[LLVMdev] About commit TILE-Gx backend to community repository and default disabled

[Chris Lattner]

On Mar 19, 2013, at 8:58 PM, Jiong Wang <jiwang at tilera.com> wrote:

> Hi Chris,
> 
> could you please comment on committing TILE-Gx backend into community?

Hi Jiong,

I don't have any special advice here.  It sounds like the general functionality level is high enough.  Taking it into mainline sounds great, so long as it is reviewed by someone. 

-Chris


> 
> 
> ========== TILE-Gx Status ===========
> 
> Features Supported
> ===
> 1. general function.
> 2. PIC/TLS/JumpTable.
> 3. Instructoin Bundling for VLIW.
> 4. Asm Parser
> 5. MC Layer (aware of instruction bundle), MCJIT support.
> 6. Initial regression tests for CodeGen & MC Layer.
> 
> Regression Result
> ===
> Expected Passes : 13363
> Expected Failures : 84
> Unsupported Tests : 76
> Unexpected Failures: 21
> 
> 20 of the unexpected failures are caused by lack of old jit support.
> 
> Test-Suite Result
> ===
> Expected Passes : 953
> Unexpected Failures: 14
> (all 14 failures has the same output as tilegx gcc,
> most of them are about float precision issue)
> 
> 
> 
> 于 2013/3/16 2:37, Tom Stellard 写道:
>> On Fri, Mar 15, 2013 at 05:38:28PM +0800, Jiong Wang wrote:
>>> Hi Chandler,
>>> 
>>> on 2013/3/15 17:15, Chandler Carruth wrote:
>>>> On Fri, Mar 15, 2013 at 1:54 AM, Jiong Wang <jiwang at tilera.com
>>>> <mailto:jiwang at tilera.com>> wrote:
>>>> 
>>>>    I agree that everyone should contribute to keep the community
>>>>    active and vigorious. But I think there are difference between
>>>>    contributors.
>>>> 
>>>>       some are focused and with expertise on middle end, while others
>>>>    may on front or back end. so the normal way for a new contributor
>>>>    is, find a familiar point to start, then go deeper and wider and
>>>>    expand one's contribution during this process.
>>>> 
>>>>       For Tilera corporation and me, we would like the keep tilegx
>>>>    backend actively maintained and improved, and wish it could be a
>>>>    good target to improve and test VLIW, many core features etc.
>>>> 
>>>> 
>>>> Inevitably, if you care about LLVM's support for your platform,
>>>> you will care about bugs and features in the target-independent
>>>> stack of LLVM's software (including the target independent parts
>>>> of Clang, etc). There will be bugs which are priorities only for
>>>> you and/or your users, features most pressing only for you or your
>>>> users, etc. As such, it seems likely that you will need at least a
>>>> basic working familiarity with the stack, and the ability to
>>>> improve it in at least basic ways. Simple things like finding and
>>>> fixing bugs in the target independent code generator, improving
>>>> DAG combine optimizations, or looking for compile time hotspots
>>>> are almost always approachable by nearly anyone on the project.
>>>     I am not quite understand your points here, for a backend
>>> maintainer, of course, these things are daily work.
>>> 
>>>> Essentially, you don't need to invent a new middle end
>>>> optimization, or improve the subtleties of our alias analysis.
>>>> Most of LLVM, and most of our bugs/missing features/areas for
>>>> improvement, involve straightforward C++ code and systems that
>>>> anyone who cares can go in, hack, and improve. That's what
>>>> occupies the majority of the maintenance work as it happens. There
>>>> is relatively little deeply detailed work requiring specialized
>>>> knowledge.
>>>     Anyway, my question is "how could tilegx backend included in
>>> community repository?", I am here to seek answer for this. I
>>> searched mailinglist archive, and learned R600 and AArch64's request
>>> post, but find few things.
>>> 
>> Hi Jiong,
>> 
>> I am the maintainer of the R600 backend so I have some experience trying to
>> get a new backend included in the community repository.
>> 
>> My recommendation is to contact the project lead, Chris Lattner,
>> directly and point him to the reviews that have already been done and ask
>> him specifically what you need to do to get the backend approved.
>> 
>> There is no formal process in place for adding a new backend, and I
>> think if you keep posing your question to the list you will have trouble
>> getting a clear answer on what needs to be done
>> 
>> -Tom
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev