Hyun Gyu Kim

Introduction

Education

B.S. in Computer Science and Engineering, Sogang University, Cum Laude

Research

Accept

Reviewed

• Scalable High-Order Stencil Acceleration via Sender-Receiver Dataflow in Spatial Architectures @ IEEE Transactions on Parallel and Distributed Systems (TPDS)

Open Source Contribution

Merged

• [TIR][Schedule] Add FuseReductionEpilogue primitive to fuse epilogue @ TVM Post

• Add ONNXToLinalg conversion pass with MatMul support @ ONNX-MLIR Post

• [TIR][Schedule] FuseReductionEpilogue: Add Clipping pattern support @ TVM Post

• ONNX To LLVM Pass Backend using linalg @ ONNX-MLIR Post

• Enable One-Shot Bufferization for Mixed Linalg and ONNX Operations @ ONNX-MLIR Post

• [TIR][Schedule]Generalize fuseReductionEpilogue to support arbitrary epilogue expressions @ TVM Post

• Fix memory corruption in LiftCFGToSCF when processing empty regions @ IREE

• Implement Selective Linalg Conversion with –linalg-ops Option @ ONNX-MLIR

• Refactor ONNXToLinalg pass to use TableGen @ ONNX-MLIR

• Implement ONNX Basic Conv to Linalg lowering @ ONNX-MLIR Post

Reviewed

• [Optimization][Operator] Implement and enable Conv2d-Reshape-Add-ReLU fusion @ TVM Post

• Propagate binding correlation information to LLVM codegen for mutable binding fusion @ IREE

• Fix thread-safety issue in ONNX to Linalg conversion @ ONNX-MLIR

• Add dynamic shape support for ONNX Conv to Linalg lowering @ ONNX-MLIR