Fixed performance issues in consistent/assemble, new and improved distributed sparse matrix multiplication algorithms.

src/PartitionedArrays.jl

@@ -13,6 +13,7 @@ using BlockArrays
 
 export length_to_ptrs!
 export rewind_ptrs!
+export jagged_range
 export jagged_array
 export GenericJaggedArray
 export JaggedArray
@@ -24,6 +25,12 @@ export compresscoo
 export indextype
 export sparse_matrix
 export sparse_matrix!
+export index_array
+export pointer_array
+export halfperm
+export halfperm!
+export symbolic_halfperm
+export symbolic_halfperm!
 include("sparse_utils.jl")
 
 export linear_indices
@@ -169,9 +176,17 @@ export spmv!
 export spmtv!
 export spmm
 export spmm!
+export spmmm
+export spmmm!
 export spmtm
 export spmtm!
+export spmtmm
+export spmtmm!
 export centralize
+export explicit_transpose
+export explicit_transpose!
+export add
+export add!
 include("p_sparse_matrix.jl")
 
 export BRange
@@ -193,6 +208,16 @@ export node_coordinates_unit_cube
 export nullspace_linear_elasticity
 export nullspace_linear_elasticity!
 export near_nullspace_linear_elasticity
+export prolongator
 include("gallery.jl")
 
+export add
+export subtract
+export mul
+export matmul
+export matmul!
+export rap
+export rap!
+include("sequential_implementations.jl")
+
 end # module

src/gallery.jl

@@ -587,6 +587,3 @@ function nullspace_linear_elasticity!(B,x)
     end
     B
 end
-
-
-

src/jagged_array.jl

@@ -154,6 +154,15 @@ function JaggedArray{T,Ti}(a::AbstractArray{<:AbstractArray}) where {T,Ti}
   JaggedArray(data,ptrs)
 end
 
+# New
+function jagged_range(a::Union{JaggedArray,GenericJaggedArray},i::Integer)
+  u = one(eltype(a.ptrs))
+  pini = a.ptrs[i]
+  pend = a.ptrs[i+1]-u
+  pini:pend
+end
+
+###########
 
 Base.size(a::Union{JaggedArray,GenericJaggedArray}) = (length(a.ptrs)-1,)
 function Base.getindex(a::Union{JaggedArray,GenericJaggedArray},i::Int)

src/p_range.jl

@@ -408,7 +408,7 @@ end
 """
     neigs_snd, neigs_rcv = assembly_neighbors(index_partition;kwargs...)
 
-Return the ids of the neighbor parts from we send and receive data respectively
+Return the ids of the neighbor parts from which we send and receive data respectively
 in the assembly of distributed vectors defined on the index
 partition `index_partition`.
 partition `index_partition`. `kwargs` are delegated to [`ExchangeGraph`](@ref)
@@ -470,7 +470,7 @@ end
 
 function assembly_local_indices(indices,neighbors_snd,neighbors_rcv)
     cache = map(assembly_cache,indices)
-    mask =  map(cache) do mycache
+    mask = map(cache) do mycache
         isassigned(mycache.local_indices_snd) && isassigned(mycache.local_indices_rcv)
     end
     if ! getany(mask)