[release-0.9] Rebuild reshaped and permuted arrays without validation in @Const

[maleadt]

Backport of #793 to 0.9, which CUDA.jl and the other back-ends currently use. The testset on this branch had no constarg2d kernel, so that is added along with the test.

@Const adapts its argument inside the kernel, and Adapt.jl's rules for ReshapedArray (via reshape, which recomputes the SignedMultiplicativeInverses) and PermutedDimsArray (whose constructor checks isperm_inverse) go through error paths that build strings, which fails to compile on GPU back-ends. This adds ConstAdaptor-specific rules that swap in the constified parent and reuse the existing fields, which also avoids recomputing the inverses in every thread. The rules are specific to ConstAdaptor because host-side adapt relies on reshape collapsing wrappers into plain arrays.

Fixes #792

[github-actions]

Benchmark Results

	main	589d21d...	main / 589d21d...
const/@Const/Float32/262144	0.149 ± 0.021 ms	0.0827 ± 0.00071 ms	1.8 ± 0.25
const/@Const/Float32/65536	0.0557 ± 0.017 ms	19.5 ± 0.22 μs	2.86 ± 0.89
const/@Const/Float64/262144	0.239 ± 0.016 ms	0.166 ± 0.00088 ms	1.43 ± 0.096
const/@Const/Float64/65536	0.0809 ± 0.015 ms	0.0398 ± 0.00041 ms	2.03 ± 0.38
const/unmarked/Float32/262144	0.19 ± 0.013 ms	1.59 ± 0.0053 ms	0.12 ± 0.008
const/unmarked/Float32/65536	0.066 ± 0.016 ms	0.393 ± 0.0061 ms	0.168 ± 0.042
const/unmarked/Float64/262144	0.348 ± 0.027 ms	1.59 ± 0.0055 ms	0.219 ± 0.017
const/unmarked/Float64/65536	0.102 ± 0.011 ms	0.398 ± 0.006 ms	0.255 ± 0.028
launch/3D static workgroup, dynamic ndrange	21 ± 21 μs	0.0912 ± 0.0029 μs	230 ± 2.3e+02
launch/3D static workgroup, static ndrange	20.3 ± 20 μs	0.0445 ± 0.002 μs	456 ± 4.5e+02
launch/dynamic workgroup, dynamic ndrange	18.7 ± 18 μs	0.0516 ± 0.0014 μs	363 ± 3.6e+02
launch/dynamic workgroup, dynamic ndrange, workgroupsize given	18.1 ± 17 μs	0.0517 ± 0.0012 μs	351 ± 3.2e+02
launch/static workgroup, dynamic ndrange	19.2 ± 18 μs	28 ± 1.1 ns	687 ± 6.5e+02
launch/static workgroup, static ndrange	18.9 ± 18 μs	26.7 ± 1.2 ns	709 ± 6.7e+02
partition/dynamic workgroup, dynamic ndrange	0.0327 ± 0.0018 μs	31.5 ± 1.2 ns	1.04 ± 0.07
partition/static workgroup, dynamic ndrange	0.0414 ± 0.0088 μs	0.0388 ± 0.00087 μs	1.07 ± 0.23
partition/static workgroup, static ndrange	1.12 ± 0.001 ns	8.02 ± 3.7 ns	0.14 ± 0.064
saxpy/default/Float16/1024	21.5 ± 23 μs	0.394 ± 0.019 μs	54.6 ± 58
saxpy/default/Float16/1048576	0.2 ± 0.012 ms	0.0827 ± 0.0042 ms	2.41 ± 0.19
saxpy/default/Float16/16384	28.4 ± 18 μs	1.46 ± 0.0051 μs	19.4 ± 12
saxpy/default/Float16/2048	22.9 ± 22 μs	0.455 ± 0.016 μs	50.4 ± 49
saxpy/default/Float16/256	23.1 ± 23 μs	0.296 ± 0.0037 μs	77.8 ± 78
saxpy/default/Float16/262144	0.0674 ± 0.016 ms	19.1 ± 0.61 μs	3.52 ± 0.84
