1 // Copyright 2022 The Abseil Authors.
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 //      https://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 
15 #ifndef ABSL_CRC_INTERNAL_CRC_INTERNAL_H_
16 #define ABSL_CRC_INTERNAL_CRC_INTERNAL_H_
17 
18 #include <cstdint>
19 #include <memory>
20 #include <vector>
21 
22 #include "absl/base/internal/raw_logging.h"
23 #include "absl/crc/internal/crc.h"
24 
25 namespace absl {
26 ABSL_NAMESPACE_BEGIN
27 
28 namespace crc_internal {
29 
30 // Prefetch constants used in some Extend() implementations
31 constexpr int kPrefetchHorizon = ABSL_CACHELINE_SIZE * 4;  // Prefetch this far
32 // Shorter prefetch distance for smaller buffers
33 constexpr int kPrefetchHorizonMedium = ABSL_CACHELINE_SIZE * 1;
34 static_assert(kPrefetchHorizon >= 64, "CRCPrefetchHorizon less than loop len");
35 
36 // We require the Scramble() function:
37 //  - to be reversible (Unscramble() must exist)
38 //  - to be non-linear in the polynomial's Galois field (so the CRC of a
39 //    scrambled CRC is not linearly affected by the scrambled CRC, even if
40 //    using the same polynomial)
41 //  - not to be its own inverse.  Preferably, if X=Scramble^N(X) and N!=0, then
42 //    N is large.
43 //  - to be fast.
44 //  - not to change once defined.
45 // We introduce non-linearity in two ways:
46 //     Addition of a constant.
47 //         - The carries introduce non-linearity; we use bits of an irrational
48 //           (phi) to make it unlikely that we introduce no carries.
49 //     Rotate by a constant number of bits.
50 //         - We use floor(degree/2)+1, which does not divide the degree, and
51 //           splits the bits nearly evenly, which makes it less likely the
52 //           halves will be the same or one will be all zeroes.
53 // We do both things to improve the chances of non-linearity in the face of
54 // bit patterns with low numbers of bits set, while still being fast.
55 // Below is the constant that we add.  The bits are the first 128 bits of the
56 // fractional part of phi, with a 1 ored into the bottom bit to maximize the
57 // cycle length of repeated adds.
58 constexpr uint64_t kScrambleHi = (static_cast<uint64_t>(0x4f1bbcdcU) << 32) |
59                                  static_cast<uint64_t>(0xbfa53e0aU);
60 constexpr uint64_t kScrambleLo = (static_cast<uint64_t>(0xf9ce6030U) << 32) |
61                                  static_cast<uint64_t>(0x2e76e41bU);
62 
63 class CRCImpl : public CRC {  // Implemention of the abstract class CRC
64  public:
65   using Uint32By256 = uint32_t[256];
66 
CRCImpl()67   CRCImpl() {}
68   ~CRCImpl() override = default;
69 
70   // The internal version of CRC::New().
71   static CRCImpl* NewInternal();
72 
73   void Empty(uint32_t* crc) const override;
74 
75   // Fill in a table for updating a CRC by one word of 'word_size' bytes
76   // [last_lo, last_hi] contains the answer if the last bit in the word
77   // is set.
78   static void FillWordTable(uint32_t poly, uint32_t last, int word_size,
79                             Uint32By256* t);
80 
81   // Build the table for extending by zeroes, returning the number of entries.
82   // For a in {1, 2, ..., ZEROES_BASE-1}, b in {0, 1, 2, 3, ...},
83   // entry j=a-1+(ZEROES_BASE-1)*b
84   // contains a polynomial Pi such that multiplying
85   // a CRC by Pi mod P, where P is the CRC polynomial, is equivalent to
86   // appending a*2**(ZEROES_BASE_LG*b) zero bytes to the original string.
87   static int FillZeroesTable(uint32_t poly, Uint32By256* t);
88 
89   virtual void InitTables() = 0;
90 
91  private:
92   CRCImpl(const CRCImpl&) = delete;
93   CRCImpl& operator=(const CRCImpl&) = delete;
94 };
95 
96 // This is the 32-bit implementation.  It handles all sizes from 8 to 32.
