Hi Przemek, >> With rebuilt hbvmmt, the 25s went down to 17s, which >> seems okay on win platform. > > Better though still the difference is quite huge. > In my Linux box it's > ST: > [ total application time: ]....................................14.69 > MT: > [ total application time: ]....................................18.42 > > (GCC4.4) > > I also tested MinGW/GCC 3.4.5 with WINE and > ST: > [ total application time: ]....................................18.68 > MT: > [ total application time: ]....................................26.31
That's somewhat better than I get, maybe AMD is better here than C2D. (or my VM influences it) >> Is there a way to turn inline ASM stuff on? >> (sacrificing Win9x compatibility) >> I'd make a test with it to see the diff. > > I also do not now it they will work with Win64. Probably not though > we can update then for 64 bit mode (here Win9x is not a problem at all :)) That's right, I forgot it for an instant :) BTW, we can also enable it for MSVC 2008 and upper (32-bit), since they don't support Win9x anymore. > I'll commit small cleanup and build time macro to make such tests ASAP. > Anyhow when HB_STACK_PRELOAD is enabled the difference is not such huge. > In my test above MinGW build with stack ASM inline function and > HB_STACK_PRELOAD macro enabled together gives for MT mode: > [ total application time: ]....................................23.95 > and such difference is close the results in Linux builds. > >>> BTW it will be also good to check how works native thread local >>> variables in current MinGW builds. My last test are quite old and >>> it's highly possible that now the situation is much better. >> How to check this? > > Try to recompile Harbour with HB_USER_CFLAGS=-DHB_USE_TLS and check > the results. Maybe I will not have to touch this ASM code at all :) Tried it with mingw 4.5.0 and got 33s, so the speed halved. Strange to me. Brgds, Viktor _______________________________________________ Harbour mailing list (attachment size limit: 40KB) Harbour@harbour-project.org http://lists.harbour-project.org/mailman/listinfo/harbour
