From: Philippe E. <ph...@us...> - 2007-12-17 16:47:13
|
Update of /cvsroot/oprofile/oprofile In directory sc8-pr-cvs3.sourceforge.net:/tmp/cvs-serv25546 Modified Files: Tag: JIT_SUPPORT ChangeLog TODO.jit Log Message: change again the way anon samples are handled, use relative samples in samples files but rather to try to do the fixup in op_bfd, do it only in profile.cpp Index: ChangeLog =================================================================== RCS file: /cvsroot/oprofile/oprofile/ChangeLog,v retrieving revision 1.1749.2.55 retrieving revision 1.1749.2.56 diff -u -p -d -r1.1749.2.55 -r1.1749.2.56 --- ChangeLog 14 Dec 2007 22:46:29 -0000 1.1749.2.55 +++ ChangeLog 17 Dec 2007 16:47:14 -0000 1.1749.2.56 @@ -1,3 +1,21 @@ +2007-12-17 Philippe Elie <ph...@wa...> + + change again the way anon samples are handled, use relative samples + in samples files but rather to try to do the fixup in op_bfd, do it + only in profile.cpp, this is simpler and more reliable. I needed + again to add a hack in op_bfd.cpp because we need to know if a bfd + object is a anon file by looking the file extension, there is probably + a better way todo that. + * libpp/profile.cpp: + * libutil++/op_bfd.cpp: + * libutil++/op_bfd.h: + * libpp/callgraph_container.cpp: + * libpp/profile_container.cpp: + * libpp/profile_container.h: + * libutil++/bfd_support.cpp: + * libutil++/bfd_support.h: + * pp/opgprof.cpp: + 2007-12-14 Maynard Johnson <may...@us...> * libutil/op_file.c: Index: TODO.jit =================================================================== RCS file: /cvsroot/oprofile/oprofile/Attic/TODO.jit,v retrieving revision 1.1.2.14 retrieving revision 1.1.2.15 diff -u -p -d -r1.1.2.14 -r1.1.2.15 --- TODO.jit 14 Dec 2007 22:55:27 -0000 1.1.2.14 +++ TODO.jit 17 Dec 2007 16:47:14 -0000 1.1.2.15 @@ -81,20 +81,3 @@ Massi added than for mono, code on the s no byte code, code is always compiled to native machine code, this mean than for mono at least we can do callgraph if we can fix this samples filename problem. - -------- - -<gisle> Hm, I'm getting a strange result with a very simple program: A sequence of samples show's up in two different code regions! -<gisle> This is using opannotate -a -<gisle> Example:17509 26.1774 : 400018a81b0: mr r30,r5 -<gisle> And then: 17509 26.1774 : 40001e231b0: addi r3,r29,-8 -<gisle> The second one is correct, the first is wrong -<gisle> Interestingly, they both end in the same three digits -<gisle> it's opannotate -a on a .jo file -<gisle> (generated from an anonoymous region) -<gisle> Something similar happens in opreport -la -<gisle> samples cum. samples % cum. % image name app name symbol name -<gisle> 33427 8396533 0.3945 99.0988 13628.jo dynamite GB_0xbefed15d_BB_0xbefed17c_0x917c_0xbefed188_0x9188_/lib/ld-linux.so.2 -<gisle> 33410 8429943 0.3943 99.4931 13628.jo dynamite GB_0x804882f_BB_0x804882f_0x82f_0x804884a_0x84a_0x804882_loop.exe -<gisle> The first line is wrong (well I suspect there should be 33410 less samples there) -<gisle> Any idea how this can happen? Looks like samples are being mapped twice... |