Skip to content
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

Jump to bottom

Mark cuarray as noalias #2395

Merged
merged 2 commits into from
May 28, 2024
Merged

Mark cuarray as noalias #2395

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b9bc10e
Mark cuarray as noalias
wsmoses May 24, 2024
66ff78c
Add forward broadcast test
wsmoses May 25, 2024
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
Prev Previous commit
Add forward broadcast test
  • Loading branch information
@wsmoses
wsmoses committed May 27, 2024
commit 66ff78c6389158eebb1e2f5bebd3d5bfdf2c95b9
60 changes: 60 additions & 0 deletions ext/EnzymeCoreExt.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,6 +99,32 @@ function EnzymeCore.EnzymeRules.forward(ofn::Const{Type{CT}},
end
end

function EnzymeCore.EnzymeRules.forward(ofn::Const{Type{CT}},
::Type{RT}, uval::EnzymeCore.Annotation{DR}, args...; kwargs...) where {CT <: CuArray, DR <: CUDA.DataRef, RT}
primargs = ntuple(Val(length(args))) do i
Base.@_inline_meta
args[i].val
end
if RT <: Duplicated
shadow = ofn.val(uval.val, primargs...; kwargs...)
Duplicated(ofn.val(uval.dval, primargs...; kwargs...), shadow)
elseif RT <: Const
ofn.val(uval.val, primargs...; kwargs...)
elseif RT <: DuplicatedNoNeed
ofn.val(uval.dval, primargs...; kwargs...)
else
tup = ntuple(Val(EnzymeCore.batch_size(RT))) do i
Base.@_inline_meta
shadow = ofn.val(uval.dval[i], primargs...; kwargs...)
end
if RT <: BatchDuplicated
BatchDuplicated(ofv.val(uval.val), tup)
else
tup
end
end
end

function EnzymeCore.EnzymeRules.forward(ofn::Const{typeof(synchronize)},
::Type{RT}, args::Vararg{EnzymeCore.Annotation, N}; kwargs...) where {RT, N}
pargs = ntuple(Val(N)) do i
Expand Down Expand Up @@ -270,6 +296,40 @@ function EnzymeCore.EnzymeRules.reverse(config, ofn::Const{Type{CT}}, ::Type{RT}
end
end

function EnzymeCore.EnzymeRules.augmented_primal(config, ofn::Const{Type{CT}}, ::Type{RT}, uval::EnzymeCore.Annotation{DR}, args...; kwargs...) where {CT <: CuArray, DR <: CUDA.DataRef, RT}
primargs = ntuple(Val(length(args))) do i
Base.@_inline_meta
args[i].val
end

primal = if EnzymeRules.needs_primal(config)
ofn.val(uval.val, primargs...; kwargs...)
else
nothing
end

shadow = if EnzymeRules.needs_shadow(config)
if EnzymeRules.width(config) == 1
ofn.val(uval.dval, primargs...; kwargs...)
else
ntuple(Val(EnzymeRules.width(config))) do i
Base.@_inline_meta
ofn.val(uval.dval[i], primargs...; kwargs...)
end
end
else
nothing
end
return EnzymeRules.AugmentedReturn{(EnzymeRules.needs_primal(config) ? CT : Nothing), (EnzymeRules.needs_shadow(config) ? (EnzymeRules.width(config) == 1 ? CT : NTuple{EnzymeRules.width(config), CT}) : Nothing), Nothing}(primal, shadow, nothing)
end

function EnzymeCore.EnzymeRules.reverse(config, ofn::Const{Type{CT}}, ::Type{RT}, tape, A::EnzymeCore.Annotation{DR}, args::Vararg{EnzymeCore.Annotation, N}; kwargs...) where {CT <: CuArray, DR <: CUDA.DataRef, RT, N}
ntuple(Val(N+1)) do i
Base.@_inline_meta
nothing
end
end

function EnzymeCore.EnzymeRules.noalias(::Type{CT}, ::UndefInitializer, args...) where {CT <: CuArray}
return nothing
end
Expand Down
12 changes: 12 additions & 0 deletions test/extensions/enzyme.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,6 +70,18 @@ end
@test all(shad .≈ 0.0)
end

firstsum(x, y) = first(x .+ y)
@testset "Forward broadcast" begin
x = CuArray(5*ones(5))
y = CuArray(3*ones(5))
dx = CuArray([1.0, 0.0, 0.0, 0.0, 0.0])
dy = CuArray([0.2, 0.0, 0.1, 0.0, 0.0])
f(x, y)
res = autodiff(Forward, firstsum, Duplicated, Duplicated(x, dx), Duplicated(y, dy))
@test res[1] ≈ 8
@test res[2] ≈ 1.2
end

# TODO once reverse kernels are in
# function togpu(x)
# x = CuArray(x)
Expand Down