Polyhedron shootout between g95 and gfortran

[Steven Bosscher <stevenb dot gcc at gmail dot com>]

Hi folks,

I did a Polyhedron benchmark with g95 and gfortran as compilers (see
www.polyhedron.com but I guess most of you know about it anyway).  The
result summaries are attached.  It is not a full run, only a quick
comparison on a Nocona box (Prescott-T 3.2GHz, 1GB RAM), which means
that each benchmark runs only once.

The benchmark was run with "../pbharness g95_lin_p4.par quick.par"

Executive summary:

G95 (GCC 4.0.1 (g95!) Feb 22 2006)
vs.
GNU Fortran 95 (GCC) 4.1.0 20060222 (prerelease)

g95 Geometric Mean Execution Time =      27.19 seconds
gfortran Geometric Mean Execution Time =      34.54 seconds

Benchmark	Compile time (secs)	Average runtime (secs)
Name		g95	gfortran	g95	gfortran
ac		1.18	1.46		26.03	18.63		
aermod		61.13	57.60		56.04	52.39
air		2.92	4.80		19.36	18.90
capacita	1.52	1.44		98.35	83.32
channel		0.87	0.70		25.26	10.27
doduc		7.09	5.34		68.59	62.33
fatigue		2.39	3.38		47.43	31.15
gas_dyn		1.91	2.63		30.70	22.46
induct		5.85	6.61		60.66	275.69
linpk		0.62	0.64		22.21	22.34
mdbx		2.14	1.75		27.37	24.59
nf		1.26	1.38		43.51	51.47
protein		3.36	4.12		88.06	64.86
rnflow		4.00	3.88		70.80	78.36
test_fpu	3.36	3.04		26.76	31.83
tfft		0.54	0.52		0.05	4.74

(It looks like tfft didn't work with g95 at all???)


Two things I don't understand:

1) gfortran has one aermod test failing:
> Value= 2191.1142600     Target= 2191.1145000     Tolerance=0.10000000000E-02
> Value= 34310.421880     Target= 34310.421880     Tolerance=0.30000000000E-01
> Value= 4260.9570300     Target= 4260.9716800     Tolerance=0.50000000000E-01
> Value= 37094.347660     Target= 37094.375000     Tolerance=0.20000000000E-01
FAIL <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
> Value= 7924.8808600     Target= 7924.8842800     Tolerance=0.30000000000E-01
> Value= 37094.347660     Target= 37094.375000     Tolerance= 1.0000000000

aermod FAILED    1 fails and    5 passes

Does anyone know why this happens or what is failing?


2) gfortran is _awful_ for induct... Does anyone know why?

Gr.
Steven

Attachment: g95_lin_p4.sum
Description: Text document

Attachment: gfortran_lin_p4.sum
Description: Text document