wrote:
On 08/31/2016 12:07 PM, Mark W. Krentel wrote:
Ok, I've been sufficiently taunted. Speaking as someone who builds a
subset of Dyninst (parseapi + symtabapi) from source and applications
that use it (Rice HPCToolkit) on a daily basis
Thanks for your input!
I'm a bit unsure of just what
Ok, I've been sufficiently taunted. Speaking as someone who builds a
subset of Dyninst (parseapi + symtabapi) from source and applications
that use it (Rice HPCToolkit) on a daily basis
I'm a bit unsure of just what you're suggesting.
First, I think it's perfectly ok to require Dyninst
antics are a work-in-progress: I'm continuing to work on
adding support for more instructions and fixing whatever bugs I find. Do
let me know about other asserts and errors you see!
Thanks,
Sunny
On 9/12/2016 5:59 PM, Mark W. Krentel wrote:
I think instructionAPI (or maybe parseAPI) is broken in m
I think instructionAPI (or maybe parseAPI) is broken in master branch
on ARM (aarch64). I'm seeing two problems. The small one is a spew
of error messages, "Instruction processing threw exception for
instruction: ldr X17, [X16 + 3c0]". The bigger one is raising an
abort.
I'm running this on
I'm seeing a very bad bug in recent dyninst master. "Bad" means that
I get a segfault in Symtab::openFile().
For me, all I need in a test program is basically:
Symtab * the_symtab = NULL;
Symtab::openFile(the_symtab, argv[1]);
I get a segfault and stack trace like this:
#0
Bill,
Did you indicate that the fix for using the old libelf was merged into
master? I'm still seeing the same problem.
I just tried this version from master.
commit 64edc81ee01131502e2568749f006e3511c44fb6
Merge: d747211 3c12b19
Author: Bill Williams
Date: Tue Nov 15
Two more data points:
1. it works for me on power8/le (po) with elfutils-0.167, building and
running the same way that it failed on power7.
2. it fails for me on power7 with elfutils-0.158.
So, it looks like a bug specific to big-endian powerpc and not related
to elfutils version. And it
I'm seeing a build failure in IA_power.C in parseAPI. This happens on
both big-endian power7 and little-endian power8.
/home/krentel/release/BUILD/symtabAPI/dyninst/parseAPI/src/IA_power.C:
In member function 'virtual bool
Dyninst::InsnAdapter::IA_IAPI::sliceReturn(Dyninst::ParseAPI::Block*,
Ok, that seems to fix the problem. At least, it doesn't throw an
assert running on complicated binaries (eg, hpcstruct analyzing
itself).
But I'm struggling to understand your fix. Maybe I'm being dense or
missig something obvious. It seems to me that your patch simply
changes the body of
As Matt points out, Blue Gene doesn't get any updates for front-end
compilers. They continue to sit at RH 6.x and gcc/g++ 4.4.
The local admins could install later FE compilers, but they don't.
Red Hat could do better by providing packages for later compilers.
But that's part of the point.
I ran into this problem while trying to build Dyninst 9.3.2 on Blue
Gene (mira, vesta) at Argonne.
Dyninst 9.3.x won't build with GNU 4.4 because the typeCollection
destructor in Collections.C uses the C++11 range iterator feature
which is not supported in 4.4.
318
As we've previously discussed, I've also had problems with cotire.
Something over parallel builds when building both static and dynamic
libraries.
So yes, I'd either like to turn off cotire or else provide an option
to turn it off. Master has USE_COTIRE, but it's not exposed, so
currently I
Sasha,
I've tested this and I now get the same answers for file paths with
master and libdw as with the old 9.3.2 and libdwarf.
So, thanks!
--Mark
On 03/27/18 16:28, Sasha Da Rocha Pinheiro wrote:
Hi Mark,
I pushed a commit that fixed it, in the way that you suggested at:
(1) Declare
Sasha,
I see where you added a method to Module for getCompDir().
Thanks, but the implementation is badly broken. This has to be a
per-module answer. But the code for getCompDir() has a single static
string comp_dir_str. As a result, every Module returns the same comp
dir.
Instead, I think
Ok, with this fix to master, now I'm getting sensible results for
getCompDir() on a per-module basis.
Btw, although we don't use this right now, I can tell that a couple
months down the road, this will become useful to better understand the
layout of the source files as we present them in
I see. I guess nobody told me. Sorry,
Thanks,
--Mark
On 04/23/18 16:33, Xiaozhu Meng wrote:
For parallel parsing branch, testsuite is now a submodule. So, please do
git submodule init
git submodule update
___
Dyninst-api mailing list
The new-parallel-parsing branch of dyninst has this in the top-level
CMakeLists.txt.
add_subdirectory(testsuite)
But the testsuite directory is empty in the n-p-p branch. As a
result, cmake falls over with:
-- Found dependency location /home/krentel/config/build-dyninst/dyninstAPI
CMake Error
Xiaozhu,
I was able to adapt your patch on the overlapping instructions to our
build system. (There was one tiny issue over ah.retreat() versus
ah->retreat(), but that was easy.)
Now, I don't see any more non-determinism in our hpcstruct answer on
x86. I've run it a bunch of times on the
You may also need to turn off cotire. In the past, we've seen
problems with cotire and parallel builds. Something about cotire
generating some header files twice. In a parallel build, this could
result in one process writing a file while another one is reading it
and crashing the build.
Bill,
Stan,
As Xiaozhu said, one of the major additions to 10.x will be the merge
of the new-parallel-parsing branch.
If you want to get a jump on things, you could try building and
running the new-parallel-parsing branch. Fairly recent master has
been merged into new-par-par, so it should be pretty
Well, spack won't concretize an old version unless it has
preferred=True or you specifically ask for it. So, TBB isn't at any
greater risk than any other package.
