llvm dev - May 2016 - [Openmp-dev] [cfe-dev] RFC: Proposing an LLVM subproject for parallelism runtime and support libraries

On Mon, May 9, 2016 at 12:04 PM Andrey Bokhanko via cfe-dev <
cfe-dev at lists.llvm.org> wrote:
> Thanks Jason (and Chandler) to tell Chandler's opinion, though it still
> doesn't answer my original question:
>
>
> > Both SE and libomptarget are libraries that handle offloading, not
> parallelism. I understand other libraries, to be added in the future, might
> deal with parallelism, but maybe we need a separate project for them?
> (Something Chris already hinted.)
>
> The example simply adds confusion
>
>
> > One example he brought up was AVX 512. He thinks that code explicitly
> targeting CPU parallelism should also be included in this project, even
> though it doesn't fit in the category of "offloading".
>
> Do we want to add a vectorizer to this project as well?
>
I would not expect a vectorizer to really fit here, no.

But I would like to have this be a viable home for support libraries used
by vectorizers if that proves useful.

I thought this was usefully captured by the '_rt' or '_libs'
suffix.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160509/4f152675/attachment.html>

On Mon, May 9, 2016 at 9:07 PM, Chandler Carruth <chandlerc at gmail.com>
wrote:
>
> I would not expect a vectorizer to really fit here, no.
>
I didn't actually suggested to include vectorizer (sorry if wasn't
clear)
-- my whole point that "parallel" is too broad and frankly confusing
word.

Yours,
Andrey
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160510/0bacc190/attachment.html>

Hi

A comment on the naming, while re-reading this thread: it **may** be that
"acceleration" is a better adjective for the project you have in mind,
rather than "offloading" (not applicable to AVX and homogeneous
clusters of
Intel Xeon Phi) or "parallel" (everything is parallel at this point).

I actually think that acceleration is too generic as a term, and misleading
in some cases, but it may fit your needs here.

Thanks

-- Carlo



From:	Andrey Bokhanko via Openmp-dev <openmp-dev at lists.llvm.org>
To:	Chandler Carruth <chandlerc at gmail.com>
Cc:	llvm-dev <llvm-dev at lists.llvm.org>, cfe-dev
            <cfe-dev at lists.llvm.org>, openmp-dev at lists.llvm.org
Date:	05/10/2016 11:32 AM
Subject:	Re: [Openmp-dev] [cfe-dev] [llvm-dev] RFC: Proposing an LLVM
            subproject for parallelism runtime and support libraries
Sent by:	"Openmp-dev" <openmp-dev-bounces at lists.llvm.org>



On Mon, May 9, 2016 at 9:07 PM, Chandler Carruth <chandlerc at gmail.com>
wrote:

  I would not expect a vectorizer to really fit here, no.

I didn't actually suggested to include vectorizer (sorry if wasn't
clear)
-- my whole point that "parallel" is too broad and frankly confusing
word.

Yours,
Andrey_______________________________________________
Openmp-dev mailing list
Openmp-dev at lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/openmp-dev


-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160510/55aea10d/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: graycol.gif
Type: image/gif
Size: 105 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160510/55aea10d/attachment.gif>

I think both "accelerator" and "offload" have the same
problem for me --
they really seem too narrow.

There are a lot of use cases for runtime libraries that are targeted at the
host CPU. Perhaps your host CPU has a nice SIMD unit or it has lots of
cores. Essentially, I feel like at least some of these runtimes are
fundamentally about exploiting parallelism, which is why I argue for that
naming convention.

A concrete example that none have really addressed: a serious use case I
would like to see the project address is to have an LLVM hosted vectorized
math library that we can teach the compiler to leverage when vectorizing
code (and allow users to call directly).

Another example use case that I don't think would be well addressed with
"offload" naming: producing the core runtime support libraries used to
implement the new C++ parallelism TS in libc++.

Now, we could put these libraries into compiler-rt, but I really feel like
they will have more in common with the things in this repository than
compiler-rt.

I would very much like us to go with a *broad* name that allows many use
cases to work effectively in this project, rather than a narrow and
specific name. The libraries themselves should have specific and clear
names, but I don't really want to encourage a proliferation of top level
projects by their charters being overly narrow.

-Chandler

On Tue, May 10, 2016 at 12:14 PM Carlo Bertolli <cbertol at us.ibm.com>
wrote:
> Hi
>
> A comment on the naming, while re-reading this thread: it **may** be that
> "acceleration" is a better adjective for the project you have in
mind,
> rather than "offloading" (not applicable to AVX and homogeneous
clusters of
> Intel Xeon Phi) or "parallel" (everything is parallel at this
point).
>
> I actually think that acceleration is too generic as a term, and
> misleading in some cases, but it may fit your needs here.
>
> Thanks
>
> -- Carlo
>
> [image: Inactive hide details for Andrey Bokhanko via Openmp-dev
> ---05/10/2016 11:32:45 AM---On Mon, May 9, 2016 at 9:07 PM, Chandler
C]Andrey
> Bokhanko via Openmp-dev ---05/10/2016 11:32:45 AM---On Mon, May 9, 2016 at
> 9:07 PM, Chandler Carruth <chandlerc at gmail.com> wrote:
>
> From: Andrey Bokhanko via Openmp-dev <openmp-dev at lists.llvm.org>
>
>
> To: Chandler Carruth <chandlerc at gmail.com>
> Cc: llvm-dev <llvm-dev at lists.llvm.org>, cfe-dev <cfe-dev at
lists.llvm.org>,
> openmp-dev at lists.llvm.org
>
> Date: 05/10/2016 11:32 AM
>
>
> Subject: Re: [Openmp-dev] [cfe-dev] [llvm-dev] RFC: Proposing an LLVM
> subproject for parallelism runtime and support libraries
> Sent by: "Openmp-dev" <openmp-dev-bounces at
lists.llvm.org>
> ------------------------------
>
>
>
> On Mon, May 9, 2016 at 9:07 PM, Chandler Carruth <*chandlerc at
gmail.com*
> <chandlerc at gmail.com>> wrote:
>
>
>    I would not expect a vectorizer to really fit here, no.
>
>
> I didn't actually suggested to include vectorizer (sorry if wasn't
clear)
> -- my whole point that "parallel" is too broad and frankly
confusing word.
>
> Yours,
>
> Andrey_______________________________________________
> Openmp-dev mailing list
> Openmp-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/openmp-dev
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160510/ca36c53b/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: graycol.gif
Type: image/gif
Size: 105 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160510/ca36c53b/attachment-0002.gif>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: graycol.gif
Type: image/gif
Size: 105 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160510/ca36c53b/attachment-0003.gif>

