CUDA Programming for Dummies --- in ML perspective

CUDA examples and exercises focused on performance optimization, parallel algorithms, and their application to fundamental Deep Learning components.

Project Goals

This repository serves as an interactive learning environment to master key parallel computing concepts:

CUDA concepts: High level CUDA concepts including threads, synchronisation, shared memory and tiling.
Thrust Proficiency: Use NVIDIA's Thrust library for highly-optimized parallel patterns (e.g., sort, reduce, transform).
Application: Apply CUDA to Matrix Multiplication (GEMM) and basic Neural Network architectures.

File/Area	Concept Learned	Primary Task
`optimized_max_displacement.cu`	Fused Operations	Analyze the memory access pattern of the zip iterator.
`performance_comparison.cu`	Benchmarking	Benchmark naive vs. optimized code across varying data sizes.
`matmul/`	Tiled Kernels	Implement and test a tiled GEMM kernel for cache reuse.
`neural_nets/`	Element-wise Transforms	Use `thrust::transform` to implement custom ReLU/Sigmoid activation functions.