Absoft F77 for Linux (v3.4), optimized "-O" --- 211 seconds Portland Group (beta) F77, optimized "-O4" --- 215 seconds f2c/gcc (driver: fort77), optimized "-O4" --- 259 seconds brem90 Absoft F90 beta, optimized "-O" --- 155 seconds(!) brem90 Absoft F90 beta, optimized "-O -B100" 154 seconds(!) NOTE: -B100 is Pentium Pro optimization Absoft F77 for Linux (v4.4) optimized "-O" --- 197 seconds(!)Very very interesting that the F90 version is actually faster!
Peter sent a fairly detailed list of results, so this has been placed on its own subpage. Peter also sent in a bug fix for the brem package; one of the arrays was never initialized to zero, and the program returned garbage for systems which do not automatically initialize variables to zero at beginning of program.
Both tests are run on PowerPC model 43P, running AIX 4.2.1 both systems with 128MB RAM. The compiler is IBM XL fortran version 4 (level 4.1.0.6). optimization flags: -O3 -qarch=ppc -qtune=604 On a 200Mhz machine the results are: xlf : user 145.14, sys 0.20 xlf90: user 91.37, sys 0.14 on a 332 Mhz machine the program runs as: xlf: user 86.98, sys 0.27 xlf90: user 52.67, sys 0.18
Configuration: Pentium 200MHz, 64 MB SDRAM, 512K L2 cache FreeBSD-2.2.2 a) precompiled binaries (in Linux compatibility mode) Portland Group (beta) F77, optimized "-O4" --- 229 seconds f2c/gcc (driver: fort77), optimized "-O4" --- 269 seconds brem90 Absoft F90 beta, optimized "-O" --- 180 seconds Absoft F77 for Linux (v4.4) optimized "-O" --- 196 seconds b) native FreeBSD binaries f2c (f77reorder-2.27, f2c, gcc-2.7.2.3) (-O4) -- 328 seconds g77-0.5.19.1, gcc-2.7.2.1 (-O) --- 297 seconds g77/egcs-980205 (-O) --- 298 seconds g77/pgcc-1.0.1 (-O6 -mpentium) --- 292 seconds 2) on some machines of our Alpha cluster (f77 -O brem.f+gammln.f) Puget Alpha 500/533MHz 256 MB SDRAM --- 42.7 sec Digital Alpha 255/300MHz 192 MB DRAM --- 118.7 sec DEC 3000 AXP 400/175MHz 64 MB DRAM --- 270.8 sec
Configuration: Pentium II 233 MHz, 128 MB RAM, 512k cache, SCSI Linux 2.0.31 (RedHat 4.2) *Portland Group (1.6-4) F77, optimized "-O" --- 142 seconds Portland Group (beta) F77, optimized "-O4" --- 146 seconds Absoft F77 for Linux, optimized "-O" --- 127 seconds f2c/gcc, optimized "-O4" --- 181 seconds
xterm console Absoft binary: 1245s 1108s PG binary: 1366s F2c binary: 1527s g77-0.5.21: 1446s (1) 1260s (1) egcs-2.90.14: 1440s (1) 1266s (1)Remarks from Mr. Taeschner.
Absoft F77 for Linux (v3.4), optimized "-O" --- 171 seconds Portland Group (beta) F77, optimized "-O4" --- 204 seconds g77 --- 232 seconds f2c/gcc (driver: fort77), optimized "-O4" --- n/a
Absoft F77 for Linux (v3.4), optimized "-O" --- 443 seconds Portland Group (beta) F77, optimized "-O4" --- 483 g77 --- 549 seconds f2c/gcc (driver: fort77), optimized "-O4" --- 581
Absoft F77 for Linux 171 sec Portland Group 198 sec SNI F77 (Siemens) 210 sec (new) g77-0-5-18 249 sec
f2c/gcc (driver: fort77) optimized "-O4" 249 sec Absoft f77 (v3.4) optimized "-O" 170 sec Absoft f77 (v4.4) optimized -O -B100 193 sec Absoft F90 optimized -O -B100 154 sec egcs-1.0.1 (g77) optimized -O 259 sec More results for Linux comparisons on Alpha: Alpha 21164-600MHz box with 1GB of RAM g77 -O3 -funroll-loops -ffast-math (Linux/Alpha) 145 sec f77 -O3 41 sec (DEC Fortran v3.6, statically linked on an OSF/1 machine, and then run on the Linux/Alpha box above)
SparcUltra 170 (Sun Fortran 3.0.1, -O3) --- 162 secondsFrom Georgy Jikia (Universitaet Freiburg)
DEC f77 -O: 53.92s user 0.17s system 98% cpu 54.681 total GNU g77-O: 68.72s user 0.13s system 93% cpu 1:13.78 total DEC f90 -O: 17.02s user 0.13s system 99% cpu 17.293 total