Tests for host interface functions

[christiangnrd]

supports_atomics in Metal.jl is broken and I realized we should probably test the interface definitions

All Claude and I reviewed

[github-actions]

Benchmark Results

Show table

	main	1fc4a11...	main / 1fc4a11...
saxpy/default/Float32/1024	26.4 ± 19 μs	30.7 ± 22 μs	0.858 ± 0.87
saxpy/default/Float32/1048576	0.347 ± 0.04 ms	0.348 ± 0.044 ms	0.997 ± 0.17
saxpy/default/Float32/16384	0.0391 ± 0.012 ms	0.0402 ± 0.019 ms	0.973 ± 0.55
saxpy/default/Float32/2048	0.0448 ± 0.021 ms	0.0445 ± 0.021 ms	1.01 ± 0.68
saxpy/default/Float32/256	25.5 ± 7 μs	24.5 ± 4.2 μs	1.04 ± 0.34
saxpy/default/Float32/262144	0.11 ± 0.02 ms	0.108 ± 0.022 ms	1.02 ± 0.27
saxpy/default/Float32/32768	0.0437 ± 0.011 ms	0.0415 ± 0.012 ms	1.05 ± 0.41
saxpy/default/Float32/4096	0.0446 ± 0.021 ms	0.0498 ± 0.022 ms	0.894 ± 0.57
saxpy/default/Float32/512	26.8 ± 16 μs	25.7 ± 12 μs	1.04 ± 0.79
saxpy/default/Float32/64	24.5 ± 3.4 μs	24.3 ± 2.7 μs	1.01 ± 0.18
saxpy/default/Float32/65536	0.0546 ± 0.0059 ms	0.0546 ± 0.006 ms	1 ± 0.15
saxpy/default/Float64/1024	0.0358 ± 0.023 ms	0.0324 ± 0.024 ms	1.11 ± 1.1
saxpy/default/Float64/1048576	0.621 ± 0.085 ms	0.609 ± 0.074 ms	1.02 ± 0.19
saxpy/default/Float64/16384	0.04 ± 0.014 ms	0.0387 ± 0.012 ms	1.03 ± 0.48
saxpy/default/Float64/2048	0.0464 ± 0.024 ms	0.0456 ± 0.023 ms	1.02 ± 0.74
saxpy/default/Float64/256	24 ± 5.3 μs	24.3 ± 5.4 μs	0.987 ± 0.31
saxpy/default/Float64/262144	0.181 ± 0.033 ms	0.175 ± 0.031 ms	1.04 ± 0.26
saxpy/default/Float64/32768	0.053 ± 0.007 ms	0.0523 ± 0.0069 ms	1.01 ± 0.19
saxpy/default/Float64/4096	0.0511 ± 0.02 ms	0.0421 ± 0.021 ms	1.21 ± 0.76
saxpy/default/Float64/512	24.4 ± 7.4 μs	25.4 ± 10 μs	0.96 ± 0.48
saxpy/default/Float64/64	24.6 ± 3.8 μs	24.5 ± 3.1 μs	1 ± 0.2
saxpy/default/Float64/65536	0.0673 ± 0.011 ms	0.066 ± 0.0099 ms	1.02 ± 0.23
saxpy/static workgroup=(1024,)/Float32/1024	31.4 ± 19 μs	0.0341 ± 0.02 ms	0.921 ± 0.77
saxpy/static workgroup=(1024,)/Float32/1048576	0.464 ± 0.068 ms	0.442 ± 0.071 ms	1.05 ± 0.23
saxpy/static workgroup=(1024,)/Float32/16384	0.0455 ± 0.013 ms	0.046 ± 0.0095 ms	0.989 ± 0.35
saxpy/static workgroup=(1024,)/Float32/2048	0.044 ± 0.019 ms	0.0432 ± 0.02 ms	1.02 ± 0.65
saxpy/static workgroup=(1024,)/Float32/256	29.5 ± 16 μs	31.1 ± 17 μs	0.949 ± 0.72
saxpy/static workgroup=(1024,)/Float32/262144	0.149 ± 0.02 ms	0.149 ± 0.024 ms	0.998 ± 0.21
saxpy/static workgroup=(1024,)/Float32/32768	0.051 ± 0.0053 ms	0.0536 ± 0.0061 ms	0.952 ± 0.15
saxpy/static workgroup=(1024,)/Float32/4096	0.0491 ± 0.017 ms	0.0489 ± 0.017 ms	1.01 ± 0.49
saxpy/static workgroup=(1024,)/Float32/512	30.5 ± 16 μs	29.7 ± 17 μs	1.03 ± 0.81
saxpy/static workgroup=(1024,)/Float32/64	29.2 ± 14 μs	29.1 ± 11 μs	1 ± 0.62
saxpy/static workgroup=(1024,)/Float32/65536	0.0623 ± 0.0055 ms	0.0631 ± 0.006 ms	0.987 ± 0.13
saxpy/static workgroup=(1024,)/Float64/1024	0.0336 ± 0.021 ms	0.0327 ± 0.02 ms	1.03 ± 0.89
saxpy/static workgroup=(1024,)/Float64/1048576	0.688 ± 0.078 ms	0.679 ± 0.097 ms	1.01 ± 0.18
saxpy/static workgroup=(1024,)/Float64/16384	0.0452 ± 0.01 ms	0.0436 ± 0.012 ms	1.04 ± 0.37
saxpy/static workgroup=(1024,)/Float64/2048	0.0461 ± 0.023 ms	0.0444 ± 0.022 ms	1.04 ± 0.73
saxpy/static workgroup=(1024,)/Float64/256	29.4 ± 15 μs	29.4 ± 12 μs	0.997 ± 0.65
saxpy/static workgroup=(1024,)/Float64/262144	0.196 ± 0.02 ms	0.188 ± 0.022 ms	1.04 ± 0.16
saxpy/static workgroup=(1024,)/Float64/32768	0.0548 ± 0.0059 ms	0.0549 ± 0.0057 ms	0.997 ± 0.15
saxpy/static workgroup=(1024,)/Float64/4096	0.0389 ± 0.011 ms	0.0491 ± 0.018 ms	0.792 ± 0.36
saxpy/static workgroup=(1024,)/Float64/512	28.9 ± 14 μs	29.1 ± 14 μs	0.996 ± 0.69
saxpy/static workgroup=(1024,)/Float64/64	29.4 ± 15 μs	28.9 ± 11 μs	1.02 ± 0.67
saxpy/static workgroup=(1024,)/Float64/65536	0.0715 ± 0.01 ms	0.072 ± 0.011 ms	0.992 ± 0.2
time_to_load	0.973 ± 0.05 s	0.907 ± 0.012 s	1.07 ± 0.057

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).

[christiangnrd]

Metal failure is fixed by JuliaGPU/Metal.jl#894.

oneAPI and CUDA fail because their KA.device! implementations return a device instead of nothing but that's not specified in the interface so probably just remove the tests?

[christiangnrd]

Similarly, priority! in AMDGPU returns the symbol instead of nothing, but that's not specified in the interface

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 64.82%. Comparing base (65690fd) to head (1fc4a11).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #739      +/-   ##
==========================================
+ Coverage   64.77%   64.82%   +0.04%     
==========================================
  Files          24       24              
  Lines        2024     2024              
==========================================
+ Hits         1311     1312       +1     
+ Misses        713      712       -1     

☔ 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.