John, Yes, there is indeed a memory corruption problem but extremelly hard to track down. GCC issues an ICE, through a tree_check macro, which then tried to print a string and fails due to a garbage collection read of 1 byte uninitialized in a strlen attempt. This is two bugs in one. First it shouldn't ICE and then once it ICEs it shouldn't segfault.
Paulo Matos > -----Original Message----- > From: John Regehr [mailto:reg...@cs.utah.edu] > Sent: 03 October 2013 22:24 > To: Paulo Matos; creduce-dev@flux.utah.edu > Subject: Re: [creduce-dev] Increase of size during reduction process > > Paulo, it seems to me that if none of the line-based passes can remove > anything from a 750 KB test case, there is something weird going on. > > My rough guess is that there is a memory corruption problem in the compiler > and the ICE is a secondary effect of a very coincidental set of > circumstances. It is possible that C-Reduce simply cannot help you very > much here. > > So how about this: rather than doing this reduction, just run the compiler > using valgrind (using the --follow-children=yes option of course) and see > if there any any smoking guns. > > John > > > > On 10/03/2013 07:35 AM, Paulo Matos wrote: > > Hi, > > > > Just to let you guys know of the following: After getting c-reduce git > HEAD, and running it on the same example I see pass_lines 0 reducing one byte > at a time starting from ~751000 bytes. Since this would take the rest of my > life to reduce I decided to remove pass_lines 0 from creduce.in and make > install as per John's suggestion. Now I see: > > > > pmatos@poseidon:~/tmp/builds/creduce-build/creduce$ ./creduce > > ./min.sh PsdInfo_6cpe.i > > ===< 10766 >=== > > running 4 interestingness test(s) in parallel > > ===< pass_blank :: 0 >=== > > ===< pass_clang_binsrch :: replace-function-def-with-decl >=== > > ===< pass_clang_binsrch :: remove-unused-function >=== > > (0.0 %, 751215 bytes) > > (0.1 %, 750836 bytes) > > ===< pass_lines :: 1 >=== > > (-1.7 %, 764127 bytes) > > ===< pass_lines :: 1 >=== > > (-3.5 %, 777418 bytes) > > ===< pass_lines :: 1 >=== > > (-5.2 %, 790709 bytes) > > ===< pass_lines :: 2 >=== > > (-7.0 %, 804073 bytes) > > ===< pass_lines :: 2 >=== > > (-8.8 %, 817437 bytes) > > ===< pass_lines :: 2 >=== > > (-10.6 %, 830801 bytes) > > > > Will try to remove all pass lines and see what happens. > > > > Paulo Matos > > > > > >> -----Original Message----- > >> From: creduce-dev-boun...@flux.utah.edu [mailto:creduce-dev- > >> boun...@flux.utah.edu] On Behalf Of John Regehr > >> Sent: 02 October 2013 15:09 > >> To: creduce-dev@flux.utah.edu > >> Subject: Re: [creduce-dev] Increase of size during reduction process > >> > >> Hi Paulo, > >> > >> This is odd behavior, I haven't seen it happen where a temporary size > >> increase keeps going. > >> > >> You can easily disable individual passes that aren't being helpful: just > >> modify creduce.in and then "make install" again. > >> > >> C-Reduce is deterministic so it doesn't take any kind of seed. > >> > >> John > >> > >> > >> On 10/02/2013 08:02 AM, Paulo Matos wrote: > >>> > >>>> -----Original Message----- > >>>> From: Eitan Adler [mailto:li...@eitanadler.com] > >>>> Sent: 02 October 2013 14:55 > >>>> To: Paulo Matos > >>>> Cc: creduce-dev@flux.utah.edu > >>>> Subject: Re: [creduce-dev] Increase of size during reduction process > >>>> > >>>> On Wed, Oct 2, 2013 at 5:41 AM, Paulo Matos <pma...@broadcom.com> wrote: > >>>>> Hi, > >>>> > >>>>> Shouldn't size increase be dismissed? > >>>> > >>>> It is possible to get into local minimums which are not global minimums. > >>>> > >>>> > >>> > >>> This testcase I am trying is driving me nuts. > >>> It's currently at > >>> (-3.3 %, 473260 bytes) > >>> > >>> Each time a pass_blank starts it increases the size more than it > decreases > >> during its process, therefore even though the testcase started with > >>> ===< pass_blank :: 0 >=== > >>> (0.0 %, 458340 bytes) > >>> it is now way larger. > >>> > >>> Is there a way to set a seed to creduce, so that I can try to avoid those > >> local minimums? > >>> Maybe if I can set a seed, I can set up multiple parallel creduce and > grab > >> the one that generates the smallest example. > >>> > >>> Paulo Matos > >>> > >
