1. Martin Kruliš 26. 11. 2015 by Martin Kruliš (v1.0)1

2.  Interoperability ◦ Allows CUDA code to read/write graphical buffers  Works with OpenGL and Direct3D libraries ◦ Motivation  Direct visualization of complex simulations  Augmenting 3D rendering with visualization routines which are difficult to implement in shaders ◦ How it works  The graphics resource is registered and represented by struct cudaGraphicResource  The resource may be mapped to CUDA memory space cudaGraphicsMapResources(), … 26. 11. 2015 by Martin Kruliš (v1.0)2

3.  Initialization ◦ Device must be selected by cudaGLSetGLDevice()  Resources ◦ cudaGraphicsGLRegisterBuffer() for buffers  The mapped buffers can be accessed in the same way as CUDA allocated memory ◦ cudaGraphicsGLRegisterImage() for images and render buffers  The image buffers can be also accessed through texture and surface mechanisms 26. 11. 2015 by Martin Kruliš (v1.0)3 Examples

4.  Direct3D Support ◦ Versions 9, 10, and 11 are supported  Each version has its own API ◦ CUDA context may operate with one Direct3D device at a time  And special HW mode must be set on the device ◦ Initialization is similar to OpenGL cudaD3D[9|10|11]SetDirect3DDevice() ◦ Available Direct3D resources  Buffers, textures, and surfaces  All using cudaGraphicsD3DXXRegisterResource() 26. 11. 2015 by Martin Kruliš (v1.0)4

5.  GPU SLI Mode ◦ Multiple GPUs are interconnected (physically) and cooperating in rendering the scene  AFR mode – different GPUs render subsequent frames ◦ CUDA interoperability issues  Any CUDA allocation on one GPU is automatically performed on all SLI-connected GPUs  CUDA has to use separate contexts for each GPU  cudaGLGetDevices() – identify, which devices are in SLI  cudaGLDeviceListAll  cudaGLDeviceListCurrentFrame  cudaGLDeviceListNextFrame 26. 11. 2015 by Martin Kruliš (v1.0)5

6. 26. 11. 2015 by Martin Kruliš (v1.0)6

7. Martin Kruliš 26. 11. 2015 by Martin Kruliš (v1.0)7

8.  CUDA Basic Linear Algebra Subroutines ◦ CUDA implementation of standard BLAS library ◦ Complete support of all 152 functions on vectors/matrices  copy, move, rotate, swap  maximum, minimum, multiply by scalar  sum, dot products, Euclidean norms  matrix multiplications, inverses, linear combinations ◦ Some operations have batch versions ◦ Supports floats, doubles, and complex numbers 26. 11. 2015 by Martin Kruliš (v1.0)8

9.  CUDA Sparse Linear Algebra ◦ Open source C++ library for sparse linear structures (matrices, linear systems, …) ◦ Key features  Sparse matrix operations (add, substraction, max independent set, polynomial relaxation, …)  Supports various matrix formats  COO, CSR, DIA, ELL, and HYB ◦ Require CUDA CC 2.0 or higher 26. 11. 2015 by Martin Kruliš (v1.0)9

10.  CUDA Fast Fourier Transform ◦ Decompose signal to frequency spectrum ◦ 1-3D transforms (up to 128M elements) ◦ Many variations (precision, complex/real types, …) ◦ API similar to FFTW library  Create plan ( cufftHandle ) which holds the configuration  Associate/allocate work space (buffers)  cufftExecC2C() (or R2C, C2R ) starts execution ◦ FFT plan can be associated with CUDA stream  For synchronization and overlapping 26. 11. 2015 by Martin Kruliš (v1.0)10

11.  CUDA Thrust ◦ C++ template library based on STL API ◦ Basic idea is to develop C++ parallel applications with minimal overhead ◦ STL like vectors (for devices) and vector operations  copy, fill, create sequences, reordering, sorting, … ◦ Algorithms  Transformations  Reductions  Prefix-sums 26. 11. 2015 by Martin Kruliš (v1.0)11

12.  GPU AI for Board Games ◦ Specific AI library designed for games with large, but well-defined configuration space ◦ Requires CUDA CC 2.0 ◦ Currently supports  Game Tree Split – alpha/beta pruning  Single and multiple recursion (with large depths)  Zero-sum games (3D Tic-Tac-Toe, Reversi, …)  Sudoku backtracking generator and solver  Statistical simulations (Monte Carlo for Go) 26. 11. 2015 by Martin Kruliš (v1.0)12

13.  PhysX ◦ Realtime physics engine ◦ Originally developed by Ageia for PPU card  NVIDIA bought it and re-implemented it for CUDA ◦ Most important features  Simulation of rigid bodies (collisions, destruction)  Cloths and fluid particle systems  APEX ◦ Framework built on top of PhysX ◦ Designed for easy usage (artists, games, …) 26. 11. 2015 by Martin Kruliš (v1.0)13

14. 26. 11. 2015 by Martin Kruliš (v1.0)14