Follow up on __cyg_ ... functions

Follow up on __cyg_ ... functions

[Bill Mahoney]

Brian:

Thanks for the reply - I appreciate it.

This much I knew. However! Let me rephrase my question. Suppose I 
compile something with -finstrument-functions and don't supply my own 
function. The linker must resolve it to something, so I assumed that 
there is an empty function (e.g. a "ret" with a global label on it) 
somewhere in the runtime library source; otherwise "ld" would barf. Is 
this not the case, and the linker is just supplying that silently, or what?

The reason I ask is that a couple of years ago I instrumented blocks as 
well. (I am guilty of not posting these to the GCC community.) So I also 
have __cyg_profile_block as an additional extern. I wanted to supply a 
"default, do nothing" function for this as well.

Thoughts?  Thanks again.

Bill


Brian Dessent wrote:
> Bill Mahoney wrote:
>
>   
>> Hey, I would like to know on a new 4.3.2 install, where the source is
>> for the portion of the RT library that contains __cyg_profile_func_enter
>> / _exit.  Can somebody just email me back and aim me at the correct
>> file? I've find/grep'd until my fingers are tired.
>>     
>
> The compiler will insert calls to those functions when you use
> -finstrument-functions but you are meant to provide an implementation of
> them in your own code to receive the data and process it as you wish. 
> So you won't find what you're looking for in gcc anywhere because it
> doesn't exist.
>
> Brian
>   

>   

Building GCC without an assembler

[Gobaan_Raveendran]

Hello,

I am working on building gcc as a cross compiler and I do not have a
working assembler, thus it fails when trying to build the libraries. I
think I can build a simple program to read the input and spit it back out
so no errors are produced, but I was wondering if there is any way to avoid
building the libraries so that I can experiment with code that does not
need to be assembled/linked. I currently just install the broken cross
compiler as it allows me to experiment, but I am wondering if there is a
better way.

Thanks,

- Gobi

Re: Follow up on __cyg_ ... functions

[Brian Dessent]

Bill Mahoney wrote:

> This much I knew. However! Let me rephrase my question. Suppose I
> compile something with -finstrument-functions and don't supply my own
> function. The linker must resolve it to something, so I assumed that
> there is an empty function (e.g. a "ret" with a global label on it)
> somewhere in the runtime library source; otherwise "ld" would barf. Is
> this not the case, and the linker is just supplying that silently, or what?

It depends on the target.  glibc has an implementation which writes the
profiling data to a FIFO named in the environment variable
PCPROFILE_OUTPUT (otherwise a no-op if not set.)  Other targets don't
provide anything:

$ gcc -finstrument-functions tc.c
/tmp/cc92y1eZ.o:tc.c:(.text+0x15): undefined reference to
`___cyg_profile_func_enter'
/tmp/cc92y1eZ.o:tc.c:(.text+0x28): undefined reference to
`___cyg_profile_func_exit'
/tmp/cc92y1eZ.o:tc.c:(.text+0x63): undefined reference to
`___cyg_profile_func_enter'
/tmp/cc92y1eZ.o:tc.c:(.text+0x85): undefined reference to
`___cyg_profile_func_exit'
collect2: ld returned 1 exit status

Brian

Re: Building GCC without an assembler

[Brian Dessent]

Gobaan_Raveendran@amis.com wrote:

> I am working on building gcc as a cross compiler and I do not have a
> working assembler, thus it fails when trying to build the libraries. I
> think I can build a simple program to read the input and spit it back out
> so no errors are produced, but I was wondering if there is any way to avoid
> building the libraries so that I can experiment with code that does not
> need to be assembled/linked. I currently just install the broken cross
> compiler as it allows me to experiment, but I am wondering if there is a
> better way.

To build just the compiler and not the target libraries use the make
target "all-gcc".  In theory you should be able to install this with
"install-gcc" as well.

Brian

Re: Follow up on __cyg_ ... functions

[Andrew Haley]

Bill Mahoney wrote:
> Brian:
> 
> Thanks for the reply - I appreciate it.
> 
> This much I knew. However! Let me rephrase my question. Suppose I
> compile something with -finstrument-functions and don't supply my own
> function.

