Skip to content

Commit c0b6660

Browse files
authored
Merge pull request #5023 from MartinNowak/faster_crc
faster CRC32 implementation
2 parents 2e5b4f0 + 382f9d2 commit c0b6660

File tree

1 file changed

+75
-42
lines changed

1 file changed

+75
-42
lines changed

std/digest/crc.d

Lines changed: 75 additions & 42 deletions
Original file line numberDiff line numberDiff line change
@@ -100,46 +100,39 @@ unittest
100100
hash = crc.finish();
101101
}
102102

103-
private immutable uint[256] crc32_table =
104-
[
105-
0x00000000,0x77073096,0xee0e612c,0x990951ba,0x076dc419,0x706af48f,0xe963a535,
106-
0x9e6495a3,0x0edb8832,0x79dcb8a4,0xe0d5e91e,0x97d2d988,0x09b64c2b,0x7eb17cbd,
107-
0xe7b82d07,0x90bf1d91,0x1db71064,0x6ab020f2,0xf3b97148,0x84be41de,0x1adad47d,
108-
0x6ddde4eb,0xf4d4b551,0x83d385c7,0x136c9856,0x646ba8c0,0xfd62f97a,0x8a65c9ec,
109-
0x14015c4f,0x63066cd9,0xfa0f3d63,0x8d080df5,0x3b6e20c8,0x4c69105e,0xd56041e4,
110-
0xa2677172,0x3c03e4d1,0x4b04d447,0xd20d85fd,0xa50ab56b,0x35b5a8fa,0x42b2986c,
111-
0xdbbbc9d6,0xacbcf940,0x32d86ce3,0x45df5c75,0xdcd60dcf,0xabd13d59,0x26d930ac,
112-
0x51de003a,0xc8d75180,0xbfd06116,0x21b4f4b5,0x56b3c423,0xcfba9599,0xb8bda50f,
113-
0x2802b89e,0x5f058808,0xc60cd9b2,0xb10be924,0x2f6f7c87,0x58684c11,0xc1611dab,
114-
0xb6662d3d,0x76dc4190,0x01db7106,0x98d220bc,0xefd5102a,0x71b18589,0x06b6b51f,
115-
0x9fbfe4a5,0xe8b8d433,0x7807c9a2,0x0f00f934,0x9609a88e,0xe10e9818,0x7f6a0dbb,
116-
0x086d3d2d,0x91646c97,0xe6635c01,0x6b6b51f4,0x1c6c6162,0x856530d8,0xf262004e,
117-
0x6c0695ed,0x1b01a57b,0x8208f4c1,0xf50fc457,0x65b0d9c6,0x12b7e950,0x8bbeb8ea,
118-
0xfcb9887c,0x62dd1ddf,0x15da2d49,0x8cd37cf3,0xfbd44c65,0x4db26158,0x3ab551ce,
119-
0xa3bc0074,0xd4bb30e2,0x4adfa541,0x3dd895d7,0xa4d1c46d,0xd3d6f4fb,0x4369e96a,
120-
0x346ed9fc,0xad678846,0xda60b8d0,0x44042d73,0x33031de5,0xaa0a4c5f,0xdd0d7cc9,
121-
0x5005713c,0x270241aa,0xbe0b1010,0xc90c2086,0x5768b525,0x206f85b3,0xb966d409,
122-
0xce61e49f,0x5edef90e,0x29d9c998,0xb0d09822,0xc7d7a8b4,0x59b33d17,0x2eb40d81,
123-
0xb7bd5c3b,0xc0ba6cad,0xedb88320,0x9abfb3b6,0x03b6e20c,0x74b1d29a,0xead54739,
124-
0x9dd277af,0x04db2615,0x73dc1683,0xe3630b12,0x94643b84,0x0d6d6a3e,0x7a6a5aa8,
125-
0xe40ecf0b,0x9309ff9d,0x0a00ae27,0x7d079eb1,0xf00f9344,0x8708a3d2,0x1e01f268,
126-
0x6906c2fe,0xf762575d,0x806567cb,0x196c3671,0x6e6b06e7,0xfed41b76,0x89d32be0,
127-
0x10da7a5a,0x67dd4acc,0xf9b9df6f,0x8ebeeff9,0x17b7be43,0x60b08ed5,0xd6d6a3e8,
128-
0xa1d1937e,0x38d8c2c4,0x4fdff252,0xd1bb67f1,0xa6bc5767,0x3fb506dd,0x48b2364b,
129-
0xd80d2bda,0xaf0a1b4c,0x36034af6,0x41047a60,0xdf60efc3,0xa867df55,0x316e8eef,
130-
0x4669be79,0xcb61b38c,0xbc66831a,0x256fd2a0,0x5268e236,0xcc0c7795,0xbb0b4703,
131-
0x220216b9,0x5505262f,0xc5ba3bbe,0xb2bd0b28,0x2bb45a92,0x5cb36a04,0xc2d7ffa7,
132-
0xb5d0cf31,0x2cd99e8b,0x5bdeae1d,0x9b64c2b0,0xec63f226,0x756aa39c,0x026d930a,
133-
0x9c0906a9,0xeb0e363f,0x72076785,0x05005713,0x95bf4a82,0xe2b87a14,0x7bb12bae,
134-
0x0cb61b38,0x92d28e9b,0xe5d5be0d,0x7cdcefb7,0x0bdbdf21,0x86d3d2d4,0xf1d4e242,
135-
0x68ddb3f8,0x1fda836e,0x81be16cd,0xf6b9265b,0x6fb077e1,0x18b74777,0x88085ae6,
136-
0xff0f6a70,0x66063bca,0x11010b5c,0x8f659eff,0xf862ae69,0x616bffd3,0x166ccf45,
137-
0xa00ae278,0xd70dd2ee,0x4e048354,0x3903b3c2,0xa7672661,0xd06016f7,0x4969474d,
138-
0x3e6e77db,0xaed16a4a,0xd9d65adc,0x40df0b66,0x37d83bf0,0xa9bcae53,0xdebb9ec5,
139-
0x47b2cf7f,0x30b5ffe9,0xbdbdf21c,0xcabac28a,0x53b39330,0x24b4a3a6,0xbad03605,
140-
0xcdd70693,0x54de5729,0x23d967bf,0xb3667a2e,0xc4614ab8,0x5d681b02,0x2a6f2b94,
141-
0xb40bbe37,0xc30c8ea1,0x5a05df1b,0x2d02ef8d
142-
];
103+
private uint[256][8] genTables(uint polynomial)
104+
{
105+
uint[256][8] res = void;
106+
107+
foreach (i; 0 .. 0x100)
108+
{
109+
uint crc = i;
110+
foreach (_; 0 .. 8)
111+
crc = (crc >> 1) ^ (-int(crc & 1) & polynomial);
112+
res[0][i] = crc;
113+
}
114+
115+
foreach (i; 0 .. 0x100)
116+
{
117+
res[1][i] = (res[0][i] >> 8) ^ res[0][res[0][i] & 0xFF];
118+
res[2][i] = (res[1][i] >> 8) ^ res[0][res[1][i] & 0xFF];
119+
res[3][i] = (res[2][i] >> 8) ^ res[0][res[2][i] & 0xFF];
120+
121+
res[4][i] = (res[3][i] >> 8) ^ res[0][res[3][i] & 0xFF];
122+
res[5][i] = (res[4][i] >> 8) ^ res[0][res[4][i] & 0xFF];
123+
res[6][i] = (res[5][i] >> 8) ^ res[0][res[5][i] & 0xFF];
124+
res[7][i] = (res[6][i] >> 8) ^ res[0][res[6][i] & 0xFF];
125+
}
126+
return res;
127+
}
128+
129+
private static immutable uint[256][8] crc32Tables = genTables(0xEDB88320);
130+
131+
unittest
132+
{
133+
auto tables = genTables(0xEDB88320);
134+
assert(tables[0][0] == 0x00000000 && tables[0][$ - 1] == 0x2d02ef8d && tables[7][$ - 1] == 0x264b06e6);
135+
}
143136

