[libc-commits] [libc] c25a5e0 - [libc][math] Refactor expf16 implementation to header-only in src/__support/math folder. (#147428)
via libc-commits
libc-commits at lists.llvm.org
Thu Jul 10 19:21:52 PDT 2025
Author: Muhammad Bassiouni
Date: 2025-07-10T22:21:48-04:00
New Revision: c25a5e08ba8242bd4eca387adb557ee4185a0c1e
URL: https://github.com/llvm/llvm-project/commit/c25a5e08ba8242bd4eca387adb557ee4185a0c1e
DIFF: https://github.com/llvm/llvm-project/commit/c25a5e08ba8242bd4eca387adb557ee4185a0c1e.diff
LOG: [libc][math] Refactor expf16 implementation to header-only in src/__support/math folder. (#147428)
Part of #147386
in preparation for:
https://discourse.llvm.org/t/rfc-make-clang-builtin-math-functions-constexpr-with-llvm-libc-to-support-c-23-constexpr-math-functions/86450
Added:
libc/shared/math/expf16.h
libc/src/__support/math/expf16.h
libc/src/__support/math/expf16_utils.h
Modified:
libc/shared/math.h
libc/src/__support/math/CMakeLists.txt
libc/src/math/generic/CMakeLists.txt
libc/src/math/generic/expf16.cpp
libc/src/math/generic/expxf16.h
utils/bazel/llvm-project-overlay/libc/BUILD.bazel
Removed:
################################################################################
diff --git a/libc/shared/math.h b/libc/shared/math.h
index 4ddc29c7ae834..9db53b69041d0 100644
--- a/libc/shared/math.h
+++ b/libc/shared/math.h
@@ -12,5 +12,6 @@
#include "libc_common.h"
#include "math/expf.h"
+#include "math/expf16.h"
#endif // LLVM_LIBC_SHARED_MATH_H
diff --git a/libc/shared/math/expf16.h b/libc/shared/math/expf16.h
new file mode 100644
index 0000000000000..a6a3e89e680d4
--- /dev/null
+++ b/libc/shared/math/expf16.h
@@ -0,0 +1,29 @@
+//===-- Shared expf16 function ----------------------------------*- 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_SHARED_MATH_EXPF16_H
+#define LLVM_LIBC_SHARED_MATH_EXPF16_H
+
+#include "include/llvm-libc-macros/float16-macros.h"
+#include "shared/libc_common.h"
+
+#ifdef LIBC_TYPES_HAS_FLOAT16
+
+#include "src/__support/math/expf16.h"
+
+namespace LIBC_NAMESPACE_DECL {
+namespace shared {
+
+using math::expf16;
+
+} // namespace shared
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LIBC_TYPES_HAS_FLOAT16
+
+#endif // LLVM_LIBC_SHARED_MATH_EXPF16_H
diff --git a/libc/src/__support/math/CMakeLists.txt b/libc/src/__support/math/CMakeLists.txt
index 66c1d19a1cab0..4c73fba6613fa 100644
--- a/libc/src/__support/math/CMakeLists.txt
+++ b/libc/src/__support/math/CMakeLists.txt
@@ -22,3 +22,36 @@ add_header_library(
libc.src.__support.macros.config
libc.src.__support.macros.optimization
)
+
+add_header_library(
+ expf16_utils
+ HDRS
+ expf16_utils.h
+ DEPENDS
+ libc.src.__support.CPP.array
+ libc.src.__support.FPUtil.nearest_integer
+ libc.src.__support.FPUtil.polyeval
+ libc.src.__support.macros.attributes
+ libc.include.llvm-libc-macros.float16_macros
+)
+
+add_header_library(
+ expf16
+ HDRS
+ expf16.h
+ DEPENDS
+ .expf16_utils
+ libc.hdr.errno_macros
+ libc.hdr.fenv_macros
+ libc.src.__support.CPP.array
+ libc.src.__support.FPUtil.cast
+ libc.src.__support.FPUtil.except_value_utils
