Fix GC race in CuRef getindex causing intermittent CUDA errors

[maleadt]

The getindex methods for CuRefValue and CuRefArray only preserved the CPU Ref with GC.@preserve, but not the GPU reference. After extracting the raw device pointer via unsafe_convert, the GC could collect the CuRefValue (running its pool_free finalizer) before the unsafe_copyto! memcpy completed, resulting in use-after-free.

Fixes #3012

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.42%. Comparing base (a9a687c) to head (e49bd1b).
⚠️ Report is 5 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #3087   +/-   ##
=======================================
  Coverage   90.41%   90.42%           
=======================================
  Files         141      141           
  Lines       11993    11993           
=======================================
+ Hits        10844    10845    +1     
+ Misses       1149     1148    -1     

☔ 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.

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: e49bd1b	Previous: 5f45772	Ratio
array/accumulate/Float32/1d	101132 ns	101495 ns	1.00
array/accumulate/Float32/dims=1	76479 ns	76898 ns	0.99
array/accumulate/Float32/dims=1L	1583539 ns	1585143.5 ns	1.00
array/accumulate/Float32/dims=2	143853 ns	143801 ns	1.00
array/accumulate/Float32/dims=2L	657727 ns	657240 ns	1.00
array/accumulate/Int64/1d	118411 ns	118623 ns	1.00
array/accumulate/Int64/dims=1	80131 ns	80572.5 ns	0.99
array/accumulate/Int64/dims=1L	1695177 ns	1693852 ns	1.00
array/accumulate/Int64/dims=2	155949 ns	156484 ns	1.00
array/accumulate/Int64/dims=2L	961597.5 ns	961603 ns	1.00
array/broadcast	20514 ns	20294 ns	1.01
array/construct	1331.2 ns	1320.4 ns	1.01
array/copy	18720 ns	18780 ns	1.00
array/copyto!/cpu_to_gpu	215223.5 ns	214684 ns	1.00
array/copyto!/gpu_to_cpu	283574 ns	282072 ns	1.01
array/copyto!/gpu_to_gpu	11408 ns	11361 ns	1.00
array/iteration/findall/bool	131383 ns	131719.5 ns	1.00
array/iteration/findall/int	148780 ns	148883 ns	1.00
array/iteration/findfirst/bool	80906 ns	81470.5 ns	0.99
array/iteration/findfirst/int	83533.5 ns	83414 ns	1.00
array/iteration/findmin/1d	88432.5 ns	89419 ns	0.99
array/iteration/findmin/2d	117090.5 ns	117365 ns	1.00
array/iteration/logical	199596 ns	207612 ns	0.96
array/iteration/scalar	67301 ns	66780 ns	1.01
array/permutedims/2d	52399 ns	52471.5 ns	1.00
array/permutedims/3d	52919 ns	53137 ns	1.00
array/permutedims/4d	52303 ns	52429 ns	1.00
array/random/rand/Float32	13180 ns	13089 ns	1.01
array/random/rand/Int64	37361 ns	37236 ns	1.00
array/random/rand!/Float32	8520.333333333334 ns	8527.666666666666 ns	1.00
array/random/rand!/Int64	34437.5 ns	34109.5 ns	1.01
array/random/randn/Float32	44084 ns	38147 ns	1.16
array/random/randn!/Float32	31665 ns	31640 ns	1.00
array/reductions/mapreduce/Float32/1d	35194 ns	34735.5 ns	1.01
array/reductions/mapreduce/Float32/dims=1	40837 ns	40760 ns	1.00
array/reductions/mapreduce/Float32/dims=1L	51944 ns	51917 ns	1.00
array/reductions/mapreduce/Float32/dims=2	56712 ns	56503.5 ns	1.00
array/reductions/mapreduce/Float32/dims=2L	69408 ns	69496.5 ns	1.00
array/reductions/mapreduce/Int64/1d	42852 ns	42820 ns	1.00
array/reductions/mapreduce/Int64/dims=1	42338.5 ns	44181 ns	0.96
array/reductions/mapreduce/Int64/dims=1L	87836 ns	87798 ns	1.00
array/reductions/mapreduce/Int64/dims=2	59872 ns	59808 ns	1.00
array/reductions/mapreduce/Int64/dims=2L	85157 ns	85232 ns	1.00
array/reductions/reduce/Float32/1d	35522 ns	34883 ns	1.02
array/reductions/reduce/Float32/dims=1	43353.5 ns	39758 ns	1.09
array/reductions/reduce/Float32/dims=1L	52272 ns	52166 ns	1.00
array/reductions/reduce/Float32/dims=2	57289 ns	56925 ns	1.01
array/reductions/reduce/Float32/dims=2L	70052 ns	69909 ns	1.00
array/reductions/reduce/Int64/1d	43171 ns	42673 ns	1.01
array/reductions/reduce/Int64/dims=1	51441.5 ns	42123 ns	1.22
array/reductions/reduce/Int64/dims=1L	87779 ns	87782 ns	1.00
array/reductions/reduce/Int64/dims=2	59682 ns	59551 ns	1.00
array/reductions/reduce/Int64/dims=2L	84985.5 ns	84796 ns	1.00
array/reverse/1d	18621 ns	18432.5 ns	1.01
array/reverse/1dL	69138 ns	69025 ns	1.00
array/reverse/1dL_inplace	65918 ns	65968 ns	1.00
array/reverse/1d_inplace	8552.333333333334 ns	10240.666666666666 ns	0.84
array/reverse/2d	20669 ns	20709 ns	1.00
array/reverse/2dL	72745 ns	72815 ns	1.00
array/reverse/2dL_inplace	66016 ns	65992 ns	1.00
array/reverse/2d_inplace	10186 ns	11117.5 ns	0.92
array/sorting/1d	2734661 ns	2754859 ns	0.99
array/sorting/2d	1068455 ns	1075967 ns	0.99
array/sorting/by	3303203 ns	3328240 ns	0.99
cuda/synchronization/context/auto	1158.4 ns	1192.4 ns	0.97
cuda/synchronization/context/blocking	947.8181818181819 ns	947.7391304347826 ns	1.00
cuda/synchronization/context/nonblocking	7681.6 ns	7660.1 ns	1.00
cuda/synchronization/stream/auto	1018.4545454545455 ns	1032.5 ns	0.99
cuda/synchronization/stream/blocking	824.8139534883721 ns	841.5588235294117 ns	0.98
cuda/synchronization/stream/nonblocking	7679.9 ns	7189.6 ns	1.07
integration/byval/reference	144062 ns	143997 ns	1.00
integration/byval/slices=1	145976 ns	145776 ns	1.00
integration/byval/slices=2	284694 ns	284427 ns	1.00
integration/byval/slices=3	423386 ns	423129 ns	1.00
integration/cudadevrt	102770 ns	102598 ns	1.00
integration/volumerhs	9440456 ns	9429742.5 ns	1.00
kernel/indexing	13404 ns	13331 ns	1.01
kernel/indexing_checked	14139 ns	14116 ns	1.00
kernel/launch	2199.222222222222 ns	2147 ns	1.02
kernel/occupancy	679.9019607843137 ns	660.5723270440252 ns	1.03
kernel/rand	16496 ns	15598 ns	1.06
latency/import	3837634365 ns	3807044359.5 ns	1.01
latency/precompile	4602411854.5 ns	4590923492 ns	1.00
latency/ttfp	4415248089.5 ns	4392969126 ns	1.01

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