NIR Continues Successfully Serving The Needs Of Mesa, Better Suited Than LLVM

[phoronix]

Phoronix: NIR Continues Successfully Serving The Needs Of Mesa, Better Suited Than LLVM

Prominent Mesa developer Jason Ekstrand who formerly led Intel's "ANV" Vulkan driver effort and being one of their open-source driver developers originally involved with the NIR intermediate representation work wrote a detailed and excellent blog post outlining its successes eight years running. While it still gets brought up into discussions from time to time (including quite recently stemming from a RISC-V graphics thread) why Mesa doesn't use LLVM IR or SPIR-V directly as its intermediate representation, NIR continues as a striking success and used by all major Mesa drivers...

https://www.phoronix.com/scan.php?page=news_item&px=Mesa-NIR-Success-Story

[oleid]

Would it make sense to use NIR within ROCm?

[Maxzor]

Would it make sense to use NIR within ROCm?

The AMDGPU LLVM backend of ROCm has been praised pretty high by everyone involved whom I talked to, and an IR tailored for graphics might not be that suited for compute?

I wonder about this topic: if NIR is the reason ROCm has difficulties currently with Mesa interop, and how CUDA does on its side.
Hello bridgman, you chiming in and transmitting some vulgarization would be much appreciated

[bridgman]

I wonder about this topic: if NIR is the reason ROCm has difficulties currently with Mesa interop, and how CUDA does on its side.
Hello bridgman, you chiming in and transmitting some vulgarization would be much appreciated

I suspect that the main problem with interop between ROCm (and OpenCL in general) and Mesa GL is that our QA team tests compute/graphics interop with the closed source GL driver rather than with Mesa GL. We are working on changing that, as well as extending interop coverage in general.

[Maxzor]

I suspect that the main problem with interop between ROCm (and OpenCL in general) and Mesa GL is that our QA team tests compute/graphics interop with the closed source GL driver rather than with Mesa GL. We are working on changing that, as well as extending interop coverage in general.

Knowing how corporate pace can differ significantly from a consumer horizon, would it make any beginning of some sense if I proposed the subject to GSoC 2022?

[ResponseWriter]

He says quite a bit about the issues with using LLVM for radeonsi. Is there any mention recently of moving to ACO for radeonsi instead?

[shmerl]

He says quite a bit about the issues with using LLVM for radeonsi. Is there any mention recently of moving to ACO for radeonsi instead?

ACO developers said in the past it was the eventual plan, but I haven't seen any recent updates on this.

[geearf]

How about using SPIR-V instead of NIR? Would there any benefit in this?

[gerddie]

How about using SPIR-V instead of NIR? Would there any benefit in this?

Short answer: There wouldn't be any benefit, the long answer can be found in this thread on Mesa-dev: https://lists.freedesktop.org/archiv...ry/225640.html

[geearf]

Short answer: There wouldn't be any benefit, the long answer can be found in this thread on Mesa-dev: https://lists.freedesktop.org/archiv...ry/225640.html

Great answer, thank you!