Texplorer: Efficient Tensor Program Optimization for GPUs Using a Highly Constrained Search Space

Wei Li

doi:10.1109/TPDS.2026.3675079

Return

Texplorer: Efficient Tensor Program Optimization for GPUs Using a Highly Constrained Search Space

2026-03-18

PRE

DOI：10.1109/TPDS.2026.3675079

Original

Original request for help

Save

Abstract

En 中文

Optimizing tensor programs for deep learning inference is costly, often requiring hours of search over massive search spaces ($>\! 10^{10}$). Prior machine learning–guided approaches are stochastic and inefficient because they treat both the search and programs as black boxes, overlooking two key properties: 1) the search space is highly structured, with large regions of invalid or low-performing configurations, and 2) tensor programs exhibit analyzable patterns in memory access and pipelined execution. We present Texplorer, a deterministic tensor program optimizer that exploits these structures for robust and efficient search. Texplorerintroduces a three-stage pruning algorithm to eliminate invalid, redundant, and inefficient candidates, shrinking the search space from billions to thousands. It replaces costly learned models with a lightweight analytical cost model that captures GPU pipeline and memory behavior, enabling accurate, training-free performance prediction and strong cross-operator generalization. Finally, Texplorerintegrates pruning and cost modeling into a one-pass analytical evaluation pipeline that ranks all remaining candidates, enabling rapid convergence without iterative retraining or stochastic sampling. Across 144 operators and 11 DNNs on NVIDIA V100, A100, and H100 GPUs, Texplorermatches state-of-the-art performance while reducing search time by up to $300\times$. An enhanced variant, Texplorer$^{+}$+, further improves performance by up to 3.1% with up to $72\times$ faster tuning.

Keywords:

Tensor program tuning

Compiler optimization