The following example demonstrates the use of sprof
. The example
consists of a main program that calls two functions in a shared
object. First, the code of the main program:
$ cat prog.c
#include <stdlib.h>
void x1(void);
void x2(void);
int
main(int argc, char *argv[])
{
x1();
x2();
exit(EXIT_SUCCESS);
}
The functions x1() and x2() are defined in the following source
file that is used to construct the shared object:
$ cat libdemo.c
#include <unistd.h>
void
consumeCpu1(int lim)
{
for (int j = 0; j < lim; j++)
getppid();
}
void
x1(void) {
for (int j = 0; j < 100; j++)
consumeCpu1(200000);
}
void
consumeCpu2(int lim)
{
for (int j = 0; j < lim; j++)
getppid();
}
void
x2(void)
{
for (int j = 0; j < 1000; j++)
consumeCpu2(10000);
}
Now we construct the shared object with the real name
libdemo.so.1.0.1, and the soname libdemo.so.1:
$ cc -g -fPIC -shared -Wl,-soname,libdemo.so.1 \
-o libdemo.so.1.0.1 libdemo.c
Then we construct symbolic links for the library soname and the
library linker name:
$ ln -sf libdemo.so.1.0.1 libdemo.so.1
$ ln -sf libdemo.so.1 libdemo.so
Next, we compile the main program, linking it against the shared
object, and then list the dynamic dependencies of the program:
$ cc -g -o prog prog.c -L. -ldemo
$ ldd prog
linux-vdso.so.1 => (0x00007fff86d66000)
libdemo.so.1 => not found
libc.so.6 => /lib64/libc.so.6 (0x00007fd4dc138000)
/lib64/ld-linux-x86-64.so.2 (0x00007fd4dc51f000)
In order to get profiling information for the shared object, we
define the environment variable LD_PROFILE
with the soname of the
library:
$ export LD_PROFILE=libdemo.so.1
We then define the environment variable LD_PROFILE_OUTPUT
with
the pathname of the directory where profile output should be
written, and create that directory if it does not exist already:
$ export LD_PROFILE_OUTPUT=$(pwd)/prof_data
$ mkdir -p $LD_PROFILE_OUTPUT
LD_PROFILE
causes profiling output to be appended to the output
file if it already exists, so we ensure that there is no
preexisting profiling data:
$ rm -f $LD_PROFILE_OUTPUT/$LD_PROFILE.profile
We then run the program to produce the profiling output, which is
written to a file in the directory specified in
LD_PROFILE_OUTPUT
:
$ LD_LIBRARY_PATH=. ./prog
$ ls prof_data
libdemo.so.1.profile
We then use the sprof -p
option to generate a flat profile with
counts and ticks:
$ sprof -p libdemo.so.1 $LD_PROFILE_OUTPUT/libdemo.so.1.profile
Flat profile:
Each sample counts as 0.01 seconds.
% cumulative self self total
time seconds seconds calls us/call us/call name
60.00 0.06 0.06 100 600.00 consumeCpu1
40.00 0.10 0.04 1000 40.00 consumeCpu2
0.00 0.10 0.00 1 0.00 x1
0.00 0.10 0.00 1 0.00 x2
The sprof -q
option generates a call graph:
$ sprof -q libdemo.so.1 $LD_PROFILE_OUTPUT/libdemo.so.1.profile
index % time self children called name
0.00 0.00 100/100 x1 [1]
[0] 100.0 0.00 0.00 100 consumeCpu1 [0]
-----------------------------------------------
0.00 0.00 1/1 <UNKNOWN>
[1] 0.0 0.00 0.00 1 x1 [1]
0.00 0.00 100/100 consumeCpu1 [0]
-----------------------------------------------
0.00 0.00 1000/1000 x2 [3]
[2] 0.0 0.00 0.00 1000 consumeCpu2 [2]
-----------------------------------------------
0.00 0.00 1/1 <UNKNOWN>
[3] 0.0 0.00 0.00 1 x2 [3]
0.00 0.00 1000/1000 consumeCpu2 [2]
-----------------------------------------------
Above and below, the "<UNKNOWN>" strings represent identifiers
that are outside of the profiled object (in this example, these
are instances of main()).
The sprof -c
option generates a list of call pairs and the number
of their occurrences:
$ sprof -c libdemo.so.1 $LD_PROFILE_OUTPUT/libdemo.so.1.profile
<UNKNOWN> x1 1
x1 consumeCpu1 100
<UNKNOWN> x2 1
x2 consumeCpu2 1000