uint64_t => float/double/long double

[Frank Klemm <pfk at fuchs dot offl dot uni-jena dot de>]

C code:

----------------------------------------------------------
long double         x;
unsigned long long  y;

int  main ( void )
{
    x = y;
    return 0;
}
----------------------------------------------------------

Generated code with:

    gcc-3.0.2-20010917 -S -O4 -fomit-frame-pointer -mpreferred-stack-boundary=2 example.c

----------------------------------------------------------


.LC0:	.long   0x0,0x80000000,0x403f

        .text
.global main
.type   main,@function

main:
        movl    y+4, %ecx
        movl    y, %eax
        pushl   %ecx
        pushl   %eax
        fildll  (%esp)
        addl    $8, %esp
        testl   %ecx, %ecx
        js      .L3
.L2:    fstpt   x
        xorl    %eax, %eax
        ret
.L3:
        fldt    .LC0			<--------------
        faddp   %st, %st(1)		<--------------
        jmp     .L2

----------------------------------------------------------

The last three lines can be changed to:

----------------------------------------------------------

	fadds	.LC1
	jmp	.L2

.LC1:	.long   0x5f800000	## (float) 2^64

----------------------------------------------------------


You do not need a long double to store the value of 2^64.
This change should be done very easily.
Who is responsible ???

===========================================================

Other optimizations are to add several constants to libm.so
and use these from this library. Such constants are:

   (float) 2**64
   (float) 0.5
   (float) 2.
   (float) 3.
   (double) 1/3.
   (long double) 1/3.L
   (double) 2.*M_PI
   (long double) 2.L*M_PIl
   (double) 0.1
   (long double) 0.1
   (float) 10.

Several library functions also need these values.

If such values are in libm.so, you can use the following code:

----------------------------------------------------------

	fadds	__twopower64
	jmp	.L2

----------------------------------------------------------

-- 
Frank Klemm