Fossil

Check-in [7338b3a9]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Use intrinsic byte-swapping functions to boost the performance of delta checksums.
Downloads: Tarball | ZIP archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA1: 7338b3a9510281e7d49643f55a5c27c24107f083
User & Date: drh 2015-12-31 15:45:10.207
Context
2015-12-31
15:46
Add MIME types for C# and VB.NET. ... (check-in: 5e81f4c0 user: drh tags: trunk)
15:45
Use intrinsic byte-swapping functions to boost the performance of delta checksums. ... (check-in: 7338b3a9 user: drh tags: trunk)
15:44
Update the built-in SQLite to 3.10.0 beta for SQLite testing. ... (check-in: ec8394e1 user: drh tags: trunk)
15:18
Use intrinsic functions to boost the performance of the delta checksum. ... (Closed-Leaf check-in: 726ab361 user: drh tags: checksum-intrinsics)
Changes
Unified Diff Ignore Whitespace Patch
Changes to src/delta.c.
207
208
209
210
211
212
213






214
215



216
217

218

219




















220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239


240
241
242
243
244
245
246
247
248
249
250
251
252
253
*/
static int digit_count(int v){
  unsigned int i, x;
  for(i=1, x=64; v>=x; i++, x <<= 6){}
  return i;
}







/*
** Compute a 32-bit checksum on the N-byte buffer.  Return the result.



*/
static unsigned int checksum(const char *zIn, size_t N){

  const unsigned char *z = (const unsigned char *)zIn;

  unsigned sum0 = 0;




















  unsigned sum1 = 0;
  unsigned sum2 = 0;
  unsigned sum3 = 0;
  while(N >= 16){
    sum0 += ((unsigned)z[0] + z[4] + z[8] + z[12]);
    sum1 += ((unsigned)z[1] + z[5] + z[9] + z[13]);
    sum2 += ((unsigned)z[2] + z[6] + z[10]+ z[14]);
    sum3 += ((unsigned)z[3] + z[7] + z[11]+ z[15]);
    z += 16;
    N -= 16;
  }
  while(N >= 4){
    sum0 += z[0];
    sum1 += z[1];
    sum2 += z[2];
    sum3 += z[3];
    z += 4;
    N -= 4;
  }
  sum3 += (sum2 << 8) + (sum1 << 16) + (sum0 << 24);


  switch(N){
    case 3:   sum3 += (z[2] << 8);
    case 2:   sum3 += (z[1] << 16);
    case 1:   sum3 += (z[0] << 24);
    default:  ;
  }
  return sum3;
}

/*
** Create a new delta.
**
** The delta is written into a preallocated buffer, zDelta, which
** should be at least 60 bytes longer than the target file, zOut.







>
>
>
>
>
>

|
>
>
>


>

>
|
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
>
>
|
|
|
|


|







207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
*/
static int digit_count(int v){
  unsigned int i, x;
  for(i=1, x=64; v>=x; i++, x <<= 6){}
  return i;
}

#ifdef __GNUC__
# define GCC_VERSION (__GNUC__*1000000+__GNUC_MINOR__*1000+__GNUC_PATCHLEVEL__)
#else
# define GCC_VERSION 0
#endif

/*
** Compute a 32-bit big-endian checksum on the N-byte buffer.  If the
** buffer is not a multiple of 4 bytes length, compute the sum that would
** have occurred if the buffer was padded with zeros to the next multiple
** of four bytes.
*/
static unsigned int checksum(const char *zIn, size_t N){
  static const int byteOrderTest = 1;
  const unsigned char *z = (const unsigned char *)zIn;
  const unsigned char *zEnd = (const unsigned char*)&zIn[N&~3];
  unsigned sum = 0;
  assert( (3&(sqlite3_uint64)z)==0 );  /* Four-byte alignment */
  if( 0==*(char*)&byteOrderTest ){
    /* This is a big-endian machine */
    while( z<zEnd ){
      sum += *(unsigned*)z;
      z += 4;
    }
  }else{
    /* A little-endian machine */
#if GCC_VERSION>=4003000
    while( z<zEnd ){
      sum += __builtin_bswap32(*(unsigned*)z);
      z += 4;
    }
#elif defined(_MSC_VER) && _MSC_VER>=1300
    while( z<zEnd ){
      sum += _byteswap_ulong(*(unsigned*)z);
      z += 4;
    }
#else    
    unsigned sum0 = 0;
    unsigned sum1 = 0;
    unsigned sum2 = 0;
    while(N >= 16){
      sum0 += ((unsigned)z[0] + z[4] + z[8] + z[12]);
      sum1 += ((unsigned)z[1] + z[5] + z[9] + z[13]);
      sum2 += ((unsigned)z[2] + z[6] + z[10]+ z[14]);
      sum  += ((unsigned)z[3] + z[7] + z[11]+ z[15]);
      z += 16;
      N -= 16;
    }
    while(N >= 4){
      sum0 += z[0];
      sum1 += z[1];
      sum2 += z[2];
      sum  += z[3];
      z += 4;
      N -= 4;
    }
    sum += (sum2 << 8) + (sum1 << 16) + (sum0 << 24);
#endif
  }
  switch(N&3){
    case 3:   sum += (z[2] << 8);
    case 2:   sum += (z[1] << 16);
    case 1:   sum += (z[0] << 24);
    default:  ;
  }
  return sum;
}

/*
** Create a new delta.
**
** The delta is written into a preallocated buffer, zDelta, which
** should be at least 60 bytes longer than the target file, zOut.