97 class CRC32 : public CRCImpl {
98  public:
CRC32()99   CRC32() {}
~CRC32()100   ~CRC32() override {}
101 
102   void Extend(uint32_t* crc, const void* bytes, size_t length) const override;
103   void ExtendByZeroes(uint32_t* crc, size_t length) const override;
104   void Scramble(uint32_t* crc) const override;
105   void Unscramble(uint32_t* crc) const override;
106   void UnextendByZeroes(uint32_t* crc, size_t length) const override;
107 
108   void InitTables() override;
109 
110  private:
111   // Common implementation guts for ExtendByZeroes and UnextendByZeroes().
112   //
113   // zeroes_table is a table as returned by FillZeroesTable(), containing
114   // polynomials representing CRCs of strings-of-zeros of various lenghts,
115   // and which can be combined by polynomial multiplication.  poly_table is
116   // a table of CRC byte extension values.  These tables are determined by
117   // the generator polynomial.
118   //
119   // These will be set to reverse_zeroes_ and reverse_table0_ for Unextend, and
120   // CRC32::zeroes_ and CRC32::table0_ for Extend.
121   void ExtendByZeroesImpl(uint32_t* crc, size_t length,
122                           const uint32_t zeroes_table[256],
123                           const uint32_t poly_table[256]) const;
124 
125   uint32_t table0_[256];  // table of byte extensions
126   uint32_t zeroes_[256];  // table of zero extensions
127 
128   // table of 4-byte extensions shifted by 12 bytes of zeroes
129   uint32_t table_[4][256];
130 
131   // Reverse lookup tables, using the alternate polynomial used by
132   // UnextendByZeroes().
133   uint32_t reverse_table0_[256];  // table of reverse byte extensions
134   uint32_t reverse_zeroes_[256];  // table of reverse zero extensions
135 
136   CRC32(const CRC32&) = delete;
137   CRC32& operator=(const CRC32&) = delete;
138 };
139 
140 // Helpers
141 
142 // Return a bit mask containing len 1-bits.
143 // Requires 0 < len <= sizeof(T)
144 template <typename T>
MaskOfLength(int len)145 T MaskOfLength(int len) {
146   // shift 2 by len-1 rather than 1 by len because shifts of wordsize
147   // are undefined.
148   return (T(2) << (len - 1)) - 1;
149 }
150 
151 // Rotate low-order "width" bits of "in" right by "r" bits,
152 // setting other bits in word to arbitrary values.
153 template <typename T>
RotateRight(T in,int width,int r)154 T RotateRight(T in, int width, int r) {
155   return (in << (width - r)) | ((in >> r) & MaskOfLength<T>(width - r));
156 }
157 
158 // RoundUp<N>(p) returns the lowest address >= p aligned to an N-byte
159 // boundary.  Requires that N is a power of 2.
160 template <int alignment>
RoundUp(const uint8_t * p)161 const uint8_t* RoundUp(const uint8_t* p) {
162   static_assert((alignment & (alignment - 1)) == 0, "alignment is not 2^n");
163   constexpr uintptr_t mask = alignment - 1;
164   const uintptr_t as_uintptr = reinterpret_cast<uintptr_t>(p);
165   return reinterpret_cast<const uint8_t*>((as_uintptr + mask) & ~mask);
166 }
167 
168 // Return a newly created CRC32AcceleratedX86ARMCombined if we can use Intel's
169 // or ARM's CRC acceleration for a given polynomial.  Return nullptr otherwise.
170 CRCImpl* TryNewCRC32AcceleratedX86ARMCombined();
171 
172 // Return all possible hardware accelerated implementations. For testing only.
173 std::vector<std::unique_ptr<CRCImpl>> NewCRC32AcceleratedX86ARMCombinedAll();
174 
175 }  // namespace crc_internal
176 ABSL_NAMESPACE_END
177 }  // namespace absl
178 
179 #endif  // ABSL_CRC_INTERNAL_CRC_INTERNAL_H_
180