54:20 +0000 (15:54 +0100)]
dist: don't automatically rebuild doxygen on each make dist

Warn if the doc doesn't exist, or just use what's already there.

contrib/dist/make_dist_tarball takes care of removing/recreating the doc,
and the doc is already correctly rebuilt in most cases, it is built by
make dist when not already available.
No need to keep complex rules for always rebuilding the doc just in case
somebody ever builds a tarball manually and ends up containing an outdated
doc for some reason.

Remove --enable-doxygen from make distcheck now that things are simple.

11 years agodoc: Fix the error PDFs are needed but doxygen isn't available
Brice Goglin [Thu, 23 Jan 2014 14:53:34 +0000 (15:53 +0100)]
doc: Fix the error PDFs are needed but doxygen isn't available

11 years agoFix make dist during make distcheck
Brice Goglin [Wed, 22 Jan 2014 15:22:57 +0000 (16:22 +0100)]
Fix make dist during make distcheck

It wants to generate doxygen docs, but it doesn't by default
since it builds from a tarball. Force --enable-doxygen to fix that.

By the way, no need to double-check that doc/readme were generated
before building the tarball. config/distscript.csh is called on dist
and it does things as expected.

Also add a better error message when make dist fails because
PDFs do not exist yet, and doxygen isn't available.

Ideally, we'd just remove distscript.csh to make things much simpler
but I don't want to break more things today.

11 years agoconfigure: Fix the disabling of doxygen when building from tarballs
Brice Goglin [Tue, 21 Jan 2014 14:31:24 +0000 (15:31 +0100)]
configure: Fix the disabling of doxygen when building from tarballs

We check if we are in a developer tree, display the result, but do nothing with it.
Add the relevant code to update enable_doxygen accordingly if it wasn't forced
on the command-line.

This was unnoticed for a long time because we distribute hwloc.tag,
which causes tarball-builds not to regenerate the doc.
But they will regenerate it as soon as a dependency changes.

Now doxygen is entirely disabled properly in tarball-builds for real
as long as --enable-doxygen is not given.

11 years agotests/linux: add a test with very buggy NUMA info
Brice Goglin [Thu, 16 Jan 2014 14:07:18 +0000 (15:07 +0100)]
tests/linux: add a test with very buggy NUMA info

Linux reports 8 NUMA nodes containing the (same) entire list of all cores,
hence all of them get merged.

Thanks to Jeff Becker for the test case.

11 years agolinux/NUMA: Work around buggy NUMA node cpusets
Brice Goglin [Thu, 16 Jan 2014 20:06:15 +0000 (21:06 +0100)]
linux/NUMA: Work around buggy NUMA node cpusets

If Linux reports multiple nodes with intersecting (or identical) cpusets,
they get merged and the Linux backend asserts false.

Don't assert when insertion returns a merge error anymore.
Just disable NUMA distances if that ever happens since one
object is missing (instead of removing the corresponding row/column).
The BIOS is buggy, the matrix would likely be meaningless anyway.

Thanks to Jeff Becker for reporting the problem on the OMPI list.

By the way, fix the number of nodes returned by look_sysfsnode()
when we failed to read/create some NUMA objects.

11 years agocore: switch more printf(stderr) to unique messages that can be silenced with HWLOC_H...
Brice Goglin [Thu, 16 Jan 2014 13:39:21 +0000 (14:39 +0100)]
core: switch more printf(stderr) to unique messages that can be silenced with HWLOC_HIDE_ERRORS=1

Useful when debugging failures to merge because of different OS indexes or levels.

11 years agoupdate-my-copyright.pl: update to git releases that don't prefix git status output...
Brice Goglin [Thu, 16 Jan 2014 13:38:10 +0000 (14:38 +0100)]
update-my-copyright.pl: update to git releases that don't prefix git status output with # by default anymore

11 years agoChange the logic in bind.c to only include <malloc.h> if we don't have posix_memalign.
Jeff Squyres [Fri, 10 Jan 2014 17:05:48 +0000 (09:05 -0800)]
Change the logic in bind.c to only include <malloc.h> if we don't have posix_memalign.

In http://www.open-mpi.org/community/lists/devel/2014/01/13619.php,
Paul Hargrove found a compiler warning on OpenBSD where <malloc.h>
exists, but is not intended to be used (and doesn't error out, so
AC_CHECK_HEADERS says its ok).

