diff options
Diffstat (limited to 'libc/src')
200 files changed, 3167 insertions, 1663 deletions
diff --git a/libc/src/__support/CMakeLists.txt b/libc/src/__support/CMakeLists.txt index 37a27cc..2196d9e 100644 --- a/libc/src/__support/CMakeLists.txt +++ b/libc/src/__support/CMakeLists.txt @@ -15,6 +15,7 @@ add_header_library( HDRS block.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.CPP.algorithm libc.src.__support.CPP.limits libc.src.__support.CPP.new @@ -86,6 +87,7 @@ add_header_library( blockstore.h DEPENDS .libc_assert + libc.hdr.stdint_proxy libc.src.__support.CPP.new ) @@ -97,6 +99,8 @@ add_header_library( macros/properties/architectures.h macros/attributes.h macros/config.h + DEPENDS + libc.hdr.stdint_proxy ) add_header_library( @@ -199,6 +203,7 @@ add_header_library( integer_to_string.h DEPENDS .big_int + libc.hdr.stdint_proxy libc.src.__support.common libc.src.__support.CPP.algorithm libc.src.__support.CPP.limits @@ -215,6 +220,7 @@ add_header_library( ryu_long_double_constants.h DEPENDS .libc_assert + libc.hdr.stdint_proxy libc.src.__support.CPP.type_traits libc.src.__support.FPUtil.fp_bits libc.src.__support.common @@ -226,6 +232,7 @@ add_header_library( high_precision_decimal.h DEPENDS .str_to_integer + libc.hdr.stdint_proxy ) add_header_library( @@ -240,6 +247,7 @@ add_header_library( .str_to_num_result .uint128 libc.hdr.errno_macros + libc.hdr.stdint_proxy libc.src.__support.common libc.src.__support.CPP.bit libc.src.__support.CPP.limits @@ -257,6 +265,7 @@ add_header_library( integer_literals.h DEPENDS .uint128 + libc.hdr.stdint_proxy libc.src.__support.CPP.limits ) @@ -290,6 +299,7 @@ add_header_library( HDRS arg_list.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.common ) @@ -329,6 +339,7 @@ add_header_library( DEPENDS .math_extras .number_pair + libc.hdr.stdint_proxy libc.src.__support.CPP.array libc.src.__support.CPP.bit libc.src.__support.CPP.type_traits @@ -361,6 +372,7 @@ add_header_library( hash.h DEPENDS .uint128 + libc.hdr.stdint_proxy libc.src.__support.CPP.bit libc.src.__support.CPP.limits libc.src.__support.macros.attributes diff --git a/libc/src/__support/CPP/CMakeLists.txt b/libc/src/__support/CPP/CMakeLists.txt index d2ba00a..8b65a8839 100644 --- a/libc/src/__support/CPP/CMakeLists.txt +++ b/libc/src/__support/CPP/CMakeLists.txt @@ -17,6 +17,7 @@ add_header_library( DEPENDS .limits .type_traits + libc.hdr.stdint_proxy libc.src.__support.macros.attributes libc.src.__support.macros.sanitizer ) @@ -39,6 +40,8 @@ add_header_library( functional HDRS functional.h + DEPENDS + libc.hdr.stdint_proxy ) add_header_library( diff --git a/libc/src/__support/CPP/bit.h b/libc/src/__support/CPP/bit.h index e491f3e..df1b177 100644 --- a/libc/src/__support/CPP/bit.h +++ b/libc/src/__support/CPP/bit.h @@ -11,14 +11,13 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_CPP_BIT_H #define LLVM_LIBC_SRC___SUPPORT_CPP_BIT_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/limits.h" // numeric_limits #include "src/__support/CPP/type_traits.h" #include "src/__support/macros/attributes.h" #include "src/__support/macros/config.h" #include "src/__support/macros/sanitizer.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace cpp { diff --git a/libc/src/__support/CPP/functional.h b/libc/src/__support/CPP/functional.h index 50cfa256..5c43d22 100644 --- a/libc/src/__support/CPP/functional.h +++ b/libc/src/__support/CPP/functional.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_CPP_FUNCTIONAL_H #define LLVM_LIBC_SRC___SUPPORT_CPP_FUNCTIONAL_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/type_traits/enable_if.h" #include "src/__support/CPP/type_traits/is_convertible.h" #include "src/__support/CPP/type_traits/is_same.h" @@ -19,8 +20,6 @@ #include "src/__support/macros/attributes.h" #include "src/__support/macros/config.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace cpp { diff --git a/libc/src/__support/FPUtil/CMakeLists.txt b/libc/src/__support/FPUtil/CMakeLists.txt index cc941f2..6e447fc 100644 --- a/libc/src/__support/FPUtil/CMakeLists.txt +++ b/libc/src/__support/FPUtil/CMakeLists.txt @@ -6,6 +6,8 @@ add_header_library( libc.hdr.types.fenv_t libc.hdr.fenv_macros libc.hdr.math_macros + libc.hdr.stdint_proxy + libc.src.__support.CPP.type_traits libc.src.__support.macros.attributes libc.src.errno.errno ) @@ -16,6 +18,7 @@ add_header_library( rounding_mode.h DEPENDS libc.hdr.fenv_macros + libc.src.__support.CPP.type_traits libc.src.__support.macros.attributes libc.src.__support.macros.properties.architectures libc.src.__support.macros.sanitizer @@ -27,6 +30,7 @@ add_header_library( HDRS FPBits.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.common libc.src.__support.CPP.bit libc.src.__support.CPP.type_traits @@ -71,6 +75,7 @@ add_header_library( NormalFloat.h DEPENDS .fp_bits + libc.hdr.stdint_proxy libc.src.__support.CPP.type_traits libc.src.__support.common ) @@ -210,6 +215,17 @@ add_header_library( ) add_header_library( + comparison_operations + HDRS + comparison_operations.h + DEPENDS + .fenv_impl + .fp_bits + libc.src.__support.CPP.type_traits + libc.src.__support.macros.config +) + +add_header_library( hypot HDRS Hypot.h @@ -236,6 +252,7 @@ add_header_library( .nearest_integer_operations .normal_float libc.hdr.math_macros + libc.hdr.stdint_proxy libc.src.errno.errno libc.src.__support.common libc.src.__support.CPP.bit @@ -263,7 +280,9 @@ add_header_library( bfloat16.h DEPENDS .cast + .comparison_operations .dyadic_float + libc.hdr.stdint_proxy libc.src.__support.CPP.bit libc.src.__support.CPP.type_traits libc.src.__support.macros.config diff --git a/libc/src/__support/FPUtil/FEnvImpl.h b/libc/src/__support/FPUtil/FEnvImpl.h index 7691088..7bd5643 100644 --- a/libc/src/__support/FPUtil/FEnvImpl.h +++ b/libc/src/__support/FPUtil/FEnvImpl.h @@ -12,6 +12,7 @@ #include "hdr/fenv_macros.h" #include "hdr/math_macros.h" #include "hdr/types/fenv_t.h" +#include "src/__support/CPP/type_traits.h" #include "src/__support/libc_errno.h" #include "src/__support/macros/attributes.h" // LIBC_INLINE #include "src/__support/macros/config.h" @@ -72,40 +73,58 @@ LIBC_INLINE int set_env(const fenv_t *) { return 0; } namespace LIBC_NAMESPACE_DECL { namespace fputil { -LIBC_INLINE static int clear_except_if_required([[maybe_unused]] int excepts) { +LIBC_INLINE static constexpr int +clear_except_if_required([[maybe_unused]] int excepts) { + if (cpp::is_constant_evaluated()) { + return 0; + } else { #ifndef LIBC_MATH_HAS_NO_EXCEPT - if (math_errhandling & MATH_ERREXCEPT) - return clear_except(excepts); + if (math_errhandling & MATH_ERREXCEPT) + return clear_except(excepts); #endif // LIBC_MATH_HAS_NO_EXCEPT - return 0; + return 0; + } } -LIBC_INLINE static int set_except_if_required([[maybe_unused]] int excepts) { +LIBC_INLINE static constexpr int +set_except_if_required([[maybe_unused]] int excepts) { + if (cpp::is_constant_evaluated()) { + return 0; + } else { #ifndef LIBC_MATH_HAS_NO_EXCEPT - if (math_errhandling & MATH_ERREXCEPT) - return set_except(excepts); + if (math_errhandling & MATH_ERREXCEPT) + return set_except(excepts); #endif // LIBC_MATH_HAS_NO_EXCEPT - return 0; + return 0; + } } -LIBC_INLINE static int raise_except_if_required([[maybe_unused]] int excepts) { +LIBC_INLINE static constexpr int +raise_except_if_required([[maybe_unused]] int excepts) { + if (cpp::is_constant_evaluated()) { + return 0; + } else { #ifndef LIBC_MATH_HAS_NO_EXCEPT - if (math_errhandling & MATH_ERREXCEPT) + if (math_errhandling & MATH_ERREXCEPT) #ifdef LIBC_TARGET_ARCH_IS_X86_64 - return raise_except</*SKIP_X87_FPU*/ true>(excepts); + return raise_except</*SKIP_X87_FPU*/ true>(excepts); #else // !LIBC_TARGET_ARCH_IS_X86 - return raise_except(excepts); + return raise_except(excepts); #endif // LIBC_TARGET_ARCH_IS_X86 #endif // LIBC_MATH_HAS_NO_EXCEPT - return 0; + return 0; + } } -LIBC_INLINE static void set_errno_if_required([[maybe_unused]] int err) { +LIBC_INLINE static constexpr void +set_errno_if_required([[maybe_unused]] int err) { + if (!cpp::is_constant_evaluated()) { #ifndef LIBC_MATH_HAS_NO_ERRNO - if (math_errhandling & MATH_ERRNO) - libc_errno = err; + if (math_errhandling & MATH_ERRNO) + libc_errno = err; #endif // LIBC_MATH_HAS_NO_ERRNO + } } } // namespace fputil diff --git a/libc/src/__support/FPUtil/FPBits.h b/libc/src/__support/FPUtil/FPBits.h index 9e21136..2f695c1 100644 --- a/libc/src/__support/FPUtil/FPBits.h +++ b/libc/src/__support/FPUtil/FPBits.h @@ -15,6 +15,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_FPUTIL_FPBITS_H #define LLVM_LIBC_SRC___SUPPORT_FPUTIL_FPBITS_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/bit.h" #include "src/__support/CPP/type_traits.h" #include "src/__support/common.h" @@ -26,8 +27,6 @@ #include "src/__support/sign.h" // Sign #include "src/__support/uint128.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace fputil { diff --git a/libc/src/__support/FPUtil/NormalFloat.h b/libc/src/__support/FPUtil/NormalFloat.h index a2f285f..b30e36f 100644 --- a/libc/src/__support/FPUtil/NormalFloat.h +++ b/libc/src/__support/FPUtil/NormalFloat.h @@ -11,12 +11,11 @@ #include "FPBits.h" +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/type_traits.h" #include "src/__support/common.h" #include "src/__support/macros/config.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace fputil { diff --git a/libc/src/__support/FPUtil/aarch64/FEnvImpl.h b/libc/src/__support/FPUtil/aarch64/FEnvImpl.h index 914155a..18182ed 100644 --- a/libc/src/__support/FPUtil/aarch64/FEnvImpl.h +++ b/libc/src/__support/FPUtil/aarch64/FEnvImpl.h @@ -18,9 +18,9 @@ #endif #include <arm_acle.h> -#include <stdint.h> #include "hdr/fenv_macros.h" +#include "hdr/stdint_proxy.h" #include "hdr/types/fenv_t.h" #include "src/__support/FPUtil/FPBits.h" diff --git a/libc/src/__support/FPUtil/aarch64/fenv_darwin_impl.h b/libc/src/__support/FPUtil/aarch64/fenv_darwin_impl.h index dcce76b..a2066d1 100644 --- a/libc/src/__support/FPUtil/aarch64/fenv_darwin_impl.h +++ b/libc/src/__support/FPUtil/aarch64/fenv_darwin_impl.h @@ -18,9 +18,9 @@ #endif #include <arm_acle.h> -#include <stdint.h> #include "hdr/fenv_macros.h" +#include "hdr/stdint_proxy.h" #include "hdr/types/fenv_t.h" #include "src/__support/FPUtil/FPBits.h" diff --git a/libc/src/__support/FPUtil/arm/FEnvImpl.h b/libc/src/__support/FPUtil/arm/FEnvImpl.h index aaf37c0..64ab4a9 100644 --- a/libc/src/__support/FPUtil/arm/FEnvImpl.h +++ b/libc/src/__support/FPUtil/arm/FEnvImpl.h @@ -10,11 +10,11 @@ #define LLVM_LIBC_SRC___SUPPORT_FPUTIL_ARM_FENVIMPL_H #include "hdr/fenv_macros.h" +#include "hdr/stdint_proxy.h" #include "hdr/types/fenv_t.h" #include "src/__support/FPUtil/FPBits.h" #include "src/__support/macros/attributes.h" // For LIBC_INLINE #include "src/__support/macros/config.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { namespace fputil { diff --git a/libc/src/__support/FPUtil/bfloat16.h b/libc/src/__support/FPUtil/bfloat16.h index bc0b8b2..fa45d73 100644 --- a/libc/src/__support/FPUtil/bfloat16.h +++ b/libc/src/__support/FPUtil/bfloat16.h @@ -9,15 +9,15 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_FPUTIL_BFLOAT16_H #define LLVM_LIBC_SRC___SUPPORT_FPUTIL_BFLOAT16_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/bit.h" #include "src/__support/CPP/type_traits.h" #include "src/__support/FPUtil/cast.h" +#include "src/__support/FPUtil/comparison_operations.h" #include "src/__support/FPUtil/dyadic_float.h" #include "src/__support/macros/config.h" #include "src/__support/macros/properties/types.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace fputil { @@ -57,6 +57,30 @@ struct BFloat16 { uint32_t x_bits = static_cast<uint32_t>(bits) << 16U; return cpp::bit_cast<float>(x_bits); } + + LIBC_INLINE bool operator==(BFloat16 other) const { + return fputil::equals(*this, other); + } + + LIBC_INLINE bool operator!=(BFloat16 other) const { + return !fputil::equals(*this, other); + } + + LIBC_INLINE bool operator<(BFloat16 other) const { + return fputil::less_than(*this, other); + } + + LIBC_INLINE bool operator<=(BFloat16 other) const { + return fputil::less_than_or_equals(*this, other); + } + + LIBC_INLINE bool operator>(BFloat16 other) const { + return fputil::greater_than(*this, other); + } + + LIBC_INLINE bool operator>=(BFloat16 other) const { + return fputil::greater_than_or_equals(*this, other); + } }; // struct BFloat16 } // namespace fputil diff --git a/libc/src/__support/FPUtil/comparison_operations.h b/libc/src/__support/FPUtil/comparison_operations.h new file mode 100644 index 0000000..ff62ce0 --- /dev/null +++ b/libc/src/__support/FPUtil/comparison_operations.h @@ -0,0 +1,114 @@ +//===-- Comparison operations on floating point numbers ---------*- 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_FPUTIL_COMPARISONOPERATIONS_H +#define LLVM_LIBC_SRC___SUPPORT_FPUTIL_COMPARISONOPERATIONS_H + +#include "FEnvImpl.h" +#include "FPBits.h" +#include "src/__support/CPP/type_traits.h" +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { +namespace fputil { + +// All predicates are hereby implemented as per IEEE Std 754-2019 +// Implements compareQuietEqual predicate +// Rules for comparison within the same floating point type +// 1. +0 = −0 +// 2. (i) +inf = +inf +// (ii) -inf = -inf +// (iii) -inf != +inf +// 3. Any comparison with NaN returns false +template <typename T> +LIBC_INLINE cpp::enable_if_t<cpp::is_floating_point_v<T>, bool> equals(T x, + T y) { + using FPBits = FPBits<T>; + FPBits x_bits(x); + FPBits y_bits(y); + + if (x_bits.is_signaling_nan() || y_bits.is_signaling_nan()) + fputil::raise_except_if_required(FE_INVALID); + + // NaN == x returns false for every x + if (x_bits.is_nan() || y_bits.is_nan()) + return false; + + // +/- 0 == +/- 0 + if (x_bits.is_zero() && y_bits.is_zero()) + return true; + + return x_bits.uintval() == y_bits.uintval(); +} + +// Implements compareSignalingLess predicate +// Section 5.11 Rules: +// 1. -inf < x (x != -inf) +// 2. x < +inf (x != +inf) +// 3. Any comparison with NaN return false +template <typename T> +LIBC_INLINE cpp::enable_if_t<cpp::is_floating_point_v<T>, bool> less_than(T x, + T y) { + using FPBits = FPBits<T>; + FPBits x_bits(x); + FPBits y_bits(y); + + // Any comparison with NaN returns false + if (x_bits.is_nan() || y_bits.is_nan()) { + fputil::raise_except_if_required(FE_INVALID); + return false; + } + + if (x_bits.is_zero() && y_bits.is_zero()) + return false; + + if (x_bits.is_neg() && y_bits.is_pos()) + return true; + + if (x_bits.is_pos() && y_bits.is_neg()) + return false; + + // since floating-point numbers are stored in the format: s | e | m + // we can directly compare the uintval's + + // both negative + if (x_bits.is_neg()) + return x_bits.uintval() > y_bits.uintval(); + + // both positive + return x_bits.uintval() < y_bits.uintval(); +} + +// Implements compareSignalingGreater predicate +// x < y => y > x +template <typename T> +LIBC_INLINE cpp::enable_if_t<cpp::is_floating_point_v<T>, bool> +greater_than(T x, T y) { + return less_than(y, x); +} + +// Implements compareSignalingLessEqual predicate +// x <= y => (x < y) || (x == y) +template <typename T> +LIBC_INLINE cpp::enable_if_t<cpp::is_floating_point_v<T>, bool> +less_than_or_equals(T x, T y) { + return less_than(x, y) || equals(x, y); +} + +// Implements compareSignalingGreaterEqual predicate +// x >= y => (x > y) || (x == y) => (y < x) || (x == y) +template <typename T> +LIBC_INLINE cpp::enable_if_t<cpp::is_floating_point_v<T>, bool> +greater_than_or_equals(T x, T y) { + return less_than(y, x) || equals(x, y); +} + +} // namespace fputil +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_FPUTIL_COMPARISONOPERATIONS_H diff --git a/libc/src/__support/FPUtil/generic/add_sub.h b/libc/src/__support/FPUtil/generic/add_sub.h index fda7029..7205d8d 100644 --- a/libc/src/__support/FPUtil/generic/add_sub.h +++ b/libc/src/__support/FPUtil/generic/add_sub.h @@ -153,8 +153,10 @@ add_or_sub(InType x, InType y) { result_mant <<= GUARD_BITS_LEN; } else { - InStorageType max_mant = max_bits.get_explicit_mantissa() << GUARD_BITS_LEN; - InStorageType min_mant = min_bits.get_explicit_mantissa() << GUARD_BITS_LEN; + InStorageType max_mant = static_cast<InStorageType>( + max_bits.get_explicit_mantissa() << GUARD_BITS_LEN); + InStorageType min_mant = static_cast<InStorageType>( + min_bits.get_explicit_mantissa() << GUARD_BITS_LEN); int alignment = (max_bits.get_biased_exponent() - max_bits.is_normal()) - (min_bits.get_biased_exponent() - min_bits.is_normal()); @@ -172,7 +174,8 @@ add_or_sub(InType x, InType y) { (static_cast<InStorageType>( min_mant << (InFPBits::STORAGE_LEN - alignment))) != 0; - InStorageType min_mant_sticky(static_cast<int>(aligned_min_mant_sticky)); + InStorageType min_mant_sticky = + static_cast<InStorageType>(static_cast<int>(aligned_min_mant_sticky)); if (is_effectively_add) result_mant = max_mant + (aligned_min_mant | min_mant_sticky); diff --git a/libc/src/__support/FPUtil/riscv/FEnvImpl.h b/libc/src/__support/FPUtil/riscv/FEnvImpl.h index 2f525eb..cb2d2d5 100644 --- a/libc/src/__support/FPUtil/riscv/FEnvImpl.h +++ b/libc/src/__support/FPUtil/riscv/FEnvImpl.h @@ -10,13 +10,12 @@ #define LLVM_LIBC_SRC___SUPPORT_FPUTIL_RISCV_FENVIMPL_H #include "hdr/fenv_macros.h" +#include "hdr/stdint_proxy.h" #include "hdr/types/fenv_t.h" #include "src/__support/FPUtil/FPBits.h" #include "src/__support/macros/attributes.h" // For LIBC_INLINE_ASM #include "src/__support/macros/config.h" // For LIBC_INLINE -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace fputil { diff --git a/libc/src/__support/FPUtil/rounding_mode.h b/libc/src/__support/FPUtil/rounding_mode.h index bc66d09..4ee0a0b 100644 --- a/libc/src/__support/FPUtil/rounding_mode.h +++ b/libc/src/__support/FPUtil/rounding_mode.h @@ -10,6 +10,7 @@ #define LLVM_LIBC_SRC___SUPPORT_FPUTIL_ROUNDING_MODE_H #include "hdr/fenv_macros.h" +#include "src/__support/CPP/type_traits.h" // is_constant_evaluated #include "src/__support/macros/attributes.h" // LIBC_INLINE #include "src/__support/macros/config.h" @@ -20,18 +21,26 @@ namespace fputil { // Using the following observation: // 1.0f + 2^-25 = 1.0f for FE_TONEAREST, FE_DOWNWARD, FE_TOWARDZERO // = 0x1.000002f for FE_UPWARD. -LIBC_INLINE bool fenv_is_round_up() { - volatile float x = 0x1.0p-25f; - return (1.0f + x != 1.0f); +LIBC_INLINE static constexpr bool fenv_is_round_up() { + if (cpp::is_constant_evaluated()) { + return false; + } else { + volatile float x = 0x1.0p-25f; + return (1.0f + x != 1.0f); + } } // Quick free-standing test whether fegetround() == FE_DOWNWARD. // Using the following observation: // -1.0f - 2^-25 = -1.0f for FE_TONEAREST, FE_UPWARD, FE_TOWARDZERO // = -0x1.000002f for FE_DOWNWARD. -LIBC_INLINE bool fenv_is_round_down() { - volatile float x = 0x1.0p-25f; - return (-1.0f - x != -1.0f); +LIBC_INLINE static constexpr bool fenv_is_round_down() { + if (cpp::is_constant_evaluated()) { + return false; + } else { + volatile float x = 0x1.0p-25f; + return (-1.0f - x != -1.0f); + } } // Quick free-standing test whether fegetround() == FE_TONEAREST. @@ -40,10 +49,14 @@ LIBC_INLINE bool fenv_is_round_down() { // = 0x1.100002p0f for FE_UPWARD, // 1.5f - 2^-24 = 1.5f for FE_TONEAREST, FE_UPWARD // = 0x1.0ffffep-1f for FE_DOWNWARD, FE_TOWARDZERO -LIBC_INLINE bool fenv_is_round_to_nearest() { - static volatile float x = 0x1.0p-24f; - float y = x; - return (1.5f + y == 1.5f - y); +LIBC_INLINE static constexpr bool fenv_is_round_to_nearest() { + if (cpp::is_constant_evaluated()) { + return true; + } else { + volatile float x = 0x1.0p-24f; + float y = 1.5f + x; + return (y == 1.5f - x); + } } // Quick free-standing test whether fegetround() == FE_TOWARDZERO. @@ -56,23 +69,31 @@ LIBC_INLINE bool fenv_is_round_to_nearest() { // (0x1.000002p0f + 2^-24) + (-1.0f - 2^-24) = 2^-23 for FE_TOWARDZERO // = 2^-22 for FE_TONEAREST, FE_UPWARD // = 0 for FE_DOWNWARD -LIBC_INLINE bool fenv_is_round_to_zero() { - static volatile float x = 0x1.0p-24f; - float y = x; - return ((0x1.000002p0f + y) + (-1.0f - y) == 0x1.0p-23f); +LIBC_INLINE static constexpr bool fenv_is_round_to_zero() { + if (cpp::is_constant_evaluated()) { + return false; + } else { + volatile float x = 0x1.0p-24f; + volatile float y = 0x1.000002p0f + x; + return (y + (-1.0f - x) == 0x1.0p-23f); + } } // Quick free standing get rounding mode based on the above observations. -LIBC_INLINE int quick_get_round() { - static volatile float x = 0x1.0p-24f; - float y = x; - float z = (0x1.000002p0f + y) + (-1.0f - y); +LIBC_INLINE static constexpr int quick_get_round() { + if (cpp::is_constant_evaluated()) { + return FE_TONEAREST; + } else { + volatile float x = 0x1.0p-24f; + volatile float y = 0x1.000002p0f + x; + float z = y + (-1.0f - x); - if (z == 0.0f) - return FE_DOWNWARD; - if (z == 0x1.0p-23f) - return FE_TOWARDZERO; - return (2.0f + y == 2.0f) ? FE_TONEAREST : FE_UPWARD; + if (z == 0.0f) + return FE_DOWNWARD; + if (z == 0x1.0p-23f) + return FE_TOWARDZERO; + return (2.0f + x == 2.0f) ? FE_TONEAREST : FE_UPWARD; + } } } // namespace fputil diff --git a/libc/src/__support/FPUtil/x86_64/FEnvImpl.h b/libc/src/__support/FPUtil/x86_64/FEnvImpl.h index 560727c..5da5097 100644 --- a/libc/src/__support/FPUtil/x86_64/FEnvImpl.h +++ b/libc/src/__support/FPUtil/x86_64/FEnvImpl.h @@ -17,8 +17,7 @@ #error "Invalid include" #endif -#include <stdint.h> - +#include "hdr/stdint_proxy.h" #include "hdr/types/fenv_t.h" #include "src/__support/macros/sanitizer.h" diff --git a/libc/src/__support/FPUtil/x86_64/NextAfterLongDouble.h b/libc/src/__support/FPUtil/x86_64/NextAfterLongDouble.h index 2e6b297..74a991d 100644 --- a/libc/src/__support/FPUtil/x86_64/NextAfterLongDouble.h +++ b/libc/src/__support/FPUtil/x86_64/NextAfterLongDouble.h @@ -16,12 +16,11 @@ #error "Invalid include" #endif +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/bit.h" #include "src/__support/FPUtil/FEnvImpl.h" #include "src/__support/FPUtil/FPBits.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace fputil { diff --git a/libc/src/__support/File/CMakeLists.txt b/libc/src/__support/File/CMakeLists.txt index 5a4af5e..253243f 100644 --- a/libc/src/__support/File/CMakeLists.txt +++ b/libc/src/__support/File/CMakeLists.txt @@ -12,13 +12,14 @@ add_object_library( HDRS file.h DEPENDS + libc.hdr.stdio_macros + libc.hdr.stdint_proxy + libc.hdr.func.realloc + libc.hdr.types.off_t libc.src.__support.CPP.new libc.src.__support.CPP.span libc.src.__support.threads.mutex libc.src.__support.error_or - libc.hdr.types.off_t - libc.hdr.stdio_macros - libc.hdr.func.realloc ) add_object_library( diff --git a/libc/src/__support/File/file.h b/libc/src/__support/File/file.h index 5c97a9c..3652e44 100644 --- a/libc/src/__support/File/file.h +++ b/libc/src/__support/File/file.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_FILE_FILE_H #define LLVM_LIBC_SRC___SUPPORT_FILE_FILE_H +#include "hdr/stdint_proxy.h" #include "hdr/stdio_macros.h" #include "hdr/types/off_t.h" #include "src/__support/CPP/new.h" @@ -18,7 +19,6 @@ #include "src/__support/threads/mutex.h" #include <stddef.h> -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/__support/File/linux/CMakeLists.txt b/libc/src/__support/File/linux/CMakeLists.txt index 84e3d56..7a7bd6c 100644 --- a/libc/src/__support/File/linux/CMakeLists.txt +++ b/libc/src/__support/File/linux/CMakeLists.txt @@ -8,16 +8,17 @@ add_object_library( lseekImpl.h DEPENDS libc.hdr.fcntl_macros + libc.hdr.stdio_macros + libc.hdr.stdint_proxy + libc.hdr.types.off_t + libc.hdr.types.FILE libc.include.sys_syscall libc.include.sys_stat libc.src.__support.CPP.new libc.src.__support.OSUtil.osutil - libc.src.errno.errno libc.src.__support.error_or libc.src.__support.File.file - libc.hdr.types.off_t - libc.hdr.types.FILE - libc.hdr.stdio_macros + libc.src.errno.errno ) add_object_library( diff --git a/libc/src/__support/File/linux/lseekImpl.h b/libc/src/__support/File/linux/lseekImpl.h index 300e5c5..c22a6c5 100644 --- a/libc/src/__support/File/linux/lseekImpl.h +++ b/libc/src/__support/File/linux/lseekImpl.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_FILE_LINUX_LSEEKIMPL_H #define LLVM_LIBC_SRC___SUPPORT_FILE_LINUX_LSEEKIMPL_H +#include "hdr/stdint_proxy.h" // For uint64_t. #include "hdr/types/off_t.h" #include "src/__support/OSUtil/syscall.h" // For internal syscall function. #include "src/__support/common.h" @@ -16,7 +17,6 @@ #include "src/__support/libc_errno.h" #include "src/__support/macros/config.h" -#include <stdint.h> // For uint64_t. #include <sys/syscall.h> // For syscall numbers. namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/__support/GPU/CMakeLists.txt b/libc/src/__support/GPU/CMakeLists.txt index 4ffee01..f8fdfeb 100644 --- a/libc/src/__support/GPU/CMakeLists.txt +++ b/libc/src/__support/GPU/CMakeLists.txt @@ -16,6 +16,7 @@ add_object_library( HDRS allocator.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.common libc.src.__support.RPC.rpc_client libc.src.__support.CPP.atomic diff --git a/libc/src/__support/GPU/allocator.cpp b/libc/src/__support/GPU/allocator.cpp index 7923fbb..866aea7 100644 --- a/libc/src/__support/GPU/allocator.cpp +++ b/libc/src/__support/GPU/allocator.cpp @@ -34,13 +34,12 @@ constexpr static uint32_t BITS_IN_WORD = sizeof(uint32_t) * 8; constexpr static uint32_t MIN_SIZE = 16; constexpr static uint32_t MIN_ALIGNMENT = MIN_SIZE - 1; +// The number of times to attempt claiming an in-progress slab allocation. +constexpr static uint32_t MAX_TRIES = 128; + // A sentinel used to indicate an invalid but non-null pointer value. constexpr static uint64_t SENTINEL = cpp::numeric_limits<uint64_t>::max(); -// The number of times we will try starting on a single index before skipping -// past it. -constexpr static uint32_t MAX_TRIES = 512; - static_assert(!(ARRAY_SIZE & (ARRAY_SIZE - 1)), "Must be a power of two"); namespace impl { @@ -92,20 +91,10 @@ static inline uint32_t xorshift32(uint32_t &state) { return state * 0x9e3779bb; } -// Final stage of murmurhash used to get a unique index for the global array -static inline uint32_t hash(uint32_t x) { - x ^= x >> 16; - x *= 0x85ebca6b; - x ^= x >> 13; - x *= 0xc2b2ae35; - x ^= x >> 16; - return x; -} - // Rounds the input value to the closest permitted chunk size. Here we accept // the sum of the closest three powers of two. For a 2MiB slab size this is 48 // different chunk sizes. This gives us average internal fragmentation of 87.5%. -static inline uint32_t get_chunk_size(uint32_t x) { +static inline constexpr uint32_t get_chunk_size(uint32_t x) { uint32_t y = x < MIN_SIZE ? MIN_SIZE : x; uint32_t pow2 = BITS_IN_WORD - cpp::countl_zero(y - 1); @@ -123,6 +112,16 @@ static inline uint32_t get_chunk_size(uint32_t x) { return (s3 + MIN_ALIGNMENT) & ~MIN_ALIGNMENT; } +// Converts a chunk size into an index suitable for a statically sized array. +static inline constexpr uint32_t get_chunk_id(uint32_t x) { + if (x <= MIN_SIZE) + return 0; + uint32_t y = x >> 4; + if (x < MIN_SIZE << 2) + return cpp::popcount(y); + return cpp::popcount(y) + 3 * (BITS_IN_WORD - cpp::countl_zero(y)) - 7; +} + // Rounds to the nearest power of two. template <uint32_t N, typename T> static inline constexpr T round_up(const T x) { @@ -143,6 +142,12 @@ static inline constexpr bool is_pow2(uint64_t x) { return x && (x & (x - 1)) == 0; } +// Where this chunk size should start looking in the global array. +static inline constexpr uint32_t start_index(uint32_t chunk_index) { + return (ARRAY_SIZE * impl::get_chunk_id(chunk_index)) / + impl::get_chunk_id(SLAB_SIZE / 2); +} + } // namespace impl /// A slab allocator used to hand out identically sized slabs of memory. @@ -251,12 +256,18 @@ struct Slab { // The uniform mask represents which lanes contain a uniform target pointer. // We attempt to place these next to each other. void *result = nullptr; + uint32_t after = ~0u; + uint32_t old_index = 0; for (uint64_t mask = lane_mask; mask; mask = gpu::ballot(lane_mask, !result)) { if (result) continue; - uint32_t start = gpu::broadcast_value(lane_mask, impl::xorshift32(state)); + // We try using any known empty bits from the previous attempt first. + uint32_t start = gpu::shuffle(mask, cpp::countr_zero(uniform & mask), + ~after ? (old_index & ~(BITS_IN_WORD - 1)) + + cpp::countr_zero(~after) + : impl::xorshift32(state)); uint32_t id = impl::lane_count(uniform & mask); uint32_t index = (start + id) % usable_bits(chunk_size); @@ -266,8 +277,9 @@ struct Slab { // Get the mask of bits destined for the same slot and coalesce it. uint64_t match = uniform & gpu::match_any(mask, slot); uint32_t length = cpp::popcount(match); - uint32_t bitmask = static_cast<uint32_t>((uint64_t(1) << length) - 1) - << bit; + uint32_t bitmask = gpu::shuffle( + mask, cpp::countr_zero(match), + static_cast<uint32_t>((uint64_t(1) << length) - 1) << bit); uint32_t before = 0; if (gpu::get_lane_id() == static_cast<uint32_t>(cpp::countr_zero(match))) @@ -278,6 +290,9 @@ struct Slab { result = ptr_from_index(index, chunk_size); else sleep_briefly(); + + after = before | bitmask; + old_index = index; } cpp::atomic_thread_fence(cpp::MemoryOrder::ACQUIRE); @@ -451,66 +466,65 @@ public: // The global array used to search for a valid slab to allocate from. static GuardPtr slots[ARRAY_SIZE] = {}; +// Keep a cache of the last successful slot for each chunk size. Initialize it +// to an even spread of the total size. Must be updated if the chunking scheme +// changes. +#define S(X) (impl::start_index(X)) +static cpp::Atomic<uint32_t> indices[] = { + S(16), S(32), S(48), S(64), S(96), S(112), S(128), + S(192), S(224), S(256), S(384), S(448), S(512), S(768), + S(896), S(1024), S(1536), S(1792), S(2048), S(3072), S(3584), + S(4096), S(6144), S(7168), S(8192), S(12288), S(14336), S(16384), + S(24576), S(28672), S(32768), S(49152), S(57344), S(65536), S(98304), + S(114688), S(131072), S(196608), S(229376), S(262144), S(393216), S(458752), + S(524288), S(786432), S(917504), S(1048576)}; +#undef S + // Tries to find a slab in the table that can support the given chunk size. static Slab *find_slab(uint32_t chunk_size) { - // We start at a hashed value to spread out different chunk sizes. - uint32_t start = impl::hash(chunk_size); - uint64_t lane_mask = gpu::get_lane_mask(); - uint64_t uniform = gpu::match_any(lane_mask, chunk_size); - - Slab *result = nullptr; - uint32_t nudge = 0; - for (uint64_t mask = lane_mask; mask; - mask = gpu::ballot(lane_mask, !result), ++nudge) { - uint32_t index = cpp::numeric_limits<uint32_t>::max(); - for (uint32_t offset = nudge / MAX_TRIES; - gpu::ballot(lane_mask, index == cpp::numeric_limits<uint32_t>::max()); - offset += cpp::popcount(uniform & lane_mask)) { - uint32_t candidate = - (start + offset + impl::lane_count(uniform & lane_mask)) % ARRAY_SIZE; - uint64_t available = - gpu::ballot(lane_mask, slots[candidate].use_count() < - Slab::available_chunks(chunk_size)); - uint32_t new_index = gpu::shuffle( - lane_mask, cpp::countr_zero(available & uniform), candidate); - - // Each uniform group will use the first empty slot they find. - if ((index == cpp::numeric_limits<uint32_t>::max() && - (available & uniform))) - index = new_index; - - // Guaruntees that this loop will eventuall exit if there is no space. - if (offset >= ARRAY_SIZE) { - result = reinterpret_cast<Slab *>(SENTINEL); - index = 0; - } - } + // We start at the index of the last successful allocation for this kind. + uint32_t chunk_id = impl::get_chunk_id(chunk_size); + uint32_t start = indices[chunk_id].load(cpp::MemoryOrder::RELAXED); + uint64_t uniform = gpu::match_any(gpu::get_lane_mask(), chunk_size); + + for (uint32_t offset = 0; offset < ARRAY_SIZE; ++offset) { + uint32_t index = + !offset ? start : (impl::start_index(chunk_size) + offset) % ARRAY_SIZE; - // Try to claim a slot for the found slot. - if (!result) { + if (slots[index].use_count() < Slab::available_chunks(chunk_size)) { + uint64_t lane_mask = gpu::get_lane_mask(); uint64_t reserved = 0; - Slab *slab = slots[index].try_lock(lane_mask & mask, uniform & mask, + + Slab *slab = slots[index].try_lock(lane_mask, uniform & lane_mask, reserved, chunk_size, index); + + // If there is a slab allocation in progress we retry a few times. + for (uint32_t retries = 0; + retries < MAX_TRIES && !slab && reserved != SENTINEL; retries++) { + uint64_t lane_mask = gpu::get_lane_mask(); + slab = slots[index].try_lock(lane_mask, uniform & lane_mask, reserved, + chunk_size, index); + sleep_briefly(); + } + // If we find a slab with a matching chunk size then we store the result. // Otherwise, we need to free the claimed lock and continue. In the case - // of out-of-memory we return a sentinel value. + // of out-of-memory we receive a sentinel value and return a failure. if (slab && reserved <= Slab::available_chunks(chunk_size) && slab->get_chunk_size() == chunk_size) { - result = slab; + if (index != start) + indices[chunk_id].store(index, cpp::MemoryOrder::RELAXED); + return slab; } else if (slab && (reserved > Slab::available_chunks(chunk_size) || slab->get_chunk_size() != chunk_size)) { - if (slab->get_chunk_size() != chunk_size) - start = index + 1; slots[index].unlock(gpu::get_lane_mask(), gpu::get_lane_mask() & uniform); - } else if (!slab && reserved == cpp::numeric_limits<uint64_t>::max()) { - result = reinterpret_cast<Slab *>(SENTINEL); - } else { - sleep_briefly(); + } else if (!slab && reserved == SENTINEL) { + return nullptr; } } } - return result; + return nullptr; } // Release the lock associated with a given slab. diff --git a/libc/src/__support/GPU/allocator.h b/libc/src/__support/GPU/allocator.h index a7cf8bc..67a7ff5 100644 --- a/libc/src/__support/GPU/allocator.h +++ b/libc/src/__support/GPU/allocator.