Intel Posts FP16 GLSL Patches For Mesa
Topi Pohjolainen of Intel has been working on adding proper FP16/half-precision support to Mesa's GLSL code.
Being pushed along by Igalia working on 16-bit storage for SPIR-V/Vulkan, Topi has been working on 16-bit GLSL support in the Mesa code. This includes patches for basic 16-bit lowering with NIR, FP16 support improvements within the Intel compiler code, and the common Mesa GLSL changes.
These 51 patches for now are on Mesa-dev. Using 16-bit/half-precision can save on storage and bandwidth requirements where supported by the hardware. On the Intel Vulkan side, their 16-bit support is further along.
Being pushed along by Igalia working on 16-bit storage for SPIR-V/Vulkan, Topi has been working on 16-bit GLSL support in the Mesa code. This includes patches for basic 16-bit lowering with NIR, FP16 support improvements within the Intel compiler code, and the common Mesa GLSL changes.
These 51 patches for now are on Mesa-dev. Using 16-bit/half-precision can save on storage and bandwidth requirements where supported by the hardware. On the Intel Vulkan side, their 16-bit support is further along.
4 Comments