<html><head></head><body lang="en-US"
style="background-color: rgb(255, 255, 255); line-height:
initial;">                                                              
<div style="width: 100%; font-size: initial; font-family: Calibri,
'Slate Pro', sans-serif, sans-serif; color: rgb(31, 73, 125);
text-align: initial; background-color: rgb(255, 255, 255);">The words to
describe what you're referring to are onload vs offloading. It's a very
frequent term in networking as well as computation.</div><div
style="width: 100%; font-size: initial; font-family: Calibri, 'Slate
Pro', sans-serif, sans-serif; color: rgb(31, 73, 125); text-align: initial;
background-color: rgb(255, 255, 255);"><br></div><div
style="width: 100%; font-size: initial; font-family: Calibri, 'Slate
Pro', sans-serif, sans-serif; color: rgb(31, 73, 125); text-align: initial;
background-color: rgb(255, 255, 255);">Accelerator lib... feels
odd</div><div style="width: 100%; font-size: initial; font-family:
Calibri, 'Slate Pro', sans-serif, sans-serif; color: rgb(31, 73, 125);
text-align: initial; background-color: rgb(255, 255,
255);"><br></div><div style="width: 100%; font-size:
initial; font-family: Calibri, 'Slate Pro', sans-serif, sans-serif;
color: rgb(31, 73, 125); text-align: initial; background-color: rgb(255, 255,
255);">What about computation lib?</div><div style="width:
100%; font-size: initial; font-family: Calibri, 'Slate Pro', sans-serif,
sans-serif; color: rgb(31, 73, 125); text-align: initial; background-color:
rgb(255, 255, 255);"><br></div><div style="width:
100%; font-size: initial; font-family: Calibri, 'Slate Pro', sans-serif,
sans-serif; color: rgb(31, 73, 125); text-align: initial; background-color:
rgb(255, 255, 255);">‎Sorry on phone and can't quote your text, but
about math libraries and c++ standard..</div><div style="width:
100%; font-size: initial; font-family: Calibri, 'Slate Pro', sans-serif,
sans-serif; color: rgb(31, 73, 125); text-align: initial; background-color:
rgb(255, 255, 255);"><br></div><div style="width:
100%; font-size: initial; font-family: Calibri, 'Slate Pro', sans-serif,
sans-serif; color: rgb(31, 73, 125); text-align: initial; background-color:
rgb(255, 255, 255);">Math libs *really* need their own home, source repo
and project page. Unix philosophy, do one thing and do it well.. I'm not a
big fan of bucket of soup project. At the target layer depending on it may be
nice but not required. This is a big and wholly complex problem on its own. Do
you write it in CUDA, OCL, asm, C.. is it just to provide vector versions of
functions or what about transcendentals... what about cuFFTW alternative. then
we go into accuracy vs speed discussions.. </div><div
style="width: 100%; font-size: initial; font-family: Calibri, 'Slate
Pro', sans-serif, sans-serif; color: rgb(31, 73, 125); text-align: initial;
background-color: rgb(255, 255, 255);"><br></div><div
style="width: 100%; font-size: initial; font-family: Calibri, 'Slate
Pro', sans-serif, sans-serif; color: rgb(31, 73, 125); text-align: initial;
background-color: rgb(255, 255, 255);">C++-standard... why would this
live anywhere except libc++? Should filesystem be in compiler-rt
instead?</div><div style="width: 100%; font-size: initial;
font-family: Calibri, 'Slate Pro', sans-serif, sans-serif; color:
rgb(31, 73, 125); text-align: initial; background-color: rgb(255, 255,
255);">----</div><div style="width: 100%; font-size:
initial; font-family: Calibri, 'Slate Pro', sans-serif, sans-serif;
color: rgb(31, 73, 125); text-align: initial; background-color: rgb(255, 255,
255);">To add more complexity to the conversation.. what about debugging
and profiling api.. from a tools perspective it should likely be exposed from
the runtime.</div><div style="width: 100%; font-size: initial;
font-family: Calibri, 'Slate Pro', sans-serif, sans-serif; color:
rgb(31, 73, 125); text-align: initial; background-color: rgb(255, 255,
255);"><br></div><div style="width: 100%; font-size:
initial; font-family: Calibri, 'Slate Pro', sans-serif, sans-serif;
color: rgb(31, 73, 125); text-align: initial; background-color: rgb(255, 255,
255);">You can call this a 'lib' but will it ever be a runtime?
</div></body></html>