If you want to draw the line at 2018.06, then whatever, go ahead.
Be sure to put that in the release notes.
But if it were me, IMO
I know that dyninst will download and build TBB, elfutils, boost if
they're not specified. That's fine that you go the extra mile, but
that's not the spack way of doing things.
Spack wants to build all the prereqs itself, through the intel-tbb
package.py file. (So, it's important to allow -D
On 11/04/18 15:46, Benjamin Welton wrote:
Mark,
Packages.yaml is site/user specific (at least as per my understanding).
> Spack will take the latest version from intel-tbb by default.
I do not believe this is true. Spack will only do this if no version
of TBB was installed by spack already.
I've been working on building Dyninst 10.0 on Blue Gene with RedHat
6.x and old glibc-headers and I ran into the following problem.
The build dies in Object-elf.C with undefined symbol EM_AARCH64.
You've seen this problem before. Elf_X.C and dwarfHandle.C have
code to work around an old elf.h
I was about to warn you that the new config option won't be available
until the next release, but I just check the elfutils web site and
0.176 (with the option) was released on Friday, Feb 15.
[DIR]0.174/14-Sep-2018 11:44
[DIR]0.175/16-Nov-2018 12:14
[DIR]0.176/
Branch: refs/heads/master
Home: https://github.com/dyninst/dyninst
Commit: 54a2debd94c3e8df3b6f1e579f1db02dae61cbb7
https://github.com/dyninst/dyninst/commit/54a2debd94c3e8df3b6f1e579f1db02dae61cbb7
Author: Mark W. Krentel
Date: 2019-02-01 (Fri, 01 Feb 2019)
Changed paths
Branch: refs/heads/master
Home: https://github.com/dyninst/dyninst
Commit: 54a2debd94c3e8df3b6f1e579f1db02dae61cbb7
https://github.com/dyninst/dyninst/commit/54a2debd94c3e8df3b6f1e579f1db02dae61cbb7
Author: Mark W. Krentel
Date: 2019-02-01T13:14:11-06:00
Changed paths
This is mostly an early heads up, I'm not blocked or anything. But at
some point, you'll need to adapt to some new boost usage.
I'll leave it to you to decide if this is needed for 10.1. Before
10.1 would be a bit easier for spack users, so that when 10.1 comes
out, then the latest release of
Will that solution work for both 1.69 and 1.70?
That would be the most robust way of handling this, if there's a
solution that will work either way, or else a config test and #if.
You'd rather not have a hard line where Dyninst 10.0 requires <= 1.69
and Dyninst 10.1 requires >= 1.70.
Thanks!
One of my cron jobs that builds Dyninst master tripped an alarm today.
With the recent commits, the build procedure for master and upcoming
10.1 has changed in a few ways.
First question: are the changes to the cmake args (see below) now
complete, or are there more to come before 10.1?
I would
orced a code change before merging this, but
it had already been tested by the person who submitted the PR, and it
was blocking their build cycle. The correct solution is to use to use
std::to_string introduced in C++11. I'll make a PR for that and re-test.
Thanks.
- Tim
On Tue, Aug
Thanks.
- Tim
On Wed, Aug 7, 2019 at 2:03 PM Mark W. Krentel <mailto:kren...@rice.edu>> wrote:
Tim,
Based on the contents of the file, I don't see where this
would be limited to 32-bit.
I have seen this on more than just power8. With RH 6.x, it's "broken"
One of my cron jobs tripped an alarm for Dyninst master on powerpc
this morning.
This one is bizarre, but the last commit about the PRIx64 macros
breaks the build on some systems, depending on the exact version of
the header file.
Here is a unit test that reproduces the problem. This is a
Branch: refs/heads/master
Home: https://github.com/dyninst/dyninst
Commit: b98a53f455e92d577f6f294732ba0712d53c36dc
https://github.com/dyninst/dyninst/commit/b98a53f455e92d577f6f294732ba0712d53c36dc
Author: Mark W. Krentel
Date: 2020-06-01 (Mon, 01 Jun 2020)
Changed paths
Uh, exactly how are you building dyninst api_breakages?
I tried building this directly from cmake using spack-built prereqs.
I get this failure about unable to find libiberty include directory
[ 9%] Building CXX object
common/CMakeFiles/common.dir/src/addrtranslate-linux.C.o
[ 9%] Building
The failure doesn't depend on STERILE build.
So, exactly how are you building dyninst master?
And why are you not seeing build failures like this?
This is a cmake error right?
Not setting an include directory?
--Mark
On 10/27/20 13:16, Mark W. Krentel wrote:
Uh, exactly how are you building
Hi, what are y'all thoughts on building Dyninst with a TBB other than
intel-tbb, specifically intel-oneapi-tbb ?
Spack has three providers for the virtural tbb: intel-tbb,
intel-oneapi-tbb and the full intel-parallel-studio. I get the sense
that intel-tbb is older, to be phased out in favor of
hub.com/dyninst/dyninst/pull/1252. That will
> likely get merged today.
>
> Thanks.
> -Tim
>
> On Fri, Apr 8, 2022 at 12:23 PM Mark W. Krentel wrote:
> I'm seeing a repeatable build failure for dyninst master on powerpc.
> This only happens on ppc64/le (llnl.cs.rice.ed
I'm seeing a repeatable build failure for dyninst master on powerpc.
This only happens on ppc64/le (llnl.cs.rice.edu), not x86, not arm.
It just started happening today with a few new commits including the
new _DEFAULT_SOURCE lines.
It looks like something has triggered some arch-specific changes
39 matches
Mail list logo