Re: Parsing timestamps?

On 10/07/2025 6:35 pm, Anton Ertl wrote:

dxf <dxforth@gmail.com> writes:

The catch with SSE is there's nothing like FCHS or FABS
so depending on how one implements them, results vary across implementations.

 
You can see in Gforth how to implement FNEGATE and FABS with SSE2:
 
see fnegate
Code fnegate 
   0x000055e6a78a8274:   add    $0x8,%rbx
   0x000055e6a78a8278:   xorpd  0x24d8f(%rip),%xmm15        # 0x55e6a78cd010
   0x000055e6a78a8281:   mov    %r15,%r9
   0x000055e6a78a8284:   mov    (%rbx),%rax
   0x000055e6a78a8287:   jmp    *%rax
end-code
 ok
0x55e6a78cd010 16 dump
55E6A78CD010: 00 00 00 00  00 00 00 80 - 00 00 00 00  00 00 00 00
 ok
see fabs
Code fabs 
   0x000055e6a78a84fe:   add    $0x8,%rbx
   0x000055e6a78a8502:   andpd  0x24b15(%rip),%xmm15        # 0x55e6a78cd020
   0x000055e6a78a850b:   mov    %r15,%r9
   0x000055e6a78a850e:   mov    (%rbx),%rax
   0x000055e6a78a8511:   jmp    *%rax
end-code
 ok
0x55e6a78cd020 16 dump
55E6A78CD020: FF FF FF FF  FF FF FF 7F - 00 00 00 00  00 00 00 00
 
The actual implementation is the xorpd instruction for FNEGATE, and in
the andpd instruction for FABS.  The memory locations contain masks:
for FNEGATE only the sign bit is set, for FABS everything but the sign
bit is set.
 
Sure you can implement FNEGATE and FABS in more complicated ways, but
you can also implement them in more complicated ways if you use the
387 instruction set.  Here's an example of more complicated
implementations:
 
see fnegate
FNEGATE
( 004C4010    4833C0 )                XOR     RAX, RAX
( 004C4013    F34D0F7EC8 )            MOVQ    XMM9, XMM8
( 004C4018    664C0F6EC0 )            MOVQ    XMM8, RAX
( 004C401D    F2450F5CC1 )            SUBSD   XMM8, XMM9
( 004C4022    C3 )                    RET/NEXT
( 19 bytes, 5 instructions )
 ok
see fabs
FABS
( 004C40B0    E8FBEFFFFF )            CALL    004C30B0  FS@
( 004C40B5    4885DB )                TEST    RBX, RBX
( 004C40B8    488B5D00 )              MOV     RBX, [RBP]
( 004C40BC    488D6D08 )              LEA     RBP, [RBP+08]
( 004C40C0    0F8D05000000 )          JNL/GE  004C40CB
( 004C40C6    E845FFFFFF )            CALL    004C4010  FNEGATE
( 004C40CB    C3 )                    RET/NEXT
( 28 bytes, 7 instructions )

The latter were basically what was existed in the implementation.  As they
don't handle -ve zero (or NANs) I swapped them out for the former ones you
mention.