| commit | 69f1b01e204f538283ef575691c35e76b99f25df | |
| author | Lauri Kasanen <cand@gmx.com> | |
| Sat, 9 Mar 2013 10:00:33 +0000 (9 12:00 +0200) | ||
| committer | Steve Bennett <steveb@workware.net.au> | |
| Mon, 11 Mar 2013 23:00:19 +0000 (12 09:00 +1000) | ||
| tree | c62213a45ca4b3647cae10d21086716d84586ed1 | treesnapshot (tar.gz zip) |
| parent | 9d4f23daa022c9ca012fb65606960b64b22febe5 | commitdiff |
Precompute operator length for JimParseExprOperator, optimize logic
No regressions, no changes in bench.tcl.
JimParseExprOperator was shockingly inefficient, to the point of taking 63% of cpu
in my own app.
It did a linear search over >60 structs, re-computing the string length each time, and
doing a string comparison for each.
This patch pre-computes the operator string length, speeding the function up by two.
Adding the first-letter fast path and evaluating length before strncmp make for
another 2x speedup, making the total speedup 4x.
Signed-off-by: Lauri Kasanen <cand@gmx.com>
No regressions, no changes in bench.tcl.
JimParseExprOperator was shockingly inefficient, to the point of taking 63% of cpu
in my own app.
It did a linear search over >60 structs, re-computing the string length each time, and
doing a string comparison for each.
This patch pre-computes the operator string length, speeding the function up by two.
Adding the first-letter fast path and evaluating length before strncmp make for
another 2x speedup, making the total speedup 4x.
Signed-off-by: Lauri Kasanen <cand@gmx.com>
| jim.c | diffblobblamehistory |