Scribd
Upload
26 views

Auto Tuning Techniques For Compiler Optimization

Its a group presented made by the students of B.P.PODDAR.INSTITUTE OF MANAGEMENT AND TECHNOLOGY. its the ppt about the COMPILER DESIGN SUBJECT.

Uploaded by

rakshitavijit27
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
26 views

Auto Tuning Techniques For Compiler Optimization

Its a group presented made by the students of B.P.PODDAR.INSTITUTE OF MANAGEMENT AND TECHNOLOGY. its the ppt about the COMPILER DESIGN SUBJECT.

Uploaded by

rakshitavijit27
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 12

PCC-CS501

Auto Tuning
Technique For
Compiler Optimization

Avijit Rakshit Paramjit Banerjee


11500122064 11500122086
Barin Ghosh Akashdeep Naha
11500122065 11500122070
Table Of Content
1) Introduction
2) Auto Tuning Approach For Compiler Optimization
a)Design Search Exploration(DSE)
b)Machine Learing Approach
4)Workflow for Auto Tuning
5)Impact Of Compiler Transformation
6)Model Assisted
7)Application
8)Conclusion
9)Reference
Introduction to Auto-tuning Technique
Auto-tuning techniques for compiler optimization are
designed to address the challenges of manually tuning
compiler optimization flags for improved runtime
performance. These techniques leverage various strategies
to efficiently search through the vast space of
optimization flag combinations, aiming to find the
optimal settings for a given program.
Design Space Exploration (DSE)
Architectural Exploration
The first step is to explore the design space of the target
hardware architecture, generating a set of promising candidate
configurations that match the characteristics of the application.
This is achieved using a multi-objective optimization approach
that balances performance and resource utilization.

Compiler Optimization
The compiled architecture properties are then fed into a
compiler framework, which systematically explores the vast
space of possible compiler transformations. Statistical analysis
techniques are employed to assess the impact of each compiler
option on performance metrics.

Insights and Decisions


The methodology provides the designer with valuable insights
into the complex interactions between the hardware architecture
and the compiler optimizations. This enables informed decisions
on the optimal configuration for a given application.
Machine Learning for Compiler Auto-tuning
Application Characterization Bayesian Network Model Predictive Optimization

The key to this approach is the A Bayesian Network is used to Given a new, unseen application,

dynamic characterization of the input model the complex probability the Bayesian Network is used to

application using microarchitecture- distribution of the optimal predict the most promising

independent features. This allows the compiler transformations. This compiler optimizations, guiding

machine learning model to capture probabilistic model is more the auto-tuning process towards

the relationship between application flexible than a rigid set of rules, high-performing solutions. This

characteristics and the most effective allowing the framework to adapt approach has demonstrated

compiler optimizations. to a wide range of applications significant speedups over


and architectures. standard optimization levels.
Auto-Tuning Workflow
Problem Definition

Clearly define the optimization objectives, constraints, and


target platform characteristics.

Search Space Exploration

Employ advanced search algorithms to efficiently navigate the


vast design space of compiler settings.

Performance Evaluation

Measure the performance of candidate solutions using


representative workloads and benchmarks.
Assessing the Impact of Compiler
Transformations
Systematic Analysis

The DSE-based methodology applies rigorous statistical techniques, such as the


Kruskal-Wallis test, to systematically assess the impact of individual compiler
transformations on application performance.

Visualizing effects

The resulting performance distributions provide valuable insights into the


effectiveness of each compiler option, revealing which transformations have a
significant positive impact and which have little to no effect.

Cross-Architecture Insights

The analysis can be performed in a cross-architecture manner, enabling the designer


to understand the generality or specificity of the compiler transformations across
different hardware platforms.
Model-Assisted Auto-Tuning

Predictive Modeling
Develop machine learning models to predict the performance of
candidate solutions.

Adaptive Exploration
Leverage the models to guide the search process and expedite the
optimization.

Continuous Refinement
Update the models with new performance data to improve
accuracy and efficiency.
Real-World Applications
High-Performance Computing Auto-tuning can unlock significant performance gains
for scientific and HPC workloads.

Embedded Systems Auto-tuning is crucial for embedded devices, where


power efficiency and performance are paramount.

Machine Learning Compilers Auto-tuning can optimize the compilation of machine


learning models for improved inference speeds.

Heterogeneous Architectures Auto-tuning is essential for efficiently utilizing the diverse


hardware resources in heterogeneous systems.
Conclusion: The Future of Auto-Tuning

Artificial Intelligence Multi-Objective Optimization Integrated Workflows


Advancements in machine learning Auto-tuning will seamlessly integrate
techniques will enable more Balancing multiple performance with the entire compiler and software
sophisticated auto-tuning solutions. metrics, such as energy, latency, and development lifecycle.
throughput, will become increasingly
important.
References
1.
https://www.researchgate.net/publication/290441528_Auto-tuning_Techniques_for_Com
piler_Optimization
2.
https://www.semanticscholar.org/paper/Auto-tuning-Techniques-for-Compiler-Optimizat
ion-Ashouri-Palermo/cee4afd97acb7c617a05df0b6120d1e47bc96854
3.https://www.geeksforgeeks.org/code-optimization-in-compiler-design/
THANK YOU

You might also like