|
# GPU Profiling Methodology
|
|
# GPU Profiling Methodology
|
|
|
|
When creating a profile report, there is a methodology which can be applied to make a thorough overview of the application's performance. You can find an example report attached to this wiki page. The sections below will cover the parts one-by-one, with information on how to acquire the information needed.
|
|
|
|
|
|
|
|
## General introduction
|
|
|
|
The introduction of a report is quite fixed. It consists of the following three parts:
|
|
|
|
|
|
|
|
1. Information on typical ways to do GPU profiling and how to address issues.
|
|
|
|
2. The 7 major performance hazards for GPU code. Each of them is explained and substantiated with scenarios that could cause the performance degradation.
|
|
|
|
3. How we diagnose the performance hazards in their code.
|
|
|
|
|
|
|
|
## Actual report
|
|
|
|
After the general introduction, we start writing the actual report. The first section will cover the code, system software, and system hardware. The second section will give a in-depth analysis of the performance of the program. The third and last section will discuss the 7 performance hazards and how the program scores on them.
|
|
|
|
|
|
|
|
### Code, system software, and system hardware
|
|
|
|
|
|
|
|
### Analysis of the GPU code
|
|
|
|
|
|
|
|
### Performance analysis discussion
|
|
|
|
|
|
# Subpages
|
|
# Subpages
|
|
- [Intel Advisor & Intel VTune](3.a.-Intel-Offload-Advisor-&-Intel-VTune)
|
|
- [Intel Advisor & Intel VTune](3.a.-Intel-Offload-Advisor-&-Intel-VTune)
|
... | | ... | |