Full workshop

on YouTube ↗03h 42m

" summary

Workshop opens with practical results: a dependent-type array system uses lift_nuts and dependent pairs to express CSR and other sparse layouts, generates OpenCL and matches or beats cuSPARSE (typical speedups ~1.2–1.5×, variable by density). GHC profiling on NUMA uses numaprof+malt to show spark threads allocate locally but the unpinned main thread floods one region, creating a hot column; GC shows static objects lack locality. Automatic-differentiation sessions show jvp/vjp decompose poorly over scans and that array indexing dualizes into a permute/scatter (write-conflicts sum), with an Accelerate prototype working but hitting memory/checkpointing limits.

AI-written summary · about

" chapters (39)

1. 0:00Start
2. 6:39Background
3. 9:08Community Update
4. 9:41Future of Fhpnc
5. 10:22Submissions
6. 11:43Presentations
7. 14:11Generating High Performance Codes for Irregular Data Structures Using Dependent Types
8. 15:54Type of a Regular Array
9. 17:24The Sparse Matrix Vector Multiplication
10. 19:36Position Dependent Arrays
11. 20:35Csr Matrices
12. 27:30What Is a Dependent Pair
13. 28:50Construct a Dependent Pair
14. 36:49Does the Compiler Rely on or Take Advantage of Types during Optimization or Are Types Only Used for Verification
15. 42:32Final Questions
16. 48:43Screen Sharing
17. 52:37Profiling Numa
18. 56:23Memory Management in Parallel Haskell
19. 57:17Haskell Execution Context
20. 59:12Benchmarks
21. 59:38Rsa Encryption Algorithm
22. 1:06:48Analyzing the Data
23. 1:13:06Ghc Garbage Collection
24. 1:15:59Locality of the Objects Touched by the Garbage Collection Threads
25. 1:19:03Allocation Locality
26. 1:19:42Thread Locality
27. 1:25:05Does Haskell Need Its Own Metric for Comparing Numer Loads
28. 1:35:24Low Carbon Computing
29. 1:37:24Emissions Gap
30. 1:54:40Homogeneous Nesting
31. 1:59:33Arrays of Known Shape
32. 2:00:33Width Loop
33. 2:09:08Index Vector Elimination
34. 2:13:31Loop Scalarization
35. 2:16:31Multi-Threaded Execution
36. 2:17:31Multi-Thread Back End
37. 2:21:55Memory Management
38. 2:25:05Relatively Free Variables
39. 2:26:01Bimodal Reference Counting

" related talks

• 23mA Haskell Auto-Parallelizer for Distributed ComputingYuxi Long, Shiyou Wu, Yingjie Xu · Haskell 2023
• 45mAPLicative Programming with Naperian FunctorsJeremy Gibbons · Lambda World 2017
• 36mWorking Hard to Keep Things Lazyraichoo · Lambda World 2016
• 02h 38mLazy EvaluationAndres Löh · ZuriHac 2023
• 36mConcrete Functional ProgrammingKamila Szewczyk · Lambda World 2024
• 01h 08mUnconferenceFlavio Corpa, Alejandro Serrano, Tomás Ruiz, Tapio Rautonen, Alfredo de Hoces · Lambda World Cádiz 2019
• 51mStretching the Glasgow Haskell Compiler: Nourishing GHC with Domain-Driven De...Sylvain Henry, Jeffrey M. Young, John Ericson · FUNARCH 2023
• 21mEfficient Parallel Functional Programming with EffectsUmut A. Acar, Jatin Arora, Sam Westrick · PLDI 2023