FP16 Half-Precision Support Added To AMDGPU LLVM Back-End
The AMDGPU backend within LLVM that's used by the open-source Radeon graphics driver stack has landed half-precision / FP16 support.
With this commit overnight is nearly five thousand lines of new code for supporting F16 half-precision support in this GPU back-end for AMD hardware. This new LLVM code was under review the past few weeks but is now ready for LLVM 4.0.
This half-precision support is capable on AMD Volcanic Islands (3rd gen GCN) hardware and newer. Now if only the open-source AMD OpenCL/compute state was much more mature and widespread...
With this commit overnight is nearly five thousand lines of new code for supporting F16 half-precision support in this GPU back-end for AMD hardware. This new LLVM code was under review the past few weeks but is now ready for LLVM 4.0.
This half-precision support is capable on AMD Volcanic Islands (3rd gen GCN) hardware and newer. Now if only the open-source AMD OpenCL/compute state was much more mature and widespread...
15 Comments