Efficient CSR-Based Sparse Matrix-Vector Multiplication on GPU

Table 1

Sparse matrices used in the experiments.

Matrix	Rows	Columns	Nonzeros (NZ)	NZ/Row
Dense	2,000	2,000	4,000,000	2000.0
Protein	36,417	36,417	4,344,765	119.3
FEM/Spheres	83,334	83,334	6,010,480	72.1
FEM/Cantilever	62,451	62,451	4,007,383	64.1
Wind Tunnel	217,918	217,918	11,634,424	53.3
FEM/Harbor	46,835	46,835	2,374,001	50.6
QCD	49,152	49,152	1,916,928	39.0
FEM/Ship	140,874	140,874	7,813,404	55.4
Economics	206,500	206,500	1,273,389	6.1
Epidemiology	525,825	525,825	2,100,225	3.9
FEM/Accelerator	121,192	121,192	2,624,331	21.6
Circuit	170,998	170,998	958,936	5.6
Webbase	1,000,005	1,000,005	3,105,536	3.1
LP	4,284	1,092,610	11,279,748	2632.9
circuit5M	5,558,326	5,558,326	59,524,291	10.7
eu-2005	862,664	862,664	19,235,140	22.3
Ga41As41H72	268,096	268,096	18,488,476	69.0
in-2004	1,382,908	1,382,908	16,917,053	12.2
mip1	66,463	66,463	10,352,819	155.8
Si41Ge41H72	185,639	185,639	15,011,265	80.9