[LLVMdev] 3.4.1 Release Plans

Jiangning Liu liujiangning1 at gmail.com
Wed Apr 2 03:09:53 PDT 2014


Hi Tom,


> In principle, I don't really have a problem with this kind of change going
> into the stable branch since it only modifies two backends
> and not common code, but I would like to know how these changes affect
> the
> 
> non-neon parts of the AArch64 and ARM backends?
>

For the changes of 

non-neon parts of the AArch64 and ARM backends, I listed them as below
previously.

* Change 64-bit integer type int64_t mapping from "long long" to "long" and
it potentially affects binary compatibility.
* va_copy run-time behavior failure for AArch64.
* Fix a silent codegen fault for atomic operations (e.g. __sync_...
Intrinsics).
* Fix an assertion failure in A15 SDOptimizer about DPair reg class by
treating DPair as QPR.
* Fix ARM back-end ld/st for v1i64 vector list failure around writeback
mode.


> Also, in order to be accepted, these changes will need to be approved by
> the code owners for both ARM and AARCH64 and we will need to verify that
> these patches don't change the ABI.
>

The owners of those patches are 

 * adibiagio :r198084
 * apazos : r197928 r197929 r198437  r198791  r199242  r201112
 * chandlerc: r199631
 * haoliu: r196748 r196749 r197135  r197361  r197897  r197969  r198001
r198188  r198192  r198193  r198682  r198684  r198730  r199461  r199463
r199861  r201061  r201287  201298  r201381  r204304  r196750  r197898
 * jiangning : r197993  r198194  r199296  r199369  r200113  r200119
r201385  r201541  r197994  r198195  r198741  r200114  r201384
 * kbeyls: r198937 r198941
 * kevinqin: r196887r196889 r196998  r197113  r197551  r197966  r197967
r198190  r198675  r198743  r199069  r199070  r199462  r199485  r199621
r199628  r199791  199858  r199978  r200109  r200110  r200111  r200179
r200180  r200365  r201395  r201841  r202775  r204424  r196888  r197112
r199866  r202004
 * mcrosier: r196831r196832 r196833 r196926  r196930  r196962  r196963
r196964  r196965  r197066  r197067  r197068  r197090  r197159  r197250
r197402  r200491  201091  r201793  r196834  r196835  r196836  r196927
r196931  r196936  r196966  r196967  r196968  r196969  r197069  r197070
r197071  r197091  r197403
 * tnorthover:  r200706  r200768  r200470  r200471  r200472  r200524
r200525  r200526  r200527  r200528  r200707  r200708  r200769

and I added them in CC list.

For ABI changes, there are two things I think,
* Some of the patches from Tim merged LLVM IR intrinsic from
llvm.aarch64.xxx to llvm.arm.xxx. e.g. 200768.
* For AArch64, int64_t is changed from "long long" to "long". C++ name
mangling could expose the issue.
i.e. 
202004
.


> I've been using:
> http://ispras.linuxbase.org/index.php/ABI_compliance_checker
> to verify ABI consistency when merging patches.  Could you test it out
> on your branch?  I think changes like r
> 
> 200768 which add or remove
> intrinsics in llvm/include/IR will end up breaking the ABI.
>

I tried this tool, but I can't make it work yet. The tool requires to use
GCC, but the tool can't use gcc 4.8, because gcc has a bug, but building
llvm requires to use gcc 4.8 or above, so this is a conflict I have to
solve before making it work. So how do you handle this issue when using
this tool?

Also, could you please share with me the xml files and command lines you
are using for this tool?

Thanks,
-Jiangning
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140402/5317de95/attachment.html>


More information about the cfe-commits mailing list