X-pop3-spooler:	POP3MAIL 2.1.0 b 4 980420 -bs-
Message-ID:	<35CDB2C6.F3D49819@kali.com.cn>
Date:	Sun, 09 Aug 1998 22:31:34 +0800
From:	Pan Xing <panxin AT kali DOT com DOT cn>
X-Mailer:	Mozilla 4.05 [en] (X11; I; Linux 2.0.30 i686)
MIME-Version:	1.0
To:	beastium-list AT Desk DOT nl
Subject:	Pgcc Slower than Gcc???(Not in mailist, re me diretly, Thx)
Sender:	Marc Lehmann <pcg AT goof DOT com>
Status:	RO
X-Status:	A
Lines:	84

Hi! Every One!
    I just plan to do some ting computation intensive work , ( wave
simulation with PDE). A month ago, I have
tested the gcc 2.7.2.3 under linux 2.0.30 in a PII233 box.
Unfortunately, I found its performance is lower than
MS's VC++5.0,  it's about 63Mflops .vs. 70 Mflops. Tested with the
wellknown flops.c (1992 version)

    I hear that Pgcc can deliver about 2-30% more power than gcc
generally, So I tested it and gcc again this evening.
My box is Pii233,64M, Slackware34/linux2.0.30, libc5.44. Test program:
flops.c  Version 2.0,  18 Dec 1992

                                             by   Al Aburto        /*
   aburto AT nosc DOT mil      */
Some weird things happened:

1) The flops.c compiled by gcc 2.7.2.3  WILL provide different results
when runing under Xwindow's xterm
and under ascii mode(I have installed kde, will it bother? ).
RESULT under ASCII MODE:

   FLOPS C Program (Double Precision), V2.0 18 Dec 1992

   Module     Error        RunTime      MFLOPS
                            (usec)
     1      4.8490e-13      0.2787     50.2297
     2      8.1272e-16      0.1322     52.9576
     3      2.7316e-14      0.1612    105.4804
     4     -3.8270e-15      0.1721     87.1572
     5     -9.5745e-15      0.3926     73.8753
     6     -1.2826e-14      0.2848    101.8108
     7      1.2524e-10      0.4779     25.1116
     8     -4.3365e-14      0.2938    102.1187

RESULT under Xterm

   FLOPS C Program (Double Precision), V2.0 18 Dec 1992

   Module     Error        RunTime      MFLOPS
                            (usec)
     1     -5.4193e-13      0.1605     87.2201
     2      8.5760e-16      0.1334     52.4795
     3      3.4567e-14      0.1615    105.2469
     4      3.6970e-13      0.1737     86.3719
     5     -5.1910e-15      0.3943     73.5532
     6      2.3930e-14      0.2932     98.9196
     7     -1.6524e-10      0.5447     22.0309
     8      1.4631e-13      0.3033     98.9282

 It seems that the result under Xterm is faster but the one under Ascii
mode is more accurate. Why, I guess
In Intel's chip only its internal parallism may affect its difference?
Does any one who have meet similar situation
or have experience with Intel 's Chip?  Thanks in advance.!

2) The Pgcc is Rather Slower than Gcc!?

        a)I was surprised by the results. But it seems true. I use the
pgcc-2.90.29 980515 (egcs-1.0.3 release),
        the result Pgcc .vs. Gcc is about 40M flops .vs. 60M flops. My
pgcc doomed my machine!
        I read the Pgcc faq provided by http://www.gcc.ml.org, use all
switches to accelarate which includ -O6,
        --funroll-all-loops -malign-double -mstack-align-double
-marg-align-double -mpentium,
        All could not save me from the bad luck.
        I installed the pgcc as the readme and install reference in the
package instructed me . ( patch, configure, make, make install under
/usr/local/), What's wrong with it?  I also got the comiled binary of
pgcc2.90, no help.

        b) Some good news to be reported: under the pgcc the weird
problem in 1) dispeared.
        the flops.c compiled by pgcc give identical results when running
under X and Ascii mode.


I know these questions may bother you. I am just new to gcc and come to
the maillist first time.

        Oh! I am not hacker but waste so much time. God save me! Where
is the devil?

- Raw text -