Introduction to CUDA Kernel Programming

In this course, you will dive into the core mechanics of heterogeneous computing using C++ and CUDA. You will learn how to allocate GPU memory safely, transfer datasets correctly between the CPU and GPU, and configure block dimensions to execute and verify your first one-dimensional kernels.

Master 2D thread indexing in CUDA by mapping multidimensional grids to linear memory. You will implement row-major indexing, build a naive matrix multiplication kernel, and apply 2D grid-stride loops. These skills allow you to process large, rectangular datasets efficiently while ensuring your kernels remain scalable and flexible across various GPU architectures.

In this course, you will tackle global memory latency by harnessing the power of fast, on-chip shared memory. You will learn to synchronize threads using shared memory, implement a boundary-safe tiled matrix multiplication algorithm, and empirically compare it against a naive implementation using validated benchmarks.