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_GPU_ALLOCATOR_H #define LLVM_LIBC_SRC___SUPPORT_GPU_ALLOCATOR_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { namespace gpu { diff --git a/libc/src/__support/HashTable/CMakeLists.txt b/libc/src/__support/HashTable/CMakeLists.txt index 698b8d0..d5861e7 100644 --- a/libc/src/__support/HashTable/CMakeLists.txt +++ b/libc/src/__support/HashTable/CMakeLists.txt @@ -5,6 +5,7 @@ add_header_library( FLAGS EXPLICIT_SIMD_OPT DEPENDS + libc.hdr.stdint_proxy libc.src.__support.common libc.src.__support.CPP.bit libc.src.__support.macros.properties.cpu_features @@ -26,6 +27,7 @@ add_header_library( table.h DEPENDS .bitmask + libc.hdr.stdint_proxy libc.hdr.types.ENTRY libc.src.__support.CPP.bit libc.src.__support.CPP.new diff --git a/libc/src/__support/HashTable/bitmask.h b/libc/src/__support/HashTable/bitmask.h index 3cac481..a4af496 100644 --- a/libc/src/__support/HashTable/bitmask.h +++ b/libc/src/__support/HashTable/bitmask.h @@ -9,11 +9,11 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_HASHTABLE_BITMASK_H #define LLVM_LIBC_SRC___SUPPORT_HASHTABLE_BITMASK_H +#include "hdr/stdint_proxy.h" // uint8_t, uint64_t #include "src/__support/CPP/bit.h" #include "src/__support/macros/config.h" #include "src/__support/macros/properties/cpu_features.h" #include <stddef.h> // size_t -#include <stdint.h> // uint8_t, uint64_t namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/__support/HashTable/table.h b/libc/src/__support/HashTable/table.h index 10dd971..966ee0f 100644 --- a/libc/src/__support/HashTable/table.h +++ b/libc/src/__support/HashTable/table.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_HASHTABLE_TABLE_H #define LLVM_LIBC_SRC___SUPPORT_HASHTABLE_TABLE_H +#include "hdr/stdint_proxy.h" #include "hdr/types/ENTRY.h" #include "src/__support/CPP/bit.h" // bit_ceil #include "src/__support/CPP/new.h" @@ -21,7 +22,6 @@ #include "src/string/memory_utils/inline_strcmp.h" #include "src/string/string_utils.h" #include <stddef.h> -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/__support/RPC/rpc_server.h b/libc/src/__support/RPC/rpc_server.h index dc3d803..beea4dd 100644 --- a/libc/src/__support/RPC/rpc_server.h +++ b/libc/src/__support/RPC/rpc_server.h @@ -15,13 +15,17 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_RPC_RPC_SERVER_H #define LLVM_LIBC_SRC___SUPPORT_RPC_RPC_SERVER_H +#include "src/__support/macros/properties/compiler.h" + // Workaround for missing __has_builtin in < GCC 10. #ifndef __has_builtin #define __has_builtin(x) 0 #endif // Workaround for missing __builtin_is_constant_evaluated in < GCC 10. -#ifndef __builtin_is_constant_evaluated +// Also this builtin is defined for GCC 9. +#if !(__has_builtin(__builtin_is_constant_evaluated) || \ + (defined(LIBC_COMPILER_IS_GCC) && (LIBC_COMPILER_GCC_VER >= 900))) #define __builtin_is_constant_evaluated(x) 0 #endif diff --git a/libc/src/__support/arg_list.h b/libc/src/__support/arg_list.h index 66afa67..1e26a5e 100644 --- a/libc/src/__support/arg_list.h +++ b/libc/src/__support/arg_list.h @@ -9,12 +9,12 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_ARG_LIST_H #define LLVM_LIBC_SRC___SUPPORT_ARG_LIST_H +#include "hdr/stdint_proxy.h" #include "src/__support/common.h" #include "src/__support/macros/config.h" #include <stdarg.h> #include <stddef.h> -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/__support/big_int.h b/libc/src/__support/big_int.h index 85db31d..10e35ef 100644 --- a/libc/src/__support/big_int.h +++ b/libc/src/__support/big_int.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_BIG_INT_H #define LLVM_LIBC_SRC___SUPPORT_BIG_INT_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/array.h" #include "src/__support/CPP/bit.h" // countl_zero #include "src/__support/CPP/limits.h" @@ -23,7 +24,6 @@ #include "src/__support/number_pair.h" #include <stddef.h> // For size_t -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/__support/block.h b/libc/src/__support/block.h index a58c38b..b0d6576 100644 --- a/libc/src/__support/block.h +++ b/libc/src/__support/block.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_BLOCK_H #define LLVM_LIBC_SRC___SUPPORT_BLOCK_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/algorithm.h" #include "src/__support/CPP/cstddef.h" #include "src/__support/CPP/limits.h" @@ -20,8 +21,6 @@ #include "src/__support/macros/config.h" #include "src/__support/math_extras.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { /// Returns the value rounded down to the nearest multiple of alignment. diff --git a/libc/src/__support/blockstore.h b/libc/src/__support/blockstore.h index efe2234..61ee0ee 100644 --- a/libc/src/__support/blockstore.h +++ b/libc/src/__support/blockstore.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_BLOCKSTORE_H #define LLVM_LIBC_SRC___SUPPORT_BLOCKSTORE_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/array.h" #include "src/__support/CPP/new.h" #include "src/__support/CPP/type_traits.h" @@ -16,7 +17,6 @@ #include "src/__support/macros/config.h" #include <stddef.h> -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/__support/detailed_powers_of_ten.h b/libc/src/__support/detailed_powers_of_ten.h index 28741b8..ab6f2be 100644 --- a/libc/src/__support/detailed_powers_of_ten.h +++ b/libc/src/__support/detailed_powers_of_ten.h @@ -9,11 +9,10 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_DETAILED_POWERS_OF_TEN_H #define LLVM_LIBC_SRC___SUPPORT_DETAILED_POWERS_OF_TEN_H +#include "hdr/stdint_proxy.h" #include "src/__support/common.h" #include "src/__support/macros/config.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/__support/endian_internal.h b/libc/src/__support/endian_internal.h index 77839ad..c78090a 100644 --- a/libc/src/__support/endian_internal.h +++ b/libc/src/__support/endian_internal.h @@ -9,11 +9,10 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_ENDIAN_INTERNAL_H #define LLVM_LIBC_SRC___SUPPORT_ENDIAN_INTERNAL_H -#include "common.h" +#include "hdr/stdint_proxy.h" +#include "src/__support/common.h" #include "src/__support/macros/config.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { // We rely on compiler preprocessor defines to allow for cross compilation. diff --git a/libc/src/__support/fixed_point/CMakeLists.txt b/libc/src/__support/fixed_point/CMakeLists.txt index 235c030..145eedb 100644 --- a/libc/src/__support/fixed_point/CMakeLists.txt +++ b/libc/src/__support/fixed_point/CMakeLists.txt @@ -3,6 +3,7 @@ add_header_library( HDRS fx_rep.h DEPENDS + libc.hdr.stdint_proxy libc.include.llvm-libc-macros.stdfix_macros libc.src.__support.macros.attributes libc.src.__support.CPP.type_traits diff --git a/libc/src/__support/fixed_point/fx_rep.h b/libc/src/__support/fixed_point/fx_rep.h index 7227fff..e68be57 100644 --- a/libc/src/__support/fixed_point/fx_rep.h +++ b/libc/src/__support/fixed_point/fx_rep.h @@ -9,13 +9,12 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_FIXED_POINT_FX_REP_H #define LLVM_LIBC_SRC___SUPPORT_FIXED_POINT_FX_REP_H +#include "hdr/stdint_proxy.h" #include "include/llvm-libc-macros/stdfix-macros.h" #include "src/__support/CPP/type_traits.h" #include "src/__support/macros/attributes.h" // LIBC_INLINE, LIBC_INLINE_VAR #include "src/__support/macros/config.h" -#include <stdint.h> - #ifdef LIBC_COMPILER_HAS_FIXED_POINT namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/__support/float_to_string.h b/libc/src/__support/float_to_string.h index d88bf84..cab146a 100644 --- a/libc/src/__support/float_to_string.h +++ b/libc/src/__support/float_to_string.h @@ -9,8 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_FLOAT_TO_STRING_H #define LLVM_LIBC_SRC___SUPPORT_FLOAT_TO_STRING_H -#include <stdint.h> - +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/limits.h" #include "src/__support/CPP/type_traits.h" #include "src/__support/FPUtil/FPBits.h" diff --git a/libc/src/__support/hash.h b/libc/src/__support/hash.h index 49138b1..6dfaadf 100644 --- a/libc/src/__support/hash.h +++ b/libc/src/__support/hash.h @@ -9,12 +9,12 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_HASH_H #define LLVM_LIBC_SRC___SUPPORT_HASH_H +#include "hdr/stdint_proxy.h" // For uint64_t #include "src/__support/CPP/bit.h" // rotl #include "src/__support/CPP/limits.h" // numeric_limits #include "src/__support/macros/attributes.h" // LIBC_INLINE #include "src/__support/macros/config.h" #include "src/__support/uint128.h" // UInt128 -#include <stdint.h> // For uint64_t namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/__support/high_precision_decimal.h b/libc/src/__support/high_precision_decimal.h index cb4b50c..08af786 100644 --- a/libc/src/__support/high_precision_decimal.h +++ b/libc/src/__support/high_precision_decimal.h @@ -15,11 +15,11 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_HIGH_PRECISION_DECIMAL_H #define LLVM_LIBC_SRC___SUPPORT_HIGH_PRECISION_DECIMAL_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/limits.h" #include "src/__support/ctype_utils.h" #include "src/__support/macros/config.h" #include "src/__support/str_to_integer.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/__support/integer_literals.h b/libc/src/__support/integer_literals.h index f68b7ef..67d241d 100644 --- a/libc/src/__support/integer_literals.h +++ b/libc/src/__support/integer_literals.h @@ -13,13 +13,13 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_INTEGER_LITERALS_H #define LLVM_LIBC_SRC___SUPPORT_INTEGER_LITERALS_H +#include "hdr/stdint_proxy.h" // uintxx_t #include "src/__support/CPP/limits.h" // CHAR_BIT #include "src/__support/ctype_utils.h" #include "src/__support/macros/attributes.h" // LIBC_INLINE #include "src/__support/macros/config.h" #include "src/__support/uint128.h" // UInt128 #include <stddef.h> // size_t -#include <stdint.h> // uintxx_t namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/__support/integer_to_string.h b/libc/src/__support/integer_to_string.h index 65bdcf1..29449bd 100644 --- a/libc/src/__support/integer_to_string.h +++ b/libc/src/__support/integer_to_string.h @@ -57,8 +57,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_INTEGER_TO_STRING_H #define LLVM_LIBC_SRC___SUPPORT_INTEGER_TO_STRING_H -#include <stdint.h> - +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/algorithm.h" // max #include "src/__support/CPP/array.h" #include "src/__support/CPP/bit.h" diff --git a/libc/src/__support/macros/CMakeLists.txt b/libc/src/__support/macros/CMakeLists.txt index 99d4f64..a639d3e 100644 --- a/libc/src/__support/macros/CMakeLists.txt +++ b/libc/src/__support/macros/CMakeLists.txt @@ -35,5 +35,4 @@ add_header_library( DEPENDS .config .optimization - .sanitizer ) diff --git a/libc/src/__support/macros/null_check.h b/libc/src/__support/macros/null_check.h index abf65c56..438ba69 100644 --- a/libc/src/__support/macros/null_check.h +++ b/libc/src/__support/macros/null_check.h @@ -11,9 +11,8 @@ #include "src/__support/macros/config.h" #include "src/__support/macros/optimization.h" -#include "src/__support/macros/sanitizer.h" -#if defined(LIBC_ADD_NULL_CHECKS) && !defined(LIBC_HAS_SANITIZER) +#if defined(LIBC_ADD_NULL_CHECKS) #define LIBC_CRASH_ON_NULLPTR(ptr) \ do { \ if (LIBC_UNLIKELY((ptr) == nullptr)) \ diff --git a/libc/src/__support/macros/properties/CMakeLists.txt b/libc/src/__support/macros/properties/CMakeLists.txt index 80ed63a..dfa2f9c 100644 --- a/libc/src/__support/macros/properties/CMakeLists.txt +++ b/libc/src/__support/macros/properties/CMakeLists.txt @@ -34,6 +34,7 @@ add_header_library( .cpu_features .os libc.hdr.float_macros + libc.hdr.stdint_proxy libc.include.llvm-libc-macros.float16_macros libc.include.llvm-libc-types.float128 ) diff --git a/libc/src/__support/macros/properties/cpu_features.h b/libc/src/__support/macros/properties/cpu_features.h index cdb2df9..fde30ea 100644 --- a/libc/src/__support/macros/properties/cpu_features.h +++ b/libc/src/__support/macros/properties/cpu_features.h @@ -81,7 +81,7 @@ #endif #if defined(__ARM_FEATURE_FMA) || (defined(__AVX2__) && defined(__FMA__)) || \ - defined(__NVPTX__) || defined(__AMDGPU__) || defined(__LIBC_RISCV_USE_FMA) + defined(__NVPTX__) || defined(__AMDGPU__) || defined(__riscv_flen) #define LIBC_TARGET_CPU_HAS_FMA // Provide a more fine-grained control of FMA instruction for ARM targets. #if defined(LIBC_TARGET_CPU_HAS_FPU_HALF) diff --git a/libc/src/__support/macros/properties/types.h b/libc/src/__support/macros/properties/types.h index aec4a48..3259c8a 100644 --- a/libc/src/__support/macros/properties/types.h +++ b/libc/src/__support/macros/properties/types.h @@ -10,7 +10,8 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_MACROS_PROPERTIES_TYPES_H #define LLVM_LIBC_SRC___SUPPORT_MACROS_PROPERTIES_TYPES_H -#include "hdr/float_macros.h" // LDBL_MANT_DIG +#include "hdr/float_macros.h" // LDBL_MANT_DIG +#include "hdr/stdint_proxy.h" // UINT64_MAX, __SIZEOF_INT128__ #include "include/llvm-libc-macros/float16-macros.h" // LIBC_TYPES_HAS_FLOAT16 #include "include/llvm-libc-types/float128.h" // float128 #include "src/__support/macros/config.h" // LIBC_NAMESPACE_DECL @@ -19,8 +20,6 @@ #include "src/__support/macros/properties/cpu_features.h" #include "src/__support/macros/properties/os.h" -#include <stdint.h> // UINT64_MAX, __SIZEOF_INT128__ - // 'long double' properties. #if (LDBL_MANT_DIG == 53) #define LIBC_TYPES_LONG_DOUBLE_IS_FLOAT64 diff --git a/libc/src/__support/macros/sanitizer.h b/libc/src/__support/macros/sanitizer.h index c20412e..84268a1 100644 --- a/libc/src/__support/macros/sanitizer.h +++ b/libc/src/__support/macros/sanitizer.h @@ -23,16 +23,6 @@ #define LIBC_HAS_MEMORY_SANITIZER #endif -#if LIBC_HAS_FEATURE(undefined_behavior_sanitizer) -#define LIBC_HAS_UNDEFINED_BEHAVIOR_SANITIZER -#endif - -#if defined(LIBC_HAS_ADDRESS_SANITIZER) || \ - defined(LIBC_HAS_MEMORY_SANITIZER) || \ - defined(LIBC_HAS_UNDEFINED_BEHAVIOR_SANITIZER) -#define LIBC_HAS_SANITIZER -#endif - #ifdef LIBC_HAS_MEMORY_SANITIZER // Only perform MSAN unpoison in non-constexpr context. #include <sanitizer/msan_interface.h> diff --git a/libc/src/__support/math/CMakeLists.txt b/libc/src/__support/math/CMakeLists.txt index 4a29c29..1050938 100644 --- a/libc/src/__support/math/CMakeLists.txt +++ b/libc/src/__support/math/CMakeLists.txt @@ -4,7 +4,6 @@ add_header_library( acos.h DEPENDS .asin_utils - libc.src.__support.math.asin_utils libc.src.__support.FPUtil.double_double libc.src.__support.FPUtil.dyadic_float libc.src.__support.FPUtil.fenv_impl @@ -18,6 +17,99 @@ add_header_library( ) add_header_library( + acosf + HDRS + acosf.h + DEPENDS + .inv_trigf_utils + libc.src.__support.FPUtil.except_value_utils + libc.src.__support.FPUtil.fp_bits + libc.src.__support.FPUtil.multiply_add + libc.src.__support.FPUtil.polyeval + libc.src.__support.FPUtil.sqrt + libc.src.__support.macros.optimization +) + +add_header_library( + acosf16 + HDRS + acosf16.h + DEPENDS + 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.polyeval + libc.src.__support.FPUtil.sqrt + libc.src.__support.macros.optimization + libc.src.__support.macros.properties.types +) + +add_header_library( + acosh_float_constants + HDRS + acosh_float_constants.h + DEPENDS + libc.src.__support.macros.config +) + +add_header_library( + acoshf_utils + HDRS + acoshf_utils.h + DEPENDS + .acosh_float_constants + libc.src.__support.FPUtil.fp_bits + libc.src.__support.FPUtil.multiply_add + libc.src.__support.FPUtil.polyeval +) + +add_header_library( + acoshf + HDRS + acoshf.h + DEPENDS + .acoshf_utils + libc.src.__support.FPUtil.fenv_impl + libc.src.__support.FPUtil.fp_bits + libc.src.__support.FPUtil.multiply_add + libc.src.__support.FPUtil.sqrt + libc.src.__support.macros.optimization +) + +add_header_library( + acoshf16 + HDRS + acoshf16.h + DEPENDS + .acoshf_utils + 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.polyeval + libc.src.__support.FPUtil.sqrt + libc.src.__support.macros.optimization +) + +add_header_library( + acospif16 + HDRS + acospif16.h + DEPENDS + libc.src.__support.FPUtil.cast + 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.sqrt + libc.src.__support.macros.optimization + libc.src.__support.macros.properties.types +) + +add_header_library( asin_utils HDRS asin_utils.h @@ -32,6 +124,65 @@ add_header_library( ) add_header_library( + asin + HDRS + asin.h + DEPENDS + .asin_utils + libc.src.__support.FPUtil.double_double + libc.src.__support.FPUtil.dyadic_float + 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.sqrt + libc.src.__support.macros.optimization + libc.src.__support.macros.properties.cpu_features +) + +add_header_library( + asinf + HDRS + asinf.h + DEPENDS + .inv_trigf_utils + libc.src.__support.FPUtil.fenv_impl + libc.src.__support.FPUtil.fp_bits + libc.src.__support.FPUtil.except_value_utils + libc.src.__support.FPUtil.multiply_add + libc.src.__support.FPUtil.sqrt + libc.src.__support.macros.config + libc.src.__support.macros.optimization + libc.src.__support.macros.properties.cpu_features +) + +add_header_library( + asinf16 + HDRS + asinf16.h + DEPENDS + libc.src.__support.FPUtil.fenv_impl + libc.src.__support.FPUtil.fp_bits + libc.src.__support.FPUtil.polyeval + libc.src.__support.FPUtil.cast + libc.src.__support.FPUtil.multiply_add + libc.src.__support.FPUtil.sqrt + libc.src.__support.macros.optimization +) + +add_header_library( + erff + HDRS + erff.h + DEPENDS + libc.src.__support.FPUtil.fp_bits + libc.src.__support.FPUtil.except_value_utils + libc.src.__support.FPUtil.multiply_add + libc.src.__support.FPUtil.polyeval + libc.src.__support.macros.optimization +) + +add_header_library( exp_float_constants HDRS exp_float_constants.h @@ -99,6 +250,16 @@ add_header_library( ) add_header_library( + inv_trigf_utils + HDRS + inv_trigf_utils.h + DEPENDS + libc.src.__support.FPUtil.multiply_add + libc.src.__support.FPUtil.polyeval + libc.src.__support.common +) + +add_header_library( frexpf16 HDRS frexpf16.h @@ -237,6 +398,7 @@ add_header_library( HDRS exp10_float16_constants.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.CPP.array ) @@ -257,11 +419,10 @@ add_header_library( DEPENDS .exp10f16_utils libc.src.__support.FPUtil.fp_bits - src.__support.FPUtil.FEnvImpl - src.__support.FPUtil.FPBits - src.__support.FPUtil.cast - src.__support.FPUtil.rounding_mode - src.__support.FPUtil.except_value_utils - src.__support.macros.optimization - src.__support.macros.properties.cpu_features + libc.src.__support.FPUtil.fenv_impl + libc.src.__support.FPUtil.cast + libc.src.__support.FPUtil.rounding_mode + libc.src.__support.FPUtil.except_value_utils + libc.src.__support.macros.optimization + libc.src.__support.macros.properties.cpu_features ) diff --git a/libc/src/__support/math/acos.h b/libc/src/__support/math/acos.h index a7287f1..0e1e413 100644 --- a/libc/src/__support/math/acos.h +++ b/libc/src/__support/math/acos.h @@ -24,10 +24,9 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -using DoubleDouble = fputil::DoubleDouble; -using Float128 = fputil::DyadicFloat<128>; - -static constexpr double acos(double x) { +LIBC_INLINE static constexpr double acos(double x) { + using DoubleDouble = fputil::DoubleDouble; + using namespace asin_internal; using FPBits = fputil::FPBits<double>; FPBits xbits(x); diff --git a/libc/src/__support/math/acosf.h b/libc/src/__support/math/acosf.h new file mode 100644 index 0000000..7a0c0e5 --- /dev/null +++ b/libc/src/__support/math/acosf.h @@ -0,0 +1,147 @@ +//===-- Implementation header for acosf -------------------------*- 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_ACOSF_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ACOSF_H + +#include "inv_trigf_utils.h" +#include "src/__support/FPUtil/FEnvImpl.h" +#include "src/__support/FPUtil/FPBits.h" +#include "src/__support/FPUtil/except_value_utils.h" +#include "src/__support/FPUtil/multiply_add.h" +#include "src/__support/FPUtil/sqrt.h" +#include "src/__support/macros/config.h" +#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +namespace acosf_internal { + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + +static constexpr size_t N_EXCEPTS = 4; + +// Exceptional values when |x| <= 0.5 +static constexpr fputil::ExceptValues<float, N_EXCEPTS> ACOSF_EXCEPTS = {{ + // (inputs, RZ output, RU offset, RD offset, RN offset) + // x = 0x1.110b46p-26, acosf(x) = 0x1.921fb4p0 (RZ) + {0x328885a3, 0x3fc90fda, 1, 0, 1}, + // x = -0x1.110b46p-26, acosf(x) = 0x1.921fb4p0 (RZ) + {0xb28885a3, 0x3fc90fda, 1, 0, 1}, + // x = 0x1.04c444p-12, acosf(x) = 0x1.920f68p0 (RZ) + {0x39826222, 0x3fc907b4, 1, 0, 1}, + // x = -0x1.04c444p-12, acosf(x) = 0x1.923p0 (RZ) + {0xb9826222, 0x3fc91800, 1, 0, 1}, +}}; + +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + +} // namespace acosf_internal + +LIBC_INLINE static constexpr float acosf(float x) { + using namespace acosf_internal; + using namespace inv_trigf_utils_internal; + using FPBits = typename fputil::FPBits<float>; + + FPBits xbits(x); + uint32_t x_uint = xbits.uintval(); + uint32_t x_abs = xbits.uintval() & 0x7fff'ffffU; + uint32_t x_sign = x_uint >> 31; + + // |x| <= 0.5 + if (LIBC_UNLIKELY(x_abs <= 0x3f00'0000U)) { + // |x| < 0x1p-10 + if (LIBC_UNLIKELY(x_abs < 0x3a80'0000U)) { + // When |x| < 2^-10, we use the following approximation: + // acos(x) = pi/2 - asin(x) + // ~ pi/2 - x - x^3 / 6 + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + // Check for exceptional values + if (auto r = ACOSF_EXCEPTS.lookup(x_uint); LIBC_UNLIKELY(r.has_value())) + return r.value(); +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + + double xd = static_cast<double>(x); + return static_cast<float>(fputil::multiply_add( + -0x1.5555555555555p-3 * xd, xd * xd, M_MATH_PI_2 - xd)); + } + + // For |x| <= 0.5, we approximate acosf(x) by: + // acos(x) = pi/2 - asin(x) = pi/2 - x * P(x^2) + // Where P(X^2) = Q(X) is a degree-20 minimax even polynomial approximating + // asin(x)/x on [0, 0.5] generated by Sollya with: + // > Q = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20|], + // [|1, D...|], [0, 0.5]); + double xd = static_cast<double>(x); + double xsq = xd * xd; + double x3 = xd * xsq; + double r = asin_eval(xsq); + return static_cast<float>(fputil::multiply_add(-x3, r, M_MATH_PI_2 - xd)); + } + + // |x| >= 1, return 0, 2pi, or NaNs. + if (LIBC_UNLIKELY(x_abs >= 0x3f80'0000U)) { + if (x_abs == 0x3f80'0000U) + return x_sign ? /* x == -1.0f */ fputil::round_result_slightly_down( + 0x1.921fb6p+1f) + : /* x == 1.0f */ 0.0f; + + if (xbits.is_signaling_nan()) { + fputil::raise_except_if_required(FE_INVALID); + return FPBits::quiet_nan().get_val(); + } + + // |x| <= +/-inf + if (x_abs <= 0x7f80'0000U) { + fputil::set_errno_if_required(EDOM); + fputil::raise_except_if_required(FE_INVALID); + } + + return x + FPBits::quiet_nan().get_val(); + } + + // When 0.5 < |x| < 1, we perform range reduction as follow: + // + // Assume further that 0.5 < x <= 1, and let: + // y = acos(x) + // We use the double angle formula: + // x = cos(y) = 1 - 2 sin^2(y/2) + // So: + // sin(y/2) = sqrt( (1 - x)/2 ) + // And hence: + // y = 2 * asin( sqrt( (1 - x)/2 ) ) + // Let u = (1 - x)/2, then + // acos(x) = 2 * asin( sqrt(u) ) + // Moreover, since 0.5 < x <= 1, + // 0 <= u < 1/4, and 0 <= sqrt(u) < 0.5, + // And hence we can reuse the same polynomial approximation of asin(x) when + // |x| <= 0.5: + // acos(x) ~ 2 * sqrt(u) * P(u). + // + // When -1 < x <= -0.5, we use the identity: + // acos(x) = pi - acos(-x) + // which is reduced to the postive case. + + xbits.set_sign(Sign::POS); + double xd = static_cast<double>(xbits.get_val()); + double u = fputil::multiply_add(-0.5, xd, 0.5); + double cv = 2 * fputil::sqrt<double>(u); + + double r3 = asin_eval(u); + double r = fputil::multiply_add(cv * u, r3, cv); + return static_cast<float>(x_sign ? M_MATH_PI - r : r); +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ACOS_H diff --git a/libc/src/__support/math/acosf16.h b/libc/src/__support/math/acosf16.h new file mode 100644 index 0000000..3f0e002 --- /dev/null +++ b/libc/src/__support/math/acosf16.h @@ -0,0 +1,164 @@ +//===-- Implementation header for acosf16 -----------------------*- 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_ACOSF16_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ACOSF16_H + +#include "include/llvm-libc-macros/float16-macros.h" + +#ifdef LIBC_TYPES_HAS_FLOAT16 + +#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/multiply_add.h" +#include "src/__support/FPUtil/sqrt.h" +#include "src/__support/macros/optimization.h" + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +LIBC_INLINE static constexpr float16 acosf16(float16 x) { + + // Generated by Sollya using the following command: + // > round(pi/2, SG, RN); + // > round(pi, SG, RN); + constexpr float PI_OVER_2 = 0x1.921fb6p0f; + constexpr float PI = 0x1.921fb6p1f; + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + constexpr size_t N_EXCEPTS = 2; + + constexpr fputil::ExceptValues<float16, N_EXCEPTS> ACOSF16_EXCEPTS{{ + // (input, RZ output, RU offset, RD offset, RN offset) + {0xacaf, 0x3e93, 1, 0, 0}, + {0xb874, 0x4052, 1, 0, 1}, + }}; +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + + using FPBits = fputil::FPBits<float16>; + FPBits xbits(x); + + uint16_t x_u = xbits.uintval(); + uint16_t x_abs = x_u & 0x7fff; + uint16_t x_sign = x_u >> 15; + + // |x| > 0x1p0, |x| > 1, or x is NaN. + if (LIBC_UNLIKELY(x_abs > 0x3c00)) { + // acosf16(NaN) = 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; + } + + // 1 < |x| <= +/-inf + fputil::raise_except_if_required(FE_INVALID); + fputil::set_errno_if_required(EDOM); + + return FPBits::quiet_nan().get_val(); + } + + float xf = x; + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + // Handle exceptional values + if (auto r = ACOSF16_EXCEPTS.lookup(x_u); LIBC_UNLIKELY(r.has_value())) + return r.value(); +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + + // |x| == 0x1p0, x is 1 or -1 + // if x is (-)1, return pi, else + // if x is (+)1, return 0 + if (LIBC_UNLIKELY(x_abs == 0x3c00)) + return fputil::cast<float16>(x_sign ? PI : 0.0f); + + float xsq = xf * xf; + + // |x| <= 0x1p-1, |x| <= 0.5 + if (x_abs <= 0x3800) { + // if x is 0, return pi/2 + if (LIBC_UNLIKELY(x_abs == 0)) + return fputil::cast<float16>(PI_OVER_2); + + // Note that: acos(x) = pi/2 + asin(-x) = pi/2 - asin(x) + // Degree-6 minimax polynomial of asin(x) generated by Sollya with: + // > P = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 0.5]); + float interm = + fputil::polyeval(xsq, 0x1.000002p0f, 0x1.554c2ap-3f, 0x1.3541ccp-4f, + 0x1.43b2d6p-5f, 0x1.a0d73ep-5f); + return fputil::cast<float16>(fputil::multiply_add(-xf, interm, PI_OVER_2)); + } + + // When |x| > 0.5, assume that 0.5 < |x| <= 1 + // + // Step-by-step range-reduction proof: + // 1: Let y = asin(x), such that, x = sin(y) + // 2: From complimentary angle identity: + // x = sin(y) = cos(pi/2 - y) + // 3: Let z = pi/2 - y, such that x = cos(z) + // 4: From double angle formula; cos(2A) = 1 - 2 * sin^2(A): + // z = 2A, z/2 = A + // cos(z) = 1 - 2 * sin^2(z/2) + // 5: Make sin(z/2) subject of the formula: + // sin(z/2) = sqrt((1 - cos(z))/2) + // 6: Recall [3]; x = cos(z). Therefore: + // sin(z/2) = sqrt((1 - x)/2) + // 7: Let u = (1 - x)/2 + // 8: Therefore: + // asin(sqrt(u)) = z/2 + // 2 * asin(sqrt(u)) = z + // 9: Recall [3]; z = pi/2 - y. Therefore: + // y = pi/2 - z + // y = pi/2 - 2 * asin(sqrt(u)) + // 10: Recall [1], y = asin(x). Therefore: + // asin(x) = pi/2 - 2 * asin(sqrt(u)) + // 11: Recall that: acos(x) = pi/2 + asin(-x) = pi/2 - asin(x) + // Therefore: + // acos(x) = pi/2 - (pi/2 - 2 * asin(sqrt(u))) + // acos(x) = 2 * asin(sqrt(u)) + // + // THE RANGE REDUCTION, HOW? + // 12: Recall [7], u = (1 - x)/2 + // 13: Since 0.5 < x <= 1, therefore: + // 0 <= u <= 0.25 and 0 <= sqrt(u) <= 0.5 + // + // Hence, we can reuse the same [0, 0.5] domain polynomial approximation for + // Step [11] as `sqrt(u)` is in range. + // When -1 < x <= -0.5, the identity: + // acos(x) = pi - acos(-x) + // allows us to compute for the negative x value (lhs) + // with a positive x value instead (rhs). + + float xf_abs = (xf < 0 ? -xf : xf); + float u = fputil::multiply_add(-0.5f, xf_abs, 0.5f); + float sqrt_u = fputil::sqrt<float>(u); + + // Degree-6 minimax polynomial of asin(x) generated by Sollya with: + // > P = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 0.5]); + float asin_sqrt_u = + sqrt_u * fputil::polyeval(u, 0x1.000002p0f, 0x1.554c2ap-3f, + 0x1.3541ccp-4f, 0x1.43b2d6p-5f, 0x1.a0d73ep-5f); + + return fputil::cast<float16>( + x_sign ? fputil::multiply_add(-2.0f, asin_sqrt_u, PI) : 2 * asin_sqrt_u); +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LIBC_TYPES_HAS_FLOAT16 + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ACOS_H diff --git a/libc/src/__support/math/acosh_float_constants.h b/libc/src/__support/math/acosh_float_constants.h new file mode 100644 index 0000000..2eb245d --- /dev/null +++ b/libc/src/__support/math/acosh_float_constants.h @@ -0,0 +1,114 @@ +//===-- Common constants for acoshf 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_SRC___SUPPORT_MATH_ACOSH_FLOAT_CONSTANTS_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ACOSH_FLOAT_CONSTANTS_H + +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { + +namespace acoshf_internal { + +// Lookup table for (1/f) where f = 1 + n*2^(-7), n = 0..127. +static constexpr double ONE_OVER_F[128] = { + 0x1.0000000000000p+0, 0x1.fc07f01fc07f0p-1, 0x1.f81f81f81f820p-1, + 0x1.f44659e4a4271p-1, 0x1.f07c1f07c1f08p-1, 0x1.ecc07b301ecc0p-1, + 0x1.e9131abf0b767p-1, 0x1.e573ac901e574p-1, 0x1.e1e1e1e1e1e1ep-1, + 0x1.de5d6e3f8868ap-1, 0x1.dae6076b981dbp-1, 0x1.d77b654b82c34p-1, + 0x1.d41d41d41d41dp-1, 0x1.d0cb58f6ec074p-1, 0x1.cd85689039b0bp-1, + 0x1.ca4b3055ee191p-1, 0x1.c71c71c71c71cp-1, 0x1.c3f8f01c3f8f0p-1, + 0x1.c0e070381c0e0p-1, 0x1.bdd2b899406f7p-1, 0x1.bacf914c1bad0p-1, + 0x1.b7d6c3dda338bp-1, 0x1.b4e81b4e81b4fp-1, 0x1.b2036406c80d9p-1, + 0x1.af286bca1af28p-1, 0x1.ac5701ac5701bp-1, 0x1.a98ef606a63bep-1, + 0x1.a6d01a6d01a6dp-1, 0x1.a41a41a41a41ap-1, 0x1.a16d3f97a4b02p-1, + 0x1.9ec8e951033d9p-1, 0x1.9c2d14ee4a102p-1, 0x1.999999999999ap-1, + 