144137
/**
145138
* Template API CRC32 implementation.
@@ -159,8 +152,48 @@ struct CRC32
159152
*/
160153
void put(scope const(ubyte)[] data...) @trusted pure nothrow @nogc
161154
{
162-
foreach (val; data)
163-
_state = (_state >> 8) ^ crc32_table[cast(ubyte)_state ^ val];
155+
uint crc = _state;
156+
// process eight bytes at once
157+
while (data.length >= 8)
158+
{
159+
// Use byte-wise reads to support architectures without HW support
160+
// for unaligned reads. This can be optimized by compilers to a single
161+
// 32-bit read if unaligned reads are supported.
162+
// DMD is not able to do this optimization though, so explicitly
163+
// do unaligned reads for DMD's architectures.
164+
version (X86)
165+
enum hasLittleEndianUnalignedReads = true;
166+
else version (X86_64)
167+
enum hasLittleEndianUnalignedReads = true;
168+
else
169+
enum hasLittleEndianUnalignedReads = false; // leave decision to optimizer
170+
static if (hasLittleEndianUnalignedReads)
171+
{
172+
uint one = (cast(uint*) data.ptr)[0] ^ crc;
173+
uint two = (cast(uint*) data.ptr)[1];
174+
}
175+
else
176+
{
177+
uint one = (data.ptr[3] << 24 | data.ptr[2] << 16 | data.ptr[1] << 8 | data.ptr[0]) ^ crc;
178+
uint two = (data.ptr[7] << 24 | data.ptr[6] << 16 | data.ptr[5] << 8 | data.ptr[4]);
179+
}
180+
181+
crc =
182+
crc32Tables[0][two >> 24] ^
183+
crc32Tables[1][(two >> 16) & 0xFF] ^
184+
crc32Tables[2][(two >> 8) & 0xFF] ^
185+
crc32Tables[3][two & 0xFF] ^
186+
crc32Tables[4][one >> 24] ^
187+
crc32Tables[5][(one >> 16) & 0xFF] ^
188+
crc32Tables[6][(one >> 8) & 0xFF] ^
189+
crc32Tables[7][one & 0xFF];
190+
191+
data = data[8 .. $];
192+
}
193+
// remaining 1 to 7 bytes
194+
foreach (d; data)
195+
crc = (crc >> 8) ^ crc32Tables[0][(crc & 0xFF) ^ d];
196+
_state = crc;
164197
}
165198
///
166199
unittest

0 commit comments

Comments
 (0)