Re: gcc 2.7.1

Pascal A. Dupuis (dupuis@lei.ucl.ac.be)
Thu, 6 Aug 1998 09:46:36 +0200 (CEST)


In digest #2336, "Etienne Lorrain" <lorrain@fb.sony.de> suggested:

> Maybe it is time to upgrade to gcc-2.7.3, so that this patch could
> be applied in the main stream :
>
>- -linux-old/arch/i386/Makefile
>+linux/arch/i386/Makefile
>
> CFLAGS_PIPE := -pipe
>- -CFLAGS_NSR := -fno-strength-reduce
>+CFLAGS_NSR :=
> CFLAGS := $(CFLAGS) $(CFLAGS_PIPE) $(CFLAGS_NSR)

Could some i386 gourou have a look on this ? I think that without
-fno-strength-reduce, gcc replaces multiplication in loop index
computation by addition,using more registers. As the number of
registers is low on i386, this lead to worse performance, as the number of
registers saving/restoring increases. I have compiled 2.0 kernels with and
without this option, and find that -fno-strength-reduce gives faster
kernels (no test done). In the same vein, what about
-fomit-frame-pointer ? You have one more register to use, but have to
restore the stack frame after each function return. Benchmarks, anyone ?

Pascal A. Dupuis

-- 
LinuxScan report : Win95 virus found on boot sector of primary partititon.
Severity : medium, it could halt your computer at random time, corrupt your
           datas and hard disk or try to be reinstalled every two days.
Disinfectant exists, apply (Yes/Yes/Yes) ?  

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html