saxpy/default/Float16/32768	29.6 ± 16 μs	2.61 ± 0.0079 μs	11.4 ± 6.3
saxpy/default/Float16/4096	23.6 ± 13 μs	0.609 ± 0.017 μs	38.7 ± 21
saxpy/default/Float16/512	24.1 ± 23 μs	0.336 ± 0.0053 μs	71.5 ± 69
saxpy/default/Float16/64	0.0321 ± 0.023 ms	0.276 ± 0.019 μs	116 ± 83
saxpy/default/Float16/65536	0.034 ± 0.017 ms	4.83 ± 0.014 μs	7.04 ± 3.5
saxpy/default/Float32/1024	22.1 ± 23 μs	0.317 ± 0.0036 μs	69.9 ± 72
saxpy/default/Float32/1048576	0.184 ± 0.028 ms	0.129 ± 0.0056 ms	1.43 ± 0.23
saxpy/default/Float32/16384	26.5 ± 14 μs	1.28 ± 0.0089 μs	20.6 ± 11
saxpy/default/Float32/2048	21.9 ± 18 μs	0.338 ± 0.016 μs	64.9 ± 53
saxpy/default/Float32/256	24.3 ± 23 μs	0.278 ± 0.0022 μs	87.4 ± 83
saxpy/default/Float32/262144	0.0604 ± 0.019 ms	29.7 ± 1.6 μs	2.03 ± 0.64
saxpy/default/Float32/32768	27.9 ± 15 μs	1.77 ± 0.046 μs	15.8 ± 8.4
saxpy/default/Float32/4096	24.2 ± 17 μs	0.481 ± 0.029 μs	50.3 ± 35
saxpy/default/Float32/512	20.6 ± 23 μs	0.307 ± 0.017 μs	67.1 ± 74
saxpy/default/Float32/64	22.6 ± 23 μs	0.277 ± 0.0076 μs	81.4 ± 82
saxpy/default/Float32/65536	0.0323 ± 0.017 ms	3.59 ± 0.57 μs	8.99 ± 5
saxpy/default/Float64/1024	21.3 ± 22 μs	0.328 ± 0.0036 μs	65.1 ± 66
saxpy/default/Float64/1048576	0.214 ± 0.023 ms	0.284 ± 0.035 ms	0.753 ± 0.13
saxpy/default/Float64/16384	25.4 ± 16 μs	1.77 ± 0.031 μs	14.4 ± 8.8
saxpy/default/Float64/2048	20.9 ± 17 μs	0.472 ± 0.021 μs	44.2 ± 36
saxpy/default/Float64/256	21.4 ± 22 μs	0.289 ± 0.0056 μs	74.1 ± 77
saxpy/default/Float64/262144	0.0764 ± 0.016 ms	0.0647 ± 0.0025 ms	1.18 ± 0.25
saxpy/default/Float64/32768	27.8 ± 16 μs	3.56 ± 0.61 μs	7.81 ± 4.6
saxpy/default/Float64/4096	22.2 ± 18 μs	0.785 ± 0.0098 μs	28.2 ± 23
saxpy/default/Float64/512	20.5 ± 21 μs	0.311 ± 0.0037 μs	65.9 ± 68
saxpy/default/Float64/64	20 ± 22 μs	0.271 ± 0.019 μs	73.9 ± 81
saxpy/default/Float64/65536	0.0326 ± 0.016 ms	13 ± 1.3 μs	2.51 ± 1.3
saxpy/static workgroup=(1024,)/Float16/1024	25.1 ± 22 μs	1.2 ± 0.012 μs	20.9 ± 19
saxpy/static workgroup=(1024,)/Float16/1048576	0.199 ± 0.012 ms	0.0861 ± 0.007 ms	2.31 ± 0.23
saxpy/static workgroup=(1024,)/Float16/16384	27.1 ± 16 μs	2.35 ± 0.012 μs	11.5 ± 7
saxpy/static workgroup=(1024,)/Float16/2048	23.1 ± 21 μs	1.29 ± 0.018 μs	17.9 ± 16
saxpy/static workgroup=(1024,)/Float16/256	29.7 ± 22 μs	1.55 ± 0.015 μs	19.1 ± 14
saxpy/static workgroup=(1024,)/Float16/262144	0.0675 ± 0.016 ms	20.8 ± 0.51 μs	3.24 ± 0.77
saxpy/static workgroup=(1024,)/Float16/32768	29 ± 17 μs	3.47 ± 0.018 μs	8.36 ± 5
saxpy/static workgroup=(1024,)/Float16/4096	23.3 ± 15 μs	1.46 ± 0.022 μs	16 ± 10
saxpy/static workgroup=(1024,)/Float16/512	22.2 ± 22 μs	1.72 ± 0.015 μs	12.9 ± 13
saxpy/static workgroup=(1024,)/Float16/64	23.1 ± 22 μs	1.41 ± 0.032 μs	16.4 ± 16
saxpy/static workgroup=(1024,)/Float16/65536	0.0334 ± 0.017 ms	5.73 ± 0.032 μs	5.82 ± 3
