fp Question

Hi,

I ran Spec2006 with -O4. All integer benchmarks passed, but only 8 out 17 of floating point benchmarks passed. Is this normal or I made a mistake in my build?

Reza

Hi Reza. Somebody on Linux should answer, but I don't think it's normal. You may have checked out the source at a moment when it had a bug in it; that unfortunately happens frequently. Which ones are failing?

Estimated Estimated
Base Base Base Peak Peak Peak
Benchmarks Ref. Run Time Ratio Ref. Run Time Ratio


410.bwaves – 0.0194 RE
416.gamess – 0.00135 RE
433.milc – 16.5 – S
434.zeusmp – 0.00146 RE
435.gromacs – 0.00138 RE
436.cactusADM – 0.00135 RE
437.leslie3d – 0.00141 RE
444.namd – 19.4 – S
447.dealII – 19.7 – S
450.soplex – 0.0380 – S
453.povray – 2.49 – S
454.calculix – 0.00135 RE
459.GemsFDTD – 0.00150 RE
465.tonto – 2.40 – S
470.lbm – 6.52 – S
481.wrf – 0.00140 RE
482.sphinx3 – 3.19 – S

Following is the list of fp benchmarks that fail. They all pass with -O3, but some fail with -O4. I did the test run.

Thanks,
Reza

Estimated Estimated
Base Base Base Peak Peak Peak
Benchmarks Ref. Run Time Ratio Ref. Run Time Ratio


410.bwaves – 0.0194 RE
416.gamess – 0.00135 RE
433.milc – 16.5 – S
434.zeusmp – 0.00146 RE
435.gromacs – 0.00138 RE
436.cactusADM – 0.00135 RE
437.leslie3d – 0.00141 RE
444.namd – 19.4 – S
447.dealII – 19.7 – S
450.soplex – 0.0380 – S
453.povray – 2.49 – S
454.calculix – 0.00135 RE
459.GemsFDTD – 0.00150 RE
465.tonto – 2.40 – S
470.lbm – 6.52 – S
481.wrf – 0.00140 RE
482.sphinx3 – 3.19 – S

433.milc, 447.deall, and 450.soplex are run as part of the nightly tests on Darwin (=MacOSX) in a way that closely approximates -O4. They are working there. I am inclined to suspect gold given that they work at -O3. Can a Linux person comment?

I haven't tried those benchmarks with gold yet, sorry.

Would be more than happy to look at a bug if you can reduce it.

Cheers,

433.milc, 447.deall, and 450.soplex are run as part of the nightly tests
on Darwin (=MacOSX) in a way that closely approximates -O4. They are
working there. I am inclined to suspect gold given that they work at
-O3. Can a Linux person comment?

I'm currently working on a miscompilation of 450.soplex by dragonegg
at -O2. Kenneth Hoste saw several SPEC2006 llvm-gcc miscompilations
when he varied the set of optimization passes run, which we will be
working on together to try to resolve.

Ciao,

Duncan.