[0.9] Forbid divergent execution of work-group barriers

[vchuravy]

#558 without the POCL backend, so hopefully we can have that as a non-breaking change.

[vchuravy]

• [0.9] Forbid divergent execution of work-group barriers #564 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[github-actions]

Benchmark Results

	main	993ef0b...	main/993ef0bf315b2a...
saxpy/default/Float16/1024	0.737 ± 0.0067 μs	0.733 ± 0.0077 μs	1.01
saxpy/default/Float16/1048576	0.177 ± 0.0083 ms	0.173 ± 0.0063 ms	1.02
saxpy/default/Float16/16384	3.33 ± 0.025 μs	3.32 ± 0.031 μs	1
saxpy/default/Float16/2048	0.916 ± 0.011 μs	0.908 ± 0.012 μs	1.01
saxpy/default/Float16/256	0.597 ± 0.0055 μs	0.588 ± 0.0072 μs	1.01
saxpy/default/Float16/262144	0.0439 ± 0.0004 ms	0.0438 ± 0.0003 ms	1
saxpy/default/Float16/32768	6.01 ± 0.052 μs	6.03 ± 0.095 μs	0.997
saxpy/default/Float16/4096	1.31 ± 0.023 μs	1.3 ± 0.025 μs	1.01
saxpy/default/Float16/512	0.651 ± 0.0063 μs	0.649 ± 0.0065 μs	1
saxpy/default/Float16/64	0.561 ± 0.0048 μs	0.559 ± 0.0058 μs	1
saxpy/default/Float16/65536	11.6 ± 0.1 μs	11.6 ± 0.15 μs	1
saxpy/default/Float32/1024	0.647 ± 0.011 μs	0.634 ± 0.01 μs	1.02
saxpy/default/Float32/1048576	0.195 ± 0.035 ms	0.227 ± 0.026 ms	0.86
saxpy/default/Float32/16384	2.76 ± 0.14 μs	2.81 ± 0.6 μs	0.981
saxpy/default/Float32/2048	0.76 ± 0.065 μs	0.756 ± 0.026 μs	1.01
saxpy/default/Float32/256	0.583 ± 0.0065 μs	0.562 ± 0.017 μs	1.04
saxpy/default/Float32/262144	0.0559 ± 0.0042 ms	0.0574 ± 0.0047 ms	0.974
saxpy/default/Float32/32768	5.28 ± 0.26 μs	5.39 ± 1 μs	0.98
saxpy/default/Float32/4096	1.13 ± 0.093 μs	1.11 ± 0.042 μs	1.01
saxpy/default/Float32/512	0.614 ± 0.0077 μs	0.596 ± 0.0092 μs	1.03
saxpy/default/Float32/64	0.568 ± 0.0046 μs	0.547 ± 0.0065 μs	1.04
saxpy/default/Float32/65536	12.5 ± 0.72 μs	12.4 ± 0.96 μs	1.01
saxpy/default/Float64/1024	0.747 ± 0.04 μs	0.743 ± 0.031 μs	1.01
saxpy/default/Float64/1048576	0.474 ± 0.052 ms	0.493 ± 0.026 ms	0.962
saxpy/default/Float64/16384	5.26 ± 0.31 μs	5.37 ± 0.6 μs	0.98
saxpy/default/Float64/2048	1.15 ± 0.097 μs	1.12 ± 0.084 μs	1.03
saxpy/default/Float64/256	0.583 ± 0.0066 μs	0.564 ± 0.011 μs	1.03
saxpy/default/Float64/262144	0.0931 ± 0.0098 ms	0.115 ± 0.0079 ms	0.812
saxpy/default/Float64/32768	11.9 ± 0.91 μs	12.8 ± 1.3 μs	0.931
saxpy/default/Float64/4096	1.69 ± 0.19 μs	1.68 ± 0.13 μs	1.01
saxpy/default/Float64/512	0.634 ± 0.011 μs	0.614 ± 0.013 μs	1.03
saxpy/default/Float64/64	0.559 ± 0.0059 μs	0.546 ± 0.0096 μs	1.02
saxpy/default/Float64/65536	24 ± 2.1 μs	28.7 ± 1.8 μs	0.838
saxpy/static workgroup=(1024,)/Float16/1024	2.19 ± 0.025 μs	2.17 ± 0.027 μs	1.01
