Intel Looking To Upstream A Proper SPIR-V Compute Back-End For LLVM

[phoronix]

Phoronix: Intel Looking To Upstream A Proper SPIR-V Compute Back-End For LLVM

It's been talked about many times from various parties but so far has remained elusive from the mainline LLVM code-base: a SPIR-V back-end for LLVM that would go from LLVM into this Khronos intermediate representation most notably used by OpenCL and Vulkan drivers. Intel engineers are stepping up and hope to help get a proper SPIR-V back-end upstreamed into LLVM...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-2021-LLVM-SPIR-V-Backend

[Orcblood]

Good. Intel is stepping up and doing what AMD has struggled to do manly cause they didn't have the talent or resources to. Glad to see Intel bringing the challenge to closed-source-minded Nvidia and hopefully improve things for AMD in the process & the open-source community. Cuda needs to die or be made open-source.

[StillStuckOnSI]

Highly recommend reading the linked mailing list discussion. I hope they come to a consensus on this soon though, because lack of good SPIR-V compiler infrastructure is really hurting alternative (i.e. non CUDA) compute platforms. Currently, everyone has to implement their own language IR -> SPIR-V backend or write GLSL (mostly for Vulkan). The former is no small effort and the latter sucks for compute. The LLVM-SPIRV translator kinda works, but it can only be run as a CLI tool and doesn't support Vulkan-flavoured SPIR-V. MLIR is probably the most ideal path for this, but it's still very immature and doesn't expose enough useful bits in its C API for other languages (plain C, Rust, Julia, Nim, Go, etc.) to use.