An effective fusion and tile size model for optimizing image processing pipelines

Jangda, Abhinav; Bondhugula, Uday

doi:10.1145/3178487.3178507

Cited by 22 publications

(41 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The polyhedral compiler PolyMage [Mullapudi et al 2015] uses an algorithm similar to that of the current Halide autoscheduler. This has also been extended with backtracking and a richer cost model to some benefit [Jangda and Bondhugula 2018], but without changing the fundamentally restricted search space. Sioutas et al [2018] improved local scheduling of individual Halide operations by defining a richer manual cost model to capture cache and prefetcher behaviour on CPUs.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Learning to optimize halide with tree search and random programs

Adams

Anderson³

et al. 2019

ACM Trans. Graph.

176

View full text Add to dashboard Cite

We present a new algorithm to automatically schedule Halide programs for high-performance image processing and deep learning. We significantly improve upon the performance of previous methods, which considered a limited subset of schedules. We define a parameterization of possible schedules much larger than prior methods and use a variant of beam search to search over it. The search optimizes runtime predicted by a cost model based on a combination of new derived features and machine learning. We train the cost model by generating and featurizing hundreds of thousands of random programs and schedules. We show that this approach operates effectively with or without autotuning. It produces schedules which are on average almost twice as fast as the existing Halide autoscheduler without autotuning, or more than twice as fast with, and is the first automatic scheduling algorithm to significantly outperform human experts on average. CCS Concepts: • Computing methodologies → Image processing; • Software and its engineering → Domain specific languages.

show abstract

Section: Related Workmentioning

confidence: 99%

“…PolyMage. The closest related language with a fully-automatic scheduler is PolyMage [Jangda and Bondhugula 2018]. The Poly-Mage repository reproduces a subset of this benchmark suite, though the code has decayed over time.…”

Section: Additional Comparisons and Ablationsmentioning

confidence: 99%

Learning to optimize halide with tree search and random programs

Adams

Anderson³

et al. 2019

ACM Trans. Graph.

176

View full text Add to dashboard Cite

show abstract

“…The Halide Auto-Scheduler [15] attempts to group stages together and evaluates an effective tiling in each group. PolyMage can use both auto-tuning to search parts of the design space as well as a recently introduced model-driven approach [10]. This new approach quickly attempts to fuse stages and extends the search space to cover more solutions than the previous auto-tuning method.…”

Section: Introductionmentioning

confidence: 99%

“…This new approach quickly attempts to fuse stages and extends the search space to cover more solutions than the previous auto-tuning method. However, all three techniques [10,15,16] focus on the interleaving of the computation of each stage using overlapping tiles and therefore lead to solutions with limited reuse possibilities and often miss sliding window opportunities.…”

Section: Introductionmentioning

confidence: 99%

“…We implement it as a tool to be used with the Halide DSL as an alternative cost model and analysis to the Halide Auto-Scheduler and evaluate it across a variety of benchmarks and target platforms. We compare our solutions to the ones produced by the Halide Auto-Scheduler, the manual solutions given for the Halide DSL on the same benchmarks when applicable, as well as the ones produced by PolyMage (using both the original auto-tuned method, as well as the DP-fusion technique implemented in Reference [10]) on the same target architectures. We observe a substantial performance improvement across all platforms and architectures.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Schedule Synthesis for Halide Pipelines through Reuse Analysis

Sioutas

Stuijk

Waeijen

et al. 2019

ACM Trans. Archit. Code Optim.

View full text Add to dashboard Cite

Efficient code generation for image processing applications continues to pose a challenge in a domain where high performance is often necessary to meet real-time constraints. The inherently complex structure found in most image-processing pipelines, the plethora of transformations that can be applied to optimize the performance of an implementation, as well as the interaction of these optimizations with locality, redundant computation and parallelism, can be indentified as the key reasons behind this issue. Recent domain-specific languages (DSL) such as the Halide DSL and compiler attempt to encourage high-level design-space exploration to facilitate the optimization process. We propose a novel optimization strategy that aims to maximize producer-consumer locality by exploiting reuse in image-processing pipelines. We implement our analysis as a tool that can be used alongside the Halide DSL to automatically generate schedules for pipelines implemented in Halide and test it on a variety of benchmarks. Experimental results on three different multi-core architectures show an average performance improvement of 40% over the Halide Auto-Scheduler and 75% over a state-of-the art approach that targets the PolyMage DSL.

show abstract