[libc][math][c23] Add exp2m1f C23 math function

overmighty · overmighty · commit 79c6351d3e27 · 2024-04-03T23:06:52.000+01:00
Fixes #86502.
diff --git a/libc/config/linux/x86_64/entrypoints.txt b/libc/config/linux/x86_64/entrypoints.txt
@@ -370,6 +370,7 @@ set(TARGET_LIBM_ENTRYPOINTS
     libc.src.math.exp10f
     libc.src.math.exp2
     libc.src.math.exp2f
+    libc.src.math.exp2m1f
     libc.src.math.expm1
     libc.src.math.expm1f
     libc.src.math.fabs
diff --git a/libc/spec/stdc.td b/libc/spec/stdc.td
@@ -535,6 +535,8 @@ def StdC : StandardSpec<"stdc"> {
           FunctionSpec<"exp2", RetValSpec<DoubleType>, [ArgSpec<DoubleType>]>,
           FunctionSpec<"exp2f", RetValSpec<FloatType>, [ArgSpec<FloatType>]>,
 
+          FunctionSpec<"exp2m1f", RetValSpec<FloatType>, [ArgSpec<FloatType>]>,
+
           FunctionSpec<"expm1", RetValSpec<DoubleType>, [ArgSpec<DoubleType>]>,
           FunctionSpec<"expm1f", RetValSpec<FloatType>, [ArgSpec<FloatType>]>,
 
diff --git a/libc/src/math/CMakeLists.txt b/libc/src/math/CMakeLists.txt
@@ -88,6 +88,8 @@ add_math_entrypoint_object(expf)
 add_math_entrypoint_object(exp2)
 add_math_entrypoint_object(exp2f)
 
+add_math_entrypoint_object(exp2m1f)
+
 add_math_entrypoint_object(exp10)
 add_math_entrypoint_object(exp10f)
 
diff --git a/libc/src/math/exp2m1f.h b/libc/src/math/exp2m1f.h
@@ -0,0 +1,18 @@
+//===-- Implementation header for exp2m1f -----------------------*- 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_MATH_EXP2M1F_H
+#define LLVM_LIBC_SRC_MATH_EXP2M1F_H
+
+namespace LIBC_NAMESPACE {
+
+float exp2m1f(float x);
+
+} // namespace LIBC_NAMESPACE
+
+#endif // LLVM_LIBC_SRC_MATH_EXP2M1F_H
diff --git a/libc/src/math/generic/CMakeLists.txt b/libc/src/math/generic/CMakeLists.txt
@@ -837,6 +837,27 @@ add_entrypoint_object(
     -O3
 )
 
