Prism is the first symbolic superoptimizer for tensor programs that uses sGraph for compact representation of program families, two-level search, e-graph equivalence checking, and auto-tuning to achieve up to 2.2x speedup over prior superoptimizers on LLM workloads.
TVM: An Automated End-to-End Optimizing Compiler for Deep Learning
10 Pith papers cite this work, alongside 141 external citations. Polarity classification is still indexing.
10
Pith papers citing it
141
external citations · Pith
abstract
There is an increasing need to bring machine learning to a wide diversity of hardware devices. Current frameworks rely on vendor-specific operator libraries and optimize for a narrow range of server-class GPUs. Deploying workloads to new platforms -- such as mobile phones, embedded devices, and accelerators (e.g., FPGAs, ASICs) -- requires significant manual effort. We propose TVM, a compiler that exposes graph-level and operator-level optimizations to provide performance portability to deep learning workloads across diverse hardware back-ends. TVM solves optimization challenges specific to deep learning, such as high-level operator fusion, mapping to arbitrary hardware primitives, and memory latency hiding. It also automates optimization of low-level programs to hardware characteristics by employing a novel, learning-based cost modeling method for rapid exploration of code optimizations. Experimental results show that TVM delivers performance across hardware back-ends that are competitive with state-of-the-art, hand-tuned libraries for low-power CPU, mobile GPU, and server-class GPUs. We also demonstrate TVM's ability to target new accelerator back-ends, such as the FPGA-based generic deep learning accelerator. The system is open sourced and in production use inside several major companies.
citation-role summary
background 3
citation-polarity summary
roles
background 3polarities
background 3representative citing papers
Nautilus auto-compiles math-like tensor descriptions into optimized GPU kernels, delivering up to 42% higher throughput than prior compilers on transformer models across NVIDIA GPUs.
Neptune introduces dependency-breaking fusion with algebraic corrections for reduction sequences, generating FlashAttention-like kernels from plain attention code with 1.35x average speedup across ten benchmarks and four GPU architectures.
A co-design framework using approximate matrix decomposition and genetic algorithms delivers 33% average latency reduction in TinyML CNN FPGA accelerators with 1.3% average accuracy loss versus standard systolic arrays.
R^3 optimizes full scientific applications on GPUs better than tuning kernel parameters or compiler flags alone while running nearly an order of magnitude faster than modern evolutionary search methods.
An empirical study decomposes the LLVM -O3 pipeline into cumulative prefixes and quantifies per-pass effects on 30 kernels, finding non-monotonic behavior, back-loaded gains, and a 46.35% idealized upper bound on phase-interference losses.
Piper decouples user-defined distributed training strategies from runtime execution using transformations on a unified global training DAG IR, achieving parity on ZeRO and gains on composed strategies like DualPipe.
IO-aware GPU kernels for SpMM convolutions, degree-aware reductions, and fused attention layers deliver median speedups of 1.6-2.6x (up to 10x) and memory reductions up to 76x over DGL/PyG baselines on realistic graphs.
HTAM builds a Hierarchical Transition Graph to organize coarse global directions and detailed local strategies for guiding LLM-based CUDA kernel optimization, improving results on KernelBench.
MLIR concepts such as match-and-rewrite and staged lowering correspond to established formal theories, providing a basis for more principled abstraction design in compiler infrastructure.
citing papers explorer
-
Prism: Symbolic Superoptimization of Tensor Programs
Prism is the first symbolic superoptimizer for tensor programs that uses sGraph for compact representation of program families, two-level search, e-graph equivalence checking, and auto-tuning to achieve up to 2.2x speedup over prior superoptimizers on LLM workloads.
-
Nautilus: An Auto-Scheduling Tensor Compiler for Efficient Tiled GPU Kernels
Nautilus auto-compiles math-like tensor descriptions into optimized GPU kernels, delivering up to 42% higher throughput than prior compilers on transformer models across NVIDIA GPUs.
-
Neptune: Advanced ML Operator Fusion for Locality and Parallelism on GPUs
Neptune introduces dependency-breaking fusion with algebraic corrections for reduction sequences, generating FlashAttention-like kernels from plain attention code with 1.35x average speedup across ten benchmarks and four GPU architectures.
-
Co-Design of CNN Accelerators for TinyML using Approximate Matrix Decomposition
A co-design framework using approximate matrix decomposition and genetic algorithms delivers 33% average latency reduction in TinyML CNN FPGA accelerators with 1.3% average accuracy loss versus standard systolic arrays.
-
Record-Remix-Replay: Hierarchical GPU Kernel Optimization using Evolutionary Search
R^3 optimizes full scientific applications on GPUs better than tuning kernel parameters or compiler flags alone while running nearly an order of magnitude faster than modern evolutionary search methods.
-
A Multi-Dimensional, Per-Pass Empirical Study of the LLVM Optimization Pipeline
An empirical study decomposes the LLVM -O3 pipeline into cumulative prefixes and quantifies per-pass effects on 30 kernels, finding non-monotonic behavior, back-loaded gains, and a 46.35% idealized upper bound on phase-interference losses.
-
Piper: A Programmable Distributed Training System
Piper decouples user-defined distributed training strategies from runtime execution using transformations on a unified global training DAG IR, achieving parity on ZeRO and gains on composed strategies like DualPipe.
-
On Efficient Scaling of GNNs via IO-Aware Layers Implementations
IO-aware GPU kernels for SpMM convolutions, degree-aware reductions, and fused attention layers deliver median speedups of 1.6-2.6x (up to 10x) and memory reductions up to 76x over DGL/PyG baselines on realistic graphs.
-
HTAM: Hierarchical Transition-Attended Memory for Operator Optimization
HTAM builds a Hierarchical Transition Graph to organize coarse global directions and detailed local strategies for guiding LLM-based CUDA kernel optimization, improving results on KernelBench.
-
Reading AI Model Compilation in MLIR Through the Lens of Formal Theories
MLIR concepts such as match-and-rewrite and staged lowering correspond to established formal theories, providing a basis for more principled abstraction design in compiler infrastructure.