ckolivas/lrzip
1
0
Fork 0
mirror of https://github.com/ckolivas/lrzip.git synced 2025-12-06 07:12:00 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Make match_len a function completely removing all indirect calls to get_sb, significantly speeding up the single_get_sb case

Browse source
This commit is contained in:
ckolivas 2015-03-08 01:21:53 +11:00
parent cb7b0dd339
commit f378595dce
2 changed files with 42 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
lrzip_private.h
View file

@ -432,10 +432,10 @@ struct rzip_control {
void *log_data; void *log_data;
struct sliding_buffer sb; struct sliding_buffer sb;
uchar *(*get_sb)(rzip_control *control, i64 p);
void (*do_mcpy)(rzip_control *, unsigned char *, i64, i64); void (*do_mcpy)(rzip_control *, unsigned char *, i64, i64);
void (*next_tag)(rzip_control *, struct rzip_state *, i64, tag *); void (*next_tag)(rzip_control *, struct rzip_state *, i64, tag *);
tag (*full_tag)(rzip_control *, struct rzip_state *, i64); tag (*full_tag)(rzip_control *, struct rzip_state *, i64);
i64 (*match_len)(rzip_control *, struct rzip_state *, i64, i64, i64, i64 *);
}; };
struct stream { struct stream {

57
rzip.c
View file

@ -159,11 +159,6 @@ static uchar *sliding_get_sb(rzip_control *control, i64 p)
return (sb->buf_high + (p - sb->offset_high)); return (sb->buf_high + (p - sb->offset_high));
} }
static uchar *single_get_sb(__maybe_unused rzip_control *control, i64 p)
{
return (control->sb.buf_low + p);
}
/* The length of continous range of the sliding buffer, /* The length of continous range of the sliding buffer,
* starting from the offset P. * starting from the offset P.
*/ */
@ -471,19 +466,17 @@ static tag sliding_full_tag(rzip_control *control, struct rzip_state *st, i64 p)
return ret; return ret;
} }
static inline i64 static i64
match_len(rzip_control *control, struct rzip_state *st, i64 p0, i64 op, single_match_len(rzip_control *control, struct rzip_state *st, i64 p0, i64 op,
i64 end, i64 *rev) i64 end, i64 *rev)
{ {
uchar *(*csb)(rzip_control *, i64);
i64 p, len; i64 p, len;
if (op >= p0) if (op >= p0)
return 0; return 0;
p = p0; p = p0;
csb = control->get_sb; while (p < end && control->sb.buf_low[p] == control->sb.buf_low[op]) {
while (p < end && (*csb(control, p) == *csb(control, op))) {
p++; p++;
op++; op++;
} }
@ -493,7 +486,39 @@ match_len(rzip_control *control, struct rzip_state *st, i64 p0, i64 op,
end = MAX(0, st->last_match); end = MAX(0, st->last_match);
while (p > end && op > 0 && *csb(control, op - 1) == *csb(control, p - 1)) { while (p > end && op > 0 && control->sb.buf_low[op - 1] == control->sb.buf_low[p - 1]) {
op--;
p--;
}
len += *rev = p0 - p;
if (len < MINIMUM_MATCH)
return 0;
return len;
}
static i64
sliding_match_len(rzip_control *control, struct rzip_state *st, i64 p0, i64 op,
i64 end, i64 *rev)
{
i64 p, len;
if (op >= p0)
return 0;
p = p0;
while (p < end && *sliding_get_sb(control, p) == *sliding_get_sb(control, op)) {
p++;
op++;
}
len = p - p0;
p = p0;
op -= len;
end = MAX(0, st->last_match);
while (p > end && op > 0 && *sliding_get_sb(control, op - 1) == *sliding_get_sb(control, p - 1)) {
op--; op--;
p--; p--;
} }
@ -525,8 +550,8 @@ find_best_match(rzip_control *control, struct rzip_state *st, tag t, i64 p,
i64 mlen; i64 mlen;
if (t == he->t) { if (t == he->t) {
mlen = match_len(control, st, p, he->offset, end, mlen = control->match_len(control, st, p, he->offset, end,
&rev); &rev);
if (mlen) if (mlen)
st->stats.tag_hits++; st->stats.tag_hits++;
@ -1013,10 +1038,10 @@ bool rzip_fd(rzip_control *control, int fd_in, int fd_out)
gettimeofday(&start, NULL); gettimeofday(&start, NULL);
prepare_streamout_threads(control); prepare_streamout_threads(control);
control->get_sb = single_get_sb;
control->do_mcpy = single_mcpy; control->do_mcpy = single_mcpy;
control->next_tag = &single_next_tag; control->next_tag = &single_next_tag;
control->full_tag = &single_full_tag; control->full_tag = &single_full_tag;
control->match_len = &single_match_len;
while (!pass || len > 0 || (STDIN && !st->stdin_eof)) { while (!pass || len > 0 || (STDIN && !st->stdin_eof)) {
double pct_base, pct_multiple; double pct_base, pct_multiple;
@ -1084,10 +1109,10 @@ retry:
} }
if (st->mmap_size < st->chunk_size) { if (st->mmap_size < st->chunk_size) {
print_maxverbose("Enabling sliding mmap mode and using mmap of %lld bytes with window of %lld bytes\n", st->mmap_size, st->chunk_size); print_maxverbose("Enabling sliding mmap mode and using mmap of %lld bytes with window of %lld bytes\n", st->mmap_size, st->chunk_size);
control->get_sb = &sliding_get_sb;
control->do_mcpy = &sliding_mcpy; control->do_mcpy = &sliding_mcpy;
control->next_tag = &sliding_next_tag; control->next_tag = &sliding_next_tag;
control->full_tag = &sliding_full_tag; control->full_tag = &sliding_full_tag;
control->match_len = &sliding_match_len;
} }
} }
print_maxverbose("Succeeded in testing %lld sized mmap for rzip pre-processing\n", st->mmap_size); print_maxverbose("Succeeded in testing %lld sized mmap for rzip pre-processing\n", st->mmap_size);