Is it possible to high performance computing by Golang and CUDA?

go cuda opencl archlinux hpc
Yang picture Yang · Feb 18, 2016 · Viewed 15k times · Source

I've googled for a while and the only useful infos are:

  • github.com/barnex/cuda5
  • mumax.github.io/

Unfortunately, the latest Arch Linux only provides CUDA 7.5 package, so the barnex's project may be not supported.

Arne Vansteenkiste recommends concurrency rather than pure Golang or Golang plus CUDA. What's more, there's someone says the same idea that "Wouldn't it be cool to start a goroutine on a GPU and communicate with it via channels?". I think both of these ideas are wonderful since I would like to change the existing code as little as possible instead of refactoring the whole program. Is the idea possible, or is there some documents introducing this topic in details?

Update

It seems that there's two bindings to HPC in Golang:

  • CUDA (< 6.0): github.com/barnex/cuda5
  • OpenCL: github.com/rainliu/gocl

Both of them are less documented, currently what I got is only Macro13's answer, very helpful, but it's more about java . So please help me some detailed materials in Golang. Thanks!

Answer

Related questions

How to check if a map contains a key in Go?

I know I can iterate over a map m by, for k, v := range m { ... } and look for a key but is there a more efficient way of testing a key's existence in a map? I couldn't find the answer …

Read More
How can I convert a zero-terminated byte array to string?

I need to read [100]byte to transfer a bunch of string data. Because not all of the strings are precisely 100 characters long, the remaining part of the byte array is padded with 0s. If I convert [100]byte to string by: …

Read More
How to efficiently concatenate strings in go

In Go, a string is a primitive type, which means it is read-only, and every manipulation of it will create a new string. So if I want to concatenate strings many times without knowing the length of the resulting string, …

Read More