saxpy/static workgroup=(1024,)/Float32/1024	21.5 ± 22 μs	1.14 ± 0.02 μs	18.9 ± 20
saxpy/static workgroup=(1024,)/Float32/1048576	0.181 ± 0.028 ms	0.133 ± 0.0073 ms	1.36 ± 0.22
saxpy/static workgroup=(1024,)/Float32/16384	25.7 ± 14 μs	2.09 ± 0.025 μs	12.3 ± 6.6
saxpy/static workgroup=(1024,)/Float32/2048	22.4 ± 21 μs	1.17 ± 0.026 μs	19.2 ± 18
saxpy/static workgroup=(1024,)/Float32/256	27 ± 22 μs	1.47 ± 0.018 μs	18.4 ± 15
saxpy/static workgroup=(1024,)/Float32/262144	0.0594 ± 0.019 ms	30.9 ± 1.4 μs	1.92 ± 0.62
saxpy/static workgroup=(1024,)/Float32/32768	27.4 ± 16 μs	2.61 ± 0.034 μs	10.5 ± 6.2
saxpy/static workgroup=(1024,)/Float32/4096	23.4 ± 16 μs	1.32 ± 0.034 μs	17.8 ± 12
saxpy/static workgroup=(1024,)/Float32/512	25.1 ± 22 μs	1.52 ± 0.015 μs	16.5 ± 14
saxpy/static workgroup=(1024,)/Float32/64	22.3 ± 23 μs	1.44 ± 0.028 μs	15.4 ± 16
saxpy/static workgroup=(1024,)/Float32/65536	0.0319 ± 0.018 ms	4.86 ± 0.82 μs	6.57 ± 4
saxpy/static workgroup=(1024,)/Float64/1024	21.6 ± 23 μs	1.22 ± 0.023 μs	17.7 ± 19
saxpy/static workgroup=(1024,)/Float64/1048576	0.213 ± 0.032 ms	0.29 ± 0.031 ms	0.735 ± 0.13
saxpy/static workgroup=(1024,)/Float64/16384	24.9 ± 15 μs	2.63 ± 0.033 μs	9.47 ± 5.8
saxpy/static workgroup=(1024,)/Float64/2048	21.1 ± 18 μs	1.36 ± 0.033 μs	15.6 ± 13
saxpy/static workgroup=(1024,)/Float64/256	0.0334 ± 0.021 ms	1.61 ± 0.015 μs	20.7 ± 13
saxpy/static workgroup=(1024,)/Float64/262144	0.0732 ± 0.016 ms	0.0654 ± 0.0023 ms	1.12 ± 0.25
saxpy/static workgroup=(1024,)/Float64/32768	27.2 ± 15 μs	5.03 ± 0.94 μs	5.41 ± 3.2
saxpy/static workgroup=(1024,)/Float64/4096	22.5 ± 19 μs	1.65 ± 0.1 μs	13.6 ± 11
saxpy/static workgroup=(1024,)/Float64/512	26.7 ± 22 μs	1.62 ± 0.023 μs	16.4 ± 14
saxpy/static workgroup=(1024,)/Float64/64	22 ± 22 μs	1.58 ± 0.016 μs	13.9 ± 14
saxpy/static workgroup=(1024,)/Float64/65536	0.0322 ± 0.017 ms	14.8 ± 1.2 μs	2.18 ± 1.2
time_to_load	0.494 ± 0.0038 s	0.15 ± 0.00042 s	3.28 ± 0.027

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).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.69%. Comparing base (49dee88) to head (589d21d).
⚠️ Report is 1 commits behind head on release-0.9.

Additional details and impacted files

@@               Coverage Diff               @@
##           release-0.9     #794      +/-   ##
===============================================
+ Coverage        79.38%   79.69%   +0.31%     
===============================================
  Files               14       14              
  Lines              907      911       +4     
===============================================
+ Hits               720      726       +6     
+ Misses             187      185       -2     

☔ 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.