[maleadt]

The launch configuration tests asserted which kind of launch KernelAbstractions picks for a few shapes. That choice is KernelAbstractions' policy, not something the back end controls, and the assertions break whenever it changes: JuliaGPU/KernelAbstractions.jl#919 launches 4-d ranges on the N-d grid, which fails them on every back end.

This keeps the checks of what the back end does affect: that a dynamic 3-d range is indexed without divisions, and that iteration spaces too large for 32-bit indices are indexed correctly. The latter now covers a 4-d range that is launched on the N-d grid with JuliaGPU/KernelAbstractions.jl#919, and one that is only large along its last dimension.

The tests pass with KernelAbstractions main and with JuliaGPU/KernelAbstractions.jl#919, so this can be merged first.

[github-actions]

AMDGPU.jl Benchmarks

Details

Benchmark suite	Current: fef3c64	Previous: 7689e9d	Ratio
amdgpu/synchronization/context/device	1807.5 ns	1795 ns	1.01
amdgpu/synchronization/stream/blocking	227.5 ns	227.5 ns	1
amdgpu/synchronization/stream/nonblocking	347.5 ns	342.5 ns	1.01
applications/bitonic_sort	1117945.25 ns	1112856.25 ns	1.00
applications/convolution	76231.25 ns	76553.5 ns	1.00
applications/floyd_warshall	9073870.75 ns	9077180 ns	1.00
applications/histogram	789972.5 ns	800526.75 ns	0.99
applications/prefix_sum	190653 ns	194390.25 ns	0.98
array/accumulate/Float32/1d	89684 ns	90223.75 ns	0.99
array/accumulate/Float32/dims=1	69748.75 ns	70451 ns	0.99
array/accumulate/Float32/dims=1L	104636.75 ns	104944 ns	1.00
array/accumulate/Float32/dims=2	104344.25 ns	104844 ns	1.00
array/accumulate/Float32/dims=2L	126994.75 ns	127731.75 ns	0.99
array/accumulate/Int64/1d	94044 ns	114961.5 ns	0.82
array/accumulate/Int64/dims=1	83241.25 ns	69886 ns	1.19
array/accumulate/Int64/dims=1L	110929.5 ns	111521.5 ns	0.99
array/accumulate/Int64/dims=2	106139.5 ns	108984 ns	0.97
array/accumulate/Int64/dims=2L	186563 ns	188030.25 ns	0.99
array/broadcast	57431 ns	33030.5 ns	1.74
array/construct	2105.25 ns	2225.25 ns	0.95
array/copy	37568 ns	39400.5 ns	0.95
array/copyto!/cpu_to_gpu	85844 ns	134109.5 ns	0.64
array/copyto!/gpu_to_cpu	85711.5 ns	86758.75 ns	0.99
array/copyto!/gpu_to_gpu	58268.5 ns	58358.25 ns	1.00
array/iteration/findall/bool	93536.5 ns	93526.5 ns	1.00
array/iteration/findall/int	105119 ns	105324 ns	1.00
array/iteration/findfirst/bool	49843.25 ns	55503.5 ns	0.90
array/iteration/findfirst/int	56993.5 ns	55410.75 ns	1.03
array/iteration/findmin/1d	92174 ns	96369 ns	0.96
array/iteration/findmin/2d	67938.75 ns	83236.25 ns	0.82
array/iteration/logical	89603.75 ns	94576.25 ns	0.95
array/iteration/scalar	296424.75 ns	288479 ns	1.03
array/permutedims/2d	61168.5 ns	48830.75 ns	1.25
array/permutedims/3d	58783.75 ns	58641 ns	1.00
array/permutedims/4d	62706 ns	62333.5 ns	1.01
array/random/rand/Float32	46163.25 ns	46288.25 ns	1.00
