Skip to content

CUDA: Fallback to UnsafeAtomics #56

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 4 commits into
base: vc/cuda_atomicswap
Choose a base branch
from
Open

CUDA: Fallback to UnsafeAtomics #56

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
15d75b9
CUDA: Fallback to UnsafeAtomics
vchuravy May 13, 2025
19bd807
use monotonic order
vchuravy May 13, 2025
16bd0fe
UnsafeAtomics defined in Internal
vchuravy May 13, 2025
a8ec317
enable test
vchuravy May 13, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
57 changes: 54 additions & 3 deletions ext/AtomixCUDAExt.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,16 +2,67 @@
module AtomixCUDAExt

using Atomix: Atomix, IndexableRef
using Atomix.Internal: UnsafeAtomics
using CUDA: CUDA, CuDeviceArray

const CuIndexableRef{Indexable<:CuDeviceArray} = IndexableRef{Indexable}

# from https://github.com/JuliaGPU/CUDA.jl/pull/1644
# function atomic_load(ptr::LLVMPtr{T}, order, scope::System=System()) where T
# if order == Acq_Rel() || order == Release()
# assert(false)
# end
# if compute_capability() >= sv"7.0"
# if order == Relaxed()
# val = __load(ptr, Relaxed(), scope)
# return val
# end
# if order == Seq_Cst()
# atomic_thread_fence(Seq_Cst(), scope)
# end
# val = __load(ptr, Acquire(), scope)
# return val
# else
# if order == Seq_Cst()
# atomic_thread_fence(Seq_Cst(), scope)
# end
# val = __load_volatile(ptr)
# if order == Relaxed()
# return val
# end
# atomic_thread_fence(order, scope)
# return val
# end
# end

function Atomix.get(ref::CuIndexableRef, order)
error("not implemented")
ptr = Atomix.pointer(ref)
return UnsafeAtomics.load(ptr, UnsafeAtomics.monotonic)
end

# function atomic_store!(ptr::LLVMPtr{T}, val::T, order, scope::System=System()) where T
# if order == Acq_Rel() || order == Consume() || order == Acquire()
# assert(false)
# end
# if compute_capability() >= sv"7.0"
# if order == Release()
# __store!(ptr, val, Release(), scope)
# return
# end
# if order == Seq_Cst()
# atomic_thread_fence(Seq_Cst(), scope)
# end
# __store!(ptr, val, Relaxed(), scope)
# else
# if order == Seq_Cst()
# atomic_thread_fence(Seq_Cst(), scope)
# end
# __store_volatile!(ptr, val)
# end
# end
function Atomix.set!(ref::CuIndexableRef, v, order)
error("not implemented")
ptr = Atomix.pointer(ref)
return UnsafeAtomics.store!(ptr, v, UnsafeAtomics.monotonic)
end

@inline function Atomix.replace!(
Expand Down Expand Up @@ -51,7 +102,7 @@ end
elseif op === Atomix.right
CUDA.atomic_xchg!(ptr, x)
else
error("not implemented")
return UnsafeAtomics.modify(ptr, op, x, UnsafeAtomics.monotonic)
end
end
return old => op(old, x)
Expand Down
7 changes: 1 addition & 6 deletions test/test_atomix_cuda.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,10 +15,7 @@ function cuda(f)
CUDA.@cuda g()
end


# Not implemented:
#=
function test_get_set()
@testset "AtomixCUDAExt:test_get_set" begin
A = CUDA.ones(Int, 3)
cuda() do
GC.@preserve A begin
Expand All @@ -29,8 +26,6 @@ function test_get_set()
end
@test collect(A) == [-1, 1, 1]
end
=#


@testset "AtomixCUDAExt:test_cas" begin
idx = (
Expand Down
Loading