================
@@ -43,8 +48,10 @@ _CLC_DEF _CLC_OVERLOAD half __clc_fmin(half x, half y) {
return (y < x) ? y : x;
----------------
frasercrmck wrote:
It doesn't generate the same IR, because SPIR-V isn't optimized. I wonder if we
in fact want to have `half` use `__builtin_fminf16`? We can simplify the
definitions if all types are using a builtin. What do you think? That would
optimize the `half` case, rather than pessimize the `float`/`double` case.
https://github.com/llvm/llvm-project/pull/128506
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
