Improve Int128 ABI mismatch detection.

[maleadt]

For some context on why we started doing this in the first place: The NVPTX back-end aligns 128-bit integers to 16 bytes. Julia only started doing that in 1.12; on 1.10 and 1.11, Int128 is 8-byte aligned. So on those versions an aggregate with an (U)Int128 field has a different layout on the host than on the device. struct {Int64; Int128} is 24 bytes on the host (field at offset 8) and 32 bytes on the device (offset 16), and codegen always uses the device layout, so the two disagree.

[github-actions]

CUDA.jl Benchmarks

Details

Benchmark suite	Current: 34dc577	Previous: 1caa3ad	Ratio
array/accumulate/Float32/1d	99383 ns	100770 ns	0.99
array/accumulate/Float32/dims=1	75149 ns	76381 ns	0.98
array/accumulate/Float32/dims=1L	1629231 ns	1630799 ns	1.00
array/accumulate/Float32/dims=2	140654 ns	141553 ns	0.99
array/accumulate/Float32/dims=2L	653360 ns	653429 ns	1.00
array/accumulate/Int64/1d	118282 ns	119282 ns	0.99
array/accumulate/Int64/dims=1	78853 ns	80346 ns	0.98
array/accumulate/Int64/dims=1L	1724276 ns	1726167 ns	1.00
array/accumulate/Int64/dims=2	153119 ns	154865 ns	0.99
array/accumulate/Int64/dims=2L	959403 ns	960762 ns	1.00
array/broadcast	18258 ns	18636 ns	0.98
array/construct	1213.6 ns	1242.5 ns	0.98
array/copy	16518 ns	16748 ns	0.99
array/copyto!/cpu_to_gpu	212993 ns	214495 ns	0.99
array/copyto!/gpu_to_cpu	280084 ns	281315 ns	1.00
array/copyto!/gpu_to_gpu	10334 ns	10356 ns	1.00
array/iteration/findall/bool	132639 ns	134688 ns	0.98
array/iteration/findall/int	146104 ns	148663 ns	0.98
array/iteration/findfirst/bool	68874 ns	70356 ns	0.98
array/iteration/findfirst/int	70427 ns	71778 ns	0.98
array/iteration/findmin/1d	65071 ns	68513 ns	0.95
array/iteration/findmin/2d	101226 ns	101687 ns	1.00
array/iteration/logical	189699 ns	195448 ns	0.97
array/iteration/scalar	64535 ns	66072 ns	0.98
array/permutedims/2d	49387 ns	50059 ns	0.99
array/permutedims/3d	50674 ns	50890 ns	1.00
array/permutedims/4d	50651 ns	51304 ns	0.99
array/random/rand/Float32	11381 ns	11937 ns	0.95
array/random/rand/Int64	22105 ns	24454 ns	0.90
array/random/rand!/Float32	7847 ns	8092.666666666667 ns	0.97
array/random/rand!/Int64	17412 ns	20949 ns	0.83
array/random/randn/Float32	34683 ns	36329 ns	0.95
array/random/randn!/Float32	23723 ns	24624 ns	0.96
array/reductions/mapreduce/Float32/1d	33101 ns	33603 ns	0.99
array/reductions/mapreduce/Float32/dims=1	38329 ns	38517 ns	1.00
array/reductions/mapreduce/Float32/dims=1L	50058 ns	50114 ns	1.00
array/reductions/mapreduce/Float32/dims=2	55785 ns	56468 ns	0.99
array/reductions/mapreduce/Float32/dims=2L	66656 ns	67374 ns	0.99
array/reductions/mapreduce/Int64/1d	38514 ns	39976 ns	0.96
array/reductions/mapreduce/Int64/dims=1	41371 ns	41660 ns	0.99
array/reductions/mapreduce/Int64/dims=1L	86330 ns	86590 ns	1.00
array/reductions/mapreduce/Int64/dims=2	58183 ns	59096 ns	0.98
array/reductions/mapreduce/Int64/dims=2L	81985 ns	82999 ns	0.99
array/reductions/reduce/Float32/1d	33246 ns	33460 ns	0.99
array/reductions/reduce/Float32/dims=1	38133 ns	38736 ns	0.98
array/reductions/reduce/Float32/dims=1L	50073 ns	50547 ns	0.99
array/reductions/reduce/Float32/dims=2	55632 ns	56305 ns	0.99
array/reductions/reduce/Float32/dims=2L	66888 ns	67363 ns	0.99
array/reductions/reduce/Int64/1d	39180 ns	40096 ns	0.98
array/reductions/reduce/Int64/dims=1	40997 ns	41717 ns	0.98
array/reductions/reduce/Int64/dims=1L	86233 ns	86750 ns	0.99
array/reductions/reduce/Int64/dims=2	57731 ns	58479 ns	0.99
array/reductions/reduce/Int64/dims=2L	82615 ns	83983 ns	0.98
array/reverse/1d	16748 ns	17202 ns	0.97
array/reverse/1dL	67810 ns	68074 ns	1.00
array/reverse/1dL_inplace	65177 ns	65359 ns	1.00
array/reverse/1d_inplace	8217.666666666666 ns	8376.666666666666 ns	0.98
array/reverse/2d	19882 ns	20409 ns	0.97
array/reverse/2dL	71655 ns	72447 ns	0.99
array/reverse/2dL_inplace	65002 ns	64948 ns	1.00
array/reverse/2d_inplace	9597 ns	9770 ns	0.98
array/sorting/1d	2655304 ns	2653489 ns	1.00
array/sorting/2d	1032380 ns	1034091 ns	1.00
array/sorting/by	3180105 ns	3181225 ns	1.00
cuda/synchronization/context/auto	1128.8 ns	1135.5 ns	0.99
cuda/synchronization/context/blocking	894.1276595744681 ns	939.7407407407408 ns	0.95
cuda/synchronization/context/nonblocking	6043 ns	6181.8 ns	0.98
cuda/synchronization/stream/auto	994.8181818181819 ns	991.6 ns	1.00
cuda/synchronization/stream/blocking	799.989247311828 ns	841.3783783783783 ns	0.95
cuda/synchronization/stream/nonblocking	6045 ns	6139.8 ns	0.98
integration/byval/reference	143105 ns	143264 ns	1.00
integration/byval/slices=1	145192 ns	145361 ns	1.00
integration/byval/slices=2	283765 ns	283971 ns	1.00
integration/byval/slices=3	422310 ns	422479 ns	1.00
integration/cudadevrt	101577 ns	101731 ns	1.00
integration/volumerhs	8988630 ns	8999650 ns	1.00
kernel/indexing	12590 ns	12703 ns	0.99
kernel/indexing_checked	13321 ns	13354 ns	1.00
kernel/launch	2089.3333333333335 ns	2086 ns	1.00
kernel/occupancy	734.3741007194244 ns	693.3698630136986 ns	1.06
kernel/rand	15863 ns	15890 ns	1.00
latency/import	3864457489 ns	3853824190 ns	1.00
latency/precompile	4639386595 ns	4635422433 ns	1.00
latency/ttfp	4559920169 ns	4550893912 ns	1.00

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 16.31%. Comparing base (1caa3ad) to head (34dc577).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3170      +/-   ##
==========================================
+ Coverage   10.82%   16.31%   +5.48%     
==========================================
  Files         123      124       +1     
  Lines        9479     9875     +396     
==========================================
+ Hits         1026     1611     +585     
+ Misses       8453     8264     -189     

☔ 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.