Pattern-Based Optimization (...)

on YouTube ↗23m

" summary

Describes extending high-level functional primitives to express dense linear algebra on multidimensional arrays and presents rewrite rules that transform those expressions to improve data locality for better performance on modern hardware, including GPUs. Focuses on using automatic tools to recognize and optimize parts of expression trees for dense arithmetic common in scientific simulations and machine learning.

AI-written summary · about

" chapters (11)

1. 0:00Start
2. 1:21Hardware hierarchies
3. 2:07Memory hierarchies
4. 2:34Specific example: linear algebra
5. 4:45Naperian Functors
6. 6:25Multidimensional tensors
7. 8:07Higher order function primitives
8. 12:286 Rearrangements of the matrix-vector multiplication at 1 level of subdivision
9. 12:56Rearrangements of the matrix-matrix multiplication
10. 14:02What have we gained?
11. 16:46What next?

" related talks

• 01h 37mAn Overview of Program Optimization TechniquesMathias Gaunard · ACCU 2017
• 02h 50mFunctional Array Funhouse Intensive - Part 3 of 5Aaron W Hsu · LambdaConf 2017
• 28mCommunication Avoiding Algorithms for Linear Algebra and BeyondJim Demmel · Parallel and Distributed Algorithms for Inference and Optimization (Simons Institute, Oct. 21–24, 2013)
• 44mFunctional Array Funhouse Intensive - Part 5 of 5Aaron W Hsu · LambdaConf 2017
• 01h 04mDesign for PerformanceFedor Pikus · CppCon 2018
• 58mFunctional Array Funhouse Intensive - Part 2 of 5Aaron W Hsu · LambdaConf 2017
• 01h 06mAutomatic Task-based Code Generation for High Performance DSELJoel Falcou · Meeting C++ 2014
• 43mDotty Linker: Precise Types Bring PerformanceDmitry Petrashko · Scala Days Berlin 2016