[maleadt]

Split out of #906.

The CPU backend gives PoCL as many threads as Julia has, and the main CI job uses one, so work-groups only ran in parallel in the POCL session test's 4-thread subprocess. This adds a job running the suite with JULIA_NUM_THREADS=4 on Linux and macOS, with Julia 1.10 and 1.13.

[github-actions]

Benchmark results

Compared with c10ff8cf4, the merge base.
The environment differs from the baseline's: GPUCompiler 2.18.0 → 2.19.0.

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	−0.5%	0	0	32
Latency	+1.0%	0	0	13
Applications	+1.0%	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	290 µs	284 µs	−2.0%	noise unknown
julia/serial/narrow/masked-reference	689 µs	692 µs	+0.5%	noise unknown
pocl/parallel/atomics/histogram	5.13 ms	5.11 ms	−0.5%	noise unknown
pocl/parallel/groupreduce/add-f64	1.65 ms	1.65 ms	−0.1%	noise unknown
pocl/parallel/groupreduce/fallback	18.1 ms	18.1 ms	+0.0%	noise unknown
pocl/parallel/groupscan/add-i32	2.50 ms	2.52 ms	+0.5%	noise unknown
pocl/parallel/indexing/cartesian3d	73.7 µs	82.3 µs	+11.6%	noise unknown
pocl/parallel/indexing/linear3d	81.0 µs	79.1 µs	−2.4%	noise unknown
pocl/parallel/indexing/ntuple3d	80.7 µs	80.1 µs	−0.7%	noise unknown
pocl/parallel/indexing/ntuple4d	162 µs	175 µs	+8.1%	noise unknown
pocl/parallel/localmem/transpose-naive	455 µs	456 µs	+0.2%	noise unknown
pocl/parallel/localmem/transpose-tiled	125 µs	106 µs	−15.1%	noise unknown
pocl/parallel/private/across-barrier	164 µs	164 µs	+0.4%	noise unknown
pocl/parallel/scheduling/default	185 µs	181 µs	−1.9%	noise unknown
pocl/parallel/scheduling/explicit	181 µs	181 µs	−0.3%	noise unknown
pocl/parallel/stream/copy/cache	786 ns	697 ns	−11.3%	noise unknown
pocl/parallel/stream/copy/dram	6.16 ms	6.31 ms	+2.5%	noise unknown
pocl/parallel/stream/triad/cache	6.84 µs	6.64 µs	−3.0%	noise unknown
pocl/parallel/stream/triad/dram	9.34 ms	9.38 ms	+0.4%	noise unknown
pocl/parallel/util/copyto-small	24.0 ns	24.0 ns	+0.0%	noise unknown
pocl/parallel/util/zeros-large	995 µs	1000 µs	+0.4%	noise unknown
pocl/parallel/util/zeros-small	651 ns	635 ns	−2.5%	noise unknown
pocl/serial/arithmetic/contract	1.91 ms	1.93 ms	+0.9%	noise unknown
pocl/serial/arithmetic/muladd	764 µs	758 µs	−0.8%	noise unknown
pocl/serial/const/marked	722 µs	732 µs	+1.3%	noise unknown
pocl/serial/const/unmarked	922 µs	889 µs	−3.5%	noise unknown
pocl/serial/math/exp-f64	146 µs	145 µs	−0.3%	noise unknown
pocl/serial/math/sin-f64	186 µs	190 µs	+2.4%	noise unknown
pocl/serial/narrow/masked	824 µs	829 µs	+0.5%	noise unknown
pocl/serial/narrow/unmasked	789 µs	794 µs	+0.7%	noise unknown
pocl/serial/stream/triad/dram	8.59 ms	8.62 ms	+0.3%	noise unknown
Latency
host/import/ka	509 ms	505 ms	−0.9%	noise unknown
host/precompile/ka	44.6 s	44.7 s	+0.1%	noise unknown
host/startup	132 ms	130 ms	−1.7%	noise unknown
pocl/parallel/launch/dynamic-default	4.30 µs	4.44 µs	+3.3%	noise unknown
pocl/parallel/launch/groupreduce-f64	5.03 µs	5.90 µs	+17.2%	noise unknown
pocl/parallel/launch/many-args	12.4 µs	12.3 µs	−0.3%	noise unknown
pocl/parallel/launch/partial-workgroup	4.19 µs	4.10 µs	−2.0%	noise unknown
pocl/serial/compile/groupreduce	1.88 s	1.87 s	−0.5%	noise unknown
pocl/serial/compile/localmem-barrier	719 ms	718 ms	−0.1%	noise unknown
pocl/serial/compile/new-kernel	110 ms	109 ms	−0.0%	noise unknown
pocl/serial/compile/new-workgroupsize	13.6 ms	13.8 ms	+1.5%	noise unknown
pocl/serial/first-kernel/cold-cache	2.05 s	2.04 s	−0.3%	noise unknown
pocl/serial/first-kernel/warm-cache	392 ms	385 ms	−1.7%	noise unknown
Applications
pocl/parallel/dgsem/fluxdiff-node	291 µs	290 µs	−0.4%	noise unknown
pocl/parallel/dgsem/weak-element	203 µs	209 µs	+2.8%	noise unknown
pocl/parallel/dgsem/weak-node-localmem	148 µs	147 µs	−0.6%	noise unknown
pocl/parallel/stencil/timestep	7.80 ms	8.00 ms	+2.5%	noise unknown
pocl/serial/stencil/timestep	11.1 ms	11.2 ms	+0.6%	noise unknown

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.76%. Comparing base (c10ff8c) to head (2a4516a).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #940      +/-   ##
==========================================
- Coverage   83.46%   82.76%   -0.71%     
==========================================
  Files          31       31              
  Lines        2571     2524      -47     
==========================================
- Hits         2146     2089      -57     
- Misses        425      435      +10     

☔ 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.