+add_entrypoint_object(
+  exp2m1f
+  SRCS
+    exp2m1f.cpp
+  HDRS
+    ../exp2m1f.h
+  DEPENDS
+    .explogxf
+    libc.src.errno.errno
+    libc.src.__support.common
+    libc.src.__support.FPUtil.fenv_impl
+    libc.src.__support.FPUtil.fp_bits
+    libc.src.__support.FPUtil.multiply_add
+    libc.src.__support.FPUtil.polyeval
+    libc.src.__support.FPUtil.rounding_mode
+    libc.src.__support.macros.optimization
+    libc.src.__support.macros.properties.cpu_features
+  COMPILE_OPTIONS
+    -O3
+)
+
 add_entrypoint_object(
   exp10
   SRCS
diff --git a/libc/src/math/generic/exp2m1f.cpp b/libc/src/math/generic/exp2m1f.cpp
@@ -0,0 +1,144 @@
+//===-- Implementation of exp2m1f 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "src/math/exp2m1f.h"
+#include "src/__support/FPUtil/FEnvImpl.h"
+#include "src/__support/FPUtil/FPBits.h"
+#include "src/__support/FPUtil/PolyEval.h"
+#include "src/__support/FPUtil/multiply_add.h"
+#include "src/__support/FPUtil/rounding_mode.h"
+#include "src/__support/common.h"
+#include "src/__support/macros/optimization.h"
+#include "src/__support/macros/properties/cpu_features.h"
+#include "src/errno/libc_errno.h"
+
+#include "explogxf.h"
+
+namespace LIBC_NAMESPACE {
+
+LLVM_LIBC_FUNCTION(float, exp2m1f, (float x)) {
+  using FPBits = fputil::FPBits<float>;
+  FPBits xbits(x);
+
+  uint32_t x_u = xbits.uintval();
+  uint32_t x_abs = x_u & 0x7fff'ffffU;
+
+  // When |x| >= 128, or x is nan, or |x| <= 2^-5
+  if (LIBC_UNLIKELY(x_abs >= 0x4300'0000U || x_abs <= 0x3d00'0000U)) {
+    // |x| <= 2^-5
+    if (x_abs <= 0x3d00'0000) {
+      // TODO: Handle exceptional values.
+
+      // Minimax polynomial generated by Sollya with:
+      // > display = hexadecimal;
+      // > fpminimax((2^x - 1)/x, 5, [|D...|], [-2^-5, 2^-5]);
+      constexpr double COEFFS[] = {
+          0x1.62e42fefa39f3p-1, 0x1.ebfbdff82c57bp-3,  0x1.c6b08d6f2d7aap-5,
+          0x1.3b2ab6fc92f5dp-7, 0x1.5d897cfe27125p-10, 0x1.43090e61e6af1p-13};
+      double xd = x;
+      double xsq = xd * xd;
+      double c0 = fputil::multiply_add(xd, COEFFS[1], COEFFS[0]);
+      double c1 = fputil::multiply_add(xd, COEFFS[3], COEFFS[2]);
+      double c2 = fputil::multiply_add(xd, COEFFS[5], COEFFS[4]);
+      double p = fputil::polyeval(xsq, c0, c1, c2);
+      return static_cast<float>(p * xd);
+    }
+
+    // x >= 128, or x is nan
+    if (xbits.is_pos()) {
+      if (xbits.is_finite()) {
+        int rounding = fputil::quick_get_round();
+        if (rounding == FE_DOWNWARD || rounding == FE_TOWARDZERO)
+          return FPBits::max_normal().get_val();
+
+        fputil::set_errno_if_required(ERANGE);
+        fputil::raise_except_if_required(FE_OVERFLOW);
+      }
+
+      // x >= 128 and 2^x - 1 rounds to +inf, or x is +inf or nan
+      return x + FPBits::inf().get_val();
+    }
+
+    // x <= -25
+    if (x <= -25.0f) {
+      // 2^(-inf) - 1 = -1
+      if (xbits.is_inf())
+        return -1.0f;
+      // 2^nan - 1 = nan
+      if (xbits.is_nan())
+        return x;
+
+      int rounding = fputil::quick_get_round();
+      if (rounding == FE_UPWARD || rounding == FE_TOWARDZERO)
+        return -0x1.ffff'fep-1f; // -1.0f + 0x1.0p-24f
+
+      fputil::set_errno_if_required(ERANGE);
+      fputil::raise_except_if_required(FE_UNDERFLOW);
+      return -1.0f;
+    }
+  }
+
+  // For -25 < x < 128, to compute 2^x, we perform the following range
+  // reduction: find hi, mid, lo such that:
+  //   x = hi + mid + lo, in which:
+  //     hi is an integer,
+  //     0 <= mid * 2^5 < 32 is an integer,
+  //     -2^(-6) <= lo <= 2^(-6).
+  // In particular,
+  //   hi + mid = round(x * 2^5) * 2^(-5).
+  // Then,
+  //   2^x = 2^(hi + mid + lo) = 2^hi * 2^mid * 2^lo.
+  // 2^mid is stored in the lookup table of 32 elements.
+  // 2^lo is computed using a degree-4 minimax polynomial generated by Sollya.
+  // We perform 2^hi * 2^mid by simply add hi to the exponent field of 2^mid.
+
+  // kf = (hi + mid) * 2^5 = round(x * 2^5)
+  float kf;
+  int k;
+#ifdef LIBC_TARGET_CPU_HAS_NEAREST_INT
+  kf = fputil::nearest_integer(x * 32.0f);
+  k = static_cast<int>(kf);
+#else
+  constexpr float HALF[2] = {0.5f, -0.5f};
+  k = static_cast<int>(fputil::multiply_add(x, 32.0f, HALF[x < 0.0f]));
+  kf = static_cast<float>(k);
+#endif // LIBC_TARGET_CPU_HAS_NEAREST_INT
+
+  // dx = lo = x - (hi + mid) = x - kf * 2^(-5)
+  double dx = fputil::multiply_add(-0x1.0p-5f, kf, x);
+
+  // hi = floor(kf * 2^(-4))
+  // exp_hi = shift hi to the exponent field of double precision.
+  int64_t exp_hi =
+      static_cast<int64_t>(static_cast<uint64_t>(k >> ExpBase::MID_BITS)
+                           << fputil::FPBits<double>::FRACTION_LEN);
+  // mh = 2^hi * 2^mid
+  // mh_bits = bit field of mh
+  int64_t mh_bits = ExpBase::EXP_2_MID[k & ExpBase::MID_MASK] + exp_hi;
+  double mh = fputil::FPBits<double>(static_cast<uint64_t>(mh_bits)).get_val();
+
+  // Degree-4 polynomial approximating (2^x - 1)/x generated by Sollya with:
+  // > display = hexadecimal;
+  // > fpminimax((2^x - 1)/x, 4, [|D...|], [-2^-6, 2^-6]);
+  constexpr double COEFFS[5] = {0x1.62e42fefa39efp-1, 0x1.ebfbdff8131c4p-3,
+                                0x1.c6b08d7061695p-5, 0x1.3b2b1bee74b2ap-7,
+                                0x1.5d88091198529p-10};
+  double dx_sq = dx * dx;
+  double c1 = fputil::multiply_add(dx, COEFFS[0], 1.0);
+  double c2 = fputil::multiply_add(dx, COEFFS[2], COEFFS[1]);
+  double c3 = fputil::multiply_add(dx, COEFFS[4], COEFFS[3]);
+  double p = fputil::multiply_add(dx_sq, c3, c2);
+  // 2^x = 2^(hi + mid + lo)
+  //     = 2^(hi + mid) * 2^lo
+  //     ~ mh * (1 + lo * P(lo))
+  //     = mh + (mh*lo) * P(lo)
+  double exp2_lo = fputil::multiply_add(p, dx_sq, c1);
+  return static_cast<float>(fputil::multiply_add(exp2_lo, mh, -1.0));
+}
+
+} // namespace LIBC_NAMESPACE
diff --git a/libc/test/src/math/exhaustive/CMakeLists.txt b/libc/test/src/math/exhaustive/CMakeLists.txt
@@ -142,6 +142,21 @@ add_fp_unittest(
     -lpthread
 )
 
