[maleadt]

Same class of bug as JuliaGPU/Metal.jl#567 (fixed there in JuliaGPU/Metal.jl#1070, and in OpenCL.jl by JuliaGPU/OpenCL.jl#568).

On Julia 1.12, inference forms a cycle through the exception runtime:

throw_inexacterror (overlay) → `@gputhrow` → setproperty!(::ExceptionInfo, ::Symbol, ::Ptr{UInt8})
  → unsafe_store!(::Ptr{Int32}, ::Ptr{UInt8}) → Int32(::UInt64) → checked_trunc_sint → throw_inexacterror

setproperty! is also inferred without a constant sym. In that case the :status/:output_lock branches convert whatever value is stored to Int32. Julia 1.12 merges the effects of all frames in a cycle, and the overlays the runtime reaches (kernel_state(), the atomics, threadIdx()) are not consistent, so Int32(::UInt64) ends up with nonoverlayed = ALWAYS_FALSE. Any :foldable code that uses it then can't be concrete-evaluated anymore. x < π happens to avoid that conversion on CUDA, but the new test shows a foldable function that doesn't fold.

The fix type-asserts those two stores instead of converting. Every setter stores an Int32 there anyway. With the fix, no cycle passes through an overlay on 1.12 or 1.13.

The new concrete evaluation codegen test fails on 1.12 without the fix. core/codegen, core/exceptions and core/execution pass with it.