+ libc.src.__support.FPUtil.fenv_impl
+ libc.src.__support.FPUtil.fp_bits
+ libc.src.__support.FPUtil.multiply_add
+ libc.src.__support.FPUtil.nearest_integer
+ libc.src.__support.FPUtil.polyeval
+ libc.src.__support.FPUtil.rounding_mode
+ libc.src.__support.macros.optimization
+ libc.include.llvm-libc-macros.float16_macros
+)
diff --git a/libc/src/__support/math/expf16.h b/libc/src/__support/math/expf16.h
new file mode 100644
index 0000000000000..ded28c7dba500
--- /dev/null
+++ b/libc/src/__support/math/expf16.h
@@ -0,0 +1,141 @@
+//===-- Implementation header for expf16 ------------------------*- 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_EXPF16_H
+#define LLVM_LIBC_SRC___SUPPORT_MATH_EXPF16_H
+
+#include "include/llvm-libc-macros/float16-macros.h"
+
+#ifdef LIBC_TYPES_HAS_FLOAT16
+
+#include "hdr/errno_macros.h"
+#include "hdr/fenv_macros.h"
+#include "src/__support/FPUtil/FEnvImpl.h"
+#include "src/__support/FPUtil/FPBits.h"
+#include "src/__support/FPUtil/PolyEval.h"
+#include "src/__support/FPUtil/cast.h"
+#include "src/__support/FPUtil/except_value_utils.h"
+#include "src/__support/FPUtil/rounding_mode.h"
+#include "src/__support/common.h"
+#include "src/__support/macros/config.h"
+#include "src/__support/macros/optimization.h"
+
+#include "expf16_utils.h"
+
+namespace LIBC_NAMESPACE_DECL {
+
+namespace math {
+
+static constexpr float16 expf16(float16 x) {
+#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+ constexpr fputil::ExceptValues<float16, 2> EXPF16_EXCEPTS_LO = {{
+ // (input, RZ output, RU offset, RD offset, RN offset)
+ // x = 0x1.de4p-8, expf16(x) = 0x1.01cp+0 (RZ)
+ {0x1f79U, 0x3c07U, 1U, 0U, 0U},
+ // x = 0x1.73cp-6, expf16(x) = 0x1.05cp+0 (RZ)
+ {0x25cfU, 0x3c17U, 1U, 0U, 0U},
+ }};
+
+ constexpr fputil::ExceptValues<float16, 3> EXPF16_EXCEPTS_HI = {{
+ // (input, RZ output, RU offset, RD offset, RN offset)
+ // x = 0x1.c34p+0, expf16(x) = 0x1.74cp+2 (RZ)
+ {0x3f0dU, 0x45d3U, 1U, 0U, 1U},
+ // x = -0x1.488p-5, expf16(x) = 0x1.ebcp-1 (RZ)
+ {0xa922U, 0x3bafU, 1U, 0U, 0U},
+ // x = -0x1.55p-5, expf16(x) = 0x1.ebp-1 (RZ)
+ {0xa954U, 0x3bacU, 1U, 0U, 0U},
+ }};
+#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+
+ using FPBits = fputil::FPBits<float16>;
+ FPBits x_bits(x);
+
+ uint16_t x_u = x_bits.uintval();
+ uint16_t x_abs = x_u & 0x7fffU;
+
+ // When 0 < |x| <= 2^(-5), or |x| >= 12, or x is NaN.
+ if (LIBC_UNLIKELY(x_abs <= 0x2800U || x_abs >= 0x4a00U)) {
+ // exp(NaN) = NaN
+ if (x_bits.is_nan()) {
+ if (x_bits.is_signaling_nan()) {
+ fputil::raise_except_if_required(FE_INVALID);
+ return FPBits::quiet_nan().get_val();
+ }
+
+ return x;
+ }
+
+ // When x >= 12.
+ if (x_bits.is_pos() && x_abs >= 0x4a00U) {
+ // exp(+inf) = +inf
+ if (x_bits.is_inf())
+ return FPBits::inf().get_val();
+
+ switch (fputil::quick_get_round()) {
+ case FE_TONEAREST:
+ case FE_UPWARD:
+ fputil::set_errno_if_required(ERANGE);
+ fputil::raise_except_if_required(FE_OVERFLOW);
+ return FPBits::inf().get_val();
+ default:
+ return FPBits::max_normal().get_val();
+ }
+ }
+
+ // When x <= -18.