saxpy/static workgroup=(1024,)/Float16/1048576	0.157 ± 0.0073 ms	0.158 ± 0.0088 ms	0.999
saxpy/static workgroup=(1024,)/Float16/16384	4.43 ± 0.083 μs	4.41 ± 0.07 μs	1
saxpy/static workgroup=(1024,)/Float16/2048	2.36 ± 0.026 μs	2.35 ± 0.031 μs	1
saxpy/static workgroup=(1024,)/Float16/256	2.85 ± 0.032 μs	2.82 ± 0.033 μs	1.01
saxpy/static workgroup=(1024,)/Float16/262144	0.0418 ± 0.00094 ms	0.0419 ± 0.00082 ms	0.998
saxpy/static workgroup=(1024,)/Float16/32768	6.85 ± 0.18 μs	6.84 ± 0.17 μs	1
saxpy/static workgroup=(1024,)/Float16/4096	2.68 ± 0.039 μs	2.67 ± 0.039 μs	1
saxpy/static workgroup=(1024,)/Float16/512	3.29 ± 0.033 μs	3.26 ± 0.035 μs	1.01
saxpy/static workgroup=(1024,)/Float16/64	2.54 ± 0.21 μs	2.51 ± 0.21 μs	1.01
saxpy/static workgroup=(1024,)/Float16/65536	12.5 ± 0.39 μs	12.4 ± 0.25 μs	1.01
saxpy/static workgroup=(1024,)/Float32/1024	2.23 ± 0.029 μs	2.21 ± 0.036 μs	1.01
saxpy/static workgroup=(1024,)/Float32/1048576	0.239 ± 0.019 ms	0.237 ± 0.02 ms	1.01
saxpy/static workgroup=(1024,)/Float32/16384	4.34 ± 0.23 μs	4.43 ± 0.32 μs	0.978
saxpy/static workgroup=(1024,)/Float32/2048	2.37 ± 0.059 μs	2.38 ± 0.062 μs	0.995
saxpy/static workgroup=(1024,)/Float32/256	2.69 ± 0.053 μs	2.68 ± 0.053 μs	1
saxpy/static workgroup=(1024,)/Float32/262144	0.0605 ± 0.0035 ms	0.0603 ± 0.0038 ms	1
saxpy/static workgroup=(1024,)/Float32/32768	7.39 ± 0.38 μs	7.49 ± 0.49 μs	0.987
saxpy/static workgroup=(1024,)/Float32/4096	2.66 ± 0.079 μs	2.65 ± 0.069 μs	1
saxpy/static workgroup=(1024,)/Float32/512	2.7 ± 0.032 μs	2.69 ± 0.034 μs	1
saxpy/static workgroup=(1024,)/Float32/64	2.73 ± 5.1 μs	2.71 ± 4.5 μs	1.01
saxpy/static workgroup=(1024,)/Float32/65536	15.3 ± 0.98 μs	15.4 ± 1.2 μs	0.991
saxpy/static workgroup=(1024,)/Float64/1024	2.33 ± 0.059 μs	2.32 ± 0.078 μs	1
saxpy/static workgroup=(1024,)/Float64/1048576	0.518 ± 0.029 ms	0.504 ± 0.025 ms	1.03
saxpy/static workgroup=(1024,)/Float64/16384	7.31 ± 0.39 μs	7.31 ± 0.35 μs	1
saxpy/static workgroup=(1024,)/Float64/2048	2.62 ± 0.09 μs	2.6 ± 0.078 μs	1.01
saxpy/static workgroup=(1024,)/Float64/256	2.66 ± 0.058 μs	2.65 ± 0.061 μs	1.01
saxpy/static workgroup=(1024,)/Float64/262144	0.0985 ± 0.0094 ms	0.0948 ± 0.0098 ms	1.04
saxpy/static workgroup=(1024,)/Float64/32768	15.3 ± 1 μs	14.6 ± 1.4 μs	1.05
saxpy/static workgroup=(1024,)/Float64/4096	3.17 ± 0.19 μs	3.16 ± 0.22 μs	1
saxpy/static workgroup=(1024,)/Float64/512	2.67 ± 0.065 μs	2.66 ± 0.071 μs	1
saxpy/static workgroup=(1024,)/Float64/64	2.62 ± 0.065 μs	2.6 ± 0.06 μs	1.01
saxpy/static workgroup=(1024,)/Float64/65536	29.2 ± 4.8 μs	26.7 ± 2.6 μs	1.1
time_to_load	0.316 ± 0.0061 s	0.315 ± 0.0059 s	1

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).