Switch POCL backend to CompilerCaching.jl

[vchuravy]

Adopts the new GPUCompiler 2.0 caching API for the POCL backend, x-ref JuliaGPU/GPUCompiler.jl#794, following the same approach as JuliaGPU/OpenCL.jl#431:

• Compilation artifacts (SPIR-V bytes, entry name, device_rng flag) now live in an OpenCLResults struct managed by GPUCompiler.cached_results, i.e. attached to the CodeInstance in Julia's integrated code cache on 1.11+ (persisting through precompilation) instead of the legacy per-context cached_compilation dictionaries.
• Linked cl.Kernel handles are session-local and kept in a small per-context linear cache on the results struct; they are not populated while precompiling, so no dangling handles end up in package images.
• GPUCompiler.reset_runtime() is gone (the runtime library is cached through the same mechanism now).
• Adds a compilation-cache regression test mirroring the one from the OpenCL.jl PR.

GPUCompiler 2.0 is now registered, so the [sources] branch pin is dropped and the compat bound is "2". compile_or_lookup is adapted to the released cached_results API, which returns nothing until code exists for the job and is re-fetched after compiling — matching the final form merged in JuliaGPU/OpenCL.jl#431.

🤖 Generated with Claude Code

[github-actions]

Benchmark Results

Show table

	main	4de90e0...	main / 4de90e0...
saxpy/default/Float32/1024	0.0844 ± 0.0082 ms	0.0844 ± 0.011 ms	1 ± 0.16
saxpy/default/Float32/1048576	0.492 ± 0.023 ms	0.496 ± 0.024 ms	0.993 ± 0.067
saxpy/default/Float32/16384	0.0661 ± 0.033 ms	0.0668 ± 0.033 ms	0.99 ± 0.7
saxpy/default/Float32/2048	0.082 ± 0.029 ms	0.0848 ± 0.024 ms	0.967 ± 0.44
saxpy/default/Float32/256	0.0846 ± 0.014 ms	0.0849 ± 0.013 ms	0.997 ± 0.22
saxpy/default/Float32/262144	0.172 ± 0.031 ms	0.176 ± 0.031 ms	0.975 ± 0.24
saxpy/default/Float32/32768	0.0729 ± 0.031 ms	0.0732 ± 0.031 ms	0.996 ± 0.59
saxpy/default/Float32/4096	0.0814 ± 0.031 ms	0.0807 ± 0.03 ms	1.01 ± 0.54
saxpy/default/Float32/512	0.0842 ± 0.01 ms	0.0844 ± 0.011 ms	0.998 ± 0.18
saxpy/default/Float32/64	0.084 ± 0.015 ms	0.0844 ± 0.013 ms	0.995 ± 0.24
saxpy/default/Float32/65536	0.0869 ± 0.03 ms	0.087 ± 0.031 ms	0.999 ± 0.5
saxpy/default/Float64/1024	0.0845 ± 0.013 ms	0.0845 ± 0.011 ms	1 ± 0.2
saxpy/default/Float64/1048576	0.578 ± 0.068 ms	0.575 ± 0.065 ms	1 ± 0.17
saxpy/default/Float64/16384	0.0673 ± 0.029 ms	0.0694 ± 0.03 ms	0.97 ± 0.6
saxpy/default/Float64/2048	0.0813 ± 0.031 ms	0.0832 ± 0.028 ms	0.978 ± 0.5
saxpy/default/Float64/256	0.0848 ± 0.012 ms	0.0846 ± 0.013 ms	1 ± 0.2
