Various performance and memory optimizations

.buildkite/pipeline.yml

@@ -69,6 +69,20 @@ steps:
  julia_args: "--threads=1"
  - JuliaCI/julia-coverage#v1:
  codecov: true
+ - label: Julia 1.9 (macOS)
+ timeout_in_minutes: 90
+ <<: *test
+ agents:
+ queue: "juliaecosystem"
+ os: macos
+ arch: x86_64
+ plugins:
+ - JuliaCI/julia#v1:
+ version: "1.9"
+ - JuliaCI/julia-test#v1:
+ julia_args: "--threads=1"
+ - JuliaCI/julia-coverage#v1:
+ codecov: true
  - label: Julia 1.8 - TimespanLogging
  timeout_in_minutes: 20
  <<: *test

docs/src/task-spawning.md

@@ -39,7 +39,6 @@ key-value pairs, which can be any of:
 
 There are also some extra options that can be passed, although they're considered advanced options to be used only by developers or library authors:
 - `get_result::Bool` -- return the actual result to the scheduler instead of [`Chunk`](@ref) objects. Used when `f` explicitly constructs a [`Chunk`](@ref) or when return value is small (e.g. in case of reduce)
-- `persist::Bool` -- the result of this Thunk should not be released after it becomes unused in the DAG
 - `cache::Bool` -- cache the result of this Thunk such that if the thunk is evaluated again, one can just reuse the cached value. If it’s been removed from cache, recompute the value.
 
 ## Simple example

src/Dagger.jl

@@ -21,6 +21,7 @@ if !isdefined(Base, :ScopedValues)
 else
  import Base.ScopedValues: ScopedValue, with
 end
+import TaskLocalValues: TaskLocalValue
 
 if !isdefined(Base, :get_extension)
  import Requires: @require
@@ -31,27 +32,36 @@ import TimespanLogging: timespan_start, timespan_finish
 
 import Adapt
 
+import MacroTools: @capture
+
 include("lib/util.jl")
 include("utils/dagdebug.jl")
 
+# Logging Basics
+include("utils/logging.jl")
+
 # Distributed data
 include("utils/locked-object.jl")
 include("utils/tasks.jl")
-
-import MacroTools: @capture
-include("options.jl")
+include("utils/reuse.jl")
 include("processor.jl")
 include("threadproc.jl")
+include("sch_options.jl")
 include("context.jl")
 include("utils/processors.jl")
 include("task-tls.jl")
 include("scopes.jl")
 include("utils/scopes.jl")
+include("chunks.jl")
+include("utils/signature.jl")
+include("options.jl")
 include("dtask.jl")
+include("argument.jl")
 include("queue.jl")
 include("thunk.jl")
+include("utils/fetch.jl")
+include("utils/chunks.jl")
 include("submission.jl")
-include("chunks.jl")
 include("memory-spaces.jl")
 
 # Task scheduling
@@ -64,34 +74,34 @@ include("sch/Sch.jl"); using .Sch
 # Data dependency task queue
 include("datadeps.jl")
 
+# File IO
+include("file-io.jl")
+
 # Array computations
 include("array/darray.jl")
 include("array/alloc.jl")
 include("array/map-reduce.jl")
 include("array/copy.jl")
-
-# File IO
-include("file-io.jl")
-
+include("array/random.jl")
 include("array/operators.jl")
 include("array/indexing.jl")
 include("array/setindex.jl")
 include("array/matrix.jl")
 include("array/sparse_partition.jl")
+include("array/parallel-blocks.jl")
 include("array/sort.jl")
 include("array/linalg.jl")
 include("array/mul.jl")
 include("array/cholesky.jl")
 
+# Custom Logging Events
+include("utils/logging-events.jl")
+
 # Visualization
 include("visualization.jl")
 include("ui/gantt-common.jl")
 include("ui/gantt-text.jl")
 
