|
| From: | Christian Jullien |
| Subject: | Re: [Tinycc-devel] modern c++ compiler written in C (food for thought) |
| Date: | Thu, 29 Oct 2015 06:32:45 +0100 |
Ooops! Don't misunderstand me. No doubt to me, tcc *IS* a real C compiler and, even it does not have all optimization phases of its competitor, it does not behaves too bad. I love its compilation speed and, with reasonable efforts, its code is understandable. My only concern is to see efforts is to have C++ based on tcc. Cfront was made on an age where C standard was K&R (no const, no prototypes ...). C++ compilers evolved a lot to be smart enough to compile complex C++ code to something efficient. Once again, I deeply respect the efforts to have a Cfront(tcc) working with minimal changes in Cfront source code 'just for the fun'. But, IMHO, it's a waste of time any attempt to do more. I bet the result will be used by nobody because poor performances and lack of support for more recent C++ features. M2c. Christian PS. Here are my own benchmarks made on my Lisp compiler using different C compilers, tcc lack of optimization makes the Lisp compiler SLOWER than interpreter!!! I don’t want to imagine the performance loss on a C++ benchmark using Cfront(tcc). However, I repeat I LOVE tcc TRUE C Compiler :o) GNU GCC 5.x (Mgw) Win32 (LAPtoCC) | 0.202 Microsoft C 18.0 Win32 (LAPtoCC) | 0.217 GNU GCC 5.x (Mgw) Win64 (LAPtoCC) | 0.237 Microsoft C 18.0 Win64 (LAPtoCC) | 0.237 Digital Mars 8.57 Win32 (LAPtoCC) | 0.248 Tiny C Compiler Win32 (LAPtoCC) | 0.339 Tiny C Compiler Win64 (LAPtoCC) | 0.382 GNU GCC 5.x (Mgw) Win32 (Compile) | 1.082 Watcom C 1.90 Win32 (Compile) | 1.142 Microsoft C 18.0 Win64 (Compile) | 1.216 Microsoft C 18.0 Win32 (Compile) | 1.253 Digital Mars 8.57 Win32 (Compile) | 1.353 LCC C Compiler Win32 (Compile) | 1.467 GNU GCC 5.x (Mgw) Win64 (Compile) | 1.526 IBM Visual Age C Win32 (Compile) | 1.580 Microsoft C 18.0 Win32 (32 bits) | 1.802 Microsoft C 18.0 Win32 (Unicode) | 1.819 Microsoft C 18.0 Win64 (64 bits) | 1.838 LCC C Compiler Win64 (Compile) | 1.866 Microsoft C 18.0 Win64 (Unicode) | 2.084 GNU GCC 5.x (Mgw) Win32 (32 bits) | 2.274 GNU GCC 5.x (Mgw) Win64 (64 bits) | 2.373 Watcom C 1.90 Win32 (32 bits) | 2.703 IBM Visual Age C Win32 (32 bits) | 2.859 Pelles C Compiler Win32 (32 bits) | 3.563 Digital Mars 8.57 Win32 (32 bits) | 3.630 LCC C Compiler Win32 (32 bits) | 3.763 Pelles C Compiler Win64 (64 bits) | 4.071 LCC C Compiler Win64 (64 bits) | 4.887 Tiny C Compiler Win32 (32 bits) | 5.899 Tiny C Compiler Win64 (64 bits) | 5.903 Tiny C Compiler Win32 (Compile) | 10.500 Tiny C Compiler Win64 (Compile) | 10.790 -----Original Message----- Hi Christian Jullien! > I'm really afraid to see discussion on about to optimize Cfront tcc backend! There is no such discussion. But Basile and you trying to say tcc is not a real compiler, a tcc can not be used to perform some task. Only because it is not a optimizer compiler. This is not true. We don't need an optimizer compiler all the time. Some words from the MELT build instructions: - 6 GByte of the RAM - 20 minutes for build Only because a C++ must be compiled > It has the reputation to be slower than C and, in some sense, it is > true when used by non-highly experimenter C++ programmers. (In real > life, C++ can outperform C > - I have code that run >2x as fast in C++ --thanks to templates-- than in C). I looked in OW STL some words about comparison a C++ way programming with plain C: in some cases C++ slower a plain C by 1.5 It is not a fail of the "highly expnon-erimented C++ programmers". But yes: C++ can be as fast as a plain C if you don't use C++ features or use only a templates w/o classes. PS: I can't fast find this article: a test was a C way list and iterators against a C++ list, iterators and so... _______________________________________________ Tinycc-devel mailing list |
| [Prev in Thread] | Current Thread | [Next in Thread] |