Memory Access Basics:
Vector Add - grid, block, thread indexing basics.
Matrix Addition - 2D matrix, row major and column major.
Julia set - graphical representation of equation.
Mandelbrot Fractals - graphical representation.
Tree Fractal - graphical represention with rendering.
Tiling and coalesing:
Naive Matrix Multiplication - 1 thread per resultant matrix element.
Tiled Matrix Multiplication - shared memory access.
Coalescing Matrix Multiplication - choosing indexing of the matrix so that they are coalesced.
Tiled and Coalesced matrix multiplication.
Bench marking of all the matrix multiplication.
Control flow and divergence:
Naive Reduction
Shared Memory Reduction
Unroll last warp Reduction
Bank Conflict:
Naive Matrix Transpose - reads coalesced, writes strided
Tiled Matrix Transpose - loads and stores coalesced with shared memory (bank conflicts)
Tiled and Padded Transpose - fixing banking conflicts
Parallel Patterns:
Convolution 1 D using constant memory
Convolution 2 D using constant memory
Prefix sum - reduction but produces partial sums. (divergence and synchronization)
Histogram computation
Sparse Matrix computation
Merge Sort
Graph Search
Dynamic Parallelism
All the bench marking are done on Jetson TX2 and Nvidia Orin Nano development board.
