This is the mail archive of the
java@gcc.gnu.org
mailing list for the Java project.
Re: compacting _Jv_Utf8Const
Tom Tromey wrote:
fleche. nm .libs/libgcj.so|grep _Utf|wc -l
66488
Per> I assume this is without duplicate removal. The _UtfNN symbols are
Per> local symbols; I'm not sure how to count unique symbols.
I thought we did duplicate removal in the linker.
Perhaps - but it doesn't follow the symbols are eliminated.
In fact if the local symbols are visible in the output, it would
be *wrong* to remove duplicates, even if they resolve to the same
location.
$ nm libgcj.so|grep ' _Utf'|awk '{print $1}'>/tmp/bar
$ wc /tmp/bar
60208 60208 541872 /tmp/bar
$ uniq </tmp/bar|wc
54395 54395 489555
It seems duplicate elimination removes about 10%.
Not a huge win. Other metadata compaction strategies
might win, even if they're incompatible with duplicate
elimination, especially if they reduce runtime relocation.
This is interesting. More later.
--
--Per Bothner
per@bothner.com http://per.bothner.com/