Adapt to the library-based NVPTX_LLVM_Backend_jll 23

[maleadt]

JuliaPackaging/Yggdrasil#14727 replaced the llc, lld and llvm-downgrade executables in the GPU LLVM JLLs with shared libraries exposing a small C API, moving the back-ends to LLVM 23 at the same time. JuliaGPU/GPUCompiler.jl#930 makes GPUCompiler call those libraries in-process, and bumps its compat to the new JLL majors. Since this package pins the same JLL, it needs its compat bumped in lockstep, which is what this PR does.

Changes:

• CUDACore: NVPTX_LLVM_Backend_jll compat 22 → 23, GPUCompiler compat → 2.7 (the library API needs the new GPUCompiler). No code references the JLL's products; the compatibility tables already know LLVM 23.
• The integer abs PTX test now accepts max.s32/max.s64: since LLVM 23 ([SelectionDAG] Add ISD::ABS_MIN_POISON to preserve poision semantics of llvm.abs llvm/llvm-project#183851) only llvm.abs with the poison flag lowers to PTX abs.s, which is undefined for INT_MIN, and Julia's abs must map INT_MIN to itself, so it becomes neg + max.

Note that NVPTX_LLVM_Backend_jll 23.1.1+0 had a bug in its FMA contraction option (fused in LLVM instead of leaving pairs contractible for ptxas); 23.1.1+1 fixes it, and compat cannot exclude the +0 build.

