On Apr 22, 8:02 am, BobG <[email protected]> wrote: > [...] I timed a bunch of floating > point adds and multiplies in a loop, and I was getting about a million > per sec (dev phone 2, 528mhz cpu?). (I see the Linpack app in the > market gives the same result) This is many times faster than sw > floating point on a 20mhz AVR, but I bet its way slower than hw fp. > Does the fp multiply in java multiply 2 32 bit floats, or promote them > to doubles like c does?
Java and Dalvik bytecode have independent mul-float and mul-double instructions. On the VFP-enabled hardware I've had my hands on, the performance difference is negligible. On devices with soft float, it's more substantial. The VM will use VFP if present. (This is one of the basic arguments in favor of just-in-time vs. static compilation -- if you did this in native code you'd have to provide two versions.) The core of the interpreter is hand-coded ARM assembly, so on VFP devices it would actually be slower to call out to a library. We're always interested in good JIT compiler benchmarks. If you have an APK or stand-alone executable that does a bunch of computationally intensive stuff and spits out numbers, put it up somewhere and we'll poke at it. I don't know if we're allowed to talk about performance numbers yet (the JIT hasn't shipped yet), but we can hand-wave something. :-) -- You received this message because you are subscribed to the Google Groups "Android Developers" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/android-developers?hl=en