Then your program won't link.

> The linker must resolve it to something, so I assumed that
> there is an empty function (e.g. a "ret" with a global label on it)
> somewhere in the runtime library source; otherwise "ld" would barf. Is
> this not the case, and the linker is just supplying that silently, or what?

If your program links you have a definition of __cyg_ ... from
somewhere.  A linker map will tell you where it comes from.

Andrew.

Re: Follow up on __cyg_ ... functions

[Bill Mahoney]

Brian and (Andrew Haley):

But, you see, it is not unresolved for me:

[bmahoney@IDEA play]$ gcc -finstrument-functions -o test test.c
[bmahoney@IDEA play]

And a "nm" shows that they're going to get sucked in as part of the library:

                 U __cyg_profile_func_enter@@GLIBC_2.2.5
                 U __cyg_profile_func_exit@@GLIBC_2.2.5

which brings me back to... Where are they in the lib source? No amount 
of grepping seems to find it. The doc's for GCC say that the runtime is 
in the distribution sources, so I assume that is true. This is from a 
brand new 4.3.2 install.

Bill


Brian Dessent wrote:
> Bill Mahoney wrote:
>
>   
>> This much I knew. However! Let me rephrase my question. Suppose I
>> compile something with -finstrument-functions and don't supply my own
>> function. The linker must resolve it to something, so I assumed that
>> there is an empty function (e.g. a "ret" with a global label on it)
>> somewhere in the runtime library source; otherwise "ld" would barf. Is
>> this not the case, and the linker is just supplying that silently, or what?
>>     
>
> It depends on the target.  glibc has an implementation which writes the
> profiling data to a FIFO named in the environment variable
> PCPROFILE_OUTPUT (otherwise a no-op if not set.)  Other targets don't
> provide anything:
>
> $ gcc -finstrument-functions tc.c
> /tmp/cc92y1eZ.o:tc.c:(.text+0x15): undefined reference to
> `___cyg_profile_func_enter'
> /tmp/cc92y1eZ.o:tc.c:(.text+0x28): undefined reference to
> `___cyg_profile_func_exit'
> /tmp/cc92y1eZ.o:tc.c:(.text+0x63): undefined reference to
> `___cyg_profile_func_enter'
> /tmp/cc92y1eZ.o:tc.c:(.text+0x85): undefined reference to
> `___cyg_profile_func_exit'
> collect2: ld returned 1 exit status
>
> Brian
>   

Re: Follow up on __cyg_ ... functions

[Andrew Haley]

Bill Mahoney wrote:
> Brian and (Andrew Haley):
> 
> But, you see, it is not unresolved for me:
> 
> [bmahoney@IDEA play]$ gcc -finstrument-functions -o test test.c
> [bmahoney@IDEA play]
> 
> And a "nm" shows that they're going to get sucked in as part of the
> library:
> 
>                 U __cyg_profile_func_enter@@GLIBC_2.2.5
>                 U __cyg_profile_func_exit@@GLIBC_2.2.5
> 
> which brings me back to... Where are they in the lib source? No amount
> of grepping seems to find it. The doc's for GCC say that the runtime is
> in the distribution sources, so I assume that is true. This is from a
> brand new 4.3.2 install.

Did you generate a link map, as I suggested?  That would immediately
tell you where it was coming from: it's glibc.

  gcc -Wl,-Map,map

Andrew.

Re: Follow up on __cyg_ ... functions

[Brian Dessent]

Bill Mahoney wrote:

> which brings me back to... Where are they in the lib source? No amount
> of grepping seems to find it. The doc's for GCC say that the runtime is
> in the distribution sources, so I assume that is true. This is from a
> brand new 4.3.2 install.

As I said, glibc provides an implementation.  glibc is not part of gcc. 
The runtime referred to is the compiler support runtime (i.e. libgcc,
libstdc++, libgfortran, etc.) not libc.  gcc does not include a C
library.

Brian