This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
SPEC95 results for the DFA scheduler branch on SPARC
- From: Dan Nicolaescu <dann at godzilla dot ICS dot UCI dot EDU>
- To: gcc at gcc dot gnu dot org
- Cc: Vladimir Makarov <vmakarov at redhat dot com>, "David S. Miller" <davem at redhat dot com>
- Date: Fri, 16 Nov 2001 09:37:16 -0800
- Subject: SPEC95 results for the DFA scheduler branch on SPARC
Hi!
I have run SPEC95 comparing gcc on the DFA branch before and after
David Miller changed the sparc pipeline description to be dfa based.
(ie the only change in the compiler between the runs was the SPARC
pipeline description)
In the following tables the "base" configuration was compiled with gcc
the gcc before dfa and the "peak" with the dfa gcc.
In both cases the flags used were: -O2 -mcpu=v8 -mtune=ultrasparc
The machine is a Sun Blade 100 with a 500MHz UltraSPARCIIe, and 1GB
RAM running solaris 2.8
Each benchmark was run 3 times for all configurations.
Disclaimer: I've _only_ run the benchmarks, all the credit for the
work goes to Vladimir Makarov and David Miller.
Also David said that the DFA SPARC pipeline to is still not tuned,
it can still be improved.
The results are good, all the "peak" results are slightly better than
the "base" results.
So the DFA scheduler is in a pretty good shape.
SpecINT:
(the 126.gcc benchmark from specint does not work in any
configuration)
SPEC Benchmark CINT95 Summary
Base Base Base Peak Peak Peak
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
------------ -------- -------- -------- -------- -------- --------
099.go 4600 277 16.6 4600 275 16.7
099.go 4600 279 16.5 4600 272 16.9
099.go 4600 278 16.6* 4600 273 16.8*
124.m88ksim 1900 155 12.2* 1900 152 12.5
124.m88ksim 1900 155 12.3 1900 152 12.5
124.m88ksim 1900 157 12.1 1900 152 12.5*
126.gcc 1700 114 --X 1700 115 --X
126.gcc 1700 114 --X 1700 116 --X
126.gcc 1700 114 --X 1700 114 --X
129.compress 1800 172 10.5 1800 171 10.5
129.compress 1800 172 10.5* 1800 171 10.5*
129.compress 1800 173 10.4 1800 172 10.5
130.li 1900 174 10.9 1900 174 10.9
130.li 1900 174 10.9* 1900 173 11.0
130.li 1900 174 10.9 1900 174 10.9*
132.ijpeg 2400 157 15.3 2400 153 15.7*
132.ijpeg 2400 157 15.3 2400 153 15.7
132.ijpeg 2400 157 15.3* 2400 153 15.7
134.perl 1900 122 15.6 1900 124 15.3
134.perl 1900 122 15.6* 1900 121 15.7*
134.perl 1900 122 15.6 1900 121 15.7
147.vortex 2700 220 12.3* 2700 217 12.4*
147.vortex 2700 220 12.3 2700 217 12.4
147.vortex 2700 222 12.2 2700 218 12.4
========================================================================
099.go 4600 278 16.6 4600 273 16.8
124.m88ksim 1900 155 12.2 1900 152 12.5
126.gcc 1700 -- -- 1700 -- --
129.compress 1800 172 10.5 1800 171 10.5
130.li 1900 174 10.9 1900 174 10.9
132.ijpeg 2400 157 15.3 2400 153 15.7
134.perl 1900 122 15.6 1900 121 15.7
147.vortex 2700 220 12.3 2700 217 12.4
Est. SPECint_base95 (Geom. Mean) --
Est. SPECint95 (Geom. Mean) --
SPEC Benchmark CFP95 Summary
Base Base Base Peak Peak Peak
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
------------ -------- -------- -------- -------- -------- --------
101.tomcatv 3700 293 12.6 * 3700 279 13.3
101.tomcatv 3700 293 12.6 3700 277 13.4
101.tomcatv 3700 293 12.6 3700 277 13.4 *
102.swim 8600 386 22.3 8600 379 22.7 *