+ if (x_u >= 0xcc80U) {
+ // exp(-inf) = +0
+ if (x_bits.is_inf())
+ return FPBits::zero().get_val();
+
+ fputil::set_errno_if_required(ERANGE);
+ fputil::raise_except_if_required(FE_UNDERFLOW | FE_INEXACT);
+
+ switch (fputil::quick_get_round()) {
+ case FE_UPWARD:
+ return FPBits::min_subnormal().get_val();
+ default:
+ return FPBits::zero().get_val();
+ }
+ }
+
+ // When 0 < |x| <= 2^(-5).
+ if (x_abs <= 0x2800U && !x_bits.is_zero()) {
+#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+ if (auto r = EXPF16_EXCEPTS_LO.lookup(x_u); LIBC_UNLIKELY(r.has_value()))
+ return r.value();
+#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+
+ float xf = x;
+ // Degree-3 minimax polynomial generated by Sollya with the following
+ // commands:
+ // > display = hexadecimal;
+ // > P = fpminimax(expm1(x)/x, 2, [|SG...|], [-2^-5, 2^-5]);
+ // > 1 + x * P;
+ return fputil::cast<float16>(
+ fputil::polyeval(xf, 0x1p+0f, 0x1p+0f, 0x1.0004p-1f, 0x1.555778p-3f));
+ }
+ }
+
+#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+ if (auto r = EXPF16_EXCEPTS_HI.lookup(x_u); LIBC_UNLIKELY(r.has_value()))
+ return r.value();
+#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+
+ // exp(x) = exp(hi + mid) * exp(lo)
+ auto [exp_hi_mid, exp_lo] = exp_range_reduction(x);
+ return fputil::cast<float16>(exp_hi_mid * exp_lo);
+}
+
+} // namespace math
+
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LIBC_TYPES_HAS_FLOAT16
+
+#endif // LLVM_LIBC_SRC___SUPPORT_MATH_EXPF16_H
diff --git a/libc/src/__support/math/expf16_utils.h b/libc/src/__support/math/expf16_utils.h
new file mode 100644
index 0000000000000..bebb72b09b886
--- /dev/null
+++ b/libc/src/__support/math/expf16_utils.h
@@ -0,0 +1,89 @@
+//===-- Common utils for expf16 functions -----------------------*- 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_EXPF16_UTILS_H
+#define LLVM_LIBC_SRC___SUPPORT_MATH_EXPF16_UTILS_H
+
+#include "include/llvm-libc-macros/float16-macros.h"
+
+#ifdef LIBC_TYPES_HAS_FLOAT16
+
+#include "src/__support/CPP/array.h"
+#include "src/__support/FPUtil/PolyEval.h"
+#include "src/__support/FPUtil/nearest_integer.h"
+#include "src/__support/macros/properties/types.h"
+
+namespace LIBC_NAMESPACE_DECL {
+
+// Generated by Sollya with the following commands:
+// > display = hexadecimal;
+// > for i from -18 to 12 do print(round(exp(i), SG, RN));
+static constexpr cpp::array<float, 31> EXP_HI = {
+ 0x1.05a628p-26f, 0x1.639e32p-25f, 0x1.e355bcp-24f, 0x1.4875cap-22f,
+ 0x1.be6c7p-21f, 0x1.2f6054p-19f, 0x1.9c54c4p-18f, 0x1.183542p-16f,
+ 0x1.7cd79cp-15f, 0x1.02cf22p-13f, 0x1.5fc21p-12f, 0x1.de16bap-11f,
+ 0x1.44e52p-9f, 0x1.b993fep-8f, 0x1.2c155cp-6f, 0x1.97db0cp-5f,
+ 0x1.152aaap-3f, 0x1.78b564p-2f, 0x1p+0f, 0x1.5bf0a8p+1f,
+ 0x1.d8e64cp+2f, 0x1.415e5cp+4f, 0x1.b4c902p+5f, 0x1.28d38ap+7f,
+ 0x1.936dc6p+8f, 0x1.122886p+10f, 0x1.749ea8p+11f, 0x1.fa7158p+12f,
+ 0x1.5829dcp+14f, 0x1.d3c448p+15f, 0x1.3de166p+17f,
+};
+
+// Generated by Sollya with the following commands:
+// > display = hexadecimal;
+// > for i from 0 to 7 do print(round(exp(i * 2^-3), SG, RN));
+static constexpr cpp::array<float, 8> EXP_MID = {
+ 0x1p+0f, 0x1.221604p+0f, 0x1.48b5e4p+0f, 0x1.747a52p+0f,
+ 0x1.a61298p+0f, 0x1.de455ep+0f, 0x1.0ef9dcp+1f, 0x1.330e58p+1f,
+};
+
+struct ExpRangeReduction {
+ float exp_hi_mid;
+ float exp_lo;
+};
+
+static constexpr ExpRangeReduction exp_range_reduction(float16 x) {
+ // For -18 < x < 12, to compute exp(x), we perform the following range
+ // reduction: find hi, mid, lo, such that:
+ // x = hi + mid + lo, in which
+ // hi is an integer,
+ // mid * 2^3 is an integer,
+ // -2^(-4) <= lo < 2^(-4).