0x1.970e4f80cb872p-1, 0x1.948b0fcd6e9e0p-1, 0x1.920fb49d0e229p-1, + 0x1.8f9c18f9c18fap-1, 0x1.8d3018d3018d3p-1, 0x1.8acb90f6bf3aap-1, + 0x1.886e5f0abb04ap-1, 0x1.8618618618618p-1, 0x1.83c977ab2beddp-1, + 0x1.8181818181818p-1, 0x1.7f405fd017f40p-1, 0x1.7d05f417d05f4p-1, + 0x1.7ad2208e0ecc3p-1, 0x1.78a4c8178a4c8p-1, 0x1.767dce434a9b1p-1, + 0x1.745d1745d1746p-1, 0x1.724287f46debcp-1, 0x1.702e05c0b8170p-1, + 0x1.6e1f76b4337c7p-1, 0x1.6c16c16c16c17p-1, 0x1.6a13cd1537290p-1, + 0x1.6816816816817p-1, 0x1.661ec6a5122f9p-1, 0x1.642c8590b2164p-1, + 0x1.623fa77016240p-1, 0x1.6058160581606p-1, 0x1.5e75bb8d015e7p-1, + 0x1.5c9882b931057p-1, 0x1.5ac056b015ac0p-1, 0x1.58ed2308158edp-1, + 0x1.571ed3c506b3ap-1, 0x1.5555555555555p-1, 0x1.5390948f40febp-1, + 0x1.51d07eae2f815p-1, 0x1.5015015015015p-1, 0x1.4e5e0a72f0539p-1, + 0x1.4cab88725af6ep-1, 0x1.4afd6a052bf5bp-1, 0x1.49539e3b2d067p-1, + 0x1.47ae147ae147bp-1, 0x1.460cbc7f5cf9ap-1, 0x1.446f86562d9fbp-1, + 0x1.42d6625d51f87p-1, 0x1.4141414141414p-1, 0x1.3fb013fb013fbp-1, + 0x1.3e22cbce4a902p-1, 0x1.3c995a47babe7p-1, 0x1.3b13b13b13b14p-1, + 0x1.3991c2c187f63p-1, 0x1.3813813813814p-1, 0x1.3698df3de0748p-1, + 0x1.3521cfb2b78c1p-1, 0x1.33ae45b57bcb2p-1, 0x1.323e34a2b10bfp-1, + 0x1.30d190130d190p-1, 0x1.2f684bda12f68p-1, 0x1.2e025c04b8097p-1, + 0x1.2c9fb4d812ca0p-1, 0x1.2b404ad012b40p-1, 0x1.29e4129e4129ep-1, + 0x1.288b01288b013p-1, 0x1.27350b8812735p-1, 0x1.25e22708092f1p-1, + 0x1.2492492492492p-1, 0x1.23456789abcdfp-1, 0x1.21fb78121fb78p-1, + 0x1.20b470c67c0d9p-1, 0x1.1f7047dc11f70p-1, 0x1.1e2ef3b3fb874p-1, + 0x1.1cf06ada2811dp-1, 0x1.1bb4a4046ed29p-1, 0x1.1a7b9611a7b96p-1, + 0x1.19453808ca29cp-1, 0x1.1811811811812p-1, 0x1.16e0689427379p-1, + 0x1.15b1e5f75270dp-1, 0x1.1485f0e0acd3bp-1, 0x1.135c81135c811p-1, + 0x1.12358e75d3033p-1, 0x1.1111111111111p-1, 0x1.0fef010fef011p-1, + 0x1.0ecf56be69c90p-1, 0x1.0db20a88f4696p-1, 0x1.0c9714fbcda3bp-1, + 0x1.0b7e6ec259dc8p-1, 0x1.0a6810a6810a7p-1, 0x1.0953f39010954p-1, + 0x1.0842108421084p-1, 0x1.073260a47f7c6p-1, 0x1.0624dd2f1a9fcp-1, + 0x1.05197f7d73404p-1, 0x1.0410410410410p-1, 0x1.03091b51f5e1ap-1, + 0x1.0204081020408p-1, 0x1.0101010101010p-1}; + +// Lookup table for log(f) = log(1 + n*2^(-7)) where n = 0..127. +static constexpr double LOG_F[128] = { + 0x0.0000000000000p+0, 0x1.fe02a6b106788p-8, 0x1.fc0a8b0fc03e3p-7, + 0x1.7b91b07d5b11ap-6, 0x1.f829b0e783300p-6, 0x1.39e87b9febd5fp-5, + 0x1.77458f632dcfcp-5, 0x1.b42dd711971bep-5, 0x1.f0a30c01162a6p-5, + 0x1.16536eea37ae0p-4, 0x1.341d7961bd1d0p-4, 0x1.51b073f06183fp-4, + 0x1.6f0d28ae56b4bp-4, 0x1.8c345d6319b20p-4, 0x1.a926d3a4ad563p-4, + 0x1.c5e548f5bc743p-4, 0x1.e27076e2af2e5p-4, 0x1.fec9131dbeabap-4, + 0x1.0d77e7cd08e59p-3, 0x1.1b72ad52f67a0p-3, 0x1.29552f81ff523p-3, + 0x1.371fc201e8f74p-3, 0x1.44d2b6ccb7d1ep-3, 0x1.526e5e3a1b437p-3, + 0x1.5ff3070a793d3p-3, 0x1.6d60fe719d21cp-3, 0x1.7ab890210d909p-3, + 0x1.87fa06520c910p-3, 0x1.9525a9cf456b4p-3, 0x1.a23bc1fe2b563p-3, + 0x1.af3c94e80bff2p-3, 0x1.bc286742d8cd6p-3, 0x1.c8ff7c79a9a21p-3, + 0x1.d5c216b4fbb91p-3, 0x1.e27076e2af2e5p-3, 0x1.ef0adcbdc5936p-3, + 0x1.fb9186d5e3e2ap-3, 0x1.0402594b4d040p-2, 0x1.0a324e27390e3p-2, + 0x1.1058bf9ae4ad5p-2, 0x1.1675cababa60ep-2, 0x1.1c898c16999fap-2, + 0x1.22941fbcf7965p-2, 0x1.2895a13de86a3p-2, 0x1.2e8e2bae11d30p-2, + 0x1.347dd9a987d54p-2, 0x1.3a64c556945e9p-2, 0x1.404308686a7e3p-2, + 0x1.4618bc21c5ec2p-2, 0x1.4be5f957778a0p-2, 0x1.51aad872df82dp-2, + 0x1.5767717455a6cp-2, 0x1.5d1bdbf5809cap-2, 0x1.62c82f2b9c795p-2, + 0x1.686c81e9b14aep-2, 0x1.6e08eaa2ba1e3p-2, 0x1.739d7f6bbd006p-2, + 0x1.792a55fdd47a2p-2, 0x1.7eaf83b82afc3p-2, 0x1.842d1da1e8b17p-2, + 0x1.89a3386c1425ap-2, 0x1.8f11e873662c7p-2, 0x1.947941c2116fap-2, + 0x1.99d958117e08ap-2, 0x1.9f323ecbf984bp-2, 0x1.a484090e5bb0ap-2, + 0x1.a9cec9a9a0849p-2, 0x1.af1293247786bp-2, 0x1.b44f77bcc8f62p-2, + 0x1.b9858969310fbp-2, 0x1.beb4d9da71b7bp-2, 0x1.c3dd7a7cdad4dp-2, + 0x1.c8ff7c79a9a21p-2, 0x1.ce1af0b85f3ebp-2, 0x1.d32fe7e00ebd5p-2, + 0x1.d83e7258a2f3ep-2, 0x1.dd46a04c1c4a0p-2, 0x1.e24881a7c6c26p-2, + 0x1.e744261d68787p-2, 0x1.ec399d2468cc0p-2, 0x1.f128f5faf06ecp-2, + 0x1.f6123fa7028acp-2, 0x1.faf588f78f31ep-2, 0x1.ffd2e0857f498p-2, + 0x1.02552a5a5d0fep-1, 0x1.04bdf9da926d2p-1, 0x1.0723e5c1cdf40p-1, + 0x1.0986f4f573520p-1, 0x1.0be72e4252a82p-1, 0x1.0e44985d1cc8bp-1, + 0x1.109f39e2d4c96p-1, 0x1.12f719593efbcp-1, 0x1.154c3d2f4d5e9p-1, + 0x1.179eabbd899a0p-1, 0x1.19ee6b467c96ep-1, 0x1.1c3b81f713c24p-1, + 0x1.1e85f5e7040d0p-1, 0x1.20cdcd192ab6dp-1, 0x1.23130d7bebf42p-1, + 0x1.2555bce98f7cbp-1, 0x1.2795e1289b11ap-1, 0x1.29d37fec2b08ap-1, + 0x1.2c0e9ed448e8bp-1, 0x1.2e47436e40268p-1, 0x1.307d7334f10bep-1, + 0x1.32b1339121d71p-1, 0x1.34e289d9ce1d3p-1, 0x1.37117b54747b5p-1, + 0x1.393e0d3562a19p-1, 0x1.3b68449fffc22p-1, 0x1.3d9026a7156fap-1, + 0x1.3fb5b84d16f42p-1, 0x1.41d8fe84672aep-1, 0x1.43f9fe2f9ce67p-1, + 0x1.4618bc21c5ec2p-1, 0x1.48353d1ea88dfp-1, 0x1.4a4f85db03ebbp-1, + 0x1.4c679afccee39p-1, 0x1.4e7d811b75bb0p-1, 0x1.50913cc01686bp-1, + 0x1.52a2d265bc5aap-1, 0x1.54b2467999497p-1, 0x1.56bf9d5b3f399p-1, + 0x1.58cadb5cd7989p-1, 0x1.5ad404c359f2cp-1, 0x1.5cdb1dc6c1764p-1, + 0x1.5ee02a9241675p-1, 0x1.60e32f44788d8p-1}; + +} // namespace acoshf_internal + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ACOSH_FLOAT_CONSTANTS_H diff --git a/libc/src/__support/math/acoshf.h b/libc/src/__support/math/acoshf.h new file mode 100644 index 0000000..4e00311 --- /dev/null +++ b/libc/src/__support/math/acoshf.h @@ -0,0 +1,86 @@ +//===-- Implementation header for acoshf ------------------------*- 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_ACOSHF_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ACOSHF_H + +#include "acoshf_utils.h" +#include "src/__support/FPUtil/FEnvImpl.h" +#include "src/__support/FPUtil/FPBits.h" +#include "src/__support/FPUtil/multiply_add.h" +#include "src/__support/FPUtil/sqrt.h" +#include "src/__support/macros/config.h" +#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +LIBC_INLINE static constexpr float acoshf(float x) { + using namespace acoshf_internal; + using FPBits_t = typename fputil::FPBits<float>; + FPBits_t xbits(x); + + if (LIBC_UNLIKELY(x <= 1.0f)) { + if (x == 1.0f) + return 0.0f; + // x < 1. + fputil::set_errno_if_required(EDOM); + fputil::raise_except_if_required(FE_INVALID); + return FPBits_t::quiet_nan().get_val(); + } + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + uint32_t x_u = xbits.uintval(); + if (LIBC_UNLIKELY(x_u >= 0x4f8ffb03)) { + if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) + return x; + + // Helper functions to set results for exceptional cases. + auto round_result_slightly_down = [](float r) -> float { + volatile float tmp = r; + tmp = tmp - 0x1.0p-25f; + return tmp; + }; + auto round_result_slightly_up = [](float r) -> float { + volatile float tmp = r; + tmp = tmp + 0x1.0p-25f; + return tmp; + }; + + switch (x_u) { + case 0x4f8ffb03: // x = 0x1.1ff606p32f + return round_result_slightly_up(0x1.6fdd34p4f); + case 0x5c569e88: // x = 0x1.ad3d1p57f + return round_result_slightly_up(0x1.45c146p5f); + case 0x5e68984e: // x = 0x1.d1309cp61f + return round_result_slightly_up(0x1.5c9442p5f); + case 0x655890d3: // x = 0x1.b121a6p75f + return round_result_slightly_down(0x1.a9a3f2p5f); + case 0x6eb1a8ec: // x = 0x1.6351d8p94f + return round_result_slightly_down(0x1.08b512p6f); + case 0x7997f30a: // x = 0x1.2fe614p116f + return round_result_slightly_up(0x1.451436p6f); + } + } +#else + if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) + return x; +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + + double x_d = static_cast<double>(x); + // acosh(x) = log(x + sqrt(x^2 - 1)) + return static_cast<float>(log_eval( + x_d + fputil::sqrt<double>(fputil::multiply_add(x_d, x_d, -1.0)))); +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ACOSHF_H diff --git a/libc/src/__support/math/acoshf16.h b/libc/src/__support/math/acoshf16.h new file mode 100644 index 0000000..e5be2a8 --- /dev/null +++ b/libc/src/__support/math/acoshf16.h @@ -0,0 +1,123 @@ +//===-- Implementation header for acoshf16 ----------------------*- 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_ACOSHF16_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ACOSHF16_H + +#include "include/llvm-libc-macros/float16-macros.h" + +#ifdef LIBC_TYPES_HAS_FLOAT16 + +#include "acoshf_utils.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/multiply_add.h" +#include "src/__support/FPUtil/sqrt.h" +#include "src/__support/macros/config.h" +#include "src/__support/macros/optimization.h" + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +LIBC_INLINE static constexpr float16 acoshf16(float16 x) { + + using namespace acoshf_internal; + constexpr size_t N_EXCEPTS = 2; + constexpr fputil::ExceptValues<float16, N_EXCEPTS> ACOSHF16_EXCEPTS{{ + // (input, RZ output, RU offset, RD offset, RN offset) + // x = 0x1.6dcp+1, acoshf16(x) = 0x1.b6p+0 (RZ) + {0x41B7, 0x3ED8, 1, 0, 0}, + // x = 0x1.39p+0, acoshf16(x) = 0x1.4f8p-1 (RZ) + {0x3CE4, 0x393E, 1, 0, 1}, + }}; + + using FPBits = fputil::FPBits<float16>; + FPBits xbits(x); + uint16_t x_u = xbits.uintval(); + + // Check for NaN input first. + if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) { + if (xbits.is_signaling_nan()) { + fputil::raise_except_if_required(FE_INVALID); + return FPBits::quiet_nan().get_val(); + } + if (xbits.is_neg()) { + fputil::set_errno_if_required(EDOM); + fputil::raise_except_if_required(FE_INVALID); + return FPBits::quiet_nan().get_val(); + } + return x; + } + + // Domain error for inputs less than 1.0. + if (LIBC_UNLIKELY(x <= 1.0f)) { + if (x == 1.0f) + return FPBits::zero().get_val(); + fputil::set_errno_if_required(EDOM); + fputil::raise_except_if_required(FE_INVALID); + return FPBits::quiet_nan().get_val(); + } + + if (auto r = ACOSHF16_EXCEPTS.lookup(xbits.uintval()); + LIBC_UNLIKELY(r.has_value())) + return r.value(); + + float xf = x; + // High-precision polynomial approximation for inputs close to 1.0 + // ([1, 1.25)). + // + // Brief derivation: + // 1. Expand acosh(1 + delta) using Taylor series around delta=0: + // acosh(1 + delta) ≈ sqrt(2 * delta) * [1 - delta/12 + 3*delta^2/160 + // - 5*delta^3/896 + 35*delta^4/18432 + ...] + // 2. Truncate the series to fit accurately for delta in [0, 0.25]. + // 3. Polynomial coefficients (from sollya) used here are: + // P(delta) ≈ 1 - 0x1.555556p-4 * delta + 0x1.333334p-6 * delta^2 + // - 0x1.6db6dcp-8 * delta^3 + 0x1.f1c71cp-10 * delta^4 + // 4. The Sollya commands used to generate these coefficients were: + // > display = hexadecimal; + // > round(1/12, SG, RN); + // > round(3/160, SG, RN); + // > round(5/896, SG, RN); + // > round(35/18432, SG, RN); + // With hexadecimal display mode enabled, the outputs were: + // 0x1.555556p-4 + // 0x1.333334p-6 + // 0x1.6db6dcp-8 + // 0x1.f1c71cp-10 + // 5. The maximum absolute error, estimated using: + // dirtyinfnorm(acosh(1 + x) - sqrt(2*x) * P(x), [0, 0.25]) + // is: + // 0x1.d84281p-22 + if (LIBC_UNLIKELY(x_u < 0x3D00U)) { + float delta = xf - 1.0f; + float sqrt_2_delta = fputil::sqrt<float>(2.0 * delta); + float pe = fputil::polyeval(delta, 0x1p+0f, -0x1.555556p-4f, 0x1.333334p-6f, + -0x1.6db6dcp-8f, 0x1.f1c71cp-10f); + float approx = sqrt_2_delta * pe; + return fputil::cast<float16>(approx); + } + + // acosh(x) = log(x + sqrt(x^2 - 1)) + float sqrt_term = fputil::sqrt<float>(fputil::multiply_add(xf, xf, -1.0f)); + float result = static_cast<float>(log_eval(xf + sqrt_term)); + + return fputil::cast<float16>(result); +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LIBC_TYPES_HAS_FLOAT16 + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ACOSHF16_H diff --git a/libc/src/__support/math/acoshf_utils.h b/libc/src/__support/math/acoshf_utils.h new file mode 100644 index 0000000..808c3dd --- /dev/null +++ b/libc/src/__support/math/acoshf_utils.h @@ -0,0 +1,60 @@ +//===-- Collection of utils for acoshf --------------------------*- 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_ACOSHF_UTILS_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ACOSHF_UTILS_H + +#include "acosh_float_constants.h" +#include "src/__support/FPUtil/FPBits.h" +#include "src/__support/FPUtil/PolyEval.h" +#include "src/__support/FPUtil/multiply_add.h" + +namespace LIBC_NAMESPACE_DECL { + +namespace acoshf_internal { + +// x should be positive, normal finite value +LIBC_INLINE static double log_eval(double x) { + // For x = 2^ex * (1 + mx) + // log(x) = ex * log(2) + log(1 + mx) + using FPB = fputil::FPBits<double>; + FPB bs(x); + + double ex = static_cast<double>(bs.get_exponent()); + + // p1 is the leading 7 bits of mx, i.e. + // p1 * 2^(-7) <= m_x < (p1 + 1) * 2^(-7). + int p1 = static_cast<int>(bs.get_mantissa() >> (FPB::FRACTION_LEN - 7)); + + // Set bs to (1 + (mx - p1*2^(-7)) + bs.set_uintval(bs.uintval() & (FPB::FRACTION_MASK >> 7)); + bs.set_biased_exponent(FPB::EXP_BIAS); + // dx = (mx - p1*2^(-7)) / (1 + p1*2^(-7)). + double dx = (bs.get_val() - 1.0) * ONE_OVER_F[p1]; + + // Minimax polynomial of log(1 + dx) generated by Sollya with: + // > P = fpminimax(log(1 + x)/x, 6, [|D...|], [0, 2^-7]); + const double COEFFS[6] = {-0x1.ffffffffffffcp-2, 0x1.5555555552ddep-2, + -0x1.ffffffefe562dp-3, 0x1.9999817d3a50fp-3, + -0x1.554317b3f67a5p-3, 0x1.1dc5c45e09c18p-3}; + double dx2 = dx * dx; + double c1 = fputil::multiply_add(dx, COEFFS[1], COEFFS[0]); + double c2 = fputil::multiply_add(dx, COEFFS[3], COEFFS[2]); + double c3 = fputil::multiply_add(dx, COEFFS[5], COEFFS[4]); + + double p = fputil::polyeval(dx2, dx, c1, c2, c3); + double result = + fputil::multiply_add(ex, /*log(2)*/ 0x1.62e42fefa39efp-1, LOG_F[p1] + p); + return result; +} + +} // namespace acoshf_internal + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ACOSHF_UTILS_H diff --git a/libc/src/__support/math/acospif16.h b/libc/src/__support/math/acospif16.h new file mode 100644 index 0000000..cf29c76 --- /dev/null +++ b/libc/src/__support/math/acospif16.h @@ -0,0 +1,147 @@ +//===-- Implementation header for acospif16 ---------------------*- 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_ACOSPIF16_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ACOSPIF16_H + +#include "include/llvm-libc-macros/float16-macros.h" + +#ifdef LIBC_TYPES_HAS_FLOAT16 + +#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/multiply_add.h" +#include "src/__support/FPUtil/sqrt.h" +#include "src/__support/macros/optimization.h" + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +LIBC_INLINE static constexpr float16 acospif16(float16 x) { + using FPBits = fputil::FPBits<float16>; + FPBits xbits(x); + + uint16_t x_u = xbits.uintval(); + uint16_t x_abs = x_u & 0x7fff; + uint16_t x_sign = x_u >> 15; + + // |x| > 0x1p0, |x| > 1, or x is NaN. + if (LIBC_UNLIKELY(x_abs > 0x3c00)) { + // acospif16(NaN) = 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; + } + + // 1 < |x| <= +inf + fputil::raise_except_if_required(FE_INVALID); + fputil::set_errno_if_required(EDOM); + + return FPBits::quiet_nan().get_val(); + } + + // |x| == 0x1p0, x is 1 or -1 + // if x is (-)1, return 1 + // if x is (+)1, return 0 + if (LIBC_UNLIKELY(x_abs == 0x3c00)) + return fputil::cast<float16>(x_sign ? 1.0f : 0.0f); + + float xf = x; + float xsq = xf * xf; + + // Degree-6 minimax polynomial coefficients of asin(x) generated by Sollya + // with: > P = fpminimax(asin(x)/(pi * x), [|0, 2, 4, 6, 8|], [|SG...|], [0, + // 0.5]); + constexpr float POLY_COEFFS[5] = {0x1.45f308p-2f, 0x1.b2900cp-5f, + 0x1.897e36p-6f, 0x1.9efafcp-7f, + 0x1.06d884p-6f}; + // |x| <= 0x1p-1, |x| <= 0.5 + if (x_abs <= 0x3800) { + // if x is 0, return 0.5 + if (LIBC_UNLIKELY(x_abs == 0)) + return fputil::cast<float16>(0.5f); + + // Note that: acos(x) = pi/2 + asin(-x) = pi/2 - asin(x), then + // acospi(x) = 0.5 - asin(x)/pi + float interm = + fputil::polyeval(xsq, POLY_COEFFS[0], POLY_COEFFS[1], POLY_COEFFS[2], + POLY_COEFFS[3], POLY_COEFFS[4]); + + return fputil::cast<float16>(fputil::multiply_add(-xf, interm, 0.5f)); + } + + // When |x| > 0.5, assume that 0.5 < |x| <= 1 + // + // Step-by-step range-reduction proof: + // 1: Let y = asin(x), such that, x = sin(y) + // 2: From complimentary angle identity: + // x = sin(y) = cos(pi/2 - y) + // 3: Let z = pi/2 - y, such that x = cos(z) + // 4: From double angle formula; cos(2A) = 1 - 2 * sin^2(A): + // z = 2A, z/2 = A + // cos(z) = 1 - 2 * sin^2(z/2) + // 5: Make sin(z/2) subject of the formula: + // sin(z/2) = sqrt((1 - cos(z))/2) + // 6: Recall [3]; x = cos(z). Therefore: + // sin(z/2) = sqrt((1 - x)/2) + // 7: Let u = (1 - x)/2 + // 8: Therefore: + // asin(sqrt(u)) = z/2 + // 2 * asin(sqrt(u)) = z + // 9: Recall [3]; z = pi/2 - y. Therefore: + // y = pi/2 - z + // y = pi/2 - 2 * asin(sqrt(u)) + // 10: Recall [1], y = asin(x). Therefore: + // asin(x) = pi/2 - 2 * asin(sqrt(u)) + // 11: Recall that: acos(x) = pi/2 + asin(-x) = pi/2 - asin(x) + // Therefore: + // acos(x) = pi/2 - (pi/2 - 2 * asin(sqrt(u))) + // acos(x) = 2 * asin(sqrt(u)) + // acospi(x) = 2 * (asin(sqrt(u)) / pi) + // + // THE RANGE REDUCTION, HOW? + // 12: Recall [7], u = (1 - x)/2 + // 13: Since 0.5 < x <= 1, therefore: + // 0 <= u <= 0.25 and 0 <= sqrt(u) <= 0.5 + // + // Hence, we can reuse the same [0, 0.5] domain polynomial approximation for + // Step [11] as `sqrt(u)` is in range. + // When -1 < x <= -0.5, the identity: + // acos(x) = pi - acos(-x) + // acospi(x) = 1 - acos(-x)/pi + // allows us to compute for the negative x value (lhs) + // with a positive x value instead (rhs). + + float xf_abs = (xf < 0 ? -xf : xf); + float u = fputil::multiply_add(-0.5f, xf_abs, 0.5f); + float sqrt_u = fputil::sqrt<float>(u); + + float asin_sqrt_u = + sqrt_u * fputil::polyeval(u, POLY_COEFFS[0], POLY_COEFFS[1], + POLY_COEFFS[2], POLY_COEFFS[3], POLY_COEFFS[4]); + + // Same as acos(x), but devided the expression with pi + return fputil::cast<float16>( + x_sign ? fputil::multiply_add(-2.0f, asin_sqrt_u, 1.0f) + : 2.0f * asin_sqrt_u); +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LIBC_TYPES_HAS_FLOAT16 + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ACOSPIF16_H diff --git a/libc/src/__support/math/asin.h b/libc/src/__support/math/asin.h new file mode 100644 index 0000000..5e06d04 --- /dev/null +++ b/libc/src/__support/math/asin.h @@ -0,0 +1,297 @@ +//===-- Implementation header for asin --------------------------*- 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_ASIN_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ASIN_H + +#include "asin_utils.h" +#include "src/__support/FPUtil/FEnvImpl.h" +#include "src/__support/FPUtil/FPBits.h" +#include "src/__support/FPUtil/double_double.h" +#include "src/__support/FPUtil/dyadic_float.h" +#include "src/__support/FPUtil/multiply_add.h" +#include "src/__support/FPUtil/sqrt.h" +#include "src/__support/macros/config.h" +#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY +#include "src/__support/macros/properties/cpu_features.h" // LIBC_TARGET_CPU_HAS_FMA +#include "src/__support/math/asin_utils.h" + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +LIBC_INLINE static constexpr double asin(double x) { + using namespace asin_internal; + using FPBits = fputil::FPBits<double>; + + FPBits xbits(x); + int x_exp = xbits.get_biased_exponent(); + + // |x| < 0.5. + if (x_exp < FPBits::EXP_BIAS - 1) { + // |x| < 2^-26. + if (LIBC_UNLIKELY(x_exp < FPBits::EXP_BIAS - 26)) { + // When |x| < 2^-26, the relative error of the approximation asin(x) ~ x + // is: + // |asin(x) - x| / |asin(x)| < |x^3| / (6|x|) + // = x^2 / 6 + // < 2^-54 + // < epsilon(1)/2. + // So the correctly rounded values of asin(x) are: + // = x + sign(x)*eps(x) if rounding mode = FE_TOWARDZERO, + // or (rounding mode = FE_UPWARD and x is + // negative), + // = x otherwise. + // To simplify the rounding decision and make it more efficient, we use + // fma(x, 2^-54, x) instead. + // Note: to use the formula x + 2^-54*x to decide the correct rounding, we + // do need fma(x, 2^-54, x) to prevent underflow caused by 2^-54*x when + // |x| < 2^-1022. For targets without FMA instructions, when x is close to + // denormal range, we normalize x, +#if defined(LIBC_MATH_HAS_SKIP_ACCURATE_PASS) + return x; +#elif defined(LIBC_TARGET_CPU_HAS_FMA_DOUBLE) + return fputil::multiply_add(x, 0x1.0p-54, x); +#else + if (xbits.abs().uintval() == 0) + return x; + // Get sign(x) * min_normal. + FPBits eps_bits = FPBits::min_normal(); + eps_bits.set_sign(xbits.sign()); + double eps = eps_bits.get_val(); + double normalize_const = (x_exp == 0) ? eps : 0.0; + double scaled_normal = + fputil::multiply_add(x + normalize_const, 0x1.0p54, eps); + return fputil::multiply_add(scaled_normal, 0x1.0p-54, -normalize_const); +#endif // LIBC_MATH_HAS_SKIP_ACCURATE_PASS + } + +#ifdef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + return x * asin_eval(x * x); +#else + using Float128 = fputil::DyadicFloat<128>; + using DoubleDouble = fputil::DoubleDouble; + + unsigned idx = 0; + DoubleDouble x_sq = fputil::exact_mult(x, x); + double err = xbits.abs().get_val() * 0x1.0p-51; + // Polynomial approximation: + // p ~ asin(x)/x + + DoubleDouble p = asin_eval(x_sq, idx, err); + // asin(x) ~ x * (ASIN_COEFFS[idx][0] + p) + DoubleDouble r0 = fputil::exact_mult(x, p.hi); + double r_lo = fputil::multiply_add(x, p.lo, r0.lo); + + // Ziv's accuracy test. + + double r_upper = r0.hi + (r_lo + err); + double r_lower = r0.hi + (r_lo - err); + + if (LIBC_LIKELY(r_upper == r_lower)) + return r_upper; + + // Ziv's accuracy test failed, perform 128-bit calculation. + + // Recalculate mod 1/64. + idx = static_cast<unsigned>(fputil::nearest_integer(x_sq.hi * 0x1.0p6)); + + // Get x^2 - idx/64 exactly. When FMA is available, double-double + // multiplication will be correct for all rounding modes. Otherwise we use + // Float128 directly. + Float128 x_f128(x); + +#ifdef LIBC_TARGET_CPU_HAS_FMA_DOUBLE + // u = x^2 - idx/64 + Float128 u_hi( + fputil::multiply_add(static_cast<double>(idx), -0x1.0p-6, x_sq.hi)); + Float128 u = fputil::quick_add(u_hi, Float128(x_sq.lo)); +#else + Float128 x_sq_f128 = fputil::quick_mul(x_f128, x_f128); + Float128 u = fputil::quick_add( + x_sq_f128, Float128(static_cast<double>(idx) * (-0x1.0p-6))); +#endif // LIBC_TARGET_CPU_HAS_FMA_DOUBLE + + Float128 p_f128 = asin_eval(u, idx); + Float128 r = fputil::quick_mul(x_f128, p_f128); + + return static_cast<double>(r); +#endif // LIBC_MATH_HAS_SKIP_ACCURATE_PASS + } + // |x| >= 0.5 + + double x_abs = xbits.abs().get_val(); + + // Maintaining the sign: + constexpr double SIGN[2] = {1.0, -1.0}; + double x_sign = SIGN[xbits.is_neg()]; + + // |x| >= 1 + if (LIBC_UNLIKELY(x_exp >= FPBits::EXP_BIAS)) { + // x = +-1, asin(x) = +- pi/2 + if (x_abs == 1.0) { + // return +- pi/2 + return fputil::multiply_add(x_sign, PI_OVER_TWO.hi, + x_sign * PI_OVER_TWO.lo); + } + // |x| > 1, return NaN. + if (xbits.is_quiet_nan()) + return x; + + // Set domain error for non-NaN input. + if (!xbits.is_nan()) + fputil::set_errno_if_required(EDOM); + + fputil::raise_except_if_required(FE_INVALID); + return FPBits::quiet_nan().get_val(); + } + + // When |x| >= 0.5, we perform range reduction as follow: + // + // Assume further that 0.5 <= x < 1, and let: + // y = asin(x) + // We will use the double angle formula: + // cos(2y) = 1 - 2 sin^2(y) + // and the complement angle identity: + // x = sin(y) = cos(pi/2 - y) + // = 1 - 2 sin^2 (pi/4 - y/2) + // So: + // sin(pi/4 - y/2) = sqrt( (1 - x)/2 ) + // And hence: + // pi/4 - y/2 = asin( sqrt( (1 - x)/2 ) ) + // Equivalently: + // asin(x) = y = pi/2 - 2 * asin( sqrt( (1 - x)/2 ) ) + // Let u = (1 - x)/2, then: + // asin(x) = pi/2 - 2 * asin( sqrt(u) ) + // Moreover, since 0.5 <= x < 1: + // 0 < u <= 1/4, and 0 < sqrt(u) <= 0.5, + // And hence we can reuse the same polynomial approximation of asin(x) when + // |x| <= 0.5: + // asin(x) ~ pi/2 - 2 * sqrt(u) * P(u), + + // u = (1 - |x|)/2 + double u = fputil::multiply_add(x_abs, -0.5, 0.5); + // v_hi + v_lo ~ sqrt(u). + // Let: + // h = u - v_hi^2 = (sqrt(u) - v_hi) * (sqrt(u) + v_hi) + // Then: + // sqrt(u) = v_hi + h / (sqrt(u) + v_hi) + // ~ v_hi + h / (2 * v_hi) + // So we can use: + // v_lo = h / (2 * v_hi). + // Then, + // asin(x) ~ pi/2 - 2*(v_hi + v_lo) * P(u) + double v_hi = fputil::sqrt<double>(u); + +#ifdef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + double p = asin_eval(u); + double r = x_sign * fputil::multiply_add(-2.0 * v_hi, p, PI_OVER_TWO.hi); + return r; +#else + +#ifdef LIBC_TARGET_CPU_HAS_FMA_DOUBLE + double h = fputil::multiply_add(v_hi, -v_hi, u); +#else + DoubleDouble v_hi_sq = fputil::exact_mult(v_hi, v_hi); + double h = (u - v_hi_sq.hi) - v_hi_sq.lo; +#endif // LIBC_TARGET_CPU_HAS_FMA_DOUBLE + + // Scale v_lo and v_hi by 2 from the formula: + // vh = v_hi * 2 + // vl = 2*v_lo = h / v_hi. + double vh = v_hi * 2.0; + double vl = h / v_hi; + + // Polynomial approximation: + // p ~ asin(sqrt(u))/sqrt(u) + unsigned idx = 0; + double err = vh * 0x1.0p-51; + + DoubleDouble p = asin_eval(DoubleDouble{0.0, u}, idx, err); + + // Perform computations in double-double arithmetic: + // asin(x) = pi/2 - (v_hi + v_lo) * (ASIN_COEFFS[idx][0] + p) + DoubleDouble r0 = fputil::quick_mult(DoubleDouble{vl, vh}, p); + DoubleDouble r = fputil::exact_add(PI_OVER_TWO.hi, -r0.hi); + + double r_lo = PI_OVER_TWO.lo - r0.lo + r.lo; + + // Ziv's accuracy test. + +#ifdef LIBC_TARGET_CPU_HAS_FMA_DOUBLE + double r_upper = fputil::multiply_add( + r.hi, x_sign, fputil::multiply_add(r_lo, x_sign, err)); + double r_lower = fputil::multiply_add( + r.hi, x_sign, fputil::multiply_add(r_lo, x_sign, -err)); +#else + r_lo *= x_sign; + r.hi *= x_sign; + double r_upper = r.hi + (r_lo + err); + double r_lower = r.hi + (r_lo - err); +#endif // LIBC_TARGET_CPU_HAS_FMA_DOUBLE + + if (LIBC_LIKELY(r_upper == r_lower)) + return r_upper; + + // Ziv's accuracy test failed, we redo the computations in Float128. + // Recalculate mod 1/64. + idx = static_cast<unsigned>(fputil::nearest_integer(u * 0x1.0p6)); + + // After the first step of Newton-Raphson approximating v = sqrt(u), we have + // that: + // sqrt(u) = v_hi + h / (sqrt(u) + v_hi) + // v_lo = h / (2 * v_hi) + // With error: + // sqrt(u) - (v_hi + v_lo) = h * ( 1/(sqrt(u) + v_hi) - 1/(2*v_hi) ) + // = -h^2 / (2*v * (sqrt(u) + v)^2). + // Since: + // (sqrt(u) + v_hi)^2 ~ (2sqrt(u))^2 = 4u, + // we can add another correction term to (v_hi + v_lo) that is: + // v_ll = -h^2 / (2*v_hi * 4u) + // = -v_lo * (h / 4u) + // = -vl * (h / 8u), + // making the errors: + // sqrt(u) - (v_hi + v_lo + v_ll) = O(h^3) + // well beyond 128-bit precision needed. + + // Get the rounding error of vl = 2 * v_lo ~ h / vh + // Get full product of vh * vl +#ifdef LIBC_TARGET_CPU_HAS_FMA_DOUBLE + double vl_lo = fputil::multiply_add(-v_hi, vl, h) / v_hi; +#else + DoubleDouble vh_vl = fputil::exact_mult(v_hi, vl); + double vl_lo = ((h - vh_vl.hi) - vh_vl.lo) / v_hi; +#endif // LIBC_TARGET_CPU_HAS_FMA_DOUBLE + // vll = 2*v_ll = -vl * (h / (4u)). + double t = h * (-0.25) / u; + double vll = fputil::multiply_add(vl, t, vl_lo); + // m_v = -(v_hi + v_lo + v_ll). + Float128 m_v = fputil::quick_add( + Float128(vh), fputil::quick_add(Float128(vl), Float128(vll))); + m_v.sign = Sign::NEG; + + // Perform computations in Float128: + // asin(x) = pi/2 - (v_hi + v_lo + vll) * P(u). + Float128 y_f128(fputil::multiply_add(static_cast<double>(idx), -0x1.0p-6, u)); + + Float128 p_f128 = asin_eval(y_f128, idx); + Float128 r0_f128 = fputil::quick_mul(m_v, p_f128); + Float128 r_f128 = fputil::quick_add(PI_OVER_TWO_F128, r0_f128); + + if (xbits.is_neg()) + r_f128.sign = Sign::NEG; + + return static_cast<double>(r_f128); +#endif // LIBC_MATH_HAS_SKIP_ACCURATE_PASS +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ASIN_H diff --git a/libc/src/__support/math/asin_utils.h b/libc/src/__support/math/asin_utils.h index 3146444..e0c9096 100644 --- a/libc/src/__support/math/asin_utils.h +++ b/libc/src/__support/math/asin_utils.h @@ -19,7 +19,7 @@ namespace LIBC_NAMESPACE_DECL { -namespace { +namespace asin_internal { using DoubleDouble = fputil::DoubleDouble; using Float128 = fputil::DyadicFloat<128>; @@ -567,7 +567,7 @@ LIBC_INLINE static constexpr Float128 asin_eval(const Float128 &u, #endif // LIBC_MATH_HAS_SKIP_ACCURATE_PASS -} // anonymous namespace +} // namespace asin_internal } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/__support/math/asinf.h b/libc/src/__support/math/asinf.h new file mode 100644 index 0000000..bfa0dc3 --- /dev/null +++ b/libc/src/__support/math/asinf.h @@ -0,0 +1,175 @@ +//===-- Implementation header for asinf -------------------------*- 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_ASINF_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ASINF_H + +#include "inv_trigf_utils.h" +#include "src/__support/FPUtil/FEnvImpl.h" +#include "src/__support/FPUtil/FPBits.h" +#include "src/__support/FPUtil/except_value_utils.h" +#include "src/__support/FPUtil/multiply_add.h" +#include "src/__support/FPUtil/sqrt.h" +#include "src/__support/macros/config.h" +#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY +#include "src/__support/macros/properties/cpu_features.h" // LIBC_TARGET_CPU_HAS_FMA + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +LIBC_INLINE static constexpr float asinf(float x) { + using namespace inv_trigf_utils_internal; + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + constexpr size_t N_EXCEPTS = 2; + + // Exceptional values when |x| <= 0.5 + constexpr fputil::ExceptValues<float, N_EXCEPTS> ASINF_EXCEPTS_LO = {{ + // (inputs, RZ output, RU offset, RD offset, RN offset) + // x = 0x1.137f0cp-5, asinf(x) = 0x1.138c58p-5 (RZ) + {0x3d09bf86, 0x3d09c62c, 1, 0, 1}, + // x = 0x1.cbf43cp-4, asinf(x) = 0x1.cced1cp-4 (RZ) + {0x3de5fa1e, 0x3de6768e, 1, 0, 0}, + }}; + + // Exceptional values when 0.5 < |x| <= 1 + constexpr fputil::ExceptValues<float, N_EXCEPTS> ASINF_EXCEPTS_HI = {{ + // (inputs, RZ output, RU offset, RD offset, RN offset) + // x = 0x1.107434p-1, asinf(x) = 0x1.1f4b64p-1 (RZ) + {0x3f083a1a, 0x3f0fa5b2, 1, 0, 0}, + // x = 0x1.ee836cp-1, asinf(x) = 0x1.4f0654p0 (RZ) + {0x3f7741b6, 0x3fa7832a, 1, 0, 0}, + }}; +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + + using namespace inv_trigf_utils_internal; + using FPBits = typename fputil::FPBits<float>; + + FPBits xbits(x); + uint32_t x_uint = xbits.uintval(); + uint32_t x_abs = xbits.uintval() & 0x7fff'ffffU; + constexpr double SIGN[2] = {1.0, -1.0}; + uint32_t x_sign = x_uint >> 31; + + // |x| <= 0.5-ish + if (x_abs < 0x3f04'471dU) { + // |x| < 0x1.d12edp-12 + if (LIBC_UNLIKELY(x_abs < 0x39e8'9768U)) { + // When |x| < 2^-12, the relative error of the approximation asin(x) ~ x + // is: + // |asin(x) - x| / |asin(x)| < |x^3| / (6|x|) + // = x^2 / 6 + // < 2^-25 + // < epsilon(1)/2. + // So the correctly rounded values of asin(x) are: + // = x + sign(x)*eps(x) if rounding mode = FE_TOWARDZERO, + // or (rounding mode = FE_UPWARD and x is + // negative), + // = x otherwise. + // To simplify the rounding decision and make it more efficient, we use + // fma(x, 2^-25, x) instead. + // An exhaustive test shows that this formula work correctly for all + // rounding modes up to |x| < 0x1.d12edp-12. + // Note: to use the formula x + 2^-25*x to decide the correct rounding, we + // do need fma(x, 2^-25, x) to prevent underflow caused by 2^-25*x when + // |x| < 2^-125. For targets without FMA instructions, we simply use + // double for intermediate results as it is more efficient than using an + // emulated version of FMA. +#if defined(LIBC_TARGET_CPU_HAS_FMA_FLOAT) + return fputil::multiply_add(x, 0x1.0p-25f, x); +#else + double xd = static_cast<double>(x); + return static_cast<float>(fputil::multiply_add(xd, 0x1.0p-25, xd)); +#endif // LIBC_TARGET_CPU_HAS_FMA_FLOAT + } + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + // Check for exceptional values + if (auto r = ASINF_EXCEPTS_LO.lookup_odd(x_abs, x_sign); + LIBC_UNLIKELY(r.has_value())) + return r.value(); +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + + // For |x| <= 0.5, we approximate asinf(x) by: + // asin(x) = x * P(x^2) + // Where P(X^2) = Q(X) is a degree-20 minimax even polynomial approximating + // asin(x)/x on [0, 0.5] generated by Sollya with: + // > Q = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20|], + // [|1, D...