Matrix multiplication in Scala

First take with idiomatic scala(77ms):

	def singleThreadedMultiplication2(m1:Seq[Array[Double]], m2:Array[Array[Double]] ) ={
	val res = Array.fill(m1.length, m2(0).length)(0.0)
	for(row <- 0 until m1.length;
	col <- 0 until m2(0).length;
	i <- 0 until m1(0).length){
	res(row)(col) += m1(row)(i) * m2(i)(col)
	}
	res
	}

view raw gistfile1.scala hosted with ❤ by GitHub

Second take - replacing fors with whiles(63ms):

	@inline def singleThreadedMultiplication1(m1:Seq[Array[Double]], m2:Array[Array[Double]] ) ={
	val res = Array.ofDim[Double](m1.length, m2(0).length)
	var col, i = 0
	var row = 0
	// while statements are much faster than for statements
	while(row < m1.length){ col = 0
	while(col < m2(0).length){ i = 0
	while(i < m1(0).length){
	res(row)(col) += m1(row)(i) * m2(i)(col)
	i+=1
	}
	col += 1
	}
	row += 1
	}
	res
	}

view raw gistfile1.scala hosted with ❤ by GitHub

Third take - sum in a variable(55ms):

	@inline def singleThreadedMultiplicationFAST(m1:Seq[Array[Double]], m2:Array[Array[Double]] ) ={
	val res = Array.ofDim[Double](m1.length, m2(0).length)
	val M1_COLS = m1(0).length
	val M1_ROWS = m1.length
	val M2_COLS = m2(0).length
	var col, i = 0
	var sum = 0.0
	var row = 0
	while(row < M1_ROWS){ col = 0
	while(col < M2_COLS){ i = 0; sum = 0
	while(i<M1_COLS){
	// calculating the sum in a variable seems to be faster than updating the array all the time
	sum += m1(row)(i) * m2(i)(col)
	i+=1
	}
	res(row)(col) = sum
	col += 1
	}; row += 1
	}
	res
	}

view raw gistfile1.scala hosted with ❤ by GitHub

Multi-threaded with parallel collections(15ms):

	override def multiply(m1: Array[Array[Double]], m2: Array[Array[Double]]) : Array[Array[Double]] = {
	val res = Array.ofDim[Double](m1.length, m2(0).length)
	val M1_COLS = m1(0).length
	val M1_ROWS = m1.length
	val M2_COLS = m2(0).length
	@inline def singleThreadedMultiplicationFAST(start_row:Int, end_row:Int) {
	var col, i = 0
	var sum = 0.0
	var row = start_row
	// while statements are much faster than for statements
	while(row < end_row){ col = 0
	while(col < M2_COLS){ i = 0; sum = 0
	while(i<M1_COLS){
	sum += m1(row)(i) * m2(i)(col)
	i+=1
	}
	res(row)(col) = sum
	col += 1
	}; row += 1
	}
	}
	(0 until M1_ROWS).par.foreach( i =>
	singleThreadedMultiplicationFAST(i, i+1)
	)
	res
	}

view raw gistfile1.scala hosted with ❤ by GitHub

Multi-threaded with parallel collections idiomatic scala(32ms):

	def multiThreadedIdiomatic(m1:Seq[Array[Double]], m2:Array[Array[Double]] ) ={
	val res = Array.fill(m1.length, m2(0).length)(0.0)
	for(row <- (0 until m1.length).par;
	col <- (0 until m2(0).length).par;
	i <- 0 until m1(0).length){
	res(row)(col) += m1(row)(i) * m2(i)(col)
	}
	res
	}

view raw gistfile1.scala hosted with ❤ by GitHub