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-exception
6 //
7 //===----------------------------------------------------------------------===//
8 #ifndef LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_MEMCMP_H
9 #define LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_MEMCMP_H
10
11 #include "src/__support/macros/attributes.h" // LIBC_INLINE
12 #include "src/__support/macros/optimization.h" // LIBC_UNLIKELY
13 #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" // MemcmpReturnType
16
17 namespace LIBC_NAMESPACE_DECL {
18
19 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
inline_memcmp_generic_gt16(CPtr p1,CPtr p2,size_t count)20 inline_memcmp_generic_gt16(CPtr p1, CPtr p2, size_t count) {
21 if (LIBC_UNLIKELY(count >= 384)) {
22 if (auto value = generic::Memcmp<uint8x16_t>::block(p1, p2))
23 return value;
24 align_to_next_boundary<16, Arg::P1>(p1, p2, count);
25 }
26 return generic::Memcmp<uint8x16_t>::loop_and_tail(p1, p2, count);
27 }
28
29 [[maybe_unused]] LIBC_INLINE MemcmpReturnType
inline_memcmp_aarch64_neon_gt16(CPtr p1,CPtr p2,size_t count)30 inline_memcmp_aarch64_neon_gt16(CPtr p1, CPtr p2, size_t count) {
31 if (LIBC_UNLIKELY(count >= 128)) { // [128, ∞]
32 if (auto value = generic::Memcmp<uint8x16_t>::block(p1, p2))
33 return value;
34 align_to_next_boundary<16, Arg::P1>(p1, p2, count);
35 return generic::Memcmp<uint8x16x2_t>::loop_and_tail(p1, p2, count);
36 }
37 if (generic::Bcmp<uint8x16_t>::block(p1, p2)) // [16, 16]
38 return generic::Memcmp<uint8x16_t>::block(p1, p2);
39 if (count < 32) // [17, 31]
40 return generic::Memcmp<uint8x16_t>::tail(p1, p2, count);
41 if (generic::Bcmp<uint8x16_t>::block(p1 + 16, p2 + 16)) // [32, 32]
42 return generic::Memcmp<uint8x16_t>::block(p1 + 16, p2 + 16);
43 if (count < 64) // [33, 63]
44 return generic::Memcmp<uint8x16x2_t>::tail(p1, p2, count);
45 // [64, 127]
46 return generic::Memcmp<uint8x16_t>::loop_and_tail(p1 + 32, p2 + 32,
47 count - 32);
48 }
49
inline_memcmp_aarch64(CPtr p1,CPtr p2,size_t count)50 LIBC_INLINE MemcmpReturnType inline_memcmp_aarch64(CPtr p1, CPtr p2,
51 size_t count) {
52 if (count == 0)
53 return MemcmpReturnType::zero();
54 if (count == 1)
55 return generic::Memcmp<uint8_t>::block(p1, p2);
56 if (count == 2)
57 return generic::Memcmp<uint16_t>::block(p1, p2);
58 if (count == 3)
59 return generic::MemcmpSequence<uint16_t, uint8_t>::block(p1, p2);
60 if (count <= 8)
61 return generic::Memcmp<uint32_t>::head_tail(p1, p2, count);
62 if (count <= 16)
63 return generic::Memcmp<uint64_t>::head_tail(p1, p2, count);
64 if constexpr (aarch64::kNeon)
65 return inline_memcmp_aarch64_neon_gt16(p1, p2, count);
66 else
67 return inline_memcmp_generic_gt16(p1, p2, count);
68 }
69 } // namespace LIBC_NAMESPACE_DECL
70
71 #endif // LIBC_SRC_STRING_MEMORY_UTILS_X86_64_INLINE_MEMCMP_H
72