[llvm-dev] SPIRV-LLVM as an external tool

[Anastasia Stulova via llvm-dev]

Hi Chris,


The main benefit for LLVM to include SPIRV support directly is to increase the number of users and developers in the area of heterogeneous computing, e.g. GPUs, FPGAs, DSPs.


We want to increase the number of such devices that LLVM natively supports by adding compilation to SPIRV due to the shortage of proprietary backends in upstream LLVM.

Just to clarify we are currently suggesting to integrate the converter as a subproject of LLVM, similar to Clang or libclc, to reduce the overhead for the overall community in maintaining it and running tests.


One more thing to be mentioned, the latest OpenCL standards evolve towards off-line compilation from OpenCL C++ to SPIRV. So having SPIRV generation directly in LLVM would allow us to deliver fully complete and compliant OpenCL C++ support, without using any external tools. See discussion with Tom earlier.


Thanks!

Anastasia


________________________________
From: Chris Lattner <clattner at nondot.org>
Sent: 01 March 2018 01:52
To: tstellar at redhat.com
Cc: Anastasia Stulova; Tomeu Vizoso; llvm-dev at lists.llvm.org; nd
Subject: Re: [llvm-dev] SPIRV-LLVM as an external tool

On Feb 27, 2018, at 10:25 AM, Tom Stellard via llvm-dev <llvm-dev at lists.llvm.org> wrote:
> On 02/27/2018 05:07 AM, Anastasia Stulova wrote:
>>> SPIR-V does not have to be a part of LLVM for you to do this.  You can add
>>> the SPIR-V target to clang and then define a SPIR-V toolchain (i.e. clang/Driver/Toolchains)
>>> that uses the external tool to translate LLVM IR to SPIR-V.
>>
>>
>> Ok. I guess if Clang community accepts this way, it would be better to set up the SPIRV converter as a tool of LLVM.
>>
>> So the question is are there any downsides of this? Or would anyone object if we add the converter to the LLVM project as an optional tool? We would of course take care of configuring and maintaining it ourselves.
>
> There is no requirement that the tool needs to be an official part of the LLVM
> project to implement to use it this way  For example, the cuda toolchain
> in clang relies on a few proprietary tools from NVIDIA.
>
> I think too much emphasis is being placed on having this tool be part of the
> LLVM project.

Agreed.  Why is it good for LLVM to include this tool?

If there was a strong rationale for doing so, it would probably make sense to be a new subproject of some sort rather than included in the main llvm repo (not sure if that is what was being proposed).

-Chris


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20180306/1f58dfa7/attachment.html>