[LLVMdev] LLVM Weekly - #32, Aug 11th 2014

[Alex Bradbury]

LLVM Weekly - #32, Aug 11th 2014
================================

If you prefer, you can read a HTML version of this email at
<http://llvmweekly.org/issue/32>.

Welcome to the thirty-second issue of LLVM Weekly, a weekly newsletter
(published every Monday) covering developments in LLVM, Clang, and related
projects. LLVM Weekly is brought to you by [Alex
Bradbury](http://asbradbury.org).Subscribe to future issues at
<http://llvmweekly.org> and pass it on to anyone else you think may be
interested. Please send any tips or feedback to <asb at asbradbury.org>, or
@llvmweekly or @asbradbury on Twitter.

Some readers may be interested to know that [lowRISC](http://lowrisc.org/), a
project to produce a fully open-source SoC started by a number of us at the
University of Cambridge Computer Lab has been announced. [We are
hiring](http://www.jobs.cam.ac.uk/job/4665/).

## News and articles from around the web

Codeplay contributed the LLDB MI (Machine Interface) frontend a while ago, and
have now committed some additional features. To coincide with that, they've
published a [series of blog
posts](http://lists.cs.uiuc.edu/pipermail/lldb-dev/2014-August/004768.html)
covering the MI driver's implementation, how to set it up from within Eclipse,
and how to add support for new MI commands.

McSema, a framework for transforming x86 programs to LLVM bitcode has [now
been
open-sourced](http://blog.trailofbits.com/2014/08/07/mcsema-is-officially-open-source/).
The talk about McSema from the ReCON conference is also now online.

Registration for the LLVM Developer's Meeting 2014 is [now
open](http://llvm.org/devmtg/2014-10/#registration). The event will take place
in San Jose on October 28th-29th. You have until September 1st to [submit your
talk/BoF/poster/tutorial
proposal](http://article.gmane.org/gmane.comp.compilers.llvm.devel/75567).


## On the mailing lists

* Robin Morisset, currently an intern at Google has [written about his plan to
optimize C11/C++11 atomics in
LLVM](http://article.gmane.org/gmane.comp.compilers.llvm.devel/75534). This
resulted in a discussion on [whether some of these transformation should be
done at the IR
level](http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-August/075573.html).

* Dan Liew has written a long post about [re-organising and improving
zorg](http://article.gmane.org/gmane.comp.compilers.llvm.devel/75575) (LLVM's
testing and buildbot infrastructure).

* Eric Christopher has written to the mailing list to warn us of [incoming API
changes](http://article.gmane.org/gmane.comp.compilers.llvm.devel/75475).
These changes include modifying getSubtarget/getSubtargetImpl to take a
Function/MachineFunction, so sub-targets could be used based on attributes on
the function.

* Sergey Ostanevich from Intel has
[shared](http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-August/075606.html) a
[proposal for the implementation of OpenMP offloading in
LLVM](http://lists.cs.uiuc.edu/pipermail/llvmdev/attachments/20140809/cd6c7f7a/attachment-0001.pdf).
This proposal was created by contributors from IBM, Intel, ANL, TI, and AMD.
The example given would allow code to be offloaded to e.g. a Xeon Phi
co-processor if present, or a GPU.

* Delesley Hutchins has [shared an update on his work on thread safety
analysis](http://article.gmane.org/gmane.comp.compilers.clang.devel/38248).
This details the recently added negative capabilities patch. He is looking for
feedback on how to limit the propagation of negative capabilities.


## LLVM commits

* Initial work on the MachineCombiner pass landed. This estimates critical
path length of the original instruction sequence vs a transformed (combined)
instruction sequence and chooses the faster code. An example given in the
commit message is choosing between add+mul vs madd on AArch64, and a followup
commit implements MachineCombiner for this target.
[r214666](http://reviews.llvm.org/rL214666),
[r214669](http://reviews.llvm.org/rL214669).

* A few useful helper functions were added to the LLVM C API:
`LLVM{IsConstantString, GetAsString, GetElementAsConstant}`.
[r214976](http://reviews.llvm.org/rL214676).

* A whole load of AVX512 instructions were added.
[r214719](http://reviews.llvm.org/rL214719).

* FastISel for AArch64 now support basic argument lowering.
[r214846](http://reviews.llvm.org/rL214846).

* A flag has been added to experiment with running the loop vectorizer before
the SLP vectorizer. According to the commit message, eventually this should be
the default.
[r214963](http://reviews.llvm.org/rL214963).

* The old JIT is almost dead, it has been removed (for those not paying close
attention, 3.5 has already been branched so still contains the old JIT).
However the patch was then reverted, so it's in zombie status.
[r215111](http://reviews.llvm.org/rL215111).

* AArch64 gained a load balancing pass for the Cortex-A57, which tries to make
maximum use of available resources by balancing use of even and odd FP
registers. [r215199](http://reviews.llvm.org/rL215199).


## Clang commits

* Thread safety analysis gained support for negative requirements to be
specified. [r214725](http://reviews.llvm.org/rL214725).

* Coverage mapping generation has been committed. The `-fcoverage-mapping`
command line option can be used to generate coverage mapping information,
which can then be combined with execution counts from instrumentation-based
profiling to perform code coverage analysis.
[r214752](http://reviews.llvm.org/rL214752).

* A command line option to limit the alignment that the compiler can assume
for an arbitrary pointer. [r214911](http://reviews.llvm.org/rL214911).


## Other project commits

* LLDB's FileSpec class learned to understand Windows paths.
[r215123](http://reviews.llvm.org/rL215123).

* LLDB learned a whole bunch of new commands and features for its Machine
Interface. [r215223](http://reviews.llvm.org/rL215223).

* OpenMP gained PowerPC64 support.
[r215093](http://reviews.llvm.org/rL215093).