|
Title:
|
UNIFIED OPENCL PROFILING FOR GPU-BASED HETEROGENEOUS EMBEDDED APPLICATIONS |
|
Author(s):
|
Gul N. Khan and Mohid Tayyub |
|
ISBN:
|
978-989-8704-62 |
|
Editors:
|
Paula Miranda and Pedro Isaías |
|
Year:
|
2024 |
|
Edition:
|
Single |
|
Keywords:
|
Cross-Platform Profiling, Opencl, Performance Tuning for Heterogeneous Systems, Unified Programming Model |
|
Type:
|
Full |
|
First Page:
|
167 |
|
Last Page:
|
174 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Due to the growth of accessible computing, GPU based heterogeneous computing requires portable standards to
streamline the design of heterogeneous embedded applications. One such standard is OpenCL that enables programs to
work across multiple devices and platforms. Due to abstract model, it is possible for applications to misuse the hardware
resulting in degraded performance. In-depth profiling is necessary to identify the performance bottlenecks. Current tools
do not provide a dynamic interface for comprehensive metric collection. Moreover, they are device/vendor-specific
negating the portability of OpenCL. This paper presents unified flexible profiling for OpenCL applications that maintains
portability and unlocks new metric collection capabilities by utilizing the new features of OpenCL 2 |
|
|
|
|
|
|
