Generic GPU Kernels in Julia
mikeinnes.github.io
Generic GPU Kernels in Julia
1–10 of 15 posts
Re: Generic GPU Kernels in Julia
#2Generic GPU kernels... for just NVIDIA devices.
Re: Generic GPU Kernels in Julia
#3Generic GPU kernels... for just NVIDIA devices.
The `generic` in the title applies to `kernel` rather than `GPU`. However, since LLVM does have an AMDGPU backend I suspect it's possible to do the same for AMD GPUs as well using the same technique and without changing the frontend code (assuming the AMD driver accepts whatever comes out of the AMDGPU backend). However, I can hardly blame folks for focusing on NVIDIA devices. For better or worse, they're dominating the GPGPU space right now. Just as an example, all the cloud providers have NVIDIA GPUs on offer, but the same is not true for AMD GPUs (though IIRC, Google is planning to offer them in the next generation).
Re: Generic GPU Kernels in Julia
#4Generic GPU kernels... for just NVIDIA devices.
To add to Keno's comment, a bunch of this stuff actually can work with OpenCL and OpenGL, and you can see it in action in GPUArrays.jl [1]. In general, there's a bunch of work to make Julia as hardware-agnostic as possible, but you gotta start somewhere.
Re: Generic GPU Kernels in Julia
#5Generic GPU kernels... for just NVIDIA devices.
Yeah, I hate vendor lockin and would love to see this extended to OpenCL
Re: Generic GPU Kernels in Julia
#6Generic GPU kernels... for just NVIDIA devices.
To add to Keno's comment, a bunch of this stuff actually can work with OpenCL and OpenGL, and you can see it in action in GPUArrays.jl [1]. In general, there's a bunch of work to make Julia as hardware-agnostic as possible, but you gotta start somewhere. [1]: https://github.com/JuliaGPU/GPUArrays.jl
Very cool!
My comment came off as more snarky than I intended. I just like choice. I'm an HPC guy so I follow Julia with great interest.
Re: Generic GPU Kernels in Julia
#7I hadn't really looked at Julia before, so I started digging, and its really nice! https://julialang.org/
Given this addition of GPGPU required for contemporary compute applications, Julia looks like a compelling alternative to other languages in the space.
Re: Generic GPU Kernels in Julia
#8Generic GPU kernels... for just NVIDIA devices.
Yeah, I hate vendor lockin and would love to see this extended to OpenCL
As replied in the thread above, the GPUArrays package can use OpenCL: https://github.com/JuliaGPU/GPUArrays.jl
More details in this package: https://github.com/SimonDanisch/Transpiler.jl
Re: Generic GPU Kernels in Julia
#9It may not be immediately obvious where the code for this package is. For those who want to follow along:
Re: Generic GPU Kernels in Julia
#10> This creates a single kernel call, with no memory allocation or temporary arrays required. Pretty cool – and well out of the reach any other system I know of.
D with DCompute can do that as well.