|], [0, 0.5]); + // An exhaustive test shows that this approximation works well up to a + // little more than 0.5. + double xd = static_cast<double>(x); + double xsq = xd * xd; + double x3 = xd * xsq; + double r = asin_eval(xsq); + return static_cast<float>(fputil::multiply_add(x3, r, xd)); + } + + // |x| > 1, return NaNs. + if (LIBC_UNLIKELY(x_abs > 0x3f80'0000U)) { + if (xbits.is_signaling_nan()) { + fputil::raise_except_if_required(FE_INVALID); + return FPBits::quiet_nan().get_val(); + } + + if (x_abs <= 0x7f80'0000U) { + fputil::set_errno_if_required(EDOM); + fputil::raise_except_if_required(FE_INVALID); + } + + return FPBits::quiet_nan().get_val(); + } + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + // Check for exceptional values + if (auto r = ASINF_EXCEPTS_HI.lookup_odd(x_abs, x_sign); + LIBC_UNLIKELY(r.has_value())) + return r.value(); +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + + // When |x| > 0.5, we perform range reduction as follow: + // + // Assume further that 0.5 < x <= 1, and let: + // y = asin(x) + // We will use the double angle formula: + // cos(2y) = 1 - 2 sin^2(y) + // and the complement angle identity: + // x = sin(y) = cos(pi/2 - y) + // = 1 - 2 sin^2 (pi/4 - y/2) + // So: + // sin(pi/4 - y/2) = sqrt( (1 - x)/2 ) + // And hence: + // pi/4 - y/2 = asin( sqrt( (1 - x)/2 ) ) + // Equivalently: + // asin(x) = y = pi/2 - 2 * asin( sqrt( (1 - x)/2 ) ) + // Let u = (1 - x)/2, then: + // asin(x) = pi/2 - 2 * asin( sqrt(u) ) + // Moreover, since 0.5 < x <= 1: + // 0 <= u < 1/4, and 0 <= sqrt(u) < 0.5, + // And hence we can reuse the same polynomial approximation of asin(x) when + // |x| <= 0.5: + // asin(x) ~ pi/2 - 2 * sqrt(u) * P(u), + + xbits.set_sign(Sign::POS); + double sign = SIGN[x_sign]; + double xd = static_cast<double>(xbits.get_val()); + double u = fputil::multiply_add(-0.5, xd, 0.5); + double c1 = sign * (-2 * fputil::sqrt<double>(u)); + double c2 = fputil::multiply_add(sign, M_MATH_PI_2, c1); + double c3 = c1 * u; + + double r = asin_eval(u); + return static_cast<float>(fputil::multiply_add(c3, r, c2)); +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ASINF_H diff --git a/libc/src/__support/math/asinf16.h b/libc/src/__support/math/asinf16.h new file mode 100644 index 0000000..3d032a4 --- /dev/null +++ b/libc/src/__support/math/asinf16.h @@ -0,0 +1,146 @@ +//===-- Implementation header for asinf16 -----------------------*- 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_ASINF16_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ASINF16_H + +#include "include/llvm-libc-macros/float16-macros.h" + +#ifdef LIBC_TYPES_HAS_FLOAT16 + +#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/multiply_add.h" +#include "src/__support/FPUtil/sqrt.h" +#include "src/__support/macros/optimization.h" + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +LIBC_INLINE static constexpr float16 asinf16(float16 x) { + + // Generated by Sollya using the following command: + // > round(pi/2, D, RN); + constexpr float PI_2 = 0x1.921fb54442d18p0f; + + using FPBits = fputil::FPBits<float16>; + FPBits xbits(x); + + uint16_t x_u = xbits.uintval(); + uint16_t x_abs = x_u & 0x7fff; + float xf = x; + + // |x| > 0x1p0, |x| > 1, or x is NaN. + if (LIBC_UNLIKELY(x_abs > 0x3c00)) { + // asinf16(NaN) = 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; + } + + // 1 < |x| <= +/-inf + fputil::raise_except_if_required(FE_INVALID); + fputil::set_errno_if_required(EDOM); + + return FPBits::quiet_nan().get_val(); + } + + float xsq = xf * xf; + + // |x| <= 0x1p-1, |x| <= 0.5 + if (x_abs <= 0x3800) { + // asinf16(+/-0) = +/-0 + if (LIBC_UNLIKELY(x_abs == 0)) + return x; + + // Exhaustive tests show that, + // for |x| <= 0x1.878p-9, when: + // x > 0, and rounding upward, or + // x < 0, and rounding downward, then, + // asin(x) = x * 2^-11 + x + // else, in other rounding modes, + // asin(x) = x + if (LIBC_UNLIKELY(x_abs <= 0x1a1e)) { + int rounding = fputil::quick_get_round(); + + if ((xbits.is_pos() && rounding == FE_UPWARD) || + (xbits.is_neg() && rounding == FE_DOWNWARD)) + return fputil::cast<float16>(fputil::multiply_add(xf, 0x1.0p-11f, xf)); + return x; + } + + // Degree-6 minimax odd polynomial of asin(x) generated by Sollya with: + // > P = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 0.5]); + float result = + fputil::polyeval(xsq, 0x1.000002p0f, 0x1.554c2ap-3f, 0x1.3541ccp-4f, + 0x1.43b2d6p-5f, 0x1.a0d73ep-5f); + return fputil::cast<float16>(xf * result); + } + + // When |x| > 0.5, assume that 0.5 < |x| <= 1, + // + // Step-by-step range-reduction proof: + // 1: Let y = asin(x), such that, x = sin(y) + // 2: From complimentary angle identity: + // x = sin(y) = cos(pi/2 - y) + // 3: Let z = pi/2 - y, such that x = cos(z) + // 4: From double angle formula; cos(2A) = 1 - sin^2(A): + // z = 2A, z/2 = A + // cos(z) = 1 - 2 * sin^2(z/2) + // 5: Make sin(z/2) subject of the formula: + // sin(z/2) = sqrt((1 - cos(z))/2) + // 6: Recall [3]; x = cos(z). Therefore: + // sin(z/2) = sqrt((1 - x)/2) + // 7: Let u = (1 - x)/2 + // 8: Therefore: + // asin(sqrt(u)) = z/2 + // 2 * asin(sqrt(u)) = z + // 9: Recall [3], z = pi/2 - y. Therefore: + // y = pi/2 - z + // y = pi/2 - 2 * asin(sqrt(u)) + // 10: Recall [1], y = asin(x). Therefore: + // asin(x) = pi/2 - 2 * asin(sqrt(u)) + // + // WHY? + // 11: Recall [7], u = (1 - x)/2 + // 12: Since 0.5 < x <= 1, therefore: + // 0 <= u <= 0.25 and 0 <= sqrt(u) <= 0.5 + // + // Hence, we can reuse the same [0, 0.5] domain polynomial approximation for + // Step [10] as `sqrt(u)` is in range. + + // 0x1p-1 < |x| <= 0x1p0, 0.5 < |x| <= 1.0 + float xf_abs = (xf < 0 ? -xf : xf); + float sign = (xbits.uintval() >> 15 == 1 ? -1.0 : 1.0); + float u = fputil::multiply_add(-0.5f, xf_abs, 0.5f); + float u_sqrt = fputil::sqrt<float>(u); + + // Degree-6 minimax odd polynomial of asin(x) generated by Sollya with: + // > P = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 0.5]); + float asin_sqrt_u = + u_sqrt * fputil::polyeval(u, 0x1.000002p0f, 0x1.554c2ap-3f, + 0x1.3541ccp-4f, 0x1.43b2d6p-5f, 0x1.a0d73ep-5f); + + return fputil::cast<float16>(sign * + fputil::multiply_add(-2.0f, asin_sqrt_u, PI_2)); +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LIBC_TYPES_HAS_FLOAT16 + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ASINF16_H diff --git a/libc/src/__support/math/erff.h b/libc/src/__support/math/erff.h new file mode 100644 index 0000000..b81be30 --- /dev/null +++ b/libc/src/__support/math/erff.h @@ -0,0 +1,193 @@ +//===-- Implementation header for erff --------------------------*- 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_ERFF_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_ERFF_H + +#include "src/__support/FPUtil/FPBits.h" +#include "src/__support/FPUtil/except_value_utils.h" +#include "src/__support/FPUtil/multiply_add.h" +#include "src/__support/macros/config.h" +#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY + +namespace LIBC_NAMESPACE_DECL { + +namespace math { + +LIBC_INLINE static constexpr float erff(float x) { + + // Polynomials approximating erf(x)/x on ( k/8, (k + 1)/8 ) generated by + // Sollya with: > P = fpminimax(erf(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14|], + // [|D...|], + // [k/8, (k + 1)/8]); + // for k = 0..31. + constexpr double COEFFS[32][8] = { + {0x1.20dd750429b6dp0, -0x1.812746b037753p-2, 0x1.ce2f219e8596ap-4, + -0x1.b82cdacb78fdap-6, 0x1.56479297dfda5p-8, -0x1.8b3ac5455ef02p-11, + -0x1.126fcac367e3bp-8, 0x1.2d0bdb3ba4984p-4}, + {0x1.20dd750429b6dp0, -0x1.812746b0379a8p-2, 0x1.ce2f21a03cf2ap-4, + -0x1.b82ce30de083ep-6, 0x1.565bcad3eb60fp-8, -0x1.c02c66f659256p-11, + 0x1.f92f673385229p-14, -0x1.def402648ae9p-17}, + {0x1.20dd750429b34p0, -0x1.812746b032dcep-2, 0x1.ce2f219d84aaep-4, + -0x1.b82ce22dcf139p-6, 0x1.565b9efcd4af1p-8, -0x1.c021f1af414bcp-11, + 0x1.f7c6d177eff82p-14, -0x1.c9e4410dcf865p-17}, + {0x1.20dd750426eabp0, -0x1.812746ae592c7p-2, 0x1.ce2f211525f14p-4, + -0x1.b82ccc125e63fp-6, 0x1.56596f261cfd3p-8, -0x1.bfde1ff8eeecfp-11, + 0x1.f31a9d15dc5d8p-14, -0x1.a5a4362844b3cp-17}, + {0x1.20dd75039c705p0, -0x1.812746777e74dp-2, 0x1.ce2f17af98a1bp-4, + -0x1.b82be4b817cbep-6, 0x1.564bec2e2962ep-8, -0x1.bee86f9da3558p-11, + 0x1.e9443689dc0ccp-14, -0x1.79c0f230805d8p-17}, + {0x1.20dd74f811211p0, -0x1.81274371a3e8fp-2, 0x1.ce2ec038262e5p-4, + -0x1.b8265b82c5e1fp-6, 0x1.5615a2e239267p-8, -0x1.bc63ae023dcebp-11, + 0x1.d87c2102f7e06p-14, -0x1.49584bea41d62p-17}, + {0x1.20dd746d063e3p0, -0x1.812729a8a950fp-2, 0x1.ce2cb0a2df232p-4, + -0x1.b80eca1f51278p-6, 0x1.5572e26c46815p-8, -0x1.b715e5638b65ep-11, + 0x1.bfbb195484968p-14, -0x1.177a565c15c52p-17}, + {0x1.20dd701b44486p0, -0x1.812691145f237p-2, 0x1.ce23a06b8cfd9p-4, + -0x1.b7c1dc7245288p-6, 0x1.53e92f7f397ddp-8, -0x1.ad97cc4acf0b2p-11, + 0x1.9f028b2b09b71p-14, -0x1.cdc4da08da8c1p-18}, + {0x1.20dd5715ac332p0, -0x1.8123e680bd0ebp-2, 0x1.ce0457aded691p-4, + -0x1.b6f52d52bed4p-6, 0x1.50c291b84414cp-8, -0x1.9ea246b1ad4a9p-11, + 0x1.77654674e0cap-14, -0x1.737c11a1bcebbp-18}, + {0x1.20dce6593e114p0, -0x1.811a59c02eadcp-2, 0x1.cdab53c7cd7d5p-4, + -0x1.b526d2e321eedp-6, 0x1.4b1d32cd8b994p-8, -0x1.8963143ec0a1ep-11, + 0x1.4ad5700e4db91p-14, -0x1.231e100e43ef2p-18}, + {0x1.20db48bfd5a62p0, -0x1.80fdd84f9e308p-2, 0x1.ccd340d462983p-4, + -0x1.b196a2928768p-6, 0x1.4210c2c13a0f7p-8, -0x1.6dbdfb4ff71aep-11, + 0x1.1bca2d17fbd71p-14, -0x1.bca36f90c7cf5p-19}, + {0x1.20d64b2f8f508p0, -0x1.80b4d4f19fa8bp-2, 0x1.cb088197262e3p-4, + -0x1.ab51fd02e5b99p-6, 0x1.34e1e5e81a632p-8, -0x1.4c66377b502cep-11, + 0x1.d9ad25066213cp-15, -0x1.4b0df7dd0cfa1p-19}, + {0x1.20c8fc1243576p0, -0x1.8010cb2009e27p-2, 0x1.c7a47e9299315p-4, + -0x1.a155be5683654p-6, 0x1.233502694997bp-8, -0x1.26c94b7d813p-11, + 0x1.8094f1de25fb9p-15, -0x1.e0e3d776c6eefp-20}, + {0x1.20a9bd1611bc1p0, -0x1.7ec7fbce83f9p-2, 0x1.c1d757d7317b7p-4, + -0x1.92c160cd589fp-6, 0x1.0d307269cc5c2p-8, -0x1.fda5b0d2d1879p-12, + 0x1.2fdd7b3b14a7fp-15, -0x1.54eed4a26af5ap-20}, + {0x1.20682834f943dp0, -0x1.7c73f747bf5a9p-2, 0x1.b8c2db4a9ffd1p-4, + -0x1.7f0e4ffe989ecp-6, 0x1.e7061eae4166ep-9, -0x1.ad36e873fff2dp-12, + 0x1.d39222396128ep-16, -0x1.d83dacec5ea6bp-21}, + {0x1.1feb8d12676d7p0, -0x1.7898347284afep-2, 0x1.aba3466b34451p-4, + -0x1.663adc573e2f9p-6, 0x1.ae99fb17c3e08p-9, -0x1.602f950ad5535p-12, + 0x1.5e9717490609dp-16, -0x1.3fca107bbc8d5p-21}, + {0x1.1f12fe3c536fap0, -0x1.72b1d1f22e6d3p-2, 0x1.99fc0eed4a896p-4, + -0x1.48db0a87bd8c6p-6, 0x1.73e368895aa61p-9, -0x1.19b35d5301fc8p-12, + 0x1.007987e4bb033p-16, -0x1.a7edcd4c2dc7p-22}, + {0x1.1db7b0df84d5dp0, -0x1.6a4e4a41cde02p-2, 0x1.83bbded16455dp-4, + -0x1.2809b3b36977ep-6, 0x1.39c08bab44679p-9, -0x1.b7b45a70ed119p-13, + 0x1.6e99b36410e7bp-17, -0x1.13619bb7ebc0cp-22}, + {0x1.1bb1c85c4a527p0, -0x1.5f23b99a249a3p-2, 0x1.694c91fa0d12cp-4, + -0x1.053e1ce11c72dp-6, 0x1.02bf72c50ea78p-9, -0x1.4f478fb56cb02p-13, + 0x1.005f80ecbe213p-17, -0x1.5f2446bde7f5bp-23}, + {0x1.18dec3bd51f9dp0, -0x1.5123f58346186p-2, 0x1.4b8a1ca536ab4p-4, + -0x1.c4243015cc723p-7, 0x1.a1a8a01d351efp-10, -0x1.f466b34f1d86bp-14, + 0x1.5f835eea0bf6ap-18, -0x1.b83165b939234p-24}, + {0x1.152804c3369f4p0, -0x1.4084cd4afd4bcp-2, 0x1.2ba2e836e47aap-4, + -0x1.800f2dfc6904bp-7, 0x1.4a6daf0669c59p-10, -0x1.6e326ab872317p-14, + 0x1.d9761a6a755a5p-19, -0x1.0fca33f9dd4b5p-24}, + {0x1.1087ad68356aap0, -0x1.2dbb044707459p-2, 0x1.0aea8ceaa0384p-4, + -0x1.40b516d52b3d2p-7, 0x1.00c9e05f01d22p-10, -0x1.076afb0dc0ff7p-14, + 0x1.39fadec400657p-19, -0x1.4b5761352e7e3p-25}, + {0x1.0b0a7a8ba4a22p0, -0x1.196990d22d4a1p-2, 0x1.d5551e6ac0c4dp-5, + -0x1.07cce1770bd1ap-7, 0x1.890347b8848bfp-11, -0x1.757ec96750b6ap-15, + 0x1.9b258a1e06bcep-20, -0x1.8fc6d22da7572p-26}, + {0x1.04ce2be70fb47p0, -0x1.0449e4b0b9cacp-2, 0x1.97f7424f4b0e7p-5, + -0x1.ac825439c42f4p-8, 0x1.28f5f65426dfbp-11, -0x1.05b699a90f90fp-15, + 0x1.0a888eecf4593p-20, -0x1.deace2b32bb31p-27}, + {0x1.fbf9fb0e11cc8p-1, -0x1.de2640856545ap-3, 0x1.5f5b1f47f851p-5, + -0x1.588bc71eb41b9p-8, 0x1.bc6a0a772f56dp-12, -0x1.6b9fad1f1657ap-16, + 0x1.573204ba66504p-21, -0x1.1d38065c94e44p-27}, + {0x1.ed8f18c99e031p-1, -0x1.b4cb6acd903b4p-3, 0x1.2c7f3dddd6fc1p-5, + -0x1.13052067df4ep-8, 0x1.4a5027444082fp-12, -0x1.f672bab0e2554p-17, + 0x1.b83c756348cc9p-22, -0x1.534f1a1079499p-28}, + {0x1.debd33044166dp-1, -0x1.8d7cd9053f7d8p-3, 0x1.ff9957fb3d6e7p-6, + -0x1.b50be55de0f36p-9, 0x1.e92c8ec53a628p-13, -0x1.5a4b88d508007p-17, + 0x1.1a27737559e26p-22, -0x1.942ae62cb2c14p-29}, + {0x1.cfdbf0386f3bdp-1, -0x1.68e33d93b0dc4p-3, 0x1.b2683d58f53dep-6, + -0x1.5a9174e70d26fp-9, 0x1.69ddd326d49cdp-13, -0x1.dd8f397a8219cp-18, + 0x1.6a755016ad4ddp-23, -0x1.e366e0139187dp-30}, + {0x1.c132adb8d7464p-1, -0x1.475a899f61b46p-3, 0x1.70a431397a77cp-6, + -0x1.12e3d35beeee2p-9, 0x1.0c16b05738333p-13, -0x1.4a47f873e144ep-18, + 0x1.d3d494c698c02p-24, -0x1.2302c59547fe5p-30}, + {0x1.b2f5fd05555e7p-1, -0x1.28feefbe03ec7p-3, 0x1.3923acbb3a676p-6, + -0x1.b4ff793cd6358p-10, 0x1.8ea0eb8c913bcp-14, -0x1.cb31ec2baceb1p-19, + 0x1.30011e7e80c04p-24, -0x1.617710635cb1dp-31}, + {0x1.a54853cd9593ep-1, -0x1.0dbdbaea4dc8ep-3, 0x1.0a93e2c20a0fdp-6, + -0x1.5c969ff401ea8p-10, 0x1.29e0cc64fe627p-14, -0x1.4160d8e9d3c2ap-19, + 0x1.8e7b67594624ap-25, -0x1.b1cf2c975b09bp-32}, + {0x1.983ceece09ff8p-1, -0x1.eacc78f7a2dp-4, 0x1.c74418410655fp-7, + -0x1.1756a050e441ep-10, 0x1.bff3650f7f548p-15, -0x1.c56c0217d3adap-20, + 0x1.07b4918d0b489p-25, -0x1.0d4be8c1c50f8p-32}, + }; + + using FPBits = typename fputil::FPBits<float>; + FPBits xbits(x); + + uint32_t x_u = xbits.uintval(); + uint32_t x_abs = x_u & 0x7fff'ffffU; + + if (LIBC_UNLIKELY(x_abs >= 0x4080'0000U)) { + constexpr float ONE[2] = {1.0f, -1.0f}; + constexpr float SMALL[2] = {-0x1.0p-25f, 0x1.0p-25f}; + + int sign = xbits.is_neg() ? 1 : 0; + + if (LIBC_UNLIKELY(x_abs >= 0x7f80'0000U)) { + if (xbits.is_signaling_nan()) { + fputil::raise_except_if_required(FE_INVALID); + return FPBits::quiet_nan().get_val(); + } + return (x_abs > 0x7f80'0000) ? x : ONE[sign]; + } + + return ONE[sign] + SMALL[sign]; + } + +#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS + // Exceptional mask = common 0 bits of 2 exceptional values. + constexpr uint32_t EXCEPT_MASK = 0x809a'6184U; + + if (LIBC_UNLIKELY((x_abs & EXCEPT_MASK) == 0)) { + // Exceptional values + if (LIBC_UNLIKELY(x_abs == 0x3f65'9229U)) // |x| = 0x1.cb2452p-1f + return x < 0.0f ? fputil::round_result_slightly_down(-0x1.972ea8p-1f) + : fputil::round_result_slightly_up(0x1.972ea8p-1f); + if (LIBC_UNLIKELY(x_abs == 0x4004'1e6aU)) // |x| = 0x1.083cd4p+1f + return x < 0.0f ? fputil::round_result_slightly_down(-0x1.fe3462p-1f) + : fputil::round_result_slightly_up(0x1.fe3462p-1f); + if (x_abs == 0U) + return x; + } +#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS + + // Polynomial approximation: + // erf(x) ~ x * (c0 + c1 * x^2 + c2 * x^4 + ... + c7 * x^14) + double xd = static_cast<double>(x); + double xsq = xd * xd; + + constexpr uint32_t EIGHT = 3 << FPBits::FRACTION_LEN; + int idx = static_cast<int>(FPBits(x_abs + EIGHT).get_val()); + + double x4 = xsq * xsq; + double c0 = fputil::multiply_add(xsq, COEFFS[idx][1], COEFFS[idx][0]); + double c1 = fputil::multiply_add(xsq, COEFFS[idx][3], COEFFS[idx][2]); + double c2 = fputil::multiply_add(xsq, COEFFS[idx][5], COEFFS[idx][4]); + double c3 = fputil::multiply_add(xsq, COEFFS[idx][7], COEFFS[idx][6]); + + double x8 = x4 * x4; + double p0 = fputil::multiply_add(x4, c1, c0); + double p1 = fputil::multiply_add(x4, c3, c2); + + return static_cast<float>(xd * fputil::multiply_add(x8, p1, p0)); +} + +} // namespace math + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_ERFF_H diff --git a/libc/src/__support/math/exp.h b/libc/src/__support/math/exp.h index ff59ff7..83638e8 100644 --- a/libc/src/__support/math/exp.h +++ b/libc/src/__support/math/exp.h @@ -67,7 +67,7 @@ namespace { // Return expm1(dx) / x ~ 1 + dx / 2 + dx^2 / 6 + dx^3 / 24. // For |dx| < 2^-13 + 2^-30: // | output - expm1(dx) / dx | < 2^-51. -static double poly_approx_d(double dx) { +LIBC_INLINE static double poly_approx_d(double dx) { // dx^2 double dx2 = dx * dx; // c0 = 1 + dx / 2 @@ -85,7 +85,7 @@ static double poly_approx_d(double dx) { // Return exp(dx) ~ 1 + dx + dx^2 / 2 + ... + dx^6 / 720 // For |dx| < 2^-13 + 2^-30: // | output - exp(dx) | < 2^-101 -static DoubleDouble poly_approx_dd(const DoubleDouble &dx) { +LIBC_INLINE static DoubleDouble poly_approx_dd(const DoubleDouble &dx) { // Taylor polynomial. constexpr DoubleDouble COEFFS[] = { {0, 0x1p0}, // 1 @@ -106,7 +106,7 @@ static DoubleDouble poly_approx_dd(const DoubleDouble &dx) { // Return exp(dx) ~ 1 + dx + dx^2 / 2 + ... + dx^7 / 5040 // For |dx| < 2^-13 + 2^-30: // | output - exp(dx) | < 2^-126. -static Float128 poly_approx_f128(const Float128 &dx) { +LIBC_INLINE static Float128 poly_approx_f128(const Float128 &dx) { constexpr Float128 COEFFS_128[]{ {Sign::POS, -127, 0x80000000'00000000'00000000'00000000_u128}, // 1.0 {Sign::POS, -127, 0x80000000'00000000'00000000'00000000_u128}, // 1.0 @@ -127,7 +127,7 @@ static Float128 poly_approx_f128(const Float128 &dx) { // Compute exp(x) using 128-bit precision. // TODO(lntue): investigate triple-double precision implementation for this // step. -static Float128 exp_f128(double x, double kd, int idx1, int idx2) { +LIBC_INLINE static Float128 exp_f128(double x, double kd, int idx1, int idx2) { // Recalculate dx: double t1 = fputil::multiply_add(kd, MLOG_2_EXP2_M12_HI, x); // exact @@ -160,8 +160,8 @@ static Float128 exp_f128(double x, double kd, int idx1, int idx2) { } // Compute exp(x) with double-double precision. -static DoubleDouble exp_double_double(double x, double kd, - const DoubleDouble &exp_mid) { +LIBC_INLINE static DoubleDouble exp_double_double(double x, double kd, + const DoubleDouble &exp_mid) { // Recalculate dx: // dx = x - k * 2^-12 * log(2) double t1 = fputil::multiply_add(kd, MLOG_2_EXP2_M12_HI, x); // exact @@ -184,7 +184,7 @@ static DoubleDouble exp_double_double(double x, double kd, // Check for exceptional cases when // |x| <= 2^-53 or x < log(2^-1075) or x >= 0x1.6232bdd7abcd3p+9 -static double set_exceptional(double x) { +LIBC_INLINE static double set_exceptional(double x) { using FPBits = typename fputil::FPBits<double>; FPBits xbits(x); @@ -234,7 +234,7 @@ static double set_exceptional(double x) { namespace math { -static double exp(double x) { +LIBC_INLINE static double exp(double x) { using FPBits = typename fputil::FPBits<double>; FPBits xbits(x); diff --git a/libc/src/__support/math/exp10.h b/libc/src/__support/math/exp10.h index fa60e40c..12a09d7 100644 --- a/libc/src/__support/math/exp10.h +++ b/libc/src/__support/math/exp10.h @@ -83,7 +83,8 @@ LIBC_INLINE static double exp10_poly_approx_d(double dx) { // > P = fpminimax((10^x - 1)/x, 5, [|DD...|], [-2^-14, 2^-14]); // Error bounds: // | output - 10^(dx) | < 2^-101 -static constexpr DoubleDouble exp10_poly_approx_dd(const DoubleDouble &dx) { +LIBC_INLINE static constexpr DoubleDouble +exp10_poly_approx_dd(const DoubleDouble &dx) { // Taylor polynomial. constexpr DoubleDouble COEFFS[] = { {0, 0x1p0}, @@ -105,7 +106,8 @@ static constexpr DoubleDouble exp10_poly_approx_dd(const DoubleDouble &dx) { // Return exp(dx) ~ 1 + a0 * dx + a1 * dx^2 + ... + a6 * dx^7 // For |dx| < 2^-14: // | output - 10^dx | < 1.5 * 2^-124. -static constexpr Float128 exp10_poly_approx_f128(const Float128 &dx) { +LIBC_INLINE static constexpr Float128 +exp10_poly_approx_f128(const Float128 &dx) { constexpr Float128 COEFFS_128[]{ {Sign::POS, -127, 0x80000000'00000000'00000000'00000000_u128}, // 1.0 {Sign::POS, -126, 0x935d8ddd'aaa8ac16'ea56d62b'82d30a2d_u128}, @@ -126,7 +128,8 @@ static constexpr Float128 exp10_poly_approx_f128(const Float128 &dx) { // Compute 10^(x) using 128-bit precision. // TODO(lntue): investigate triple-double precision implementation for this // step. -static Float128 exp10_f128(double x, double kd, int idx1, int idx2) { +LIBC_INLINE static Float128 exp10_f128(double x, double kd, int idx1, + int idx2) { double t1 = fputil::multiply_add(kd, MLOG10_2_EXP2_M12_HI, x); // exact double t2 = kd * MLOG10_2_EXP2_M12_MID_32; // exact double t3 = kd * MLOG10_2_EXP2_M12_LO; // Error < 2^-144 @@ -157,8 +160,8 @@ static Float128 exp10_f128(double x, double kd, int idx1, int idx2) { } // Compute 10^x with double-double precision. -static DoubleDouble exp10_double_double(double x, double kd, - const DoubleDouble &exp_mid) { +LIBC_INLINE static DoubleDouble +exp10_double_double(double x, double kd, const DoubleDouble &exp_mid) { // Recalculate dx: // dx = x - k * 2^-12 * log10(2) double t1 = fputil::multiply_add(kd, MLOG10_2_EXP2_M12_HI, x); // exact @@ -180,7 +183,7 @@ static DoubleDouble exp10_double_double(double x, double kd, #endif // LIBC_MATH_HAS_SKIP_ACCURATE_PASS // When output is denormal. -static double exp10_denorm(double x) { +LIBC_INLINE static double exp10_denorm(double x) { // Range reduction. double tmp = fputil::multiply_add(x, LOG2_10, 0x1.8000'0000'4p21); int k = static_cast<int>(cpp::bit_cast<uint64_t>(tmp) >> 19); @@ -234,7 +237,7 @@ static double exp10_denorm(double x) { // * x >= log10(2^1024) // * x <= log10(2^-1022) // * x is inf or nan -static constexpr double exp10_set_exceptional(double x) { +LIBC_INLINE static constexpr double exp10_set_exceptional(double x) { using FPBits = typename fputil::FPBits<double>; FPBits xbits(x); @@ -285,7 +288,7 @@ static constexpr double exp10_set_exceptional(double x) { namespace math { -static constexpr double exp10(double x) { +LIBC_INLINE static constexpr double exp10(double x) { using FPBits = typename fputil::FPBits<double>; FPBits xbits(x); diff --git a/libc/src/__support/math/exp10_float16_constants.h b/libc/src/__support/math/exp10_float16_constants.h index f5928db..ef50185 100644 --- a/libc/src/__support/math/exp10_float16_constants.h +++ b/libc/src/__support/math/exp10_float16_constants.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_MATH_EXP10_FLOAT16_CONSTANTS_H #define LLVM_LIBC_SRC___SUPPORT_MATH_EXP10_FLOAT16_CONSTANTS_H +#include "hdr/stdint_proxy.h" #include "include/llvm-libc-macros/float16-macros.h" -#include <stdint.h> #ifdef LIBC_TYPES_HAS_FLOAT16 diff --git a/libc/src/__support/math/exp10f.h b/libc/src/__support/math/exp10f.h index 807b4f0..76ae197 100644 --- a/libc/src/__support/math/exp10f.h +++ b/libc/src/__support/math/exp10f.h @@ -20,7 +20,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float exp10f(float x) { +LIBC_INLINE static constexpr float exp10f(float x) { using FPBits = typename fputil::FPBits<float>; FPBits xbits(x); diff --git a/libc/src/__support/math/exp10f16.h b/libc/src/__support/math/exp10f16.h index 0d8b125..3eca867 100644 --- a/libc/src/__support/math/exp10f16.h +++ b/libc/src/__support/math/exp10f16.h @@ -57,7 +57,7 @@ static constexpr fputil::ExceptValues<float16, N_EXP10F16_EXCEPTS> }}; #endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS -static constexpr float16 exp10f16(float16 x) { +LIBC_INLINE static constexpr float16 exp10f16(float16 x) { using FPBits = fputil::FPBits<float16>; FPBits x_bits(x); diff --git a/libc/src/__support/math/exp10f16_utils.h b/libc/src/__support/math/exp10f16_utils.h index bffb81b..5952a41 100644 --- a/libc/src/__support/math/exp10f16_utils.h +++ b/libc/src/__support/math/exp10f16_utils.h @@ -19,8 +19,7 @@ namespace LIBC_NAMESPACE_DECL { -LIBC_INLINE static constexpr ExpRangeReduction -exp10_range_reduction(float16 x) { +LIBC_INLINE static ExpRangeReduction exp10_range_reduction(float16 x) { // For -8 < x < 5, to compute 10^x, we perform the following range reduction: // find hi, mid, lo, such that: // x = (hi + mid) * log2(10) + lo, in which diff --git a/libc/src/__support/math/exp10f_utils.h b/libc/src/__support/math/exp10f_utils.h index c30def9..010a2f1 100644 --- a/libc/src/__support/math/exp10f_utils.h +++ b/libc/src/__support/math/exp10f_utils.h @@ -89,7 +89,7 @@ struct Exp10Base : public ExpBase { 0x1.0470591dff149p1, 0x1.2bd7c0a9fbc4dp0, 0x1.1429e74a98f43p-1}; - static double powb_lo(double dx) { + LIBC_INLINE static double powb_lo(double dx) { using fputil::multiply_add; double dx2 = dx * dx; // c0 = 1 + COEFFS[0] * dx diff --git a/libc/src/__support/math/exp_utils.h b/libc/src/__support/math/exp_utils.h index fc9ab10..ef408ed 100644 --- a/libc/src/__support/math/exp_utils.h +++ b/libc/src/__support/math/exp_utils.h @@ -22,8 +22,8 @@ namespace LIBC_NAMESPACE_DECL { // So if we scale x up by 2^1022, we can use // double(1.0 + 2^1022 * x) - 1.0 to test how x is rounded in denormal range. template <bool SKIP_ZIV_TEST = false> -static constexpr cpp::optional<double> ziv_test_denorm(int hi, double mid, - double lo, double err) { +LIBC_INLINE static constexpr cpp::optional<double> +ziv_test_denorm(int hi, double mid, double lo, double err) { using FPBits = typename fputil::FPBits<double>; // Scaling factor = 1/(min normal number) = 2^1022 diff --git a/libc/src/__support/math/expf.h b/libc/src/__support/math/expf.h index 88c1514..f7e11be 100644 --- a/libc/src/__support/math/expf.h +++ b/libc/src/__support/math/expf.h @@ -24,7 +24,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float expf(float x) { +LIBC_INLINE static constexpr float expf(float x) { using FPBits = typename fputil::FPBits<float>; FPBits xbits(x); diff --git a/libc/src/__support/math/expf16.h b/libc/src/__support/math/expf16.h index ded28c7..14302a7 100644 --- a/libc/src/__support/math/expf16.h +++ b/libc/src/__support/math/expf16.h @@ -31,7 +31,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float16 expf16(float16 x) { +LIBC_INLINE 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) diff --git a/libc/src/__support/math/expf16_utils.h b/libc/src/__support/math/expf16_utils.h index 8a2fc94..4204dab7 100644 --- a/libc/src/__support/math/expf16_utils.h +++ b/libc/src/__support/math/expf16_utils.h @@ -47,7 +47,8 @@ struct ExpRangeReduction { float exp_lo; }; -[[maybe_unused]] static ExpRangeReduction exp_range_reduction(float16 x) { +[[maybe_unused]] LIBC_INLINE static 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 diff --git a/libc/src/__support/math/frexpf.h b/libc/src/__support/math/frexpf.h index 4d2f494..7834a12 100644 --- a/libc/src/__support/math/frexpf.h +++ b/libc/src/__support/math/frexpf.h @@ -17,7 +17,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float frexpf(float x, int *exp) { +LIBC_INLINE static constexpr float frexpf(float x, int *exp) { return fputil::frexp(x, *exp); } diff --git a/libc/src/__support/math/frexpf128.h b/libc/src/__support/math/frexpf128.h index 2fd5bc4..5218b26 100644 --- a/libc/src/__support/math/frexpf128.h +++ b/libc/src/__support/math/frexpf128.h @@ -21,7 +21,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float128 frexpf128(float128 x, int *exp) { +LIBC_INLINE static constexpr float128 frexpf128(float128 x, int *exp) { return fputil::frexp(x, *exp); } diff --git a/libc/src/__support/math/frexpf16.h b/libc/src/__support/math/frexpf16.h index 8deeba0..530b61a 100644 --- a/libc/src/__support/math/frexpf16.h +++ b/libc/src/__support/math/frexpf16.h @@ -21,7 +21,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float16 frexpf16(float16 x, int *exp) { +LIBC_INLINE static constexpr float16 frexpf16(float16 x, int *exp) { return fputil::frexp(x, *exp); } diff --git a/libc/src/math/generic/inv_trigf_utils.cpp b/libc/src/__support/math/inv_trigf_utils.h index f23028b..4a8fbec 100644 --- a/libc/src/math/generic/inv_trigf_utils.cpp +++ b/libc/src/__support/math/inv_trigf_utils.h @@ -1,4 +1,4 @@ -//===-- Single-precision general exp/log functions ------------------------===// +//===-- Single-precision general inverse trigonometric functions ----------===// // // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. // See https://llvm.org/LICENSE.txt for license information. @@ -6,11 +6,22 @@ // //===----------------------------------------------------------------------===// -#include "inv_trigf_utils.h" +#ifndef LLVM_LIBC_SRC___SUPPORT_MATH_INV_TRIGF_UTILS_H +#define LLVM_LIBC_SRC___SUPPORT_MATH_INV_TRIGF_UTILS_H + +#include "src/__support/FPUtil/PolyEval.h" +#include "src/__support/FPUtil/multiply_add.h" +#include "src/__support/common.h" #include "src/__support/macros/config.h" namespace LIBC_NAMESPACE_DECL { +namespace inv_trigf_utils_internal { + +// PI and PI / 2 +static constexpr double M_MATH_PI = 0x1.921fb54442d18p+1; +static constexpr double M_MATH_PI_2 = 0x1.921fb54442d18p+0; + // Polynomial approximation for 0 <= x <= 1: // atan(x) ~ atan((i/16) + (x - (i/16)) * Q(x - i/16) // = P(x - i/16) @@ -29,7 +40,7 @@ namespace LIBC_NAMESPACE_DECL { // Notice that degree-7 is good enough for atanf, but degree-8 helps reduce the // error bounds for atan2f's fast pass 16 times, and it does not affect the // performance of atanf much. -double ATAN_COEFFS[17][9] = { +static constexpr double ATAN_COEFFS[17][9] = { {0.0, 1.0, 0x1.3f8d76d26d61bp-47, -0x1.5555555574cd8p-2, 0x1.0dde5d06878eap-29, 0x1.99997738acc77p-3, 0x1.2c43eac9797cap-16, -0x1.25fb020007dbdp-3, 0x1.c1b6c31d7b0aep-7}, @@ -83,4 +94,91 @@ double ATAN_COEFFS[17][9] = { 0x1.555e31a1e15e9p-6, -0x1.245240d65e629p-7, -0x1.fa9ba66478903p-11}, }; +// Look-up table for atan(k/16) with k = 0..16. +static constexpr double ATAN_K_OVER_16[17] = { + 0.0, + 0x1.ff55bb72cfdeap-5, + 0x1.fd5ba9aac2f6ep-4, + 0x1.7b97b4bce5b02p-3, + 0x1.f5b75f92c80ddp-3, + 0x1.362773707ebccp-2, + 0x1.6f61941e4def1p-2, + 0x1.a64eec3cc23fdp-2, + 0x1.dac670561bb4fp-2, + 0x1.0657e94db30dp-1, + 0x1.1e00babdefeb4p-1, + 0x1.345f01cce37bbp-1, + 0x1.4978fa3269ee1p-1, + 0x1.5d58987169b18p-1, + 0x1.700a7c5784634p-1, + 0x1.819d0b7158a4dp-1, + 0x1.921fb54442d18p-1, +}; + +// For |x| <= 1/32 and 0 <= i <= 16, return Q(x) such that: +// Q(x) ~ (atan(x + i/16) - atan(i/16)) / x. +LIBC_INLINE static double atan_eval(double x, unsigned i) { + double x2 = x * x; + + double c0 = fputil::multiply_add(x, ATAN_COEFFS[i][2], ATAN_COEFFS[i][1]); + double c1 = fputil::multiply_add(x, ATAN_COEFFS[i][4], ATAN_COEFFS[i][3]); + double c2 = fputil::multiply_add(x, ATAN_COEFFS[i][6], ATAN_COEFFS[i][5]); + double c3 = fputil::multiply_add(x, ATAN_COEFFS[i][8], ATAN_COEFFS[i][7]); + + double x4 = x2 * x2; + double d1 = fputil::multiply_add(x2, c1, c0); + double d2 = fputil::multiply_add(x2, c3, c2); + double p = fputil::multiply_add(x4, d2, d1); + return p; +} + +// Evaluate atan without big lookup table. +// atan(n/d) - atan(k/16) = atan((n/d - k/16) / (1 + (n/d) * (k/16))) +// = atan((n - d * k/16)) / (d + n * k/16)) +// So we let q = (n - d * k/16) / (d + n * k/16), +// and approximate with Taylor polynomial: +// atan(q) ~ q - q^3/3 + q^5/5 - q^7/7 + q^9/9 +LIBC_INLINE static double atan_eval_no_table(double num, double den, + double k_over_16) { + double num_r = fputil::multiply_add(den, -k_over_16, num); + double den_r = fputil::multiply_add(num, k_over_16, den); + double q = num_r / den_r; + + constexpr double ATAN_TAYLOR[] = { + -0x1.5555555555555p-2, + 0x1.999999999999ap-3, + -0x1.2492492492492p-3, + 0x1.c71c71c71c71cp-4, + }; + double q2 = q * q; + double q3 = q2 * q; + double q4 = q2 * q2; + double c0 = fputil::multiply_add(q2, ATAN_TAYLOR[1], ATAN_TAYLOR[0]); + double c1 = fputil::multiply_add(q2, ATAN_TAYLOR[3], ATAN_TAYLOR[2]); + double d = fputil::multiply_add(q4, c1, c0); + return fputil::multiply_add(q3, d, q); +} + +// > Q = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20|], +// [|1, D...