Repository navigation
Switch to POCL #600
Description
The outstanding
- POCL@v7 Switch to pocl_jll@v7 #599
- printf support [nanoPOCL] Re-enable error reporting in 1.12+ #681 Use Printf to report errors from POCL #592
- Switch to pocl_standalone Use pocl_standalone_jll #704
- PoCL: Add standalone (OpenCL-less) variant JuliaPackaging/Yggdrasil#11166
- Renaming issue in build
- Distentagle SPIRVIntrinsics: Currently we inherit quirks from OpenCL https://github.com/JuliaGPU/KernelAbstractions.jl/blob/main/src/pocl/device/quirks.jl but this uses the same method table as OpenCL
- Test OpenCL and KA together to ensure that we have no hidden conflicts.
- Downstream CI ala Enable downstream CI for 0.10 #608
- [nanoOpenCL] illegal addrspace cast in Julia 1.12 #624
- [pocl] Device side RNG #641 ([pocl] RNG support #659)
- float atomics
- Float16 support
- Windows compilation failure
cc: @maleadt
Activity
JamesWrigley commentedon Sep 15, 2025
Gentle bump, just checking in to see if there's anything a relative noob could help with :) #542 is hitting me again so it'd be cool to get this over the line.
christiangnrd commentedon Nov 7, 2025
#624 should probably be added to this list
christiangnrd commentedon Nov 19, 2025
Seems like I missed method table stuff in #635 as uncovered by the buildkite OpenCL run.
vchuravy commentedon Nov 19, 2025
Huh? That should be fixed post #615 and JuliaGPU/OpenCL.jl#325 each package should use it's own
method_tableand then stack that on top of the SPIRV method table.vchuravy commentedon Nov 19, 2025
@simeonschaub could I bother you to take a look at what it would take to implement device side rand for pocl?
simeonschaub commentedon Nov 19, 2025
Sure, I can look into it! Shouldn't be too difficult to adapt JuliaGPU/OpenCL.jl#365. It's a little unfortunate to have two copies of the same code, I wonder whether we should eventually move some of the implementation to SPIRVIntrinsics? Not really sure that makes sense though, since the parts are all intertwined.
christiangnrd commentedon Sep 12, 2026
Close?
vchuravy commentedon Sep 12, 2026
Yeah I think we are pretty much there