[maleadt]

GPUCompiler 2.15 (JuliaGPU/GPUCompiler.jl#963) lowers atomics on local memory to plain loads and stores instead of rejecting them, so the atomic operation in address space 5 test in test/core/device/intrinsics.jl now fails. It breaks GPUCompiler's downstream CUDA.jl job on main and on every PR.

The test dates from when CUDA.jl's atomic functions asserted the address space themselves. Since #3350 it only re-checks a GPUCompiler rule, which GPUCompiler covers in its own tests. The atomics testset (test/core/device/intrinsics/atomics.jl) already checks that GPUCompiler's atomic validation errors surface through CUDA.jl, with a backtrace into the kernel, for every atomic function and for UnsafeAtomics. So this removes the test rather than moving it to another invalid address space. Removing it works across the whole GPUCompiler = "2.13" compat range.

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 1bab36b	Previous: d906dd4	Ratio
array/accumulate/Float32/1d	54371 ns	54610 ns	1.00
array/accumulate/Float32/dims=1	39176 ns	39595 ns	0.99
array/accumulate/Float32/dims=1L	82211 ns	82179 ns	1.00
array/accumulate/Float32/dims=2	49577 ns	49690 ns	1.00
array/accumulate/Float32/dims=2L	397854 ns	399898 ns	0.99
array/accumulate/Int64/1d	88457 ns	88538 ns	1.00
array/accumulate/Int64/dims=1	59652 ns	59894 ns	1.00
array/accumulate/Int64/dims=1L	139878 ns	139741 ns	1.00
array/accumulate/Int64/dims=2	68854 ns	68697 ns	1.00
array/accumulate/Int64/dims=2L	797054 ns	795555 ns	1.00
array/broadcast	10180 ns	10266 ns	0.99
array/broadcast launch	6642.4 ns	6671.8 ns	1.00
array/construct	969.3846153846154 ns	970.4285714285714 ns	1.00
array/copy	16973 ns	16671 ns	1.02
array/copyto!/cpu_to_gpu	208622 ns	208691 ns	1.00
array/copyto!/gpu_to_cpu	242680 ns	241416 ns	1.01
array/copyto!/gpu_to_gpu	8878 ns	8901 ns	1.00
array/iteration/findall/bool	49988 ns	50292 ns	0.99
array/iteration/findall/int	66209 ns	65694 ns	1.01
array/iteration/findfirst/bool	33850 ns	34156 ns	0.99
array/iteration/findfirst/int	33668 ns	34135 ns	0.99
array/iteration/findmin/1d	51279 ns	51295 ns	1.00
array/iteration/findmin/2d	51932 ns	51568 ns	1.01
array/iteration/logical	50683 ns	51191 ns	0.99
array/iteration/scalar	60745 ns	60491 ns	1.00
array/permutedims/2d	47148 ns	47182 ns	1.00
array/permutedims/3d	44647 ns	44299 ns	1.01
array/permutedims/4d	47749 ns	47627 ns	1.00
array/random/rand/Float32	11024 ns	10941 ns	1.01
array/random/rand/Int64	17818 ns	17362 ns	1.03
array/random/rand!/Float32	8118.666666666667 ns	8365.666666666666 ns	0.97
array/random/rand!/Int64	15339 ns	15492 ns	0.99
array/random/randn/Float32	34201 ns	34079 ns	1.00
array/random/randn!/Float32	24934 ns	24936 ns	1.00
array/reductions/mapreduce/Float32/1d	34522 ns	34206 ns	1.01
array/reductions/mapreduce/Float32/dims=1	18603 ns	18940 ns	0.98
array/reductions/mapreduce/Float32/dims=1L	56680 ns	56520 ns	1.00
