[maleadt]

KI.copyto! between two arrays always launched a kernel, which takes 2.5 µs for 64 Float32s that the host copies in 10 ns (the performance/pocl/parallel/util/copyto-small benchmark).

Like zeros and ones since #927, this copies with a kernel only from 1 MiB, where it spreads the destination's pages over PoCL's threads (and their memory domains); smaller copies and copies of bits-union arrays use Base.copyto!. Overlapping arrays are copied from the host too, which handles the overlap, instead of throwing "Arrays may not alias".

Measured on an M3 Pro only, where the host copy is faster at every size; the 1 MiB threshold is kept for the many-core and NUMA machines that #927 and #928 were about.

[github-actions]

Benchmark results

Compared with d2ada37c6, the latest measured ancestor of the merge base (1 commit before it).
The environment differs from the baseline's: Parsers 3.0.1 → 3.0.2.

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	−20.7%	0	0	32
Latency	+0.2%	0	0	13
Applications	+10.8%	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	284 µs	286 µs	+0.7%	noise unknown
julia/serial/narrow/masked-reference	702 µs	726 µs	+3.4%	noise unknown
pocl/parallel/atomics/histogram	5.16 ms	5.15 ms	−0.1%	noise unknown
pocl/parallel/groupreduce/add-f64	1.66 ms	1.65 ms	−0.5%	noise unknown
pocl/parallel/groupreduce/fallback	18.1 ms	18.9 ms	+4.5%	noise unknown
pocl/parallel/groupscan/add-i32	2.53 ms	2.53 ms	−0.0%	noise unknown
pocl/parallel/indexing/cartesian3d	84.8 µs	79.5 µs	−6.3%	noise unknown
pocl/parallel/indexing/linear3d	84.4 µs	78.5 µs	−7.0%	noise unknown
pocl/parallel/indexing/ntuple3d	81.4 µs	84.1 µs	+3.4%	noise unknown
pocl/parallel/indexing/ntuple4d	164 µs	168 µs	+2.6%	noise unknown
pocl/parallel/localmem/transpose-naive	459 µs	453 µs	−1.3%	noise unknown
pocl/parallel/localmem/transpose-tiled	101 µs	104 µs	+3.0%	noise unknown
pocl/parallel/private/across-barrier	164 µs	165 µs	+0.8%	noise unknown
pocl/parallel/scheduling/default	181 µs	181 µs	−0.3%	noise unknown
pocl/parallel/scheduling/explicit	181 µs	181 µs	−0.0%	noise unknown
pocl/parallel/stream/copy/cache	6.06 µs	686 ns	−88.7%	noise unknown
pocl/parallel/stream/copy/dram	6.32 ms	6.26 ms	−0.9%	noise unknown
pocl/parallel/stream/triad/cache	6.90 µs	6.96 µs	+0.9%	noise unknown
pocl/parallel/stream/triad/dram	8.94 ms	9.41 ms	+5.2%	noise unknown
pocl/parallel/util/copyto-small	4.59 µs	24.0 ns	−99.5%	noise unknown
pocl/parallel/util/zeros-large	998 µs	1.01 ms	+1.4%	noise unknown
pocl/parallel/util/zeros-small	665 ns	635 ns	−4.5%	noise unknown
pocl/serial/arithmetic/contract	1.91 ms	1.91 ms	+0.1%	noise unknown
pocl/serial/arithmetic/muladd	759 µs	771 µs	+1.5%	noise unknown
pocl/serial/const/marked	728 µs	722 µs	−0.8%	noise unknown
pocl/serial/const/unmarked	915 µs	915 µs	−0.1%	noise unknown
pocl/serial/math/exp-f64	147 µs	147 µs	−0.2%	noise unknown
pocl/serial/math/sin-f64	202 µs	194 µs	−4.2%	noise unknown
pocl/serial/narrow/masked	854 µs	819 µs	−4.0%	noise unknown
pocl/serial/narrow/unmasked	765 µs	798 µs	+4.3%	noise unknown
pocl/serial/stream/triad/dram	8.63 ms	8.62 ms	−0.2%	noise unknown
Latency
host/import/ka	506 ms	505 ms	−0.2%	noise unknown
host/precompile/ka	44.9 s	44.4 s	−0.9%	noise unknown
host/startup	133 ms	131 ms	−2.0%	noise unknown
pocl/parallel/launch/dynamic-default	4.39 µs	4.45 µs	+1.3%	noise unknown
pocl/parallel/launch/groupreduce-f64	18.5 µs	18.4 µs	−0.5%	noise unknown
pocl/parallel/launch/many-args	12.0 µs	12.0 µs	−0.3%	noise unknown
pocl/parallel/launch/partial-workgroup	4.18 µs	4.42 µs	+5.6%	noise unknown
pocl/serial/compile/groupreduce	2.06 s	2.08 s	+1.1%	noise unknown
pocl/serial/compile/localmem-barrier	808 ms	802 ms	−0.8%	noise unknown
pocl/serial/compile/new-kernel	118 ms	119 ms	+0.5%	noise unknown
pocl/serial/compile/new-workgroupsize	13.6 ms	13.6 ms	+0.7%	noise unknown
pocl/serial/first-kernel/cold-cache	2.02 s	2.01 s	−0.3%	noise unknown
pocl/serial/first-kernel/warm-cache	360 ms	356 ms	−0.9%	noise unknown
Applications
pocl/parallel/dgsem/fluxdiff-node	291 µs	291 µs	+0.1%	noise unknown
pocl/parallel/dgsem/weak-element	160 µs	198 µs	+24.2%	noise unknown
pocl/parallel/dgsem/weak-node-localmem	118 µs	147 µs	+24.5%	noise unknown
pocl/parallel/stencil/timestep	7.91 ms	7.86 ms	−0.6%	noise unknown
pocl/serial/stencil/timestep	11.3 ms	12.2 ms	+8.7%	noise unknown

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.27%. Comparing base (d2ada37) to head (d20ab13).
⚠️ Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #935      +/-   ##
==========================================
+ Coverage   82.87%   83.27%   +0.39%     
==========================================
  Files          31       31              
  Lines        2564     2565       +1     
==========================================
+ Hits         2125     2136      +11     
+ Misses        439      429      -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.