Skip to content

improve performance using CUDA performance tips #69

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
matthieugomez wants to merge 1 commit into from
Closed

improve performance using CUDA performance tips #69

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
27 changes: 16 additions & 11 deletions ext/CUDAExt.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,10 +52,12 @@ function FixedEffects.gather!(fecoef::CuVector, refs::CuVector, α::Number, y::C
end

function gather_kernel!(fecoef, refs, α, y, cache)
index = (blockIdx().x - 1) * blockDim().x + threadIdx().x
index = (blockIdx().x - Int32(1)) * blockDim().x + threadIdx().x
stride = blockDim().x * gridDim().x
@inbounds for i = index:stride:length(y)
i = index
@inbounds while i <= length(y)
CUDA.@atomic fecoef[refs[i]] += α * y[i] * cache[i]
i += stride
end
end

Expand All @@ -65,10 +67,12 @@ function FixedEffects.scatter!(y::CuVector, α::Number, fecoef::CuVector, refs::
end

function scatter_kernel!(y, α, fecoef, refs, cache)
index = (blockIdx().x - 1) * blockDim().x + threadIdx().x
index = (blockIdx().x - Int32(1)) * blockDim().x + threadIdx().x
stride = blockDim().x * gridDim().x
@inbounds for i = index:stride:length(y)
i = index
@inbounds while i <= length(y)
y[i] += α * fecoef[refs[i]] * cache[i]
i += stride
end
end

Expand Down Expand Up @@ -128,10 +132,12 @@ function scale!(scale::CuVector, refs::CuVector, interaction::CuVector, weights:
end

function scale_kernel!(scale, refs, interaction, weights)
index = (blockIdx().x - 1) * blockDim().x + threadIdx().x
index = (blockIdx().x - Int32(1)) * blockDim().x + threadIdx().x
stride = blockDim().x * gridDim().x
@inbounds for i = index:stride:length(interaction)
i = index
@inbounds while i <= length(interaction)
CUDA.@atomic scale[refs[i]] += abs2(interaction[i]) * weights[i]
i += stride
end
end

Expand All @@ -141,13 +147,12 @@ function cache!(cache::CuVector, refs::CuVector, interaction::CuVector, weights:
end

function cache!_kernel!(cache, refs, interaction, weights, scale)
index = (blockIdx().x - 1) * blockDim().x + threadIdx().x
index = (blockIdx().x - Int32(1)) * blockDim().x + threadIdx().x
stride = blockDim().x * gridDim().x
@inbounds for i = index:stride:length(cache)
i = index
@inbounds while i <= length(cache)
cache[i] = interaction[i] * sqrt(weights[i]) * scale[refs[i]]
i += stride
end
end



end
Loading