RMA improvements

docs/src/reference/onesided.md

@@ -3,9 +3,13 @@
 ```@docs
 MPI.Win_create
 MPI.Win_create_dynamic
+MPI.Win_allocate
 MPI.Win_allocate_shared
 MPI.Win_shared_query
 MPI.Win_flush
+MPI.Win_flush_local
+MPI.Win_lock_all
+MPI.Win_unlock_all
 MPI.Win_lock
 MPI.Win_unlock
 MPI.Get!

src/onesided.jl

@@ -69,6 +69,42 @@ function Win_create(base::SubArray{T}, comm::Comm; infokws...) where T
     Win_create(base, sizeof(base), sizeof(T), comm; infokws...)
 end
 
+"""
+    win, array = MPI.Win_allocate(Array{T}, dims, comm::Comm; infokws...)
+
+Create and allocate a window for objects of type `T` of dimension
+`dims` (either an integer or tuple of integers), returning a `Win` and the
+`Array{T}` attached to the local process.
+
+This is a collective call over `comm`, but `dims` can differ for each call (and
+can be zero).
+
+If the underlying implementation supports MPI 4.1, 
+use [`MPI.Win_shared_query`](@ref) to obtain the `Array` attached to a different
+process in the same shared memory space.
+
+`infokws` are info keys providing optimization hints.
+
+[`MPI.free`](@ref) should be called on the `Win` object once operations have
+been completed.
+"""
+function Win_allocate(::Type{Ptr{T}}, len::Integer, comm::Comm; kwargs...) where T
+    win = Win()
+    out_baseptr = Ref{Ptr{T}}()
+    # int MPI_Win_allocate(MPI_Aint size, int disp_unit, MPI_Info info,
+    #                      MPI_Comm comm, void *baseptr, MPI_Win *win)
+    API.MPI_Win_allocate(len*sizeof(T), sizeof(T), Info(kwargs...), comm, out_baseptr, win)
+    finalizer(free, win)
+    return win, out_baseptr[]
+end
+function Win_allocate(::Type{Array{T}}, dims, comm::Comm; kwargs...) where T
+    win, ptr = Win_allocate(Ptr{T}, prod(dims), comm; kwargs...)
+    array = unsafe_wrap(Array, ptr, dims)
+    win.object = array
+    finalizer(free, win)
+    return win, array
+end
+
 """
     win, array = MPI.Win_allocate_shared(Array{T}, dims, comm::Comm; infokws...)
 
@@ -184,6 +220,18 @@ function Win_fence(win::Win; nostore=false, noput=false, noprecede=false, nosucc
     Win_fence(assert, win)
 end
 
+"""
+    Win_flush_local(win::Win; rank)
+
+Completes locally all outstanding RMA operations initiated by the calling process to the
+target rank on the specified window.
+
+# External links
+$(_doc_external("MPI_Win_flush_local"))
+"""
+Win_flush_local(win::Win; rank) = Win_flush_local(rank, win)
+Win_flush_local(rank::Integer, win::Win) = API.MPI_Win_flush_local(rank, win)
+
 """
     Win_flush(win::Win; rank)
 
@@ -198,6 +246,41 @@ Win_flush(rank::Integer, win::Win) = API.MPI_Win_flush(rank, win)
 
 Win_sync(win::Win) = API.MPI_Win_sync(win)
 
+"""
+    Win_lock_all(win::Win; nocheck=false)
+
+Starts an RMA access epoch. The window can be
+accessed by RMA operations at all processes on `win` during that epoch.
+This operation is equivalent to a call to Win_lock(shared) for all processes
+associted with the window.
+
+If `nocheck=true`, no other process holds, or will attempt to acquire, a
+conflicting lock, while the caller holds the window lock. This is useful when
+mutual exclusion is achieved by other means, but the coherence operations that
+may be attached to the lock and unlock calls are still required.
+
+# External links
+$(_doc_external("MPI_Win_lock_all"))
+"""
+function Win_lock_all(win::Win; nocheck::Bool = false)
+    Win_lock_all(nocheck * API.MPI_MODE_NOCHECK[], win)
+end
+function Win_lock_all(assert::Integer, win::Win)
+    # int MPI_Win_lock_all(int assert, MPI_Win win)
+    API.MPI_Win_lock_all(assert, win)
+end
+
+"""
+    Win_unlock_all(win::Win)
+
+Completes an RMA access epoch started by a call to [`Win_lock_all`](@ref).
+
+# External links
+$(_doc_external("MPI_Win_unlock_all"))
+"""
+Win_unlock_all(win::Win) = Win_unlock_all(win)
+Win_unlock_all(win::Win) = API.MPI_Win_unlock_all(win)
+
 """
     Win_lock(win::Win; rank::Integer, type=:exclusive/:shared, nocheck=false)