[libc-commits] [libc] [llvm] [libc][math][c23] Add tanbf16 math function (PR #185100)

[via libc-commits]

================
@@ -0,0 +1,84 @@
+//===-- Implementation of tanbf16 function --------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIBC_SRC___SUPPORT_MATH_TANBF16_H
+#define LLVM_LIBC_SRC___SUPPORT_MATH_TANBF16_H
+
+#include "hdr/errno_macros.h"
+#include "hdr/fenv_macros.h"
+#include "sincosf_utils.h"
+#include "src/__support/FPUtil/FEnvImpl.h"
+#include "src/__support/FPUtil/FPBits.h"
+#include "src/__support/FPUtil/bfloat16.h"
+#include "src/__support/FPUtil/cast.h"
+#include "src/__support/FPUtil/multiply_add.h"
+#include "src/__support/macros/optimization.h"
+
+namespace LIBC_NAMESPACE_DECL {
+
+namespace math {
+
+LIBC_INLINE bfloat16 tanbf16(bfloat16 x) {
+  using namespace sincosf_utils_internal;
+  using FPBits = fputil::FPBits<bfloat16>;
+  FPBits xbits(x);
+
+  uint16_t x_u = xbits.uintval();
+  uint16_t x_abs = x_u & 0x7fff;
+  float xf = x;
+
+  // NaN or -/+ INF
+  if (x_abs >= 0x7F80) {
+    // NaN
+    if (xbits.is_nan()) {
+      if (xbits.is_signaling_nan()) {
+        fputil::raise_except_if_required(FE_INVALID);
+        return FPBits::quiet_nan().get_val();
+      }
+      return x;
+    }
+    //|x| is +/- INF
+    fputil::set_errno_if_required(EDOM);
+    fputil::raise_except_if_required(FE_INVALID);
+    return x + FPBits::quiet_nan().get_val();
+  }
+
+  // |x| = {0}
+  if (LIBC_UNLIKELY(x_abs == 0)) {
+    return x;
+  }
+
+  // Through Exhaustive testing -
+  // The last value where tan(x) ~ x is 0x3db8
+  if (LIBC_UNLIKELY(x_abs <= 0x3db8)) {
+    int rounding = fputil::quick_get_round();
+    // separate case handles it with magnitude of 2^-11
+    if ((xbits.is_pos() && rounding == FE_UPWARD) ||
+        (xbits.is_neg() && rounding == FE_DOWNWARD))
+      return fputil::cast<bfloat16>(fputil::multiply_add(xf, 0x1.0p-11f, xf));
----------------
lntue wrote:

Why don't we just do `fputil::multiply_add(xf, 0x1.0p-13f, xf)` for all the rounding modes, and don't need to do `quick_get_round()`.

https://github.com/llvm/llvm-project/pull/185100