> The current benchmarks are more comprehensive than my example: the main > reason I brought it up is that my perfectly horrible bit of code below is > still good enough to do what is needed. > > ... and it does it without having to identify processors or eliminate the > non-Intel world from consideration. > > -- Lynn
Ok, agreed that CPUID idea turn out to be more complex for development and support where many different platforms included, really thought only about x86 when wrote first comment on this topic. But if benchmark can be shortened and simplified, and, even much more important, if it can be done not on schedule but on event basis, it's worth to code IMO. Just one more unneeded overhead will gone. And please, take into consideration not only cost of benchmark by itself. MUCH more important that benchmark preemts and interrupts current task in processing. This overhead, especially on Multi-CPU, multy-GPU hosts will be MUCH more bigger than just benchmark itself. _______________________________________________ boinc_dev mailing list boinc_dev@ssl.berkeley.edu http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
