01-SGEMM Tutorial

This module is the entry point of CUDA Kernel Academy, focusing on the progressive optimization of SGEMM.

Learning Path

Build

This module is a standalone tutorial and does not participate in the root CMake build.

cd 01-sgemm-tutorial
make GPU_ARCH=sm_86
./build/sgemm_benchmark

Sub-topics

• Naive SGEMM
• Tiled SGEMM
• Bank Conflict Free
• Double Buffer SGEMM
• Tensor Core SGEMM

References