kadircet added a comment.
Moreover, the most time consuming(well this might be subject to ltrace not
counting IO wait times, as an openat above was only 5microsecs/call and a
tolower call is 56 microsecs/call :D) bit actually seems like the string
comparisons/manipulations done while parsing the command line args:
% time seconds usecs/call calls function
------ ----------- ----------- --------- --------------------
48.73 2.244680 56 39747 tolower
21.88 1.008019 54 18522 strlen
7.18 0.330929 54 6055
_ZNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEE9_M_appendEPKcm
5.45 0.251138 54 4605 _ZdlPv
4.69 0.216144 54 3961 bcmp
4.61 0.212397 54 3890 memcpy
1.44 0.066164 54 1204 _Znwm
So caching the result of command line parsing might actually be better than
caching IO. This also seems quite easy, as that part of the code is already
isolated into a `clang::driver::Compilation`.
We can cache the result of command line parsing and invalidate it whenever
compile commands changes(this is safe, as it is a pure function, (well maybe
minus env variables, haven't checked in detail)). And later on use this for
creating a compiler invocation.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D81719/new/
https://reviews.llvm.org/D81719
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
