Avoid relying on a RNG in the back-end.

[maleadt]

GPUArrays only provides Random.rand!(::RNG, ::AnyGPUArray), not Random.rand!(::AnyGPUArray), so the examples only work when the back-end provides a native RNG. Avoid that by copying random memory from the CPU.

Ran into this when trying to move out the cuRAND stuff from CUDA.jl.

[github-actions]

Benchmark Results

	main	469aa70...	main / 469aa70...
saxpy/default/Float32/1024	0.082 ± 0.01 ms	0.641 ± 0.0092 μs	128 ± 16
saxpy/default/Float32/1048576	0.457 ± 0.021 ms	0.248 ± 0.015 ms	1.84 ± 0.14
saxpy/default/Float32/16384	0.0759 ± 0.031 ms	2.78 ± 0.15 μs	27.4 ± 11
saxpy/default/Float32/2048	0.0821 ± 0.026 ms	0.752 ± 0.066 μs	109 ± 36
saxpy/default/Float32/256	0.082 ± 0.013 ms	0.573 ± 0.0063 μs	143 ± 23
saxpy/default/Float32/262144	0.174 ± 0.03 ms	0.0635 ± 0.0041 ms	2.74 ± 0.5
saxpy/default/Float32/32768	0.0755 ± 0.031 ms	5.31 ± 0.28 μs	14.2 ± 5.9
saxpy/default/Float32/4096	0.0782 ± 0.033 ms	1.14 ± 0.096 μs	68.4 ± 29
saxpy/default/Float32/512	0.0816 ± 0.012 ms	0.61 ± 0.0057 μs	134 ± 20
saxpy/default/Float32/64	0.0821 ± 0.009 ms	0.563 ± 0.0058 μs	146 ± 16
saxpy/default/Float32/65536	0.0922 ± 0.031 ms	13.3 ± 1.2 μs	6.94 ± 2.4
saxpy/default/Float64/1024	0.0815 ± 0.022 ms	0.774 ± 0.064 μs	105 ± 30
saxpy/default/Float64/1048576	0.577 ± 0.086 ms	0.512 ± 0.044 ms	1.13 ± 0.19
saxpy/default/Float64/16384	0.0818 ± 0.031 ms	5.31 ± 0.29 μs	15.4 ± 5.9
saxpy/default/Float64/2048	0.0827 ± 0.027 ms	1.15 ± 0.092 μs	71.6 ± 24
saxpy/default/Float64/256	0.0813 ± 0.021 ms	0.59 ± 0.0072 μs	138 ± 35
saxpy/default/Float64/262144	0.199 ± 0.033 ms	0.123 ± 0.0086 ms	1.62 ± 0.29
saxpy/default/Float64/32768	0.0905 ± 0.03 ms	13.2 ± 1.3 μs	6.88 ± 2.4
saxpy/default/Float64/4096	0.0802 ± 0.031 ms	1.71 ± 0.11 μs	47 ± 18
saxpy/default/Float64/512	0.0813 ± 0.012 ms	0.646 ± 0.0097 μs	126 ± 18
saxpy/default/Float64/64	0.082 ± 0.013 ms	0.569 ± 0.0058 μs	144 ± 23
saxpy/default/Float64/65536	0.107 ± 0.031 ms	31.5 ± 2.9 μs	3.39 ± 1
saxpy/static workgroup=(1024,)/Float32/1024	0.0796 ± 0.0085 ms	2.1 ± 0.029 μs	37.9 ± 4.1
saxpy/static workgroup=(1024,)/Float32/1048576	0.46 ± 0.021 ms	0.251 ± 0.018 ms	1.84 ± 0.16
saxpy/static workgroup=(1024,)/Float32/16384	0.0764 ± 0.031 ms	4.22 ± 0.2 μs	18.1 ± 7.3
saxpy/static workgroup=(1024,)/Float32/2048	0.0799 ± 0.024 ms	2.25 ± 0.072 μs	35.4 ± 11
saxpy/static workgroup=(1024,)/Float32/256	0.0803 ± 0.016 ms	2.69 ± 0.019 μs	29.8 ± 5.9
saxpy/static workgroup=(1024,)/Float32/262144	0.172 ± 0.031 ms	0.0658 ± 0.0044 ms	2.61 ± 0.5
saxpy/static workgroup=(1024,)/Float32/32768	0.0736 ± 0.031 ms	7.23 ± 0.35 μs	10.2 ± 4.3
saxpy/static workgroup=(1024,)/Float32/4096	0.0788 ± 0.031 ms	2.56 ± 0.086 μs	30.8 ± 12
saxpy/static workgroup=(1024,)/Float32/512	0.0797 ± 0.01 ms	2.8 ± 0.028 μs	28.5 ± 3.7
saxpy/static workgroup=(1024,)/Float32/64	0.0801 ± 0.011 ms	2.63 ± 0.019 μs	30.5 ± 4.1
saxpy/static workgroup=(1024,)/Float32/65536	0.095 ± 0.032 ms	15.9 ± 1.3 μs	5.96 ± 2.1
saxpy/static workgroup=(1024,)/Float64/1024	0.0786 ± 0.022 ms	2.39 ± 0.076 μs	32.9 ± 9.4
saxpy/static workgroup=(1024,)/Float64/1048576	0.534 ± 0.079 ms	0.526 ± 0.065 ms	1.01 ± 0.2
saxpy/static workgroup=(1024,)/Float64/16384	0.0785 ± 0.03 ms	7.55 ± 0.48 μs	10.4 ± 4.1
saxpy/static workgroup=(1024,)/Float64/2048	0.0798 ± 0.026 ms	2.7 ± 0.09 μs	29.6 ± 9.8
saxpy/static workgroup=(1024,)/Float64/256	0.0797 ± 0.016 ms	2.84 ± 0.026 μs	28.1 ± 5.6
saxpy/static workgroup=(1024,)/Float64/262144	0.2 ± 0.032 ms	0.127 ± 0.0086 ms	1.58 ± 0.27
saxpy/static workgroup=(1024,)/Float64/32768	0.0823 ± 0.03 ms	16.8 ± 1.7 μs	4.91 ± 1.9
saxpy/static workgroup=(1024,)/Float64/4096	0.0786 ± 0.031 ms	3.22 ± 0.12 μs	24.4 ± 9.7
saxpy/static workgroup=(1024,)/Float64/512	0.0791 ± 0.014 ms	2.95 ± 0.031 μs	26.8 ± 4.9
saxpy/static workgroup=(1024,)/Float64/64	0.0796 ± 0.012 ms	2.76 ± 0.021 μs	28.8 ± 4.3
saxpy/static workgroup=(1024,)/Float64/65536	0.104 ± 0.032 ms	0.0349 ± 0.0028 ms	2.97 ± 0.95
time_to_load	0.97 ± 0.0043 s	0.284 ± 0.0011 s	3.42 ± 0.02

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 71.85%. Comparing base (4224ad9) to head (469aa70).
⚠️ Report is 2 commits behind head on release-0.9.

Additional details and impacted files

@@             Coverage Diff              @@
##           release-0.9     #686   +/-   ##
============================================
  Coverage        71.85%   71.85%           
============================================
  Files               14       14           
  Lines              906      906           
============================================
  Hits               651      651           
  Misses             255      255           

☔ View full report in Codecov by Sentry.
📢 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.