Hey guys... im doing some floating point stuff in my 64 bit program and i need to compare two flloating point numbers and jump for a condition. What i tried but didnt work was the following:

faddp st1,st0
fcomp [ex2] ; ex2 is the value 4.0
fstsw ax
sahf ;ILLEGAL instruction
ja finished

i would really appreciate it if anyone has an alternative to this or a way around the problems.

Thanks, Mark

Use fcomi.

I wonder what version of FASM are you using? Both 1.67.21 and 1.67.29 (newest) accepts SAHF and LAHF.

BTW, don't forget that when comparing two FP numbers which are product of some computation, you should compare with some margin, eg:
if num1-margin < num2 < num1+margin then numbers are equal.

Of course this may not always apply, but mostly it does

MazeGen,

They do accept, but…

barld: yes (and ref.x86asm.net/coder64.html says the same) but only early steppings of EM64T architecture don't support LAHF and SAHF and as long as we can't set some "100%-64-bit-compatibility" option in FASM, LAHF and SAHF should be enabled.

i used fcomi instead... thanks revolution.



Fasm does allow it to assembler (SAHF and LAHF)... no problems. But if you run it through a debugger for example fdbg001B it will still come up as an illegal instruction and hence the program will just skip over it.

mehble,

FDBG decode 0x9E as "illegal sahf" but F7/F8 doesn't generate #UD fault and sahf works as specified? Probably FDBG's authors should check aforementioned CPUID flag…

Yay, another spelling of my nick! Only 117 left…

thanks guys, I'll update that in fdbg

Use TEST ah,condition_bitmask.

Hi, I implemented the checking of LAHF / SAHF instructions support into fdbg so now it doesn't output such confusing informations.
http://board.flatassembler.net/topic.php?p=85361#85361