llvm dev - Jul 2017 - [RFC][GlobalISel] Making GlobalISel non-optional in the build

Hi,

I would like to get rid of the logic around disabling GlobalISel from the build.


*** Context ***

So far, GlobalISel has been this framework built up on the side for about two
years. Now, we are close to a point where having it being optional impedes our
ability to implement the right things. For instance, checks that would fit
naturally in the MachineVerifier are directly done during the execution of the
related GISel generic passes to make sure we don’t end up with ifdefs all around
of CodeGen.


*** What Changed? ***

* GlobalISel is more mature

We now have a better idea of what the framework should look like and the scope
of the changes what will happen on the CodeGen library.

* GlobalISel is getting to be the default for AArch64 O0

We are now on the verge of having GlobalISel being the default for AArch64 at
O0. I would like to make sure it is not possible to build a configuration that
wouldn’t support that default.

* GlobalISel is built by default since roughly 6 months

Unless people actively disabled GlobalISel while building LLVM, they already
build it since end of January!
http://lists.llvm.org/pipermail/llvm-dev/2017-January/109029.html
<http://lists.llvm.org/pipermail/llvm-dev/2017-January/109029.html>

 
*** What Do You Think? ***

As a remainder, the overhead of building GISel is about 1 min on a 8-core
machine and the code size impact for llc is about 1 MB. Data form yesterday ToT
attached r307176. (Format of the title of each section is <built target>
<build type> <gisel or not>.)

As far as I know people don’t actively disable the building of GISel, so
removing that feature shouldn’t be a big deal but I can be wrong.

Cheers,
-Quentin
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170706/571e43c1/attachment-0002.html>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: build_nums.txt
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170706/571e43c1/attachment-0001.txt>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170706/571e43c1/attachment-0003.html>

Gentle ping.

Hi,

I haven’t heard any push back on that.

Does this mean people agree or just missed this message because of July 4th? ;)

Cheers,
-Quentin
> On Jul 6, 2017, at 4:50 PM, Quentin Colombet via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> Hi,
> 
> I would like to get rid of the logic around disabling GlobalISel from the
build.
> 
> 
> *** Context ***
> 
> So far, GlobalISel has been this framework built up on the side for about
two years. Now, we are close to a point where having it being optional impedes
our ability to implement the right things. For instance, checks that would fit
naturally in the MachineVerifier are directly done during the execution of the
related GISel generic passes to make sure we don’t end up with ifdefs all around
of CodeGen.
> 
> 
> *** What Changed? ***
> 
> * GlobalISel is more mature
> 
> We now have a better idea of what the framework should look like and the
scope of the changes what will happen on the CodeGen library.
> 
> * GlobalISel is getting to be the default for AArch64 O0
> 
> We are now on the verge of having GlobalISel being the default for AArch64
at O0. I would like to make sure it is not possible to build a configuration
that wouldn’t support that default.
> 
> * GlobalISel is built by default since roughly 6 months
> 
> Unless people actively disabled GlobalISel while building LLVM, they
already build it since end of January!
> http://lists.llvm.org/pipermail/llvm-dev/2017-January/109029.html
<http://lists.llvm.org/pipermail/llvm-dev/2017-January/109029.html>
> 
>  
> *** What Do You Think? ***
> 
> As a remainder, the overhead of building GISel is about 1 min on a 8-core
machine and the code size impact for llc is about 1 MB. Data form yesterday ToT
attached r307176. (Format of the title of each section is <built target>
<build type> <gisel or not>.)
> 
> As far as I know people don’t actively disable the building of GISel, so
removing that feature shouldn’t be a big deal but I can be wrong.
> 
> Cheers,
> -Quentin
> <build_nums.txt>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170712/18cdeca3/attachment-0001.html>

On Wed, Jul 12, 2017 at 11:34 AM, Quentin Colombet via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Gentle ping.
>
> Hi,
>
> I haven’t heard any push back on that.
>
> Does this mean people agree or just missed this message because of July
> 4th? ;)
>
Both. :)

I'll give you a +1.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170712/5f87b5ac/attachment.html>