Export gputhrow

[termi-official]

@gputhrow was exported on v5

[maleadt]

It was not? https://github.com/JuliaGPU/CUDA.jl/blob/release-5.11/src/device/quirks.jl

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 45b627c	Previous: af6961a	Ratio
array/accumulate/Float32/1d	101055 ns	101502.5 ns	1.00
array/accumulate/Float32/dims=1	76936 ns	76629.5 ns	1.00
array/accumulate/Float32/dims=1L	1585505 ns	1585691.5 ns	1.00
array/accumulate/Float32/dims=2	144151 ns	143840 ns	1.00
array/accumulate/Float32/dims=2L	658023 ns	657866.5 ns	1.00
array/accumulate/Int64/1d	118692 ns	119085 ns	1.00
array/accumulate/Int64/dims=1	79686 ns	79723 ns	1.00
array/accumulate/Int64/dims=1L	1694741 ns	1694532 ns	1.00
array/accumulate/Int64/dims=2	156062 ns	156114 ns	1.00
array/accumulate/Int64/dims=2L	961855 ns	961892 ns	1.00
array/broadcast	20347 ns	20629 ns	0.99
array/construct	1287.6 ns	1256.5 ns	1.02
array/copy	18146 ns	17886 ns	1.01
array/copyto!/cpu_to_gpu	214789 ns	212623 ns	1.01
array/copyto!/gpu_to_cpu	285256 ns	283142 ns	1.01
array/copyto!/gpu_to_gpu	10932 ns	10702 ns	1.02
array/iteration/findall/bool	135084 ns	134713 ns	1.00
array/iteration/findall/int	150895 ns	149574 ns	1.01
array/iteration/findfirst/bool	82571 ns	81484 ns	1.01
array/iteration/findfirst/int	84149.5 ns	83536 ns	1.01
array/iteration/findmin/1d	85286.5 ns	85967.5 ns	0.99
array/iteration/findmin/2d	117254 ns	116482 ns	1.01
array/iteration/logical	200524.5 ns	199753 ns	1.00
array/iteration/scalar	67866 ns	67105.5 ns	1.01
array/permutedims/2d	52533 ns	52568 ns	1.00
array/permutedims/3d	52716.5 ns	53057 ns	0.99
array/permutedims/4d	51875 ns	51868.5 ns	1.00
array/random/rand/Float32	13217 ns	12842 ns	1.03
array/random/rand/Int64	25251 ns	25396 ns	0.99
array/random/rand!/Float32	9853.5 ns	8414.666666666666 ns	1.17
array/random/rand!/Int64	21914 ns	21981 ns	1.00
array/random/randn/Float32	43758 ns	37382.5 ns	1.17
array/random/randn!/Float32	31002 ns	30580 ns	1.01
array/reductions/mapreduce/Float32/1d	35214.5 ns	34346 ns	1.03
array/reductions/mapreduce/Float32/dims=1	39305 ns	39918.5 ns	0.98
array/reductions/mapreduce/Float32/dims=1L	51269 ns	51509 ns	1.00
array/reductions/mapreduce/Float32/dims=2	56724.5 ns	56563 ns	1.00
array/reductions/mapreduce/Float32/dims=2L	69541 ns	69236 ns	1.00
array/reductions/mapreduce/Int64/1d	43334 ns	42783 ns	1.01
array/reductions/mapreduce/Int64/dims=1	43236.5 ns	42869 ns	1.01
array/reductions/mapreduce/Int64/dims=1L	87164 ns	87545 ns	1.00
array/reductions/mapreduce/Int64/dims=2	59640 ns	59325 ns	1.01
array/reductions/mapreduce/Int64/dims=2L	85148 ns	85131 ns	1.00
array/reductions/reduce/Float32/1d	35062 ns	34338 ns	1.02
array/reductions/reduce/Float32/dims=1	42837 ns	48895 ns	0.88
array/reductions/reduce/Float32/dims=1L	51460 ns	51567 ns	1.00
array/reductions/reduce/Float32/dims=2	56801 ns	56622 ns	1.00
array/reductions/reduce/Float32/dims=2L	70088 ns	69625 ns	1.01
array/reductions/reduce/Int64/1d	43395 ns	42697 ns	1.02
array/reductions/reduce/Int64/dims=1	42548 ns	47176.5 ns	0.90
array/reductions/reduce/Int64/dims=1L	87073 ns	87361 ns	1.00
array/reductions/reduce/Int64/dims=2	59960.5 ns	59337 ns	1.01
array/reductions/reduce/Int64/dims=2L	85238 ns	84717 ns	1.01
array/reverse/1d	17777 ns	18051 ns	0.98
array/reverse/1dL	68336 ns	68628 ns	1.00
array/reverse/1dL_inplace	65822 ns	65835 ns	1.00
array/reverse/1d_inplace	10346.166666666668 ns	8676.333333333334 ns	1.19
array/reverse/2d	20928 ns	20541 ns	1.02
array/reverse/2dL	72881 ns	72496 ns	1.01
array/reverse/2dL_inplace	65775 ns	65804 ns	1.00
array/reverse/2d_inplace	10414 ns	10104 ns	1.03
array/sorting/1d	2736888 ns	2734890.5 ns	1.00
array/sorting/2d	1069374 ns	1069333 ns	1.00
array/sorting/by	3306761 ns	3305442 ns	1.00
cuda/synchronization/context/auto	1174 ns	1142.4 ns	1.03
cuda/synchronization/context/blocking	945.3255813953489 ns	918.9428571428572 ns	1.03
cuda/synchronization/context/nonblocking	7649 ns	7080.9 ns	1.08
cuda/synchronization/stream/auto	974.7777777777778 ns	983.3846153846154 ns	0.99
cuda/synchronization/stream/blocking	791.6862745098039 ns	819.060606060606 ns	0.97
cuda/synchronization/stream/nonblocking	7291.200000000001 ns	7951.1 ns	0.92
integration/byval/reference	143814 ns	143938 ns	1.00
integration/byval/slices=1	145593 ns	145713 ns	1.00
integration/byval/slices=2	284337 ns	284653 ns	1.00
integration/byval/slices=3	422732 ns	422978 ns	1.00
integration/cudadevrt	102289 ns	102487 ns	1.00
integration/volumerhs	23451596 ns	23490453 ns	1.00
kernel/indexing	13244 ns	13409 ns	0.99
kernel/indexing_checked	14046 ns	14040 ns	1.00
kernel/launch	2248.8888888888887 ns	2267.4444444444443 ns	0.99
kernel/occupancy	702.8958333333334 ns	828.1463414634146 ns	0.85
kernel/rand	14343 ns	14466 ns	0.99
latency/import	3824776930.5 ns	3809615751 ns	1.00
latency/precompile	4599001518 ns	4560767959.5 ns	1.01
latency/ttfp	4432304043.5 ns	4401798424 ns	1.01

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

[termi-official]

Oh, right. Sorry. I was really just took CUDA: @gputhrow from somewhere and was assuming it was part of the public API.

We would like to do the following in our package: @device_override @noinline Ferrite.throw_detJ_not_pos(detJ) = @gputhrow("ArgumentError", "det(J) is not positive. Please check the value on CPU.") to still give reasonable error messages. Would it be possible to add @gputhrow to the public API or should we do something else?

[maleadt]

The functionality is too much tied to internal state for it to become public, I think. Especially because I plan to rework exception handling, albeit at some undetermined point in the future. Would it be fine if you lose some of the reporting accuracy and simply do what we used to do before:

@cuprintln "ERROR: " $(args...) "."
throw(nothong)

[termi-official]

I see. We will stick with whatever is currently recommended.