11 years agoIf a git command fails, also print the repo in the error message
Jeff Squyres [Thu, 9 Jan 2014 14:34:24 +0000 (09:34 -0500)]
If a git command fails, also print the repo in the error message

11 years agoKeep a copy of the script that we use for sending git diff emails.
Jeff Squyres [Wed, 8 Jan 2014 23:09:46 +0000 (18:09 -0500)]
Keep a copy of the script that we use for sending git diff emails.

11 years agohwloc-ps: don't report locations as caches unless there is nothing equivalent
Brice Goglin [Wed, 18 Dec 2013 19:46:03 +0000 (20:46 +0100)]
hwloc-ps: don't report locations as caches unless there is nothing equivalent

We report the higher object matching the binding.
Go down if it's a cache that has a single identical child.

11 years agohwloc-bind: Add -e as a synonym to --get-last-cpu-location
Brice Goglin [Wed, 18 Dec 2013 19:42:57 +0000 (20:42 +0100)]
hwloc-bind: Add -e as a synonym to --get-last-cpu-location

To match what hwloc-ps supports.

11 years agohwloc-ps: Add -e/--get-last-cpu-location
Brice Goglin [Wed, 18 Dec 2013 19:22:32 +0000 (20:22 +0100)]
hwloc-ps: Add -e/--get-last-cpu-location

11 years agolinux: export hwloc_linux_get_tid_last_cpu_location()
Brice Goglin [Wed, 18 Dec 2013 19:12:17 +0000 (20:12 +0100)]
linux: export hwloc_linux_get_tid_last_cpu_location()

This is equivalent to hwloc_get_proc_last_cpu_location(THREAD) on Linux,
but it's better for clarity to have this Linux specific call as well
(and it was already implemented internally)

11 years agohwloc.h: Fix proc_cpubind() proc_last_cpu_location() documentation when using Linux TID
Brice Goglin [Wed, 18 Dec 2013 19:03:57 +0000 (20:03 +0100)]
hwloc.h: Fix proc_cpubind() proc_last_cpu_location() documentation when using Linux TID

and document that hwloc_linux_*_tid_cpubind() are equivalent.

11 years agoutils: fix cpubind support checks
Brice Goglin [Wed, 18 Dec 2013 18:31:03 +0000 (19:31 +0100)]
utils: fix cpubind support checks

11 years agoautomake subdir-object mode needs check for -c -o
Rob Latham [Fri, 6 Dec 2013 00:09:06 +0000 (09:09 +0900)]
automake subdir-object mode needs check for -c -o

Older automake versions (at least 1.13.1) will complain if we asked for subdir
mode and we did not verify that both -c and -o flags work (AM_PROG_CC_C_O)

11 years agoupdate master NEWS from v1.8
Brice Goglin [Mon, 18 Nov 2013 12:27:15 +0000 (05:27 -0700)]
update master NEWS from v1.8

11 years agoAPI: fix some doxygen refs
Brice Goglin [Mon, 18 Nov 2013 11:45:04 +0000 (04:45 -0700)]
API: fix some doxygen refs

11 years agoScript that is run on the build server to make nightly snapshot tarballs
Jeff Squyres [Fri, 15 Nov 2013 17:02:25 +0000 (09:02 -0800)]
Script that is run on the build server to make nightly snapshot tarballs

11 years agocustom: document that objects above assembled topologies have NULL cpuset/nodeset
Brice Goglin [Tue, 12 Nov 2013 13:40:34 +0000 (14:40 +0100)]
custom: document that objects above assembled topologies have NULL cpuset/nodeset

Thanks to Jérôme Clet-Ortega for reporting the need for clarification.

11 years agotests/hwloc_insert_misc: more assertions
Brice Goglin [Thu, 7 Nov 2013 10:20:49 +0000 (11:20 +0100)]
tests/hwloc_insert_misc: more assertions

To detect spurious failures on AIX because of ENOMEM.

11 years agotest-hwloc-diffpatch: exit the temp directory before removing it
Brice Goglin [Thu, 7 Nov 2013 08:32:05 +0000 (09:32 +0100)]
test-hwloc-diffpatch: exit the temp directory before removing it

Otherwise rmdir fails on AIX

11 years agoFix unused parameter warning
Samuel Thibault [Wed, 6 Nov 2013 16:27:35 +0000 (17:27 +0100)]
Fix unused parameter warning

11 years agoFix types of RGB colors.
Samuel Thibault [Wed, 6 Nov 2013 16:26:11 +0000 (17:26 +0100)]
Fix types of RGB colors.