+add_fp_unittest(
+  exp2m1f_test
+  NO_RUN_POSTBUILD
+  NEED_MPFR
+  SUITE
+    libc_math_exhaustive_tests
+  SRCS
+    exp2m1f_test.cpp
+  DEPENDS
+    .exhaustive_test
+    libc.src.math.exp2m1f
+  LINK_LIBRARIES
+    -lpthread
+)
+
 add_fp_unittest(
   exp10f_test
   NO_RUN_POSTBUILD
diff --git a/libc/test/src/math/exhaustive/exp2m1f_test.cpp b/libc/test/src/math/exhaustive/exp2m1f_test.cpp
@@ -0,0 +1,33 @@
+//===-- Exhaustive test for exp2m1f ---------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "exhaustive_test.h"
+#include "src/math/exp2m1f.h"
+#include "utils/MPFRWrapper/MPFRUtils.h"
+
+namespace mpfr = LIBC_NAMESPACE::testing::mpfr;
+
+using LlvmLibcExp2m1fExhaustiveTest =
+    LlvmLibcUnaryOpExhaustiveMathTest<float, mpfr::Operation::Exp2m1,
+                                      LIBC_NAMESPACE::exp2m1f>;
+
+// Range: [0, Inf];
+static constexpr uint32_t POS_START = 0x0000'0000U;
+static constexpr uint32_t POS_STOP = 0x7f80'0000U;
+
+TEST_F(LlvmLibcExp2m1fExhaustiveTest, PostiveRange) {
+  test_full_range_all_roundings(POS_START, POS_STOP);
+}
+
+// Range: [-Inf, 0];
+static constexpr uint32_t NEG_START = 0x8000'0000U;
+static constexpr uint32_t NEG_STOP = 0xff80'0000U;
+
+TEST_F(LlvmLibcExp2m1fExhaustiveTest, NegativeRange) {
+  test_full_range_all_roundings(NEG_START, NEG_STOP);
+}
diff --git a/libc/test/src/math/smoke/CMakeLists.txt b/libc/test/src/math/smoke/CMakeLists.txt
@@ -773,6 +773,18 @@ add_fp_unittest(
    libc.src.__support.FPUtil.fp_bits
 )
 
