Hi,
There are some huge compile time regressions between 16/7 and 17/7, most
likely due to the IPA work from Kenny and Dan. These are the build times
in seconds, taken from Diego Novillo's nightly SPEC tester box:
http://people.redhat.com/dnovillo/spec2000/gcc/log/20050716/spec-20050716.stats
http://people.redhat.com/dnovillo/spec2000/gcc/log/20050717/spec-20050717.stats
Sadly, the SPEC box only measures rounded numbers for the build times, so
the places where there is no measured slowdown may actually have slowed
down as well, but there is no way to tell from this data.
CFP2000 BASE
168.wupwise 2 2 100.00%
171.swim 1 1 100.00%
172.mgrid 1 1 100.00%
173.applu 3 3 100.00%
177.mesa 31 40 129.03%
178.galgel 13 13 100.00%
179.art 1 1 100.00%
183.equake 2 1 50.00%
187.facerec 4 4 100.00%
188.ammp 9 10 111.11%
189.lucas 3 3 100.00%
191.fma3d 80 83 103.75%
200.sixtrack 70 70 100.00%
301.apsi 7 7 100.00%
TOTAL 227 239 105.29%
CFP2000 PEAK
168.wupwise 3 3 100.00%
171.swim 0 0 ???
172.mgrid 1 1 100.00%
173.applu 5 4 80.00%
177.mesa 37 37 100.00%
178.galgel 13 13 100.00%
179.art 1 1 100.00%
183.equake 2 2 100.00%
187.facerec 4 4 100.00%
188.ammp 11 11 100.00%
189.lucas 3 4 133.33%
191.fma3d 130 133 102.31%
200.sixtrack 74 258 348.65%
301.apsi 7 7 100.00%
TOTAL 291 478 164.26%
CINT2000 BASE
164.gzip 4 4 100.00%
175.vpr 9 9 100.00%
176.gcc 83 83 100.00%
181.mcf 2 2 100.00%
186.crafty 16 16 100.00%
197.parser 7 7 100.00%
252.eon 101 105 103.96%
253.perlbmk 38 38 100.00%
254.gap 32 32 100.00%
255.vortex 31 31 100.00%
256.bzip2 2 2 100.00%
300.twolf 20 20 100.00%
TOTAL 345 349 101.16%
CINT2000 PEAK
164.gzip 4 4 100.00%
175.vpr 12 11 91.67%
176.gcc 105 127 120.95%
181.mcf 2 2 100.00%
186.crafty 17 17 100.00%
197.parser 11 19 172.73%
252.eon 105 109 103.81%
253.perlbmk 45 54 120.00%
254.gap 37 38 102.70%
255.vortex 32 32 100.00%
256.bzip2 3 3 100.00%
300.twolf 21 21 100.00%
TOTAL 394 437 110.91%
Note especially the sixtrack and parser increases. I have not looked
at the SPEC scores, but eon miscompares now.
Gr.
Steven
