Hi Przemek, Is it also true that C++ mode enhances speed for x86 code? I see it slightly does for x86_64.
Brgds, Viktor On 2010 Mar 10, at 16:43, Przemysław Czerpak wrote: > On Wed, 10 Mar 2010, francesco perillo wrote: > > Hi, > >>> BCC is a joke these days. >> ooops... I have a production site working with a bbc version of Harbour... > > It works but it's not very good choice. > There are some very serious bugs in BCC but for the ones known for us > we have workarounds, i.e. we do not use any floating point BCC low level > functions like _fpclass[l]() because they break 'double' number > calculations reducing the precision to 'float' level or we use > DLMALLOC as default memory manager because the one in BCC can cause > memory corruption in some situations. > But more important is the fact that BCC was not extended in last years > and now in comparison to other modern C compilers it does not have many > important features and optimization methods. > In contrast nearly each new GCC version has some important extensions > which improve optimization code and performance of final applications, > i.e. GCC 4.4 generates nearly twice faster code then 2.9x. (see results > below). > > Harbour core code is written in a way which should help in optimization > process for compilers using some advanced optimization methods like > automatic autoinlining or automatic detecting of restricted areas due > to constant declarations or strict aliasing rules. It even does not have > such simple optimization methods like using assembler inline code for > commonly used CTRL functions like memset(), memcmp(), memmove(), > str*(), etc. > The results can be visible in the speed of final binaries. > You can use harbour/tests/speedtst.prg to compare the speed of Harbour > compiled with different C compilers. > You can also use it with other xbase/clipper compatible languages like > xHarbour, Clipper, CLIP, FlagShip, xBase++. > > Below I'm attaching results for current Harbour SVN code. I cannot make > any tests with MSVC because I cannot use it with WINE. The speedtst.prg > code does not make any IO operations which may effect execution speed > so WINE and DOSEMU can execute final binaries naively without any time > consuming interrupts which may the results. > I used "official" MinGW with GCC 3.4.5 which gives slower code then GCC4.4. > I also do not have the newest PelleC 6.0. > > GCC4.5 was used with -flto which enable link time optimization (LTO). > As you can see GCC4.4 in C++ mode and GCC4.5 gives the fastest code > and old PelleC compiler used by xHarbour.com (XCC) gives the slowest > code. > I haven't installed yet C++ compiler for GCC4.5 which potentially should > give the fastest code. > Please note that all OpenWatcom builds were created with disable IPO > optimization in HVM code. Due to compilation time we disabled HB_HVM_ALL=yes > optimization which improve the performance of final OW binaries. > > Here are sorted the results. > > Harbour 2.1.0dev (Rev. 14060) GNU C 4.5 (64-bit) x86-64 > [ total application time: ]....................................14.66 > > Harbour 2.1.0dev (Rev. 14060) GNU C++ 4.4 (64-bit) x86-64 > [ total application time: ]....................................14.65 > > Harbour 2.1.0dev (Rev. 14060) GNU C 4.4 (64-bit) x86-64 > [ total application time: ]....................................14.95 > > Harbour 2.1.0dev (Rev. 14060) Intel(R) (ICC) C 11.0 (64-bit) x86-64 > [ total application time: ]....................................16.22 > > Harbour 2.1.0dev (Rev. 14060) Open64 C 4.2.1 (64-bit) x86-64 > [ total application time: ]....................................16.29 > > Harbour 2.1.0dev (Rev. 14060) LLVM/Clang C 4.2.1 (64-bit) x86-64 > [ total application time: ]....................................17.57 > > Harbour 2.1.0dev (Rev. 14060) GNU C 4.4 (32-bit) x86 > [ total application time: ]....................................18.13 > > Harbour 2.1.0dev (Rev. 14060) Sun C (ident 0x5100) (64-bit) x86-64 > [ total application time: ]....................................18.21 > > Harbour 2.1.0dev (Rev. 14060) MinGW GNU C 3.4.5 (32-bit) x86 > [ total application time: ]....................................19.09 > > Harbour 2.1.0dev (Rev. 14060) Delorie GNU C 4.3.2 (DJGPP 2.04) (32-bit) x86 > [ total application time: ]....................................23.18 > > Harbour 2.1.0dev (Rev. 14060) Open Watcom C 12.80 (32-bit) x86 (LINUX) > [ total application time: ]....................................23.54 > > Harbour 2.1.0dev (Rev. 14060) Open Watcom C++ 12.80.8 (32-bit) x86 (WIN) > [ total application time: ]....................................25.65 > > Harbour 2.1.0dev (Rev. 14060) Open Watcom C 12.80 (32-bit) x86 (DOS) > [ total application time: ]....................................25.76 > > Harbour 2.1.0dev (Rev. 14060) Borland C++ 5.5 (32-bit) x86 > [ total application time: ]....................................27.48 > > Harbour 2.1.0dev (Rev. 14060) GNU C 2.96 (32-bit) x86 > [ total application time: ]....................................29.39 > > Harbour 2.1.0dev (Rev. 14060) Pelles ISO C Compiler 4.50 (32-bit) x86 > [ total application time: ]....................................30.83 > > Harbour 2.1.0dev (Rev. 14060) Pelles ISO C Compiler 2.70 (32-bit) x86 (XCC) > [ total application time: ]....................................34.70 > > > For comparison I'm attaching result for some other compatible compilers. > xHarbour was compiled from current CVS with some small improvements and > fixes. > > xHarbour build 1.2.1 Intl. (SimpLex) (Rev. 6691) GNU C 4.4 (64 bit) ? > [ total application time: ]....................................32.42 > > xHarbour build 1.2.1 Intl. (SimpLex) (Rev. 6691) MinGW GNU C 3.4.5 (32 bit) ? > [ total application time: ]....................................33.78 > > xHarbour build 1.2.1 Intl. (SimpLex) (Rev. 6691) Borland C++ 5.5 (32 bit) ? > [ total application time: ]....................................44.02 > > FlagShip 4.48.2460 (1024 users) ? > [ total application time: ]....................................80.04 > > best regards, > Przemek > _______________________________________________ > Harbour mailing list (attachment size limit: 40KB) > Harbour@harbour-project.org > http://lists.harbour-project.org/mailman/listinfo/harbour _______________________________________________ Harbour mailing list (attachment size limit: 40KB) Harbour@harbour-project.org http://lists.harbour-project.org/mailman/listinfo/harbour
