Segmentation fault when converting BFloat16 to Float32

[findmyway]

julia> versioninfo()
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: 192 × Intel(R) Xeon(R) Platinum 8468
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, sapphirerapids)
Threads: 1 default, 0 interactive, 1 GC (on 192 virtual cores)
Environment:
  LD_LIBRARY_PATH = /usr/local/nccl-rdma-sharp-plugins/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64                  |

julia> using BFloat16s

julia> x = BFloat16[0]
1-element Vector{BFloat16}:
 0.0

julia> Float32.(x)
[1]    26383 segmentation fault (core dumped)  julia

[inkydragon]

Cannot be reproduced on Win11 and WSL.
Maybe the issue is related to the system or the CPU.

Details

Win11

julia> using BFloat16s

julia> x = BFloat16[0]
1-element Vector{BFloat16}:
 0.0

julia> Float32.(x)
1-element Vector{Float32}:
 0.0

julia> versioninfo()
Julia Version 1.11.2
Commit 5e9a32e7af (2024-12-01 20:02 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 32 × 13th Gen Intel(R) Core(TM) i9-13900HX
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, alderlake)
Threads: 1 default, 0 interactive, 1 GC (on 32 virtual cores)

WSL (Ubuntu 22.04.5 LTS)

julia> using BFloat16s

julia> x = BFloat16[0]
1-element Vector{BFloat16}:
 0.0

julia> Float32.(x)
1-element Vector{Float32}:
 0.0

julia> versioninfo()
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: 32 × 13th Gen Intel(R) Core(TM) i9-13900HX
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, alderlake)
Threads: 1 default, 0 interactive, 1 GC (on 32 virtual cores)

BFloat16s.jl/src/bfloat16.jl

Lines 210 to 223 in c468ba4

	if llvm_arithmetic
	Base.Float32(x::BFloat16) = Base.fpext(Float32, x)
	Base.Float64(x::BFloat16) = Base.fpext(Float64, x)
	else
	# Expansion to Float32
	function Base.Float32(x::BFloat16)
	reinterpret(Float32, UInt32(reinterpret(Unsigned, x)) << 16)
	end
	# Expansion to Float64
	function Base.Float64(x::BFloat16)
	Float64(Float32(x))
	end
	end

Does the following code trigger the problem and what is your output?

Float32(x[1])
BFloat16s.llvm_arithmetic
Base.fpext.(Float32, x)
@. reinterpret(Float32, UInt32(reinterpret(UInt16, x)) << 16)

Base.fpext isa Core.IntrinsicFunction.
I'm guessing this is a julia bug.

[findmyway]

This only occurs on my Linux server. On my local mac environment, it works fine. Should we create an issue under the Julia repo?