11 years agoFix mixed declaration/code warning
Samuel Thibault [Wed, 6 Nov 2013 16:25:43 +0000 (17:25 +0100)]
Fix mixed declaration/code warning

11 years agoWe're 100% git these days.
Jeff Squyres [Wed, 6 Nov 2013 14:39:48 +0000 (06:39 -0800)]
We're 100% git these days.

11 years agomaster is now v1.9
Brice Goglin [Tue, 5 Nov 2013 21:00:04 +0000 (22:00 +0100)]
master is now v1.9

11 years agoRemove Cairo dependency on X11 and factorize X11 tests
Brice Goglin [Wed, 6 Nov 2013 08:49:00 +0000 (09:49 +0100)]
Remove Cairo dependency on X11 and factorize X11 tests

Per discussion on the mailing list, it doesn't look like Cairo needs
X11 any more (it doesn't on OS X to generate PDFs and PNGs).  So
remove the dependency in configure.ac.

Also, move the X11 tests from outside of the Cairo tests (and outside
of the GL tests) to be in the main configure logic.

By the way, rename HWLOC_HAVE_X11 into HWLOC_HAVE_X11_KEYSYM to avoid
confusion between Xlib.h/XOpenDisplay being available (what GL needs)
and Xutil.h+keysym.h being available too (what lstopo/X11 needs).

Based on a patch from Jeff, tweaked by me.

11 years agocleanup v1.8 NEWS bullets
Brice Goglin [Sun, 3 Nov 2013 22:57:25 +0000 (23:57 +0100)]
cleanup v1.8 NEWS bullets

11 years agoBump AM to 1.14 for v1.8
Brice Goglin [Sat, 26 Oct 2013 07:34:36 +0000 (09:34 +0200)]
Bump AM to 1.14 for v1.8

11 years agolinux/mic: add a MICSerialNumber info attribute when running inside the MIC
Brice Goglin [Thu, 17 Oct 2013 17:04:50 +0000 (19:04 +0200)]
linux/mic: add a MICSerialNumber info attribute when running inside the MIC

OMPI people want an easy way to recognize MICs and nodes using hwloc.
We already have MICSerialNumber inside OS devices in the host topology,
add the same to the root object of MIC topologies.
Unfortunately, we couldn't find anything better than parsing /proc/elog
to find that number.

11 years agohwloc-compress-dir: allow to uncompress with -R
Brice Goglin [Thu, 17 Oct 2013 14:40:10 +0000 (16:40 +0200)]
hwloc-compress-dir: allow to uncompress with -R

Modify the test case a bit because the input XMLs weren't totally perfect.

11 years agohwloc-patch: add support for automatically guessing the input topology filename from...
Brice Goglin [Thu, 17 Oct 2013 13:10:42 +0000 (15:10 +0200)]
hwloc-patch: add support for automatically guessing the input topology filename from the XML refname

You may do "hwloc-patch refname foo.diff.xml", it will replace "refname"
with the topology that was used as a reference when generating the diff.

11 years agotools: Add --whole-system and --restrict for uniformity
Brice Goglin [Tue, 15 Oct 2013 09:13:40 +0000 (11:13 +0200)]
tools: Add --whole-system and --restrict for uniformity

Add --whole-system to hwloc-bind, hwloc-calc, hwloc-distances and
hwloc-distrib, and add --restrict to hwloc-bind for uniformity among
tools.

Based on http://www.open-mpi.org/community/lists/hwloc-users/2013/10/0908.php

11 years agocontrib/update-my-copyright.pl: fix git status parsing for non-english locales
Brice Goglin [Tue, 15 Oct 2013 08:45:11 +0000 (10:45 +0200)]
contrib/update-my-copyright.pl: fix git status parsing for non-english locales

Pass LANG=C to git status so that we properly recognize modified files.

11 years agoFix make_nightly_snapshot processing.
Jeff Squyres [Fri, 4 Oct 2013 02:19:48 +0000 (19:19 -0700)]
Fix make_nightly_snapshot processing.

Leave *only one place* where "git desribe" is invoked and set into the
snapshot_version field in VERSION: the make_nightly_snapshot script.
distscript.csh no longer twiddles the VERSION file.

11 years agoUpdates for tarball generation to handle new git infrastructure.
Jeff Squyres [Thu, 3 Oct 2013 14:50:55 +0000 (10:50 -0400)]
Updates for tarball generation to handle new git infrastructure.

