On Jun 8, 7:18 am, Mario Zechner <badlogicga...@gmail.com> wrote: > If you work in Java it does not pay of using fixed point math compared > to using floating point math. For some reasons integer divisions are > extremely costly and at the end of the day you will need to get the > non-fractional part of the fixed point number which involves a > division (or right shift). If you use fixed point math in native code > via the NDK you might see a bit of a speed up on older devices. If you > stay in Java land i strongly suggest to staying with floating point > math.
Two reasons for fixed-point difficulties: (1) there is no hardware integer divide instruction on ARM. They included it in ARMv7-R and ARMv7-M, but that doesn't help for mobile devices. (2) You really want a 32x32 -> 64-bit multiply, with a shift to reduce it back to 32. There is no such operation in Java, so you have to convert both args to 64-bit, do a 64x64 -> 64 multiply, and then shift down. The instruction overhead kills you. With the JIT you may find that the balance shifts, and you actually get a good speed boost with fixed-point on hardware without an FPU. However, given the way the handset manufacturers seem to be going, sticking with floating point in your Java code is probably a good strategy. -- You received this message because you are subscribed to the Google Groups "Android Developers" group. To post to this group, send email to android-developers@googlegroups.com To unsubscribe from this group, send email to android-developers+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/android-developers?hl=en