+ // In particular,
+ // hi + mid = round(x * 2^3) * 2^(-3).
+ // Then,
+ // exp(x) = exp(hi + mid + lo) = exp(hi) * exp(mid) * exp(lo).
+ // We store exp(hi) and exp(mid) in the lookup tables EXP_HI and EXP_MID
+ // respectively. exp(lo) is computed using a degree-3 minimax polynomial
+ // generated by Sollya.
+
+ float xf = x;
+ float kf = fputil::nearest_integer(xf * 0x1.0p+3f);
+ int x_hi_mid = static_cast<int>(kf);
+ int x_hi = x_hi_mid >> 3;
+ int x_mid = x_hi_mid & 0x7;
+ // lo = x - (hi + mid) = round(x * 2^3) * (-2^(-3)) + x
+ float lo = fputil::multiply_add(kf, -0x1.0p-3f, xf);
+
+ float exp_hi = EXP_HI[x_hi + 18];
+ float exp_mid = EXP_MID[x_mid];
+ // Degree-3 minimax polynomial generated by Sollya with the following
+ // commands:
+ // > display = hexadecimal;
+ // > P = fpminimax(expm1(x)/x, 2, [|SG...|], [-2^-4, 2^-4]);
+ // > 1 + x * P;
+ float exp_lo =
+ fputil::polyeval(lo, 0x1p+0f, 0x1p+0f, 0x1.001p-1f, 0x1.555ddep-3f);
+ return {exp_hi * exp_mid, exp_lo};
+}
+
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LIBC_TYPES_HAS_FLOAT16
+
+#endif // LLVM_LIBC_SRC___SUPPORT_MATH_EXPF16_UTILS_H
diff --git a/libc/src/math/generic/CMakeLists.txt b/libc/src/math/generic/CMakeLists.txt
index fd1e6c0d648aa..6c3f28f423c7b 100644
--- a/libc/src/math/generic/CMakeLists.txt
+++ b/libc/src/math/generic/CMakeLists.txt
@@ -1332,19 +1332,8 @@ add_entrypoint_object(
HDRS
../expf16.h
DEPENDS
- .expxf16
- libc.hdr.errno_macros
- libc.hdr.fenv_macros
- libc.src.__support.CPP.array
- libc.src.__support.FPUtil.cast
- libc.src.__support.FPUtil.except_value_utils
- libc.src.__support.FPUtil.fenv_impl
- libc.src.__support.FPUtil.fp_bits
- libc.src.__support.FPUtil.multiply_add
- libc.src.__support.FPUtil.nearest_integer
- libc.src.__support.FPUtil.polyeval
- libc.src.__support.FPUtil.rounding_mode
- libc.src.__support.macros.optimization
+ libc.src.__support.math.expf16
+ libc.src.errno.errno
)
add_entrypoint_object(
@@ -5075,11 +5064,10 @@ add_header_library(
HDRS
expxf16.h
DEPENDS
- libc.src.__support.CPP.array
libc.src.__support.FPUtil.cast
libc.src.__support.FPUtil.fp_bits
libc.src.__support.FPUtil.multiply_add
libc.src.__support.FPUtil.nearest_integer
- libc.src.__support.FPUtil.polyeval
libc.src.__support.macros.attributes
+ libc.src.__support.math.expf16_utils
)
diff --git a/libc/src/math/generic/expf16.cpp b/libc/src/math/generic/expf16.cpp
index 1af9b3ec9ad6e..ad213e237f021 100644
--- a/libc/src/math/generic/expf16.cpp
+++ b/libc/src/math/generic/expf16.cpp
@@ -7,120 +7,11 @@
//===----------------------------------------------------------------------===//
#include "src/math/expf16.h"
-#include "expxf16.h"
