CUDA programming. CUDA requirements. CUDA Querying. CUDA Querying. A CUDA-capable GPU (NVIDIA) NVIDIA driver A CUDA SDK

Transcription

1 CUDA programming Bedrich Benes, Ph.D. Purdue University Department of Computer Graphics CUDA requirements A CUDA-capable GPU (NVIDIA) NVIDIA driver A CUDA SDK Standard C compiler What do I have? cudadeviceprop prop; int n; cudagetdevicecount(&n);//how many GPUs? for (int i=0;i<n;i++) { cudagetdeviceproperties(&prop,i); printf( %i %i,prop.major,prop.minor); } Device 0: "GeForce GTX 780" CUDA Driver Version / Runtime Version 6.5 / 6.5 CUDA Capability Major/Minor version number: 3.5 Total amount of global memory: 3072 MBytes ( bytes) (12) Multiprocessors, (192) CUDA Cores/MP: 2304 CUDA Cores GPU Clock rate: 902 MHz (0.90 GHz) Memory Clock rate: 3004 Mhz Memory Bus Width: 384-bit L2 Cache Size: bytes Maximum Texture Dimension Size (x,y,z) 1D=(65536), 2D=(65536, 65536),3D=(4096, 4096, 4096) Maximum Layered 1D Texture Size, (num) layers 1D=(16384), 2048 layers Maximum Layered 2D Texture Size, (num) layers 2D=(16384, 16384), 2048 layers Total amount of constant memory: bytes Total amount of shared memory per block: bytes Total number of registers available per block: Warp size: 32 1

2 Maximum number of threads per multiprocessor: 2048 Maximum number of threads per block: 1024 Max dimension size of a thread block (x,y,z): (1024, 1024, 64) Max dimension size of a grid size (x,y,z): ( , 65535, 65535) Maximum memory pitch: bytes Texture alignment: 512 bytes CUDA query example Concurrent copy and kernel execution: Yes with 1 copy engine(s) Run time limit on kernels: Yes Integrated GPU sharing Host Memory: No Support host page-locked memory mapping: Yes Alignment requirement for Surfaces: Yes Device has ECC support: Disabled CUDA Device Driver Mode (TCC or WDDM): WDDM (Windows Display Driver Model) Device supports Unified Addressing (UVA): No Device PCI Bus ID / PCI location ID: 2 / 0 Compute Mode: < Default (multiple host threads can use ::cudasetdevice() with device simu ltaneously) > devicequery, CUDA Driver = CUDART, CUDA Driver Version = 6.5, CUDA Runtime Version = 6.5, NumDevs = 1, Device0 = GeForce GTX 780 CUDA instruction set is called Parallel Thread Execution PTX A kernel can be written in PTX, but it is tedious Nvidia C compiler - nvcc makes this task easier. CUDA & C source code (*.cu, *.cpp) CUDA libraries (FFT, BLAS, etc.) nvcc.exe (NVIDIA compiler) PTX (NVIDIA assembly code) ASM (CPU host code) CUDA Driver C compiler C libraries GPU CPU Offline compilation Compiles the device code into the assembly code PTX is a device independent object code PTX is compiled to a particular device PTX can be executed on a device different than the device that has generated it cubin cuda binary 2

3 Just in time compilation Virtual layer CUDA & C source code (*.cu) Physical layer PTX2Target Compiler PTX is further compiled to binary The binary code is cached in compute cache nvcc.exe (NVIDIA compiler) PTX (NVIDIA assembly code) GPU1 GPU2 GPUn CUDA C Runtime The runtime is in library: cudart.lib static link cudart.dll dynamic link Build configurations nvcc <filename>.cu [-o executable] generates release mode nvcc -g <filename>.cu debug mode (for host code only, not for device) nvcc -deviceemu <filename>.cu builds device emulation mode all runs on the CPU, no debug symbols nvcc -deviceemu -g <filename>.cu debug device emulation mode (all cpu & debug symbols) 3

4 Create empty project (Win32 console application) Select Build Customizations Select CUDA Add your file.cu to the project Select the source *.cpp or *.cu file and assign the custom compile to it multiple files in a project *.cpp files can be normally used and will be linked 4

5 .net Debugging You can step in read values etc. using standard.net tools Reading NVIDIA CUDA Programming Guide Kirk, D.B., Hwu, W.W., Programming Massively Parallel Processors, NVIDIA, Morgan Kaufmann 2010 C:\ProgramData\NVIDIA Corporation\CUDA Samples\v6.5 5