Sender:	root AT meer DOT meer DOT net
Message-ID:	<37F32789.CF843725@netfall.com>
Date:	Thu, 30 Sep 1999 02:04:09 -0700
From:	Andrew Sharp <andy AT netfall DOT com>
Organization:	Sharp Programmers
X-Mailer:	Mozilla 4.61 [en] (X11; U; Linux 2.2.7 i686)
X-Accept-Language:	en
MIME-Version:	1.0
To:	pgcc AT delorie DOT com
Subject:	2.95.1 fp slower than 2.91.6?
Reply-To:	pgcc AT delorie DOT com

Anyone have an idea on this?  See the results of the byte benchmark
below.  2.91.6 came from Epic 0.6, and 2.95.1 I got the old fashioned
way, I got the code and applied the patch.  Both run on the same
PentiumII-333.  Why does the FP related tests scores lag significantly
on the FP, while the integer and logic stuff shows a definite
improvement.  Both nbench binaries were built with the same make file,
hence the same options:

CFLAGS = -s -static -Wall -O6 -fomit-frame-pointer -funroll-loops
-malign-loops=2 -malign-functions=2 -malign-jumps=2 -mpentiumpro

Results:


BYTEmark* Native Mode Benchmark ver. 2 (10/95)
Index-split by Andrew D. Balsa (11/97)
Linux/Unix* port by Uwe F. Mayer (12/96,11/97)
TEST                : Iterations/sec.  : Old Index   : New Index
                    :                  : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT        :          76.289  :       1.96  :       0.64
STRING SORT         :          8.3736  :       3.74  :       0.58
v BITFIELD            :        3.29e+07  :       5.64  :       1.18
FP EMULATION        :          8.5614  :       4.11  :       0.95
FOURIER             :          1990.7  :       2.26  :       1.27
ASSIGNMENT          :          1.3477  :       5.13  :       1.33
IDEA                :          214.18  :       3.28  :       0.97
HUFFMAN             :          89.832  :       2.49  :       0.80
NEURAL NET          :          2.4354  :       3.91  :       1.65
LU DECOMPOSITION    :           80.92  :       4.19  :       3.03
==========================ORIGINAL BYTEMARK
RESULTS==========================
INTEGER INDEX       : 3.550
FLOATING-POINT INDEX: 3.336
Baseline (MSDOS*)   : Pentium* 90, 256 KB L2-cache, Watcom* compiler
10.0
==============================LINUX DATA
BELOW===============================
C compiler          : gcc version pgcc-2.91.60 19981201 (egcs-1.1.1
release)
libc                : ld-2.1.1.so
MEMORY INDEX        : 0.968
INTEGER INDEX       : 0.829
FLOATING-POINT INDEX: 1.850
Baseline (LINUX)    : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3,
libc-5.4.38
* Trademarks are property of their respective holder.

corwin:/tmp/nbench-byte-2.1# ./nbench,ppro
BYTEmark* Native Mode Benchmark ver. 2 (10/95)
Index-split by Andrew D. Balsa (11/97)
Linux/Unix* port by Uwe F. Mayer (12/96,11/97)

TEST                : Iterations/sec.  : Old Index   : New Index
                    :                  : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT        :          117.32  :       3.01  :       0.99
STRING SORT         :          8.3467  :       3.73  :       0.58
BITFIELD            :      3.2744e+07  :       5.62  :       1.17
FP EMULATION        :          8.5284  :       4.09  :       0.94
FOURIER             :          2140.6  :       2.43  :       1.37
ASSIGNMENT          :          1.5209  :       5.79  :       1.50
IDEA                :          281.45  :       4.30  :       1.28
HUFFMAN             :          106.92  :       2.96  :       0.95
NEURAL NET          :          2.1277  :       3.42  :       1.44
LU DECOMPOSITION    :          55.236  :       2.86  :       2.07
==========================ORIGINAL BYTEMARK
RESULTS==========================
INTEGER INDEX       : 4.087
FLOATING-POINT INDEX: 2.877
Baseline (MSDOS*)   : Pentium* 90, 256 KB L2-cache, Watcom* compiler
10.0
==============================LINUX DATA
BELOW===============================
C compiler          : gcc version pgcc-2.95.1 19990816 (release)
libc                : ld-2.0.7.so
MEMORY INDEX        : 1.006
INTEGER INDEX       : 1.031
FLOATING-POINT INDEX: 1.596
Baseline (LINUX)    : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3,
libc-5.4.38
* Trademarks are property of their respective holder.

- Raw text -