-#include "hdr/errno_macros.h"
-#include "hdr/fenv_macros.h"
-#include "src/__support/FPUtil/FEnvImpl.h"
-#include "src/__support/FPUtil/FPBits.h"
-#include "src/__support/FPUtil/PolyEval.h"
-#include "src/__support/FPUtil/cast.h"
-#include "src/__support/FPUtil/except_value_utils.h"
-#include "src/__support/FPUtil/rounding_mode.h"
-#include "src/__support/common.h"
-#include "src/__support/macros/config.h"
-#include "src/__support/macros/optimization.h"
-namespace LIBC_NAMESPACE_DECL {
-
-#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
-static constexpr fputil::ExceptValues<float16, 2> EXPF16_EXCEPTS_LO = {{
- // (input, RZ output, RU offset, RD offset, RN offset)
- // x = 0x1.de4p-8, expf16(x) = 0x1.01cp+0 (RZ)
- {0x1f79U, 0x3c07U, 1U, 0U, 0U},
- // x = 0x1.73cp-6, expf16(x) = 0x1.05cp+0 (RZ)
- {0x25cfU, 0x3c17U, 1U, 0U, 0U},
-}};
-
-static constexpr fputil::ExceptValues<float16, 3> EXPF16_EXCEPTS_HI = {{
- // (input, RZ output, RU offset, RD offset, RN offset)
- // x = 0x1.c34p+0, expf16(x) = 0x1.74cp+2 (RZ)
- {0x3f0dU, 0x45d3U, 1U, 0U, 1U},
- // x = -0x1.488p-5, expf16(x) = 0x1.ebcp-1 (RZ)
- {0xa922U, 0x3bafU, 1U, 0U, 0U},
- // x = -0x1.55p-5, expf16(x) = 0x1.ebp-1 (RZ)
- {0xa954U, 0x3bacU, 1U, 0U, 0U},
-}};
-#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS
-
-LLVM_LIBC_FUNCTION(float16, expf16, (float16 x)) {
- using FPBits = fputil::FPBits<float16>;
- FPBits x_bits(x);
-
- uint16_t x_u = x_bits.uintval();
- uint16_t x_abs = x_u & 0x7fffU;
-
- // When 0 < |x| <= 2^(-5), or |x| >= 12, or x is NaN.
- if (LIBC_UNLIKELY(x_abs <= 0x2800U || x_abs >= 0x4a00U)) {
- // exp(NaN) = NaN
- if (x_bits.is_nan()) {
- if (x_bits.is_signaling_nan()) {
- fputil::raise_except_if_required(FE_INVALID);
- return FPBits::quiet_nan().get_val();
- }
-
- return x;
- }
-
- // When x >= 12.
- if (x_bits.is_pos() && x_abs >= 0x4a00U) {
- // exp(+inf) = +inf
- if (x_bits.is_inf())
- return FPBits::inf().get_val();
+#include "src/__support/math/expf16.h"
- switch (fputil::quick_get_round()) {
- case FE_TONEAREST:
- case FE_UPWARD:
- fputil::set_errno_if_required(ERANGE);
- fputil::raise_except_if_required(FE_OVERFLOW);
- return FPBits::inf().get_val();
- default:
- return FPBits::max_normal().get_val();
- }
- }
-
- // When x <= -18.
- if (x_u >= 0xcc80U) {
- // exp(-inf) = +0
- if (x_bits.is_inf())
- return FPBits::zero().get_val();
-
- fputil::set_errno_if_required(ERANGE);
- fputil::raise_except_if_required(FE_UNDERFLOW | FE_INEXACT);
-
- switch (fputil::quick_get_round()) {
- case FE_UPWARD:
- return FPBits::min_subnormal().get_val();
- default:
- return FPBits::zero().get_val();
- }
- }
-
- // When 0 < |x| <= 2^(-5).