-# Logging
-include("utils/logging-events.jl")
-include("utils/logging.jl")
-
 # Precompilation
 import PrecompileTools: @compile_workload
 include("precompile.jl")

src/argument.jl

@@ -0,0 +1,44 @@
+mutable struct ArgPosition
+ positional::Bool
+ idx::Int
+ kw::Symbol
+end
+ArgPosition() = ArgPosition(true, 0, :NULL)
+ArgPosition(pos::ArgPosition) = ArgPosition(pos.positional, pos.idx, pos.kw)
+ispositional(pos::ArgPosition) = pos.positional
+iskw(pos::ArgPosition) = !pos.positional
+function pos_idx(pos::ArgPosition)
+ @assert pos.positional
+ @assert pos.idx > 0
+ @assert pos.kw == :NULL
+ return pos.idx
+end
+function pos_kw(pos::ArgPosition)
+ @assert !pos.positional
+ @assert pos.idx == 0
+ @assert pos.kw != :NULL
+ return pos.kw
+end
+mutable struct Argument
+ pos::ArgPosition
+ value
+end
+Argument(pos::Integer, value) = Argument(ArgPosition(true, pos, :NULL), value)
+Argument(kw::Symbol, value) = Argument(ArgPosition(false, 0, kw), value)
+ispositional(arg::Argument) = ispositional(arg.pos)
+iskw(arg::Argument) = iskw(arg.pos)
+pos_idx(arg::Argument) = pos_idx(arg.pos)
+pos_kw(arg::Argument) = pos_kw(arg.pos)
+value(arg::Argument) = arg.value
+valuetype(arg::Argument) = typeof(arg.value)
+Base.iterate(arg::Argument) = (arg.pos, true)
+function Base.iterate(arg::Argument, state::Bool)
+ if state
+ return (arg.value, false)
+ else
+ return nothing
+ end
+end
+
+Base.copy(arg::Argument) = Argument(ArgPosition(arg.pos), arg.value)
+chunktype(arg::Argument) = chunktype(value(arg))

src/array/darray.jl

@@ -173,7 +173,7 @@ domainchunks(d::DArray) = d.subdomains
 size(x::DArray) = size(domain(x))
 stage(ctx, c::DArray) = c
 
-function Base.collect(d::DArray; tree=false)
+function Base.collect(d::DArray{T,N}; tree=false, copyto=false) where {T,N}
  a = fetch(d)
  if isempty(d.chunks)
  return Array{eltype(d)}(undef, size(d)...)
@@ -183,6 +183,13 @@ function Base.collect(d::DArray; tree=false)
  return fetch(a.chunks[1])
  end
 
+ if copyto
+ C = Array{T,N}(undef, size(a))
+ DC = view(C, Blocks(size(a)...))
+ copyto!(DC, a)
+ return C
+ end
+
  dimcatfuncs = [(x...) -> d.concat(x..., dims=i) for i in 1:ndims(d)]
  if tree
  collect(fetch(treereduce_nd(map(x -> ((args...,) -> Dagger.@spawn x(args...)) , dimcatfuncs), a.chunks)))

