[libc-commits] [libc] [llvm] [libc][math][c23] Add atanbf16 function (PR #184019)

[via libc-commits]

================
@@ -67,6 +67,12 @@ LIBC_INLINE bfloat16 atanbf16(bfloat16 x) {
   // > display = hexadecimal ;
   // > P = fpminimax(atan(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14|], [|SG, SG,
   // SG..SG|], [0, 1]);
+  //
+  // absolute error for the polynomial given by:
+  // dirtyinfnorm(atan(x) - x * P(x), [0, 1]) gives
----------------
lntue wrote:

Try relative errors estimate by
```
> dirtyinfnorm(atan(x)/x - P(x), [0, 1]);
```

https://github.com/llvm/llvm-project/pull/184019