nona sometime crashes with gpu, for single image

[jpe]

Nona, when using gpu (ubuntu, [GeForce MX150) sometimes crashes.
The error is typically

Feb 15 19:02:42 apollo19 kernel: [302516.363894] nona[32290]: segfault at 56100dbacda8 ip 00007fd2e21ba0dd sp 00007ffdfe26ea00 error 6 in libc-2.27.so[7fd2e2125000+1e7000]

It's a malloc problem.

If people are interested, I can do more tests

The error seems to depend on how one cuts a image, it seems to me.

[tmodes (tmodes)]

"sometimes crashes" is very vague. So further test would be fine.
Ideally a testcase which always crashes would be optimal. Otherwise the debug is very difficult.
(I'm using nona-gpu for a long time and had no crash until now.)

[jpe (jean-pierre-eckmann)]

here is the output from a broken run. After the last line it says segmentation fault.
(NB : The "old official" ubuntu version DOES work,

second attachment follows

if you need more debug, let me know how to produce that...

ddd?

[jpe (jean-pierre-eckmann)]

the output

[jpe (jean-pierre-eckmann)]

here the output from ddd

[tmodes (tmodes)]

I can't reproduce the issue here. It runs fine here: Windows + ATI Mobility Radeon. Tested also (L)Ubuntu in a virtual machine. It took some time but finished without crash.
And backtrace does not give me a hint. The mentioned lines are some very general one, and the interesting ones (#7 and #8) are truncated.

[jpe (jean-pierre-eckmann)]

ok, more info

let me know what you need. There is something bizarre with the allocation.

(are you sure that the virtual linux really uses the gpu? without gpu, it DOES work)

from the backtrace
#8 0x00007ffff79aabbf in HuginBase::Nona::remapImage<vigra::BasicImage<vigra::RGBValue<unsigned char, 0u, 1u, 2u>, std::allocator<vigra::RGBValue<unsigned char, 0u, 1u, 2u> > >, vigra::BasicImage<float, std::allocator<float> >, vigra::BasicImage<vigra::RGBValue<unsigned char, 0u, 1u, 2u>, std::allocator<vigra::RGBValue<unsigned char, 0u, 1u, 2u> > >, vigra::BasicImage<unsigned char, std::allocator<unsigned char> > > (srcImg=..., srcAlpha=..., srcFlat=..., src=..., dest=..., outputROI=..., remapped=warning: RTTI symbol not found for class 'HuginBase::Nona::RemappedPanoImage<vigra::BasicImage<vigra::RGBValue<unsigned char, 0u, 1u, 2u>, std::allocator<vigra::RGBValue<unsigned char, 0u, 1u, 2u> > >, vigra::BasicImage<unsigned char, std::allocator<unsigned char> > >'
..., progress=0x55555577cc00) at /home/eckmann/HUGIN/hugin-hugin/src/hugin_base/nona/RemappedPanoImage.h:855

#7 0x00007ffff7810c0a in HuginBase::Nona::RemappedPanoImage<vigra::BasicImage<vigra::RGBValue<unsigned char, 0u, 1u, 2u>, std::allocator<vigra::RGBValue<unsigned char, 0u, 1u, 2u> > >, vigra::BasicImage<unsigned char, std::allocator<unsigned char> > >::remapImage<vigra::ConstBasicImageIterator<vigra::RGBValue<unsigned char, 0u, 1u, 2u>, vigra::RGBValue<unsigned char, 0u, 1u, 2u>**>, vigra::RGBAccessor<vigra::RGBValue<unsigned char, 0u, 1u, 2u> > > (this=0x5555559a4ad0, srcImg=..., interpol=<optimized out>, progress=0x55555577cc00, singleThreaded=<optimized out>) at /home/eckmann/HUGIN/hugin-hugin/src/hugin_base/nona/RemappedPanoImage.h:618

#6 vigra::BasicImage<vigra::RGBValue<unsigned char, 0u, 1u, 2u>, std::allocator<vigra::RGBValue<unsigned char, 0u, 1u, 2u> > >::operator= (this=0x5555559a4ad8, rhs=...) at /usr/local/include/vigra/basicimage.hxx:1102

#5 0x00007ffff77dfd0f in vigra::BasicImage<vigra::RGBValue<unsigned char, 0u, 1u, 2u>, std::allocator<vigra::RGBValue<unsigned char, 0u, 1u, 2u> > >::resizeCopy (rhs=..., this=0x5555559a4ad8) at /usr/local/include/vigra/basicimage.hxx:831

#4 vigra::BasicImage<vigra::RGBValue<unsigned char, 0u, 1u, 2u>, std::allocator<vigra::RGBValue<unsigned char, 0u, 1u, 2u> > >::resizeCopy (this=this@entry=0x5555559a4ad8, width=2082, height=2135, data=0x555555a34f00) at /usr/local/include/vigra/basicimage.hxx:1204

#3 0x00007ffff77dfb84 in __gnu_cxx::new_allocator<vigra::RGBValue<unsigned char, 0u, 1u, 2u> >::allocate (this=0x5555559a4af8, __n=4445070) at /usr/include/c++/7/ext/new_allocator.h:111

[tmodes (tmodes)]

Could you please try the attached patch if it fixes the crash?
(It just a guess.)

[jpe (jean-pierre-eckmann)]

still not ok. if you wish, we can communicate directly. I would give you my email

[tmodes (tmodes)]

Okay, found the culprit. Committed the changes to repository.