|], [0, 0.5]); +static constexpr double ASIN_COEFFS[10] = { + 0x1.5555555540fa1p-3, 0x1.333333512edc2p-4, 0x1.6db6cc1541b31p-5, + 0x1.f1caff324770ep-6, 0x1.6e43899f5f4f4p-6, 0x1.1f847cf652577p-6, + 0x1.9b60f47f87146p-7, 0x1.259e2634c494fp-6, -0x1.df946fa875ddp-8, + 0x1.02311ecf99c28p-5}; + +// Evaluate P(x^2) - 1, where P(x^2) ~ asin(x)/x +LIBC_INLINE static double asin_eval(double xsq) { + double x4 = xsq * xsq; + double r1 = fputil::polyeval(x4, ASIN_COEFFS[0], ASIN_COEFFS[2], + ASIN_COEFFS[4], ASIN_COEFFS[6], ASIN_COEFFS[8]); + double r2 = fputil::polyeval(x4, ASIN_COEFFS[1], ASIN_COEFFS[3], + ASIN_COEFFS[5], ASIN_COEFFS[7], ASIN_COEFFS[9]); + return fputil::multiply_add(xsq, r2, r1); +} + +} // namespace inv_trigf_utils_internal + } // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC___SUPPORT_MATH_INV_TRIGF_UTILS_H diff --git a/libc/src/__support/math/ldexpf.h b/libc/src/__support/math/ldexpf.h index 3a5ec1d..9ef5d96 100644 --- a/libc/src/__support/math/ldexpf.h +++ b/libc/src/__support/math/ldexpf.h @@ -17,7 +17,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float ldexpf(float x, int exp) { +LIBC_INLINE static constexpr float ldexpf(float x, int exp) { return fputil::ldexp(x, exp); } diff --git a/libc/src/__support/math/ldexpf128.h b/libc/src/__support/math/ldexpf128.h index 3625830..4fba20c 100644 --- a/libc/src/__support/math/ldexpf128.h +++ b/libc/src/__support/math/ldexpf128.h @@ -21,7 +21,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float128 ldexpf128(float128 x, int exp) { +LIBC_INLINE static constexpr float128 ldexpf128(float128 x, int exp) { return fputil::ldexp(x, exp); } diff --git a/libc/src/__support/math/ldexpf16.h b/libc/src/__support/math/ldexpf16.h index fbead87..d978d22 100644 --- a/libc/src/__support/math/ldexpf16.h +++ b/libc/src/__support/math/ldexpf16.h @@ -21,7 +21,7 @@ namespace LIBC_NAMESPACE_DECL { namespace math { -static constexpr float16 ldexpf16(float16 x, int exp) { +LIBC_INLINE static constexpr float16 ldexpf16(float16 x, int exp) { return fputil::ldexp(x, exp); } diff --git a/libc/src/__support/ryu_constants.h b/libc/src/__support/ryu_constants.h index 1ecb34e..09bc235 100644 --- a/libc/src/__support/ryu_constants.h +++ b/libc/src/__support/ryu_constants.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_RYU_CONSTANTS_H #define LLVM_LIBC_SRC___SUPPORT_RYU_CONSTANTS_H +#include "hdr/stdint_proxy.h" #include <stddef.h> -#include <stdint.h> constexpr size_t TABLE_SHIFT_CONST = 120; constexpr size_t IDX_SIZE = 16; diff --git a/libc/src/__support/ryu_long_double_constants.h b/libc/src/__support/ryu_long_double_constants.h index 8ebb297..487fd4a 100644 --- a/libc/src/__support/ryu_long_double_constants.h +++ b/libc/src/__support/ryu_long_double_constants.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_RYU_LONG_DOUBLE_CONSTANTS_H #define LLVM_LIBC_SRC___SUPPORT_RYU_LONG_DOUBLE_CONSTANTS_H +#include "hdr/stdint_proxy.h" #include <stddef.h> -#include <stdint.h> constexpr size_t TABLE_SHIFT_CONST = 120; constexpr size_t IDX_SIZE = 128; diff --git a/libc/src/__support/str_to_float.h b/libc/src/__support/str_to_float.h index a7dd7ce..3d35d8a 100644 --- a/libc/src/__support/str_to_float.h +++ b/libc/src/__support/str_to_float.h @@ -16,6 +16,7 @@ #define LLVM_LIBC_SRC___SUPPORT_STR_TO_FLOAT_H #include "hdr/errno_macros.h" // For ERANGE +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/bit.h" #include "src/__support/CPP/limits.h" #include "src/__support/CPP/optional.h" @@ -33,8 +34,6 @@ #include "src/__support/str_to_num_result.h" #include "src/__support/uint128.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/__support/threads/CMakeLists.txt b/libc/src/__support/threads/CMakeLists.txt index bd49bbb..b084346 100644 --- a/libc/src/__support/threads/CMakeLists.txt +++ b/libc/src/__support/threads/CMakeLists.txt @@ -49,6 +49,7 @@ add_header_library( HDRS thread.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.common libc.src.__support.CPP.atomic libc.src.__support.CPP.optional @@ -64,6 +65,7 @@ if(TARGET libc.src.__support.threads.${LIBC_TARGET_OS}.thread) DEPENDS .mutex .${LIBC_TARGET_OS}.thread + libc.hdr.stdint_proxy libc.src.__support.common libc.src.__support.fixedvector libc.src.__support.CPP.array diff --git a/libc/src/__support/threads/CndVar.h b/libc/src/__support/threads/CndVar.h index e42fa14..7b2a712 100644 --- a/libc/src/__support/threads/CndVar.h +++ b/libc/src/__support/threads/CndVar.h @@ -9,13 +9,12 @@ #ifndef LLVM_LIBC___SUPPORT_SRC_THREADS_LINUX_CNDVAR_H #define LLVM_LIBC___SUPPORT_SRC_THREADS_LINUX_CNDVAR_H +#include "hdr/stdint_proxy.h" // uint32_t #include "src/__support/macros/config.h" #include "src/__support/threads/linux/futex_utils.h" // Futex #include "src/__support/threads/linux/raw_mutex.h" // RawMutex #include "src/__support/threads/mutex.h" // Mutex -#include <stdint.h> // uint32_t - namespace LIBC_NAMESPACE_DECL { class CndVar { diff --git a/libc/src/__support/threads/linux/CMakeLists.txt b/libc/src/__support/threads/linux/CMakeLists.txt index 364e7e2..cbb7886 100644 --- a/libc/src/__support/threads/linux/CMakeLists.txt +++ b/libc/src/__support/threads/linux/CMakeLists.txt @@ -2,6 +2,8 @@ add_header_library( futex_word_type HDRS futex_word.h + DEPENDS + libc.hdr.stdint_proxy ) if(NOT TARGET libc.src.__support.OSUtil.osutil) @@ -114,6 +116,7 @@ add_object_library( HDRS ../CndVar.h DEPENDS + libc.hdr.stdint_proxy libc.include.sys_syscall libc.src.__support.OSUtil.osutil libc.src.__support.threads.linux.futex_word_type diff --git a/libc/src/__support/threads/linux/futex_word.h b/libc/src/__support/threads/linux/futex_word.h index a5a6a0c..1cf7bef 100644 --- a/libc/src/__support/threads/linux/futex_word.h +++ b/libc/src/__support/threads/linux/futex_word.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_THREADS_LINUX_FUTEX_WORD_H #define LLVM_LIBC_SRC___SUPPORT_THREADS_LINUX_FUTEX_WORD_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" -#include <stdint.h> #include <sys/syscall.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/__support/threads/linux/thread.cpp b/libc/src/__support/threads/linux/thread.cpp index baad26a..d9e479ee 100644 --- a/libc/src/__support/threads/linux/thread.cpp +++ b/libc/src/__support/threads/linux/thread.cpp @@ -23,10 +23,10 @@ #endif #include "hdr/fcntl_macros.h" +#include "hdr/stdint_proxy.h" #include <linux/param.h> // For EXEC_PAGESIZE. #include <linux/prctl.h> // For PR_SET_NAME #include <linux/sched.h> // For CLONE_* flags. -#include <stdint.h> #include <sys/mman.h> // For PROT_* and MAP_* definitions. #include <sys/syscall.h> // For syscall numbers. diff --git a/libc/src/__support/threads/thread.h b/libc/src/__support/threads/thread.h index f2b1f6b..114ab49 100644 --- a/libc/src/__support/threads/thread.h +++ b/libc/src/__support/threads/thread.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_THREADS_THREAD_H #define LLVM_LIBC_SRC___SUPPORT_THREADS_THREAD_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/atomic.h" #include "src/__support/CPP/optional.h" #include "src/__support/CPP/string_view.h" @@ -21,7 +22,6 @@ #include <linux/param.h> // for exec_pagesize. #include <stddef.h> // For size_t -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/__support/wchar/CMakeLists.txt b/libc/src/__support/wchar/CMakeLists.txt index 802441d..e363ad3 100644 --- a/libc/src/__support/wchar/CMakeLists.txt +++ b/libc/src/__support/wchar/CMakeLists.txt @@ -3,7 +3,8 @@ add_header_library( HDRS mbstate.h DEPENDS - libc.hdr.types.char32_t + libc.hdr.stdint_proxy + libc.hdr.types.char32_t ) add_header_library( @@ -68,3 +69,20 @@ add_object_library( .character_converter .mbstate ) + +add_header_library( + wcsnrtombs + HDRS + wcsnrtombs.h + DEPENDS + libc.hdr.errno_macros + libc.hdr.types.char8_t + libc.hdr.types.char32_t + libc.hdr.types.size_t + libc.hdr.types.wchar_t + libc.src.__support.error_or + libc.src.__support.common + .string_converter + .character_converter + .mbstate +) diff --git a/libc/src/__support/wchar/mbrtowc.cpp b/libc/src/__support/wchar/mbrtowc.cpp index 90ba934..0f730d6 100644 --- a/libc/src/__support/wchar/mbrtowc.cpp +++ b/libc/src/__support/wchar/mbrtowc.cpp @@ -37,7 +37,8 @@ ErrorOr<size_t> mbrtowc(wchar_t *__restrict pwc, const char *__restrict s, } auto wc = char_conv.pop_utf32(); if (wc.has_value()) { - *pwc = wc.value(); + if (pwc != nullptr) + *pwc = wc.value(); // null terminator -> return 0 if (wc.value() == L'\0') return 0; diff --git a/libc/src/__support/wchar/mbstate.h b/libc/src/__support/wchar/mbstate.h index 32304a5..96256ca 100644 --- a/libc/src/__support/wchar/mbstate.h +++ b/libc/src/__support/wchar/mbstate.h @@ -9,9 +9,9 @@ #ifndef LLVM_LIBC_SRC___SUPPORT_MBSTATE_H #define LLVM_LIBC_SRC___SUPPORT_MBSTATE_H +#include "hdr/stdint_proxy.h" #include "hdr/types/char32_t.h" #include "src/__support/common.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/__support/wchar/wcsnrtombs.h b/libc/src/__support/wchar/wcsnrtombs.h new file mode 100644 index 0000000..433097c --- /dev/null +++ b/libc/src/__support/wchar/wcsnrtombs.h @@ -0,0 +1,69 @@ +//===-- Implementation header for wcsnrtombs ------------------------------===// +// +// 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_WCHAR_WCSNRTOMBS_H +#define LLVM_LIBC_SRC__SUPPORT_WCHAR_WCSNRTOMBS_H + +#include "hdr/types/char32_t.h" +#include "hdr/types/size_t.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/common.h" +#include "src/__support/libc_errno.h" +#include "src/__support/macros/config.h" +#include "src/__support/macros/null_check.h" +#include "src/__support/wchar/mbstate.h" +#include "src/__support/wchar/string_converter.h" + +namespace LIBC_NAMESPACE_DECL { +namespace internal { + +LIBC_INLINE static ErrorOr<size_t> +wcsnrtombs(char *__restrict dest, const wchar_t **__restrict ptr_to_src, + size_t num_src_widechars, size_t dest_len, mbstate *ps) { + LIBC_CRASH_ON_NULLPTR(ptr_to_src); + LIBC_CRASH_ON_NULLPTR(ps); + + CharacterConverter cr(ps); + if (!cr.isValidState()) + return Error(EINVAL); + + if (dest == nullptr) + dest_len = SIZE_MAX; + + StringConverter<char32_t> str_conv( + reinterpret_cast<const char32_t *>(*ptr_to_src), ps, dest_len, + num_src_widechars); + size_t dst_idx = 0; + ErrorOr<char8_t> converted = str_conv.popUTF8(); + while (converted.has_value()) { + if (dest != nullptr) + dest[dst_idx] = converted.value(); + + if (converted.value() == '\0') { + if (dest != nullptr) + *ptr_to_src = nullptr; + return dst_idx; + } + + dst_idx++; + converted = str_conv.popUTF8(); + } + + if (dest != nullptr) + *ptr_to_src += str_conv.getSourceIndex(); + + if (converted.error() == -1) // if we hit conversion limit + return dst_idx; + + return Error(converted.error()); +} + +} // namespace internal +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC__SUPPORT_WCHAR_WCSNRTOMBS_H diff --git a/libc/src/arpa/inet/CMakeLists.txt b/libc/src/arpa/inet/CMakeLists.txt index cdd53e9..1f39a07 100644 --- a/libc/src/arpa/inet/CMakeLists.txt +++ b/libc/src/arpa/inet/CMakeLists.txt @@ -5,6 +5,7 @@ add_entrypoint_object( HDRS htonl.h DEPENDS + libc.hdr.stdint_proxy libc.include.arpa_inet libc.src.__support.common ) @@ -16,6 +17,7 @@ add_entrypoint_object( HDRS htons.h DEPENDS + libc.hdr.stdint_proxy libc.include.arpa_inet libc.src.__support.common ) @@ -27,6 +29,7 @@ add_entrypoint_object( HDRS ntohl.h DEPENDS + libc.hdr.stdint_proxy libc.include.arpa_inet libc.src.__support.common ) @@ -38,6 +41,7 @@ add_entrypoint_object( HDRS ntohs.h DEPENDS + libc.hdr.stdint_proxy libc.include.arpa_inet libc.src.__support.common ) diff --git a/libc/src/arpa/inet/htonl.h b/libc/src/arpa/inet/htonl.h index e444972..34f017d 100644 --- a/libc/src/arpa/inet/htonl.h +++ b/libc/src/arpa/inet/htonl.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC_ARPA_INET_HTONL_H #define LLVM_LIBC_SRC_ARPA_INET_HTONL_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/arpa/inet/htons.h b/libc/src/arpa/inet/htons.h index 35c2acdc..5e283cd 100644 --- a/libc/src/arpa/inet/htons.h +++ b/libc/src/arpa/inet/htons.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC_ARPA_INET_HTONS_H #define LLVM_LIBC_SRC_ARPA_INET_HTONS_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/arpa/inet/ntohl.h b/libc/src/arpa/inet/ntohl.h index 4007965..76ba02b 100644 --- a/libc/src/arpa/inet/ntohl.h +++ b/libc/src/arpa/inet/ntohl.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC_ARPA_INET_NTOHL_H #define LLVM_LIBC_SRC_ARPA_INET_NTOHL_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/arpa/inet/ntohs.h b/libc/src/arpa/inet/ntohs.h index 5fe3ebc..b98c852 100644 --- a/libc/src/arpa/inet/ntohs.h +++ b/libc/src/arpa/inet/ntohs.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC_ARPA_INET_NTOHS_H #define LLVM_LIBC_SRC_ARPA_INET_NTOHS_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/compiler/generic/CMakeLists.txt b/libc/src/compiler/generic/CMakeLists.txt index 2fc8f7f..d9ad42e 100644 --- a/libc/src/compiler/generic/CMakeLists.txt +++ b/libc/src/compiler/generic/CMakeLists.txt @@ -5,6 +5,7 @@ add_entrypoint_object( HDRS ../__stack_chk_fail.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.OSUtil.osutil libc.src.stdlib.abort ) diff --git a/libc/src/compiler/generic/__stack_chk_fail.cpp b/libc/src/compiler/generic/__stack_chk_fail.cpp index 00e976a..7edd16c 100644 --- a/libc/src/compiler/generic/__stack_chk_fail.cpp +++ b/libc/src/compiler/generic/__stack_chk_fail.cpp @@ -7,9 +7,9 @@ //===----------------------------------------------------------------------===// #include "src/compiler/__stack_chk_fail.h" +#include "hdr/stdint_proxy.h" // For uintptr_t #include "src/__support/OSUtil/io.h" #include "src/stdlib/abort.h" -#include <stdint.h> // For uintptr_t extern "C" { diff --git a/libc/src/inttypes/CMakeLists.txt b/libc/src/inttypes/CMakeLists.txt index c3111ed..3a48c9a 100644 --- a/libc/src/inttypes/CMakeLists.txt +++ b/libc/src/inttypes/CMakeLists.txt @@ -5,6 +5,7 @@ add_entrypoint_object( HDRS strtoimax.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.str_to_integer libc.src.errno.errno ) @@ -16,6 +17,7 @@ add_entrypoint_object( HDRS strtoumax.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.str_to_integer libc.src.errno.errno ) diff --git a/libc/src/inttypes/strtoimax.h b/libc/src/inttypes/strtoimax.h index 804d07c..fedc458b 100644 --- a/libc/src/inttypes/strtoimax.h +++ b/libc/src/inttypes/strtoimax.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC_INTTYPES_STRTOIMAX_H #define LLVM_LIBC_SRC_INTTYPES_STRTOIMAX_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/inttypes/strtoumax.h b/libc/src/inttypes/strtoumax.h index 4c53c03..d5b82af 100644 --- a/libc/src/inttypes/strtoumax.h +++ b/libc/src/inttypes/strtoumax.h @@ -9,8 +9,8 @@ #ifndef LLVM_LIBC_SRC_INTTYPES_STRTOUMAX_H #define LLVM_LIBC_SRC_INTTYPES_STRTOUMAX_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/link/CMakeLists.txt b/libc/src/link/CMakeLists.txt index f68950e..55f5edf 100644 --- a/libc/src/link/CMakeLists.txt +++ b/libc/src/link/CMakeLists.txt @@ -4,4 +4,6 @@ add_entrypoint_object( dl_iterate_phdr.cpp HDRS dl_iterate_phdr.h + DEPENDS + libc.hdr.stdint_proxy ) diff --git a/libc/src/math/docs/add_math_function.md b/libc/src/math/docs/add_math_function.md index daaf1a3..d2c85ec 100644 --- a/libc/src/math/docs/add_math_function.md +++ b/libc/src/math/docs/add_math_function.md @@ -183,8 +183,8 @@ implementation (which is very often glibc). - Build and Run exhaustive test (might take hours to run): ``` - $ ninja libc.test.src.math.exhaustive.<func>_test - $ projects/libc/test/src/math/exhaustive/libc.test.src.math.exhaustive.<func>_test + $ ninja libc.test.src.math.exhaustive.<func>_test.__unit__ + $ projects/libc/test/src/math/exhaustive/libc.test.src.math.exhaustive.<func>_test.__unit__ ``` - Build and Run performance test: diff --git a/libc/src/math/generic/CMakeLists.txt b/libc/src/math/generic/CMakeLists.txt index 7e6a32b..d4d268c 100644 --- a/libc/src/math/generic/CMakeLists.txt +++ b/libc/src/math/generic/CMakeLists.txt @@ -1295,12 +1295,8 @@ add_entrypoint_object( HDRS ../erff.h DEPENDS - .common_constants - libc.src.__support.FPUtil.fp_bits - libc.src.__support.FPUtil.except_value_utils - libc.src.__support.FPUtil.multiply_add - libc.src.__support.FPUtil.polyeval - libc.src.__support.macros.optimization + libc.src.__support.math.erff + libc.src.errno.errno ) add_entrypoint_object( @@ -1898,6 +1894,7 @@ add_object_library( common_constants.cpp DEPENDS libc.src.__support.math.exp_constants + libc.src.__support.math.acosh_float_constants libc.src.__support.number_pair ) @@ -3761,7 +3758,7 @@ add_header_library( DEPENDS .common_constants libc.src.__support.math.exp_utils - libc.src.__support.math.exp10f_utils + libc.src.__support.math.acoshf_utils libc.src.__support.macros.properties.cpu_features libc.src.errno.errno ) @@ -3871,12 +3868,7 @@ add_entrypoint_object( ../acoshf.h DEPENDS .explogxf - 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.sqrt - libc.src.__support.macros.optimization + libc.src.__support.math.acoshf ) add_entrypoint_object( @@ -3886,18 +3878,8 @@ add_entrypoint_object( HDRS ../acoshf16.h DEPENDS - .explogxf - libc.hdr.errno_macros - libc.hdr.fenv_macros - 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.polyeval - libc.src.__support.FPUtil.sqrt - libc.src.__support.macros.optimization - libc.src.__support.macros.properties.types + libc.src.__support.math.acoshf16 + libc.src.errno.errno ) add_entrypoint_object( @@ -3969,18 +3951,6 @@ add_entrypoint_object( libc.src.__support.macros.properties.types ) -add_object_library( - inv_trigf_utils - HDRS - inv_trigf_utils.h - SRCS - inv_trigf_utils.cpp - DEPENDS - libc.src.__support.FPUtil.multiply_add - libc.src.__support.FPUtil.polyeval - libc.src.__support.common -) - add_entrypoint_object( asinf SRCS @@ -3988,13 +3958,7 @@ add_entrypoint_object( HDRS ../asinf.h DEPENDS - libc.src.__support.FPUtil.except_value_utils - libc.src.__support.FPUtil.fp_bits - libc.src.__support.FPUtil.multiply_add - libc.src.__support.FPUtil.polyeval - libc.src.__support.FPUtil.sqrt - libc.src.__support.macros.optimization - .inv_trigf_utils + libc.src.__support.math.asinf ) add_entrypoint_object( @@ -4004,16 +3968,7 @@ add_entrypoint_object( HDRS ../asinf16.h DEPENDS - libc.hdr.errno_macros - libc.hdr.fenv_macros - libc.src.__support.FPUtil.cast - 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.sqrt - libc.src.__support.macros.optimization - libc.src.__support.macros.properties.types + libc.src.__support.math.asinf16 ) add_entrypoint_object( @@ -4023,16 +3978,7 @@ add_entrypoint_object( HDRS ../asin.h DEPENDS - libc.src.__support.math.asin_utils - libc.src.__support.FPUtil.double_double - libc.src.__support.FPUtil.dyadic_float - 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.sqrt - libc.src.__support.macros.optimization - libc.src.__support.macros.properties.cpu_features + libc.src.__support.math.asin ) add_entrypoint_object( @@ -4042,13 +3988,7 @@ add_entrypoint_object( HDRS ../acosf.h DEPENDS - libc.src.__support.FPUtil.except_value_utils - libc.src.__support.FPUtil.fp_bits - libc.src.__support.FPUtil.multiply_add - libc.src.__support.FPUtil.polyeval - libc.src.__support.FPUtil.sqrt - libc.src.__support.macros.optimization - .inv_trigf_utils + libc.src.__support.math.acosf ) add_entrypoint_object( @@ -4058,17 +3998,8 @@ add_entrypoint_object( HDRS ../acosf16.h DEPENDS - libc.hdr.errno_macros - libc.hdr.fenv_macros - 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.polyeval - libc.src.__support.FPUtil.sqrt - libc.src.__support.macros.optimization - libc.src.__support.macros.properties.types + libc.src.__support.math.acosf16 + libc.src.errno.errno ) add_entrypoint_object( @@ -4088,16 +4019,8 @@ add_entrypoint_object( HDRS ../acospif16.h DEPENDS - libc.hdr.errno_macros - libc.hdr.fenv_macros - libc.src.__support.FPUtil.cast - 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.sqrt - libc.src.__support.macros.optimization - libc.src.__support.macros.properties.types + libc.src.__support.math.acospif16 + libc.src.errno.errno ) add_header_library( @@ -4120,7 +4043,6 @@ add_entrypoint_object( HDRS ../atanf.h DEPENDS - .inv_trigf_utils libc.src.__support.FPUtil.except_value_utils libc.src.__support.FPUtil.fp_bits libc.src.__support.FPUtil.multiply_add @@ -4128,6 +4050,7 @@ add_entrypoint_object( libc.src.__support.FPUtil.polyeval libc.src.__support.FPUtil.rounding_mode libc.src.__support.macros.optimization + libc.src.__support.math.inv_trigf_utils ) add_entrypoint_object( @@ -4176,7 +4099,6 @@ add_entrypoint_object( ../atan2f.h atan2f_float.h DEPENDS - .inv_trigf_utils libc.hdr.fenv_macros libc.src.__support.FPUtil.double_double libc.src.__support.FPUtil.fenv_impl @@ -4186,6 +4108,7 @@ add_entrypoint_object( libc.src.__support.FPUtil.polyeval libc.src.__support.FPUtil.rounding_mode libc.src.__support.macros.optimization + libc.src.__support.math.inv_trigf_utils ) add_entrypoint_object( @@ -4987,6 +4910,7 @@ add_header_library( HDRS expxf16.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.FPUtil.fp_bits libc.src.__support.FPUtil.cast libc.src.__support.FPUtil.multiply_add diff --git a/libc/src/math/generic/acosf.cpp b/libc/src/math/generic/acosf.cpp index 8dd6de2..7afc7d6 100644 --- a/libc/src/math/generic/acosf.cpp +++ b/libc/src/math/generic/acosf.cpp @@ -7,127 +7,10 @@ //===----------------------------------------------------------------------===// #include "src/math/acosf.h" -#include "src/__support/FPUtil/FEnvImpl.h" -#include "src/__support/FPUtil/FPBits.h" -#include "src/__support/FPUtil/PolyEval.h" -#include "src/__support/FPUtil/except_value_utils.h" -#include "src/__support/FPUtil/multiply_add.h" -#include "src/__support/FPUtil/sqrt.h" -#include "src/__support/macros/config.h" -#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY - -#include "inv_trigf_utils.h" +#include "src/__support/math/acosf.h" namespace LIBC_NAMESPACE_DECL { -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS -static constexpr size_t N_EXCEPTS = 4; - -// Exceptional values when |x| <= 0.5 -static constexpr fputil::ExceptValues<float, N_EXCEPTS> ACOSF_EXCEPTS = {{ - // (inputs, RZ output, RU offset, RD offset, RN offset) - // x = 0x1.110b46p-26, acosf(x) = 0x1.921fb4p0 (RZ) - {0x328885a3, 0x3fc90fda, 1, 0, 1}, - // x = -0x1.110b46p-26, acosf(x) = 0x1.921fb4p0 (RZ) - {0xb28885a3, 0x3fc90fda, 1, 0, 1}, - // x = 0x1.04c444p-12, acosf(x) = 0x1.920f68p0 (RZ) - {0x39826222, 0x3fc907b4, 1, 0, 1}, - // x = -0x1.04c444p-12, acosf(x) = 0x1.923p0 (RZ) - {0xb9826222, 0x3fc91800, 1, 0, 1}, -}}; -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS - -LLVM_LIBC_FUNCTION(float, acosf, (float x)) { - using FPBits = typename fputil::FPBits<float>; - - FPBits xbits(x); - uint32_t x_uint = xbits.uintval(); - uint32_t x_abs = xbits.uintval() & 0x7fff'ffffU; - uint32_t x_sign = x_uint >> 31; - - // |x| <= 0.5 - if (LIBC_UNLIKELY(x_abs <= 0x3f00'0000U)) { - // |x| < 0x1p-10 - if (LIBC_UNLIKELY(x_abs < 0x3a80'0000U)) { - // When |x| < 2^-10, we use the following approximation: - // acos(x) = pi/2 - asin(x) - // ~ pi/2 - x - x^3 / 6 - -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS - // Check for exceptional values - if (auto r = ACOSF_EXCEPTS.lookup(x_uint); LIBC_UNLIKELY(r.has_value())) - return r.value(); -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS - - double xd = static_cast<double>(x); - return static_cast<float>(fputil::multiply_add( - -0x1.5555555555555p-3 * xd, xd * xd, M_MATH_PI_2 - xd)); - } - - // For |x| <= 0.5, we approximate acosf(x) by: - // acos(x) = pi/2 - asin(x) = pi/2 - x * P(x^2) - // Where P(X^2) = Q(X) is a degree-20 minimax even polynomial approximating - // asin(x)/x on [0, 0.5] generated by Sollya with: - // > Q = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20|], - // [|1, D...|], [0, 0.5]); - double xd = static_cast<double>(x); - double xsq = xd * xd; - double x3 = xd * xsq; - double r = asin_eval(xsq); - return static_cast<float>(fputil::multiply_add(-x3, r, M_MATH_PI_2 - xd)); - } - - // |x| >= 1, return 0, 2pi, or NaNs. - if (LIBC_UNLIKELY(x_abs >= 0x3f80'0000U)) { - if (x_abs == 0x3f80'0000U) - return x_sign ? /* x == -1.0f */ fputil::round_result_slightly_down( - 0x1.921fb6p+1f) - : /* x == 1.0f */ 0.0f; - - if (xbits.is_signaling_nan()) { - fputil::raise_except_if_required(FE_INVALID); - return FPBits::quiet_nan().get_val(); - } - - // |x| <= +/-inf - if (x_abs <= 0x7f80'0000U) { - fputil::set_errno_if_required(EDOM); - fputil::raise_except_if_required(FE_INVALID); - } - - return x + FPBits::quiet_nan().get_val(); - } - - // When 0.5 < |x| < 1, we perform range reduction as follow: - // - // Assume further that 0.5 < x <= 1, and let: - // y = acos(x) - // We use the double angle formula: - // x = cos(y) = 1 - 2 sin^2(y/2) - // So: - // sin(y/2) = sqrt( (1 - x)/2 ) - // And hence: - // y = 2 * asin( sqrt( (1 - x)/2 ) ) - // Let u = (1 - x)/2, then - // acos(x) = 2 * asin( sqrt(u) ) - // Moreover, since 0.5 < x <= 1, - // 0 <= u < 1/4, and 0 <= sqrt(u) < 0.5, - // And hence we can reuse the same polynomial approximation of asin(x) when - // |x| <= 0.5: - // acos(x) ~ 2 * sqrt(u) * P(u). - // - // When -1 < x <= -0.5, we use the identity: - // acos(x) = pi - acos(-x) - // which is reduced to the postive case. - - xbits.set_sign(Sign::POS); - double xd = static_cast<double>(xbits.get_val()); - double u = fputil::multiply_add(-0.5, xd, 0.5); - double cv = 2 * fputil::sqrt<double>(u); - - double r3 = asin_eval(u); - double r = fputil::multiply_add(cv * u, r3, cv); - return static_cast<float>(x_sign ? M_MATH_PI - r : r); -} +LLVM_LIBC_FUNCTION(float, acosf, (float x)) { return math::acosf(x); } } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/math/generic/acosf16.cpp b/libc/src/math/generic/acosf16.cpp index 202a950..0bf85f8 100644 --- a/libc/src/math/generic/acosf16.cpp +++ b/libc/src/math/generic/acosf16.cpp @@ -8,144 +8,10 @@ //===----------------------------------------------------------------------===// #include "src/math/acosf16.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/multiply_add.h" -#include "src/__support/FPUtil/sqrt.h" -#include "src/__support/macros/optimization.h" +#include "src/__support/math/acosf16.h" namespace LIBC_NAMESPACE_DECL { -// Generated by Sollya using the following command: -// > round(pi/2, SG, RN); -// > round(pi, SG, RN); -static constexpr float PI_OVER_2 = 0x1.921fb6p0f; -static constexpr float PI = 0x1.921fb6p1f; +LLVM_LIBC_FUNCTION(float16, acosf16, (float16 x)) { return math::acosf16(x); } -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS -static constexpr size_t N_EXCEPTS = 2; - -static constexpr fputil::ExceptValues<float16, N_EXCEPTS> ACOSF16_EXCEPTS{{ - // (input, RZ output, RU offset, RD offset, RN offset) - {0xacaf, 0x3e93, 1, 0, 0}, - {0xb874, 0x4052, 1, 0, 1}, -}}; -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS - -LLVM_LIBC_FUNCTION(float16, acosf16, (float16 x)) { - using FPBits = fputil::FPBits<float16>; - FPBits xbits(x); - - uint16_t x_u = xbits.uintval(); - uint16_t x_abs = x_u & 0x7fff; - uint16_t x_sign = x_u >> 15; - - // |x| > 0x1p0, |x| > 1, or x is NaN. - if (LIBC_UNLIKELY(x_abs > 0x3c00)) { - // acosf16(NaN) = 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; - } - - // 1 < |x| <= +/-inf - fputil::raise_except_if_required(FE_INVALID); - fputil::set_errno_if_required(EDOM); - - return FPBits::quiet_nan().get_val(); - } - - float xf = x; - -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS - // Handle exceptional values - if (auto r = ACOSF16_EXCEPTS.lookup(x_u); LIBC_UNLIKELY(r.has_value())) - return r.value(); -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS - - // |x| == 0x1p0, x is 1 or -1 - // if x is (-)1, return pi, else - // if x is (+)1, return 0 - if (LIBC_UNLIKELY(x_abs == 0x3c00)) - return fputil::cast<float16>(x_sign ? PI : 0.0f); - - float xsq = xf * xf; - - // |x| <= 0x1p-1, |x| <= 0.5 - if (x_abs <= 0x3800) { - // if x is 0, return pi/2 - if (LIBC_UNLIKELY(x_abs == 0)) - return fputil::cast<float16>(PI_OVER_2); - - // Note that: acos(x) = pi/2 + asin(-x) = pi/2 - asin(x) - // Degree-6 minimax polynomial of asin(x) generated by Sollya with: - // > P = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 0.5]); - float interm = - fputil::polyeval(xsq, 0x1.000002p0f, 0x1.554c2ap-3f, 0x1.3541ccp-4f, - 0x1.43b2d6p-5f, 0x1.a0d73ep-5f); - return fputil::cast<float16>(fputil::multiply_add(-xf, interm, PI_OVER_2)); - } - - // When |x| > 0.5, assume that 0.5 < |x| <= 1 - // - // Step-by-step range-reduction proof: - // 1: Let y = asin(x), such that, x = sin(y) - // 2: From complimentary angle identity: - // x = sin(y) = cos(pi/2 - y) - // 3: Let z = pi/2 - y, such that x = cos(z) - // 4: From double angle formula; cos(2A) = 1 - 2 * sin^2(A): - // z = 2A, z/2 = A - // cos(z) = 1 - 2 * sin^2(z/2) - // 5: Make sin(z/2) subject of the formula: - // sin(z/2) = sqrt((1 - cos(z))/2) - // 6: Recall [3]; x = cos(z). Therefore: - // sin(z/2) = sqrt((1 - x)/2) - // 7: Let u = (1 - x)/2 - // 8: Therefore: - // asin(sqrt(u)) = z/2 - // 2 * asin(sqrt(u)) = z - // 9: Recall [3]; z = pi/2 - y. Therefore: - // y = pi/2 - z - // y = pi/2 - 2 * asin(sqrt(u)) - // 10: Recall [1], y = asin(x). Therefore: - // asin(x) = pi/2 - 2 * asin(sqrt(u)) - // 11: Recall that: acos(x) = pi/2 + asin(-x) = pi/2 - asin(x) - // Therefore: - // acos(x) = pi/2 - (pi/2 - 2 * asin(sqrt(u))) - // acos(x) = 2 * asin(sqrt(u)) - // - // THE RANGE REDUCTION, HOW? - // 12: Recall [7], u = (1 - x)/2 - // 13: Since 0.5 < x <= 1, therefore: - // 0 <= u <= 0.25 and 0 <= sqrt(u) <= 0.5 - // - // Hence, we can reuse the same [0, 0.5] domain polynomial approximation for - // Step [11] as `sqrt(u)` is in range. - // When -1 < x <= -0.5, the identity: - // acos(x) = pi - acos(-x) - // allows us to compute for the negative x value (lhs) - // with a positive x value instead (rhs). - - float xf_abs = (xf < 0 ? -xf : xf); - float u = fputil::multiply_add(-0.5f, xf_abs, 0.5f); - float sqrt_u = fputil::sqrt<float>(u); - - // Degree-6 minimax polynomial of asin(x) generated by Sollya with: - // > P = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 0.5]); - float asin_sqrt_u = - sqrt_u * fputil::polyeval(u, 0x1.000002p0f, 0x1.554c2ap-3f, - 0x1.3541ccp-4f, 0x1.43b2d6p-5f, 0x1.a0d73ep-5f); - - return fputil::cast<float16>( - x_sign ? fputil::multiply_add(-2.0f, asin_sqrt_u, PI) : 2 * asin_sqrt_u); -} } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/math/generic/acoshf.cpp b/libc/src/math/generic/acoshf.cpp index c4927fa..5c04583 100644 --- a/libc/src/math/generic/acoshf.cpp +++ b/libc/src/math/generic/acoshf.cpp @@ -7,73 +7,11 @@ //===----------------------------------------------------------------------===// #include "src/math/acoshf.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/sqrt.h" -#include "src/__support/macros/config.h" -#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY -#include "src/math/generic/common_constants.h" -#include "src/math/generic/explogxf.h" -namespace LIBC_NAMESPACE_DECL { - -LLVM_LIBC_FUNCTION(float, acoshf, (float x)) { - using FPBits_t = typename fputil::FPBits<float>; - FPBits_t xbits(x); - - if (LIBC_UNLIKELY(x <= 1.0f)) { - if (x == 1.0f) - return 0.0f; - // x < 1. - fputil::set_errno_if_required(EDOM); - fputil::raise_except_if_required(FE_INVALID); - return FPBits_t::quiet_nan().get_val(); - } +#include "src/__support/math/acoshf.h" -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS - uint32_t x_u = xbits.uintval(); - if (LIBC_UNLIKELY(x_u >= 0x4f8ffb03)) { - if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) - return x; - - // Helper functions to set results for exceptional cases. - auto round_result_slightly_down = [](float r) -> float { - volatile float tmp = r; - tmp = tmp - 0x1.0p-25f; - return tmp; - }; - auto round_result_slightly_up = [](float r) -> float { - volatile float tmp = r; - tmp = tmp + 0x1.0p-25f; - return tmp; - }; - - switch (x_u) { - case 0x4f8ffb03: // x = 0x1.1ff606p32f - return round_result_slightly_up(0x1.6fdd34p4f); - case 0x5c569e88: // x = 0x1.ad3d1p57f - return round_result_slightly_up(0x1.45c146p5f); - case 0x5e68984e: // x = 0x1.d1309cp61f - return round_result_slightly_up(0x1.5c9442p5f); - case 0x655890d3: // x = 0x1.b121a6p75f - return round_result_slightly_down(0x1.a9a3f2p5f); - case 0x6eb1a8ec: // x = 0x1.6351d8p94f - return round_result_slightly_down(0x1.08b512p6f); - case 0x7997f30a: // x = 0x1.2fe614p116f - return round_result_slightly_up(0x1.451436p6f); - } - } -#else - if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) - return x; -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS +namespace LIBC_NAMESPACE_DECL { - double x_d = static_cast<double>(x); - // acosh(x) = log(x + sqrt(x^2 - 1)) - return static_cast<float>(log_eval( - x_d + fputil::sqrt<double>(fputil::multiply_add(x_d, x_d, -1.0)))); -} +LLVM_LIBC_FUNCTION(float, acoshf, (float x)) { return math::acoshf(x); } } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/math/generic/acoshf16.cpp b/libc/src/math/generic/acoshf16.cpp index 44783a8..bb3a91f 100644 --- a/libc/src/math/generic/acoshf16.cpp +++ b/libc/src/math/generic/acoshf16.cpp @@ -7,104 +7,10 @@ //===----------------------------------------------------------------------===// #include "src/math/acoshf16.h" -#include "explogxf.