Valgrind (notite)

VALGRIND

- tools available: memcheck (for memory leaks), cachegrind (detects cache misses), massif (heap and stack profiling), helgrind, callgrind, DRD

- cannot be attached to a already running process (Valgrind wants to have full control from the very start)

- results are given for the whole process but you can find the errors coming from a specific shared object

Memcheck

> valgrind --tool=memcheck --leak-check=yes --track-children=yes --log-file=log.txt java Example

Types of errors (counting the occurrences):

• illegal read/write

    cat log.txt | grep "Invalid" | wc -l

• use of uninitialised values

    cat log.txt | grep "uninitialised" | wc -l 
Results at the end of log file:

• definitely lost = program is leaking memory
• indirectly = leaking memory in a pointer-based structure
• possibly = leaking memory unless pointers do something awesome
• still rechable = it just didn't free some memory it could have freed
• suppressed = to be ignored

Cachegrind

> valgrind --tool=cachegrind --log-file=log.txt java -Djava.library.path=/home/gtache/workspace/MKL/lib Example

Results in the log file, miss rates for L1, L2, L3. 

Massif

> valgrind --tool=massif --stacks=yes --time-unit=B --log-file=log.txt java --massif-out-file=massif.out.txt -Djava.library.path=/home/gtache/workspace/MKL/lib Example

Results in massif.out

> vim massif.out  or  ms_print massif.out # snapshots of total memory consumption for heap/stack at different moments of time

• mem_heap_B = the number of useful heap bytes allocated at that point
• mem_heap_extra_B = the number of bytes allocated in the excess of what the program asked for (might be associated with alignment)
• mem_stacks_B = size of the stack

Sometimes it gives the detailed heap tree for a snapshot.

> massif_visualizer massif.out # graphical output