[Embench] Question on soft-float versus hard-float

Brian Grayson brian.grayson at sifive.com
Wed Jun 17 20:47:39 CEST 2020


The benchmark is intended for implementations without hardware floating
point, but what is the proper thing to do on an implementation that has
hardware floating point?

One thing I've seen is, the benchmarks with a lot of FP, like nbody and
minver, end up running very fast on machines with hardware FP, and are very
susceptible to minor variances in timing.

I can adjust for this on a per-benchmark basis by changing
LOCAL_SCALE_FACTOR (as changing CPU_MHZ would affect all benchmarks, not
just the FP ones), but then the relative scoring report is incorrect,
because my LOCAL_SCALE_FACTOR is different from what the baseline run used.
(In other words, if one doubles LOCAL_SCALE_FACTOR, one halves the obtained
relative score compared to the baseline.)

Thanks.

Brian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.librecores.org/pipermail/embench/attachments/20200617/fe28467d/attachment.htm>


More information about the Embench mailing list