[clang] Add support for __builtin_amdgcn_frexp_expf (PR #242167)

Steffen Larsen via cfe-commits cfe-commits at lists.llvm.org
Sun Nov 15 22:10:25 PST 2026


================
@@ -655,8 +655,11 @@ CIRGenFunction::emitAMDGPUBuiltinExpr(unsigned builtinId,
     return emitBuiltinWithOneOverloadedType<1>(expr, "amdgcn.frexp.mant")
         .getValue();
   }
-  case AMDGPU::BI__builtin_amdgcn_frexp_exp:
   case AMDGPU::BI__builtin_amdgcn_frexp_expf:
----------------
steffenlarsen wrote:

I believe all the `exp` variants should be able to go through this path the same way.
```suggestion
  case AMDGPU::BI__builtin_amdgcn_frexp_exp:
  case AMDGPU::BI__builtin_amdgcn_frexp_expf:
  case AMDGPU::BI__builtin_amdgcn_frexp_exph:
```
This would of course need more testing, but you could take inspiration from existing CodeGenHIP tests.

https://github.com/llvm/llvm-project/pull/242167


More information about the cfe-commits mailing list