[pocl] Back @private Scratchpad with GPUCompiler.alloca

[vchuravy]

Summary

Replaces the POCL back-end's MArray-backed @private scratchpad with a direct per-workitem stack allocation via GPUCompiler.alloca. The returned Ptr is wrapped in a CLDeviceArray over OpenCL "Function" storage (LLVM addrspace 0), which is where the SPIR-V target places allocas.

@device_override @inline function KA.Scratchpad(ctx, ::Type{T}, ::Val{Dims}) where {T, Dims}
    ptr = POCL.GPUCompiler.alloca(T, Val(prod(Dims)))
    CLDeviceArray(Dims, reinterpret(POCL.LLVMPtr{T, POCL.AS.Function}, ptr))
end

This drops the StaticArrays dependency from the POCL back-end (StaticArrays is still used by the CPU back-end).

Why

GPUCompiler.alloca emits a real entry-block alloca that the optimizer can promote, in the target's alloca address space — avoiding the unsoundness of llvmcall + alloca and the overhead/semantics of MArray. See the motivation in the companion GPUCompiler PR.

Alignment

The alloca is aligned to Base.datatype_alignment(T), which is exactly the alignment CLDeviceArray uses for its element loads/stores (alignment(::CLDeviceArray{T})), so accesses are consistent. isbits-union element types are intentionally unsupported (GPUCompiler.alloca guards on isbitstype(T)).

Status

Draft — depends on JuliaGPU/GPUCompiler.jl#859 (adds the alloca intrinsic). Project.toml compat is bumped to GPUCompiler = "1.23"; this can be un-drafted once that is merged and released.

Testing

Verified end-to-end against the local GPUCompiler branch: @private Float32 (4,) lowers to alloca [16 x i8], align 4 in addrspace 0 with no surviving julia.gpu.alloca, and the kernel runs correctly on the POCL CPU device.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 0.00%. Comparing base (a8022b2) to head (0b58f9b).

Files with missing lines	Patch %	Lines
src/pocl/backend.jl	0.00%	2 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (a8022b2) and HEAD (0b58f9b). Click for more details.

HEAD has 28 uploads less than BASE

Flag	BASE (a8022b2)	HEAD (0b58f9b)
	48	20

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #714       +/-   ##
==========================================
- Coverage   62.51%   0.00%   -62.52%     
==========================================
  Files          23      22        -1     
  Lines        1926    1737      -189     
==========================================
- Hits         1204       0     -1204     
- Misses        722    1737     +1015     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

