Add preference to use CUDA_Runtime_Discovery for the compiler binaries

[apozharski]

This works for my use case of ahead of time compilation, but now that I think about it may break if you use a local runtime and the artifact compiler, which I need to check, as well as not having any checks for compatibility of the compiler version and runtime version. It also might need some docs. As such, keeping this as [wip] for now.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 10.16%. Comparing base (a79b516) to head (58f4c8e).

Additional details and impacted files

@@             Coverage Diff             @@
##           master    #3080       +/-   ##
===========================================
- Coverage   90.42%   10.16%   -80.27%     
===========================================
  Files         141      114       -27     
  Lines       11993     9123     -2870     
===========================================
- Hits        10845      927     -9918     
- Misses       1148     8196     +7048     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[apozharski]

The above concerns about mixing the local compiler and local runtime preferences has been addressed and everything should work correctly now.

[maleadt]

Needs a bump in Project.toml for CUDA_Runtime_Discovery

[maleadt]

Compat needs to be bumped in subpackages as well.

[apozharski]

Hmmm it seems that a GPUArrays test fails consistently in CI and I can't repro it locally 🤔

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 9a7c2c6	Previous: e0e295f	Ratio
array/accumulate/Float32/1d	100746 ns	100878 ns	1.00
array/accumulate/Float32/dims=1	77158 ns	75855 ns	1.02
array/accumulate/Float32/dims=1L	1586087.5 ns	1585504 ns	1.00
array/accumulate/Float32/dims=2	144338 ns	143115.5 ns	1.01
array/accumulate/Float32/dims=2L	658356 ns	657101 ns	1.00
array/accumulate/Int64/1d	118165.5 ns	118250 ns	1.00
array/accumulate/Int64/dims=1	80273 ns	79820.5 ns	1.01
array/accumulate/Int64/dims=1L	1695183 ns	1694871 ns	1.00
array/accumulate/Int64/dims=2	156182 ns	155746 ns	1.00
array/accumulate/Int64/dims=2L	961534 ns	961802 ns	1.00
array/broadcast	20370 ns	20486 ns	0.99
array/construct	1258.7 ns	1263.9 ns	1.00
array/copy	18040 ns	17962 ns	1.00
array/copyto!/cpu_to_gpu	213158 ns	214197 ns	1.00
array/copyto!/gpu_to_cpu	282748 ns	281343 ns	1.00
array/copyto!/gpu_to_gpu	10585.666666666666 ns	10794 ns	0.98
array/iteration/findall/bool	134671 ns	134478 ns	1.00
array/iteration/findall/int	149809 ns	149314.5 ns	1.00
array/iteration/findfirst/bool	81561 ns	81113 ns	1.01
array/iteration/findfirst/int	83914 ns	83293 ns	1.01
array/iteration/findmin/1d	87680 ns	84555 ns	1.04
array/iteration/findmin/2d	117130.5 ns	116516 ns	1.01
array/iteration/logical	199715.5 ns	197262.5 ns	1.01
array/iteration/scalar	67619 ns	67092 ns	1.01
array/permutedims/2d	51960 ns	52211 ns	1.00
array/permutedims/3d	52466.5 ns	52764 ns	0.99
array/permutedims/4d	51059.5 ns	51452 ns	0.99
array/random/rand/Float32	12481 ns	12943 ns	0.96
array/random/rand/Int64	24755 ns	24996 ns	0.99
array/random/rand!/Float32	8327.666666666666 ns	8402.333333333334 ns	0.99
array/random/rand!/Int64	21598 ns	21937 ns	0.98
array/random/randn/Float32	36533 ns	36954 ns	0.99
array/random/randn!/Float32	30546 ns	30982 ns	0.99
array/reductions/mapreduce/Float32/1d	34038.5 ns	34678 ns	0.98