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 304a683	Previous: e94446c	Ratio
array/accumulate/Float32/1d	98840 ns	98260 ns	1.01
array/accumulate/Float32/dims=1	72953 ns	71580 ns	1.02
array/accumulate/Float32/dims=1L	1590723 ns	1599406 ns	0.99
array/accumulate/Float32/dims=2	138427 ns	136501 ns	1.01
array/accumulate/Float32/dims=2L	663008 ns	660905 ns	1.00
array/accumulate/Int64/1d	118638 ns	118400 ns	1.00
array/accumulate/Int64/dims=1	76388 ns	75812 ns	1.01
array/accumulate/Int64/dims=1L	1702219 ns	1714599 ns	0.99
array/accumulate/Int64/dims=2	151533 ns	148622 ns	1.02
array/accumulate/Int64/dims=2L	987966 ns	987047 ns	1.00
array/broadcast	18485 ns	18404 ns	1.00
array/broadcast launch	9967 ns	9654 ns	1.03
array/construct	922.8064516129032 ns	939.0967741935484 ns	0.98
array/copy	16679 ns	16406 ns	1.02
array/copyto!/cpu_to_gpu	209469 ns	206627 ns	1.01
array/copyto!/gpu_to_cpu	242644 ns	240222 ns	1.01
array/copyto!/gpu_to_gpu	8865.333333333334 ns	8693 ns	1.02
array/iteration/findall/bool	133860 ns	133493 ns	1.00
array/iteration/findall/int	148475 ns	147703 ns	1.01
array/iteration/findfirst/bool	70987 ns	69631 ns	1.02
array/iteration/findfirst/int	72642 ns	70675 ns	1.03
array/iteration/findmin/1d	68410 ns	65468 ns	1.04
array/iteration/findmin/2d	101412 ns	99926 ns	1.01
array/iteration/logical	195567 ns	191094 ns	1.02
array/iteration/scalar	63263 ns	62400 ns	1.01
array/permutedims/2d	49582 ns	48890 ns	1.01
array/permutedims/3d	51034 ns	50580 ns	1.01
array/permutedims/4d	51461 ns	50191 ns	1.03
array/random/rand/Float32	11222 ns	10859 ns	1.03
array/random/rand/Int64	21496 ns	21058 ns	1.02
array/random/rand!/Float32	7965 ns	7807.5 ns	1.02
array/random/rand!/Int64	19481 ns	20484 ns	0.95
array/random/randn/Float32	34202 ns	33556 ns	1.02
array/random/randn!/Float32	23320 ns	23685 ns	0.98
array/reductions/mapreduce/Float32/1d	34093 ns	33179 ns	1.03
array/reductions/mapreduce/Float32/dims=1	38831 ns	37934 ns	1.02
array/reductions/mapreduce/Float32/dims=1L	51336 ns	50708 ns	1.01
array/reductions/mapreduce/Float32/dims=2	56634 ns	55105 ns	1.03
array/reductions/mapreduce/Float32/dims=2L	68176 ns	67222 ns	1.01
array/reductions/mapreduce/Int64/1d	41849 ns	41213 ns	1.02
array/reductions/mapreduce/Int64/dims=1	41219 ns	41011 ns	1.01
array/reductions/mapreduce/Int64/dims=1L	88977 ns	88752 ns	1.00
array/reductions/mapreduce/Int64/dims=2	58409 ns	57557 ns	1.01
array/reductions/mapreduce/Int64/dims=2L	84458 ns	83504 ns	1.01
array/reductions/reduce/Float32/1d	33730 ns	33340 ns	1.01
array/reductions/reduce/Float32/dims=1	38026 ns	37782 ns	1.01
array/reductions/reduce/Float32/dims=1L	50997 ns	50647 ns	1.01
array/reductions/reduce/Float32/dims=2	55660 ns	55026 ns	1.01
array/reductions/reduce/Float32/dims=2L	68535 ns	68497 ns	1.00
array/reductions/reduce/Int64/1d	42221 ns	41017 ns	1.03
array/reductions/reduce/Int64/dims=1	41328 ns	40606 ns	1.02
array/reductions/reduce/Int64/dims=1L	89036 ns	88762 ns	1.00
array/reductions/reduce/Int64/dims=2	58770 ns	57379 ns	1.02
array/reductions/reduce/Int64/dims=2L	84561 ns	83389 ns	1.01
array/reverse/1d	16159 ns	17082 ns	0.95
array/reverse/1dL	69941 ns	69888 ns	1.00
array/reverse/1dL_inplace	67777 ns	67721 ns	1.00
array/reverse/1d_inplace	9004 ns	8855.666666666666 ns	1.02
array/reverse/2d	20499 ns	20119 ns	1.02
array/reverse/2dL	73804 ns	73467 ns	1.00
array/reverse/2dL_inplace	67684 ns	67398 ns	1.00
array/reverse/2d_inplace	10323 ns	10085 ns	1.02
array/sorting/1d	2639306 ns	2640661 ns	1.00
array/sorting/2d	1019230 ns	1029199 ns	0.99
array/sorting/by	3175659 ns	3155779 ns	1.01
cuda/synchronization/context/auto	1035.1 ns	1033.7 ns	1.00
cuda/synchronization/context/blocking	801.86 ns	799.7415730337078 ns	1.00
cuda/synchronization/context/nonblocking	5827.666666666667 ns	5869.833333333333 ns	0.99
cuda/synchronization/stream/auto	878.8260869565217 ns	869.0980392156863 ns	1.01
cuda/synchronization/stream/blocking	669.421052631579 ns	672.4144736842105 ns	1.00
cuda/synchronization/stream/nonblocking	5724.166666666667 ns	5623.714285714285 ns	1.02
integration/byval/reference	148212 ns	148331 ns	1.00
integration/byval/slices=1	149413 ns	149217 ns	1.00
integration/byval/slices=2	292362 ns	291903 ns	1.00
integration/byval/slices=3	435554 ns	435113 ns	1.00
integration/cudadevrt	105483 ns	105462 ns	1.00
integration/volumerhs	9145621 ns	9138309 ns	1.00
kernel/indexing	13316 ns	13181 ns	1.01
kernel/indexing_checked	14058 ns	13900 ns	1.01
kernel/launch	2349.5555555555557 ns	2293.1111111111113 ns	1.02
kernel/occupancy	880.9137931034483 ns	841.4430379746835 ns	1.05
kernel/rand	14308 ns	16464 ns	0.87
latency/import	4201203057 ns	4216998657 ns	1.00
latency/precompile	4955314654 ns	4950855891 ns	1.00
latency/ttfp	4654533020 ns	4698355034 ns	0.99

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.56%. Comparing base (c34b904) to head (304a683).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3267      +/-   ##
==========================================
- Coverage   92.58%   92.56%   -0.02%     
==========================================
  Files         188      188              
  Lines       16471    16471              
==========================================
- Hits        15249    15246       -3     
- Misses       1222     1225       +3     

☔ 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.