Sign in Subscribe

Mini Projects

Each mini project has a clear set of goals. Explore and understand different concepts practically while working towards achieving those goals.

Mini Projects
Photo by Erik Mclean / Unsplash

xGeMM

GitHub - tgautam03/xGeMM: Accelerated General (FP32) Matrix Multiplication
Accelerated General (FP32) Matrix Multiplication. Contribute to tgautam03/xGeMM development by creating an account on GitHub.
GitHubtgautam03

Goal: Code matrix multiplication from scratch and (try to) match the performance of cuBLAS SGeMM.

Content Index

  1. Why care about matrix multiplication?
  2. What is SGeMM?
  3. Naive matrix multiplication on a GPU.
  4. Matrix multiplication on a GPU with coalesced memory accesses.
  5. Matrix multiplication on a GPU using shared memory.
  6. Matrix multiplication on a GPU using registers.
  7. Matrix multiplication on a GPU using even more registers.
  8. Matrix multiplication on a GPU with vectorized memory accesses.