SPLASH 2021
Sun 17 - Fri 22 October 2021 Chicago, Illinois, United States
Thu 21 Oct 2021 14:05 - 14:20 at Zurich C - Implementation of special Paradigms Chair(s): Frank Tip
Thu 21 Oct 2021 22:05 - 22:20 at Zurich C - Implementation of special Paradigms - mirror Chair(s): Steve Blackburn

We present a new algorithm to quickly generate high-performance GPU implementations of complex imaging and vision pipelines, directly from high-level Halide algorithm code. It is fully automatic, requiring no schedule templates or hand-optimized kernels. We address the scalability challenge of extending search-based automatic scheduling to map large real-world programs to the deep hierarchies of memory and parallelism on GPU architectures in reasonable compile time. We achieve this using (1) a two-phase search algorithm that first ‘freezes’ decisions for the lowest cost sections of a program, allowing relatively more time to be spent on the important stages, (2) a hierarchical sampling strategy that groups schedules based on their structural similarity, then samples representatives to be evaluated, allowing us to explore a large space with few samples, and (3) memoization of repeated partial schedules, amortizing their cost over all their occurrences. We guide the process with an efficient cost model combining machine learning, program analysis, and GPU architecture knowledge.
We evaluate our method’s performance on a diverse suite of real-world imaging and vision pipelines. Our scalability optimizations lead to average compile time speedups of 49x (up to 530x). We find schedules that are on average 1.7x faster than existing automatic solutions (up to 5x), and competitive with what the best human experts were able to achieve in an active effort to beat our automatic results.

Thu 21 Oct

Displayed time zone: Central Time (US & Canada) change

13:50 - 15:10
Implementation of special ParadigmsOOPSLA at Zurich C +8h
Chair(s): Frank Tip Northeastern University
13:50
15m
Talk
Coarsening Optimization for Differentiable ProgrammingVirtual
OOPSLA
Xipeng Shen North Carolina State University; Facebook, Guoqiang Zhang North Carolina State University; Facebook, Irene Dea Facebook, Samantha Andow Facebook, Emilio Arroyo-Fang Facebook, Neal Gafter Facebook, Johann George Facebook, Melissa Grueter Facebook, Erik Meijer Facebook, Olin Grigsby Shivers Facebook, Steffi Stumpos Facebook, Alanna Tempest Facebook, Christy Warden Facebook, Shannon Yang Facebook
DOI
14:05
15m
Talk
Efficient Automatic Scheduling of Imaging and Vision Pipelines for the GPUVirtual
OOPSLA
Luke Anderson Massachusetts Institute of Technology, Andrew Adams Adobe, Karima Ma Massachusetts Institute of Technology, Tzu-Mao Li Massachusetts Institute of Technology; University of California at San Diego, Tian Jin Massachusetts Institute of Technology, Jonathan Ragan-Kelley Massachusetts Institute of Technology
DOI
14:20
15m
Talk
Statically Bounded-Memory Delayed Sampling for Probabilistic StreamsIn-Person
OOPSLA
Eric Atkinson Massachusetts Institute of Technology, Guillaume Baudart IBM Research, USA, Louis Mandel IBM Research, Charles Yuan Massachusetts Institute of Technology, Michael Carbin Massachusetts Institute of Technology
DOI
14:35
15m
Talk
Compilation of Sparse Array Programming ModelsIn-Person
OOPSLA
Rawn Henry Massachusetts Institute of Technology, Olivia Hsu Stanford University, Rohan Yadav Stanford University, Stephen Chou Massachusetts Institute of Technology, Kunle Olukotun Stanford University, Saman Amarasinghe Massachusetts Institute of Technology, Fredrik Kjolstad Stanford University
DOI
14:50
20m
Live Q&A
Discussion, Questions and Answers
OOPSLA

21:50 - 23:10
Implementation of special Paradigms - mirrorOOPSLA at Zurich C
Chair(s): Steve Blackburn Australian National University
21:50
15m
Talk
Coarsening Optimization for Differentiable ProgrammingVirtual
OOPSLA
Xipeng Shen North Carolina State University; Facebook, Guoqiang Zhang North Carolina State University; Facebook, Irene Dea Facebook, Samantha Andow Facebook, Emilio Arroyo-Fang Facebook, Neal Gafter Facebook, Johann George Facebook, Melissa Grueter Facebook, Erik Meijer Facebook, Olin Grigsby Shivers Facebook, Steffi Stumpos Facebook, Alanna Tempest Facebook, Christy Warden Facebook, Shannon Yang Facebook
DOI
22:05
15m
Talk
Efficient Automatic Scheduling of Imaging and Vision Pipelines for the GPUVirtual
OOPSLA
Luke Anderson Massachusetts Institute of Technology, Andrew Adams Adobe, Karima Ma Massachusetts Institute of Technology, Tzu-Mao Li Massachusetts Institute of Technology; University of California at San Diego, Tian Jin Massachusetts Institute of Technology, Jonathan Ragan-Kelley Massachusetts Institute of Technology
DOI
22:20
15m
Talk
Statically Bounded-Memory Delayed Sampling for Probabilistic StreamsIn-Person
OOPSLA
Eric Atkinson Massachusetts Institute of Technology, Guillaume Baudart IBM Research, USA, Louis Mandel IBM Research, Charles Yuan Massachusetts Institute of Technology, Michael Carbin Massachusetts Institute of Technology
DOI
22:35
15m
Talk
Compilation of Sparse Array Programming ModelsIn-Person
OOPSLA
Rawn Henry Massachusetts Institute of Technology, Olivia Hsu Stanford University, Rohan Yadav Stanford University, Stephen Chou Massachusetts Institute of Technology, Kunle Olukotun Stanford University, Saman Amarasinghe Massachusetts Institute of Technology, Fredrik Kjolstad Stanford University
DOI
22:50
20m
Live Q&A
Discussion, Questions and Answers
OOPSLA