- if (x_abs <= 0x2800U && !x_bits.is_zero()) {
-#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
- if (auto r = EXPF16_EXCEPTS_LO.lookup(x_u); LIBC_UNLIKELY(r.has_value()))
- return r.value();
-#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS
-
- float xf = x;
- // Degree-3 minimax polynomial generated by Sollya with the following
- // commands:
- // > display = hexadecimal;
- // > P = fpminimax(expm1(x)/x, 2, [|SG...|], [-2^-5, 2^-5]);
- // > 1 + x * P;
- return fputil::cast<float16>(
- fputil::polyeval(xf, 0x1p+0f, 0x1p+0f, 0x1.0004p-1f, 0x1.555778p-3f));
- }
- }
-
-#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
- if (auto r = EXPF16_EXCEPTS_HI.lookup(x_u); LIBC_UNLIKELY(r.has_value()))
- return r.value();
-#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+namespace LIBC_NAMESPACE_DECL {
- // exp(x) = exp(hi + mid) * exp(lo)
- auto [exp_hi_mid, exp_lo] = exp_range_reduction(x);
- return fputil::cast<float16>(exp_hi_mid * exp_lo);
-}
+LLVM_LIBC_FUNCTION(float16, expf16, (float16 x)) { return math::expf16(x); }
} // namespace LIBC_NAMESPACE_DECL
diff --git a/libc/src/math/generic/expxf16.h b/libc/src/math/generic/expxf16.h
index 67bb248307519..05ac95d586823 100644
--- a/libc/src/math/generic/expxf16.h
+++ b/libc/src/math/generic/expxf16.h
@@ -9,9 +9,7 @@
#ifndef LLVM_LIBC_SRC_MATH_GENERIC_EXPXF16_H
#define LLVM_LIBC_SRC_MATH_GENERIC_EXPXF16_H
-#include "src/__support/CPP/array.h"
#include "src/__support/FPUtil/FPBits.h"
-#include "src/__support/FPUtil/PolyEval.h"
#include "src/__support/FPUtil/cast.h"
#include "src/__support/FPUtil/multiply_add.h"
#include "src/__support/FPUtil/nearest_integer.h"
@@ -19,69 +17,9 @@
#include "src/__support/macros/config.h"
#include <stdint.h>
-namespace LIBC_NAMESPACE_DECL {
-
-// Generated by Sollya with the following commands:
-// > display = hexadecimal;
-// > for i from -18 to 12 do print(round(exp(i), SG, RN));
-static constexpr cpp::array<float, 31> EXP_HI = {
- 0x1.05a628p-26f, 0x1.639e32p-25f, 0x1.e355bcp-24f, 0x1.4875cap-22f,
- 0x1.be6c7p-21f, 0x1.2f6054p-19f, 0x1.9c54c4p-18f, 0x1.183542p-16f,
- 0x1.7cd79cp-15f, 0x1.02cf22p-13f, 0x1.5fc21p-12f, 0x1.de16bap-11f,
- 0x1.44e52p-9f, 0x1.b993fep-8f, 0x1.2c155cp-6f, 0x1.97db0cp-5f,
- 0x1.152aaap-3f, 0x1.78b564p-2f, 0x1p+0f, 0x1.5bf0a8p+1f,
- 0x1.d8e64cp+2f, 0x1.415e5cp+4f, 0x1.b4c902p+5f, 0x1.28d38ap+7f,
- 0x1.936dc6p+8f, 0x1.122886p+10f, 0x1.749ea8p+11f, 0x1.fa7158p+12f,
- 0x1.5829dcp+14f, 0x1.d3c448p+15f, 0x1.3de166p+17f,
-};
-
-// Generated by Sollya with the following commands:
-// > display = hexadecimal;
-// > for i from 0 to 7 do print(round(exp(i * 2^-3), SG, RN));
-static constexpr cpp::array<float, 8> EXP_MID = {
- 0x1p+0f, 0x1.221604p+0f, 0x1.48b5e4p+0f, 0x1.747a52p+0f,
- 0x1.a61298p+0f, 0x1.de455ep+0f, 0x1.0ef9dcp+1f, 0x1.330e58p+1f,
-};
-
-struct ExpRangeReduction {
- float exp_hi_mid;
- float exp_lo;
-};
+#include "src/__support/math/expf16_utils.h"
-LIBC_INLINE ExpRangeReduction exp_range_reduction(float16 x) {
- // For -18 < x < 12, to compute exp(x), we perform the following range
- // reduction: find hi, mid, lo, such that:
- // x = hi + mid + lo, in which
- // hi is an integer,
- // mid * 2^3 is an integer,
- // -2^(-4) <= lo < 2^(-4).
