←back to thread

ML needs a new programming language – Interview with Chris Lattner

(signalsandthreads.com)
311 points melodyogonna | 7 comments | 05 Sep 25 11:33 UTC | HN request time: 0s | source | bottom
Show context
Cynddl ◴[05 Sep 25 12:46 UTC] No.45137883[source]▶
Anyone knows what Mojo is doing that Julia cannot do? I appreciate that Julia is currently limited by its ecosystem (although it does interface nicely with Python), but I don't see how Mojo is any better then.
replies(8): >>45137977 #>>45137999 #>>45138029 #>>45138039 #>>45138393 #>>45138420 #>>45138513 #>>45143187 #
thetwentyone ◴[05 Sep 25 12:58 UTC] No.45137999[source]▶
Especially because Julia has pretty user friendly and robust GPU capabilities such as JuliaGPU and Reactant[2] among other generic-Julia-code to GPU options.

1: https://enzymead.github.io/Reactant.jl/dev/ 2: https://enzymead.github.io/Reactant.jl/dev/

replies(1): >>45138080 #
1. jb1991 ◴[05 Sep 25 13:05 UTC] No.45138080[source]▶
I get the impression that most of the comments in this thread don't understand what a GPU kernel is. These high-level languages like Python and Julia are not running on the kernel, they are calling into other kernels usually written in C++. The goal is different with Mojo, it says at the top of the article:

> write state of the art kernels

You don't write kernels in Julia.

replies(5): >>45138165 #>>45138182 #>>45138372 #>>45138558 #>>45138820 #
2. jakobnissen ◴[05 Sep 25 13:14 UTC] No.45138165[source]▶
Im pretty sure Julia does JIT compilation of pure Julia to the GPU: https://github.com/JuliaGPU/GPUCompiler.jl
replies(1): >>45138233 #
3. ssfrr ◴[05 Sep 25 13:15 UTC] No.45138182[source]▶
It doesn’t make sense to lump python and Julia together in this high-level/low-level split. Julia is like python if numba were built-in - your code gets jit compiled to native code so you can (for example) write for loops to process an array without the interpreter overhead you get with python.

People have used the same infrastructure to allow you to compile Julia code (with restrictions) into GPU kernels

4. actionfromafar ◴[05 Sep 25 13:19 UTC] No.45138233[source]▶
” you should use one of the packages that builds on GPUCompiler.jl, such as CUDA.jl, AMDGPU.jl, Metal.jl, oneAPI.jl, or OpenCL.jl”

Not sure how that organization compares to Mojo.

5. arbitrandomuser ◴[05 Sep 25 13:34 UTC] No.45138372[source]▶
>You don't write kernels in Julia.

The package https://github.com/JuliaGPU/KernelAbstractions.jl was specifically designed so that julia can be compiled down to kernels.

Julia's is high level yes, but Julia's semantics allow it to be compiled down to machine code without a "runtime interpretter" . This is a core differentiating feature from Python. Julia can be used to write gpu kernels.

6. pjmlp ◴[05 Sep 25 13:49 UTC] No.45138558[source]▶
See new cu tile architecture on CUDA, designed from the ground up with Python in mind.
7. adgjlsfhk1 ◴[05 Sep 25 14:11 UTC] No.45138820[source]▶
Julia's GPU stack doesn't compile to C++. it compiles Julia straight to GPU assembly.