大家都用什么CFLAGS?

zhuqin_83

顺便再说一下，native生成的包明显要比其他的小一点。

jarryson

按照你的结果来说确实，针对cpu优化的结果居然不如默认的。

不过我做这个测试的时候不知道为何，每次结果相差都很大，10s到8s的都有，甚至貌似和重复的间隔都有关系，继续关注中。

而且这个是取随机数的测试其实也不能代表全部结果。貌似都是那什么-fomit-frame-pointer的功劳

zhuqin_83

最终报告：

-O2 -march=pentium-m -pipe＝＝＝＝＝＝==============>9.3"
-O2 -march=pentium-m -pipe -fomit-frame-pointer========>9.4+"
-O2 -march=i686 -mtune=generic  -pipe===============>9.3"
-O2 -march=i686 -mtune=generic  -pipe -fomit-frame-pointer==>8.8"
-O2 -march=native -pipe========================>9.3"
-O2 -march=native -pipe -fomit-frame-pointer===========>8.8"

这样终于明朗化了，-fomit-frame-pointer可以起到关键的作用，但第二组的结果让我很意外。

结论：i686下，非多媒体程序（即采用sse指令的如mplayer, audacious等），除了上面所列的第二组是例外，源里所打的包与自己用所谓的-march最优化参数编译的并无区别，-fomit-frame-pointer参数在不少情况下会有提升，但不明显（%5左右），而该参数在x86_64下默认是开启的，也即是说，64位的源已经对绝大多数程序作出了优化，包括默认打开了sse2。据我所知，好像还没有一个程序支持sse2+的指令，用64位系统的同学们欢呼吧～不必羡慕什么gentoo榨干机器的油水，arch万岁！

希望看到64位系统下的测试，如果wiki所说属实，那么应该不会有什么区别。

fallening

用64位的测试了一下：
$ g++ -o test test.cc ran.cc $CFLAGS
[feng@enlightenment test]$ ./test
It costs 3.1 second(s) to generate 100000000 random numbers.
Sum of them is 4.99974e+07
[feng@enlightenment test]$ g++ -o test test.cc ran.cc -O2 -march=nocona -pipe -fomit-frame-pointer -msse2 -mmmx -mfpmath=sse
[feng@enlightenment test]$ ./test
It costs 3.07 second(s) to generate 100000000 random numbers.
Sum of them is 5.00028e+07
[feng@enlightenment test]$ g++ -o test test.cc ran.cc -O3 -march=nocona -pipe -fomit-frame-pointer -msse2 -mmmx -mfpmath=sse
[feng@enlightenment test]$ ./test
It costs 3.08 second(s) to generate 100000000 random numbers.
Sum of them is 5.00023e+07
[feng@enlightenment test]$ g++ -o test test.cc ran.cc -O3 -march=native -pipe -fomit-frame-pointer -msse2 -mmmx -mfpmath=sse
[feng@enlightenment test]$ ./test
It costs 3.08 second(s) to generate 100000000 random numbers.
Sum of them is 5.00038e+07
[feng@enlightenment test]$ g++ -o test test.cc ran.cc -O2 -march=native -pipe -fomit-frame-pointer -msse2 -mmmx -mfpmath=sse
[feng@enlightenment test]$ ./test
It costs 3.09 second(s) to generate 100000000 random numbers.
Sum of them is 5.00012e+07
[feng@enlightenment test]$ g++ -o test test.cc ran.cc -O2 -march=native -pipe -fomit-frame-pointer
[feng@enlightenment test]$ ./test
It costs 3.08 second(s) to generate 100000000 random numbers.
Sum of them is 4.99989e+07
[feng@enlightenment test]$ g++ -o test test.cc ran.cc -O3 -march=native -pipe -fomit-frame-pointer
[feng@enlightenment test]$ ./test
It costs 3.09 second(s) to generate 100000000 random numbers.
Sum of them is 4.99966e+07
[feng@enlightenment test]$ g++ -o test test.cc ran.cc
[feng@enlightenment test]$ ./test
It costs 4.87 second(s) to generate 100000000 random numbers.
Sum of them is 4.99989e+07
[feng@enlightenment test]$ icpc -o test test.cc ran.cc
ran.cc(36): warning #279: controlling expression is constant
        if("Random Number Generation: Impossible Thing Happened! EXIT...\n");
           ^
[feng@enlightenment test]$ ./test
It costs 4.24 second(s) to generate 100000000 random numbers.
Sum of them is 5.00023e+07
[feng@enlightenment test]$ icpc -o test test.cc ran.cc -fast
ran.cc(36): warning #279: controlling expression is constant
        if("Random Number Generation: Impossible Thing Happened! EXIT...\n");
           ^

ipo: remark #11000: performing multi-file optimizations
ipo: remark #11005: generating object file /tmp/ipo_icpcbu3gnn.o
[feng@enlightenment test]$ ./test
It costs 3 second(s) to generate 100000000 random numbers.
Sum of them is 5.00055e+07

xiaorui

不行的话。可以使用 time makepkg
看看每次的参数改变所用的时间。

ashadow

CFLAGS="-march=k8-sse3 -O2 -pipe -fomit-frame-pointer"

hejux

Post by tpxc;1819238
CFLAGS="-march=k8 -O3 -pipe -fomit-frame-pointer -msse3"

我的。

GCC 4.3 可以用k8-sse3这个集合了

ashadow

Post by hejux;1892164
GCC 4.3 可以用k8-sse3这个集合了

觉得还是直接native算了……

firefoxmmx

-match=core2

crazyatom

CHOST="i686-pc-linux-gnu"
CFLAGS="-march=core2 -mtune=core2 -O2 -pipe -mssse3 -fomit-frame-pointer"

cpu: core2 duo t5870