profiling C shared libraries loaded into R

Simon:

Sorry, please disregard my last email.  Shark was crashing because  
some of something with Shark, not R.  But unfortunately, the Unknown  
Libraries remain.

Here is how I compiled R.  In the config.site file, I set:

R_PAPERSIZE=letter
R_BATCHSAVE=--no-save
CC='gcc-4.2 -arch x86_64 -m64 -march=core2 -msse3 -g -pg -O2'
OBJC='gcc-4.2 -arch x86_64 -m64 -march=core2 -msse3 -g -pg -O2'
DEFS=-DHAVE_INLINE
F77='gfortran -arch x86_64 -g -pg -O2'
LIBnn=lib
CXX='g++-4.2 -arch x86_64 -march=core2 -p -pg -O2'
FC='gfortran -arch x86_64 -m64 -march=core2 -g -pg -O2'
TCLTK_LIBS=/usr/lib

To configure, I ran the following script

#!/bin/bash

./configure \
SHELL='/bin/bash' \
r_arch=x86_64 \
--with-system-zlib \
--with-blas='-framework vecLib' --with-lapack='-framework vecLib' \
--with-ICU \
--with-aqua \
--enable-R-framework \
--enable-BLAS-shlib \
--with-x \
--with-readline=/usr/local \
--with-tcltk \
--with-iconv=no \
--with-cairo \
--disable-R-profiling \
--with-recommended-packages=no


Any idea what I did wrong?

Thanks,

Michael



On Jun 12, 2009, at 7:27 PM, Simon Urbanek wrote:

Michael,

On Jun 12, 2009, at 4:20 PM, Michael Braun wrote:

Hi.  I have a shared library that I would like to profile.  I see  
that this issue has been touched on in some previous posts, but I  
am still having trouble, and I am hoping someone here can help.

First things first...  I am running R 2.8.0 patched on OS X  
10.5.7, on a dual quad-core Mac Pro.   It is the 64-bit binary  
build for Leopard, as downloaded from CRAN.

I am compiling my code using gcc-4.2, and I am including the -g -p  
-pg flags.

The -p -pg part (they are mutually-exclusive aren't they? so you  
get just -pg) won't really help unless you also compile R that way  
since the gprof initialization happens in the executable, not in  
the library (and don't forget it at link time).

Optimization is turned off (-O0).

Just to make sure - using PKG_CFLAGS doesn't have the desired  
effect, because R's -O2 will override it, so better be sure that  
the optimization is off. If you run the steps manually, make sure  
you have -g (and -pg if you wish) everywhere.

I AM able to debug the code using gdb (called as R -d gdb).  I  
always run R from the command line.

I have tried profiling using Shark, which recognizes functions in  
libR.dylib, but none of the functions in mylib.so.  For each  
function call, Shark refers to "Unknown Library", and after the  
function address, there is an indicator of [unknown].  Also, when  
I use Shark, it appears to want to profile the entire R session.   
All I care about is my library.

I'm unable to replicate that. Shark correctly recognizes the  
function and the loaded library for me. Try using default flags -  
they do include -g -O2 (which worked for me).

I am also open to using gprof instead, but I cannot find the  
requisite gmon.out file.

See above - you won't find it unless you compile R with -pg since  
the gprof init code must be in the executable (more precisely you  
don't have to re-compile the entire R with that, it should be  
sufficient to use -pg in the link step for exec/R).

Cheers,
Simon

Thanks in advance for any suggestions.

Michael



-------------------------------------------
Michael Braun
Homer A. Burnell Career Development Professor, and
	Assistant Professor of Management Science (Marketing Group)
MIT Sloan School of Management
One Amherst St., E40-169
Cambridge, MA 02142
braunm at mit.edu
617-253-3436

-------------------------------------------
Michael Braun
Homer A. Burnell Career Development Professor, and
	Assistant Professor of Management Science (Marketing Group)
MIT Sloan School of Management
One Amherst St., E40-169
Cambridge, MA 02142
braunm at mit.edu
617-253-3436