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/multiply_add.h" -#include "src/__support/FPUtil/sqrt.h" -#include "src/__support/common.h" -#include "src/__support/macros/config.h" -#include "src/__support/macros/optimization.h" +#include "src/__support/math/acoshf16.h" namespace LIBC_NAMESPACE_DECL { -static constexpr size_t N_EXCEPTS = 2; -static constexpr fputil::ExceptValues<float16, N_EXCEPTS> ACOSHF16_EXCEPTS{{ - // (input, RZ output, RU offset, RD offset, RN offset) - // x = 0x1.6dcp+1, acoshf16(x) = 0x1.b6p+0 (RZ) - {0x41B7, 0x3ED8, 1, 0, 0}, - // x = 0x1.39p+0, acoshf16(x) = 0x1.4f8p-1 (RZ) - {0x3CE4, 0x393E, 1, 0, 1}, -}}; - -LLVM_LIBC_FUNCTION(float16, acoshf16, (float16 x)) { - using FPBits = fputil::FPBits<float16>; - FPBits xbits(x); - uint16_t x_u = xbits.uintval(); - - // Check for NaN input first. - if (LIBC_UNLIKELY(xbits.is_inf_or_nan())) { - if (xbits.is_signaling_nan()) { - fputil::raise_except_if_required(FE_INVALID); - return FPBits::quiet_nan().get_val(); - } - if (xbits.is_neg()) { - fputil::set_errno_if_required(EDOM); - fputil::raise_except_if_required(FE_INVALID); - return FPBits::quiet_nan().get_val(); - } - return x; - } - - // Domain error for inputs less than 1.0. - if (LIBC_UNLIKELY(x <= 1.0f)) { - if (x == 1.0f) - return FPBits::zero().get_val(); - fputil::set_errno_if_required(EDOM); - fputil::raise_except_if_required(FE_INVALID); - return FPBits::quiet_nan().get_val(); - } - - if (auto r = ACOSHF16_EXCEPTS.lookup(xbits.uintval()); - LIBC_UNLIKELY(r.has_value())) - return r.value(); - - float xf = x; - // High-precision polynomial approximation for inputs close to 1.0 - // ([1, 1.25)). - // - // Brief derivation: - // 1. Expand acosh(1 + delta) using Taylor series around delta=0: - // acosh(1 + delta) ≈ sqrt(2 * delta) * [1 - delta/12 + 3*delta^2/160 - // - 5*delta^3/896 + 35*delta^4/18432 + ...] - // 2. Truncate the series to fit accurately for delta in [0, 0.25]. - // 3. Polynomial coefficients (from sollya) used here are: - // P(delta) ≈ 1 - 0x1.555556p-4 * delta + 0x1.333334p-6 * delta^2 - // - 0x1.6db6dcp-8 * delta^3 + 0x1.f1c71cp-10 * delta^4 - // 4. The Sollya commands used to generate these coefficients were: - // > display = hexadecimal; - // > round(1/12, SG, RN); - // > round(3/160, SG, RN); - // > round(5/896, SG, RN); - // > round(35/18432, SG, RN); - // With hexadecimal display mode enabled, the outputs were: - // 0x1.555556p-4 - // 0x1.333334p-6 - // 0x1.6db6dcp-8 - // 0x1.f1c71cp-10 - // 5. The maximum absolute error, estimated using: - // dirtyinfnorm(acosh(1 + x) - sqrt(2*x) * P(x), [0, 0.25]) - // is: - // 0x1.d84281p-22 - if (LIBC_UNLIKELY(x_u < 0x3D00U)) { - float delta = xf - 1.0f; - float sqrt_2_delta = fputil::sqrt<float>(2.0 * delta); - float pe = fputil::polyeval(delta, 0x1p+0f, -0x1.555556p-4f, 0x1.333334p-6f, - -0x1.6db6dcp-8f, 0x1.f1c71cp-10f); - float approx = sqrt_2_delta * pe; - return fputil::cast<float16>(approx); - } - - // acosh(x) = log(x + sqrt(x^2 - 1)) - float sqrt_term = fputil::sqrt<float>(fputil::multiply_add(xf, xf, -1.0f)); - float result = static_cast<float>(log_eval(xf + sqrt_term)); - - return fputil::cast<float16>(result); -} +LLVM_LIBC_FUNCTION(float16, acoshf16, (float16 x)) { return math::acoshf16(x); } } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/math/generic/acospif16.cpp b/libc/src/math/generic/acospif16.cpp index bfdf169..09cbd99 100644 --- a/libc/src/math/generic/acospif16.cpp +++ b/libc/src/math/generic/acospif16.cpp @@ -7,128 +7,12 @@ //===----------------------------------------------------------------------===// #include "src/math/acospif16.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/multiply_add.h" -#include "src/__support/FPUtil/sqrt.h" -#include "src/__support/macros/optimization.h" +#include "src/__support/math/acospif16.h" namespace LIBC_NAMESPACE_DECL { LLVM_LIBC_FUNCTION(float16, acospif16, (float16 x)) { - using FPBits = fputil::FPBits<float16>; - FPBits xbits(x); - - uint16_t x_u = xbits.uintval(); - uint16_t x_abs = x_u & 0x7fff; - uint16_t x_sign = x_u >> 15; - - // |x| > 0x1p0, |x| > 1, or x is NaN. - if (LIBC_UNLIKELY(x_abs > 0x3c00)) { - // acospif16(NaN) = 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; - } - - // 1 < |x| <= +inf - fputil::raise_except_if_required(FE_INVALID); - fputil::set_errno_if_required(EDOM); - - return FPBits::quiet_nan().get_val(); - } - - // |x| == 0x1p0, x is 1 or -1 - // if x is (-)1, return 1 - // if x is (+)1, return 0 - if (LIBC_UNLIKELY(x_abs == 0x3c00)) - return fputil::cast<float16>(x_sign ? 1.0f : 0.0f); - - float xf = x; - float xsq = xf * xf; - - // Degree-6 minimax polynomial coefficients of asin(x) generated by Sollya - // with: > P = fpminimax(asin(x)/(pi * x), [|0, 2, 4, 6, 8|], [|SG...|], [0, - // 0.5]); - constexpr float POLY_COEFFS[5] = {0x1.45f308p-2f, 0x1.b2900cp-5f, - 0x1.897e36p-6f, 0x1.9efafcp-7f, - 0x1.06d884p-6f}; - // |x| <= 0x1p-1, |x| <= 0.5 - if (x_abs <= 0x3800) { - // if x is 0, return 0.5 - if (LIBC_UNLIKELY(x_abs == 0)) - return fputil::cast<float16>(0.5f); - - // Note that: acos(x) = pi/2 + asin(-x) = pi/2 - asin(x), then - // acospi(x) = 0.5 - asin(x)/pi - float interm = - fputil::polyeval(xsq, POLY_COEFFS[0], POLY_COEFFS[1], POLY_COEFFS[2], - POLY_COEFFS[3], POLY_COEFFS[4]); - - return fputil::cast<float16>(fputil::multiply_add(-xf, interm, 0.5f)); - } - - // When |x| > 0.5, assume that 0.5 < |x| <= 1 - // - // Step-by-step range-reduction proof: - // 1: Let y = asin(x), such that, x = sin(y) - // 2: From complimentary angle identity: - // x = sin(y) = cos(pi/2 - y) - // 3: Let z = pi/2 - y, such that x = cos(z) - // 4: From double angle formula; cos(2A) = 1 - 2 * sin^2(A): - // z = 2A, z/2 = A - // cos(z) = 1 - 2 * sin^2(z/2) - // 5: Make sin(z/2) subject of the formula: - // sin(z/2) = sqrt((1 - cos(z))/2) - // 6: Recall [3]; x = cos(z). Therefore: - // sin(z/2) = sqrt((1 - x)/2) - // 7: Let u = (1 - x)/2 - // 8: Therefore: - // asin(sqrt(u)) = z/2 - // 2 * asin(sqrt(u)) = z - // 9: Recall [3]; z = pi/2 - y. Therefore: - // y = pi/2 - z - // y = pi/2 - 2 * asin(sqrt(u)) - // 10: Recall [1], y = asin(x). Therefore: - // asin(x) = pi/2 - 2 * asin(sqrt(u)) - // 11: Recall that: acos(x) = pi/2 + asin(-x) = pi/2 - asin(x) - // Therefore: - // acos(x) = pi/2 - (pi/2 - 2 * asin(sqrt(u))) - // acos(x) = 2 * asin(sqrt(u)) - // acospi(x) = 2 * (asin(sqrt(u)) / pi) - // - // THE RANGE REDUCTION, HOW? - // 12: Recall [7], u = (1 - x)/2 - // 13: Since 0.5 < x <= 1, therefore: - // 0 <= u <= 0.25 and 0 <= sqrt(u) <= 0.5 - // - // Hence, we can reuse the same [0, 0.5] domain polynomial approximation for - // Step [11] as `sqrt(u)` is in range. - // When -1 < x <= -0.5, the identity: - // acos(x) = pi - acos(-x) - // acospi(x) = 1 - acos(-x)/pi - // allows us to compute for the negative x value (lhs) - // with a positive x value instead (rhs). - - float xf_abs = (xf < 0 ? -xf : xf); - float u = fputil::multiply_add(-0.5f, xf_abs, 0.5f); - float sqrt_u = fputil::sqrt<float>(u); - - float asin_sqrt_u = - sqrt_u * fputil::polyeval(u, POLY_COEFFS[0], POLY_COEFFS[1], - POLY_COEFFS[2], POLY_COEFFS[3], POLY_COEFFS[4]); - - // Same as acos(x), but devided the expression with pi - return fputil::cast<float16>( - x_sign ? fputil::multiply_add(-2.0f, asin_sqrt_u, 1.0f) - : 2.0f * asin_sqrt_u); + return math::acospif16(x); } + } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/math/generic/asin.cpp b/libc/src/math/generic/asin.cpp index c033597..b5ba9ea 100644 --- a/libc/src/math/generic/asin.cpp +++ b/libc/src/math/generic/asin.cpp @@ -7,24 +7,12 @@ //===----------------------------------------------------------------------===// #include "src/math/asin.h" -#include "src/__support/FPUtil/FEnvImpl.h" -#include "src/__support/FPUtil/FPBits.h" -#include "src/__support/FPUtil/PolyEval.h" -#include "src/__support/FPUtil/double_double.h" -#include "src/__support/FPUtil/dyadic_float.h" -#include "src/__support/FPUtil/multiply_add.h" -#include "src/__support/FPUtil/sqrt.h" -#include "src/__support/macros/config.h" -#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY -#include "src/__support/macros/properties/cpu_features.h" // LIBC_TARGET_CPU_HAS_FMA -#include "src/__support/math/asin_utils.h" +#include "src/__support/math/asin.h" namespace LIBC_NAMESPACE_DECL { -using DoubleDouble = fputil::DoubleDouble; -using Float128 = fputil::DyadicFloat<128>; - LLVM_LIBC_FUNCTION(double, asin, (double x)) { + using namespace asin_internal; using FPBits = fputil::FPBits<double>; FPBits xbits(x); diff --git a/libc/src/math/generic/asinf.cpp b/libc/src/math/generic/asinf.cpp index 12383bf..9c6766f 100644 --- a/libc/src/math/generic/asinf.cpp +++ b/libc/src/math/generic/asinf.cpp @@ -7,160 +7,10 @@ //===----------------------------------------------------------------------===// #include "src/math/asinf.h" -#include "src/__support/FPUtil/FEnvImpl.h" -#include "src/__support/FPUtil/FPBits.h" -#include "src/__support/FPUtil/PolyEval.h" -#include "src/__support/FPUtil/except_value_utils.h" -#include "src/__support/FPUtil/multiply_add.h" -#include "src/__support/FPUtil/sqrt.h" -#include "src/__support/macros/config.h" -#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY -#include "src/__support/macros/properties/cpu_features.h" // LIBC_TARGET_CPU_HAS_FMA - -#include "inv_trigf_utils.h" +#include "src/__support/math/asinf.h" namespace LIBC_NAMESPACE_DECL { -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS -static constexpr size_t N_EXCEPTS = 2; - -// Exceptional values when |x| <= 0.5 -static constexpr fputil::ExceptValues<float, N_EXCEPTS> ASINF_EXCEPTS_LO = {{ - // (inputs, RZ output, RU offset, RD offset, RN offset) - // x = 0x1.137f0cp-5, asinf(x) = 0x1.138c58p-5 (RZ) - {0x3d09bf86, 0x3d09c62c, 1, 0, 1}, - // x = 0x1.cbf43cp-4, asinf(x) = 0x1.cced1cp-4 (RZ) - {0x3de5fa1e, 0x3de6768e, 1, 0, 0}, -}}; - -// Exceptional values when 0.5 < |x| <= 1 -static constexpr fputil::ExceptValues<float, N_EXCEPTS> ASINF_EXCEPTS_HI = {{ - // (inputs, RZ output, RU offset, RD offset, RN offset) - // x = 0x1.107434p-1, asinf(x) = 0x1.1f4b64p-1 (RZ) - {0x3f083a1a, 0x3f0fa5b2, 1, 0, 0}, - // x = 0x1.ee836cp-1, asinf(x) = 0x1.4f0654p0 (RZ) - {0x3f7741b6, 0x3fa7832a, 1, 0, 0}, -}}; -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS - -LLVM_LIBC_FUNCTION(float, asinf, (float x)) { - using FPBits = typename fputil::FPBits<float>; - - FPBits xbits(x); - uint32_t x_uint = xbits.uintval(); - uint32_t x_abs = xbits.uintval() & 0x7fff'ffffU; - constexpr double SIGN[2] = {1.0, -1.0}; - uint32_t x_sign = x_uint >> 31; - - // |x| <= 0.5-ish - if (x_abs < 0x3f04'471dU) { - // |x| < 0x1.d12edp-12 - if (LIBC_UNLIKELY(x_abs < 0x39e8'9768U)) { - // When |x| < 2^-12, the relative error of the approximation asin(x) ~ x - // is: - // |asin(x) - x| / |asin(x)| < |x^3| / (6|x|) - // = x^2 / 6 - // < 2^-25 - // < epsilon(1)/2. - // So the correctly rounded values of asin(x) are: - // = x + sign(x)*eps(x) if rounding mode = FE_TOWARDZERO, - // or (rounding mode = FE_UPWARD and x is - // negative), - // = x otherwise. - // To simplify the rounding decision and make it more efficient, we use - // fma(x, 2^-25, x) instead. - // An exhaustive test shows that this formula work correctly for all - // rounding modes up to |x| < 0x1.d12edp-12. - // Note: to use the formula x + 2^-25*x to decide the correct rounding, we - // do need fma(x, 2^-25, x) to prevent underflow caused by 2^-25*x when - // |x| < 2^-125. For targets without FMA instructions, we simply use - // double for intermediate results as it is more efficient than using an - // emulated version of FMA. -#if defined(LIBC_TARGET_CPU_HAS_FMA_FLOAT) - return fputil::multiply_add(x, 0x1.0p-25f, x); -#else - double xd = static_cast<double>(x); - return static_cast<float>(fputil::multiply_add(xd, 0x1.0p-25, xd)); -#endif // LIBC_TARGET_CPU_HAS_FMA_FLOAT - } - -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS - // Check for exceptional values - if (auto r = ASINF_EXCEPTS_LO.lookup_odd(x_abs, x_sign); - LIBC_UNLIKELY(r.has_value())) - return r.value(); -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS - - // For |x| <= 0.5, we approximate asinf(x) by: - // asin(x) = x * P(x^2) - // Where P(X^2) = Q(X) is a degree-20 minimax even polynomial approximating - // asin(x)/x on [0, 0.5] generated by Sollya with: - // > Q = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20|], - // [|1, D...|], [0, 0.5]); - // An exhaustive test shows that this approximation works well up to a - // little more than 0.5. - double xd = static_cast<double>(x); - double xsq = xd * xd; - double x3 = xd * xsq; - double r = asin_eval(xsq); - return static_cast<float>(fputil::multiply_add(x3, r, xd)); - } - - // |x| > 1, return NaNs. - if (LIBC_UNLIKELY(x_abs > 0x3f80'0000U)) { - if (xbits.is_signaling_nan()) { - fputil::raise_except_if_required(FE_INVALID); - return FPBits::quiet_nan().get_val(); - } - - if (x_abs <= 0x7f80'0000U) { - fputil::set_errno_if_required(EDOM); - fputil::raise_except_if_required(FE_INVALID); - } - - return FPBits::quiet_nan().get_val(); - } - -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS - // Check for exceptional values - if (auto r = ASINF_EXCEPTS_HI.lookup_odd(x_abs, x_sign); - LIBC_UNLIKELY(r.has_value())) - return r.value(); -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS - - // When |x| > 0.5, we perform range reduction as follow: - // - // Assume further that 0.5 < x <= 1, and let: - // y = asin(x) - // We will use the double angle formula: - // cos(2y) = 1 - 2 sin^2(y) - // and the complement angle identity: - // x = sin(y) = cos(pi/2 - y) - // = 1 - 2 sin^2 (pi/4 - y/2) - // So: - // sin(pi/4 - y/2) = sqrt( (1 - x)/2 ) - // And hence: - // pi/4 - y/2 = asin( sqrt( (1 - x)/2 ) ) - // Equivalently: - // asin(x) = y = pi/2 - 2 * asin( sqrt( (1 - x)/2 ) ) - // Let u = (1 - x)/2, then: - // asin(x) = pi/2 - 2 * asin( sqrt(u) ) - // Moreover, since 0.5 < x <= 1: - // 0 <= u < 1/4, and 0 <= sqrt(u) < 0.5, - // And hence we can reuse the same polynomial approximation of asin(x) when - // |x| <= 0.5: - // asin(x) ~ pi/2 - 2 * sqrt(u) * P(u), - - xbits.set_sign(Sign::POS); - double sign = SIGN[x_sign]; - double xd = static_cast<double>(xbits.get_val()); - double u = fputil::multiply_add(-0.5, xd, 0.5); - double c1 = sign * (-2 * fputil::sqrt<double>(u)); - double c2 = fputil::multiply_add(sign, M_MATH_PI_2, c1); - double c3 = c1 * u; - - double r = asin_eval(u); - return static_cast<float>(fputil::multiply_add(c3, r, c2)); -} +LLVM_LIBC_FUNCTION(float, asinf, (float x)) { return math::asinf(x); } } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/math/generic/asinf16.cpp b/libc/src/math/generic/asinf16.cpp index 518c384..af8dbfe 100644 --- a/libc/src/math/generic/asinf16.cpp +++ b/libc/src/math/generic/asinf16.cpp @@ -7,127 +7,10 @@ //===----------------------------------------------------------------------===// #include "src/math/asinf16.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/multiply_add.h" -#include "src/__support/FPUtil/sqrt.h" -#include "src/__support/macros/optimization.h" +#include "src/__support/math/asinf16.h" namespace LIBC_NAMESPACE_DECL { -// Generated by Sollya using the following command: -// > round(pi/2, D, RN); -static constexpr float PI_2 = 0x1.921fb54442d18p0f; - -LLVM_LIBC_FUNCTION(float16, asinf16, (float16 x)) { - using FPBits = fputil::FPBits<float16>; - FPBits xbits(x); - - uint16_t x_u = xbits.uintval(); - uint16_t x_abs = x_u & 0x7fff; - float xf = x; - - // |x| > 0x1p0, |x| > 1, or x is NaN. - if (LIBC_UNLIKELY(x_abs > 0x3c00)) { - // asinf16(NaN) = 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; - } - - // 1 < |x| <= +/-inf - fputil::raise_except_if_required(FE_INVALID); - fputil::set_errno_if_required(EDOM); - - return FPBits::quiet_nan().get_val(); - } - - float xsq = xf * xf; - - // |x| <= 0x1p-1, |x| <= 0.5 - if (x_abs <= 0x3800) { - // asinf16(+/-0) = +/-0 - if (LIBC_UNLIKELY(x_abs == 0)) - return x; - - // Exhaustive tests show that, - // for |x| <= 0x1.878p-9, when: - // x > 0, and rounding upward, or - // x < 0, and rounding downward, then, - // asin(x) = x * 2^-11 + x - // else, in other rounding modes, - // asin(x) = x - if (LIBC_UNLIKELY(x_abs <= 0x1a1e)) { - int rounding = fputil::quick_get_round(); - - if ((xbits.is_pos() && rounding == FE_UPWARD) || - (xbits.is_neg() && rounding == FE_DOWNWARD)) - return fputil::cast<float16>(fputil::multiply_add(xf, 0x1.0p-11f, xf)); - return x; - } - - // Degree-6 minimax odd polynomial of asin(x) generated by Sollya with: - // > P = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 0.5]); - float result = - fputil::polyeval(xsq, 0x1.000002p0f, 0x1.554c2ap-3f, 0x1.3541ccp-4f, - 0x1.43b2d6p-5f, 0x1.a0d73ep-5f); - return fputil::cast<float16>(xf * result); - } - - // When |x| > 0.5, assume that 0.5 < |x| <= 1, - // - // Step-by-step range-reduction proof: - // 1: Let y = asin(x), such that, x = sin(y) - // 2: From complimentary angle identity: - // x = sin(y) = cos(pi/2 - y) - // 3: Let z = pi/2 - y, such that x = cos(z) - // 4: From double angle formula; cos(2A) = 1 - sin^2(A): - // z = 2A, z/2 = A - // cos(z) = 1 - 2 * sin^2(z/2) - // 5: Make sin(z/2) subject of the formula: - // sin(z/2) = sqrt((1 - cos(z))/2) - // 6: Recall [3]; x = cos(z). Therefore: - // sin(z/2) = sqrt((1 - x)/2) - // 7: Let u = (1 - x)/2 - // 8: Therefore: - // asin(sqrt(u)) = z/2 - // 2 * asin(sqrt(u)) = z - // 9: Recall [3], z = pi/2 - y. Therefore: - // y = pi/2 - z - // y = pi/2 - 2 * asin(sqrt(u)) - // 10: Recall [1], y = asin(x). Therefore: - // asin(x) = pi/2 - 2 * asin(sqrt(u)) - // - // WHY? - // 11: Recall [7], u = (1 - x)/2 - // 12: Since 0.5 < x <= 1, therefore: - // 0 <= u <= 0.25 and 0 <= sqrt(u) <= 0.5 - // - // Hence, we can reuse the same [0, 0.5] domain polynomial approximation for - // Step [10] as `sqrt(u)` is in range. - - // 0x1p-1 < |x| <= 0x1p0, 0.5 < |x| <= 1.0 - float xf_abs = (xf < 0 ? -xf : xf); - float sign = (xbits.uintval() >> 15 == 1 ? -1.0 : 1.0); - float u = fputil::multiply_add(-0.5f, xf_abs, 0.5f); - float u_sqrt = fputil::sqrt<float>(u); - - // Degree-6 minimax odd polynomial of asin(x) generated by Sollya with: - // > P = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8|], [|SG...|], [0, 0.5]); - float asin_sqrt_u = - u_sqrt * fputil::polyeval(u, 0x1.000002p0f, 0x1.554c2ap-3f, - 0x1.3541ccp-4f, 0x1.43b2d6p-5f, 0x1.a0d73ep-5f); - - return fputil::cast<float16>(sign * - fputil::multiply_add(-2.0f, asin_sqrt_u, PI_2)); -} +LLVM_LIBC_FUNCTION(float16, asinf16, (float16 x)) { return math::asinf16(x); } } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/math/generic/asinhf.cpp b/libc/src/math/generic/asinhf.cpp index 0bb7065..3aed3bc 100644 --- a/libc/src/math/generic/asinhf.cpp +++ b/libc/src/math/generic/asinhf.cpp @@ -19,6 +19,7 @@ namespace LIBC_NAMESPACE_DECL { LLVM_LIBC_FUNCTION(float, asinhf, (float x)) { + using namespace acoshf_internal; using FPBits_t = typename fputil::FPBits<float>; FPBits_t xbits(x); uint32_t x_u = xbits.uintval(); diff --git a/libc/src/math/generic/asinhf16.cpp b/libc/src/math/generic/asinhf16.cpp index 7878632..0a0b471 100644 --- a/libc/src/math/generic/asinhf16.cpp +++ b/libc/src/math/generic/asinhf16.cpp @@ -49,6 +49,7 @@ static constexpr fputil::ExceptValues<float16, N_EXCEPTS> ASINHF16_EXCEPTS{{ #endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS LLVM_LIBC_FUNCTION(float16, asinhf16, (float16 x)) { + using namespace acoshf_internal; using FPBits = fputil::FPBits<float16>; FPBits xbits(x); diff --git a/libc/src/math/generic/atan2f.cpp b/libc/src/math/generic/atan2f.cpp index c04b0eb..32b977f 100644 --- a/libc/src/math/generic/atan2f.cpp +++ b/libc/src/math/generic/atan2f.cpp @@ -8,7 +8,6 @@ #include "src/math/atan2f.h" #include "hdr/fenv_macros.h" -#include "inv_trigf_utils.h" #include "src/__support/FPUtil/FEnvImpl.h" #include "src/__support/FPUtil/FPBits.h" #include "src/__support/FPUtil/PolyEval.h" @@ -18,6 +17,7 @@ #include "src/__support/FPUtil/rounding_mode.h" #include "src/__support/macros/config.h" #include "src/__support/macros/optimization.h" // LIBC_UNLIKELY +#include "src/__support/math/inv_trigf_utils.h" #if defined(LIBC_MATH_HAS_SKIP_ACCURATE_PASS) && \ defined(LIBC_MATH_HAS_INTERMEDIATE_COMP_IN_FLOAT) @@ -236,6 +236,7 @@ float atan2f_double_double(double num_d, double den_d, double q_d, int idx, // which is about rounding errors of double-double (2^-104). LLVM_LIBC_FUNCTION(float, atan2f, (float y, float x)) { + using namespace inv_trigf_utils_internal; using FPBits = typename fputil::FPBits<float>; constexpr double IS_NEG[2] = {1.0, -1.0}; constexpr double PI = 0x1.921fb54442d18p1; diff --git a/libc/src/math/generic/atanf.cpp b/libc/src/math/generic/atanf.cpp index 46196dbe..22f962e 100644 --- a/libc/src/math/generic/atanf.cpp +++ b/libc/src/math/generic/atanf.cpp @@ -7,7 +7,6 @@ //===----------------------------------------------------------------------===// #include "src/math/atanf.h" -#include "inv_trigf_utils.h" #include "src/__support/FPUtil/FPBits.h" #include "src/__support/FPUtil/PolyEval.h" #include "src/__support/FPUtil/except_value_utils.h" @@ -16,10 +15,12 @@ #include "src/__support/FPUtil/rounding_mode.h" #include "src/__support/macros/config.h" #include "src/__support/macros/optimization.h" // LIBC_UNLIKELY +#include "src/__support/math/inv_trigf_utils.h" namespace LIBC_NAMESPACE_DECL { LLVM_LIBC_FUNCTION(float, atanf, (float x)) { + using namespace inv_trigf_utils_internal; using FPBits = typename fputil::FPBits<float>; constexpr double FINAL_SIGN[2] = {1.0, -1.0}; diff --git a/libc/src/math/generic/atanhf.cpp b/libc/src/math/generic/atanhf.cpp index f6fde76..602a8f0 100644 --- a/libc/src/math/generic/atanhf.cpp +++ b/libc/src/math/generic/atanhf.cpp @@ -16,6 +16,7 @@ namespace LIBC_NAMESPACE_DECL { LLVM_LIBC_FUNCTION(float, atanhf, (float x)) { + using namespace acoshf_internal; using FPBits = typename fputil::FPBits<float>; FPBits xbits(x); diff --git a/libc/src/math/generic/common_constants.cpp b/libc/src/math/generic/common_constants.cpp index 4dcf84d..42e3ff0 100644 --- a/libc/src/math/generic/common_constants.cpp +++ b/libc/src/math/generic/common_constants.cpp @@ -51,52 +51,6 @@ const float ONE_OVER_F_FLOAT[128] = { 0x1.08421p-1f, 0x1.07326p-1f, 0x1.0624dep-1f, 0x1.05198p-1f, 0x1.041042p-1f, 0x1.03091cp-1f, 0x1.020408p-1f, 0x1.010102p-1f}; -// Lookup table for (1/f) where f = 1 + n*2^(-7), n = 0..127. -const double ONE_OVER_F[128] = { - 0x1.0000000000000p+0, 0x1.fc07f01fc07f0p-1, 0x1.f81f81f81f820p-1, - 0x1.f44659e4a4271p-1, 0x1.f07c1f07c1f08p-1, 0x1.ecc07b301ecc0p-1, - 0x1.e9131abf0b767p-1, 0x1.e573ac901e574p-1, 0x1.e1e1e1e1e1e1ep-1, - 0x1.de5d6e3f8868ap-1, 0x1.dae6076b981dbp-1, 0x1.d77b654b82c34p-1, - 0x1.d41d41d41d41dp-1, 0x1.d0cb58f6ec074p-1, 0x1.cd85689039b0bp-1, - 0x1.ca4b3055ee191p-1, 0x1.c71c71c71c71cp-1, 0x1.c3f8f01c3f8f0p-1, - 0x1.c0e070381c0e0p-1, 0x1.bdd2b899406f7p-1, 0x1.bacf914c1bad0p-1, - 0x1.b7d6c3dda338bp-1, 0x1.b4e81b4e81b4fp-1, 0x1.b2036406c80d9p-1, - 0x1.af286bca1af28p-1, 0x1.ac5701ac5701bp-1, 0x1.a98ef606a63bep-1, - 0x1.a6d01a6d01a6dp-1, 0x1.a41a41a41a41ap-1, 0x1.a16d3f97a4b02p-1, - 0x1.9ec8e951033d9p-1, 0x1.9c2d14ee4a102p-1, 0x1.999999999999ap-1, - 0x1.970e4f80cb872p-1, 0x1.948b0fcd6e9e0p-1, 0x1.920fb49d0e229p-1, - 0x1.8f9c18f9c18fap-1, 0x1.8d3018d3018d3p-1, 0x1.8acb90f6bf3aap-1, - 0x1.886e5f0abb04ap-1, 0x1.8618618618618p-1, 0x1.83c977ab2beddp-1, - 0x1.8181818181818p-1, 0x1.7f405fd017f40p-1, 0x1.7d05f417d05f4p-1, - 0x1.7ad2208e0ecc3p-1, 0x1.78a4c8178a4c8p-1, 0x1.767dce434a9b1p-1, - 0x1.745d1745d1746p-1, 0x1.724287f46debcp-1, 0x1.702e05c0b8170p-1, - 0x1.6e1f76b4337c7p-1, 0x1.6c16c16c16c17p-1, 0x1.6a13cd1537290p-1, - 0x1.6816816816817p-1, 0x1.661ec6a5122f9p-1, 0x1.642c8590b2164p-1, - 0x1.623fa77016240p-1, 0x1.6058160581606p-1, 0x1.5e75bb8d015e7p-1, - 0x1.5c9882b931057p-1, 0x1.5ac056b015ac0p-1, 0x1.58ed2308158edp-1, - 0x1.571ed3c506b3ap-1, 0x1.5555555555555p-1, 0x1.5390948f40febp-1, - 0x1.51d07eae2f815p-1, 0x1.5015015015015p-1, 0x1.4e5e0a72f0539p-1, - 0x1.4cab88725af6ep-1, 0x1.4afd6a052bf5bp-1, 0x1.49539e3b2d067p-1, - 0x1.47ae147ae147bp-1, 0x1.460cbc7f5cf9ap-1, 0x1.446f86562d9fbp-1, - 0x1.42d6625d51f87p-1, 0x1.4141414141414p-1, 0x1.3fb013fb013fbp-1, - 0x1.3e22cbce4a902p-1, 0x1.3c995a47babe7p-1, 0x1.3b13b13b13b14p-1, - 0x1.3991c2c187f63p-1, 0x1.3813813813814p-1, 0x1.3698df3de0748p-1, - 0x1.3521cfb2b78c1p-1, 0x1.33ae45b57bcb2p-1, 0x1.323e34a2b10bfp-1, - 0x1.30d190130d190p-1, 0x1.2f684bda12f68p-1, 0x1.2e025c04b8097p-1, - 0x1.2c9fb4d812ca0p-1, 0x1.2b404ad012b40p-1, 0x1.29e4129e4129ep-1, - 0x1.288b01288b013p-1, 0x1.27350b8812735p-1, 0x1.25e22708092f1p-1, - 0x1.2492492492492p-1, 0x1.23456789abcdfp-1, 0x1.21fb78121fb78p-1, - 0x1.20b470c67c0d9p-1, 0x1.1f7047dc11f70p-1, 0x1.1e2ef3b3fb874p-1, - 0x1.1cf06ada2811dp-1, 0x1.1bb4a4046ed29p-1, 0x1.1a7b9611a7b96p-1, - 0x1.19453808ca29cp-1, 0x1.1811811811812p-1, 0x1.16e0689427379p-1, - 0x1.15b1e5f75270dp-1, 0x1.1485f0e0acd3bp-1, 0x1.135c81135c811p-1, - 0x1.12358e75d3033p-1, 0x1.1111111111111p-1, 0x1.0fef010fef011p-1, - 0x1.0ecf56be69c90p-1, 0x1.0db20a88f4696p-1, 0x1.0c9714fbcda3bp-1, - 0x1.0b7e6ec259dc8p-1, 0x1.0a6810a6810a7p-1, 0x1.0953f39010954p-1, - 0x1.0842108421084p-1, 0x1.073260a47f7c6p-1, 0x1.0624dd2f1a9fcp-1, - 0x1.05197f7d73404p-1, 0x1.0410410410410p-1, 0x1.03091b51f5e1ap-1, - 0x1.0204081020408p-1, 0x1.0101010101010p-1}; - // Lookup table for log(f) = log(1 + n*2^(-7)) where n = 0..127, // computed and stored as float precision constants. // Generated by Sollya with the following commands: @@ -136,52 +90,6 @@ const float LOG_F_FLOAT[128] = { 0x1.52a2d2p-1f, 0x1.54b246p-1f, 0x1.56bf9ep-1f, 0x1.58cadcp-1f, 0x1.5ad404p-1f, 0x1.5cdb1ep-1f, 0x1.5ee02ap-1f, 0x1.60e33p-1f}; -// Lookup table for log(f) = log(1 + n*2^(-7)) where n = 0..127. -const double LOG_F[128] = { - 0x0.0000000000000p+0, 0x1.fe02a6b106788p-8, 0x1.fc0a8b0fc03e3p-7, - 0x1.7b91b07d5b11ap-6, 0x1.f829b0e783300p-6, 0x1.39e87b9febd5fp-5, - 0x1.77458f632dcfcp-5, 0x1.b42dd711971bep-5, 0x1.f0a30c01162a6p-5, - 0x1.16536eea37ae0p-4, 0x1.341d7961bd1d0p-4, 0x1.51b073f06183fp-4, - 0x1.6f0d28ae56b4bp-4, 0x1.8c345d6319b20p-4, 0x1.a926d3a4ad563p-4, - 0x1.c5e548f5bc743p-4, 0x1.e27076e2af2e5p-4, 0x1.fec9131dbeabap-4, - 0x1.0d77e7cd08e59p-3, 0x1.1b72ad52f67a0p-3, 0x1.29552f81ff523p-3, - 0x1.371fc201e8f74p-3, 0x1.44d2b6ccb7d1ep-3, 0x1.526e5e3a1b437p-3, - 0x1.5ff3070a793d3p-3, 0x1.6d60fe719d21cp-3, 0x1.7ab890210d909p-3, - 0x1.87fa06520c910p-3, 0x1.9525a9cf456b4p-3, 0x1.a23bc1fe2b563p-3, - 0x1.af3c94e80bff2p-3, 0x1.bc286742d8cd6p-3, 0x1.c8ff7c79a9a21p-3, - 0x1.d5c216b4fbb91p-3, 0x1.e27076e2af2e5p-3, 0x1.ef0adcbdc5936p-3, - 0x1.fb9186d5e3e2ap-3, 0x1.0402594b4d040p-2, 0x1.0a324e27390e3p-2, - 0x1.1058bf9ae4ad5p-2, 0x1.1675cababa60ep-2, 0x1.1c898c16999fap-2, - 0x1.22941fbcf7965p-2, 0x1.2895a13de86a3p-2, 0x1.2e8e2bae11d30p-2, - 0x1.347dd9a987d54p-2, 0x1.3a64c556945e9p-2, 0x1.404308686a7e3p-2, - 0x1.4618bc21c5ec2p-2, 0x1.4be5f957778a0p-2, 0x1.51aad872df82dp-2, - 0x1.5767717455a6cp-2, 0x1.5d1bdbf5809cap-2, 0x1.62c82f2b9c795p-2, - 0x1.686c81e9b14aep-2, 0x1.6e08eaa2ba1e3p-2, 0x1.739d7f6bbd006p-2, - 0x1.792a55fdd47a2p-2, 0x1.7eaf83b82afc3p-2, 0x1.842d1da1e8b17p-2, - 0x1.89a3386c1425ap-2, 0x1.8f11e873662c7p-2, 0x1.947941c2116fap-2, - 0x1.99d958117e08ap-2, 0x1.9f323ecbf984bp-2, 0x1.a484090e5bb0ap-2, - 0x1.a9cec9a9a0849p-2, 0x1.af1293247786bp-2, 0x1.b44f77bcc8f62p-2, - 0x1.b9858969310fbp-2, 0x1.beb4d9da71b7bp-2, 0x1.c3dd7a7cdad4dp-2, - 0x1.c8ff7c79a9a21p-2, 0x1.ce1af0b85f3ebp-2, 0x1.d32fe7e00ebd5p-2, - 0x1.d83e7258a2f3ep-2, 0x1.dd46a04c1c4a0p-2, 0x1.e24881a7c6c26p-2, - 0x1.e744261d68787p-2, 0x1.ec399d2468cc0p-2, 0x1.f128f5faf06ecp-2, - 0x1.f6123fa7028acp-2, 0x1.faf588f78f31ep-2, 0x1.ffd2e0857f498p-2, - 0x1.02552a5a5d0fep-1, 0x1.04bdf9da926d2p-1, 0x1.0723e5c1cdf40p-1, - 0x1.0986f4f573520p-1, 0x1.0be72e4252a82p-1, 0x1.0e44985d1cc8bp-1, - 0x1.109f39e2d4c96p-1, 0x1.12f719593efbcp-1, 0x1.154c3d2f4d5e9p-1, - 0x1.179eabbd899a0p-1, 0x1.19ee6b467c96ep-1, 0x1.1c3b81f713c24p-1, - 0x1.1e85f5e7040d0p-1, 0x1.20cdcd192ab6dp-1, 0x1.23130d7bebf42p-1, - 0x1.2555bce98f7cbp-1, 0x1.2795e1289b11ap-1, 0x1.29d37fec2b08ap-1, - 0x1.2c0e9ed448e8bp-1, 0x1.2e47436e40268p-1, 0x1.307d7334f10bep-1, - 0x1.32b1339121d71p-1, 0x1.34e289d9ce1d3p-1, 0x1.37117b54747b5p-1, - 0x1.393e0d3562a19p-1, 0x1.3b68449fffc22p-1, 0x1.3d9026a7156fap-1, - 0x1.3fb5b84d16f42p-1, 0x1.41d8fe84672aep-1, 0x1.43f9fe2f9ce67p-1, - 0x1.4618bc21c5ec2p-1, 0x1.48353d1ea88dfp-1, 0x1.4a4f85db03ebbp-1, - 0x1.4c679afccee39p-1, 0x1.4e7d811b75bb0p-1, 0x1.50913cc01686bp-1, - 0x1.52a2d265bc5aap-1, 0x1.54b2467999497p-1, 0x1.56bf9d5b3f399p-1, - 0x1.58cadb5cd7989p-1, 0x1.5ad404c359f2cp-1, 0x1.5cdb1dc6c1764p-1, - 0x1.5ee02a9241675p-1, 0x1.60e32f44788d8p-1}; - // Range reduction constants for logarithms. // r(0) = 1, r(127) = 0.5 // r(k) = 2^-8 * ceil(2^8 * (1 - 2^-8) / (1 + k*2^-7)) diff --git a/libc/src/math/generic/common_constants.h b/libc/src/math/generic/common_constants.h index 291816a..72b1d564 100644 --- a/libc/src/math/generic/common_constants.h +++ b/libc/src/math/generic/common_constants.h @@ -11,6 +11,7 @@ #include "src/__support/FPUtil/triple_double.h" #include "src/__support/macros/config.h" +#include "src/__support/math/acosh_float_constants.h" #include "src/__support/math/exp_constants.h" #include "src/__support/number_pair.h" @@ -20,16 +21,10 @@ namespace LIBC_NAMESPACE_DECL { // computed and stored as float precision constants. extern const float ONE_OVER_F_FLOAT[128]; -// Lookup table for (1/f) where f = 1 + n*2^(-7), n = 0..127. -extern const double ONE_OVER_F[128]; - // Lookup table for log(f) = log(1 + n*2^(-7)) where n = 0..127, // computed and stored as float precision constants. extern const float LOG_F_FLOAT[128]; -// Lookup table for log(f) = log(1 + n*2^(-7)) where n = 0..127. -extern const double LOG_F[128]; - // Lookup table for range reduction constants r for logarithms. extern const float R[128]; diff --git a/libc/src/math/generic/erff.cpp b/libc/src/math/generic/erff.cpp index 44607a5..003b346 100644 --- a/libc/src/math/generic/erff.cpp +++ b/libc/src/math/generic/erff.cpp @@ -7,180 +7,10 @@ //===----------------------------------------------------------------------===// #include "src/math/erff.h" -#include "src/__support/FPUtil/FPBits.h" -#include "src/__support/FPUtil/PolyEval.h" -#include "src/__support/FPUtil/except_value_utils.h" -#include "src/__support/FPUtil/multiply_add.h" -#include "src/__support/common.h" -#include "src/__support/macros/config.h" -#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY +#include "src/__support/math/erff.h" namespace LIBC_NAMESPACE_DECL { -// Polynomials approximating erf(x)/x on ( k/8, (k + 1)/8 ) generated by Sollya -// with: -// > P = fpminimax(erf(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14|], [|D...