support multiple devices per backend

[vchuravy]

Fixes #458

Minimal API proposal. Should we have a query for if the backends implement this?

[github-actions]

Benchmark Results

	main	211a658...	main/211a658eb6ce3d...
saxpy/default/Float16/1024	0.736 ± 0.0073 μs	0.731 ± 0.0069 μs	1.01
saxpy/default/Float16/1048576	0.174 ± 0.0078 ms	0.181 ± 0.014 ms	0.964
saxpy/default/Float16/16384	3.34 ± 0.025 μs	3.33 ± 0.046 μs	1
saxpy/default/Float16/2048	0.912 ± 0.012 μs	0.905 ± 0.011 μs	1.01
saxpy/default/Float16/256	0.588 ± 0.0075 μs	0.58 ± 0.0076 μs	1.01
saxpy/default/Float16/262144	0.0441 ± 0.00058 ms	0.0445 ± 0.0012 ms	0.991
saxpy/default/Float16/32768	6.02 ± 0.05 μs	6.02 ± 0.092 μs	1
saxpy/default/Float16/4096	1.31 ± 0.028 μs	1.29 ± 0.026 μs	1.01
saxpy/default/Float16/512	0.649 ± 0.0076 μs	0.641 ± 0.007 μs	1.01
saxpy/default/Float16/64	0.553 ± 0.005 μs	0.546 ± 0.0057 μs	1.01
saxpy/default/Float16/65536	11.7 ± 0.11 μs	11.6 ± 0.18 μs	1
saxpy/default/Float32/1024	0.652 ± 0.0097 μs	0.639 ± 0.0098 μs	1.02
saxpy/default/Float32/1048576	0.231 ± 0.019 ms	0.245 ± 0.034 ms	0.943
saxpy/default/Float32/16384	2.75 ± 0.16 μs	2.79 ± 0.25 μs	0.984
saxpy/default/Float32/2048	0.757 ± 0.064 μs	0.751 ± 0.015 μs	1.01
saxpy/default/Float32/256	0.578 ± 0.0063 μs	0.57 ± 0.0054 μs	1.01
saxpy/default/Float32/262144	0.0557 ± 0.0086 ms	0.0462 ± 0.0048 ms	1.2
saxpy/default/Float32/32768	5.24 ± 0.33 μs	5.38 ± 1.3 μs	0.973
saxpy/default/Float32/4096	1.13 ± 0.082 μs	1.11 ± 0.075 μs	1.01
saxpy/default/Float32/512	0.618 ± 0.0072 μs	0.609 ± 0.007 μs	1.01
saxpy/default/Float32/64	0.565 ± 0.0055 μs	0.561 ± 0.0052 μs	1.01
saxpy/default/Float32/65536	12.9 ± 1.4 μs	12.6 ± 1.7 μs	1.03
saxpy/default/Float64/1024	0.757 ± 0.064 μs	0.745 ± 0.022 μs	1.02
saxpy/default/Float64/1048576	0.503 ± 0.04 ms	0.543 ± 0.057 ms	0.926
saxpy/default/Float64/16384	5.25 ± 0.33 μs	5.48 ± 1.4 μs	0.958
saxpy/default/Float64/2048	1.13 ± 0.081 μs	1.12 ± 0.074 μs	1.01
saxpy/default/Float64/256	0.579 ± 0.0063 μs	0.577 ± 0.0067 μs	1
saxpy/default/Float64/262144	0.115 ± 0.01 ms	0.116 ± 0.012 ms	0.991
saxpy/default/Float64/32768	12.6 ± 0.86 μs	12.9 ± 1.4 μs	0.976
saxpy/default/Float64/4096	1.68 ± 0.1 μs	1.74 ± 0.28 μs	0.967
saxpy/default/Float64/512	0.631 ± 0.0094 μs	0.629 ± 0.011 μs	1
saxpy/default/Float64/64	0.557 ± 0.0055 μs	0.557 ± 0.0069 μs	1
saxpy/default/Float64/65536	28.9 ± 2.4 μs	28.9 ± 2.1 μs	1
saxpy/static workgroup=(1024,)/Float16/1024	2.18 ± 0.029 μs	2.19 ± 0.03 μs	0.992
saxpy/static workgroup=(1024,)/Float16/1048576	0.165 ± 0.011 ms	0.174 ± 0.017 ms	0.951
