Fast inference engine for Transformer models

Tuned OpenCL BLAS

Serial and parallel implementations of matrix multiplication

Specialized Parallel Linear Algebra, providing distributed GEMM functionality for specific matrix distributions with optional GPU acceleration.

Multiple GEMM operators are constructed with cutlass to support LLM inference.

DGEMM on KNL, achieve 75% MKL

My gemm optimization on RPi (ARM) achieved a 170x performance boost, showing speeds faster than Eigen and close to OpenBLAS.

fast 2D convolution implementation benchmark

Manually optimize the GEMM (GEneral Matrix Multiply) operation. There is a long way to go.

CUDA Gemm Convolution implementation

mixed-precision GEMM library

Development of deep learning inference code by OpenCL kerenl function.

Low Precision Arithmetic for Convolutional Neural Network Inference

WMMA GEMM in ROCm for RDNA GPUs

My experiments with convolution

Yet Another Machine Inference framework

gemm baseline code.

OpenMP Matrix Multiplication Offloading Playground