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

Make vectorized store convert and perform multiple stores if required #111

Merged
merged 4 commits into from
Jun 29, 2023
Merged

Make vectorized store convert and perform multiple stores if required #111

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
e692d53
Convert values when doing a vectorized store.
maleadt Jun 27, 2023
17e1a0d
Perform multiple vectorized stores when there's a type mismatch.
maleadt Jun 27, 2023
69e97c1
Move the loop into vstorea.
maleadt Jun 29, 2023
2b25c99
Manually unroll generated code.
maleadt Jun 29, 2023
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
22 changes: 17 additions & 5 deletions src/layout.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@ module Layout
using CUDA
using LLVMLoopInfo: @loopinfo
using GemmKernels.Tiling
using Base.Cartesian: @ntuple

# ---------------------
# Customise computation
Expand All @@ -17,7 +18,8 @@ using GemmKernels.Tiling

struct Vec{N, T} end

@inline @generated function vloada(::Type{Vec{N, T}}, ptr::Core.LLVMPtr{T, AS}, i::Integer = 1) where {N, T, AS}
@inline @generated function vloada(::Type{Vec{N, T}}, ptr::Core.LLVMPtr{T, AS},
i::Integer = 1) where {N, T, AS}
alignment = sizeof(T) * N

return quote
Expand All @@ -26,13 +28,23 @@ struct Vec{N, T} end
end
end

@inline @generated function vstorea!(::Type{Vec{N, T}}, ptr::Core.LLVMPtr{T, AS}, x, i::Integer = 1) where {N, T, AS}
@inline @generated function vstorea!(::Type{Vec{N, T}}, ptr::Core.LLVMPtr{T, AS},
x::NTuple{M,<:Any}, i::Integer = 1) where {N, T, AS, M}
alignment = sizeof(T) * N

return quote
vec_ptr = Base.bitcast(Core.LLVMPtr{NTuple{N, VecElement{T}}, AS}, ptr)
return unsafe_store!(vec_ptr, x, (i-1) ÷ N + 1, Val($alignment))
ex = quote end

# we may be storing more values than we can using a single vectorized operation
# (e.g., when types mismatch, storing 8 Float16s in a Float32 shared memory layout)
for offset = 0:N:M-1
append!(ex.args, (quote
y = @ntuple $N j -> VecElement{T}(x[j+$offset].value)
vec_ptr = Base.bitcast(Core.LLVMPtr{NTuple{N, VecElement{T}}, AS}, ptr)
unsafe_store!(vec_ptr, y, (i+$offset-1) ÷ N + 1, Val($alignment))
end).args)
end

return ex
end

# -----------
Expand Down