- // In particular,
- // hi + mid = round(x * 2^3) * 2^(-3).
- // Then,
- // exp(x) = exp(hi + mid + lo) = exp(hi) * exp(mid) * exp(lo).
- // We store exp(hi) and exp(mid) in the lookup tables EXP_HI and EXP_MID
- // respectively. exp(lo) is computed using a degree-3 minimax polynomial
- // generated by Sollya.
-
- float xf = x;
- float kf = fputil::nearest_integer(xf * 0x1.0p+3f);
- int x_hi_mid = static_cast<int>(kf);
- int x_hi = x_hi_mid >> 3;
- int x_mid = x_hi_mid & 0x7;
- // lo = x - (hi + mid) = round(x * 2^3) * (-2^(-3)) + x
- float lo = fputil::multiply_add(kf, -0x1.0p-3f, xf);
-
- float exp_hi = EXP_HI[x_hi + 18];
- float exp_mid = EXP_MID[x_mid];
- // Degree-3 minimax polynomial generated by Sollya with the following
- // commands:
- // > display = hexadecimal;
- // > P = fpminimax(expm1(x)/x, 2, [|SG...|], [-2^-4, 2^-4]);
- // > 1 + x * P;
- float exp_lo =
- fputil::polyeval(lo, 0x1p+0f, 0x1p+0f, 0x1.001p-1f, 0x1.555ddep-3f);
- return {exp_hi * exp_mid, exp_lo};
-}
+namespace LIBC_NAMESPACE_DECL {
// Generated by Sollya with the following commands:
// > display = hexadecimal;
diff --git a/utils/bazel/llvm-project-overlay/libc/BUILD.bazel b/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
index a80d8d430ec5c..d259f391069a4 100644
--- a/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
+++ b/utils/bazel/llvm-project-overlay/libc/BUILD.bazel
@@ -2083,11 +2083,10 @@ libc_support_library(
name = "expxf16",
hdrs = ["src/math/generic/expxf16.h"],
deps = [
- ":__support_cpp_array",
":__support_fputil_cast",
":__support_fputil_fp_bits",
":__support_fputil_nearest_integer",
- ":__support_fputil_polyeval",
+ ":__support_math_expf16_utils",
],
)
@@ -2117,6 +2116,39 @@ libc_support_library(
],
)
+libc_support_library(
+ name = "__support_math_expf16_utils",
+ hdrs = ["src/__support/math/expf16_utils.h"],
+ deps = [
+ ":__support_cpp_array",
+ ":__support_fputil_nearest_integer",
+ ":__support_fputil_polyeval",
+ ":__support_macros_attributes",
+ ":llvm_libc_macros_float16_macros"
+ ],
+)
+
+libc_support_library(
+ name = "__support_math_expf16",
+ hdrs = ["src/__support/math/expf16.h"],
+ deps = [
+ ":__support_common",
+ ":__support_libc_errno",
+ ":__support_cpp_array",
+ ":__support_fputil_cast",
+ ":__support_fputil_except_value_utils",
+ ":__support_fputil_fenv_impl",
+ ":__support_fputil_fp_bits",
+ ":__support_fputil_multiply_add",
+ ":__support_fputil_nearest_integer",
+ ":__support_fputil_polyeval",
+ ":__support_fputil_rounding_mode",
+ ":__support_macros_optimization",
+ ":__support_math_expf16_utils",
+ ":llvm_libc_macros_float16_macros"
+ ],
+)
+
############################### complex targets ################################
libc_function(
@@ -2722,6 +2754,7 @@ libc_math_function(
libc_math_function(
name = "expf16",
additional_deps = [
+ ":__support_math_expf16",
":expxf16",
],
)
More information about the libc-commits
mailing list