[PATCH] Disable passes on optnone functions

[Philip Reames]

Code wise, this seems fine.  If this style was already accepted for IR 
passes, it should work for backend passes as well.

w.r.t. the reg allocator passes, leaving it as is seems fine. Choosing a 
different register allocator is less about possible semantic changes 
(optimization) and more about running time vs code quality.  I wouldn't 
expect to find too many examples of code which are sensitive to the 
register allocator used.  (One exception: a buggy garbage collector 
implementation might be effected by a more or precise liveness 
analysis)  Probably not worth worrying about for the moment.

p.s. I don't have commit rights, so someone else will need to commit this.

Philip

On 03/27/2014 12:26 PM, Robinson, Paul wrote:
> +Philip Reames who offered to do the review.
> Restating the original email message:
>
> This patch disables a number of Machine Function Passes for functions
> with the 'optnone' attribute.
>
> To come up with the list of passes, I looked at CodeGen/Passes.cpp,
> and disabled whatever would be excluded under CodeGenOpt::None.
> Except, I did not disable anything listed in addOptimizedRegAlloc(),
> as I can't see how to substitute one RA pass sequence for another on
> a per-function basis.  Maybe there's a way to deal with that safely,
> but for now it seemed prudent (and not horribly harmful to the goals
> of 'optnone') to leave well enough alone.
> If anybody has an idea, we're open to it.
> --paulr
>
>> -----Original Message-----
>> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-
>> bounces at cs.uiuc.edu] On Behalf Of Robinson, Paul
>> Sent: Tuesday, March 25, 2014 9:53 AM
>> To: llvm-commits at cs.uiuc.edu
>> Cc: Chandler Carruth (chandlerc at gmail.com)
>> Subject: RE: [PATCH] Disable passes on optnone functions
>>
>> Ping^3.  +chandlerc on the theory that nobody else is
>> willing to step on the toes of the Pass Manager Meister,
>> but I'll take review from anybody willing to do it.
>>
>> This is the last of the LLVM pieces of the 'optnone' work.
>> It would be really really nice to put this to bed.
>> --paulr
>>
>>
>>> -----Original Message-----
>>> From: Robinson, Paul
>>> Sent: Monday, March 17, 2014 10:35 AM
>>> To: Robinson, Paul; llvm-commits at cs.uiuc.edu
>>> Subject: RE: [PATCH] Disable passes on optnone functions
>>>
>>> Ping^2.
>>> --paulr
>>>
>>>> -----Original Message-----
>>>> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-
>>>> bounces at cs.uiuc.edu] On Behalf Of Robinson, Paul
>>>> Sent: Friday, March 07, 2014 9:47 AM
>>>> To: llvm-commits at cs.uiuc.edu
>>>> Subject: RE: [PATCH] Disable passes on optnone functions
>>>>
>>>> Ping. Somebody willing to own MachineFunctionPass behavior?
>>>> --paulr
>>>>
>>>>> -----Original Message-----
>>>>> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-
>>>>> bounces at cs.uiuc.edu] On Behalf Of Robinson, Paul
>>>>> Sent: Monday, March 03, 2014 6:25 PM
>>>>> To: Nick Lewycky
>>>>> Cc: llvm-commits at cs.uiuc.edu
>>>>> Subject: RE: [PATCH] Disable passes on optnone functions
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Nick Lewycky [mailto:nicholas at mxc.ca]
>>>>>> Sent: Saturday, March 01, 2014 1:17 AM
>>>>>> To: Robinson, Paul
>>>>>> Cc: llvm-commits at cs.uiuc.edu
>>>>>> Subject: Re: [PATCH] Disable passes on optnone functions
>>>>>>
>>>>>> Robinson, Paul wrote:
>>>>>>> This patch disables a number of Machine Function Passes for
>>>>> functions
>>>>>>> with the 'optnone' attribute.
>>>>>>>
>>>>>>> To come up with the list of passes, I looked at
>>>> CodeGen/Passes.cpp,
>>>>>>> and disabled whatever would be excluded under
>> CodeGenOpt::None.
>>>>>>> Except, I did not disable anything listed in
>>>> addOptimizedRegAlloc(),
>>>>>>> as I can't see how to substitute one RA pass sequence for
>>> another
>>>> on
>>>>>>> a per-function basis.  Maybe there's a way to deal with that
>>>> safely,
>>>>>>> but for now it seemed prudent (and not horribly harmful to the
>>>> goals
>>>>>>> of 'optnone') to leave well enough alone.
>>>>>>> If anybody has an idea, we're open to it.
>>>>>> I note that I'm on the 'to' line, but I am certainly not the
>> right
>>>>>> person to review this. I know some high-level design of the
>>> codegen
>>>>>> parts, but that's about it.
>>>>> Ah, sorry.  You were there for me on the
>>>> {Function|BasicBlock|Loop}Pass
>>>>> part, I shouldn't have presumed about MachineFunctionPass.
>>>>>
>>>>>> Would someone else please take a look?
>>>>>>
>>>>>> Nick
>>>>> --paulr
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> llvm-commits mailing list
>>>>> llvm-commits at cs.uiuc.edu
>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits