David Chase wrote:

<snipped>
 
> If you compile with the "-S" flag, you can look at the
> generated assembly language, and maybe see what happened
> there.

Here it is:

sr at Klaus:~/tmp > cat testDoubles.c
#include <stdio.h>
#include <math.h>

testDoubles () {
  double arg1 = 1.0e20;
  printf("\nsin(arg1): %e", sin(arg1));
  printf("\n");
}

int main() { testDoubles(); return 0; }

sr at Klaus:~/tmp > gcc -S testDoubles.c
(working, but with possibly wrong result -7.469219e-01)
sr at Klaus:~/tmp > less testDoubles.s 

        .file   "testDoubles.c"
        .version        "01.01"
gcc2_compiled.:
.section        .rodata
.LC0:
        .string "\nsin(arg1): %e"
.LC1:
        .string "\n"
.text
        .align 16
.globl testDoubles
        .type    testDoubles, at function
testDoubles:
        pushl %ebp
        movl %esp,%ebp
        subl $8,%esp
        movl $2025163840,-8(%ebp)
        movl $1142271773,-4(%ebp)
        fldl -8(%ebp)
        subl $8,%esp
        fstpl (%esp)
        call sin
        addl $8,%esp
        subl $8,%esp
        fstpl (%esp)
        pushl $.LC0
        call printf
        addl $12,%esp
        pushl $.LC1
        call printf
        addl $4,%esp
.L1:
        movl %ebp,%esp
        popl %ebp
        ret
.Lfe1:
        .size    testDoubles,.Lfe1-testDoubles
        .align 16
.globl main
        .type    main, at function
main:
        pushl %ebp
        movl %esp,%ebp
        call testDoubles
        xorl %eax,%eax
        jmp .L2
        .align 16
.L2:
        movl %ebp,%esp
        popl %ebp
        ret
.Lfe2:
        .size    main,.Lfe2-main
        .ident  "GCC: (GNU) egcs-2.91.66 19990314/Linux (egcs-1.1.2
release)"


sr at Klaus:~/tmp > gcc -O -S testDoubles.c
(not working, result == arg)
sr at Klaus:~/tmp > less testDoubles.s 

        .file   "testDoubles.c"
        .version        "01.01"
gcc2_compiled.:
.section        .rodata
.LC47:
        .string "\nsin(arg1): %e"
.LC48:
        .string "\n"
        .align 8
.LC46:
        .long 0x78b58c40,0x4415af1d
.text
        .align 16
.globl testDoubles
        .type    testDoubles, at function
testDoubles:
        pushl %ebp
        movl %esp,%ebp
        fldl .LC46
#APP
        fsin
#NO_APP
        subl $8,%esp
        fstpl (%esp)
        pushl $.LC47
        call printf
        pushl $.LC48
        call printf
        movl %ebp,%esp
        popl %ebp
        ret
.Lfe1:
        .size    testDoubles,.Lfe1-testDoubles
        .align 16
.globl main
        .type    main, at function
main:
        pushl %ebp
        movl %esp,%ebp
        call testDoubles
        xorl %eax,%eax
        movl %ebp,%esp
        popl %ebp
        ret
.Lfe2:
        .size    main,.Lfe2-main
        .ident  "GCC: (GNU) egcs-2.91.66 19990314/Linux (egcs-1.1.2
release)"


sr at Klaus:~/tmp > gcc -O -D__NO_MATH_INLINES -S testDoubles.c
(working, but with possibly wrong result -7.469219e-01)
sr at Klaus:~/tmp > less testDoubles.s

        .file   "testDoubles.c"
        .version        "01.01"
gcc2_compiled.:
.section        .rodata
.LC1:
        .string "\nsin(arg1): %e"
.LC2:
        .string "\n"
        .align 8
.LC0:
        .long 0x78b58c40,0x4415af1d
.text
        .align 16
.globl testDoubles
        .type    testDoubles, at function
testDoubles:
        pushl %ebp
        movl %esp,%ebp
        pushl $1142271773
        pushl $2025163840
        call sin
        subl $8,%esp
        fstpl (%esp)
        pushl $.LC1
        call printf
        pushl $.LC2
        call printf
        movl %ebp,%esp
        popl %ebp
        ret
.Lfe1:
        .size    testDoubles,.Lfe1-testDoubles
        .align 16
.globl main
        .type    main, at function
main:
        pushl %ebp
        movl %esp,%ebp
        call testDoubles
        xorl %eax,%eax
        movl %ebp,%esp
        popl %ebp
        ret
.Lfe2:
        .size    main,.Lfe2-main
        .ident  "GCC: (GNU) egcs-2.91.66 19990314/Linux (egcs-1.1.2
release)"


Observation: 'Working' versions use
        call sin
not working
	fsin
.
Don't know how to interpret this (haven't hacked Intel assembler in my
life).


Greetings,

Stephan
-- 
Stephan Rudlof (sr at evolgo.de)
   "Genius doesn't work on an assembly line basis.
    You can't simply say, 'Today I will be brilliant.'"
    -- Kirk, "The Ultimate Computer", stardate 4731.3