Cyril的网志

一些浮点运算可能出现一些意外状况，譬如同事使用的ceil:

g++ -O3  结果总是会有些出入，原因如下：

-ffloat-store

                Do not store floating point variables in registers, and inhibit other options that might change whether a floating point value is taken from a register or memory.

                This option prevents undesirable excess precision on machines such as the 68000 where the floating registers (of the 68881) keep more precision than a double is supposed to have. Similarly for the x86 architecture. For most programs, the excess precision does only good, but a few programs rely on the precise definition of IEEE floating point. Use -ffloat-store for such programs, after modifying them to store all pertinent intermediate computations into variables.

-ffloat-store

            Might help a Fortran program that depends on exact IEEE conformance on some machines, but might slow down a program that doesn't.

            This option is effective when the floating-point unit is set to work in IEEE 854 `extended precision'--as it typically is on x86 and m68k GNU systems--rather than IEEE 754 double precision. -ffloat-store tries to remove the extra precision by spilling data from floating-point registers into memory and this typically involves a big performance hit. However, it doesn't affect intermediate results, so that it is only partially effective. `Excess precision' is avoided in code like:

          a = b + c
          d = a * e
          

                d = (b + c) * e
          

For another, potentially better, way of controlling the precision, see Floating-point precision.

 但即便使用该参数也有可能会出问题:

Floating-point precision

If your program depends on exact IEEE 754 floating-point handling it may help on some systems--specifically x86 or m68k hardware--to use the -ffloat-store option or to reset the precision flag on the floating-point unit. See Optimize Options.

However, it might be better simply to put the FPU into double precision mode and not take the performance hit of -ffloat-store. On x86 and m68k GNU systems you can do this with a technique similar to that for turning on floating-point exceptions (see Floating-point Exception Handling). The control word could be set to double precision by some code like this one:

     #include <fpu_control.h>
     {
       fpu_control_t cw = (_FPU_DEFAULT & ~_FPU_EXTENDED) | _FPU_DOUBLE;
       _FPU_SETCW(cw);
     }