Date: Sat, 6 May 2000 14:48:57 -0700 (PDT) From: Jean-Marc Zucconi <jmz@FreeBSD.org> To: dnelson@emsphone.com Cc: current@FreeBSD.ORG Subject: Re: Can someone explain this? Message-ID: <200005062148.OAA18135@freefall.freebsd.org> In-Reply-To: <20000506002203.A6363@dan.emsphone.com> (message from Dan Nelson on Sat, 6 May 2000 00:22:03 -0500) References: <200005060030.RAA11795@freefall.freebsd.org> <20000506002203.A6363@dan.emsphone.com>
next in thread | previous in thread | raw e-mail | index | archive | help
>>>>> Dan Nelson writes:
> In the last episode (May 05), Jean-Marc Zucconi said:
>> Here is something I don't understand:
>>
>> $ sh -c '/usr/bin/time ./a.out'
>> 2.40 real 2.38 user 0.01 sys
>> $ /usr/bin/time ./a.out
>> 7.19 real 7.19 user 0.00 sys
>>
>> The same program is 3 times slower in the second case. The effect is
>> systematic but depends on the program being run. I have seen inverse
>> behavior with another program. Using time -l, I note that this seems
>> to be related with a higher value of 'involuntary context switches'
>> (3 times more switches in the slower case).
> It has to do with your stack. Calling the program via /bin/sh sets up
> your environment differently, so your program's stack starts at a
> different place. Try running this:
> main (int argc, char **argv)
> {
> int i;
> double x=2, y=2, z=2;
> printf ("%p\n",&i);
> for (i = 0; i < 10000000; i++) z = y*x;
> return 0;
> }
> Run this commandline:
> STR= ; export STR ; while : ; do ; STR=z$STR ; /usr/bin/time ./a,out ; done
> And watch your execution time flip flop every 4 runs.
OK. The effect is indeed very clear.
> Here are some bits from the gcc infopage explaining your options if you
> want consistant speed from programs using doubles:
> `-mpreferred-stack-boundary=NUM'
> Attempt to keep the stack boundary aligned to a 2 raised to NUM
> byte boundary. If `-mpreferred-stack-boundary' is not specified,
> the default is 4 (16 bytes or 128 bits).
> The stack is required to be aligned on a 4 byte boundary. On
> Pentium and PentiumPro, `double' and `long double' values should be
> aligned to an 8 byte boundary (see `-malign-double') or suffer
> significant run time performance penalties. On Pentium III, the
> Streaming SIMD Extention (SSE) data type `__m128' suffers similar
> penalties if it is not 16 byte aligned.
> `-mno-align-double'
> Control whether GCC aligns `double', `long double', and `long
> long' variables on a two word boundary or a one word boundary.
> Aligning `double' variables on a two word boundary will produce
> code that runs somewhat faster on a `Pentium' at the expense of
> more memory.
> *Warning:* if you use the `-malign-double' switch, structures
> containing the above types will be aligned differently than the
> published application binary interface specifications for the 386.
Now the problem is that the -mpreferred-stack-boundary=NUM option does
not solve the problem :-( I still get a penalty in 50% of the cases.
Jean-Marc
--
Jean-Marc Zucconi PGP Key: finger jmz@FreeBSD.org
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200005062148.OAA18135>