• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2012 Google Inc.
3  *
4  * Use of this source code is governed by a BSD-style license that can be
5  * found in the LICENSE file.
6  */
7 
8 #ifndef SkChecksum_DEFINED
9 #define SkChecksum_DEFINED
10 
11 #include "SkTypes.h"
12 
13 /**
14  *  Computes a 32bit checksum from a blob of 32bit aligned data. This is meant
15  *  to be very very fast, as it is used internally by the font cache, in
16  *  conjuction with the entire raw key. This algorithm does not generate
17  *  unique values as well as others (e.g. MD5) but it performs much faster.
18  *  Skia's use cases can survive non-unique values (since the entire key is
19  *  always available). Clients should only be used in circumstances where speed
20  *  over uniqueness is at a premium.
21  */
22 class SkChecksum : SkNoncopyable {
23 private:
24     /*
25      *  Our Rotate and Mash helpers are meant to automatically do the right
26      *  thing depending if sizeof(uintptr_t) is 4 or 8.
27      */
28     enum {
29         ROTR = 17,
30         ROTL = sizeof(uintptr_t) * 8 - ROTR,
31         HALFBITS = sizeof(uintptr_t) * 4
32     };
33 
Mash(uintptr_t total,uintptr_t value)34     static inline uintptr_t Mash(uintptr_t total, uintptr_t value) {
35         return ((total >> ROTR) | (total << ROTL)) ^ value;
36     }
37 
38 public:
39 
40     /**
41      * Calculate 32-bit Murmur hash (murmur3).
42      * This should take 2-3x longer than SkChecksum::Compute, but is a considerably better hash.
43      * See en.wikipedia.org/wiki/MurmurHash.
44      *
45      *  @param data Memory address of the data block to be processed. Must be 32-bit aligned.
46      *  @param size Size of the data block in bytes. Must be a multiple of 4.
47      *  @param seed Initial hash seed. (optional)
48      *  @return hash result
49      */
50     static uint32_t Murmur3(const uint32_t* data, size_t bytes, uint32_t seed=0) {
51         SkASSERT(SkIsAlign4(bytes));
52         const size_t words = bytes/4;
53 
54         uint32_t hash = seed;
55         for (size_t i = 0; i < words; i++) {
56             uint32_t k = data[i];
57             k *= 0xcc9e2d51;
58             k = (k << 15) | (k >> 17);
59             k *= 0x1b873593;
60 
61             hash ^= k;
62             hash = (hash << 13) | (hash >> 19);
63             hash *= 5;
64             hash += 0xe6546b64;
65         }
66         hash ^= bytes;
67         hash ^= hash >> 16;
68         hash *= 0x85ebca6b;
69         hash ^= hash >> 13;
70         hash *= 0xc2b2ae35;
71         hash ^= hash >> 16;
72         return hash;
73     }
74 
75     /**
76      *  Compute a 32-bit checksum for a given data block
77      *
78      *  WARNING: this algorithm is tuned for efficiency, not backward/forward
79      *  compatibility.  It may change at any time, so a checksum generated with
80      *  one version of the Skia code may not match a checksum generated with
81      *  a different version of the Skia code.
82      *
83      *  @param data Memory address of the data block to be processed. Must be
84      *              32-bit aligned.
85      *  @param size Size of the data block in bytes. Must be a multiple of 4.
86      *  @return checksum result
87      */
Compute(const uint32_t * data,size_t size)88     static uint32_t Compute(const uint32_t* data, size_t size) {
89         SkASSERT(SkIsAlign4(size));
90 
91         /*
92          *  We want to let the compiler use 32bit or 64bit addressing and math
93          *  so we use uintptr_t as our magic type. This makes the code a little
94          *  more obscure (we can't hard-code 32 or 64 anywhere, but have to use
95          *  sizeof()).
96          */
97         uintptr_t result = 0;
98         const uintptr_t* ptr = reinterpret_cast<const uintptr_t*>(data);
99 
100         /*
101          *  count the number of quad element chunks. This takes into account
102          *  if we're on a 32bit or 64bit arch, since we use sizeof(uintptr_t)
103          *  to compute how much to shift-down the size.
104          */
105         size_t n4 = size / (sizeof(uintptr_t) << 2);
106         for (size_t i = 0; i < n4; ++i) {
107             result = Mash(result, *ptr++);
108             result = Mash(result, *ptr++);
109             result = Mash(result, *ptr++);
110             result = Mash(result, *ptr++);
111         }
112         size &= ((sizeof(uintptr_t) << 2) - 1);
113 
114         data = reinterpret_cast<const uint32_t*>(ptr);
115         const uint32_t* stop = data + (size >> 2);
116         while (data < stop) {
117             result = Mash(result, *data++);
118         }
119 
120         /*
121          *  smash us down to 32bits if we were 64. Note that when uintptr_t is
122          *  32bits, this code-path should go away, but I still got a warning
123          *  when I wrote
124          *      result ^= result >> 32;
125          *  since >>32 is undefined for 32bit ints, hence the wacky HALFBITS
126          *  define.
127          */
128         if (8 == sizeof(result)) {
129             result ^= result >> HALFBITS;
130         }
131         return static_cast<uint32_t>(result);
132     }
133 };
134 
135 #endif
136