saxpy/static workgroup=(1024,)/Float16/16384	4.41 ± 0.052 μs	4.41 ± 0.067 μs	0.998
saxpy/static workgroup=(1024,)/Float16/2048	2.35 ± 0.032 μs	2.36 ± 0.033 μs	0.996
saxpy/static workgroup=(1024,)/Float16/256	2.81 ± 0.039 μs	2.82 ± 0.045 μs	0.994
saxpy/static workgroup=(1024,)/Float16/262144	0.0441 ± 0.0034 ms	0.0435 ± 0.0023 ms	1.01
saxpy/static workgroup=(1024,)/Float16/32768	6.83 ± 0.14 μs	6.82 ± 0.15 μs	1
saxpy/static workgroup=(1024,)/Float16/4096	2.68 ± 0.041 μs	2.68 ± 0.042 μs	0.999
saxpy/static workgroup=(1024,)/Float16/512	3.25 ± 0.041 μs	3.27 ± 0.041 μs	0.994
saxpy/static workgroup=(1024,)/Float16/64	2.51 ± 0.22 μs	2.52 ± 0.21 μs	0.996
saxpy/static workgroup=(1024,)/Float16/65536	12.4 ± 0.28 μs	12.5 ± 0.33 μs	0.995
saxpy/static workgroup=(1024,)/Float32/1024	2.21 ± 0.041 μs	2.21 ± 0.041 μs	1
saxpy/static workgroup=(1024,)/Float32/1048576	0.239 ± 0.024 ms	0.226 ± 0.03 ms	1.06
saxpy/static workgroup=(1024,)/Float32/16384	4.34 ± 0.2 μs	4.31 ± 0.19 μs	1.01
saxpy/static workgroup=(1024,)/Float32/2048	2.37 ± 0.052 μs	2.37 ± 0.05 μs	0.999
saxpy/static workgroup=(1024,)/Float32/256	2.69 ± 0.062 μs	2.69 ± 0.048 μs	1
saxpy/static workgroup=(1024,)/Float32/262144	0.0599 ± 0.0047 ms	0.0593 ± 0.0041 ms	1.01
saxpy/static workgroup=(1024,)/Float32/32768	7.27 ± 0.34 μs	7.31 ± 0.38 μs	0.995
saxpy/static workgroup=(1024,)/Float32/4096	2.67 ± 0.077 μs	2.64 ± 0.063 μs	1.01
saxpy/static workgroup=(1024,)/Float32/512	2.72 ± 0.081 μs	2.73 ± 0.095 μs	0.997
saxpy/static workgroup=(1024,)/Float32/64	2.72 ± 5.4 μs	2.71 ± 5.5 μs	1
saxpy/static workgroup=(1024,)/Float32/65536	17 ± 1.6 μs	16 ± 1.7 μs	1.06
saxpy/static workgroup=(1024,)/Float64/1024	2.34 ± 0.073 μs	2.37 ± 0.077 μs	0.986
saxpy/static workgroup=(1024,)/Float64/1048576	0.547 ± 0.075 ms	0.582 ± 0.07 ms	0.94
saxpy/static workgroup=(1024,)/Float64/16384	7.34 ± 0.35 μs	7.31 ± 0.42 μs	1
saxpy/static workgroup=(1024,)/Float64/2048	2.62 ± 0.082 μs	2.64 ± 0.072 μs	0.995
saxpy/static workgroup=(1024,)/Float64/256	2.71 ± 0.081 μs	2.72 ± 0.082 μs	0.997
saxpy/static workgroup=(1024,)/Float64/262144	0.11 ± 0.017 ms	0.119 ± 0.012 ms	0.929
saxpy/static workgroup=(1024,)/Float64/32768	16.2 ± 1.7 μs	15.8 ± 1.6 μs	1.02
saxpy/static workgroup=(1024,)/Float64/4096	3.14 ± 0.093 μs	3.16 ± 0.12 μs	0.992
saxpy/static workgroup=(1024,)/Float64/512	2.72 ± 0.088 μs	2.71 ± 0.077 μs	1.01
saxpy/static workgroup=(1024,)/Float64/64	2.65 ± 0.078 μs	2.65 ± 0.063 μs	1
saxpy/static workgroup=(1024,)/Float64/65536	31.4 ± 2.4 μs	0.0319 ± 0.0028 ms	0.983
time_to_load	0.328 ± 0.00099 s	0.33 ± 0.0033 s	0.994

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).