Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions configs/cuda_fp16.yml
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,9 @@ Version: 0.0.1
Entry Point: ./numba_cuda/numba/cuda/include/12/cuda_fp16.h
File List:
- ./numba_cuda/numba/cuda/include/12/cuda_fp16.h
Exclude:
Struct:
- __half2
Types:
__half: Number
__half2: Type
Expand All @@ -13,3 +16,6 @@ Data Models:
__half2: StructModel
Shim Include Override: "\"cuda_fp16.h\""
Require Pynvjitlink: False
Use Separate Registry: True
GPU Arch:
- sm_86
Copy link
Contributor

@isVoid isVoid Aug 28, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

GPU Arch defines the macro CUDA_ARCH when clang is used to parse fp16 headers. We don't have a good guidance of this value yet. For float16, after sm70 it is feature complete. So any value after 70 should give us all of the API that's in the header.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ideally - Numbast generated bindings should generate different bindings for different arches based on supported arches the library is generating for. Then use difftools to conditionally enable/disable these features.

Loading