Examining AMD's RDNA 4 Changes in LLVM
chipsandcheese.com
Examining AMD's RDNA 4 Changes in LLVM
1–4 of 4 posts
Re: Examining AMD's RDNA 4 Changes in LLVM
#2I think we really need a minimal format for binary shaders (I really mean minimal). I wonder if mesa with ACO has one. GPU is all about performance, optimized assembly is more than appropriate. Still could have a reference implementation authored directly in SPIR-V (text->binary SPIR-V with a ssa checker), to avoid pulling those abominations of glsl or hlsl compilers (unless it exists plain and simple C written glsl or hlsl compilers to SPIR-V).
This web site is really good, reaaaaaaally good.
Re: Examining AMD's RDNA 4 Changes in LLVM
#3I think we really need a minimal format for binary shaders (I really mean minimal). I wonder if mesa with ACO has one. GPU is all about performance, optimized assembly is more than appropriate. Still could have a reference implementation authored directly in SPIR-V (text->binary SPIR-V with a ssa checker), to avoid pulling those abominations of glsl or hlsl compilers (unless it exists plain and simple C written glsl…
Not a glsl or hlsl compiler, but there is a C compiler that targets spirv and is written in C. Here's the repo:
https://github.com/heroseh/hcc
Re: Examining AMD's RDNA 4 Changes in LLVM
#4I think we really need a minimal format for binary shaders (I really mean minimal). I wonder if mesa with ACO has one. GPU is all about performance, optimized assembly is more than appropriate. Still could have a reference implementation authored directly in SPIR-V (text->binary SPIR-V with a ssa checker), to avoid pulling those abominations of glsl or hlsl compilers (unless it exists plain and simple C written glsl…
Not a glsl or hlsl compiler, but there is a C compiler that targets spirv and is written in C. Here's the repo: https://github.com/heroseh/hcc
Would it be that hard to make it understand glsl?
I have no idea how far apart glsl and C are.