Kernel Specialization for Improved Adaptability and Performance on Graphics Processing Units (GPUs)

Moore, Nicholas; Leeser, Miriam; King, Laurie Smith

doi:10.1109/ipdps.2013.31

Cited by 3 publications

(6 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Instead of locking a kernel implementation into one of two regimes: higher performance without adaptability or adaptability with lower performance, we use kernel specialization (KS), a method that yields both high-performance and adaptability [2], [3]. We combine adaptable kernel implementation techniques with a framework, GPU Prototyping Framework (GPU-PF), that dynamically compiles specialized kernels at application runtime.…”

Section: T He Significant Performance Gains Available With Generalmentioning

confidence: 99%

“…Combines kernel specialization with warp specialization. In [2], we demonstrated the advantages of KS applied to sliding window applications and medical image reconstruction. In this work, we show the advantages when applying KS to Particle Image Velocimetry (PIV).…”

Section: T He Significant Performance Gains Available With Generalmentioning

confidence: 99%

“…In this section, we first look at coding techniques for KS, and then the GPU-PF. Additional details can be found in [2], [3].…”

Section: Kernel Specializationmentioning

confidence: 99%

“…Our previous work describes the use of kernel specialization for lung tumor tracking and for medical X-ray cone beam computed tomography [2], [3]. Here we apply KS to improve the performance of particle image velocimetry.…”

Section: Particle Image Velocimetry: An Applicationmentioning

confidence: 99%

“…The specialized kernel is derived much like the large template matching kernel, described in [2]. A RE kernel is developed first, and then a kernel specialized version is created by replacing kernel arguments and CUDA built-in references with new macro names provided at compile time.…”

Section: Kernel Specializationmentioning

confidence: 99%

See 4 more Smart Citations

Kernel Specialization Provides Adaptable GPU Code for Particle Image Velocimetry

Moore

Leeser

King

2015

IEEE Trans. Parallel Distrib. Syst.

Self Cite

View full text Add to dashboard Cite

Graphics Processing Units (GPUs) are increasingly used to accelerate scientific applications. The state-of-the-art limits the adaptability of GPU kernels to both problem parameters and hardware characteristics. This makes writing high performance libraries for GPUs challenging. We address these challenges through Kernel Specialization (KS) which supports both user and hardware parameters and produces highly optimized GPU code. We apply KS to Particle Image Velocimetry (PIV), a technique used to obtain instantaneous velocity measurements in fluids for such diverse applications as aircraft design and artificial heart design. KS helps the user search PIV's highly non-linear design space, supports a wide range of PIV parameters, and results in improved acceleration times over existing kernels.

show abstract

Section: T He Significant Performance Gains Available With Generalmentioning

confidence: 99%

Section: T He Significant Performance Gains Available With Generalmentioning

confidence: 99%

“…In this section, we first look at coding techniques for KS, and then the GPU-PF. Additional details can be found in [2], [3].…”

Section: Kernel Specializationmentioning

confidence: 99%

Section: Particle Image Velocimetry: An Applicationmentioning

confidence: 99%

Section: Kernel Specializationmentioning

confidence: 99%