TallFurryMan commented 4 years ago

Capturing a single frame in Focus:

Thread 22 "SextractorSolve" received signal SIGFPE, Arithmetic exception.
[Switching to Thread 0xa1d7fb40 (LWP 2531)]
0xb518c6db in InternalSextractorSolver::runSEPSextractor (this=0xae178e20)
    at /var/jenkins_home/workspace/kstars-ci_i386-stellarsolver/stellarsolver/internalsextractorsolver.cpp:252
252 /var/jenkins_home/workspace/kstars-ci_i386-stellarsolver/stellarsolver/internalsextractorsolver.cpp: No such file or directory.
(gdb) bt
#0  0xb518c6db in InternalSextractorSolver::runSEPSextractor() (this=0xae178e20)
    at /var/jenkins_home/workspace/kstars-ci_i386-stellarsolver/stellarsolver/internalsextractorsolver.cpp:252
#1  0xb518e63e in InternalSextractorSolver::extract() (this=0xae178e20)
    at /var/jenkins_home/workspace/kstars-ci_i386-stellarsolver/stellarsolver/internalsextractorsolver.cpp:99
#2  0xb518e63e in InternalSextractorSolver::run() (this=<optimized out>)
    at /var/jenkins_home/workspace/kstars-ci_i386-stellarsolver/stellarsolver/internalsextractorsolver.cpp:141
#3  0xb581fdcc in  () at /usr/lib/i386-linux-gnu/sse2/libQt5Core.so.5
#4  0xb52a43bd in start_thread (arg=0xa1d7fb40) at pthread_create.c:463
#5  0xb483bfe6 in clone () at ../sysdeps/unix/sysv/linux/i386/clone.S:108

This is an Atom i386 cpu with 1GB.

TallFurryMan commented 4 years ago

This is with d1de2f7b.

TallFurryMan commented 4 years ago

https://github.com/rlancaste/stellarsolver/blob/60165c5881499a616ea6e8a4baa88c0f35e747ea/stellarsolver/internalsextractorsolver.cpp#L252

In my case QThread::idealThreadCount() returns 2.

rlancaste commented 4 years ago

Thanks for the info, this is in the code that Jasem was just working on recently.

rlancaste commented 4 years ago

We will check into it.

rlancaste commented 4 years ago

I guess this is a single core computer?

knro commented 4 years ago

Alright, minimum should be 4.

TallFurryMan commented 4 years ago

Alright, minimum should be 4.

Why this restriction? Does this impact performance or is this a hard stop for any single core host?

knro commented 4 years ago

I submitted PR #26 to fix this. This also grantees that only threads multiple of 4 are accepted to avoid any weird partitioning. If this is not met, it just runs it single-threaded like before.

TallFurryMan commented 4 years ago

My obs is testing #25 right now. I'll test #26 shortly afterwards. I suppose it's better to remain on a single thread to avoid locking kstars up on the other.

TallFurryMan commented 4 years ago

EDIT: moved to #26

25 closed, uneeded. #26 accepts frames.

This is related to KStars/Ekos, but part of my test on #26, so I'll put it here for completeness:

I observe that "Finding stars..." remains on a starless frame, but this is minor error management.
It takes 40 seconds to display a 4K frame, find 50 stars and calculate HFR on my Atom i386 (8 seconds to display the frame).
However the rest of the system is not locked up while doing this, this is a HUGE improvement 👍
~~I observe that it is not possible to open a frame with the option "Mark stars" enabled.~~

knro commented 4 years ago

How does Centeroid algorithm fares in comparison? You can hardcode it in FITSData::findStar and re-test on the same frame.

TallFurryMan commented 4 years ago

Will try. It is not that easy to do this in the UI, there should be a test doing a benchmark of each method. Any comparison done on the KStars UI Focus test? I've not done that myself.

rlancaste / stellarsolver

SIGFPE on i386 #23

25 closed, uneeded. #26 accepts frames.