saxpy/default/Float64/262144	0.193 ± 0.036 ms	0.196 ± 0.036 ms	0.987 ± 0.26
saxpy/default/Float64/32768	0.0759 ± 0.029 ms	0.0761 ± 0.029 ms	0.997 ± 0.54
saxpy/default/Float64/4096	0.0762 ± 0.03 ms	0.0784 ± 0.029 ms	0.972 ± 0.52
saxpy/default/Float64/512	0.0843 ± 0.011 ms	0.0845 ± 0.012 ms	0.997 ± 0.19
saxpy/default/Float64/64	0.0837 ± 0.019 ms	0.0845 ± 0.013 ms	0.991 ± 0.27
saxpy/default/Float64/65536	0.0927 ± 0.03 ms	0.0935 ± 0.03 ms	0.991 ± 0.45
saxpy/static workgroup=(1024,)/Float32/1024	0.0817 ± 0.015 ms	0.0823 ± 0.012 ms	0.994 ± 0.23
saxpy/static workgroup=(1024,)/Float32/1048576	0.425 ± 0.033 ms	0.428 ± 0.033 ms	0.992 ± 0.11
saxpy/static workgroup=(1024,)/Float32/16384	0.0629 ± 0.031 ms	0.064 ± 0.031 ms	0.982 ± 0.68
saxpy/static workgroup=(1024,)/Float32/2048	0.0799 ± 0.029 ms	0.0825 ± 0.025 ms	0.968 ± 0.46
saxpy/static workgroup=(1024,)/Float32/256	0.0822 ± 0.01 ms	0.0829 ± 0.015 ms	0.991 ± 0.22
saxpy/static workgroup=(1024,)/Float32/262144	0.153 ± 0.031 ms	0.153 ± 0.031 ms	0.996 ± 0.29
saxpy/static workgroup=(1024,)/Float32/32768	0.0677 ± 0.028 ms	0.0687 ± 0.029 ms	0.985 ± 0.57
saxpy/static workgroup=(1024,)/Float32/4096	0.0768 ± 0.03 ms	0.0796 ± 0.029 ms	0.965 ± 0.52
saxpy/static workgroup=(1024,)/Float32/512	0.0821 ± 0.015 ms	0.0827 ± 0.011 ms	0.993 ± 0.22
saxpy/static workgroup=(1024,)/Float32/64	0.0822 ± 0.016 ms	0.0829 ± 0.014 ms	0.992 ± 0.25
saxpy/static workgroup=(1024,)/Float32/65536	0.0798 ± 0.029 ms	0.0809 ± 0.03 ms	0.986 ± 0.51
saxpy/static workgroup=(1024,)/Float64/1024	0.0813 ± 0.026 ms	0.0821 ± 0.012 ms	0.99 ± 0.35
saxpy/static workgroup=(1024,)/Float64/1048576	0.538 ± 0.081 ms	0.545 ± 0.081 ms	0.987 ± 0.21
saxpy/static workgroup=(1024,)/Float64/16384	0.064 ± 0.028 ms	0.0641 ± 0.028 ms	0.998 ± 0.61
saxpy/static workgroup=(1024,)/Float64/2048	0.0797 ± 0.032 ms	0.081 ± 0.029 ms	0.984 ± 0.52
saxpy/static workgroup=(1024,)/Float64/256	0.0822 ± 0.012 ms	0.0825 ± 0.011 ms	0.996 ± 0.2
saxpy/static workgroup=(1024,)/Float64/262144	0.189 ± 0.034 ms	0.19 ± 0.033 ms	0.995 ± 0.25
saxpy/static workgroup=(1024,)/Float64/32768	0.0731 ± 0.027 ms	0.073 ± 0.027 ms	1 ± 0.52
saxpy/static workgroup=(1024,)/Float64/4096	0.0728 ± 0.03 ms	0.0763 ± 0.029 ms	0.954 ± 0.53
saxpy/static workgroup=(1024,)/Float64/512	0.0824 ± 0.0085 ms	0.0822 ± 0.014 ms	1 ± 0.2
saxpy/static workgroup=(1024,)/Float64/64	0.0822 ± 0.017 ms	0.0822 ± 0.017 ms	1 ± 0.29
saxpy/static workgroup=(1024,)/Float64/65536	0.0902 ± 0.027 ms	0.0907 ± 0.027 ms	0.994 ± 0.42
time_to_load	1.05 ± 0.014 s	1.09 ± 0.019 s	0.968 ± 0.021

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).