AMDGPU LLVM Adding GFX 9/10/11 "Generic Targets" To Build Once & Run On Multiple GPUs

[phoronix]

Phoronix: AMDGPU LLVM Adding GFX 9/10/11 "Generic Targets" To Build Once & Run On Multiple GPUs

Code merged today to mainline LLVM is preparing for the notion of generic targets across the GFX9, GFX10, and GFX11 GPU families. With follow-on work these generic targets are aiming to allow compiling code once and then running across multiple GPUs in the given hardware family...

https://www.phoronix.com/news/LLVM-AMDGPU-Generic-GFX

[elbar]

If there is Mesa using LLVM and there is also GCC(and maybe VS), how unfortunate this is...

LLVM is seen as top notch for today's graphic mesa stack, but something better is required... mesa multiple transitions between drivers made it somehow adoptable by possibilities not possible by other means but that does not mean that mesa architecture with LLVM is correct and when Vulkan in place this is the only solution for applications... Think especially with X11 in mind when desktop characteristics of graphic drivers is only part time job for graphic stack with Direct3D missing and problem is that requirement of C/C++ with restriction for desktop extensions built in in graphic stack mean also that such transition for corporate requirements(somehow different that common linux standards) mean huge problem with developing their apps not been able to benefit with some stack. Graphics providers like Intel and AMD may think that Mesa is helping them, but applications are then restricted to OpenGL or Vulkan or Direct3D on Windows and this is limiting especially when Mesa is in C++ and no possibility to developer their needs...

Maybe not so LLVM problem, but LLVM although modern solution to Mesa and apps problems, it does not move above Linux shadows...

[FireBurn]

If there is Mesa using LLVM and there is also GCC(and maybe VS), how unfortunate this is...

LLVM is seen as top notch for today's graphic mesa stack, but something better is required... mesa multiple transitions between drivers made it somehow adoptable by possibilities not possible by other means but that does not mean that mesa architecture with LLVM is correct and when Vulkan in place this is the only solution for applications... Think especially with X11 in mind when desktop characteristics of graphic drivers is only part time job for graphic stack with Direct3D missing and problem is that requirement of C/C++ with restriction for desktop extensions built in in graphic stack mean also that such transition for corporate requirements(somehow different that common linux standards) mean huge problem with developing their apps not been able to benefit with some stack. Graphics providers like Intel and AMD may think that Mesa is helping them, but applications are then restricted to OpenGL or Vulkan or Direct3D on Windows and this is limiting especially when Mesa is in C++ and no possibility to developer their needs...

Maybe not so LLVM problem, but LLVM although modern solution to Mesa and apps problems, it does not move above Linux shadows...

What?

[stormcrow]

What?

Skimming through the account's comment history I suspect Google (or someone else) is making a complete word salad out of the translations. None of them make sense. Either chat bot or some really bad translation services.

[cgmb]

If (or when) these targets are adopted by the ROCm libraries, it should end the era of `HSA_OVERRIDE_GFX_VERSION`. That environment variable can go back to being a niche tool for debugging, rather than the necessity that it became.