On 12/30/2014 07:10 PM, Dimitri van Heesch wrote:
> That doesn't sound right.
> Would be nice to understand what doxygen is doing but I haven't been able to
> reproduce this.
>
> What options are enabled?
> Can you do some profiling? Can you run with the '-d time' option for instance?
> Does this also happen if you use a smaller subset of the code?
Hi Dimitri,
the CPU time seems to grow dramatically with increasing project size.
The case I described has all designs I have included. To trace the
issue I've setup a test environment with the sources of a single
design, now 117 files and 28428 lines of code. Now run times for
"doxygen -d time" are
doxygen-1.8.7 real 0m07.963s user 0m08.510s sys 0m02.143s
Spent 0.849 seconds in Generating file sources...
Spent 1.106 seconds in Generating class documentation...
Spent 3.919 seconds in Running dot...
doxygen-1.8.9 real 8m32.386s user 8m32.438s sys 0m02.231s
Spent 329.133 seconds in Generating file sources...
Spent 177.439 seconds in Generating class documentation...
Spent 4.096 seconds in Running dot...
I've also done some profiling with "valgrind --tool=callgrind".
A quick looks seems to indicate that a lot of time is burned in
QGList::contains(), suggesting that a list is searched. The
pertinent part of the callgrind_annotate output for 1.8.9 is
150,241,637,969 < VhdlDocGen::findMember(...)
150,241,637,969 * VhdlDocGen::findMemberDef(...)
794,272 > QMapPrivate<QCString, MemberDef*>::insert(...)
21,869 > QGList::append(...)
1,541,745,005 > QCString::StringRep::StringRep(...)
1,824,725,951 > Definition::symbolName(...)
51,057 > ClassDef::getMemberList(...)
19,746 > QGListIterator::operator++(...)
22,775 > MemberListIterator::MemberListIterator(...)
102,891,546,408 > QGList::contains(...)
14,700,790,148 > ...:__strcmp_sse2_unaligned (...)
4,437,837,537 > QCString::operator+=(...)
812 > QCString::StringRep::data(...)
21,024 > QGListIterator::~QGListIterator(...)
1,686,722,110 > QCString::StringRep::~StringRep(...)
500,621 > QCString::StringRep::StringRep(...)
2,160 > QMap<QCString, MemberDef*>::detach(...)
8,384,846,588 > operator+(...)
1,215 > QGListIterator::toFirst(...)
Note that I aborted valgrind after about an hour, running doxygen under
valgrind is a factor 20 slower that running it directly.
The same section of a callgrind_annotate output for 1.8.7, done after
a full document creation run, looks like
14,153,463 < VhdlDocGen::writeVHDLTypeDocumentation(...)
1,722,043,706 < generateMemLink(...)
654,130,388 < VhdlDocGen::writeStringLink(...)
2,390,327,557 * VhdlDocGen::findMember(...)
510,960 > QGDictIterator::toFirst(...)
8,278,421 > QCString::QCString(...)
133,736,676 > VhdlDocGen::getClass(...)
1,955,829,369 > VhdlDocGen::findAllPackages(...)
2,402,164 > QCString::upper(...)
3,670,712 > ClassDef::className(...)
298,851 > ClassDef::protection(...)
15,384,427 > QGDict::QGDict(...)
11,590,565 > QGDictIterator::QGDictIterator(...)
13,977,784 > QGDict::~QGDict(...)
736,485 > QGDict::clear(...)
3,625,505 > QCString::assign(...)
154,335 > Definition::getOuterScope(...)
218,592,559 > VhdlDocGen::findMemberDef(...)
5,995,264 > QGDictIterator::~QGDictIterator(...)
8,458,217 > QCString::~QCString(...)
2,396,524 > QCString::lower(...)
I hope that helps a bit.
With best regards, Walter
------------------------------------------------------------------------------
Dive into the World of Parallel Programming! The Go Parallel Website,
sponsored by Intel and developed in partnership with Slashdot Media, is your
hub for all things parallel software development, from weekly thought
leadership blogs to news, videos, case studies, tutorials and more. Take a
look and join the conversation now. http://goparallel.sourceforge.net
_______________________________________________
Doxygen-users mailing list
Doxygen-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/doxygen-users
