Some notes on QuantLib
Compilation flags
Configure with following flag to slightly improve overall performance:
CXXFLAGS="-O3 -g -msse2 -msse3 -msse -march=pentium4 -mfpmath=sse -ffast-math"
The above line will not work on x64 targets. You can still pass:
CXXFLAGS="-O3 -g -mfpmath=sse -ffast-math"
How much faster?
On my testing machine the overall performance index as returned by quantlib-benchmark (part of the test-suite) moves up from 324 mflops to 371 mflops. The test with greatest improvement is RandomNumber::MersenneTwisterDescrepancy which moves up from 423 to 961 mflops.