Adding an option to sparse to disable the IR simplifications

Adding an option to sparse to disable the IR simplifications

[Dibyendu Majumdar]

Hi,

With the progress made with spars-llvm in recent weeks we are now
beginning to hit bugs in the simplifications carried out by the
linearizer. So it is useful I think to have an ability to turn off
these modifications - perhaps by a command line argument. This will
enable testing of sparse-llvm with and without the simplifications,
and help identify issues caused by the latter phase.

Thanks and Regards
Dibyendu

Re: Adding an option to sparse to disable the IR simplifications

[Dibyendu Majumdar]

On 22 March 2017 at 19:56, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
> With the progress made with spars-llvm in recent weeks we are now
> beginning to hit bugs in the simplifications carried out by the
> linearizer. So it is useful I think to have an ability to turn off
> these modifications - perhaps by a command line argument. This will
> enable testing of sparse-llvm with and without the simplifications,
> and help identify issues caused by the latter phase.
>

It appears that there is already an optimize flag that is set by -O
option. Perhaps this flag can be utilized to control the level of
simplifications attempted.

Thanks and Regards
Dibyendu

Fwd: Adding an option to sparse to disable the IR simplifications

[Dibyendu Majumdar]

Hi,

I suggested a while back that it might be useful to add an option to
control the simplifications.

In dmrC the simplification phase is turned off by default. But if you
supply -O1 option then it is turned on. This ensures that  I can test
both the unsimplified and simplified versions.


Regards
Dibyendu

---------- Forwarded message ----------
From: Dibyendu Majumdar <mobile@majumdar.org.uk>
Date: 22 March 2017 at 20:53
Subject: Re: Adding an option to sparse to disable the IR simplifications
To: Linux-Sparse <linux-sparse@vger.kernel.org>


On 22 March 2017 at 19:56, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
> With the progress made with spars-llvm in recent weeks we are now
> beginning to hit bugs in the simplifications carried out by the
> linearizer. So it is useful I think to have an ability to turn off
> these modifications - perhaps by a command line argument. This will
> enable testing of sparse-llvm with and without the simplifications,
> and help identify issues caused by the latter phase.
>

It appears that there is already an optimize flag that is set by -O
option. Perhaps this flag can be utilized to control the level of
simplifications attempted.

Thanks and Regards
Dibyendu

Re: Adding an option to sparse to disable the IR simplifications

[Luc Van Oostenryck]

On Mon, Aug 14, 2017 at 6:38 PM, Dibyendu Majumdar
<mobile@majumdar.org.uk> wrote:
> Hi,
>
> I suggested a while back that it might be useful to add an option to
> control the simplifications.
>
> In dmrC the simplification phase is turned off by default. But if you
> supply -O1 option then it is turned on. This ensures that  I can test
> both the unsimplified and simplified versions.

Yes, it's something that more useful now than it was months ago.
I think I'll first move some code around though.

-- Luc

Re: Adding an option to sparse to disable the IR simplifications

[Dibyendu Majumdar]

On 14 August 2017 at 17:46, Luc Van Oostenryck
<luc.vanoostenryck@gmail.com> wrote:
> On Mon, Aug 14, 2017 at 6:38 PM, Dibyendu Majumdar
> <mobile@majumdar.org.uk> wrote:
>> I suggested a while back that it might be useful to add an option to
>> control the simplifications.
>>
>> In dmrC the simplification phase is turned off by default. But if you
>> supply -O1 option then it is turned on. This ensures that  I can test
>> both the unsimplified and simplified versions.
>
> Yes, it's something that more useful now than it was months ago.
> I think I'll first move some code around though.
>

We can perhaps also move towards a granular level of control as in gcc
or clang - i.e. enable or disable specific simplifications.

Regards
Dibyendu

Re: Adding an option to sparse to disable the IR simplifications

[Christopher Li]

On Mon, Aug 14, 2017 at 12:59 PM, Dibyendu Majumdar
<mobile@majumdar.org.uk> wrote:
>
> We can perhaps also move towards a granular level of control as in gcc
> or clang - i.e. enable or disable specific simplifications.
>

Yes,  we can also do that. I I think the the back end should have some way
to control what kind of the optimization it want and what level it want etc.

I plan to release 0.5.1 finial on Wednesday. Then the merge windows is
open again. We can start to work on getting the IR right, getting SSA right etc.
I also have some patch want to merged as well like the ptrlist ref count.

It can even have a post release branch for experiment. After release
this branch will merge back to master.


Chris