95 lines · c
1//===-- Bcmp implementation for x86_64 --------------------------*- 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 LLVM_LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_BCMP_H9#define LLVM_LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_BCMP_H10 11#include "src/__support/macros/attributes.h" // LIBC_INLINE12#include "src/__support/macros/config.h" // LIBC_NAMESPACE_DECL13#include "src/string/memory_utils/op_generic.h"14#include "src/string/memory_utils/op_x86.h"15#include "src/string/memory_utils/utils.h" // Ptr, CPtr16 17#include <stddef.h> // size_t18 19namespace LIBC_NAMESPACE_DECL {20 21[[maybe_unused]] LIBC_INLINE BcmpReturnType22inline_bcmp_generic_gt16(CPtr p1, CPtr p2, size_t count) {23 return generic::Bcmp<uint64_t>::loop_and_tail_align_above(256, p1, p2, count);24}25 26#if defined(__SSE4_1__)27[[maybe_unused]] LIBC_INLINE BcmpReturnType28inline_bcmp_x86_sse41_gt16(CPtr p1, CPtr p2, size_t count) {29 if (count <= 32)30 return generic::Bcmp<__m128i>::head_tail(p1, p2, count);31 return generic::Bcmp<__m128i>::loop_and_tail_align_above(256, p1, p2, count);32}33#endif // __SSE4_1__34 35#if defined(__AVX__)36[[maybe_unused]] LIBC_INLINE BcmpReturnType37inline_bcmp_x86_avx_gt16(CPtr p1, CPtr p2, size_t count) {38 if (count <= 32)39 return generic::Bcmp<__m128i>::head_tail(p1, p2, count);40 if (count <= 64)41 return generic::Bcmp<__m256i>::head_tail(p1, p2, count);42 return generic::Bcmp<__m256i>::loop_and_tail_align_above(256, p1, p2, count);43}44#endif // __AVX__45 46#if defined(__AVX512BW__)47[[maybe_unused]] LIBC_INLINE BcmpReturnType48inline_bcmp_x86_avx512bw_gt16(CPtr p1, CPtr p2, size_t count) {49 if (count <= 32)50 return generic::Bcmp<__m128i>::head_tail(p1, p2, count);51 if (count <= 64)52 return generic::Bcmp<__m256i>::head_tail(p1, p2, count);53 if (count <= 128)54 return generic::Bcmp<__m512i>::head_tail(p1, p2, count);55 return generic::Bcmp<__m512i>::loop_and_tail_align_above(256, p1, p2, count);56}57#endif // __AVX512BW__58 59[[maybe_unused]] LIBC_INLINE BcmpReturnType inline_bcmp_x86(CPtr p1, CPtr p2,60 size_t count) {61 if (count == 0)62 return BcmpReturnType::zero();63 if (count == 1)64 return generic::Bcmp<uint8_t>::block(p1, p2);65 if (count == 2)66 return generic::Bcmp<uint16_t>::block(p1, p2);67 if (count == 3)68 return generic::BcmpSequence<uint16_t, uint8_t>::block(p1, p2);69 if (count == 4)70 return generic::Bcmp<uint32_t>::block(p1, p2);71 if (count == 5)72 return generic::BcmpSequence<uint32_t, uint8_t>::block(p1, p2);73 if (count == 6)74 return generic::BcmpSequence<uint32_t, uint16_t>::block(p1, p2);75 if (count == 7)76 return generic::BcmpSequence<uint32_t, uint16_t, uint8_t>::block(p1, p2);77 if (count == 8)78 return generic::Bcmp<uint64_t>::block(p1, p2);79 if (count <= 16)80 return generic::Bcmp<uint64_t>::head_tail(p1, p2, count);81#if defined(__AVX512BW__)82 return inline_bcmp_x86_avx512bw_gt16(p1, p2, count);83#elif defined(__AVX__)84 return inline_bcmp_x86_avx_gt16(p1, p2, count);85#elif defined(__SSE4_1__)86 return inline_bcmp_x86_sse41_gt16(p1, p2, count);87#else88 return inline_bcmp_generic_gt16(p1, p2, count);89#endif90}91 92} // namespace LIBC_NAMESPACE_DECL93 94#endif // LLVM_LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_BCMP_H95