array/reductions/mapreduce/Float32/dims=2	27466 ns	26651 ns	1.03
array/reductions/mapreduce/Float32/dims=2L	56884 ns	57077 ns	1.00
array/reductions/mapreduce/Int64/1d	34820 ns	34513 ns	1.01
array/reductions/mapreduce/Int64/dims=1	21812 ns	21363 ns	1.02
array/reductions/mapreduce/Int64/dims=1L	95968 ns	95697 ns	1.00
array/reductions/mapreduce/Int64/dims=2	27751 ns	27710 ns	1.00
array/reductions/mapreduce/Int64/dims=2L	82624 ns	83260 ns	0.99
array/reductions/reduce/Float32/1d	34654 ns	34650 ns	1.00
array/reductions/reduce/Float32/dims=1	18967 ns	18868 ns	1.01
array/reductions/reduce/Float32/dims=1L	56900 ns	56506 ns	1.01
array/reductions/reduce/Float32/dims=2	26973 ns	26736 ns	1.01
array/reductions/reduce/Float32/dims=2L	56927 ns	57352 ns	0.99
array/reductions/reduce/Int64/1d	34876 ns	34936 ns	1.00
array/reductions/reduce/Int64/dims=1	21623 ns	21357 ns	1.01
array/reductions/reduce/Int64/dims=1L	96157 ns	96315 ns	1.00
array/reductions/reduce/Int64/dims=2	27829 ns	27507 ns	1.01
array/reductions/reduce/Int64/dims=2L	83930 ns	82585 ns	1.02
array/reverse/1d	19575 ns	19510 ns	1.00
array/reverse/1dL	73567 ns	73486 ns	1.00
array/reverse/1dL_inplace	70072 ns	70289 ns	1.00
array/reverse/1d_inplace	11863 ns	11818 ns	1.00
array/reverse/2d	22159 ns	22108 ns	1.00
array/reverse/2dL	80486 ns	80797 ns	1.00
array/reverse/2dL_inplace	70691 ns	70790 ns	1.00
array/reverse/2d_inplace	14559 ns	14626 ns	1.00
array/sorting/1d	750080 ns	751400 ns	1.00
array/sorting/2d	194121 ns	194606 ns	1.00
array/sorting/by	928499 ns	928713 ns	1.00
cuda/graph/capture	37391 ns	35510 ns	1.05
cuda/graph/eager	33632 ns	32786 ns	1.03
cuda/graph/launch	13857 ns	13875 ns	1.00
cuda/synchronization/context/auto	2386.4444444444443 ns	2310.1111111111113 ns	1.03
cuda/synchronization/context/blocking	883.64 ns	883.9322033898305 ns	1.00
cuda/synchronization/context/nonblocking	8084 ns	8373 ns	0.97
cuda/synchronization/stream/auto	871.7402597402597 ns	826.3561643835617 ns	1.05
cuda/synchronization/stream/blocking	760.9203539823009 ns	760.0223880597015 ns	1.00
cuda/synchronization/stream/nonblocking	7696 ns	7836 ns	0.98
integration/byval/reference	147595 ns	147475 ns	1.00
integration/byval/slices=1	148801 ns	148515 ns	1.00
integration/byval/slices=2	290423 ns	290141 ns	1.00
integration/byval/slices=3	432462 ns	432030 ns	1.00
integration/cudadevrt	105089 ns	104862 ns	1.00
integration/volumerhs	9608714 ns	9608642 ns	1.00
kernel/indexing	11275 ns	11291 ns	1.00
kernel/indexing_checked	11940 ns	11884 ns	1.00
kernel/launch	2328.6666666666665 ns	2315.777777777778 ns	1.01
kernel/occupancy	704.4133333333333 ns	678.1192052980133 ns	1.04
kernel/rand	15088 ns	14928 ns	1.01
latency/import	4405853532 ns	4784183934 ns	0.92
latency/precompile	5368981855 ns	5431409471 ns	0.99
latency/ttfp	5233688297 ns	5408854891 ns	0.97

This comment was automatically generated by workflow using github-action-benchmark.