102.swim 8600 384 22.4 8600 379 22.7
102.swim 8600 384 22.4 * 8600 381 22.6
103.su2cor 1400 235 5.95* 1400 232 6.03
103.su2cor 1400 235 5.95 1400 232 6.04
103.su2cor 1400 237 5.90 1400 232 6.04*
104.hydro2d 2400 397 6.04 2400 397 6.05
104.hydro2d 2400 397 6.04 2400 397 6.05*
104.hydro2d 2400 397 6.04* 2400 399 6.02
107.mgrid 2500 323 7.73* 2500 289 8.65*
107.mgrid 2500 325 7.69 2500 289 8.65
107.mgrid 2500 323 7.74 2500 289 8.64
110.applu 2200 292 7.52 2200 288 7.65*
110.applu 2200 292 7.52* 2200 287 7.67
110.applu 2200 292 7.53 2200 289 7.60
125.turb3d 4100 299 13.7 4100 296 13.8
125.turb3d 4100 299 13.7 * 4100 296 13.8 *
125.turb3d 4100 300 13.7 4100 297 13.8
141.apsi 2100 192 11.0 2100 191 11.0
141.apsi 2100 194 10.8 2100 191 11.0
141.apsi 2100 192 10.9 * 2100 191 11.0 *
145.fpppp 9600 881 10.9 9600 845 11.4
145.fpppp 9600 869 11.1 9600 838 11.4 *
145.fpppp 9600 874 11.0 * 9600 838 11.5
146.wave5 3000 204 14.7 * 3000 200 15.0
146.wave5 3000 203 14.8 3000 198 15.2
146.wave5 3000 205 14.7 3000 200 15.0 *
=======================================================================
101.tomcatv 3700 293 12.6 3700 277 13.4
102.swim 8600 384 22.4 8600 379 22.7
103.su2cor 1400 235 5.95 1400 232 6.04
104.hydro2d 2400 397 6.04 2400 397 6.05
107.mgrid 2500 323 7.73 2500 289 8.65
110.applu 2200 292 7.52 2200 288 7.65
125.turb3d 4100 299 13.7 4100 296 13.8
141.apsi 2100 192 10.9 2100 191 11.0
145.fpppp 9600 874 11.0 9600 838 11.4
146.wave5 3000 204 14.7 3000 200 15.0
SPECfp_base95 (Geom. Mean) 10.4
SPECfp95 (Geom. Mean)
These tests were done sometime at the end of September.
In mid-October a merge from the mainline to the DFA branch was
performed. The SPECint results after the merge are also slightly
better (the results bellow are for the DFA based scheduler)
before the merge after the merge
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
------------ -------- -------- -------- -------- -------- --------
099.go 4600 275 16.7 4600 277 16.6
099.go 4600 272 16.9 4600 273 16.8*
099.go 4600 273 16.8* 4600 272 16.9
124.m88ksim 1900 152 12.5 1900 151 12.6
124.m88ksim 1900 152 12.5 1900 151 12.6*
124.m88ksim 1900 152 12.5* 1900 152 12.5
126.gcc 1700 115 --X 1700 113 --X
126.gcc 1700 116 --X 1700 113 --X
126.gcc 1700 114 --X 1700 113 --X
129.compress 1800 171 10.5 1800 169 10.6*
129.compress 1800 171 10.5* 1800 169 10.6
129.compress 1800 172 10.5 1800 175 10.3
130.li 1900 174 10.9 1900 171 11.1
130.li 1900 173 11.0 1900 172 11.1*
130.li 1900 174 10.9* 1900 172 11.1
132.ijpeg 2400 153 15.7* 2400 150 16.0
132.ijpeg 2400 153 15.7 2400 150 16.0*
132.ijpeg 2400 153 15.7 2400 150 16.0
134.perl 1900 124 15.3 1900 119 15.9
134.perl 1900 121 15.7* 1900 119 15.9
134.perl 1900 121 15.7 1900 119 15.9*
147.vortex 2700 217 12.4* 2700 214 12.6*
147.vortex 2700 217 12.4 2700 216 12.5
147.vortex 2700 218 12.4 2700 214 12.6
=========================================== ==============================
099.go 4600 273 16.8 4600 273 16.8
124.m88ksim 1900 152 12.5 1900 151 12.6
126.gcc 1700 -- -- 1700 -- --
129.compress 1800 171 10.5 1800 169 10.6
130.li 1900 174 10.9 1900 172 11.1
132.ijpeg 2400 153 15.7 2400 150 16.0
134.perl 1900 121 15.7 1900 119 15.9
147.vortex 2700 217 12.4 2700 214 12.6
Est. SPECint_base95 (Geom. Mean) --
Est. SPECint95 (Geom. Mean) --