https://github.com/kumar-shivam-ranjan/sparse-matrix-multiplication

This repository consists of sparse Matrix multiplication algorithms implemented in C/C++
https://github.com/kumar-shivam-ranjan/sparse-matrix-multiplication

c cpp14 sparse-matrix

Last synced: 2 months ago
JSON representation

This repository consists of sparse Matrix multiplication algorithms implemented in C/C++

• Host: GitHub
• URL: https://github.com/kumar-shivam-ranjan/sparse-matrix-multiplication
• Owner: kumar-shivam-ranjan
• Created: 2020-07-19T01:40:49.000Z (almost 6 years ago)
• Default Branch: master
• Last Pushed: 2020-07-19T02:02:38.000Z (almost 6 years ago)
• Last Synced: 2025-03-03T00:42:02.311Z (over 1 year ago)
• Topics: c, cpp14, sparse-matrix
• Language: C++
• Homepage:
• Size: 1.99 MB
• Stars: 3
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Sparse-Matrix-Multiplication

Code for heterogeneous computing of product of two sparse matrices


### Algorithm: Gustavson’s Row-wise SpGEMM 3


Input: Sparse matrices A and B


Output: Sparse matrix C


 

set matrix C to ∅


for all a i ∗ in matrix A in parallel do


  for all a ik in row a i ∗ do


    for all b k j in row b k ∗ do


			value ← a ik b k j


			if c i j #∈ c i ∗ then


				insert ( c i j ← v alue ) to c i ∗


			else


				c i j ← c i j + value


			end if


		end for


	end for


end for




## Algorithm 2 RowsToThreads.

Input: Sparse matrices A and B


Output: Array o f f set


{1. Set FLOP vector}

for i ← 0 to m in parallel do

	flop [ i ] ← 0

	for j ← r pts A [ i ] to r pts A [ i + 1] do

		rnz ← r pts B [ cols A [ j] + 1] − r pt B [ cols A [ j]]

		flop [ i ] ← flop [ i ] + rnz

	end for

end for

{ 2 . Assign rows to thread }

flop ps ← ParallelPrefixSum ( flop )

sum flop ← flop ps [ m ]

tnum ← omp_get_max_threads ()

a v e flop ← sum flop / tnum

o f f set[0] ← 0

for tid ← 1 to tnum in parallel do

	o f f set [ tid ] ← lowbnd ( flop ps , a v e flop ∗ tid )

end for

o f f set [ t num ] ← m



## Algorithm : Hash SpGEMM.

Input: Sparse matrices A and B
 

Output: Sparse matrix C


off set ← RowsToThreads ( A, B )

{Determine hash-table size for each thread}

tnum ← omp_get_max_threads ()

for tid ← 0 to tnum in parallel do

	size t ← 0

	for i ← o f f set [ t id] to o f f set [ t id + 1] do

		size t ← max ( size t , flop [ i ] )

	end for

	{Required maximum hash-table size is N col }

	size t ← min ( N col , size t )

	{Return minimum 2 n so that 2 n > size t }

	size t ← lowest_p2 ( size t )

end for

Symbolic ( r pts C , A, B )

Numeric ( C, A, B )