array/random/rand/Int64	45968.25 ns	45958.25 ns	1.00
array/random/rand!/Float32	49973.25 ns	42628 ns	1.17
array/random/rand!/Int64	47981 ns	47218.25 ns	1.02
array/random/randn/Float32	69868.75 ns	69713.5 ns	1.00
array/random/randn!/Float32	68091 ns	81656 ns	0.83
array/reductions/mapreduce/Float32/1d	68586 ns	68866 ns	1.00
array/reductions/mapreduce/Float32/dims=1	53981 ns	53788.25 ns	1.00
array/reductions/mapreduce/Float32/dims=1L	59778.5 ns	59428.25 ns	1.01
array/reductions/mapreduce/Float32/dims=2	60631 ns	60230.75 ns	1.01
array/reductions/mapreduce/Float32/dims=2L	59961 ns	71823.5 ns	0.83
array/reductions/mapreduce/Int64/1d	71023.75 ns	60976 ns	1.16
array/reductions/mapreduce/Int64/dims=1	51321 ns	51423.25 ns	1.00
array/reductions/mapreduce/Int64/dims=1L	71396 ns	71273.5 ns	1.00
array/reductions/mapreduce/Int64/dims=2	58418.5 ns	58898.25 ns	0.99
array/reductions/mapreduce/Int64/dims=2L	75963.75 ns	75998.5 ns	1.00
array/reductions/reduce/Float32/1d	68886 ns	65938.5 ns	1.04
array/reductions/reduce/Float32/dims=1	54490.75 ns	54135.75 ns	1.01
array/reductions/reduce/Float32/dims=1L	51088.5 ns	59218.25 ns	0.86
array/reductions/reduce/Float32/dims=2	59848.5 ns	59845.75 ns	1.00
array/reductions/reduce/Float32/dims=2L	71466 ns	72016.25 ns	0.99
array/reductions/reduce/Int64/1d	70478.5 ns	60268.25 ns	1.17
array/reductions/reduce/Int64/dims=1	51420.75 ns	51278.25 ns	1.00
array/reductions/reduce/Int64/dims=1L	70313.75 ns	70481.25 ns	1.00
array/reductions/reduce/Int64/dims=2	58396 ns	50230.5 ns	1.16
array/reductions/reduce/Int64/dims=2L	75931.25 ns	76196.25 ns	1.00
array/reverse/1d	35690.75 ns	40158 ns	0.89
array/reverse/1dL	65758.5 ns	64526 ns	1.02
array/reverse/1dL_inplace	58611 ns	78426.25 ns	0.75
array/reverse/1d_inplace	46855.75 ns	50110.75 ns	0.94
array/reverse/2d	47768.25 ns	47343.25 ns	1.01
array/reverse/2dL	75563.75 ns	57771 ns	1.31
array/reverse/2dL_inplace	94101.5 ns	94651.5 ns	0.99
array/reverse/2d_inplace	65153.75 ns	54671 ns	1.19
array/sorting/1d	335835.75 ns	336177.5 ns	1.00
gemm/tiled	1917970.5 ns	1924825.75 ns	1.00
gemm/tiled_unbounded	1944153.5 ns	1936623.25 ns	1.00
integration/byval/reference	40551 ns	40720 ns	1.00
integration/byval/slices=1	41071 ns	40730 ns	1.01
integration/byval/slices=2	129322 ns	144902 ns	0.89
integration/byval/slices=3	238243 ns	238064 ns	1.00
integration/volumerhs	4843887 ns	4842520 ns	1.00
kernel/indexing	38268 ns	44715.75 ns	0.86
kernel/indexing_checked	59256 ns	37433 ns	1.58
kernel/launch	1230 ns	1247.5 ns	0.99
kernel/rand	72368.75 ns	72283.5 ns	1.00
latency/import	1407266670 ns	1409644160 ns	1.00
latency/precompile	29090792696 ns	29133709219 ns	1.00
latency/ttfp	1790610923 ns	1794988893 ns	1.00
stencil/diffusion3d	1592050.25 ns	1595928.25 ns	1.00
stencil/diffusion3d_checked	1632435.75 ns	1614048.5 ns	1.01

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