src/array/indexing.jl

@@ -1,5 +1,3 @@
-import TaskLocalValues: TaskLocalValue
-
 ### getindex
 
 struct GetIndex{T,N} <: ArrayOp{T,N}

src/array/parallel-blocks.jl

@@ -0,0 +1,174 @@
+export ParallelBlocks
+
+using Statistics
+
+struct ParallelBlocks{N} <: Dagger.AbstractSingleBlocks{N}
+ n::Int
+end
+ParallelBlocks(n::Integer) = ParallelBlocks{0}(n)
+ParallelBlocks{N}(dist::ParallelBlocks) where N = ParallelBlocks{N}(dist.n)
+ParallelBlocks() = ParallelBlocks(Dagger.num_processors())
+
+Base.convert(::Type{ParallelBlocks{N}}, dist::ParallelBlocks) where N =
+ ParallelBlocks{N}(dist.n)
+
+wrap_chunks(chunks::Vector{<:Dagger.Chunk}, N::Integer, dist::ParallelBlocks) =
+ wrap_chunks(chunks, N, dist.n)
+wrap_chunks(chunks::Vector{<:Dagger.Chunk}, N::Integer, n::Integer) =
+ convert(Array{Any}, reshape(chunks, ntuple(i->i == 1 ? n : 1, N)))
+
+function _finish_allocation(f::Function, dist::ParallelBlocks, dims::NTuple{N,Int}) where N
+ d = ArrayDomain(map(x->1:x, dims))
+ s = reshape([d for _ in 1:dist.n],
+ ntuple(i->i == 1 ? dist.n : 1, N))
+ data = [f(dims) for _ in 1:dist.n]
+ dist = ParallelBlocks{N}(dist)
+ chunks = wrap_chunks(map(Dagger.tochunk, data), N, dist)
+ return Dagger.DArray(eltype(first(data)), d, s, chunks, dist)
+end
+
+for fn in [:rand, :randn, :zeros, :ones]
+ @eval begin
+ function Base.$fn(dist::ParallelBlocks, ::Type{ET}, dims::Dims) where {ET}
+ f(block) = $fn(ET, block)
+ _finish_allocation(f, dist, dims)
+ end
+ Base.$fn(dist::ParallelBlocks, T::Type, dims::Integer...) = $fn(dist, T, dims)
+ Base.$fn(dist::ParallelBlocks, T::Type, dims::Tuple) = $fn(dist, T, dims)
+ Base.$fn(dist::ParallelBlocks, dims::Integer...) = $fn(dist, Float64, dims)
+ Base.$fn(dist::ParallelBlocks, dims::Tuple) = $fn(dist, Float64, dims)
+ end
+end
+# FIXME: sprand
+
+function Dagger.distribute(data::AbstractArray{T,N}, dist::ParallelBlocks) where {T,N}
+ dims = size(data)
+ d = ArrayDomain(map(x->1:x, dims))
+ s = Dagger.DomainBlocks(ntuple(_->1, N),
+ ntuple(i->[dims[i]], N))
+ chunks = [Dagger.tochunk(copy(data)) for _ in 1:dist.n]
+ new_dist = ParallelBlocks{N}(dist)
+ return Dagger.DArray(T, d, s, wrap_chunks(chunks, N, dist), new_dist)
+end
+
+_invalid_call_pblocks(f::Symbol) =
+ error("`$f` is not valid for a `DArray` partitioned with `ParallelBlocks`.\nConsider `Dagger.pmap($f, x)` instead.")
+
+Base.collect(::Dagger.DArray{T,N,<:ParallelBlocks} where {T,N}) =
+ _invalid_call_pblocks(:collect)
+Base.getindex(::Dagger.DArray{T,N,<:ParallelBlocks} where {T,N}, x...) =
+ _invalid_call_pblocks(:getindex)
+Base.setindex!(::Dagger.DArray{T,N,<:ParallelBlocks} where {T,N}, value, x...) =
+ _invalid_call_pblocks(:setindex!)
+
+function pmap(f::Function, A::Dagger.DArray{T,N,ParallelBlocks{N}}) where {T,N}
+ # TODO: Chunks might not be `Array`s
+ # FIXME
+ #AT = Array{T,N}
+ #ET = eltype(Base.promote_op(f, AT))
+ ET = Any
+ new_chunks = map(A.chunks) do chunk
+ Dagger.@spawn f(chunk)
+ end
+ return DArray(ET, A.domain, A.subdomains, new_chunks, A.partitioning)
+end
+# FIXME: More useful `show` method
+Base.show(io::IO, ::MIME"text/plain", A::Dagger.DArray{T,N,ParallelBlocks{N}}) where {T,N} =
