OSL and SPIR-V

[Alex Fuller]

Something that is quite interesting:

https://github.com/KhronosGroup/SPIRV-LLVM

I wonder if OSL could output something along the lines to this without too much hassles, or it makes more sense to make a direct OSL-to-SPIR-V backend (which might be easier when you can just transform the simpler opcodes to what you really want).

Thoughts? OSL on the GPU might be fun to have, through a renderer which uses Vulkan.

Cheers

[Juhani Karlsson]

I don't really know what really goes into to making this happen, but as an idea I like it! : )

--
You received this message because you are subscribed to the Google Groups "OSL Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to osl-dev+u...@googlegroups.com.
To post to this group, send email to osl...@googlegroups.com.
Visit this group at https://groups.google.com/group/osl-dev.
For more options, visit https://groups.google.com/d/optout.

[bill_n]

I think someone has already at done this at one time. See this slide 

http://www.slideshare.net/takahiroharada/introduction-to-monte-carlo-ray-tracing-opencl-implementation-cedec-2014

see slide 65

[Larry Gritz]

I'm very interested in adapting OSL to have alternate "back ends", especially if those are achievable through our usual generation of LLVM IR. SPIR-V looks interesting as it might be a possible future gateway to multiple vendors/CPUs/technologies, whereas currently it's a little unclear how much overlap there would be for implementations of OSL in OpenGL vs DX vs OptiX/Cuda vs other technologies.

--
Larry Gritz
l...@larrygritz.com