[PATCH] GlobalOpt enhancement optimization (hoisting of initialization code into constant global initializers).

[Nick Lewycky]

On 26 August 2013 17:39, Reid Kleckner <rnk at google.com> wrote:

> See EvaluateStaticConstructor in the same file, GlobalOpt.cpp.
>
> I really haven't dug into your code because it's quite large, but it seems
> to me that you should be able to evaluate the code for your test case with
> the evaluator.  If not, you should debug why and consider extending it.
>

I know this one: because it's not a static constructor.

EvaluateStaticConstructor tries to evaluate the dynamic initializers and
bails if it can't handle anything it encounters. This patch is about doing
guarded initialization, which can occur anywhere in user-written code after
main has begun, and doesn't require the ability to evaluate the entire
function.

For a concrete example,

  bool isInit = false;
  int i;
  void foo() {
    printf("Boo!\n");
    if (isInit) {
      isInit = true;
      i = 1;
    }
  }

EvaluateStaticConstructor will bail immediately at the printf because it
can't simulate that. Of course, this makes sense for removing dynamic
initializers, but not when removing guarded initialization.

I've previously taught the evaluator to carry on when it sees a use of a
previously uncomputed value (ie., getVal no longer asserts, and can instead
fail). You can grab my implementation from
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20120206/136735.html.
As you can see, I ultimately didn't commit it.

I still want to know whether this pattern is something that actually
happens. It doesn't match the guarded initialization we get with
function-local statics in the Itanium C++ ABI.

Nick

On Mon, Aug 26, 2013 at 5:07 PM, Puyan Lotfi <plotfi at apple.com> wrote:
>
>>
>> Reid, I am not entirely clear on which Evaluator class you are referring
>> to. Could you point me to some more information on this?
>> I wasn’t really trying to special case things too heavily, was more
>> trying to match up modification of a boolean flag with stores to a global
>> to tell if things could just be hoisted.
>>
>> Thanks
>>
>> -Puyan
>>
>>
>> On Aug 26, 2013, at 11:47 AM, Reid Kleckner <rnk at google.com> wrote:
>>
>> I'm not really an owner of GlobalOpt, but this seems like a big ugly huge
>> special case, rather than something general that's part of the Evaluator
>> class.  You should be able to emulate stores to isInit in the usual way,
>> since it's internal.  Can you debug why that isn't firing?
>>
>> I think it would be much more interesting to try to look at Itanium C++
>> dynamic initializers and try to fold those away.  This would imply
>> understanding what __cxx_acquire / release mean when -fno-builtin /
>> -ffreestanding aren't set.
>>
>>
>> On Mon, Aug 26, 2013 at 1:40 AM, Puyan Lotfi <plotfi at apple.com> wrote:
>>
>>> All:
>>>
>>> Attached is a patch to GlobalOpt that adds an optimization that takes
>>> user written constant initialization code for global variables and hoists
>>> the initialization values into the global initializer. This optimization is
>>> only done on locally linked globals of integral types (scalars, arrays, and
>>> structs) that are constant initialized (and zero initialized prior to
>>> transformation).
>>>
>>> The idea is to transform the following code:
>>>
>>> A = internal global i32 0, align 4
>>> isInit = internal global i1 false
>>>
>>> define i32* foobar() {
>>>   %.b = load i1* isInit
>>>   %1 = zext i1 %.b to i8
>>>   %2 = trunc i8 %1 to i1
>>>   br i1 %2, BB4, label BB3
>>> BB3:
>>>   store i32 113, i32* A, align 4
>>>   store i1 true, i1* isInit
>>>   br label %4
>>> BB4:
>>>   ret i32* A
>>> }
>>>
>>> Into:
>>>
>>> A = internal global i32 113, align 4
>>>
>>> define i32* @_Z8initTestv() {
>>>   ret i32* A
>>> }
>>>
>>> Could someone on the list review my changes, provide feedback, and if
>>> possible submit my changes?
>>>
>>> I also have some test cases I've written but I am still trying to figure
>>> out how to add them to llvm/test/Transforms/GlobalOpt (I don't see a
>>> lit.local.cfg in that directory as the docs specify).
>>>
>>> Thanks
>>>
>>> -Puyan
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130826/a9c80248/attachment.html>