Re: floating point precision

From: Eyal Teler (et_at_nospam-et3d.com)
Date: 12/15/04

  • Next message: zak_at_secretlevel.com: "Re: floating point precision" 
    Date: Wed, 15 Dec 2004 08:22:45 +0200

    If you create the device with the D3DCREATE_FPU_PRESERVE flag, it will
    keep the FP state, instead of switching to single precision.

            Eyal

    zak@secretlevel.com wrote:
    > hello,
    >
    > i have recently discovered that d3d silently switches the floating point
    > precision to 24bits. this was giving me undue headaches!! i have my own
    > temporary solution of setting the floating point precision back to the
    > default 53bits and then restoring it after using _controlfp(...) function. i
    > am curious as to why msft does this and if it must be this way. either way i
    > think it is very important that you document this somewhere public (like in
    > the sdk documentation!!! ).
    >
    > thanks,
    > zak taylor
    > zak@secretlevel.com
    >

  • Next message: zak_at_secretlevel.com: "Re: floating point precision"

    Relevant Pages

    • Re: Best FPGA for floating point performance
      ... sure single precision is not enough but probably double precision is not ... Floating point add is a lot more expensive than fixed point. ... Glen, for this application, I'd argue that the floating point might be cheaper if he needs the dynamic range, especially if fixed point pushes him to wider than 35x35 multipliers. ... He may find that he can get away with a 17 bit significand with floating point, or at worst 4 multipliers for single precision. ...
      (comp.arch.fpga)
    • Re: Using a GPU for supercomputing?
      ... GPU, because the floating point is single-precision only, and HPC ... Exploiting Single Precision Math in HPC ...
      (comp.sys.ibm.pc.hardware.chips)
    • Re: Variable Precision Problem
      ... Most computers do it in binary, including floating point. ... In C, C++, and Java constants default to double precision unless ... That is, 0.1 is double, 0.1f is float. ... float (single precision) variable. ...
      (comp.lang.fortran)
    • Re: Spartns (fast and flexible tensor representation for Common Lisp)
      ... as a user if I give integer arguments to floating point ... functions I expect double precision result. ... I consider this as a bug in Ansi ... single precision is accurate enough for you. ...
      (comp.lang.lisp)
    • Re: Speed of floating point double vs single precision
      ... since the floating processor and all the library functions are double precision. ... > faster to do floating point arithmetic in double precision, since the hardware floating ... > units on PC's double or single precision these days, and based on this which is faster ...
      (microsoft.public.windowsxp.general)