| 1 | //===-- Fast rounding to nearest integer for floating point -----*- C++ -*-===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | #ifndef LLVM_LIBC_SRC___SUPPORT_FPUTIL_NEAREST_INTEGER_H |
| 10 | #define LLVM_LIBC_SRC___SUPPORT_FPUTIL_NEAREST_INTEGER_H |
| 11 | |
| 12 | #include "src/__support/CPP/type_traits.h" |
| 13 | #include "src/__support/macros/config.h" |
| 14 | #include "src/__support/macros/optimization.h" // LIBC_UNLIKELY |
| 15 | #include "src/__support/macros/properties/architectures.h" |
| 16 | #include "src/__support/macros/properties/cpu_features.h" |
| 17 | |
| 18 | #if (defined(LIBC_TARGET_ARCH_IS_X86_64) && defined(LIBC_TARGET_CPU_HAS_SSE4_2)) |
| 19 | #include "x86_64/nearest_integer.h" |
| 20 | #elif (defined(LIBC_TARGET_ARCH_IS_AARCH64) && defined(__ARM_FP)) |
| 21 | #include "aarch64/nearest_integer.h" |
| 22 | #elif defined(LIBC_TARGET_ARCH_IS_GPU) |
| 23 | |
| 24 | namespace LIBC_NAMESPACE_DECL { |
| 25 | namespace fputil { |
| 26 | |
| 27 | LIBC_INLINE float nearest_integer(float x) { return __builtin_rintf(x); } |
| 28 | |
| 29 | LIBC_INLINE double nearest_integer(double x) { return __builtin_rint(x); } |
| 30 | |
| 31 | } // namespace fputil |
| 32 | } // namespace LIBC_NAMESPACE_DECL |
| 33 | |
| 34 | #else |
| 35 | |
| 36 | namespace LIBC_NAMESPACE_DECL { |
| 37 | namespace fputil { |
| 38 | |
| 39 | // This is a fast implementation for rounding to a nearest integer that. |
| 40 | // |
| 41 | // Notice that for AARCH64 and x86-64 with SSE4.2 support, we will use their |
| 42 | // corresponding rounding instruction instead. And in those cases, the results |
| 43 | // are rounded to the nearest integer, tie-to-even. |
| 44 | LIBC_INLINE constexpr float nearest_integer(float x) { |
| 45 | #if __has_builtin(__builtin_roundevenf) && defined(__ARM_ARCH) && \ |
| 46 | (__ARM_ARCH >= 8) && defined(LIBC_TARGET_CPU_HAS_ARM_FPU_FLOAT) |
| 47 | if (!cpp::is_constant_evaluated()) |
| 48 | return __builtin_roundevenf(x); |
| 49 | #endif |
| 50 | if (x < 0x1p24f && x > -0x1p24f) { |
| 51 | float r = x < 0 ? (x - 0x1.0p23f) + 0x1.0p23f : (x + 0x1.0p23f) - 0x1.0p23f; |
| 52 | #ifndef LIBC_MATH_HAS_ASSUME_ROUND_NEAREST_ONLY |
| 53 | float diff = x - r; |
| 54 | // The expression above is correct for the default rounding mode, round-to- |
| 55 | // nearest, tie-to-even. For other rounding modes, it might be off by 1, |
| 56 | // which is corrected below. |
| 57 | if (LIBC_UNLIKELY(diff > 0.5f)) |
| 58 | return r + 1.0f; |
| 59 | if (LIBC_UNLIKELY(diff < -0.5f)) |
| 60 | return r - 1.0f; |
| 61 | #endif // !LIBC_MATH_HAS_ASSUME_ROUND_NEAREST_ONLY |
| 62 | return r; |
| 63 | } |
| 64 | return x; |
| 65 | } |
| 66 | |
| 67 | LIBC_INLINE constexpr double nearest_integer(double x) { |
| 68 | #if __has_builtin(__builtin_roundeven) && defined(__ARM_ARCH) && \ |
| 69 | (__ARM_ARCH >= 8) && defined(LIBC_TARGET_CPU_HAS_ARM_FPU_DOUBLE) |
| 70 | if (!cpp::is_constant_evaluated()) |
| 71 | return __builtin_roundeven(x); |
| 72 | #endif |
| 73 | if (x < 0x1p53 && x > -0x1p53) { |
| 74 | double r = x < 0 ? (x - 0x1.0p52) + 0x1.0p52 : (x + 0x1.0p52) - 0x1.0p52; |
| 75 | #ifndef LIBC_MATH_HAS_ASSUME_ROUND_NEAREST_ONLY |
| 76 | double diff = x - r; |
| 77 | // The expression above is correct for the default rounding mode, round-to- |
| 78 | // nearest, tie-to-even. For other rounding modes, it might be off by 1, |
| 79 | // which is corrected below. |
| 80 | if (LIBC_UNLIKELY(diff > 0.5)) |
| 81 | return r + 1.0; |
| 82 | if (LIBC_UNLIKELY(diff < -0.5)) |
| 83 | return r - 1.0; |
| 84 | #endif // !LIBC_MATH_HAS_ASSUME_ROUND_NEAREST_ONLY |
| 85 | return r; |
| 86 | } |
| 87 | return x; |
| 88 | } |
| 89 | |
| 90 | } // namespace fputil |
| 91 | } // namespace LIBC_NAMESPACE_DECL |
| 92 | |
| 93 | #endif |
| 94 | #endif // LLVM_LIBC_SRC___SUPPORT_FPUTIL_NEAREST_INTEGER_H |
| 95 | |