An LLM given profiler metrics and architecture details automatically generates cache-locality swizzles that speed up 9 of 10 GPU kernels on AMD MI300X, up to 2.06x.
Learning to optimize halide with tree search and random programs
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.DC 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
SwizzlePerf: Hardware-Aware LLMs for GPU Kernel Performance Optimization
An LLM given profiler metrics and architecture details automatically generates cache-locality swizzles that speed up 9 of 10 GPU kernels on AMD MI300X, up to 2.06x.