Broadcasting on arrays larger than typemax(Int32) yields truncation error

[b-fg]

Describe the bug

When broadcasting on arrays larger than typemax(Int32), a truncation error pops up:

Error

ERROR: InexactError: trunc(Int32, 2254857829)
Stacktrace:
  [1] throw_inexacterror(::Symbol, ::Vararg{Any})
    @ Core ./boot.jl:750
  [2] checked_trunc_sint
    @ ./boot.jl:764 [inlined]
  [3] toInt32
    @ ./boot.jl:801 [inlined]
  [4] Int32
    @ ./boot.jl:891 [inlined]
  [5] convert
    @ ./number.jl:7 [inlined]
  [6] cconvert
    @ ./essentials.jl:687 [inlined]
  [7] macro expansion
    @ ~/.julia-acc/packages/CUDA/1kIOw/lib/utils/call.jl:222 [inlined]
  [8] macro expansion
    @ ~/.julia-acc/packages/CUDA/1kIOw/lib/cudadrv/libcuda.jl:5139 [inlined]
  [9] #735
    @ ~/.julia-acc/packages/CUDA/1kIOw/lib/utils/call.jl:35 [inlined]
 [10] check
    @ ~/.julia-acc/packages/CUDA/1kIOw/lib/cudadrv/libcuda.jl:35 [inlined]
 [11] cuOccupancyMaxPotentialBlockSize
    @ ~/.julia-acc/packages/CUDA/1kIOw/lib/utils/call.jl:34 [inlined]
 [12] launch_configuration(fun::CuFunction; shmem::Int64, max_threads::Int64)
    @ CUDA ~/.julia-acc/packages/CUDA/1kIOw/lib/cudadrv/occupancy.jl:61
 [13] launch_configuration
    @ ~/.julia-acc/packages/CUDA/1kIOw/lib/cudadrv/occupancy.jl:56 [inlined]
 [14] (::KernelAbstractions.Kernel{…})(::CuArray{…}, ::Vararg{…}; ndrange::Tuple{…}, workgroupsize::Nothing)
    @ CUDA.CUDAKernels ~/.julia-acc/packages/CUDA/1kIOw/src/CUDAKernels.jl:107
 [15] _copyto!
    @ ~/.julia-acc/packages/GPUArrays/uiVyU/src/host/broadcast.jl:71 [inlined]
 [16] materialize!
    @ ~/.julia-acc/packages/GPUArrays/uiVyU/src/host/broadcast.jl:38 [inlined]
 [17] materialize!(dest::CuArray{Float32, 1, CUDA.DeviceMemory}, bc::Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{0}, Nothing, typeof(identity), Tuple{Float32}})
    @ Base.Broadcast ./broadcast.jl:875
 [18] top-level scope
    @ REPL[3]:1
Some type information was truncated. Use `show(err)` to see complete types.

To reproduce

using CUDA
a = CUDA.rand(Float32, floor(Int, typemax(Int32)*1.05));
a .= zero(Float32);

Note that a GPU with memory >8GB is required to reproduce this test. I am using CUDA.jl v5.6.1:

(CUDAtest) pkg> st
Project CUDAtest v0.1.0
Status `/gpfs/home/delf/delf428444/CUDAtest/Project.toml`
  [052768ef] CUDA v5.6.1

Expected behavior

No truncation error should appear when broadcasting, even for arrays larger than typemax(Int32).

Version info
Julia

Julia Version 1.11.2
Commit 5e9a32e7af2 (2024-12-01 20:02 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 160 × Intel(R) Xeon(R) Platinum 8460Y+
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, sapphirerapids)
Threads: 160 default, 0 interactive, 80 GC (on 160 virtual cores)
Environment:
  JULIA_MPI_BINARY = system
  JULIA_DEPOT_PATH = /home/delf/delf428444/.julia-acc
  LD_LIBRARY_PATH =
  JULIA_NUM_THREADS = auto

CUDA.jl

CUDA runtime 12.6, artifact installation
CUDA driver 12.6
NVIDIA driver 535.86.10

CUDA libraries:
- CUBLAS: 12.6.4
- CURAND: 10.3.7
- CUFFT: 11.3.0
- CUSOLVER: 11.7.1
- CUSPARSE: 12.5.4
- CUPTI: 2024.3.2 (API 24.0.0)
- NVML: 12.0.0+535.86.10

Julia packages:
- CUDA: 5.6.1
- CUDA_Driver_jll: 0.10.4+0
- CUDA_Runtime_jll: 0.15.5+0

Toolchain:
- Julia: 1.11.2
- LLVM: 16.0.6

4 devices:
  0: NVIDIA H100 (sm_90, 54.390 GiB / 63.718 GiB available)
  1: NVIDIA H100 (sm_90, 63.424 GiB / 63.718 GiB available)
  2: NVIDIA H100 (sm_90, 63.424 GiB / 63.718 GiB available)
  3: NVIDIA H100 (sm_90, 63.422 GiB / 63.718 GiB available)

[b-fg]

Probably related: #1968

[b-fg]

Thanks!