Improved the dot product between two vectors and a sparse matrix (#410)

Co-authored-by: Daniel Karrasch <[email protected]>

src/linalg.jl

@@ -343,28 +343,26 @@ function dot(A::AbstractSparseMatrixCSC{T1,S1},B::AbstractSparseMatrixCSC{T2,S2}
     return r
 end
 
-function dot(x::AbstractVector, A::AbstractSparseMatrixCSC, y::AbstractVector)
+function dot(x::AbstractVector{T1}, A::AbstractSparseMatrixCSC{T2}, y::AbstractVector{T3}) where {T1,T2,T3}
     require_one_based_indexing(x, y)
     m, n = size(A)
     (length(x) == m && n == length(y)) || throw(DimensionMismatch())
-    if iszero(m) || iszero(n)
-        return dot(zero(eltype(x)), zero(eltype(A)), zero(eltype(y)))
-    end
-    T = promote_type(eltype(x), eltype(A), eltype(y))
-    r = zero(T)
-    rvals = getrowval(A)
+    s = dot(zero(T1), zero(T2), zero(T3))
+    T = typeof(s)
+    (iszero(m) || iszero(n)) && return s
+
+    rowvals = getrowval(A)
     nzvals = getnzval(A)
-    @inbounds for col in 1:n
+
+    @inbounds @simd for col in 1:n
         ycol = y[col]
-        if _isnotzero(ycol)
-            temp = zero(T)
-            for k in nzrange(A, col)
-                temp += adjoint(x[rvals[k]]) * nzvals[k]
-            end
-            r += temp * ycol
+        for j in nzrange(A, col)
+            row = rowvals[j]
+            val = nzvals[j]
+            s += dot(x[row], val, ycol)
         end
     end
-    return r
+    return s
 end
 function dot(x::SparseVector, A::AbstractSparseMatrixCSC, y::SparseVector)
     m, n = size(A)