several optimizations

[firewave]

No description provided.

[danmar]

do you have some real life measurements?

[firewave]

Running with -j1 to see the actual impact - the library I used to profile and some random ones.

openni-sensor-primesense

real    0m4.441s
user    0m3.859s
sys     0m0.516s

real    0m3.844s
user    0m3.297s
sys     0m0.531s

s390-tools

real    0m5.095s
user    0m4.594s
sys     0m0.500s

real    0m4.660s
user    0m3.938s
sys     0m0.703s

gabedit

real    0m41.950s
user    0m39.781s
sys     0m2.172s

real    0m38.831s
user    0m36.672s
sys     0m2.125s

Edit: I just realized I had two local optimization still enabled. Need to re-test without them...also compared to the actual daca@home runs I forgot to adjust the --library parameters so the time will differ to those runs as well.

[firewave]

Okay, now with just the optimization ins this PR. It does not effect all scans.

openni-sensor-primesense 4.7 5.1
gabedit 44.9 44.6
s390-tools 5.5 5.5

Strangely gabedit gives about 2000 warnings less with the changes, so I need to look into this.

[danmar]

as far as I remember we try to make simplecpp threadsafe. So a static local variable should be avoided.

[firewave]

Good point. Would a global one be okay?

[danmar]

yes

[danmar]

I find it strange that this would be an optimisation. Is strpbrk optimised better?

However I think it obfuscates the code. I do not intuitively know how strpbrk works.

[firewave]

callgrind showed and there's this
https://www.codeproject.com/Articles/18473/find-first-of-A-performance-pitfall-among-the-STL#A30. I used the latest GCC so it seems this is still true although that article is 12 years old.

strpbrk is the same as find_first_of.

[firewave]

Running simplecpp on -q -Ilib lib/valueflow.cpp:
GCC 335,129,681 -> 325,111,787
Clang 323,433,151 -> 313,133,608

I filed a ticket about this https://trac.cppcheck.net/ticket/10777.

[firewave]

I tried some other measurements and those were a inconclusive. I will look into this in the next days.

[firewave]

Running simplecpp with -q -Ilib -D __GNUC__ lib/valueflow.cpp:
GCC 335,129,681 -> 318,581,229
Clang 323,433,151 -> 312,465,341

[firewave]

Using lib/valueflow.cpp from cppcheck-opensource/cppcheck@0a99e3b with -q -Ilib -D __GNUC__ lib/valueflow.cpp.

baseline
GCC 11 337,505,121
Clang 13 325,815,035

Location::adjust()
GCC 11 327,385,765
Clang 13 315,411,381

endsWith()
GCC 11 320,802,714
Clang 13 314,737,794

lastLine()
GCC 11 320,639,877
Clang 13 314,696,243

unordered_map
GCC 11 306,006,031
Clang 13 299,186,851

flags memchr
GCC 11 302,776,749
Clang 13 296,127,177

flags copy ctor
GCC 11 297,134,127
Clang 13 291,144,265

FYI about a third of the instructions is made up by the new/delete of the Token objects and the std::istream reading.