PyTorch adds FlyDSL kernels for AMD MI350-series GPUs
AIPyTorch now lets users on AMD MI350-series GPUs run FlyDSL kernels through existing APIs for dense and grouped GEMM, MXFP8/MXFP4 scaled GEMM, RMSNorm, TopK, and attention. Reported kernel-level geometric-mean speedups range from 1.10x to 4.82x against ATen or Triton, depending on the operator and baseline.
AIWhy it matters.
The post lists exact operator coverage, the dispatch path for each API, and benchmark baselines, so readers can judge where FlyDSL applies to their AMD workloads.
Source details
PyTorch Blog · AMD: Xiaobing Zhang, Yutao Xu, Rui Li, Heng Yi, Xinya Zhang, Shijie Feng, Felix Li, Jeff Daily, Jiong Gong, Hattie Wu, Peng Sun, Emad Barsoum; META: Jason Ansel and Simon Layton · pytorch.orgPublished · added here
Original title: Accelerating PyTorch on AMD GPUs with FlyDSL