Optimizing memory-bound SYMV kernel on GPU hardware accelerators

Computer Science

• Memory Bound 100%
• Hardware Accelerator 100%
• Memory Access 50%
• Concurrent Execution 50%
• Preprocessing Step 50%
• Multicore 50%
• Asymptotic Result 50%
• Single Precision 50%
• Underlying Hardware 50%
• High-Level Programming Language 50%
• Execution Context 50%
• Precision Arithmetic 50%
• Profiling Tool 50%
• Double Precision 50%
• Symmetric Matrix 50%
• Matrix Vector Product 50%

Keyphrases

• CUD A 100%
• GPU Hardware 100%
• Hardware Accelerator 100%
• Execution Environment 33%
• Extended Bandwidth 33%
• Profiling Tools 33%
• Matrix Algebra 33%
• High Performance Scientific Computing 33%
• Tridiagonalization 33%
• Eigenpair 33%
• High-level Programming Languages 33%
• Language Extension 33%
• Dense Matrix 33%
• Irregular Memory Access 33%
• NVIDIA 33%
• Matrix-vector Product 33%
• Increased Productivity 33%
• Unprecedented Level 33%