On Thursday 31 July 2008 13:53:21 Moritz Lenz wrote: > rurban ask on IRC if somebody already tested parrot with llvm-gcc. So I > just tried it out, and, *miracle*, it actually worked. Without any > modifications. > > Just a > make realclean; perl Configure.pl --cc=llvm-gcc && make test > yields > All tests successful, 17 tests and 658 subtests skipped. > > with llvm-gcc: > > $ time make benchmark_tests > ... > t/benchmark/benchmarks....ok > > All tests successful. > Files=1, Tests=37, 466 wallclock secs (454.96 cusr + 10.40 csys = 465.36 > CPU) > > real 7m47.791s > user 7m36.789s > sys 0m10.693s > > with gcc 4.2: > > Files=1, Tests=37, 382 wallclock secs (371.02 cusr + 10.83 csys = 381.85 > CPU) > > real 6m24.233s > user 6m12.827s > sys 0m11.113s > > So no speed improvements for the unoptimized build, but still good news!
I expect much better improvements for the optimized build; LLVM should be able to optimize across .o files where GCC can't. It's definitely good news though. -- c
