Ask Me Anything with Michelle Strout
summary
Explains why runtime inspectors live in libraries for sparse matrices rather than LLVM/GCC: storage formats vary (index arrays, pointer-based sparse tensors) so general-purpose compilers lack the high-level tensor/array abstractions needed for analysis. Describes two deliverables available now — the Inspector-Executor Generator on GitHub and the Sparse Polyhedral Framework — and argues that DSLs and projects like TVM, plus polyhedral techniques already in LLVM/Clang, drive practical wins for scheduling, tiling, and auto-tuning.
AI-written summary · about
chapters (6)
- 0:00Start
- 5:05What Are some of the Key Challenges You Faced When Developing the Sparse Polyhedral Framework and What Steps Have You Taken To Broaden the Impact of this Remark
- 15:28What's a Convolutional Network
- 19:24Technical Reason for Wanting Them To Be Explainable
- 43:12Our Phd Students Are Often Studious and Somewhat Introverted and Not Always Confident Do You Have any Experience To Share on How To Build Their Confidence in Ambition To Develop into Research Leaders like Yourself
- 43:41The Confidence Issue
related talks
28mFunctional Sparse Tensor Compilation
21mTowards Compiler-Guided Static Analysis
24mMapping Sparse Applications to Accelerated Computing Systems
38mFireside chat
20mA Mechanized Algebra of Verified Data Structures for Optimizing Sparse Tensor Programs
11mExtending the Scope of the Array Abstraction beyond Sparsity
19mCompiler Support for Sparse Tensor Convolutions
01h 04mJust-in-Time Compilation: The Next Big Thing?