Skip to content

Support KA warp groupreduce API #729

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

Closed
wants to merge 6 commits into from
Closed

Support KA warp groupreduce API #729

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
94fea61
Support KA groupreduce API
pxl-th Jan 30, 2025
ae026d9
Cleanup
pxl-th Jan 30, 2025
fd614f0
API changes
pxl-th Feb 3, 2025
9822426
tmp
pxl-th Feb 20, 2025
7854159
cleanup
pxl-th Feb 20, 2025
9f1c5ab
Cleanup
pxl-th Feb 28, 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
10 changes: 10 additions & 0 deletions src/ROCKernels.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -166,4 +166,14 @@ end
# TODO
end

# Reduction.

@device_override @inline KA.supports_warp_reduction() = true

KA.supports_warp_reduction(::ROCBackend) = true

function KA.shfl_down(val, offset)
AMDGPU.Device.shfl_down(val, offset)
end

end
16 changes: 7 additions & 9 deletions src/device/gcn/wavefront.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,22 +57,20 @@ end
"""
wfred(op::Function, val::T) where T -> T

Performs a wavefront-wide reduction on `val` in each lane, and returns the
result. A limited subset of functions are available to be passed as `op`. When
`op` is one of `(+, max, min, &, |, ⊻)`, `T` may be
`<:Union{Cint, Clong, Cuint, Culong}`. When `op` is one of `(+, max, min)`,
`T` may also be `<:Union{Float32, Float64}`.
Performs a wavefront-wide reduction on `val` in each lane, and returns the result.
A limited subset of functions are available to be passed as `op`.
When `op` is one of `(+, max, min, &, |, ⊻)`, `T` may be `<:Union{Cint, Clong, Cuint, Culong}`.
When `op` is one of `(+, max, min)`, `T` may also be `<:Union{Float32, Float64}`.
"""
wfred

"""
wfscan(op::Function, val::T) where T -> T

Performs a wavefront-wide scan on `val` in each lane, and returns the
result. A limited subset of functions are available to be passed as `op`. When
`op` is one of `(+, max, min, &, |, ⊻)`, `T` may be
`<:Union{Cint, Clong, Cuint, Culong}`. When `op` is one of `(+, max, min)`,
`T` may also be `<:Union{Float32, Float64}`.
result. A limited subset of functions are available to be passed as `op`.
When `op` is one of `(+, max, min, &, |, ⊻)`, `T` may be `<:Union{Cint, Clong, Cuint, Culong}`.
When `op` is one of `(+, max, min)`, `T` may also be `<:Union{Float32, Float64}`.
"""
wfscan

Expand Down