| … | |
… | |
| 45 | * it works ;) |
45 | * it works ;) |
| 46 | */ |
46 | */ |
| 47 | #ifndef FRST |
47 | #ifndef FRST |
| 48 | # define FRST(p) (((p[0]) << 8) | p[1]) |
48 | # define FRST(p) (((p[0]) << 8) | p[1]) |
| 49 | # define NEXT(v,p) (((v) << 8) | p[2]) |
49 | # define NEXT(v,p) (((v) << 8) | p[2]) |
|
|
50 | # if MULTIPLICATION_IS_SLOW |
| 50 | # if ULTRA_FAST |
51 | # if ULTRA_FAST |
| 51 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h ) & (HSIZE - 1)) |
52 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h ) & (HSIZE - 1)) |
| 52 | # elif VERY_FAST |
53 | # elif VERY_FAST |
| 53 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
54 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
|
|
55 | # else |
|
|
56 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
|
|
57 | # endif |
| 54 | # else |
58 | # else |
|
|
59 | /* this one was developed with sesse, |
|
|
60 | * and is very similar to the one in snappy. |
|
|
61 | * it does need a modern enough cpu with a fast multiplication. |
|
|
62 | */ |
| 55 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
63 | # define IDX(h) (((h * 0x1e35a7bdU) >> (32 - HLOG - 8)) & (HSIZE - 1)) |
| 56 | # endif |
64 | # endif |
| 57 | #endif |
65 | #endif |
| 58 | /* |
|
|
| 59 | * IDX works because it is very similar to a multiplicative hash, e.g. |
|
|
| 60 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
|
|
| 61 | * the latter is also quite fast on newer CPUs, and compresses similarly. |
|
|
| 62 | * |
|
|
| 63 | * the next one is also quite good, albeit slow ;) |
|
|
| 64 | * (int)(cos(h & 0xffffff) * 1e6) |
|
|
| 65 | */ |
|
|
| 66 | |
66 | |
| 67 | #if 0 |
67 | #if 0 |
| 68 | /* original lzv-like hash function, much worse and thus slower */ |
68 | /* original lzv-like hash function, much worse and thus slower */ |
| 69 | # define FRST(p) (p[0] << 5) ^ p[1] |
69 | # define FRST(p) (p[0] << 5) ^ p[1] |
| 70 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
70 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
| … | |
… | |
| 148 | if (1 |
148 | if (1 |
| 149 | #if INIT_HTAB |
149 | #if INIT_HTAB |
| 150 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
150 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
| 151 | #endif |
151 | #endif |
| 152 | && (off = ip - ref - 1) < MAX_OFF |
152 | && (off = ip - ref - 1) < MAX_OFF |
| 153 | && ip + 4 < in_end |
|
|
| 154 | && ref > (u8 *)in_data |
153 | && ref > (u8 *)in_data |
| 155 | && ref[2] == ip[2] |
154 | && ref[2] == ip[2] |
| 156 | #if STRICT_ALIGN |
155 | #if STRICT_ALIGN |
| 157 | && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) |
156 | && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) |
| 158 | #else |
157 | #else |
| … | |
… | |
| 216 | *op++ = (off >> 8) + ( 7 << 5); |
215 | *op++ = (off >> 8) + ( 7 << 5); |
| 217 | *op++ = len - 7; |
216 | *op++ = len - 7; |
| 218 | } |
217 | } |
| 219 | |
218 | |
| 220 | *op++ = off; |
219 | *op++ = off; |
|
|
220 | |
| 221 | lit = 0; op++; /* start run */ |
221 | lit = 0; op++; /* start run */ |
| 222 | |
222 | |
| 223 | ip += len + 1; |
223 | ip += len + 1; |
| 224 | |
224 | |
| 225 | if (expect_false (ip >= in_end - 2)) |
225 | if (expect_false (ip >= in_end - 2)) |
