diff options
| author | Flaviu Tamas <tamasflaviu@gmail.com> | 2018-10-28 00:52:49 -0400 |
|---|---|---|
| committer | Flaviu Tamas <tamasflaviu@gmail.com> | 2018-10-28 00:53:09 -0400 |
| commit | 9ca1c2c93035341e06d85f56edeb6a08f962a9e8 (patch) | |
| tree | b148956f5631c482ea7425abdb17f933305cf1d8 /lib | |
| parent | f2cd8ed99a7caa00b20bae2443cacf2d97700d75 (diff) | |
| download | Nim-9ca1c2c93035341e06d85f56edeb6a08f962a9e8.tar.gz | |
Eliminate floating point arithmatic in nre
Integer division is already hard enough on your CPU, using floating point here is WAY slower and can just as effectivly be done using integers. This is important because matchImpl tends to be in the center of very hot loops (like split()).
Diffstat (limited to 'lib')
| -rw-r--r-- | lib/impure/nre.nim | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/lib/impure/nre.nim b/lib/impure/nre.nim index ce3fb2aff..5155194b0 100644 --- a/lib/impure/nre.nim +++ b/lib/impure/nre.nim @@ -469,7 +469,8 @@ proc matchImpl(str: string, pattern: Regex, start, endpos: int, flags: int): Opt # 1x capture count as slack space for PCRE let vecsize = (pattern.captureCount() + 1) * 3 # div 2 because each element is 2 cints long - myResult.pcreMatchBounds = newSeq[HSlice[cint, cint]](ceil(vecsize / 2).int) + # plus 1 because we need the ceiling, not the floor + myResult.pcreMatchBounds = newSeq[HSlice[cint, cint]]((vecsize div 2) + 1) myResult.pcreMatchBounds.setLen(vecsize div 3) let strlen = if endpos == int.high: str.len else: endpos+1 |