[llvm-dev] Regarding Dependence distance dump

[Venkataramanan Kumar via llvm-dev]

Hi Sebastian,

Given that "delinerization" fails shall I file a PR in bugzilla?

regards,
Venkat,

On Thu, 20 Sep 2018 at 09:57, Venkataramanan Kumar <
venkataramanan.kumar.llvm at gmail.com> wrote:

> Hi Sebastian ,
>
> On Thu, 20 Sep 2018 at 03:57, Sebastian Pop <sebpop at gmail.com> wrote:
>
>>
>>
>> On Wed, Sep 19, 2018 at 4:58 AM Venkataramanan Kumar <
>> venkataramanan.kumar.llvm at gmail.com> wrote:
>>
>>> Hi,
>>>
>>> I tired to see when this behavior changed in LLVM.
>>> It seems to start from.
>>> --snip--
>>> commit 95e5d37d5868ebde2302bc302c1e0af407c5646d
>>> Author: Sebastian Pop <sebpop at gmail.com>
>>> Date:   Tue Mar 6 21:55:59 2018 +0000
>>>
>>>     DA: remove uses of GEP, only ask SCEV
>>> --snip--
>>>
>>> https://reviews.llvm.org/D35430
>>> Review URL says "GEP" based approach was wrong for multidimensional
>>> array references. The LLVM-DA now uses SCEV representation.  But why we are
>>> not getting the distance with SCEV?  Current DA will not work for
>>> multidimensional arrays?
>>>
>>
>> It should work with delinearization of multi-dim arrays.
>>
> I thought "delinearization"  is enabled by default on trunk now.
>
>>
>>
>>>
>>> I need to get dependence distance for a transformation that I am working
>>> on.  How do I get distance for such cases, esp multidimensional arrays?
>>>
>>> Thanks and regards,
>>> Venkat.
>>>
>>> On Tue, 18 Sep 2018 at 19:00, Venkataramanan Kumar <
>>> venkataramanan.kumar.llvm at gmail.com> wrote:
>>>
>>>> Hi,
>>>>
>>>> For the below test case, I need the dependence distance between the
>>>> array references  "a[j][i]"  and "a[j-1][i-2]".
>>>>
>>>> --Snip--
>>>> int a[5][5];
>>>> int b[10];
>>>>
>>>> void test()
>>>> {
>>>>         for (int j=1;j<5;j++)
>>>>                 for (int i=2;i<5;i++)
>>>>                 {
>>>>                         a[j][i] = 3.0;
>>>>                         b[i]= a[j-1][i-2] ;
>>>>                 }
>>>> }
>>>> --Snip-
>>>>
>>>> Compile steps
>>>> clang -O1 -emit-llvm -S test.c
>>>> opt  -analyze -da dist.ll
>>>>
>>>>
>> you are missing the -delinearize flag to opt.
>>
> Tried this flag and I am getting
> opt -analyze -delinearize -da  dist.ll
> --Snip--
> Delinearization on function test:
> Inst:  %arrayidx6 = getelementptr inbounds [5 x [5 x i32]], [5 x [5 x
> i32]]* @a, i64 0, i64 %indvars.iv30, i64 %indvars.iv
> In Loop with Header: for.body4
> AccessFunction: 0
> failed to delinearize
>
> Inst:  %arrayidx6 = getelementptr inbounds [5 x [5 x i32]], [5 x [5 x
> i32]]* @a, i64 0, i64 %indvars.iv30, i64 %indvars.iv
> In Loop with Header: for.cond1.preheader
> AccessFunction: 0
> failed to delinearize
> .....
> --Snip--
>
> regards.
> Venkat.
>
>>
>>
>>> I tried dumping the dependence information with LLVM trunk (2018_09_17),
>>>> it shows:
>>>>
>>>> Printing analysis 'Dependence Analysis' for function 'test':
>>>> da analyze - none!
>>>> da analyze - flow [< <>]! <== dumps direction vector
>>>> da analyze - none!
>>>> da analyze - none!
>>>> da analyze - none!
>>>> da analyze - consistent output [S 0]!
>>>>
>>>> LLVM Release 6.0.1 dumps the expected distance.
>>>> Printing analysis 'Dependence Analysis' for function 'test':
>>>> da analyze - none!
>>>> da analyze - consistent flow [1 2]!
>>>> da analyze - none!
>>>> da analyze - none!
>>>> da analyze - none!
>>>> da analyze - consistent output [S 0]!
>>>>
>>>> In trunk, it is now showing directions instead of distances.
>>>> How to get the dependence distance?
>>>>
>>>> regards,
>>>> Venkat.
>>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20180921/635e1ee8/attachment.html>