Xiangyu Chang
Community@xchang1121 · United States
A Ph.D student @ University of California, Riverside
Agent Skills by Xiangyu Chang
Showing 88 vetted skills indexed across 1 GitHub repositories.
xchang1121task-constructor
Extract PyTorch/Triton operator implementations into standardized KernelBench task files.
xchang1121performance-summary
Aggregate performance metrics from verify_<op_name>/result.json files across operators.
xchang1121kernel-workflow
Automate AI kernel generation, verification, and optimization across Triton, CUDA C, C++, and TileLang.
xchang1121cuda-c-basics
Explain CUDA-C kernels, memory hierarchy, and grid/block configuration to learners.
xchang1121cuda-c-examples-torch
Integrates CUDA C kernels into PyTorch via load_inline for GPU ops.
xchang1121cuda-c-api
Reference CUDA C APIs for kernel development and GPU memory management.
xchang1121cuda-c-optimization
Optimize CUDA C kernels for performance, memory access, and numerical stability.
xchang1121cuda-c-patterns
Provide CUDA-C pattern templates for vector operations, reductions, and matrix multiplication in GPU kernels.
xchang1121pypto-case-elemwise-gelu
Implement a 1D elementwise GELU activation kernel with PyPTO JIT.
xchang1121pypto-case-matmul-2d
Generate tiled 2D matrix-multiply kernels with Pypto for large-scale workloads.
xchang1121pypto-case-loss-crossentropy
Implement and evaluate a CrossEntropyLoss kernel with per-sample softmax and gather in Pypto.
xchang1121pypto-case-norm-batchnorm
Demonstrate 3D batch normalization with Pypto on batched inputs.
xchang1121pypto-case-reduction-sum
Sum a 3D tensor along a fixed axis while preserving original dimensions.
xchang1121pypto-case-norm-layernorm
Reshape input to 2D and perform looped batch-wise reductions in a PyPto kernel.
xchang1121pypto-case-matvec
Solve large-K matrix-vector multiplication with elementwise multiply-and-sum in pypto.
xchang1121pypto-optimization
Optimize PyPTO tile shapes, loop counts, and reduction-axis merging for tensor operators.
xchang1121pypto-pitfalls
Identify and fix common first-generation PyPTO errors in kernel generation workflows.
xchang1121pypto-loop-view
Enforce compile-time constant shapes for pypto.view inside loops.
xchang1121pypto-api
Inspect PyPTO API signatures and constraints for kernel implementation.
xchang1121pypto-basics
Define PyPTO programming principles for kernel development and semantic contracts.
xchang1121kernel-agent-overview
Generates and validates operator kernels via a React-based workflow with user confirmations.
xchang1121cpu-optimization-x64
Optimize x64 CPU code with SIMD/AVX vectorization and cache-aware layouts.
xchang1121cpu-optimization-arm
Identify ARM CPU bottlenecks and apply NEON vectorization with cache-friendly patterns.
xchang1121cpu-basics
Teach C++ CPU kernel design for PyTorch extensions with memory safety.