[cfe-commits] The AArch64 LLVM (& Clang) target

[Tim Northover]

Hi all,

I now finally have permission to start upstreaming the long-delayed backend
for
ARM's new 64-bit architecture, AArch64.

In summary, we're proposing this backend for inclusion into LLVM; first as an
experimental target, but with a view to being fully supported as soon as
possible.

Splitting up an entire target into smaller patches is rather tricky, but I've
isolated all changes to generic LLVM code. I'll be sending those patches out
for individual review and wouldn't consider committing until all that's left
in the backend patch is very definitely AArch64 specific. Let me know if
there's anything else I can do to help.

Meanwhile, the patch attached here *does* contain those generic changes, which
should probably be ignored for review purposes. They're present so that
someone curious can simply apply a single patch and get a working backend.

The current status of the backend is roughly as follows:
  + In a large scale: C99 and C++03 code targeting ELF on Linux should be
    supported. Problems there will be a high priority and are probably unknown
    to us.
  + Initial support for using NEON instructions is underway, being implemented
    by Ana Pazos from Qualcomm (thanks!). This is not complete though and
    should not be expected to work for generic code yet.
  + The backend produces mostly unoptimised code. Work has been mainly on
    features, with the lack of hardware making more detailed performance
    analysis less useful at present.
  + DWARF debugging should be supported.
  + Position-independent code should be supported.
  + Thread-local storage (GNU, rather than C++11) should be supported.
  + Inline assembly should be supported.
  + There is no MCJIT (or any other JIT) support.
  + Only one memory model is supported: code and statically allocated data
    must occupy a contiguous 4GB space (a.k.a. "large").

There are some areas known to need refactoring, but where we refrained to
avoid large merge conflicts. I would prefer the nastiest of these to be
delayed and form part of the Experimental -> Stable target transition, where
merges will hopefully be less painful. Most notably the ConstantIslandPass
should probably be shared between AArch32 and AArch64.

If there's anything we can do to make this process easier for people, please
let me know (a branch in svn? Upload patch to phabricator? ...). Similarly,
any other questions or comments would be welcome.

Cheers.

Tim.

-- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium.  Thank you.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: aarch64-target-clang.diff
Type: text/x-patch
Size: 238029 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130107/8112e8ef/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: aarch64-target-llvm.diff
Type: text/x-patch
Size: 2395268 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130107/8112e8ef/attachment-0001.bin>