AddressSanitizerPerformanceNumbers

Alexander Potapenko edited this page Aug 31, 2015 · 5 revisions

Spec CPU 2006 (C/C++)

These numbers are measured on SPEC 2006 (C/C++ only) using Clang3.3 (trunk) r179094 (on Google Code) (April 09 2013) on Intel Xeon W3690 @3.47GHz. 2-nd column: clang -O2 3-rd column: clang -O2 -fsanitize=address -fno-omit-frame-pointer

BENCHMARK O2 O2+asan slowdown
400.perlbench 344.00 1304.00 3.79
401.bzip2 490.00 844.00 1.72
403.gcc 322.00 608.00 1.89
429.mcf 316.00 583.00 1.84
445.gobmk 409.00 833.00 2.04
456.hmmer 605.00 1226.00 2.03
458.sjeng 456.00 982.00 2.15
462.libquantum 480.00 539.00 1.12
464.h264ref 547.00 1311.00 2.40
471.omnetpp 314.00 587.00 1.87
473.astar 403.00 655.00 1.63
483.xalancbmk 227.00 493.00 2.17
433.milc 408.00 666.00 1.63
444.namd 369.00 593.00 1.61
447.dealII 321.00 597.00 1.86
450.soplex 233.00 367.00 1.58
453.povray 189.00 425.00 2.25
470.lbm 318.00 393.00 1.24
482.sphinx3 519.00 911.00 1.76

The average slowdown is 1.93, the median slowdown is 1.86.

Older measurements: asan r496 (on Google Code) (clang -O2 -fsanitize=address) with malloc_context_size=1 (i.e. no unwinding). The baseline is clang -O2. The average slowdown introduced by full instrumentation (red) is 1.73. When only writes are instrumented (blue), the average slowdown is 1.26.

https://raw.githubusercontent.com/wiki/google/sanitizers/spec_r496.png

Clone this wiki locally
You can’t perform that action at this time.
You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session.
Press h to open a hovercard with more details.