Floating point compare

Chuck Rice

.

is this syntax valid?

if (fp.sin (fp.fsub(Lon2 , Lon1)) < 0.0)

In other words, can I compare against a floating point constant?

.

Paul Baker

It will work but only by chance, spin does not provide any support for floating point numbers except constructing constants. IOW floating point comparisons are not supported. But if you disolve what you are doing into the corresponding bit manipulations of IEEE single floating point, 0.0 is equal to $0000_0000, so 0.0 = 0 this is called a congruent representation. x < 0.0 is a negative value test, it turns out that the msb of ieee floating point is the sign bit. This also happens to be the bit in integer math which is set when the value is negative, so the comparison in integer math done within spin happens to produce the same result as a true floating point comparison for this example, but trying to use this technique will not work for anything besides > 0.0,· < 0.0 or == 0.0

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Paul Baker
Propeller Applications Engineer

Parallax, Inc.

Chuck Rice

.

So what is the correct way? I searched Using Parallax Propeller™ Floating Point Routines for compare, but got no hits.

Would it be:

if (fp.sin (fp.fsub(Lon2 , Lon1)) < constant(0.0))

Post Edited (Chuck Rice) : 4/9/2008 8:49:35 PM GMT

Paul Baker

Your original statement was fine, the compiler knows to convert a floating point constant·into the ieee representation. As far as library support, I don't know I've never found the need to use floating point in my projects. But equality testing is trivial and can actually be done in integer math using a converted floating point number as the comparator, so all that would be need is a if_above and/or if_below routine which diasects the sign, exponent and mantissa from the numbers and compare it in an iterative approach (ie if signs differ, value with bit set is less than value with bit clear; if the sign is the same then compare the exponent the one with the larger exponent is greater; if exponents are the same the one with the larger mantissa is greater)

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Paul Baker
Propeller Applications Engineer

Parallax, Inc.

Post Edited (Paul Baker (Parallax)) : 4/9/2008 9:12:37 PM GMT

John Abshier

You probably want to use FCmp(NumberA, NumberB)
returns 1 if NumberA > Number B, -1 if NumberA < NumberB, 0 if equal

Chuck Rice

John Abshier said...
You probably want to use FCmp(NumberA, NumberB)
returns 1 if NumberA > Number B, -1 if NumberA < NumberB, 0 if equal

Doh! I sure missed that one. Thanks!