array/reductions/mapreduce/Float32/dims=1	39027.5 ns	39206 ns	1.00
array/reductions/mapreduce/Float32/dims=1L	51121 ns	51259.5 ns	1.00
array/reductions/mapreduce/Float32/dims=2	56152 ns	56274 ns	1.00
array/reductions/mapreduce/Float32/dims=2L	68951.5 ns	69346 ns	0.99
array/reductions/mapreduce/Int64/1d	42066 ns	42412 ns	0.99
array/reductions/mapreduce/Int64/dims=1	42253 ns	42188 ns	1.00
array/reductions/mapreduce/Int64/dims=1L	87281 ns	87287 ns	1.00
array/reductions/mapreduce/Int64/dims=2	59211 ns	59630 ns	0.99
array/reductions/mapreduce/Int64/dims=2L	84425 ns	84743 ns	1.00
array/reductions/reduce/Float32/1d	34425 ns	34235 ns	1.01
array/reductions/reduce/Float32/dims=1	40149.5 ns	39618.5 ns	1.01
array/reductions/reduce/Float32/dims=1L	51438 ns	51305 ns	1.00
array/reductions/reduce/Float32/dims=2	56323 ns	56667 ns	0.99
array/reductions/reduce/Float32/dims=2L	69313 ns	69784 ns	0.99
array/reductions/reduce/Int64/1d	41907 ns	42369 ns	0.99
array/reductions/reduce/Int64/dims=1	42514 ns	42478 ns	1.00
array/reductions/reduce/Int64/dims=1L	87055 ns	87248 ns	1.00
array/reductions/reduce/Int64/dims=2	59346 ns	59729 ns	0.99
array/reductions/reduce/Int64/dims=2L	84476 ns	84769 ns	1.00
array/reverse/1d	17766 ns	18015.5 ns	0.99
array/reverse/1dL	68348 ns	68638 ns	1.00
array/reverse/1dL_inplace	65744 ns	65779 ns	1.00
array/reverse/1d_inplace	10264.166666666668 ns	8649.666666666666 ns	1.19
array/reverse/2d	21025 ns	20711 ns	1.02
array/reverse/2dL	73112 ns	72634 ns	1.01
array/reverse/2dL_inplace	65864 ns	65985 ns	1.00
array/reverse/2d_inplace	10000 ns	10088 ns	0.99
array/sorting/1d	2735645 ns	2734295 ns	1.00
array/sorting/2d	1069524 ns	1068343 ns	1.00
array/sorting/by	3304721 ns	3304353 ns	1.00
cuda/synchronization/context/auto	1140.3 ns	1159.9 ns	0.98
cuda/synchronization/context/blocking	924.5 ns	896.4878048780488 ns	1.03
cuda/synchronization/context/nonblocking	7086.4 ns	7409.1 ns	0.96
cuda/synchronization/stream/auto	995.3333333333334 ns	1027.578947368421 ns	0.97
cuda/synchronization/stream/blocking	844.040404040404 ns	841.2941176470588 ns	1.00
cuda/synchronization/stream/nonblocking	7321 ns	7567.799999999999 ns	0.97
integration/byval/reference	143733 ns	143876 ns	1.00
integration/byval/slices=1	145590.5 ns	145738.5 ns	1.00
integration/byval/slices=2	284469 ns	284423 ns	1.00
integration/byval/slices=3	423004 ns	423173 ns	1.00
integration/cudadevrt	102276 ns	102437 ns	1.00
integration/volumerhs	23425811 ns	23470585 ns	1.00
kernel/indexing	13108 ns	13311 ns	0.98
kernel/indexing_checked	13927 ns	14095 ns	0.99
kernel/launch	2114 ns	2235.1111111111113 ns	0.95
kernel/occupancy	669.24375 ns	693.6190476190476 ns	0.96
kernel/rand	17017 ns	18172.5 ns	0.94
latency/import	3817165248.5 ns	3820990542 ns	1.00
latency/precompile	4582580994.5 ns	4593009584 ns	1.00
latency/ttfp	4399286810.5 ns	4397252952 ns	1.00

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

[maleadt]

Could you report use of a local compiler in the versioninfo() output like we do for local toolkits?

And the failure is probably because you need to rebase on latest master in order to include #3088.

[apozharski]

Added additional info for compiler in versioninfo (exactly same way with additional line as for the runtime, let me know if you want something different).

Changes implemented.