Benchmark Results

Show table

	main	c342d87...	main / c342d87...
saxpy/default/Float32/1024	0.085 ± 0.01 ms	0.0854 ± 0.013 ms	0.996 ± 0.19
saxpy/default/Float32/1048576	0.505 ± 0.025 ms	0.502 ± 0.025 ms	1.01 ± 0.07
saxpy/default/Float32/16384	0.0673 ± 0.034 ms	0.0665 ± 0.032 ms	1.01 ± 0.71
saxpy/default/Float32/2048	0.0827 ± 0.028 ms	0.085 ± 0.027 ms	0.973 ± 0.45
saxpy/default/Float32/256	0.0853 ± 0.011 ms	0.0862 ± 0.012 ms	0.99 ± 0.19
saxpy/default/Float32/262144	0.179 ± 0.031 ms	0.179 ± 0.031 ms	1 ± 0.24
saxpy/default/Float32/32768	0.0732 ± 0.031 ms	0.073 ± 0.03 ms	1 ± 0.59
saxpy/default/Float32/4096	0.0808 ± 0.031 ms	0.0854 ± 0.031 ms	0.946 ± 0.5
saxpy/default/Float32/512	0.0847 ± 0.01 ms	0.0859 ± 0.0099 ms	0.986 ± 0.16
saxpy/default/Float32/64	0.0853 ± 0.013 ms	0.0856 ± 0.017 ms	0.996 ± 0.25
saxpy/default/Float32/65536	0.0893 ± 0.031 ms	0.0889 ± 0.031 ms	1 ± 0.5
saxpy/default/Float64/1024	0.0849 ± 0.013 ms	0.0862 ± 0.015 ms	0.985 ± 0.23
saxpy/default/Float64/1048576	0.577 ± 0.086 ms	0.515 ± 0.059 ms	1.12 ± 0.21
saxpy/default/Float64/16384	0.0672 ± 0.03 ms	0.0681 ± 0.029 ms	0.988 ± 0.61
saxpy/default/Float64/2048	0.0821 ± 0.031 ms	0.0858 ± 0.028 ms	0.957 ± 0.48
saxpy/default/Float64/256	0.0854 ± 0.011 ms	0.0873 ± 0.011 ms	0.978 ± 0.18
saxpy/default/Float64/262144	0.195 ± 0.039 ms	0.182 ± 0.033 ms	1.07 ± 0.29
saxpy/default/Float64/32768	0.0788 ± 0.03 ms	0.0779 ± 0.03 ms	1.01 ± 0.55
saxpy/default/Float64/4096	0.0801 ± 0.03 ms	0.079 ± 0.031 ms	1.01 ± 0.55
saxpy/default/Float64/512	0.0849 ± 0.011 ms	0.0863 ± 0.012 ms	0.984 ± 0.18
saxpy/default/Float64/64	0.0854 ± 0.01 ms	0.0866 ± 0.011 ms	0.986 ± 0.17
saxpy/default/Float64/65536	0.101 ± 0.031 ms	0.0964 ± 0.031 ms	1.05 ± 0.47
saxpy/static workgroup=(1024,)/Float32/1024	0.0824 ± 0.01 ms	0.0837 ± 0.013 ms	0.984 ± 0.2
saxpy/static workgroup=(1024,)/Float32/1048576	0.432 ± 0.027 ms	0.428 ± 0.031 ms	1.01 ± 0.098
saxpy/static workgroup=(1024,)/Float32/16384	0.064 ± 0.032 ms	0.0634 ± 0.031 ms	1.01 ± 0.71
saxpy/static workgroup=(1024,)/Float32/2048	0.0805 ± 0.028 ms	0.0824 ± 0.028 ms	0.977 ± 0.47
saxpy/static workgroup=(1024,)/Float32/256	0.0826 ± 0.014 ms	0.0839 ± 0.015 ms	0.986 ± 0.24
saxpy/static workgroup=(1024,)/Float32/262144	0.158 ± 0.032 ms	0.153 ± 0.031 ms	1.03 ± 0.3
saxpy/static workgroup=(1024,)/Float32/32768	0.0675 ± 0.028 ms	0.0685 ± 0.028 ms	0.985 ± 0.57
saxpy/static workgroup=(1024,)/Float32/4096	0.0797 ± 0.03 ms	0.0839 ± 0.03 ms	0.949 ± 0.49
saxpy/static workgroup=(1024,)/Float32/512	0.0825 ± 0.012 ms	0.0838 ± 0.013 ms	0.984 ± 0.21
saxpy/static workgroup=(1024,)/Float32/64	0.0826 ± 0.019 ms	0.0831 ± 0.019 ms	0.994 ± 0.32
saxpy/static workgroup=(1024,)/Float32/65536	0.0808 ± 0.031 ms	0.082 ± 0.031 ms	0.986 ± 0.53
saxpy/static workgroup=(1024,)/Float64/1024	0.0822 ± 0.022 ms	0.0832 ± 0.022 ms	0.988 ± 0.37
saxpy/static workgroup=(1024,)/Float64/1048576	0.54 ± 0.095 ms	0.493 ± 0.077 ms	1.09 ± 0.26
saxpy/static workgroup=(1024,)/Float64/16384	0.0654 ± 0.029 ms	0.0648 ± 0.027 ms	1.01 ± 0.62
saxpy/static workgroup=(1024,)/Float64/2048	0.0801 ± 0.031 ms	0.0829 ± 0.029 ms	0.967 ± 0.51
saxpy/static workgroup=(1024,)/Float64/256	0.0825 ± 0.018 ms	0.0835 ± 0.013 ms	0.989 ± 0.26
saxpy/static workgroup=(1024,)/Float64/262144	0.191 ± 0.037 ms	0.179 ± 0.036 ms	1.06 ± 0.3
saxpy/static workgroup=(1024,)/Float64/32768	0.075 ± 0.027 ms	0.0738 ± 0.026 ms	1.02 ± 0.52
saxpy/static workgroup=(1024,)/Float64/4096	0.0775 ± 0.029 ms	0.0729 ± 0.031 ms	1.06 ± 0.61
saxpy/static workgroup=(1024,)/Float64/512	0.0825 ± 0.014 ms	0.0839 ± 0.0097 ms	0.983 ± 0.2
saxpy/static workgroup=(1024,)/Float64/64	0.0829 ± 0.015 ms	0.0837 ± 0.015 ms	0.99 ± 0.25
saxpy/static workgroup=(1024,)/Float64/65536	0.0935 ± 0.028 ms	0.0923 ± 0.027 ms	1.01 ± 0.43
time_to_load	1.08 ± 0.042 s	0.89 ± 0.0083 s	1.22 ± 0.049

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).

[christiangnrd]

Julia 1.11/1.13 on linux arm test failures unrelated