LLVM Clang Lands Initial SPIR-V Toolchain Support

[phoronix]

Phoronix: LLVM Clang Lands Initial SPIR-V Toolchain Support

An exciting LLVM development has landed in time for Christmas! Complementary to the LLVM SPIR-V back-end work that could soon be mainlined, Clang has now merged the initial SPIR-V toolchain support. This allows going from Clang to the SPIR-V intermediate representation that is conformant to the OpenCL environment specification -- such as for compiling OpenCL kernels into SPIR-V...

https://www.phoronix.com/scan.php?page=news_item&px=Clang-SPIR-V-Toolchain-Lands

[trilean]

Does that mean we might see first-class support for writing shaders in Rust (and other languages using LLVM) in 2022? I can see a hype-train approaching

[coder]

Does that mean we might see first-class support for writing shaders in Rust (and other languages using LLVM) in 2022? I can see a hype-train approaching

You wouldn't use OpenCL SPIR-V for graphics shaders.

Also, you might be disappointed with the performance you get by using a HLL like Rust (or even C++) to write compute kernels that wasn't designed for this purpose. At the very least, decent efficiency would depend on using some nonstandard extensions/pragmas, restricting oneself to a limited subset of the language, and using PGO.

I should add that SYCL is probably a good place to start, if you're interested in understanding what writing GPU kernels in a HLL would look like. In this case, the language is C++, but similar restrictions would likely apply to any other language.