- Program profiling is finging out how much time
the computer is using to execute certan part of a program
- Traditional profilers usually operate at the function level
(count how many times a certain function is executed)
Some newer profilers such as tcov can perform statement-by-statement profile...
- Compile each function that you want to profile with the
profiling option:
-p
Example:
CC -c -p file1.C # Profiling functions in file1.C CC -c file2.C # NO Profiling functions in file2.C CC -p -o myProg file1.o file2.o # Always need -p option in linking
- Run your program normally
The execution will create a monitor output file mon.out
It is in a special format that is intended for a special program, so do not try to edit or view it....
- Use the prof COMMAND
command to look at the monitor ourput:
prof myProg
-
Example Program:
(profiling) ---
click here
- Compile with one of these commands:
- CC -p profile.C
- CC -O1 -p profile.C
- CC -O2 -p profile.C
- CC -O3 -p profile.C
- CC -O4 -p profile.C
- CC -fast -p profile.C
- Run a.out
- Run "prof a.out"
You will see different time results for different level of optimizations.
- Sample output (CC -p profile.C)
%Time Seconds Cumsecs #Calls msec/call Name 98.2 2.24 2.24 12750 0.1757 __1cBf6Fi_f_ // f() 1.8 0.04 2.28 50 0.8 __1cBg6Ff_f_ // g() 0.0 0.00 2.28 1 0. main
- Compile each function that you want to profile with this special
profiling option:
-xprofile=tcov
Example:
CC -c --xprofile=tcov file1.C CC -c file2.C CC -xprofile=tcov -o myProg file1.o file2.o
- Run your program normally
The execution will now create a monitor directory myProg.profile where "myProg" is the name of the executable program file
The coverage data is stored in program.profile/tcovd. You can view this file with an editor, but there is a better tool to use the data....
- Run:
tcov -x myProg.profile programFile.C
This will create a file programFile.C.tcov in the current directory and this file will contain the data from the "coverage analysis" merged with its source file.
A separate .tcov file will be created for each source file.
- Usage notes:
- Each subsequent run accumulates more coverage data
into the program.profile/tcovd file.
- Data for each object file is zeroed out
the first time the program is executed after
recompilation.
- You can zeroed the data for the entire program by removing the tcovd file.
- Each subsequent run accumulates more coverage data
into the program.profile/tcovd file.
-
Example Program:
(profiling) ---
click here
- Compile with:
- CC -xprofile=tcov profile.C
- Run a.out
- Run "tcov -x a.out.profile profile.C"
- Look at the file "profile.C.tcov"
#include#define MAX 50 float sum = 0.1234; float A[MAX] = {1, 2, 3, 4, 5, 6, 7}; // Array float f(int x) 12750 -> { int i, sum; sum = 1; for (i = 0; i < 100*x; i++) 213987500 -> sum = sum + i; 12750 -> return(sum); } float g(float x) 50 -> { int i, sum; sum = 0; for (i = 0; i < 10*x; i++) 12750 -> sum = sum + f(i); 50 -> return(sum); } int main(int argc, char *argv[]) 1 -> { int i, j; for (i = 0; i < MAX; i = i + 1) 50 -> { j = i; // Use "stop if j > 10" to stop before // the function call g(i+1) A[i] = 1/g(i+1); } 1 -> sum = 0.0; for (i = 0; i < MAX; i = i + 1) 50 -> { sum = sum + A[i]; } 1 -> cout << "sum = " << sum << "\n\n"; } Top 10 Blocks Line Count 16 213987500 11 12750 17 12750 26 12750 21 50 27 50 37 50 45 50 32 1 43 1 11 Basic blocks in this file 11 Basic blocks executed 100.00 Percent of the file executed 214025953 Total basic block executions 19456904.00 Average executions per basic block
- Compile with: