85 lines · c
1//===-- Memcmp implementation for aarch64 -----------------------*- 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-exception6//7//===----------------------------------------------------------------------===//8#ifndef LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_MEMCMP_H9#define LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_MEMCMP_H10 11#include "src/__support/macros/attributes.h" // LIBC_INLINE12#include "src/__support/macros/optimization.h" // LIBC_UNLIKELY13#include "src/string/memory_utils/op_aarch64.h"14#include "src/string/memory_utils/op_generic.h"15#include "src/string/memory_utils/utils.h" // MemcmpReturnType16 17namespace LIBC_NAMESPACE_DECL {18 19[[maybe_unused]] LIBC_INLINE MemcmpReturnType20inline_memcmp_aarch64_no_fp(CPtr p1, CPtr p2, size_t count) {21 if (count == 0)22 return MemcmpReturnType::zero();23 if (count == 1)24 return generic::Memcmp<uint8_t>::block(p1, p2);25 if (count == 2)26 return generic::Memcmp<uint16_t>::block(p1, p2);27 if (count == 3)28 return generic::MemcmpSequence<uint16_t, uint8_t>::block(p1, p2);29 if (count <= 8)30 return generic::Memcmp<uint32_t>::head_tail(p1, p2, count);31 if (count <= 16)32 return generic::Memcmp<uint64_t>::head_tail(p1, p2, count);33 34 return generic::Memcmp<uint64_t>::loop_and_tail_align_above(384, p1, p2,35 count);36}37 38#if defined(__ARM_NEON)39[[maybe_unused]] LIBC_INLINE MemcmpReturnType40inline_memcmp_aarch64_with_fp(CPtr p1, CPtr p2, size_t count) {41 if (count == 0)42 return MemcmpReturnType::zero();43 if (count == 1)44 return generic::Memcmp<uint8_t>::block(p1, p2);45 if (count == 2)46 return generic::Memcmp<uint16_t>::block(p1, p2);47 if (count == 3)48 return generic::MemcmpSequence<uint16_t, uint8_t>::block(p1, p2);49 if (count <= 8)50 return generic::Memcmp<uint32_t>::head_tail(p1, p2, count);51 if (count <= 16)52 return generic::Memcmp<uint64_t>::head_tail(p1, p2, count);53 54 if (LIBC_UNLIKELY(count >= 128)) { // [128, ∞]55 if (auto value = generic::Memcmp<uint8x16_t>::block(p1, p2))56 return value;57 align_to_next_boundary<16, Arg::P1>(p1, p2, count);58 return generic::Memcmp<uint8x16x2_t>::loop_and_tail(p1, p2, count);59 }60 if (generic::Bcmp<uint8x16_t>::block(p1, p2)) // [16, 16]61 return generic::Memcmp<uint8x16_t>::block(p1, p2);62 if (count < 32) // [17, 31]63 return generic::Memcmp<uint8x16_t>::tail(p1, p2, count);64 if (generic::Bcmp<uint8x16_t>::block(p1 + 16, p2 + 16)) // [32, 32]65 return generic::Memcmp<uint8x16_t>::block(p1 + 16, p2 + 16);66 if (count < 64) // [33, 63]67 return generic::Memcmp<uint8x16x2_t>::tail(p1, p2, count);68 // [64, 127]69 return generic::Memcmp<uint8x16_t>::loop_and_tail(p1 + 32, p2 + 32,70 count - 32);71}72#endif73 74[[gnu::flatten]] LIBC_INLINE MemcmpReturnType75inline_memcmp_aarch64_dispatch(CPtr p1, CPtr p2, size_t count) {76#if defined(__ARM_NEON)77 return inline_memcmp_aarch64_with_fp(p1, p2, count);78#else79 return inline_memcmp_aarch64_no_fp(p1, p2, count);80#endif81}82} // namespace LIBC_NAMESPACE_DECL83 84#endif // LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_MEMCMP_H85