|], -// [k/8, (k + 1)/8]); -// for k = 0..31. -constexpr double COEFFS[32][8] = { - {0x1.20dd750429b6dp0, -0x1.812746b037753p-2, 0x1.ce2f219e8596ap-4, - -0x1.b82cdacb78fdap-6, 0x1.56479297dfda5p-8, -0x1.8b3ac5455ef02p-11, - -0x1.126fcac367e3bp-8, 0x1.2d0bdb3ba4984p-4}, - {0x1.20dd750429b6dp0, -0x1.812746b0379a8p-2, 0x1.ce2f21a03cf2ap-4, - -0x1.b82ce30de083ep-6, 0x1.565bcad3eb60fp-8, -0x1.c02c66f659256p-11, - 0x1.f92f673385229p-14, -0x1.def402648ae9p-17}, - {0x1.20dd750429b34p0, -0x1.812746b032dcep-2, 0x1.ce2f219d84aaep-4, - -0x1.b82ce22dcf139p-6, 0x1.565b9efcd4af1p-8, -0x1.c021f1af414bcp-11, - 0x1.f7c6d177eff82p-14, -0x1.c9e4410dcf865p-17}, - {0x1.20dd750426eabp0, -0x1.812746ae592c7p-2, 0x1.ce2f211525f14p-4, - -0x1.b82ccc125e63fp-6, 0x1.56596f261cfd3p-8, -0x1.bfde1ff8eeecfp-11, - 0x1.f31a9d15dc5d8p-14, -0x1.a5a4362844b3cp-17}, - {0x1.20dd75039c705p0, -0x1.812746777e74dp-2, 0x1.ce2f17af98a1bp-4, - -0x1.b82be4b817cbep-6, 0x1.564bec2e2962ep-8, -0x1.bee86f9da3558p-11, - 0x1.e9443689dc0ccp-14, -0x1.79c0f230805d8p-17}, - {0x1.20dd74f811211p0, -0x1.81274371a3e8fp-2, 0x1.ce2ec038262e5p-4, - -0x1.b8265b82c5e1fp-6, 0x1.5615a2e239267p-8, -0x1.bc63ae023dcebp-11, - 0x1.d87c2102f7e06p-14, -0x1.49584bea41d62p-17}, - {0x1.20dd746d063e3p0, -0x1.812729a8a950fp-2, 0x1.ce2cb0a2df232p-4, - -0x1.b80eca1f51278p-6, 0x1.5572e26c46815p-8, -0x1.b715e5638b65ep-11, - 0x1.bfbb195484968p-14, -0x1.177a565c15c52p-17}, - {0x1.20dd701b44486p0, -0x1.812691145f237p-2, 0x1.ce23a06b8cfd9p-4, - -0x1.b7c1dc7245288p-6, 0x1.53e92f7f397ddp-8, -0x1.ad97cc4acf0b2p-11, - 0x1.9f028b2b09b71p-14, -0x1.cdc4da08da8c1p-18}, - {0x1.20dd5715ac332p0, -0x1.8123e680bd0ebp-2, 0x1.ce0457aded691p-4, - -0x1.b6f52d52bed4p-6, 0x1.50c291b84414cp-8, -0x1.9ea246b1ad4a9p-11, - 0x1.77654674e0cap-14, -0x1.737c11a1bcebbp-18}, - {0x1.20dce6593e114p0, -0x1.811a59c02eadcp-2, 0x1.cdab53c7cd7d5p-4, - -0x1.b526d2e321eedp-6, 0x1.4b1d32cd8b994p-8, -0x1.8963143ec0a1ep-11, - 0x1.4ad5700e4db91p-14, -0x1.231e100e43ef2p-18}, - {0x1.20db48bfd5a62p0, -0x1.80fdd84f9e308p-2, 0x1.ccd340d462983p-4, - -0x1.b196a2928768p-6, 0x1.4210c2c13a0f7p-8, -0x1.6dbdfb4ff71aep-11, - 0x1.1bca2d17fbd71p-14, -0x1.bca36f90c7cf5p-19}, - {0x1.20d64b2f8f508p0, -0x1.80b4d4f19fa8bp-2, 0x1.cb088197262e3p-4, - -0x1.ab51fd02e5b99p-6, 0x1.34e1e5e81a632p-8, -0x1.4c66377b502cep-11, - 0x1.d9ad25066213cp-15, -0x1.4b0df7dd0cfa1p-19}, - {0x1.20c8fc1243576p0, -0x1.8010cb2009e27p-2, 0x1.c7a47e9299315p-4, - -0x1.a155be5683654p-6, 0x1.233502694997bp-8, -0x1.26c94b7d813p-11, - 0x1.8094f1de25fb9p-15, -0x1.e0e3d776c6eefp-20}, - {0x1.20a9bd1611bc1p0, -0x1.7ec7fbce83f9p-2, 0x1.c1d757d7317b7p-4, - -0x1.92c160cd589fp-6, 0x1.0d307269cc5c2p-8, -0x1.fda5b0d2d1879p-12, - 0x1.2fdd7b3b14a7fp-15, -0x1.54eed4a26af5ap-20}, - {0x1.20682834f943dp0, -0x1.7c73f747bf5a9p-2, 0x1.b8c2db4a9ffd1p-4, - -0x1.7f0e4ffe989ecp-6, 0x1.e7061eae4166ep-9, -0x1.ad36e873fff2dp-12, - 0x1.d39222396128ep-16, -0x1.d83dacec5ea6bp-21}, - {0x1.1feb8d12676d7p0, -0x1.7898347284afep-2, 0x1.aba3466b34451p-4, - -0x1.663adc573e2f9p-6, 0x1.ae99fb17c3e08p-9, -0x1.602f950ad5535p-12, - 0x1.5e9717490609dp-16, -0x1.3fca107bbc8d5p-21}, - {0x1.1f12fe3c536fap0, -0x1.72b1d1f22e6d3p-2, 0x1.99fc0eed4a896p-4, - -0x1.48db0a87bd8c6p-6, 0x1.73e368895aa61p-9, -0x1.19b35d5301fc8p-12, - 0x1.007987e4bb033p-16, -0x1.a7edcd4c2dc7p-22}, - {0x1.1db7b0df84d5dp0, -0x1.6a4e4a41cde02p-2, 0x1.83bbded16455dp-4, - -0x1.2809b3b36977ep-6, 0x1.39c08bab44679p-9, -0x1.b7b45a70ed119p-13, - 0x1.6e99b36410e7bp-17, -0x1.13619bb7ebc0cp-22}, - {0x1.1bb1c85c4a527p0, -0x1.5f23b99a249a3p-2, 0x1.694c91fa0d12cp-4, - -0x1.053e1ce11c72dp-6, 0x1.02bf72c50ea78p-9, -0x1.4f478fb56cb02p-13, - 0x1.005f80ecbe213p-17, -0x1.5f2446bde7f5bp-23}, - {0x1.18dec3bd51f9dp0, -0x1.5123f58346186p-2, 0x1.4b8a1ca536ab4p-4, - -0x1.c4243015cc723p-7, 0x1.a1a8a01d351efp-10, -0x1.f466b34f1d86bp-14, - 0x1.5f835eea0bf6ap-18, -0x1.b83165b939234p-24}, - {0x1.152804c3369f4p0, -0x1.4084cd4afd4bcp-2, 0x1.2ba2e836e47aap-4, - -0x1.800f2dfc6904bp-7, 0x1.4a6daf0669c59p-10, -0x1.6e326ab872317p-14, - 0x1.d9761a6a755a5p-19, -0x1.0fca33f9dd4b5p-24}, - {0x1.1087ad68356aap0, -0x1.2dbb044707459p-2, 0x1.0aea8ceaa0384p-4, - -0x1.40b516d52b3d2p-7, 0x1.00c9e05f01d22p-10, -0x1.076afb0dc0ff7p-14, - 0x1.39fadec400657p-19, -0x1.4b5761352e7e3p-25}, - {0x1.0b0a7a8ba4a22p0, -0x1.196990d22d4a1p-2, 0x1.d5551e6ac0c4dp-5, - -0x1.07cce1770bd1ap-7, 0x1.890347b8848bfp-11, -0x1.757ec96750b6ap-15, - 0x1.9b258a1e06bcep-20, -0x1.8fc6d22da7572p-26}, - {0x1.04ce2be70fb47p0, -0x1.0449e4b0b9cacp-2, 0x1.97f7424f4b0e7p-5, - -0x1.ac825439c42f4p-8, 0x1.28f5f65426dfbp-11, -0x1.05b699a90f90fp-15, - 0x1.0a888eecf4593p-20, -0x1.deace2b32bb31p-27}, - {0x1.fbf9fb0e11cc8p-1, -0x1.de2640856545ap-3, 0x1.5f5b1f47f851p-5, - -0x1.588bc71eb41b9p-8, 0x1.bc6a0a772f56dp-12, -0x1.6b9fad1f1657ap-16, - 0x1.573204ba66504p-21, -0x1.1d38065c94e44p-27}, - {0x1.ed8f18c99e031p-1, -0x1.b4cb6acd903b4p-3, 0x1.2c7f3dddd6fc1p-5, - -0x1.13052067df4ep-8, 0x1.4a5027444082fp-12, -0x1.f672bab0e2554p-17, - 0x1.b83c756348cc9p-22, -0x1.534f1a1079499p-28}, - {0x1.debd33044166dp-1, -0x1.8d7cd9053f7d8p-3, 0x1.ff9957fb3d6e7p-6, - -0x1.b50be55de0f36p-9, 0x1.e92c8ec53a628p-13, -0x1.5a4b88d508007p-17, - 0x1.1a27737559e26p-22, -0x1.942ae62cb2c14p-29}, - {0x1.cfdbf0386f3bdp-1, -0x1.68e33d93b0dc4p-3, 0x1.b2683d58f53dep-6, - -0x1.5a9174e70d26fp-9, 0x1.69ddd326d49cdp-13, -0x1.dd8f397a8219cp-18, - 0x1.6a755016ad4ddp-23, -0x1.e366e0139187dp-30}, - {0x1.c132adb8d7464p-1, -0x1.475a899f61b46p-3, 0x1.70a431397a77cp-6, - -0x1.12e3d35beeee2p-9, 0x1.0c16b05738333p-13, -0x1.4a47f873e144ep-18, - 0x1.d3d494c698c02p-24, -0x1.2302c59547fe5p-30}, - {0x1.b2f5fd05555e7p-1, -0x1.28feefbe03ec7p-3, 0x1.3923acbb3a676p-6, - -0x1.b4ff793cd6358p-10, 0x1.8ea0eb8c913bcp-14, -0x1.cb31ec2baceb1p-19, - 0x1.30011e7e80c04p-24, -0x1.617710635cb1dp-31}, - {0x1.a54853cd9593ep-1, -0x1.0dbdbaea4dc8ep-3, 0x1.0a93e2c20a0fdp-6, - -0x1.5c969ff401ea8p-10, 0x1.29e0cc64fe627p-14, -0x1.4160d8e9d3c2ap-19, - 0x1.8e7b67594624ap-25, -0x1.b1cf2c975b09bp-32}, - {0x1.983ceece09ff8p-1, -0x1.eacc78f7a2dp-4, 0x1.c74418410655fp-7, - -0x1.1756a050e441ep-10, 0x1.bff3650f7f548p-15, -0x1.c56c0217d3adap-20, - 0x1.07b4918d0b489p-25, -0x1.0d4be8c1c50f8p-32}, -}; - -LLVM_LIBC_FUNCTION(float, erff, (float x)) { - using FPBits = typename fputil::FPBits<float>; - FPBits xbits(x); - - uint32_t x_u = xbits.uintval(); - uint32_t x_abs = x_u & 0x7fff'ffffU; - - if (LIBC_UNLIKELY(x_abs >= 0x4080'0000U)) { - const float ONE[2] = {1.0f, -1.0f}; - const float SMALL[2] = {-0x1.0p-25f, 0x1.0p-25f}; - - int sign = xbits.is_neg() ? 1 : 0; - - if (LIBC_UNLIKELY(x_abs >= 0x7f80'0000U)) { - if (xbits.is_signaling_nan()) { - fputil::raise_except_if_required(FE_INVALID); - return FPBits::quiet_nan().get_val(); - } - return (x_abs > 0x7f80'0000) ? x : ONE[sign]; - } - - return ONE[sign] + SMALL[sign]; - } - -#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS - // Exceptional mask = common 0 bits of 2 exceptional values. - constexpr uint32_t EXCEPT_MASK = 0x809a'6184U; - - if (LIBC_UNLIKELY((x_abs & EXCEPT_MASK) == 0)) { - // Exceptional values - if (LIBC_UNLIKELY(x_abs == 0x3f65'9229U)) // |x| = 0x1.cb2452p-1f - return x < 0.0f ? fputil::round_result_slightly_down(-0x1.972ea8p-1f) - : fputil::round_result_slightly_up(0x1.972ea8p-1f); - if (LIBC_UNLIKELY(x_abs == 0x4004'1e6aU)) // |x| = 0x1.083cd4p+1f - return x < 0.0f ? fputil::round_result_slightly_down(-0x1.fe3462p-1f) - : fputil::round_result_slightly_up(0x1.fe3462p-1f); - if (x_abs == 0U) - return x; - } -#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS - - // Polynomial approximation: - // erf(x) ~ x * (c0 + c1 * x^2 + c2 * x^4 + ... + c7 * x^14) - double xd = static_cast<double>(x); - double xsq = xd * xd; - - const uint32_t EIGHT = 3 << FPBits::FRACTION_LEN; - int idx = static_cast<int>(FPBits(x_abs + EIGHT).get_val()); - - double x4 = xsq * xsq; - double c0 = fputil::multiply_add(xsq, COEFFS[idx][1], COEFFS[idx][0]); - double c1 = fputil::multiply_add(xsq, COEFFS[idx][3], COEFFS[idx][2]); - double c2 = fputil::multiply_add(xsq, COEFFS[idx][5], COEFFS[idx][4]); - double c3 = fputil::multiply_add(xsq, COEFFS[idx][7], COEFFS[idx][6]); - - double x8 = x4 * x4; - double p0 = fputil::multiply_add(x4, c1, c0); - double p1 = fputil::multiply_add(x4, c3, c2); - - return static_cast<float>(xd * fputil::multiply_add(x8, p1, p0)); -} +LLVM_LIBC_FUNCTION(float, erff, (float x)) { return math::erff(x); } } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/math/generic/explogxf.h b/libc/src/math/generic/explogxf.h index be4328a..a2a6d60 100644 --- a/libc/src/math/generic/explogxf.h +++ b/libc/src/math/generic/explogxf.h @@ -13,6 +13,7 @@ #include "src/__support/common.h" #include "src/__support/macros/properties/cpu_features.h" +#include "src/__support/math/acoshf_utils.h" #include "src/__support/math/exp10f_utils.h" #include "src/__support/math/exp_utils.h" @@ -163,41 +164,6 @@ LIBC_INLINE static float log_eval_f(float x) { return result; } -// x should be positive, normal finite value -LIBC_INLINE static double log_eval(double x) { - // For x = 2^ex * (1 + mx) - // log(x) = ex * log(2) + log(1 + mx) - using FPB = fputil::FPBits<double>; - FPB bs(x); - - double ex = static_cast<double>(bs.get_exponent()); - - // p1 is the leading 7 bits of mx, i.e. - // p1 * 2^(-7) <= m_x < (p1 + 1) * 2^(-7). - int p1 = static_cast<int>(bs.get_mantissa() >> (FPB::FRACTION_LEN - 7)); - - // Set bs to (1 + (mx - p1*2^(-7)) - bs.set_uintval(bs.uintval() & (FPB::FRACTION_MASK >> 7)); - bs.set_biased_exponent(FPB::EXP_BIAS); - // dx = (mx - p1*2^(-7)) / (1 + p1*2^(-7)). - double dx = (bs.get_val() - 1.0) * ONE_OVER_F[p1]; - - // Minimax polynomial of log(1 + dx) generated by Sollya with: - // > P = fpminimax(log(1 + x)/x, 6, [|D...|], [0, 2^-7]); - const double COEFFS[6] = {-0x1.ffffffffffffcp-2, 0x1.5555555552ddep-2, - -0x1.ffffffefe562dp-3, 0x1.9999817d3a50fp-3, - -0x1.554317b3f67a5p-3, 0x1.1dc5c45e09c18p-3}; - double dx2 = dx * dx; - double c1 = fputil::multiply_add(dx, COEFFS[1], COEFFS[0]); - double c2 = fputil::multiply_add(dx, COEFFS[3], COEFFS[2]); - double c3 = fputil::multiply_add(dx, COEFFS[5], COEFFS[4]); - - double p = fputil::polyeval(dx2, dx, c1, c2, c3); - double result = - fputil::multiply_add(ex, /*log(2)*/ 0x1.62e42fefa39efp-1, LOG_F[p1] + p); - return result; -} - } // namespace LIBC_NAMESPACE_DECL #endif // LLVM_LIBC_SRC_MATH_GENERIC_EXPLOGXF_H diff --git a/libc/src/math/generic/expxf16.h b/libc/src/math/generic/expxf16.h index b17b14f..562a427 100644 --- a/libc/src/math/generic/expxf16.h +++ b/libc/src/math/generic/expxf16.h @@ -9,14 +9,13 @@ #ifndef LLVM_LIBC_SRC_MATH_GENERIC_EXPXF16_H #define LLVM_LIBC_SRC_MATH_GENERIC_EXPXF16_H +#include "hdr/stdint_proxy.h" #include "src/__support/FPUtil/FPBits.h" #include "src/__support/FPUtil/cast.h" #include "src/__support/FPUtil/multiply_add.h" #include "src/__support/FPUtil/nearest_integer.h" #include "src/__support/macros/attributes.h" #include "src/__support/macros/config.h" -#include <stdint.h> - #include "src/__support/math/exp10_float16_constants.h" #include "src/__support/math/expf16_utils.h" diff --git a/libc/src/math/generic/inv_trigf_utils.h b/libc/src/math/generic/inv_trigf_utils.h deleted file mode 100644 index 8b47aba..0000000 --- a/libc/src/math/generic/inv_trigf_utils.h +++ /dev/null @@ -1,110 +0,0 @@ -//===-- Single-precision general inverse trigonometric functions ----------===// -// -// 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_GENERIC_INV_TRIGF_UTILS_H -#define LLVM_LIBC_SRC_MATH_GENERIC_INV_TRIGF_UTILS_H - -#include "src/__support/FPUtil/PolyEval.h" -#include "src/__support/FPUtil/multiply_add.h" -#include "src/__support/common.h" -#include "src/__support/macros/config.h" - -namespace LIBC_NAMESPACE_DECL { - -// PI and PI / 2 -static constexpr double M_MATH_PI = 0x1.921fb54442d18p+1; -static constexpr double M_MATH_PI_2 = 0x1.921fb54442d18p+0; - -extern double ATAN_COEFFS[17][9]; - -// Look-up table for atan(k/16) with k = 0..16. -static constexpr double ATAN_K_OVER_16[17] = { - 0.0, - 0x1.ff55bb72cfdeap-5, - 0x1.fd5ba9aac2f6ep-4, - 0x1.7b97b4bce5b02p-3, - 0x1.f5b75f92c80ddp-3, - 0x1.362773707ebccp-2, - 0x1.6f61941e4def1p-2, - 0x1.a64eec3cc23fdp-2, - 0x1.dac670561bb4fp-2, - 0x1.0657e94db30dp-1, - 0x1.1e00babdefeb4p-1, - 0x1.345f01cce37bbp-1, - 0x1.4978fa3269ee1p-1, - 0x1.5d58987169b18p-1, - 0x1.700a7c5784634p-1, - 0x1.819d0b7158a4dp-1, - 0x1.921fb54442d18p-1, -}; - -// For |x| <= 1/32 and 0 <= i <= 16, return Q(x) such that: -// Q(x) ~ (atan(x + i/16) - atan(i/16)) / x. -LIBC_INLINE static double atan_eval(double x, unsigned i) { - double x2 = x * x; - - double c0 = fputil::multiply_add(x, ATAN_COEFFS[i][2], ATAN_COEFFS[i][1]); - double c1 = fputil::multiply_add(x, ATAN_COEFFS[i][4], ATAN_COEFFS[i][3]); - double c2 = fputil::multiply_add(x, ATAN_COEFFS[i][6], ATAN_COEFFS[i][5]); - double c3 = fputil::multiply_add(x, ATAN_COEFFS[i][8], ATAN_COEFFS[i][7]); - - double x4 = x2 * x2; - double d1 = fputil::multiply_add(x2, c1, c0); - double d2 = fputil::multiply_add(x2, c3, c2); - double p = fputil::multiply_add(x4, d2, d1); - return p; -} - -// Evaluate atan without big lookup table. -// atan(n/d) - atan(k/16) = atan((n/d - k/16) / (1 + (n/d) * (k/16))) -// = atan((n - d * k/16)) / (d + n * k/16)) -// So we let q = (n - d * k/16) / (d + n * k/16), -// and approximate with Taylor polynomial: -// atan(q) ~ q - q^3/3 + q^5/5 - q^7/7 + q^9/9 -LIBC_INLINE static double atan_eval_no_table(double num, double den, - double k_over_16) { - double num_r = fputil::multiply_add(den, -k_over_16, num); - double den_r = fputil::multiply_add(num, k_over_16, den); - double q = num_r / den_r; - - constexpr double ATAN_TAYLOR[] = { - -0x1.5555555555555p-2, - 0x1.999999999999ap-3, - -0x1.2492492492492p-3, - 0x1.c71c71c71c71cp-4, - }; - double q2 = q * q; - double q3 = q2 * q; - double q4 = q2 * q2; - double c0 = fputil::multiply_add(q2, ATAN_TAYLOR[1], ATAN_TAYLOR[0]); - double c1 = fputil::multiply_add(q2, ATAN_TAYLOR[3], ATAN_TAYLOR[2]); - double d = fputil::multiply_add(q4, c1, c0); - return fputil::multiply_add(q3, d, q); -} - -// > Q = fpminimax(asin(x)/x, [|0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20|], -// [|1, D...|], [0, 0.5]); -static constexpr double ASIN_COEFFS[10] = { - 0x1.5555555540fa1p-3, 0x1.333333512edc2p-4, 0x1.6db6cc1541b31p-5, - 0x1.f1caff324770ep-6, 0x1.6e43899f5f4f4p-6, 0x1.1f847cf652577p-6, - 0x1.9b60f47f87146p-7, 0x1.259e2634c494fp-6, -0x1.df946fa875ddp-8, - 0x1.02311ecf99c28p-5}; - -// Evaluate P(x^2) - 1, where P(x^2) ~ asin(x)/x -LIBC_INLINE static double asin_eval(double xsq) { - double x4 = xsq * xsq; - double r1 = fputil::polyeval(x4, ASIN_COEFFS[0], ASIN_COEFFS[2], - ASIN_COEFFS[4], ASIN_COEFFS[6], ASIN_COEFFS[8]); - double r2 = fputil::polyeval(x4, ASIN_COEFFS[1], ASIN_COEFFS[3], - ASIN_COEFFS[5], ASIN_COEFFS[7], ASIN_COEFFS[9]); - return fputil::multiply_add(xsq, r2, r1); -} - -} // namespace LIBC_NAMESPACE_DECL - -#endif // LLVM_LIBC_SRC_MATH_GENERIC_INV_TRIGF_UTILS_H diff --git a/libc/src/math/generic/log1pf.cpp b/libc/src/math/generic/log1pf.cpp index 7f61429..16b1b34 100644 --- a/libc/src/math/generic/log1pf.cpp +++ b/libc/src/math/generic/log1pf.cpp @@ -37,6 +37,7 @@ namespace internal { // We don't need to treat denormal and 0 LIBC_INLINE float log(double x) { + using namespace acoshf_internal; constexpr double LOG_2 = 0x1.62e42fefa39efp-1; using FPBits = typename fputil::FPBits<double>; diff --git a/libc/src/math/generic/range_reduction_double_common.h b/libc/src/math/generic/range_reduction_double_common.h index f3dcdb9..a93ee25 100644 --- a/libc/src/math/generic/range_reduction_double_common.h +++ b/libc/src/math/generic/range_reduction_double_common.h @@ -278,7 +278,7 @@ private: }; #ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS -static Float128 range_reduction_small_f128(double x) { +LIBC_INLINE static Float128 range_reduction_small_f128(double x) { constexpr Float128 PI_OVER_128_F128 = { Sign::POS, -133, 0xc90f'daa2'2168'c234'c4c6'628b'80dc'1cd1_u128}; constexpr double ONE_TWENTY_EIGHT_OVER_PI_D = 0x1.45f306dc9c883p5; diff --git a/libc/src/pthread/CMakeLists.txt b/libc/src/pthread/CMakeLists.txt index c8c6680..c5db6fa 100644 --- a/libc/src/pthread/CMakeLists.txt +++ b/libc/src/pthread/CMakeLists.txt @@ -99,6 +99,7 @@ add_entrypoint_object( HDRS pthread_attr_setstack.h DEPENDS + libc.hdr.stdint_proxy libc.include.pthread libc.src.pthread.pthread_attr_setstacksize libc.src.errno.errno diff --git a/libc/src/pthread/pthread_attr_setstack.cpp b/libc/src/pthread/pthread_attr_setstack.cpp index 767f959..b66072c 100644 --- a/libc/src/pthread/pthread_attr_setstack.cpp +++ b/libc/src/pthread/pthread_attr_setstack.cpp @@ -9,13 +9,13 @@ #include "pthread_attr_setstack.h" #include "pthread_attr_setstacksize.h" +#include "hdr/stdint_proxy.h" #include "src/__support/common.h" #include "src/__support/libc_errno.h" #include "src/__support/macros/config.h" #include "src/__support/threads/thread.h" // For STACK_ALIGNMENT #include <pthread.h> -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/sched/linux/CMakeLists.txt b/libc/src/sched/linux/CMakeLists.txt index 4852c90..e690e76 100644 --- a/libc/src/sched/linux/CMakeLists.txt +++ b/libc/src/sched/linux/CMakeLists.txt @@ -5,6 +5,7 @@ add_entrypoint_object( HDRS ../sched_getaffinity.h DEPENDS + libc.hdr.stdint_proxy libc.include.sched libc.src.__support.OSUtil.osutil libc.src.errno.errno diff --git a/libc/src/sched/linux/sched_getaffinity.cpp b/libc/src/sched/linux/sched_getaffinity.cpp index e005819..4a5e91a 100644 --- a/libc/src/sched/linux/sched_getaffinity.cpp +++ b/libc/src/sched/linux/sched_getaffinity.cpp @@ -8,13 +8,13 @@ #include "src/sched/sched_getaffinity.h" +#include "hdr/stdint_proxy.h" #include "src/__support/OSUtil/syscall.h" // For internal syscall function. #include "src/__support/common.h" #include "src/__support/libc_errno.h" #include "src/__support/macros/config.h" #include <sched.h> -#include <stdint.h> #include <sys/syscall.h> // For syscall numbers. namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/setjmp/aarch64/CMakeLists.txt b/libc/src/setjmp/aarch64/CMakeLists.txt index 1078af8..1299110 100644 --- a/libc/src/setjmp/aarch64/CMakeLists.txt +++ b/libc/src/setjmp/aarch64/CMakeLists.txt @@ -34,7 +34,7 @@ add_entrypoint_object( HDRS ../sigsetjmp.h DEPENDS - libc.hdr.types.jmp_buf + libc.hdr.types.sigjmp_buf libc.hdr.types.sigset_t libc.hdr.offsetof_macros libc.src.setjmp.sigsetjmp_epilogue diff --git a/libc/src/setjmp/arm/CMakeLists.txt b/libc/src/setjmp/arm/CMakeLists.txt index 77f8471..7a48da3 100644 --- a/libc/src/setjmp/arm/CMakeLists.txt +++ b/libc/src/setjmp/arm/CMakeLists.txt @@ -16,7 +16,7 @@ if (TARGET libc.src.setjmp.sigsetjmp_epilogue) HDRS ../sigsetjmp.h DEPENDS - libc.hdr.types.jmp_buf + libc.hdr.types.sigjmp_buf libc.hdr.types.sigset_t libc.hdr.offsetof_macros libc.src.setjmp.sigsetjmp_epilogue diff --git a/libc/src/setjmp/darwin/CMakeLists.txt b/libc/src/setjmp/darwin/CMakeLists.txt index b844c8c..62acec0 100644 --- a/libc/src/setjmp/darwin/CMakeLists.txt +++ b/libc/src/setjmp/darwin/CMakeLists.txt @@ -7,6 +7,6 @@ add_object_library( DEPENDS libc.src.__support.common libc.src.__support.OSUtil.osutil - libc.hdr.types.jmp_buf + libc.hdr.types.sigjmp_buf libc.hdr.types.sigset_t ) diff --git a/libc/src/setjmp/darwin/sigsetjmp_epilogue.cpp b/libc/src/setjmp/darwin/sigsetjmp_epilogue.cpp index b2ca4d9..568545a 100644 --- a/libc/src/setjmp/darwin/sigsetjmp_epilogue.cpp +++ b/libc/src/setjmp/darwin/sigsetjmp_epilogue.cpp @@ -12,7 +12,7 @@ #include "src/signal/sigprocmask.h" namespace LIBC_NAMESPACE_DECL { -[[gnu::returns_twice]] int sigsetjmp_epilogue(jmp_buf buffer, int retval) { +[[gnu::returns_twice]] int sigsetjmp_epilogue(sigjmp_buf buffer, int retval) { syscall_impl<long>(sigprocmask, SIG_SETMASK, /* set= */ retval ? &buffer->sigmask : nullptr, /* old_set= */ retval ? nullptr : &buffer->sigmask); diff --git a/libc/src/setjmp/linux/CMakeLists.txt b/libc/src/setjmp/linux/CMakeLists.txt index b844c8c..62acec0 100644 --- a/libc/src/setjmp/linux/CMakeLists.txt +++ b/libc/src/setjmp/linux/CMakeLists.txt @@ -7,6 +7,6 @@ add_object_library( DEPENDS libc.src.__support.common libc.src.__support.OSUtil.osutil - libc.hdr.types.jmp_buf + libc.hdr.types.sigjmp_buf libc.hdr.types.sigset_t ) diff --git a/libc/src/setjmp/linux/sigsetjmp_epilogue.cpp b/libc/src/setjmp/linux/sigsetjmp_epilogue.cpp index 4718623..7e14312 100644 --- a/libc/src/setjmp/linux/sigsetjmp_epilogue.cpp +++ b/libc/src/setjmp/linux/sigsetjmp_epilogue.cpp @@ -12,7 +12,7 @@ #include <sys/syscall.h> // For syscall numbers. namespace LIBC_NAMESPACE_DECL { -[[gnu::returns_twice]] int sigsetjmp_epilogue(jmp_buf buffer, int retval) { +[[gnu::returns_twice]] int sigsetjmp_epilogue(sigjmp_buf buffer, int retval) { // If set is NULL, then the signal mask is unchanged (i.e., how is // ignored), but the current value of the signal mask is nevertheless // returned in oldset (if it is not NULL). diff --git a/libc/src/setjmp/riscv/CMakeLists.txt b/libc/src/setjmp/riscv/CMakeLists.txt index 1959e9c..ee3ea28 100644 --- a/libc/src/setjmp/riscv/CMakeLists.txt +++ b/libc/src/setjmp/riscv/CMakeLists.txt @@ -19,7 +19,7 @@ if (TARGET libc.src.setjmp.sigsetjmp_epilogue) HDRS ../sigsetjmp.h DEPENDS - libc.hdr.types.jmp_buf + libc.hdr.types.sigjmp_buf libc.hdr.types.sigset_t libc.hdr.offsetof_macros libc.src.setjmp.sigsetjmp_epilogue diff --git a/libc/src/setjmp/sigsetjmp.h b/libc/src/setjmp/sigsetjmp.h index ef060c8..6e08bd6 100644 --- a/libc/src/setjmp/sigsetjmp.h +++ b/libc/src/setjmp/sigsetjmp.h @@ -9,7 +9,7 @@ #ifndef LLVM_LIBC_SRC_SETJMP_SIGSETJMP_H #define LLVM_LIBC_SRC_SETJMP_SIGSETJMP_H -#include "hdr/types/jmp_buf.h" +#include "hdr/types/sigjmp_buf.h" #include "src/__support/macros/config.h" #include "src/__support/macros/properties/compiler.h" diff --git a/libc/src/setjmp/sigsetjmp_epilogue.h b/libc/src/setjmp/sigsetjmp_epilogue.h index 88702b7..7508d4a 100644 --- a/libc/src/setjmp/sigsetjmp_epilogue.h +++ b/libc/src/setjmp/sigsetjmp_epilogue.h @@ -9,11 +9,11 @@ #ifndef LLVM_LIBC_SRC_SETJMP_SIGSETJMP_EPILOGUE_H #define LLVM_LIBC_SRC_SETJMP_SIGSETJMP_EPILOGUE_H -#include "hdr/types/jmp_buf.h" +#include "hdr/types/sigjmp_buf.h" #include "src/__support/common.h" namespace LIBC_NAMESPACE_DECL { -[[gnu::returns_twice]] int sigsetjmp_epilogue(jmp_buf buffer, int retval); +[[gnu::returns_twice]] int sigsetjmp_epilogue(sigjmp_buf buffer, int retval); } // namespace LIBC_NAMESPACE_DECL #endif // LLVM_LIBC_SRC_SETJMP_SIGSETJMP_EPILOGUE_H diff --git a/libc/src/setjmp/x86_64/CMakeLists.txt b/libc/src/setjmp/x86_64/CMakeLists.txt index 03ed5fb..5f87bc6 100644 --- a/libc/src/setjmp/x86_64/CMakeLists.txt +++ b/libc/src/setjmp/x86_64/CMakeLists.txt @@ -16,7 +16,7 @@ if (TARGET libc.src.setjmp.sigsetjmp_epilogue) HDRS ../sigsetjmp.h DEPENDS - libc.hdr.types.jmp_buf + libc.hdr.types.sigjmp_buf libc.hdr.types.sigset_t libc.hdr.offsetof_macros libc.src.setjmp.sigsetjmp_epilogue diff --git a/libc/src/spawn/CMakeLists.txt b/libc/src/spawn/CMakeLists.txt index 11621da7..fe5f81e 100644 --- a/libc/src/spawn/CMakeLists.txt +++ b/libc/src/spawn/CMakeLists.txt @@ -7,6 +7,7 @@ add_header_library( HDRS file_actions.h DEPENDS + libc.hdr.stdint_proxy libc.include.spawn ) diff --git a/libc/src/spawn/file_actions.h b/libc/src/spawn/file_actions.h index 80b9295..c69640a3d 100644 --- a/libc/src/spawn/file_actions.h +++ b/libc/src/spawn/file_actions.h @@ -9,9 +9,9 @@ #ifndef LLVM_LIBC_SRC_SPAWN_FILE_ACTIONS_H #define LLVM_LIBC_SRC_SPAWN_FILE_ACTIONS_H +#include "hdr/stdint_proxy.h" #include "src/__support/macros/config.h" #include <spawn.h> // For mode_t -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/stdio/baremetal/CMakeLists.txt b/libc/src/stdio/baremetal/CMakeLists.txt index e879230..548938f 100644 --- a/libc/src/stdio/baremetal/CMakeLists.txt +++ b/libc/src/stdio/baremetal/CMakeLists.txt @@ -72,6 +72,7 @@ add_entrypoint_object( ../scanf.h DEPENDS .scanf_internal + libc.include.inttypes libc.src.stdio.scanf_core.scanf_main libc.src.__support.arg_list libc.src.__support.OSUtil.osutil diff --git a/libc/src/stdio/gpu/CMakeLists.txt b/libc/src/stdio/gpu/CMakeLists.txt index bea1134..8412153 100644 --- a/libc/src/stdio/gpu/CMakeLists.txt +++ b/libc/src/stdio/gpu/CMakeLists.txt @@ -259,6 +259,7 @@ add_entrypoint_object( HDRS ../fgets.h DEPENDS + libc.hdr.stdint_proxy libc.hdr.types.FILE .gpu_file ) diff --git a/libc/src/stdio/gpu/fgets.cpp b/libc/src/stdio/gpu/fgets.cpp index 5447e86..e1c6088 100644 --- a/libc/src/stdio/gpu/fgets.cpp +++ b/libc/src/stdio/gpu/fgets.cpp @@ -9,12 +9,11 @@ #include "src/stdio/fgets.h" #include "file.h" +#include "hdr/stdint_proxy.h" #include "hdr/stdio_macros.h" // for EOF. #include "hdr/types/FILE.h" #include "src/__support/common.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { LLVM_LIBC_FUNCTION(char *, fgets, diff --git a/libc/src/stdio/printf_core/CMakeLists.txt b/libc/src/stdio/printf_core/CMakeLists.txt index c22f985..76eb0a2 100644 --- a/libc/src/stdio/printf_core/CMakeLists.txt +++ b/libc/src/stdio/printf_core/CMakeLists.txt @@ -44,6 +44,7 @@ add_header_library( HDRS core_structs.h DEPENDS + libc.include.inttypes libc.src.__support.CPP.string_view libc.src.__support.FPUtil.fp_bits ) @@ -97,6 +98,7 @@ add_header_library( .core_structs .printf_config .writer + libc.include.inttypes libc.src.__support.big_int libc.src.__support.common libc.src.__support.CPP.limits diff --git a/libc/src/stdio/scanf_core/CMakeLists.txt b/libc/src/stdio/scanf_core/CMakeLists.txt index dee125c..561180c 100644 --- a/libc/src/stdio/scanf_core/CMakeLists.txt +++ b/libc/src/stdio/scanf_core/CMakeLists.txt @@ -35,6 +35,7 @@ add_header_library( core_structs.h DEPENDS .scanf_config + libc.include.inttypes libc.src.__support.CPP.string_view libc.src.__support.CPP.bitset libc.src.__support.FPUtil.fp_bits @@ -97,6 +98,7 @@ add_header_library( DEPENDS .reader .core_structs + libc.include.inttypes libc.src.__support.common libc.src.__support.ctype_utils libc.src.__support.CPP.bitset diff --git a/libc/src/stdlib/CMakeLists.txt b/libc/src/stdlib/CMakeLists.txt index 74ae864..aa653c3 100644 --- a/libc/src/stdlib/CMakeLists.txt +++ b/libc/src/stdlib/CMakeLists.txt @@ -191,8 +191,9 @@ add_entrypoint_object( HDRS a64l.h DEPENDS - libc.src.__support.ctype_utils + libc.hdr.stdint_proxy libc.hdr.types.size_t + libc.src.__support.ctype_utils ) add_entrypoint_object( @@ -202,8 +203,9 @@ add_entrypoint_object( HDRS l64a.h DEPENDS - libc.src.__support.ctype_utils + libc.hdr.stdint_proxy libc.hdr.types.size_t + libc.src.__support.ctype_utils ) add_entrypoint_object( @@ -287,6 +289,7 @@ add_header_library( heap_sort.h quick_sort.h DEPENDS + libc.hdr.stdint_proxy libc.include.stdlib libc.src.__support.CPP.cstddef ) @@ -299,6 +302,7 @@ add_entrypoint_object( qsort.h DEPENDS .qsort_util + libc.hdr.stdint_proxy libc.hdr.types.size_t ) @@ -310,6 +314,7 @@ add_entrypoint_object( qsort_r.h DEPENDS .qsort_util + libc.hdr.stdint_proxy libc.hdr.types.size_t ) diff --git a/libc/src/stdlib/a64l.cpp b/libc/src/stdlib/a64l.cpp index 84be2d2..690b70d 100644 --- a/libc/src/stdlib/a64l.cpp +++ b/libc/src/stdlib/a64l.cpp @@ -7,13 +7,12 @@ //===----------------------------------------------------------------------===// #include "src/stdlib/a64l.h" +#include "hdr/stdint_proxy.h" #include "hdr/types/size_t.h" #include "src/__support/common.h" #include "src/__support/ctype_utils.h" #include "src/__support/macros/config.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { // I'm not sure this should go in ctype_utils since the specific ordering of diff --git a/libc/src/stdlib/bsearch.cpp b/libc/src/stdlib/bsearch.cpp index 69b3e74..f084805 100644 --- a/libc/src/stdlib/bsearch.cpp +++ b/libc/src/stdlib/bsearch.cpp @@ -10,7 +10,7 @@ #include "src/__support/common.h" #include "src/__support/macros/config.h" -#include <stdint.h> +#include "hdr/stdint_proxy.h" namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/stdlib/l64a.cpp b/libc/src/stdlib/l64a.cpp index b5506c3e..d59e65e 100644 --- a/libc/src/stdlib/l64a.cpp +++ b/libc/src/stdlib/l64a.cpp @@ -7,14 +7,13 @@ //===----------------------------------------------------------------------===// #include "src/stdlib/l64a.h" +#include "hdr/stdint_proxy.h" #include "hdr/types/size_t.h" #include "src/__support/common.h" #include "src/__support/ctype_utils.h" #include "src/__support/libc_assert.h" #include "src/__support/macros/config.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { // the standard says to only use up to 6 characters. Null terminator is diff --git a/libc/src/stdlib/qsort.cpp b/libc/src/stdlib/qsort.cpp index 0bf5fc7..f66b686 100644 --- a/libc/src/stdlib/qsort.cpp +++ b/libc/src/stdlib/qsort.cpp @@ -7,12 +7,11 @@ //===----------------------------------------------------------------------===// #include "src/stdlib/qsort.h" +#include "hdr/stdint_proxy.h" #include "src/__support/common.h" #include "src/__support/macros/config.h" #include "src/stdlib/qsort_util.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { LLVM_LIBC_FUNCTION(void, qsort, diff --git a/libc/src/stdlib/qsort_data.h b/libc/src/stdlib/qsort_data.h index aa6d9bb..739fce8 100644 --- a/libc/src/stdlib/qsort_data.h +++ b/libc/src/stdlib/qsort_data.h @@ -9,11 +9,10 @@ #ifndef LLVM_LIBC_SRC_STDLIB_QSORT_DATA_H #define LLVM_LIBC_SRC_STDLIB_QSORT_DATA_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/cstddef.h" #include "src/__support/macros/config.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/stdlib/qsort_r.cpp b/libc/src/stdlib/qsort_r.cpp index 4e60998..4744820 100644 --- a/libc/src/stdlib/qsort_r.cpp +++ b/libc/src/stdlib/qsort_r.cpp @@ -7,12 +7,11 @@ //===----------------------------------------------------------------------===// #include "src/stdlib/qsort_r.h" +#include "hdr/stdint_proxy.h" #include "src/__support/common.h" #include "src/__support/macros/config.h" #include "src/stdlib/qsort_util.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { LLVM_LIBC_FUNCTION(void, qsort_r, diff --git a/libc/src/stdlib/quick_sort.h b/libc/src/stdlib/quick_sort.h index 8ba0098..00115bd 100644 --- a/libc/src/stdlib/quick_sort.h +++ b/libc/src/stdlib/quick_sort.h @@ -9,13 +9,12 @@ #ifndef LLVM_LIBC_SRC_STDLIB_QUICK_SORT_H #define LLVM_LIBC_SRC_STDLIB_QUICK_SORT_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/bit.h" #include "src/__support/CPP/cstddef.h" #include "src/__support/macros/config.h" #include "src/stdlib/qsort_pivot.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace internal { diff --git a/libc/src/string/CMakeLists.txt b/libc/src/string/CMakeLists.txt index 8784bc3..809decf 100644 --- a/libc/src/string/CMakeLists.txt +++ b/libc/src/string/CMakeLists.txt @@ -17,6 +17,7 @@ add_header_library( DEPENDS libc.hdr.types.size_t libc.hdr.limits_macros + libc.hdr.stdint_proxy libc.src.__support.CPP.bitset libc.src.__support.CPP.type_traits libc.src.__support.common diff --git a/libc/src/string/allocating_string_utils.h b/libc/src/string/allocating_string_utils.h index 1dece51..e2f61f7 100644 --- a/libc/src/string/allocating_string_utils.h +++ b/libc/src/string/allocating_string_utils.h @@ -20,15 +20,15 @@ namespace LIBC_NAMESPACE_DECL { namespace internal { -LIBC_INLINE cpp::optional<char *> strdup(const char *src) { +template <typename T> LIBC_INLINE cpp::optional<T *> strdup(const T *src) { if (src == nullptr) return cpp::nullopt; size_t len = string_length(src) + 1; AllocChecker ac; - char *newstr = new (ac) char[len]; + T *newstr = new (ac) T[len]; if (!ac) return cpp::nullopt; - inline_memcpy(newstr, src, len); + inline_memcpy(newstr, src, len * sizeof(T)); return newstr; } diff --git a/libc/src/string/memory_utils/CMakeLists.txt b/libc/src/string/memory_utils/CMakeLists.txt index 8ab1c9f..670db30 100644 --- a/libc/src/string/memory_utils/CMakeLists.txt +++ b/libc/src/string/memory_utils/CMakeLists.txt @@ -34,6 +34,7 @@ add_header_library( x86_64/inline_memmove.h x86_64/inline_memset.h DEPENDS + libc.hdr.stdint_proxy libc.src.