| Issue |
160312
|
| Summary |
Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow F16C CVTPS2PH intrinsics to be used in constexpr
|
| Labels |
good first issue,
clang,
backend:X86,
constexpr,
clang:bytecode
|
| Assignees |
|
| Reporter |
RKSimon
|
```
__builtin_ia32_vcvtps2ph
__builtin_ia32_vcvtps2ph256
```
If the rounding mode argument is not using the runtime MXCSR.RC, or the conversion is exact it should be possible for the float to half truncation intrinsics to be performed in constant expressions
```
/// \param imm
/// An immediate value controlling rounding using bits [2:0]: \n
/// 000: Nearest \n
/// 001: Down \n
/// 010: Up \n
/// 011: Truncate \n
/// 1XX: Use MXCSR.RC for rounding
```
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
