1d_mgpu_c2c

cuFFT MultiGPU 1D FFT C2C example

Description

This code demonstrates the usage of single node, multiGPU cuFFT C2C functionality, on a 1D dataset. Use single gpu version as reference. Maximum FFT size limited by single GPU memory.

Note, because cuFFT 10.4.0 cufftSetStream can be used to associate a stream with multi-GPU plan. cufftXtExecDescriptor synchronizes efficiently to the stream before and after execution. Please refer to https://docs.nvidia.com/cuda/cufft/index.html#function-cufftsetstream for more information.

cuFFT by default executes multi-GPU plans in synchronous manner.

The cuFFT library doesn't guarantee that single-GPU and multi-GPU cuFFT plans will perform mathematical operations in same order. Small numerical differences are possible.

Supported SM Architectures

All GPUs supported by CUDA Toolkit (https://developer.nvidia.com/cuda-gpus)

Supported OSes

Linux
Windows

Supported CPU Architecture

x86_64
ppc64le
arm64-sbsa

CUDA APIs involved

• cufftXtSetGPUs API
• cufftMakePlan1d API
• cufftXtExecDescriptor API
• cufftXtMalloc API
• cufftXtMemcpy API

Building (make)

Prerequisites

• A Linux/Windows system with recent NVIDIA drivers.
• CMake version 3.18 minimum

Build command on Linux

$ mkdir build
$ cd build
$ cmake ..
$ make

Make sure that CMake finds expected CUDA Toolkit. If that is not the case you can add argument -DCMAKE_CUDA_COMPILER=/path/to/cuda/bin/nvcc to cmake command.

Usage 1

$  ./bin/1d_mgpu_c2c_example

Sample example output:

PASSED with L2 error = 1.25194e-07