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 88af030	Previous: 2ea18f9	Ratio
array/accumulate/Float32/1d	54704 ns	54772 ns	1.00
array/accumulate/Float32/dims=1	39663 ns	39700 ns	1.00
array/accumulate/Float32/dims=1L	82706 ns	82439 ns	1.00
array/accumulate/Float32/dims=2	49947 ns	49526 ns	1.01
array/accumulate/Float32/dims=2L	400364 ns	399506 ns	1.00
array/accumulate/Int64/1d	88661 ns	89102 ns	1.00
array/accumulate/Int64/dims=1	59819 ns	60313 ns	0.99
array/accumulate/Int64/dims=1L	139772 ns	140583 ns	0.99
array/accumulate/Int64/dims=2	68661 ns	68534 ns	1.00
array/accumulate/Int64/dims=2L	795048 ns	797122 ns	1.00
array/broadcast	10260 ns	10356 ns	0.99
array/broadcast launch	6618 ns	6753 ns	0.98
array/construct	936.0540540540541 ns	960.1785714285714 ns	0.97
array/copy	16709 ns	17189 ns	0.97
array/copyto!/cpu_to_gpu	208700 ns	210329 ns	0.99
array/copyto!/gpu_to_cpu	241402 ns	242774 ns	0.99
array/copyto!/gpu_to_gpu	8901 ns	8873 ns	1.00
array/iteration/findall/bool	49362 ns	49920 ns	0.99
array/iteration/findall/int	65664 ns	66610 ns	0.99
array/iteration/findfirst/bool	37150 ns	34868 ns	1.07
array/iteration/findfirst/int	36507 ns	34998 ns	1.04
array/iteration/findmin/1d	51575 ns	53031 ns	0.97
array/iteration/findmin/2d	53260 ns	53598 ns	0.99
array/iteration/logical	51110 ns	50904 ns	1.00
array/iteration/scalar	64090 ns	60322 ns	1.06
array/permutedims/2d	46911 ns	47093 ns	1.00
array/permutedims/3d	44482 ns	44539 ns	1.00
array/permutedims/4d	47485 ns	47896 ns	0.99
array/random/rand/Float32	11046 ns	11145 ns	0.99
array/random/rand/Int64	17614 ns	17796 ns	0.99
array/random/rand!/Float32	8281.666666666666 ns	8344.666666666666 ns	0.99
array/random/rand!/Int64	15437 ns	15558 ns	0.99
array/random/randn/Float32	33842 ns	34128 ns	0.99
array/random/randn!/Float32	24847 ns	25132 ns	0.99
array/reductions/mapreduce/Float32/1d	36521 ns	34362 ns	1.06
array/reductions/mapreduce/Float32/dims=1	19937 ns	19450 ns	1.03
array/reductions/mapreduce/Float32/dims=1L	57853 ns	57133 ns	1.01
array/reductions/mapreduce/Float32/dims=2	27748 ns	27326 ns	1.02
array/reductions/mapreduce/Float32/dims=2L	57618 ns	57650 ns	1.00
array/reductions/mapreduce/Int64/1d	36770 ns	35313 ns	1.04
array/reductions/mapreduce/Int64/dims=1	21951 ns	21761 ns	1.01
array/reductions/mapreduce/Int64/dims=1L	96968 ns	96707 ns	1.00
array/reductions/mapreduce/Int64/dims=2	29160 ns	28019 ns	1.04
array/reductions/mapreduce/Int64/dims=2L	83278 ns	83491 ns	1.00
array/reductions/reduce/Float32/1d	36667 ns	34934 ns	1.05
array/reductions/reduce/Float32/dims=1	19519 ns	19330 ns	1.01
array/reductions/reduce/Float32/dims=1L	57404 ns	57162 ns	1.00
array/reductions/reduce/Float32/dims=2	28436 ns	27376 ns	1.04
array/reductions/reduce/Float32/dims=2L	57941 ns	57509 ns	1.01
array/reductions/reduce/Int64/1d	36839 ns	35215 ns	1.05
array/reductions/reduce/Int64/dims=1	22573 ns	21685 ns	1.04
array/reductions/reduce/Int64/dims=1L	97345 ns	96542 ns	1.01
array/reductions/reduce/Int64/dims=2	29178 ns	28245 ns	1.03
array/reductions/reduce/Int64/dims=2L	84080 ns	83062 ns	1.01
array/reverse/1d	19375 ns	19915 ns	0.97
array/reverse/1dL	73382 ns	73665 ns	1.00
array/reverse/1dL_inplace	70273 ns	70464 ns	1.00
array/reverse/1d_inplace	11987 ns	12146 ns	0.99
array/reverse/2d	22374 ns	22477 ns	1.00
array/reverse/2dL	80838 ns	80858 ns	1.00
array/reverse/2dL_inplace	70920 ns	70947 ns	1.00
array/reverse/2d_inplace	14853 ns	14822 ns	1.00
array/sorting/1d	750437 ns	750787 ns	1.00
array/sorting/2d	194625 ns	194334 ns	1.00
array/sorting/by	928006 ns	928178 ns	1.00
cuda/graph/capture	37681 ns	37276 ns	1.01
cuda/graph/eager	32930 ns	34127 ns	0.96
cuda/graph/launch	13876 ns	13916 ns	1.00
cuda/synchronization/context/auto	2479.6666666666665 ns	2443.6666666666665 ns	1.01
cuda/synchronization/context/blocking	891.3409090909091 ns	906.8809523809524 ns	0.98
cuda/synchronization/context/nonblocking	8793 ns	8349 ns	1.05
cuda/synchronization/stream/auto	822.9807692307693 ns	871.8412698412699 ns	0.94
cuda/synchronization/stream/blocking	764.3076923076923 ns	789.8627450980392 ns	0.97
cuda/synchronization/stream/nonblocking	7441.666666666667 ns	8034 ns	0.93
integration/byval/reference	147967 ns	147627 ns	1.00
integration/byval/slices=1	149073 ns	148709 ns	1.00
integration/byval/slices=2	290581 ns	290270 ns	1.00
integration/byval/slices=3	432681 ns	432303 ns	1.00
integration/cudadevrt	105347 ns	104902 ns	1.00
integration/volumerhs	9613211 ns	9610172 ns	1.00
kernel/indexing	11377 ns	11573 ns	0.98
kernel/indexing_checked	12062 ns	12125 ns	0.99
kernel/launch	2271.6666666666665 ns	2307.777777777778 ns	0.98
kernel/occupancy	673.32 ns	682.8 ns	0.99
kernel/rand	14995 ns	14981 ns	1.00
latency/import	4038218390 ns	4211043895 ns	0.96
latency/precompile	4984602170 ns	5207029807 ns	0.96
latency/ttfp	4861927507 ns	5033274915 ns	0.97

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

[codecov]

Codecov Report

❌ Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.14%. Comparing base (f533f03) to head (88af030).
⚠️ Report is 20 commits behind head on main.

Files with missing lines	Patch %	Lines
CUDACore/src/device/runtime.jl	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3413      +/-   ##
==========================================
- Coverage   87.08%   85.14%   -1.94%     
==========================================
  Files         192      191       -1     
  Lines       19543    19260     -283     
==========================================
- Hits        17019    16399     -620     
- Misses       2524     2861     +337     

☔ View full report in Codecov by Harness.
📢 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.