Systematic approach in optimizing numerical memory-bound kernels on GPU

Keyphrases

• Memory Bound 100%
• General Matrix 100%
• Matrix-vector multiplication 100%
• Double Precision 60%
• Single Precision 60%
• CuBLAS 40%
• Data Access Pattern 40%
• Hessenberg Reduction 40%
• Bidiagonalization 40%
• Dense Linear Algebra 20%
• Medium Size 20%
• Commercial Software 20%
• Computing Eigenvalues 20%
• Device Architecture 20%
• Singular Values 20%
• Matrix Size 20%
• Small Matrices 20%
• GPU Computation 20%
• Reduction Algorithm 20%
• LAPACK 20%
• Compute-bound 20%
• Irregular Data 20%
• Software Solution 20%
• Symmetric Matrices 20%
• Magma 20%
• GPU Implementation 20%
• Precision Improvement 20%
• Numerical Libraries 20%
• Hybrid Computing 20%

Computer Science

• Memory Bound 100%
• Vector Multiplication 100%
• Single Precision 60%
• Double Precision 60%
• Multiplication Kernel 40%
• Data Access Pattern 40%
• High Throughput 20%
• Reduction Algorithm 20%
• Numerical Library 20%
• Computing Eigenvalue 20%
• Software Solution 20%
• Computational Kernel 20%
• Open Source 20%
• Singular Value 20%
• Symmetric Matrix 20%