Which GCC O2 optimization flags are likely to prevent an Segmentation fault?

Printable View

10-05-2020
stahta01

Which GCC O2 optimization flags are likely to prevent an Segmentation fault?

I am building an FLOSS project and getting Segmentation faults during build-time (where it runs an project created exe).

The project is created using MinGW64 GCC version (Rev3, Built by MSYS2 project) 10.2.0.

The project is SDCC; I am trying to package it under MSys2 64 bit MinGW.

It has the Segmentation faults when option "-O1" is used.
But, does not have the Segmentation faults when option "-O2" is used.

From Optimize Options (Using the GNU Compiler Collection (GCC))

The flags added by "-O2" are:

Code:

-falign-functions -falign-jumps -falign-labels -falign-loops -fcaller-saves -fcode-hoisting -fcrossjumping -fcse-follow-jumps -fcse-skip-blocks -fdelete-null-pointer-checks -fdevirtualize -fdevirtualize-speculatively -fexpensive-optimizations -ffinite-loops -fgcse -fgcse-lm -fhoist-adjacent-loads -finline-functions -finline-small-functions -findirect-inlining -fipa-bit-cp -fipa-cp -fipa-icf -fipa-ra -fipa-sra -fipa-vrp -fisolate-erroneous-paths-dereference -flra-remat -foptimize-sibling-calls -foptimize-strlen -fpartial-inlining -fpeephole2 -freorder-blocks-algorithm=stc -freorder-blocks-and-partition -freorder-functions -frerun-cse-after-loop -fschedule-insns -fschedule-insns2 -fsched-interblock -fsched-spec -fstore-merging -fstrict-aliasing -fthread-jumps -ftree-builtin-call-dce -ftree-pre -ftree-switch-conversion -ftree-tail-merge -ftree-vrp

I am trying to build with these added flags to see if the seg fault is present or not.

Code:

_debug_FLAGS+=" -finline-functions" _debug_FLAGS+=" -finline-small-functions" _debug_FLAGS+=" -findirect-inlining" _debug_FLAGS+=" -foptimize-strlen" _debug_FLAGS+=" -ftree-switch-conversion"

It takes several hours for the build to finish.

Edit: My question is which of the 5 I should keep trying and which of the others I should try? I am guessing I will hit the max command line length if I try to use all the options.

Tim S.
10-05-2020
laserlight

I think the answer is: the ones you use when you are lucky (or unlucky, as the case may be).

After all, it is probably just some bug that happens to disappear when a particular optimisation is applied; in another case it could very well have been some bug that only occurs when a particular optimisation is applied. In my understanding, the latter is more often true, i.e., it is more likely that programmers might code with flawed assumptions that don't hold when optimisations are applied, and then it is also more likely for a compiler bug to cause an optimisation to be applied incorrectly, so optimisation flags are usually more likely to result in a segmentation fault than to prevent one.
10-05-2020
stahta01

Ones of these appear to be hiding the seg fault.
My guess is "-fdelete-null-pointer-checks".
Building with the last six removed to see if one of them might be the cause.
Edit: The problem was not in the last six.

Code:

-fcaller-saves -fcode-hoisting -fdelete-null-pointer-checks -ffinite-loops -finline-functions -finline-small-functions -findirect-inlining -fisolate-erroneous-paths-dereference -foptimize-sibling-calls -foptimize-strlen -fpartial-inlining -ftree-switch-conversion

Tim S.
10-06-2020
stahta01

The winner appears to be.

Quote:

-fcaller-saves
Enable allocation of values to registers that are clobbered by function calls, by emitting extra instructions to save and restore the registers around such calls. Such allocation is done only when it seems to result in better code.

This option is always enabled by default on certain machines, usually those which have no call-preserved registers to use instead.

Enabled at levels -O2, -O3, -Os.

I had hoped for a result that gave me a hint on where to look in the source code. I still need to confirm the problem still exists after removing this option that I believe is hiding the segfault. Since, I am building an SVN project it might have added an upstream fix. Or some other minor package change I did fixed the problem.

Tim S.
10-07-2020
stahta01

Code:

-Og -ftree-bit-ccp -ftree-pta -fif-conversion2 -finline-functions-called-once -fcaller-saves

The above options avoids the segment fault error; now, I need to find how to avoid the error by fixing the project source code or just decide to use these options.

Tim S.
10-10-2020
Malcolm McLean

Quote:

Originally Posted by stahta01

Code:

-Og -ftree-bit-ccp -ftree-pta -fif-conversion2 -finline-functions-called-once -fcaller-saves

The above options avoids the segment fault error; now, I need to find how to avoid the error by fixing the project source code or just decide to use these options.

Tim S.

Comment out everything except a line in main which prints out "Hello world". If it still segfaults then the problem is not in your source. Then gradually comment in to "Hello world" and comment out in the crashing program until you can make "Hello world" segfault and the crashing program not segfault. That should enable you to toggle between the two conditions with a few simple code changes, and tell you where the fault lies.
10-10-2020
stahta01

Quote:

Originally Posted by Malcolm McLean

Comment out everything except a line in main which prints out "Hello world". If it still segfaults then the problem is not in your source. Then gradually comment in to "Hello world" and comment out in the crashing program until you can make "Hello world" segfault and the crashing program not segfault. That should enable you to toggle between the two conditions with a few simple code changes, and tell you where the fault lies.

Please read a thread before replying to them!

Tim S.
10-10-2020
stahta01

The segment faults are in sdar which is in folder sdbinutils of SDCC.
sdbinutils is a fork of binutils version 2.30 sdar is same as binutils ar.

I have decided I am going to have to read the source code of sdar to look for possible problems in the files listed in the gdb backtrace.
But, first, I am going to have to do a dozen or so backtraces. It is possible that there is many different causes of the segment faults.
And, who knows maybe only one file will be listed in common ( not counting ar.c ) and I will be able to find the problem quickly.

Tim S.
10-13-2020
stahta01

I have tried to update sdbinutils to be based on binutils version 2.31.1 and for 5 of the about 25 backends that use sdbinutils it no longer seg faults when compiled with
"-Og -fcaller-saves" I consider this to be a good enough fix; but, only once I verify that the majority of the rest of the backends are also fixed from needing these options to avoid seg faults (-ftree-bit-ccp -ftree-pta -fif-conversion2 -finline-functions-called-once).

The option "-fcaller-saves" likely will only interfere slightly if the debugger is used. While (-ftree-bit-ccp -ftree-pta -fif-conversion2 -finline-functions-called-once) would have major effect on debugging.

Tim S.