[libc-commits] [libc] [llvm] [libc][math] Implement half precision lgamma function (PR #192834)

[via libc-commits]

================
@@ -0,0 +1,311 @@
+//===-- Implementation header for lgammaf16 ---------------------*- C++ -*-===//
+//
+// 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_LGAMMAF16_H
+#define LLVM_LIBC_SRC___SUPPORT_MATH_LGAMMAF16_H
+
+#include "include/llvm-libc-macros/float16-macros.h"
+
+#ifdef LIBC_TYPES_HAS_FLOAT16
+
+#include "src/__support/CPP/bit.h"
+#include "src/__support/FPUtil/FEnvImpl.h"
+#include "src/__support/FPUtil/FPBits.h"
+#include "src/__support/FPUtil/NearestIntegerOperations.h"
+#include "src/__support/FPUtil/PolyEval.h"
+#include "src/__support/FPUtil/cast.h"
+#include "src/__support/FPUtil/multiply_add.h"
+#include "src/__support/macros/config.h"
+#include "src/__support/macros/optimization.h"
+#include "src/__support/macros/properties/types.h"
+
+namespace LIBC_NAMESPACE_DECL {
+
+namespace math {
+
+namespace lgammaf16_internal {
+
+LIBC_INLINE constexpr bool is_integer(float16 x) {
+  using FPBits = fputil::FPBits<float16>;
+  FPBits xbits(x);
+  uint16_t x_u = xbits.uintval();
+  unsigned x_e = static_cast<unsigned>(xbits.get_biased_exponent());
+  unsigned lsb = static_cast<unsigned>(
+      cpp::countr_zero(static_cast<uint16_t>(x_u | FPBits::EXP_MASK)));
+  constexpr unsigned UNIT_EXPONENT =
+      static_cast<unsigned>(FPBits::EXP_BIAS + FPBits::FRACTION_LEN);
+  return x_e + lsb >= UNIT_EXPONENT;
+}
+
+// sin(pi*x) for x in (0, 1), written as (0.25 - u^2) * P(u^2) where u = x-0.5.
+// Coefficients for P are a degree-7 polynomial in u^2
+// approximating sin(pi*(u+0.5)) / (pi*(0.25-u^2))
+// with max error ~2^{-55}, generated by Sollya with:
+//   > P = fpminimax(sin(pi*(x+0.5))/(pi*(0.25-x^2)),
+//                   [|0,2,4,6,8,10,12,14|], [|D...|], [-0.5, 0.5]);
+LIBC_INLINE constexpr double lg_sinpi(double x) {
+  constexpr double COEFFS[8] = {0x1p+2,
----------------
lntue wrote:

nit: try to play around so that clang-format can shrink this to 2-3 lines.

https://github.com/llvm/llvm-project/pull/192834