+add_fp_unittest(
+  exp2m1f_test
+  SUITE
+    libc-math-smoke-tests
+  SRCS
+    exp2m1f_test.cpp
+  DEPENDS
+    libc.src.__support.FPUtil.fp_bits
+    libc.src.errno.errno
+    libc.src.math.exp2m1f
+)
+
 add_fp_unittest(
   exp10f_test
   SUITE
diff --git a/libc/test/src/math/smoke/exp2m1f_test.cpp b/libc/test/src/math/smoke/exp2m1f_test.cpp
@@ -0,0 +1,57 @@
+//===-- Unittests for exp2m1f ---------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "src/__support/FPUtil/FPBits.h"
+#include "src/errno/libc_errno.h"
+#include "src/math/exp2m1f.h"
+#include "test/UnitTest/FPMatcher.h"
+#include "test/UnitTest/Test.h"
+
+using LlvmLibcExp2m1fTest = LIBC_NAMESPACE::testing::FPTest<float>;
+
+TEST_F(LlvmLibcExp2m1fTest, SpecialNumbers) {
+  LIBC_NAMESPACE::libc_errno = 0;
+
+  EXPECT_FP_EQ_ALL_ROUNDING(aNaN, LIBC_NAMESPACE::exp2m1f(aNaN));
+  EXPECT_MATH_ERRNO(0);
+
+  EXPECT_FP_EQ_ALL_ROUNDING(inf, LIBC_NAMESPACE::exp2m1f(inf));
+  EXPECT_MATH_ERRNO(0);
+
+  EXPECT_FP_EQ_ALL_ROUNDING(-1.0f, LIBC_NAMESPACE::exp2m1f(neg_inf));
+  EXPECT_MATH_ERRNO(0);
+
+  EXPECT_FP_EQ_ALL_ROUNDING(0.0f, LIBC_NAMESPACE::exp2m1f(0.0f));
+  EXPECT_MATH_ERRNO(0);
+
+  EXPECT_FP_EQ_ALL_ROUNDING(-0.0f, LIBC_NAMESPACE::exp2m1f(-0.0f));
+  EXPECT_MATH_ERRNO(0);
+
+  EXPECT_FP_EQ_ALL_ROUNDING(1.0f, LIBC_NAMESPACE::exp2m1f(1.0f));
+  EXPECT_FP_EQ_ALL_ROUNDING(-0.5f, LIBC_NAMESPACE::exp2m1f(-1.0f));
+  EXPECT_FP_EQ_ALL_ROUNDING(3.0f, LIBC_NAMESPACE::exp2m1f(2.0f));
+  EXPECT_FP_EQ_ALL_ROUNDING(-0.75f, LIBC_NAMESPACE::exp2m1f(-2.0f));
+}
+
+TEST_F(LlvmLibcExp2m1fTest, Overflow) {
+  LIBC_NAMESPACE::libc_errno = 0;
+  EXPECT_FP_EQ_WITH_EXCEPTION(
+      inf, LIBC_NAMESPACE::exp2m1f(FPBits(0x7f7f'ffffU).get_val()),
+      FE_OVERFLOW);
+  EXPECT_MATH_ERRNO(ERANGE);
+
+  EXPECT_FP_EQ_WITH_EXCEPTION(
+      inf, LIBC_NAMESPACE::exp2m1f(FPBits(0x4300'0000U).get_val()),
+      FE_OVERFLOW);
+  EXPECT_MATH_ERRNO(ERANGE);
+
+  EXPECT_FP_EQ_WITH_EXCEPTION(
+      inf, LIBC_NAMESPACE::exp2m1f(FPBits(0x4300'0001U).get_val()),
+      FE_OVERFLOW);
+  EXPECT_MATH_ERRNO(ERANGE);
+}
diff --git a/libc/utils/MPFRWrapper/MPFRUtils.cpp b/libc/utils/MPFRWrapper/MPFRUtils.cpp
@@ -229,6 +229,12 @@ class MPFRNumber {
     return result;
   }
 
+  MPFRNumber exp2m1() const {
+    MPFRNumber result(*this);
+    mpfr_exp2m1(result.value, value, mpfr_rounding);
+    return result;
+  }
+
   MPFRNumber exp10() const {
     MPFRNumber result(*this);
     mpfr_exp10(result.value, value, mpfr_rounding);
@@ -570,6 +576,8 @@ unary_operation(Operation op, InputType input, unsigned int precision,
     return mpfrInput.exp();
   case Operation::Exp2:
     return mpfrInput.exp2();
+  case Operation::Exp2m1:
+    return mpfrInput.exp2m1();
   case Operation::Exp10:
     return mpfrInput.exp10();
   case Operation::Expm1:
diff --git a/libc/utils/MPFRWrapper/MPFRUtils.h b/libc/utils/MPFRWrapper/MPFRUtils.h
@@ -37,6 +37,7 @@ enum class Operation : int {
   Erf,
   Exp,
   Exp2,
+  Exp2m1,
   Exp10,
   Expm1,
   Floor,

-Original file line number
+Diff line change
   Erf,
   Exp,
   Exp2,
 +  Exp2m1,
   Exp10,
   Expm1,
   Floor,