These are chat archives for elemental/chat

2nd
Dec 2016
Jack Poulson
@poulson
Dec 02 2016 00:44
@rhl- A RedHat dev is now criticizing the bug reporting :-(
Ryan H. Lewis
@rhl-
Dec 02 2016 00:45
@poulson im aware, it’s ok. I asked him to comment. I’ve talked with him before on IRC
it’s a good thing it means if we do what he wants he’ll probably look at it.
Ryan H. Lewis
@rhl-
Dec 02 2016 00:51
my PPC environment is down for maintence
but when it returns I plan on providing the info he needs
@poulson: I want to merge the linux conda recipe. Is that OK with you?
Ryan H. Lewis
@rhl-
Dec 02 2016 01:25
@poulson I added:
# Advanced options
# ----------------
if (GCC_REPORT)
 add_definitions("-v -save-temps=obj")
endif()
and it seems to be generating all the data I need neatly inside the build dir to add to the report
I think find build -name '*.i*' | xargs tar czf preprocessed_source.tgz will generate the report tarball
so we can add that as a cmake command
gotta run
Jack Poulson
@poulson
Dec 02 2016 04:23
@rhl- Is it expected that the appveyor fails?
Also, is there a way to avoid 100 one-line commits?
Ryan H. Lewis
@rhl-
Dec 02 2016 04:23
Ya
Jack Poulson
@poulson
Dec 02 2016 04:23
are you having to run build experiments through tiny tweaks?
also, I would recommend doing not having flags like GCC_REPORT and instead having something like COMPILER_REPORT and have branches for GNU
that way it is easy to extend to other compilers, such as Clang
Jack Poulson
@poulson
Dec 02 2016 04:39
I'm currently compiling with -fsanitize=undefined,address
Ryan H. Lewis
@rhl-
Dec 02 2016 04:46
I can turn off the appveyor builds if we can't fix it
Ryan H. Lewis
@rhl-
Dec 02 2016 05:12
@poulson: yes and ok
Ryan H. Lewis
@rhl-
Dec 02 2016 07:19
@poulson back a PC now. I am collecting up the PPC failure data to submit.
Yes the appveyor builds are failing. I can try disabling PMRRR, but, I think there is something also wrong with building METIS
Ryan H. Lewis
@rhl-
Dec 02 2016 07:32
if (COMPILER_REPORT)
 if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
    add_definitions("-v -save-temps=obj")
 elseif(CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
    add_definitions(“--save-temps=obj")
 endif()
 add_custom_command(TARGET collect_report POST_BUILD 
                    COMMAND find . -name "*.i*"
                    COMMAND xargs tar czf preprocessed_source.tgz)
endif()
i think this does it.
Jack Poulson
@poulson
Dec 02 2016 16:04
yes, METIS DLL issues are what I was referring to earlier
I spent some time working towards hacking METIS to fix it but didn't finish due to MSVC bugs (which I assume are fixed now)
it seems there might be some bugs in using ubsan and asan with Apple's Clang. I ran into this exact issue: http://stackoverflow.com/questions/40284307/undefined-symbols-when-using-clangs-undefined-sanitizer
also, that COMPILER_REPORT change LGTM
have to recompile clang
Ryan H. Lewis
@rhl-
Dec 02 2016 16:43
Ok well I broken my branch build trying to disable pmrrr
I think I'll turn appveyor off
And just check in the half baked yml file
So if you or I want to pick it up again later we can
Jack Poulson
@poulson
Dec 02 2016 16:46
disabling PMRRR will require some code changes
but the D&C is actually the default eigensolver since it works for all precisions
Ryan H. Lewis
@rhl-
Dec 02 2016 18:16
Ok, i’ll create a ticket for allowing PMRRR to be disabled
@poulson: https://copr-be.cloud.fedoraproject.org/results/rhl/elemental/fedora-23-i386/00483114-elemental/build.log.gz <— it looks like ComplexBKZ has started failing on a i386 slave