__support.common libc.src.__support.CPP.bit libc.src.__support.CPP.cstddef diff --git a/libc/src/string/memory_utils/op_generic.h b/libc/src/string/memory_utils/op_generic.h index 9349cfdd..3760341 100644 --- a/libc/src/string/memory_utils/op_generic.h +++ b/libc/src/string/memory_utils/op_generic.h @@ -23,6 +23,7 @@ #ifndef LLVM_LIBC_SRC_STRING_MEMORY_UTILS_OP_GENERIC_H #define LLVM_LIBC_SRC_STRING_MEMORY_UTILS_OP_GENERIC_H +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/array.h" #include "src/__support/CPP/type_traits.h" #include "src/__support/common.h" @@ -34,8 +35,6 @@ #include "src/string/memory_utils/op_builtin.h" #include "src/string/memory_utils/utils.h" -#include <stdint.h> - static_assert((UINTPTR_MAX == 4294967295U) || (UINTPTR_MAX == 18446744073709551615UL), "We currently only support 32- or 64-bit platforms"); diff --git a/libc/src/string/memory_utils/utils.h b/libc/src/string/memory_utils/utils.h index c08608c..0f9c9e3 100644 --- a/libc/src/string/memory_utils/utils.h +++ b/libc/src/string/memory_utils/utils.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC_STRING_MEMORY_UTILS_UTILS_H #define LLVM_LIBC_SRC_STRING_MEMORY_UTILS_UTILS_H +#include "hdr/stdint_proxy.h" // intptr_t / uintptr_t / INT32_MAX / INT32_MIN #include "src/__support/CPP/bit.h" #include "src/__support/CPP/cstddef.h" #include "src/__support/CPP/type_traits.h" @@ -18,7 +19,6 @@ #include "src/__support/macros/properties/architectures.h" #include <stddef.h> // size_t -#include <stdint.h> // intptr_t / uintptr_t / INT32_MAX / INT32_MIN namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/string/memory_utils/x86_64/inline_memcpy.h b/libc/src/string/memory_utils/x86_64/inline_memcpy.h index 68f64fb..bf3aa1f 100644 --- a/libc/src/string/memory_utils/x86_64/inline_memcpy.h +++ b/libc/src/string/memory_utils/x86_64/inline_memcpy.h @@ -8,6 +8,7 @@ #ifndef LLVM_LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_MEMCPY_H #define LLVM_LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_MEMCPY_H +#include "hdr/stdint_proxy.h" // SIZE_MAX #include "src/__support/macros/attributes.h" // LIBC_INLINE_VAR #include "src/__support/macros/optimization.h" // LIBC_UNLIKELY #include "src/string/memory_utils/op_builtin.h" @@ -15,7 +16,6 @@ #include "src/string/memory_utils/utils.h" #include <stddef.h> // size_t -#include <stdint.h> // SIZE_MAX #ifdef LLVM_LIBC_MEMCPY_X86_USE_ONLY_REPMOVSB #error LLVM_LIBC_MEMCPY_X86_USE_ONLY_REPMOVSB is deprecated use LIBC_COPT_MEMCPY_X86_USE_REPMOVSB_FROM_SIZE=0 instead. diff --git a/libc/src/string/string_utils.h b/libc/src/string/string_utils.h index 4f56263..80e5783 100644 --- a/libc/src/string/string_utils.h +++ b/libc/src/string/string_utils.h @@ -15,6 +15,7 @@ #define LLVM_LIBC_SRC_STRING_STRING_UTILS_H #include "hdr/limits_macros.h" +#include "hdr/stdint_proxy.h" // uintptr_t #include "hdr/types/size_t.h" #include "src/__support/CPP/bitset.h" #include "src/__support/CPP/type_traits.h" // cpp::is_same_v diff --git a/libc/src/sys/epoll/linux/epoll_create.cpp b/libc/src/sys/epoll/linux/epoll_create.cpp index 2e44e88..dcd082b 100644 --- a/libc/src/sys/epoll/linux/epoll_create.cpp +++ b/libc/src/sys/epoll/linux/epoll_create.cpp @@ -20,6 +20,11 @@ LLVM_LIBC_FUNCTION(int, epoll_create, ([[maybe_unused]] int size)) { #ifdef SYS_epoll_create int ret = LIBC_NAMESPACE::syscall_impl<int>(SYS_epoll_create, size); #elif defined(SYS_epoll_create1) + if (size == 0) { + libc_errno = EINVAL; + return -1; + } + int ret = LIBC_NAMESPACE::syscall_impl<int>(SYS_epoll_create1, 0); #else #error \ diff --git a/libc/src/sys/stat/linux/CMakeLists.txt b/libc/src/sys/stat/linux/CMakeLists.txt index 9aeb146..c99872c 100644 --- a/libc/src/sys/stat/linux/CMakeLists.txt +++ b/libc/src/sys/stat/linux/CMakeLists.txt @@ -73,6 +73,7 @@ add_header_library( HDRS kernel_statx.h DEPENDS + libc.hdr.stdint_proxy libc.include.sys_stat libc.include.sys_syscall libc.src.__support.OSUtil.osutil diff --git a/libc/src/sys/stat/linux/kernel_statx.h b/libc/src/sys/stat/linux/kernel_statx.h index d0e223a..455ab17 100644 --- a/libc/src/sys/stat/linux/kernel_statx.h +++ b/libc/src/sys/stat/linux/kernel_statx.h @@ -9,11 +9,11 @@ #ifndef LLVM_LIBC_SRC_SYS_STAT_LINUX_KERNEL_STATX_H #define LLVM_LIBC_SRC_SYS_STAT_LINUX_KERNEL_STATX_H +#include "hdr/stdint_proxy.h" #include "src/__support/OSUtil/syscall.h" // For internal syscall function. #include "src/__support/common.h" #include "src/__support/macros/config.h" -#include <stdint.h> #include <sys/stat.h> #include <sys/syscall.h> // For syscall numbers. diff --git a/libc/src/sys/time/linux/setitimer.cpp b/libc/src/sys/time/linux/setitimer.cpp index 1de0d43..fb16358 100644 --- a/libc/src/sys/time/linux/setitimer.cpp +++ b/libc/src/sys/time/linux/setitimer.cpp @@ -22,9 +22,9 @@ LLVM_LIBC_FUNCTION(int, setitimer, // There is no SYS_setitimer_time64 call, so we can't use time_t directly, // and need to convert it to long first. long new_value32[4] = {static_cast<long>(new_value->it_interval.tv_sec), - new_value->it_interval.tv_usec, + static_cast<long>(new_value->it_interval.tv_usec), static_cast<long>(new_value->it_value.tv_sec), - new_value->it_value.tv_usec}; + static_cast<long>(new_value->it_value.tv_usec)}; long old_value32[4]; ret = LIBC_NAMESPACE::syscall_impl<long>(SYS_setitimer, which, new_value32, diff --git a/libc/src/sys/time/linux/utimes.cpp b/libc/src/sys/time/linux/utimes.cpp index ed37b42..9c00ce9 100644 --- a/libc/src/sys/time/linux/utimes.cpp +++ b/libc/src/sys/time/linux/utimes.cpp @@ -59,8 +59,10 @@ LLVM_LIBC_FUNCTION(int, utimes, ts[1].tv_sec = times[1].tv_sec; // convert u-seconds to nanoseconds - ts[0].tv_nsec = times[0].tv_usec * 1000; - ts[1].tv_nsec = times[1].tv_usec * 1000; + ts[0].tv_nsec = + static_cast<decltype(ts[0].tv_nsec)>(times[0].tv_usec * 1000); + ts[1].tv_nsec = + static_cast<decltype(ts[1].tv_nsec)>(times[1].tv_usec * 1000); ts_ptr = ts; } diff --git a/libc/src/time/CMakeLists.txt b/libc/src/time/CMakeLists.txt index 3b951df..304b3f2 100644 --- a/libc/src/time/CMakeLists.txt +++ b/libc/src/time/CMakeLists.txt @@ -7,10 +7,11 @@ add_header_library( HDRS time_constants.h DEPENDS + libc.hdr.stdint_proxy + libc.hdr.types.time_t libc.include.time libc.src.__support.CPP.array libc.src.__support.CPP.string_view - libc.hdr.types.time_t ) add_object_library( @@ -29,6 +30,7 @@ add_object_library( libc.hdr.types.time_t libc.hdr.types.size_t libc.hdr.types.struct_tm + libc.hdr.stdint_proxy ) add_entrypoint_object( diff --git a/libc/src/time/linux/CMakeLists.txt b/libc/src/time/linux/CMakeLists.txt index 314623f..a6ec7c7 100644 --- a/libc/src/time/linux/CMakeLists.txt +++ b/libc/src/time/linux/CMakeLists.txt @@ -34,6 +34,7 @@ add_entrypoint_object( ../nanosleep.h DEPENDS libc.hdr.types.struct_timespec + libc.hdr.stdint_proxy libc.include.sys_syscall libc.src.__support.OSUtil.osutil libc.src.__support.CPP.limits diff --git a/libc/src/time/linux/nanosleep.cpp b/libc/src/time/linux/nanosleep.cpp index 6b97041..e5df158 100644 --- a/libc/src/time/linux/nanosleep.cpp +++ b/libc/src/time/linux/nanosleep.cpp @@ -7,13 +7,13 @@ //===----------------------------------------------------------------------===// #include "src/time/nanosleep.h" +#include "hdr/stdint_proxy.h" // For int64_t. #include "hdr/time_macros.h" #include "src/__support/OSUtil/syscall.h" // For syscall functions. #include "src/__support/common.h" #include "src/__support/libc_errno.h" #include "src/__support/macros/config.h" -#include <stdint.h> // For int64_t. #include <sys/syscall.h> // For syscall numbers. namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/time/strftime_core/CMakeLists.txt b/libc/src/time/strftime_core/CMakeLists.txt index 5e40e66..3ffd283 100644 --- a/libc/src/time/strftime_core/CMakeLists.txt +++ b/libc/src/time/strftime_core/CMakeLists.txt @@ -5,6 +5,7 @@ add_header_library( DEPENDS libc.src.__support.CPP.string_view libc.hdr.types.struct_tm + libc.hdr.stdint_proxy ) add_header_library( diff --git a/libc/src/time/strftime_core/core_structs.h b/libc/src/time/strftime_core/core_structs.h index 25bf5e6..9da57aa 100644 --- a/libc/src/time/strftime_core/core_structs.h +++ b/libc/src/time/strftime_core/core_structs.h @@ -9,11 +9,10 @@ #ifndef LLVM_LIBC_SRC_STDIO_STRFTIME_CORE_CORE_STRUCTS_H #define LLVM_LIBC_SRC_STDIO_STRFTIME_CORE_CORE_STRUCTS_H +#include "hdr/stdint_proxy.h" #include "hdr/types/struct_tm.h" #include "src/__support/CPP/string_view.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace strftime_core { diff --git a/libc/src/time/time_constants.h b/libc/src/time/time_constants.h index 0fcb7ff..32eb0a1 100644 --- a/libc/src/time/time_constants.h +++ b/libc/src/time/time_constants.h @@ -9,10 +9,10 @@ #ifndef LLVM_LIBC_SRC_TIME_TIME_CONSTANTS_H #define LLVM_LIBC_SRC_TIME_TIME_CONSTANTS_H +#include "hdr/stdint_proxy.h" #include "hdr/types/time_t.h" #include "src/__support/CPP/array.h" #include "src/__support/CPP/string_view.h" -#include <stdint.h> namespace LIBC_NAMESPACE_DECL { namespace time_constants { diff --git a/libc/src/time/time_utils.cpp b/libc/src/time/time_utils.cpp index 1c519c3..1d0daea 100644 --- a/libc/src/time/time_utils.cpp +++ b/libc/src/time/time_utils.cpp @@ -7,13 +7,12 @@ //===----------------------------------------------------------------------===// #include "src/time/time_utils.h" +#include "hdr/stdint_proxy.h" #include "src/__support/CPP/limits.h" // INT_MIN, INT_MAX #include "src/__support/common.h" #include "src/__support/macros/config.h" #include "src/time/time_constants.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace time_utils { diff --git a/libc/src/time/time_utils.h b/libc/src/time/time_utils.h index 0541c24..84d412c 100644 --- a/libc/src/time/time_utils.h +++ b/libc/src/time/time_utils.h @@ -9,6 +9,7 @@ #ifndef LLVM_LIBC_SRC_TIME_TIME_UTILS_H #define LLVM_LIBC_SRC_TIME_TIME_UTILS_H +#include "hdr/stdint_proxy.h" #include "hdr/types/size_t.h" #include "hdr/types/struct_tm.h" #include "hdr/types/time_t.h" @@ -19,8 +20,6 @@ #include "src/__support/macros/config.h" #include "time_constants.h" -#include <stdint.h> - namespace LIBC_NAMESPACE_DECL { namespace time_utils { diff --git a/libc/src/unistd/linux/CMakeLists.txt b/libc/src/unistd/linux/CMakeLists.txt index 368593a..382a61f 100644 --- a/libc/src/unistd/linux/CMakeLists.txt +++ b/libc/src/unistd/linux/CMakeLists.txt @@ -172,6 +172,7 @@ add_entrypoint_object( DEPENDS libc.hdr.types.off_t libc.hdr.fcntl_macros + libc.hdr.stdint_proxy libc.include.unistd libc.include.sys_syscall libc.src.__support.OSUtil.osutil @@ -385,6 +386,7 @@ add_entrypoint_object( libc.hdr.types.size_t libc.hdr.types.ssize_t libc.hdr.fcntl_macros + libc.hdr.stdint_proxy libc.include.unistd libc.include.sys_syscall libc.src.__support.OSUtil.osutil @@ -403,6 +405,7 @@ add_entrypoint_object( libc.hdr.types.size_t libc.hdr.types.ssize_t libc.hdr.fcntl_macros + libc.hdr.stdint_proxy libc.include.unistd libc.include.sys_syscall libc.src.__support.OSUtil.osutil @@ -546,6 +549,7 @@ add_entrypoint_object( DEPENDS libc.hdr.types.off_t libc.hdr.fcntl_macros + libc.hdr.stdint_proxy libc.include.unistd libc.include.sys_syscall libc.src.__support.OSUtil.osutil diff --git a/libc/src/unistd/linux/ftruncate.cpp b/libc/src/unistd/linux/ftruncate.cpp index f6aa6f8..b4729f8 100644 --- a/libc/src/unistd/linux/ftruncate.cpp +++ b/libc/src/unistd/linux/ftruncate.cpp @@ -11,10 +11,10 @@ #include "src/__support/OSUtil/syscall.h" // For internal syscall function. #include "src/__support/common.h" +#include "hdr/stdint_proxy.h" // For uint64_t. #include "hdr/unistd_macros.h" #include "src/__support/libc_errno.h" #include "src/__support/macros/config.h" -#include <stdint.h> // For uint64_t. #include <sys/syscall.h> // For syscall numbers. namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/unistd/linux/pread.cpp b/libc/src/unistd/linux/pread.cpp index 2f86e39..cf3152d 100644 --- a/libc/src/unistd/linux/pread.cpp +++ b/libc/src/unistd/linux/pread.cpp @@ -8,12 +8,12 @@ #include "src/unistd/pread.h" +#include "hdr/stdint_proxy.h" // For uint64_t. #include "src/__support/OSUtil/syscall.h" // For internal syscall function. #include "src/__support/common.h" #include "src/__support/libc_errno.h" #include "src/__support/macros/config.h" #include "src/__support/macros/sanitizer.h" // for MSAN_UNPOISON -#include <stdint.h> // For uint64_t. #include <sys/syscall.h> // For syscall numbers. namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/unistd/linux/pwrite.cpp b/libc/src/unistd/linux/pwrite.cpp index f4cf8e1..7672063 100644 --- a/libc/src/unistd/linux/pwrite.cpp +++ b/libc/src/unistd/linux/pwrite.cpp @@ -11,9 +11,9 @@ #include "src/__support/OSUtil/syscall.h" // For internal syscall function. #include "src/__support/common.h" +#include "hdr/stdint_proxy.h" // For uint64_t. #include "src/__support/libc_errno.h" #include "src/__support/macros/config.h" -#include <stdint.h> // For uint64_t. #include <sys/syscall.h> // For syscall numbers. namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/unistd/linux/truncate.cpp b/libc/src/unistd/linux/truncate.cpp index 6103d4b..204b27f7 100644 --- a/libc/src/unistd/linux/truncate.cpp +++ b/libc/src/unistd/linux/truncate.cpp @@ -13,8 +13,8 @@ #include "src/__support/libc_errno.h" #include "src/__support/macros/config.h" +#include "hdr/stdint_proxy.h" // For uint64_t. #include "hdr/unistd_macros.h" -#include <stdint.h> // For uint64_t. #include <sys/syscall.h> // For syscall numbers. namespace LIBC_NAMESPACE_DECL { diff --git a/libc/src/wchar/CMakeLists.txt b/libc/src/wchar/CMakeLists.txt index 7ace1a6..49f4a1b 100644 --- a/libc/src/wchar/CMakeLists.txt +++ b/libc/src/wchar/CMakeLists.txt @@ -1,3 +1,13 @@ +add_header_library( + wchar_utils + HDRS + wchar_utils.h + DEPENDS + libc.hdr.types.size_t + libc.hdr.types.wchar_t + libc.src.__support.common +) + add_entrypoint_object( wcslen SRCS @@ -127,6 +137,21 @@ add_entrypoint_object( ) add_entrypoint_object( + mbsinit + SRCS + mbsinit.cpp + HDRS + mbsinit.h + DEPENDS + libc.hdr.types.wchar_t + libc.hdr.types.mbstate_t + libc.src.__support.common + libc.src.__support.macros.config + libc.src.__support.wchar.character_converter + libc.src.__support.wchar.mbstate +) + +add_entrypoint_object( mbrtowc SRCS mbrtowc.cpp @@ -160,6 +185,78 @@ add_entrypoint_object( ) add_entrypoint_object( + wcstombs + SRCS + wcstombs.cpp + HDRS + wcstombs.h + DEPENDS + libc.hdr.types.wchar_t + libc.src.__support.wchar.mbstate + libc.src.__support.wchar.wcsnrtombs + libc.src.__support.libc_errno +) + +add_entrypoint_object( + wcsrtombs + SRCS + wcsrtombs.cpp + HDRS + wcsrtombs.h + DEPENDS + libc.hdr.types.wchar_t + libc.hdr.types.mbstate_t + libc.src.__support.wchar.mbstate + libc.src.__support.wchar.wcsnrtombs + libc.src.__support.libc_errno +) + +add_entrypoint_object( + wcsnrtombs + SRCS + wcsnrtombs.cpp + HDRS + wcsnrtombs.h + DEPENDS + libc.hdr.types.wchar_t + libc.hdr.types.mbstate_t + libc.src.__support.wchar.mbstate + libc.src.__support.wchar.wcsnrtombs + libc.src.__support.libc_errno +) + +add_entrypoint_object( + mblen + SRCS + mblen.cpp + HDRS + mblen.h + DEPENDS + libc.hdr.types.size_t + libc.src.__support.common + libc.src.__support.macros.config + libc.src.__support.libc_errno + libc.src.__support.wchar.mbrtowc + libc.src.__support.wchar.mbstate +) + +add_entrypoint_object( + mbrlen + SRCS + mbrlen.cpp + HDRS + mbrlen.h + DEPENDS + libc.hdr.types.size_t + libc.hdr.types.mbstate_t + libc.src.__support.common + libc.src.__support.macros.config + libc.src.__support.wchar.mbrtowc + libc.src.__support.libc_errno + libc.src.__support.wchar.mbstate +) + +add_entrypoint_object( wmemset SRCS wmemset.cpp @@ -215,6 +312,19 @@ add_entrypoint_object( ) add_entrypoint_object( + wcsdup + SRCS + wcsdup.cpp + HDRS + wcsdup.h + DEPENDS + libc.hdr.types.wchar_t + libc.src.__support.libc_errno + libc.src.__support.macros.config + libc.src.string.allocating_string_utils +) + +add_entrypoint_object( wcspbrk SRCS wcspbrk.cpp @@ -255,6 +365,8 @@ add_entrypoint_object( DEPENDS libc.hdr.wchar_macros libc.hdr.types.size_t + libc.src.wchar.wchar_utils + libc.src.__support.macros.null_check ) add_entrypoint_object( @@ -266,6 +378,8 @@ add_entrypoint_object( DEPENDS libc.hdr.wchar_macros libc.hdr.types.size_t + libc.src.wchar.wchar_utils + libc.src.__support.macros.null_check ) add_entrypoint_object( diff --git a/libc/src/wchar/mblen.cpp b/libc/src/wchar/mblen.cpp new file mode 100644 index 0000000..2d15b3e --- /dev/null +++ b/libc/src/wchar/mblen.cpp @@ -0,0 +1,35 @@ +//===-- Implementation of mblen -------------------------------------------===// +// +// 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/wchar/mblen.h" + +#include "hdr/types/size_t.h" +#include "src/__support/common.h" +#include "src/__support/libc_errno.h" +#include "src/__support/macros/config.h" +#include "src/__support/wchar/mbrtowc.h" +#include "src/__support/wchar/mbstate.h" + +namespace LIBC_NAMESPACE_DECL { + +LLVM_LIBC_FUNCTION(int, mblen, (const char *s, size_t n)) { + // returns 0 since UTF-8 encoding is not state-dependent + if (s == nullptr) + return 0; + internal::mbstate internal_mbstate; + auto ret = internal::mbrtowc(nullptr, s, n, &internal_mbstate); + if (!ret.has_value() || static_cast<int>(ret.value()) == -2) { + // Encoding failure + if (!ret.has_value()) + libc_errno = EILSEQ; + return -1; + } + return static_cast<int>(ret.value()); +} + +} // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/mblen.h b/libc/src/wchar/mblen.h new file mode 100644 index 0000000..a315a2f --- /dev/null +++ b/libc/src/wchar/mblen.h @@ -0,0 +1,21 @@ +//===-- Implementation header for mblen -----------------------------------===// +// +// 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_WCHAR_MBLEN_H +#define LLVM_LIBC_SRC_WCHAR_MBLEN_H + +#include "hdr/types/size_t.h" +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { + +int mblen(const char *s, size_t n); + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC_WCHAR_MBLEN_H diff --git a/libc/src/wchar/mbrlen.cpp b/libc/src/wchar/mbrlen.cpp new file mode 100644 index 0000000..8de78e0 --- /dev/null +++ b/libc/src/wchar/mbrlen.cpp @@ -0,0 +1,37 @@ +//===-- Implementation of mbrlen ------------------------------------------===// +// +// 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/wchar/mbrlen.h" + +#include "hdr/types/mbstate_t.h" +#include "hdr/types/size_t.h" +#include "src/__support/common.h" +#include "src/__support/libc_errno.h" +#include "src/__support/macros/config.h" +#include "src/__support/wchar/mbrtowc.h" +#include "src/__support/wchar/mbstate.h" + +namespace LIBC_NAMESPACE_DECL { + +LLVM_LIBC_FUNCTION(size_t, mbrlen, + (const char *__restrict s, size_t n, + mbstate_t *__restrict ps)) { + static internal::mbstate internal_mbstate; + auto ret = internal::mbrtowc(nullptr, s, n, + ps == nullptr + ? &internal_mbstate + : reinterpret_cast<internal::mbstate *>(ps)); + if (!ret.has_value()) { + // Encoding failure + libc_errno = ret.error(); + return -1; + } + return ret.value(); +} + +} // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/mbrlen.h b/libc/src/wchar/mbrlen.h new file mode 100644 index 0000000..08b59cf --- /dev/null +++ b/libc/src/wchar/mbrlen.h @@ -0,0 +1,22 @@ +//===-- Implementation header for mbrlen ----------------------------------===// +// +// 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_WCHAR_MBRLEN_H +#define LLVM_LIBC_SRC_WCHAR_MBRLEN_H + +#include "hdr/types/mbstate_t.h" +#include "hdr/types/size_t.h" +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { + +size_t mbrlen(const char *__restrict s, size_t n, mbstate_t *__restrict ps); + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC_WCHAR_MBRLEN_H diff --git a/libc/src/wchar/mbsinit.cpp b/libc/src/wchar/mbsinit.cpp new file mode 100644 index 0000000..23ba542 --- /dev/null +++ b/libc/src/wchar/mbsinit.cpp @@ -0,0 +1,26 @@ +//===-- Implementation of mbsinit -----------------------------------------===// +// +// 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/wchar/mbsinit.h" + +#include "hdr/types/mbstate_t.h" +#include "src/__support/common.h" +#include "src/__support/macros/config.h" +#include "src/__support/wchar/character_converter.h" +#include "src/__support/wchar/mbstate.h" + +namespace LIBC_NAMESPACE_DECL { + +LLVM_LIBC_FUNCTION(int, mbsinit, (mbstate_t * ps)) { + if (ps == nullptr) + return true; + internal::CharacterConverter cr(reinterpret_cast<internal::mbstate *>(ps)); + return cr.isValidState() && cr.isEmpty(); +} + +} // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/mbsinit.h b/libc/src/wchar/mbsinit.h new file mode 100644 index 0000000..fa6be0f --- /dev/null +++ b/libc/src/wchar/mbsinit.h @@ -0,0 +1,22 @@ +//===-- Implementation header for mbsinit ---------------------------------===// +// +// 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_WCHAR_MBSINIT_H +#define LLVM_LIBC_SRC_WCHAR_MBSINIT_H + +#include "hdr/types/mbstate_t.h" +#include "hdr/types/size_t.h" +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { + +int mbsinit(mbstate_t *ps); + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC_WCHAR_MBSINIT_H diff --git a/libc/src/wchar/mbtowc.cpp b/libc/src/wchar/mbtowc.cpp index eae39ba..6d099d4 100644 --- a/libc/src/wchar/mbtowc.cpp +++ b/libc/src/wchar/mbtowc.cpp @@ -25,10 +25,7 @@ LLVM_LIBC_FUNCTION(int, mbtowc, if (s == nullptr) return 0; internal::mbstate internal_mbstate; - // temp ptr to use if pwc is nullptr - wchar_t buf[1]; - auto ret = - internal::mbrtowc(pwc == nullptr ? buf : pwc, s, n, &internal_mbstate); + auto ret = internal::mbrtowc(pwc, s, n, &internal_mbstate); if (!ret.has_value() || static_cast<int>(ret.value()) == -2) { // Encoding failure libc_errno = EILSEQ; diff --git a/libc/src/wchar/wchar_utils.h b/libc/src/wchar/wchar_utils.h new file mode 100644 index 0000000..e0218c7 --- /dev/null +++ b/libc/src/wchar/wchar_utils.h @@ -0,0 +1,45 @@ +//===-- wchar utils ---------------------------------------------*- 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_WCHAR_WCHAR_UTILS_H +#define LLVM_LIBC_SRC_WCHAR_WCHAR_UTILS_H + +#include "hdr/types/size_t.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/common.h" +#include "src/__support/macros/attributes.h" // LIBC_INLINE + +namespace LIBC_NAMESPACE_DECL { +namespace internal { + +// returns true if the character exists in the string +LIBC_INLINE static bool wcschr(wchar_t c, const wchar_t *str) { + for (int n = 0; str[n]; ++n) { + if (str[n] == c) + return true; + } + return false; +} + +// bool should be true for wcscspn for complimentary span +// should be false for wcsspn since we want it to span +LIBC_INLINE static size_t wcsspn(const wchar_t *s1, const wchar_t *s2, + bool not_match_set) { + size_t i = 0; + for (; s1[i]; ++i) { + bool in_set = wcschr(s1[i], s2); + if (in_set == not_match_set) + return i; + } + return i; +} + +} // namespace internal +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC_WCHAR_WCHAR_UTILS_H diff --git a/libc/src/wchar/wcscspn.cpp b/libc/src/wchar/wcscspn.cpp index 8869d84..34f3451 100644 --- a/libc/src/wchar/wcscspn.cpp +++ b/libc/src/wchar/wcscspn.cpp @@ -12,23 +12,15 @@ #include "hdr/types/wchar_t.h" #include "src/__support/common.h" #include "src/__support/macros/config.h" +#include "src/__support/macros/null_check.h" +#include "wchar_utils.h" namespace LIBC_NAMESPACE_DECL { -bool check(wchar_t c, const wchar_t *s2) { - for (int n = 0; s2[n]; ++n) { - if (s2[n] == c) - return false; - } - return true; -} LLVM_LIBC_FUNCTION(size_t, wcscspn, (const wchar_t *s1, const wchar_t *s2)) { - size_t i = 0; - for (; s1[i]; ++i) { - if (!check(s1[i], s2)) - return i; - } - return i; + LIBC_CRASH_ON_NULLPTR(s1); + LIBC_CRASH_ON_NULLPTR(s2); + return internal::wcsspn(s1, s2, /*not_match_set=*/true); } } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/wcsdup.cpp b/libc/src/wchar/wcsdup.cpp new file mode 100644 index 0000000..d4a13d3 --- /dev/null +++ b/libc/src/wchar/wcsdup.cpp @@ -0,0 +1,27 @@ +//===-- Implementation of wcsdup -----------------------------------------===// +// +// 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/wchar/wcsdup.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/common.h" +#include "src/__support/libc_errno.h" +#include "src/__support/macros/config.h" +#include "src/string/allocating_string_utils.h" + +namespace LIBC_NAMESPACE_DECL { + +LLVM_LIBC_FUNCTION(wchar_t *, wcsdup, (const wchar_t *wcs)) { + auto dup = internal::strdup(wcs); + if (dup) + return *dup; + if (wcs != nullptr) + libc_errno = ENOMEM; + return nullptr; +} + +} // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/wcsdup.h b/libc/src/wchar/wcsdup.h new file mode 100644 index 0000000..80b3e52 --- /dev/null +++ b/libc/src/wchar/wcsdup.h @@ -0,0 +1,21 @@ +//===-- Implementation header for wcsdup ----------------------------------===// +// +// 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_WCHAR_WCSDUP_H +#define LLVM_LIBC_SRC_WCHAR_WCSDUP_H + +#include "hdr/types/wchar_t.h" +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { + +wchar_t *wcsdup(const wchar_t *wcs); + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC_WCHAR_WCSDUP_H diff --git a/libc/src/wchar/wcsnrtombs.cpp b/libc/src/wchar/wcsnrtombs.cpp new file mode 100644 index 0000000..7f25b24 --- /dev/null +++ b/libc/src/wchar/wcsnrtombs.cpp @@ -0,0 +1,40 @@ +//===-- Implementation of wcsnrtombs --------------------------------------===// +// +// 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/wchar/wcsnrtombs.h" + +#include "hdr/types/char32_t.h" +#include "hdr/types/mbstate_t.h" +#include "hdr/types/size_t.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/common.h" +#include "src/__support/libc_errno.h" +#include "src/__support/macros/config.h" +#include "src/__support/wchar/mbstate.h" +#include "src/__support/wchar/wcsnrtombs.h" + +namespace LIBC_NAMESPACE_DECL { + +LLVM_LIBC_FUNCTION(size_t, wcsnrtombs, + (char *__restrict s, const wchar_t **__restrict pwcs, + size_t nwc, size_t len, mbstate_t *ps)) { + LIBC_CRASH_ON_NULLPTR(pwcs); + static internal::mbstate internal_mbstate; + auto result = internal::wcsnrtombs( + s, pwcs, nwc, len, + ps == nullptr ? &internal_mbstate + : reinterpret_cast<internal::mbstate *>(ps)); + if (!result.has_value()) { + libc_errno = result.error(); + return -1; + } + + return result.value(); +} + +} // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/wcsnrtombs.h b/libc/src/wchar/wcsnrtombs.h new file mode 100644 index 0000000..bf8add7 --- /dev/null +++ b/libc/src/wchar/wcsnrtombs.h @@ -0,0 +1,24 @@ +//===-- Implementation header for wcsnrtombs ------------------------------===// +// +// 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_WCHAR_WCSNRTOMBS_H +#define LLVM_LIBC_SRC_WCHAR_WCSNRTOMBS_H + +#include "hdr/types/mbstate_t.h" +#include "hdr/types/size_t.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { + +size_t wcsnrtombs(char *__restrict s, const wchar_t **__restrict pwcs, + size_t nwc, size_t len, mbstate_t *ps); + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC_WCHAR_WCSNRTOMBS_H diff --git a/libc/src/wchar/wcsrtombs.cpp b/libc/src/wchar/wcsrtombs.cpp new file mode 100644 index 0000000..9d2508c --- /dev/null +++ b/libc/src/wchar/wcsrtombs.cpp @@ -0,0 +1,40 @@ +//===-- Implementation of wcsrtombs ---------------------------------------===// +// +// 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/wchar/wcsrtombs.h" + +#include "hdr/types/char32_t.h" +#include "hdr/types/mbstate_t.h" +#include "hdr/types/size_t.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/common.h" +#include "src/__support/libc_errno.h" +#include "src/__support/macros/config.h" +#include "src/__support/wchar/mbstate.h" +#include "src/__support/wchar/wcsnrtombs.h" + +namespace LIBC_NAMESPACE_DECL { + +LLVM_LIBC_FUNCTION(size_t, wcsrtombs, + (char *__restrict s, const wchar_t **__restrict pwcs, + size_t n, mbstate_t *ps)) { + LIBC_CRASH_ON_NULLPTR(pwcs); + static internal::mbstate internal_mbstate; + auto result = internal::wcsnrtombs( + s, pwcs, SIZE_MAX, n, + ps == nullptr ? &internal_mbstate + : reinterpret_cast<internal::mbstate *>(ps)); + if (!result.has_value()) { + libc_errno = result.error(); + return -1; + } + + return result.value(); +} + +} // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/wcsrtombs.h b/libc/src/wchar/wcsrtombs.h new file mode 100644 index 0000000..d23573f --- /dev/null +++ b/libc/src/wchar/wcsrtombs.h @@ -0,0 +1,24 @@ +//===-- Implementation header for wcsrtombs -------------------------------===// +// +// 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_WCHAR_WCSRTOMBS_H +#define LLVM_LIBC_SRC_WCHAR_WCSRTOMBS_H + +#include "hdr/types/mbstate_t.h" +#include "hdr/types/size_t.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { + +size_t wcsrtombs(char *__restrict s, const wchar_t **__restrict pwcs, size_t n, + mbstate_t *ps); + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC_WCHAR_WCSRTOMBS_H diff --git a/libc/src/wchar/wcsspn.cpp b/libc/src/wchar/wcsspn.cpp index 23de381..ae2cf5a 100644 --- a/libc/src/wchar/wcsspn.cpp +++ b/libc/src/wchar/wcsspn.cpp @@ -12,23 +12,15 @@ #include "hdr/types/wchar_t.h" #include "src/__support/common.h" #include "src/__support/macros/config.h" +#include "src/__support/macros/null_check.h" +#include "wchar_utils.h" namespace LIBC_NAMESPACE_DECL { -bool check(wchar_t c, const wchar_t *s2) { - for (int n = 0; s2[n]; ++n) { - if (s2[n] == c) - return true; - } - return false; -} LLVM_LIBC_FUNCTION(size_t, wcsspn, (const wchar_t *s1, const wchar_t *s2)) { - size_t i = 0; - for (; s1[i]; ++i) { - if (!check(s1[i], s2)) - return i; - } - return i; + LIBC_CRASH_ON_NULLPTR(s1); + LIBC_CRASH_ON_NULLPTR(s2); + return internal::wcsspn(s1, s2, /*not_match_set=*/false); } } // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/wcstombs.cpp b/libc/src/wchar/wcstombs.cpp new file mode 100644 index 0000000..c3793cb --- /dev/null +++ b/libc/src/wchar/wcstombs.cpp @@ -0,0 +1,38 @@ +//===-- Implementation of wcstombs ----------------------------------------===// +// +// 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/wchar/wcstombs.h" + +#include "hdr/types/char32_t.h" +#include "hdr/types/size_t.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/common.h" +#include "src/__support/libc_errno.h" +#include "src/__support/macros/config.h" +#include "src/__support/wchar/mbstate.h" +#include "src/__support/wchar/wcsnrtombs.h" + +namespace LIBC_NAMESPACE_DECL { + +LLVM_LIBC_FUNCTION(size_t, wcstombs, + (char *__restrict s, const wchar_t *__restrict wcs, + size_t n)) { + LIBC_CRASH_ON_NULLPTR(wcs); + static internal::mbstate internal_mbstate; + const wchar_t *wcs_ptr_copy = wcs; + auto result = + internal::wcsnrtombs(s, &wcs_ptr_copy, SIZE_MAX, n, &internal_mbstate); + if (!result.has_value()) { + libc_errno = result.error(); + return -1; + } + + return result.value(); +} + +} // namespace LIBC_NAMESPACE_DECL diff --git a/libc/src/wchar/wcstombs.h b/libc/src/wchar/wcstombs.h new file mode 100644 index 0000000..cd0008a --- /dev/null +++ b/libc/src/wchar/wcstombs.h @@ -0,0 +1,22 @@ +//===-- Implementation header for wcstombs --------------------------------===// +// +// 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_WCHAR_WCSTOMBS_H +#define LLVM_LIBC_SRC_WCHAR_WCSTOMBS_H + +#include "hdr/types/size_t.h" +#include "hdr/types/wchar_t.h" +#include "src/__support/macros/config.h" + +namespace LIBC_NAMESPACE_DECL { + +size_t wcstombs(char *__restrict s, const wchar_t *__restrict pwcs, size_t n); + +} // namespace LIBC_NAMESPACE_DECL + +#endif // LLVM_LIBC_SRC_WCHAR_WCSTOMBS_H |