Abstract：The geometric transformations，projection transformations，and viewport transformations involved in graphics rendering require a large number of Sparse Matrix-Vector Multiplication（SpMV） operations.How to implement SpMV high-performance computing has become one of the key issues in graphics processor design.However，the current SpMV operation has problems such as poor parallelism and large resource consumption.In order to improve the performance of hardware operations，this paper designs a special accelerator structure based on the linear combination of matrix column vectors and making full use of data parallelism.Experiments show that compared with the other two structures，the speed of the XC6VLX550T development board can be increased by 28% and 37%，respectively，and the resource occupation rate can be reduced by approximately 48% and 18%.After applying the transform operation in graphics rendering，the speed can be increased by 28% and 30%，respectively，and the resource occupation rate can be reduced by about 48% and 60%，respectively.
2021, Vol. 42 