+ print(io, typeof(A))
+pfetch(A::Dagger.DArray{T,N,ParallelBlocks{N}}) where {T,N} =
+ map(fetch, A.chunks)
+pcollect(A::Dagger.DArray{T,N,ParallelBlocks{N}}) where {T,N} =
+ map(collect, pfetch(A))
+
+function Base.map(f::Function, A::Dagger.DArray{T,N,ParallelBlocks{N}}) where {T,N}
+ ET = Base.promote_op(f, T)
+ new_chunks = map(A.chunks) do chunk
+ Dagger.@spawn map(f, chunk)
+ end
+ return DArray(ET, A.domain, A.subdomains, new_chunks, A.partitioning)
+end
+function Base.map!(f::Function,
+ x::Dagger.DArray{T1,N1,ParallelBlocks{N1}} where {T1,N1},
+ y::Dagger.DArray{T2,N2,ParallelBlocks{N2}} where {T2,N2})
+ x_dist = x.partitioning
+ y_dist = y.partitioning
+ if x_dist.n != y_dist.n
+ throw(ArgumentError("Can't `map!` over non-matching `ParallelBlocks` distributions: $(x_dist.n) != $(y_dist.n)"))
+ end
+ @sync for i in 1:x_dist.n
+ Dagger.@spawn map!(f, x.chunks[i], y.chunks[i])
+ end
+end
+
+#=
+function Base.reduce(f::Function, x::Dagger.DArray{T,N,ParallelBlocks{N}};
+ dims=:) where {T,N}
+ error("Out-of-place Reduce")
+ if dims == Base.:(:)
+ localpart = fetch(Dagger.reduce_async(f, x))
+ return MPI.Allreduce(localpart, f, comm)
+ elseif dims === nothing
+ localpart = fetch(x.chunks[1])
+ return MPI.Allreduce(localpart, f, comm)
+ else
+ error("Not yet implemented")
+ end
+end
+=#
+function allreduce!(op::Function, x::Dagger.DArray{T,N,ParallelBlocks{N}}; nchunks::Integer=0) where {T,N}
+ if nchunks == 0
+ nchunks = x.partitioning.n
+ end
+ @assert nchunks == x.partitioning.n "Number of chunks must match the number of partitions"
+
+ # Split each chunk along the last dimension
+ chunk_size = cld(size(x, ndims(x)), nchunks)
+ chunk_dist = Blocks(ntuple(i->i == N ? chunk_size : size(x, i), N))
+ chunk_ds = partition(chunk_dist, x.subdomains[1])
+ num_par_chunks = length(x.chunks)
+
+ # Allocate temporary buffer
+ y = copy(x)
+
+ # Ring-reduce into temporary buffer
+ Dagger.spawn_datadeps() do
+ for j in 1:length(chunk_ds)
+ for i in 1:num_par_chunks
+ for step in 1:(num_par_chunks-1)
+ from_idx = i
+ to_idx = mod1(i+step, num_par_chunks)
+ from_chunk = x.chunks[from_idx]
+ to_chunk = y.chunks[to_idx]
+ sd = chunk_ds[mod1(j+i-1, length(chunk_ds))].indexes
+ # FIXME: Specify aliasing based on `sd`
+ Dagger.@spawn _reduce_view!(op,
+ InOut(to_chunk), sd,
+ In(from_chunk), sd)
+ end
+ end
+ end
+
+ # Copy from temporary buffer back to origin
+ for i in 1:num_par_chunks
+ Dagger.@spawn copyto!(Out(x.chunks[i]), In(y.chunks[i]))
+ end
+ end
+
+ return x
+end
+function _reduce_view!(op, to, to_view, from, from_view)
+ to_viewed = view(to, to_view...)
+ from_viewed = view(from, from_view...)
+ reduce!(op, to_viewed, from_viewed)
+ return
+end
+function reduce!(op, to, from)
+ to .= op.(to, from)
+end
+
+function Statistics.mean!(A::Dagger.DArray{T,N,ParallelBlocks{N}}) where {T,N}
+ allreduce!(+, A)
+ len = length(A.chunks)
+ map!(x->x ./ len, A, A)
+ return A
+end

src/array/random.jl

@@ -0,0 +1,13 @@
+using Random
+
+function Random.randn!(A::DArray{T}) where T
+ Ac = A.chunks
+
+ Dagger.spawn_datadeps() do
+ for chunk in Ac
+ Dagger.@spawn randn!(InOut(chunk))
+ end
+ end
+
+ return A
+end