Skip prepare_state in GC finalizers

[luraess]

Every generated HIP wrapper starts with prepare_state(). In a GC finalizer, that created task-local state on whichever task GC interrupted, drained that task's sticky HIP error, and switched the thread back to the task's context, which undid the finalizer's HIP.context!. Item 1 of #1105.

With #1104, #1092 and #1096, the finalizer audit from #1105 drops from 1600 hits to 0 over the full suite. The new core/tls test fails on main and passes here (mi200, 2 GCDs).

🤖 Generated with Claude Code

[github-actions]

AMDGPU.jl Benchmarks

Details

Benchmark suite	Current: 9029179	Previous: 8a66547	Ratio
amdgpu/synchronization/context/device	555 ns	550 ns	1.01
amdgpu/synchronization/stream/blocking	225 ns	222.5 ns	1.01
amdgpu/synchronization/stream/nonblocking	307.5 ns	307.5 ns	1
applications/bitonic_sort	1128998.75 ns	1160103.75 ns	0.97
applications/convolution	74448.5 ns	72376 ns	1.03
applications/floyd_warshall	9036788.25 ns	9100856.25 ns	0.99
applications/histogram	795924 ns	798998.75 ns	1.00
applications/prefix_sum	189635.25 ns	190535 ns	1.00
array/accumulate/Float32/1d	74748.75 ns	79346 ns	0.94
array/accumulate/Float32/dims=1	274434 ns	271536.25 ns	1.01
array/accumulate/Float32/dims=1L	79711.25 ns	72536 ns	1.10
array/accumulate/Float32/dims=2	73951.25 ns	74041 ns	1.00
array/accumulate/Float32/dims=2L	3027629.75 ns	2834326.75 ns	1.07
array/accumulate/Int64/1d	72236.25 ns	79098.5 ns	0.91
array/accumulate/Int64/dims=1	249146.25 ns	248393.25 ns	1.00
array/accumulate/Int64/dims=1L	89091.5 ns	89128.75 ns	1.00
array/accumulate/Int64/dims=2	76356.25 ns	76041 ns	1.00
array/accumulate/Int64/dims=2L	3387365 ns	3139866 ns	1.08
array/broadcast	73443.5 ns	50443.25 ns	1.46
array/construct	1947.5 ns	2247.5 ns	0.87
array/copy	39325.5 ns	39278 ns	1.00
array/copyto!/cpu_to_gpu	85886.25 ns	85676.25 ns	1.00
array/copyto!/gpu_to_cpu	85676.25 ns	85591.25 ns	1.00
array/copyto!/gpu_to_gpu	36640.5 ns	58288.25 ns	0.63
array/iteration/findall/bool	127576.75 ns	125954.25 ns	1.01
array/iteration/findall/int	142167 ns	146617 ns	0.97
array/iteration/findfirst/bool	154669.75 ns	155292.25 ns	1.00
array/iteration/findfirst/int	138127 ns	189260 ns	0.73
array/iteration/findmin/1d	103996.5 ns	103258.75 ns	1.01
array/iteration/findmin/2d	90428.75 ns	90081.25 ns	1.00
array/iteration/logical	228603.25 ns	226565.5 ns	1.01
array/iteration/scalar	302959.25 ns	295931.5 ns	1.02
array/permutedims/2d	71483.5 ns	70736 ns	1.01
array/permutedims/3d	69423.5 ns	69253.5 ns	1.00
array/permutedims/4d	51041 ns	50278.25 ns	1.02
array/random/rand/Float32	46845.75 ns	40065.75 ns	1.17
array/random/rand/Int64	56008.25 ns	48863 ns	1.15
array/random/rand!/Float32	66213.5 ns	65541 ns	1.01
array/random/rand!/Int64	71276 ns	69623.25 ns	1.02
array/random/randn/Float32	71363.5 ns	71161 ns	1.00
array/random/randn!/Float32	82398.75 ns	58721 ns	1.40
array/reductions/mapreduce/Float32/1d	82061.25 ns	81621 ns	1.01
array/reductions/mapreduce/Float32/dims=1	73301 ns	56660.75 ns	1.29
array/reductions/mapreduce/Float32/dims=1L	840037.25 ns	844514.25 ns	0.99
array/reductions/mapreduce/Float32/dims=2	73343.5 ns	74073.5 ns	0.99
array/reductions/mapreduce/Float32/dims=2L	136339.5 ns	136524.5 ns	1.00
array/reductions/mapreduce/Int64/1d	82001.25 ns	81663.5 ns	1.00
array/reductions/mapreduce/Int64/dims=1	73566 ns	73933.5 ns	1.00
array/reductions/mapreduce/Int64/dims=1L	846449.75 ns	839174.25 ns	1.01
array/reductions/mapreduce/Int64/dims=2	74676.25 ns	75486.25 ns	0.99
array/reductions/mapreduce/Int64/dims=2L	137692 ns	137611.75 ns	1.00
array/reductions/reduce/Float32/1d	81788.5 ns	81541.25 ns	1.00
array/reductions/reduce/Float32/dims=1	72821 ns	70296 ns	1.04
array/reductions/reduce/Float32/dims=1L	838959.75 ns	836984.25 ns	1.00
array/reductions/reduce/Float32/dims=2	75286 ns	75123.5 ns	1.00
array/reductions/reduce/Float32/dims=2L	136392 ns	136774.25 ns	1.00
array/reductions/reduce/Int64/1d	81936.25 ns	81368.75 ns	1.01
array/reductions/reduce/Int64/dims=1	73213.5 ns	74831 ns	0.98
array/reductions/reduce/Int64/dims=1L	834839.5 ns	844554 ns	0.99
array/reductions/reduce/Int64/dims=2	73861 ns	75411.25 ns	0.98
array/reductions/reduce/Int64/dims=2L	137109.5 ns	137047 ns	1.00
array/reverse/1d	37023.25 ns	43458 ns	0.85
array/reverse/1dL	71553.5 ns	70606 ns	1.01
array/reverse/1dL_inplace	74546 ns	79671 ns	0.94
array/reverse/1d_inplace	61153.5 ns	60390.75 ns	1.01
array/reverse/2d	47928 ns	47625.75 ns	1.01
array/reverse/2dL	57413.25 ns	57993.5 ns	0.99
array/reverse/2dL_inplace	68601 ns	79096 ns	0.87
array/reverse/2d_inplace	40545.75 ns	61191 ns	0.66
array/sorting/1d	324740 ns	326449.5 ns	0.99
gemm/tiled	1914572.75 ns	1961659.75 ns	0.98
gemm/tiled_unbounded	1946430.75 ns	1960334.75 ns	0.99
integration/byval/reference	38721 ns	39821 ns	0.97
integration/byval/slices=1	40941 ns	40830 ns	1.00
integration/byval/slices=2	136582 ns	130842 ns	1.04
integration/byval/slices=3	241553 ns	241884 ns	1.00
integration/volumerhs	4882560 ns	4900148 ns	1.00
kernel/indexing	36475.5 ns	41880.5 ns	0.87
kernel/indexing_checked	41065.75 ns	37075.5 ns	1.11
kernel/launch	1147.5 ns	1130 ns	1.02
kernel/rand	56170.75 ns	70343.5 ns	0.80
latency/import	1457552506 ns	1456646355 ns	1.00
latency/precompile	22496228084 ns	22642954217 ns	0.99
latency/ttfp	2244637956 ns	2251455325 ns	1.00
stencil/diffusion3d	1616886 ns	1611230 ns	1.00
stencil/diffusion3d_checked	1660316.5 ns	1655470.5 ns	1.00

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