The stock app does accuracy checks on each function and won't use anything which fails those. But they're brief tests which can't catch all possible deviations, so I think this effort is a good idea.
The run on my Pentium-M has two sets of PGxxxx_v7 results from the stock app. The ref results used a -def argument (abbreviation for the " -default_functions use the safe unoptimized default functions"), and those have been slightly renamed to start ref-def- so they can be distinguished from runs where function choosing was active.
The run on Coppermine P3 is just a single set of PGxxxx_v7 results. Checking those against the ref-def- set gives this:
rescmpv5
ref-setiathome_6.98_windows_intelx86.exe-PG0009_v7.wu.res
ref-def-setiathome_6.98_windows_intelx86.exe-PG0009_v7.wu.res
Result : Strongly similar, Q= 99.53%
rescmpv5
ref-setiathome_6.98_windows_intelx86.exe-PG0395_v7.wu.res
ref-def-setiathome_6.98_windows_intelx86.exe-PG0395_v7.wu.res
Result : Strongly similar, Q= 99.88%
rescmpv5
ref-setiathome_6.98_windows_intelx86.exe-PG0444_v7.wu.res
ref-def-setiathome_6.98_windows_intelx86.exe-PG0444_v7.wu.res
Result : Strongly similar, Q= 99.78%
rescmpv5
ref-setiathome_6.98_windows_intelx86.exe-PG1327_v7.wu.res
ref-def-setiathome_6.98_windows_intelx86.exe-PG1327_v7.wu.res
Result : Strongly similar, Q= 99.29%
Joe