[maleadt]

Split out of #906; neither change needs PoCL host kernels.

• Total size first. Linearized launches put all work-items in the first dimension, so an oversized (96, 48) work-group was reported as Work-group size (4608, 1, 1) exceeds the maximum of (4096, 4096, 4096) per dimension. The total is now checked first: ... has more than 4096 work-items, the maximum for this kernel.
• No division. prod_exceeds, which runs on every launch, uses checked multiplication instead of a 64-bit division per dimension (tens of cycles on CPUs such as Skylake). Not measurable on PoCL's queue (~20k instructions per launch); it is part of POCL: run kernels on Julia's threads with PoCL host kernels #906's launch-cost work, where a launch takes ~70 ns.
• Bumps KernelInterface to 0.4.4.

[github-actions]

Benchmark results

Compared with c42b1f458, the merge base.
The environment differs from the baseline's: KernelInterface 0.4.3 → 0.4.4.

Times are the minimum time of every benchmark. A change is unusual when it exceeds both three times the benchmark's typical change between consecutive runs on the default branch and a minimum per category (performance 3%, latency 5%, applications 3%, other 3%). A benchmark that starts or stops allocating is listed however few the allocations. A single run can still be a fluke, so rerun before drawing conclusions.

	Geometric mean	Unusual: slower	Unusual: faster	Compared
Performance	−0.0%	0	0	32
Latency	+1.7%	0	0	13
Applications	+10.6%	0	0	5

No unusual changes.

All benchmarks

Benchmark	Baseline	This PR	Change
Performance
julia/serial/arithmetic/muladd-reference	655 µs	655 µs	−0.0%	noise unknown
julia/serial/math/exp-f64-reference	292 µs	284 µs	−2.7%	noise unknown
julia/serial/narrow/masked-reference	714 µs	745 µs	+4.2%	noise unknown
pocl/parallel/atomics/histogram	5.14 ms	5.14 ms	−0.1%	noise unknown
pocl/parallel/groupreduce/add-f64	1.65 ms	1.65 ms	+0.1%	noise unknown
pocl/parallel/groupreduce/fallback	19.0 ms	18.1 ms	−4.6%	noise unknown
pocl/parallel/groupscan/add-i32	2.52 ms	2.52 ms	−0.0%	noise unknown
pocl/parallel/indexing/cartesian3d	78.5 µs	80.7 µs	+2.8%	noise unknown
pocl/parallel/indexing/linear3d	80.3 µs	83.1 µs	+3.5%	noise unknown
pocl/parallel/indexing/ntuple3d	78.7 µs	64.9 µs	−17.5%	noise unknown
pocl/parallel/indexing/ntuple4d	157 µs	161 µs	+2.8%	noise unknown
pocl/parallel/localmem/transpose-naive	456 µs	459 µs	+0.6%	noise unknown
pocl/parallel/localmem/transpose-tiled	92.9 µs	123 µs	+31.9%	noise unknown
pocl/parallel/private/across-barrier	165 µs	160 µs	−3.2%	noise unknown
pocl/parallel/scheduling/default	182 µs	182 µs	−0.1%	noise unknown
pocl/parallel/scheduling/explicit	182 µs	181 µs	−0.1%	noise unknown
pocl/parallel/stream/copy/cache	781 ns	643 ns	−17.7%	noise unknown
pocl/parallel/stream/copy/dram	6.41 ms	6.29 ms	−1.9%	noise unknown
pocl/parallel/stream/triad/cache	7.17 µs	7.53 µs	+5.0%	noise unknown
pocl/parallel/stream/triad/dram	9.42 ms	9.50 ms	+0.8%	noise unknown
pocl/parallel/util/copyto-small	24.0 ns	24.0 ns	+0.0%	noise unknown
pocl/parallel/util/zeros-large	1.01 ms	953 µs	−5.2%	noise unknown
pocl/parallel/util/zeros-small	619 ns	625 ns	+1.0%	noise unknown
pocl/serial/arithmetic/contract	1.91 ms	1.90 ms	−0.5%	noise unknown
pocl/serial/arithmetic/muladd	756 µs	769 µs	+1.7%	noise unknown
pocl/serial/const/marked	728 µs	711 µs	−2.3%	noise unknown
pocl/serial/const/unmarked	910 µs	902 µs	−0.9%	noise unknown
pocl/serial/math/exp-f64	154 µs	147 µs	−4.6%	noise unknown
pocl/serial/math/sin-f64	194 µs	209 µs	+7.4%	noise unknown
pocl/serial/narrow/masked	807 µs	829 µs	+2.7%	noise unknown
pocl/serial/narrow/unmasked	766 µs	809 µs	+5.6%	noise unknown
pocl/serial/stream/triad/dram	8.62 ms	8.56 ms	−0.7%	noise unknown
Latency
host/import/ka	501 ms	507 ms	+1.1%	noise unknown
host/precompile/ka	44.4 s	44.4 s	−0.0%	noise unknown
host/startup	128 ms	129 ms	+0.1%	noise unknown
pocl/parallel/launch/dynamic-default	4.37 µs	4.45 µs	+1.9%	noise unknown
pocl/parallel/launch/groupreduce-f64	6.59 µs	6.64 µs	+0.8%	noise unknown
pocl/parallel/launch/many-args	11.2 µs	12.4 µs	+10.6%	noise unknown
pocl/parallel/launch/partial-workgroup	4.25 µs	4.29 µs	+0.8%	noise unknown
pocl/serial/compile/groupreduce	1.39 s	1.42 s	+2.1%	noise unknown
pocl/serial/compile/localmem-barrier	697 ms	694 ms	−0.4%	noise unknown
pocl/serial/compile/new-kernel	117 ms	126 ms	+7.8%	noise unknown
pocl/serial/compile/new-workgroupsize	14.1 ms	13.8 ms	−2.5%	noise unknown
pocl/serial/first-kernel/cold-cache	1.95 s	1.94 s	−0.2%	noise unknown
pocl/serial/first-kernel/warm-cache	279 ms	280 ms	+0.3%	noise unknown
Applications
pocl/parallel/dgsem/fluxdiff-node	291 µs	289 µs	−0.6%	noise unknown
pocl/parallel/dgsem/weak-element	159 µs	209 µs	+31.2%	noise unknown
pocl/parallel/dgsem/weak-node-localmem	119 µs	146 µs	+22.9%	noise unknown
pocl/parallel/stencil/timestep	7.75 ms	7.97 ms	+2.8%	noise unknown
pocl/serial/stencil/timestep	11.3 ms	11.3 ms	+0.2%	noise unknown

History: https://juliagpu.org/perf/?pkg=kernelabstractions

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.99%. Comparing base (c42b1f4) to head (81fe6df).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #938      +/-   ##
==========================================
- Coverage   84.80%   83.99%   -0.81%     
==========================================
  Files          31       31              
  Lines        2593     2543      -50     
==========================================
- Hits         2199     2136      -63     
- Misses        394      407      +13     

☔ 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.