- No longer download latest config.sub|guess in distscript.csh; the
  GNU Autotools are updated frequently enough these days that the
  bundled versions are fine
- Rename nightly/create_tarball.sh -> nightly/make_snapshot_tarball
- Add "snapshot" and "snapshot_version" fields in VERSION; removed
  want_repo_rev and repo_rev fields
- Updated make_dist_tarball and make_snapshot_tarball to handle git
- Do not handle SVN or Mercurial repos any more
- Somewhat simplified hwloc_get_version.sh:
  - Never fill in a repo version number; always just report what is in
    VERSION already (and error out if there's not enough information
    in VERSION).  Only distscript.csh will insert a value into
    VERSION:snapshot_version (when needed).
  - Removed extra CLI options (e.g., get major/minor/etc. version
    number components)

11 years agoUpdate to only handle Git repos.
Jeff Squyres [Thu, 3 Oct 2013 14:46:15 +0000 (10:46 -0400)]
Update to only handle Git repos.

No longer handle SVN or Mercurial repos.

11 years agoConvert SVN URLs to git URLs in documentation.
Jeff Squyres [Thu, 3 Oct 2013 14:29:29 +0000 (10:29 -0400)]
Convert SVN URLs to git URLs in documentation.

11 years agoUpdate .gitignore.
Jeff Squyres [Thu, 3 Oct 2013 14:27:59 +0000 (10:27 -0400)]
Update .gitignore.

11 years agotests/embedded: improve version extraction
Jeff Squyres [Thu, 3 Oct 2013 14:27:25 +0000 (10:27 -0400)]
tests/embedded: improve version extraction

Avoid troubles when the tarball name contains another instance of 'tar'.

11 years agoutils/tests: add hwloc-compress-dir
Brice Goglin [Tue, 24 Sep 2013 14:16:24 +0000 (14:16 +0000)]
utils/tests: add hwloc-compress-dir

This commit was SVN r5816.

11 years agoAdd hwloc-compress-dir utility
Brice Goglin [Tue, 24 Sep 2013 14:15:45 +0000 (14:15 +0000)]
Add hwloc-compress-dir utility

It takes a directory of XML topologies and compresses it by using
hwloc-diff to store as many topologies as diffs as possible.

This commit was SVN r5815.

11 years agoutils/manpages: fix the layout of the DESCRIPTI...
Brice Goglin [Tue, 24 Sep 2013 14:15:03 +0000 (14:15 +0000)]
utils/manpages: fix the layout of the DESCRIPTI...

utils/manpages: fix the layout of the DESCRIPTION section

This commit was SVN r5814.

11 years agohwloc-assembler-remote: Fix typo in usage and a...
Brice Goglin [Tue, 24 Sep 2013 14:14:35 +0000 (14:14 +0000)]
hwloc-assembler-remote: Fix typo in usage and a...

hwloc-assembler-remote: Fix typo in usage and abort-out on unknown option with a single dash

This commit was SVN r5813.

11 years agodoxy: sync the list of supported platforms with...
Brice Goglin [Wed, 18 Sep 2013 13:27:07 +0000 (13:27 +0000)]
doxy: sync the list of supported platforms with...

doxy: sync the list of supported platforms with the website

This commit was SVN r5812.

11 years agoBump HWLOC_API_VERSION to 0x00010800
Brice Goglin [Mon, 16 Sep 2013 20:15:08 +0000 (20:15 +0000)]
Bump HWLOC_API_VERSION to 0x00010800

We added at least  topology_dup() and the diff interface.

This commit was SVN r5811.

11 years agohwloc-calc.h: make location parsing more reliab...
Brice Goglin [Wed, 11 Sep 2013 07:46:49 +0000 (07:46 +0000)]
hwloc-calc.h: make location parsing more reliab...

hwloc-calc.h: make location parsing more reliable and add useful error messages

Use strtol() instead of atoi() when parsing type strings, depth, and ranges,
and error-out if some spurious characters are found.
Will avoid unexpected results when using wrong separators.

Thanks for Bertrand Putigny for reporting the problem by mistake.

This commit was SVN r5810.

11 years agocore: don't debug_check the topology consistenc...
Brice Goglin [Tue, 10 Sep 2013 20:50:47 +0000 (20:50 +0000)]
core: don't debug_check the topology consistenc...

core: don't debug_check the topology consistency too early during dup()

This commit was SVN r5809.

11 years agodiff: use hwloc_uint64_t instead of uint64_t
Brice Goglin [Tue, 10 Sep 2013 20:35:11 +0000 (20:35 +0000)]
diff: use hwloc_uint64_t instead of uint64_t

This commit was SVN r5808.