Issue preventing GCC from optimizing out global variable

Same issue when using ARM-GCC v4.9 (released 2015-09-25).

Hi Radu,

I cannot reproduce your issue with out 2015Q4 toolchain:

% cat main.c
static const char AppVersion[] __attribute__ ((used)) = "v3.05/10.oct.2015";

int
main (void)
{
  return 0;
}

% arm-none-eabi-gcc -o main.axf main.c -specs=rdimon.specs
% strings main.axf | grep oct
v3.05/10.oct.2015

Can you give us some more detailed instructions on how to reproduce this issue? In the meantime you can try removing the const qualifier.

Best regards.

Hi Thomas,

First, thanks for taking the time to answer.

Well, I didn't have the chance to install the new version 5.0 of the toolchain (or is it 5.2? :-)

All the test were done using the version 4.9 (released 2015-09-25).
Even if the string was definitely there in the object file main.o (or main.obj - can't remember exactly),
the string was removed after that by the linker, so I clearly wasn't able to find it in the executable *.HEX file!

So, briefly, the point is: yes, the string is there in the object file(s), yet it is wiped out by the linker...
Unfortunately, now I am not able to test it using the newer version v5.x of the toolchain.

Kind regards,
Radu.

Hi Radu,

I can not reproduce it with the 4.9 toolchain either.

$arm-none-eabi-gcc --version
arm-none-eabi-gcc (GNU Tools for ARM Embedded Processors) 4.9.3 20150529 (release) [ARM/embedded-4_9-branch revision 227977]
Copyright (C) 2014 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

$ cat t.c
static const char AppVersion[] __attribute__ ((used)) = "v3.05/10.oct.2015";

int
main (void)
{
    return 0;
}

$arm-none-eabi-gcc t.c --specs=rdimon.specs
$arm-none-eabi-strings a.out
v3.05/10.oct.2015

Can you provide us with the full compilation command line?

Cheers,
Andre

Hi Andre,

One question: "a.out" is the output of the compiler as an <object> file?
Or is it the output of the LINKER, i.e. something like a BIN or HEX file ?

I am not a guru of compilers or linkers, as a matter of fact I am using an IDE that makes all this things transparently (compile, build, etc)...
So, to answer your question: There is no "command line", the IDE named CoIDE (www.coocox.org) does all the stuff.
Yet, I am 99.99% sure that my problem is not caused by the compiler, but by the linker (please, see my previous message).
"...briefly, the point is: yes, the string is there in the object file(s), yet it is wiped out by the linker..."

OTOH, I will try tomorrow to see if there is a way to expose what exactly is CoIDE feeding the compiler and linker with.
But, again, I am pretty sure it's the linker who has to be convinced somehow not to wipe away the AppVersion string...

Best regards,
Radu.

Hi Radu,

'a.out' is the binary/executable and thus the result of the linker.

I had a look at CooCoox and it passes --gc-sections by default. Right click on the project, in the project navigation panel and choose Configuration. Then go to the tab "Link" and uncheck "Discard unused sections".

This should do the trick.

Cheers,
Andre

Hi Andre,

Unchecking that option "Discard unused sections" brings back the AppVersion string into the binary/executable file, but unfortunately with a huge penalty: the size of the binary file is now about 31% bigger than before... :(

I was thinking maybe(?) there is another way to solve the issue, something that allows me to keep only that particular string, not ALL unused strings/sections.

For example I saw some other compilers/linkers (for Atmel if I recall well) offer something like this to be used in the source code
(now I realize there's only one way to solve the problem... i.e. at source-code level) :
     #pragma keep+
      < code >
     #pragma keep-

Maybe you have a clue if this can be done also for ARM GCC by any chance...?

Many thanks in advance!
Obrigado,
Radu.

Hi Andre,

Yes, indeed ...
__asm__ ("" : : "" (AppVersion));
... solved the problem :)

I want to express my gratitude for your valuable help!

If you are a member of stackoverflow.com, please spend few seconds to copy your above answer and paste it as an answer here:
http://stackoverflow.com/questions/33351690/issue-preventing-gcc-from-optimizing-out-global-variable

Other people on stackoverflow showed real interest in this matter, so they could benefit also from your idea.
Of course, I would be very happy to vote it :-)

OTOH, if you don't have an account on stackoverflow, I will ask for your permission to post your answer there, mentioning of course you name as the author and adding a link to this site/forum.

Com os melhores cumprimentos,
Radu.

Thanks Andre Vieira, that solved my question.