On Tue, Jul 15, 2014 at 9:22 PM, Lisandro Damián Nicanor Pérez Meyer < perezme...@gmail.com> wrote:
> reassign 754894 src:qtbase-opensource-src 5.2.0+dfsg-1 > severity 754894 important > thanks > > On Tuesday 15 July 2014 19:48:06 Boris Barbulovski wrote: > > Package: libqt5core5a (actually all qt5 libraries) > > Version: 5.3.1+dfsg-3(jessie) > > > > >From Qt 5.3 SSE2 is enabled by default crashing every Qt app when run > > > > on i386 version with older(non SSE2) processor. > > > > > > Solution: Disable SSE2 when building Qt5 packages for i386. > > Not quite. > Here is the location when it crashes ( https://qt.gitorious.org/qt/qtbase/source/e0dc27d791cd2106f661bd3d949281aaf1239990:src/corelib/tools/qstring.cpp#L658 ignore the fact that it's qt git url.) The crash location is first line after line '#ifdef __SSE2__' here is the dgb output (gdb) disas 0xb7984fb0, 0xb7984fff Dump of assembler code from 0xb7984fb0 to 0xb7984fff: 0xb7984fb0: mov %edx,%esi 0xb7984fb2: shl $0x10,%esi 0xb7984fb5: or %edx,%esi 0xb7984fb7: lea 0x10(%eax),%edx 0xb7984fba: mov %esi,(%esp) 0xb7984fbd: movd (%esp),%xmm2 0xb7984fc2: cmp %edx,%ebp => 0xb7984fc4: pshufd $0x0,%xmm2,%xmm1 0xb7984fc9: jae 0xb7984fd9 0xb7984fcb: jmp 0xb7984f6e 0xb7984fcd: lea 0x0(%esi),%esi 0xb7984fd0: lea 0x10(%edx),%eax 0xb7984fd3: cmp %eax,%ebp 0xb7984fd5: jb 0xb7984f70 0xb7984fd7: mov %eax,%edx 0xb7984fd9: movdqu -0x10(%edx),%xmm0 0xb7984fde: lea -0x10(%edx),%esi 0xb7984fe1: pcmpeqw %xmm1,%xmm0 0xb7984fe5: pmovmskb %xmm0,%eax 0xb7984fe9: test %ax,%ax 0xb7984fec: je 0xb7984fd0 0xb7984fee: sub 0xc(%esp),%esi > > > I will paste paragraph from Phoronix article regarding this issue: > > > > *- Qt 5.3 by default for 32-built builds now depends upon SSE2 CPU > support. > > SSE2 CPUs have been around for the past decade and building code > > optimized for SSE2+ will yield better performance. SSE2 has been found > > since Intel Pentium 4 CPUs and for AMD hardware since the Opteron / > > Athlon 64 debut. Those building Qt 5.3 i386 binaries wishing to support > > ancient CPUs can pass -no-sse2 to still maintain pre-SSE2 compatibility.* > > > > here is the article > > <http://www.phoronix.com/scan.php?page=news_item&px=MTY3MDU>. > > Please use good data sources: upstream. > > There is a better fix pending someone to step up for doing the job. It's > not > that simple as "just disable SSE2" because that would render some parts > simply > unusable. > > I don't understand what you mean by 'render some parts simply unusable'. Do you mean some libraries will not compile? AFAIK SSE2 usage is for performance reasons only(maybe I'm wrong). I found this mail from Thiago Macieira( http://lists.qt-project.org/pipermail/development/2013-November/014085.html) where he explains the situation in more detail. > The correct fix is to build *some* of the libs twice: one with SSE2 support > and one without it. If properly installed the linker will do the correct > job > at runtime. > > Thiago(in the mail thread above) suggest something like two versions of the library. > Interestingly enough, so far no one has reproduced this bug so far. Can you > reproduce it? It would be good to know so you can tests the packages for > us. > > yes, I can reproduce the issue, and test the packages. > Kinds regards, Lisandro. > > -- > > Lisandro Damián Nicanor Pérez Meyer > http://perezmeyer.com.ar/ > http://perezmeyer.blogspot.com/ br, Boris..
