Strange behavior when passing single precision numbers into atan2

When I pass two single-precision arguments into the atan2 function, the result is different if I assign the output to a variable compared with just allowing it to be assigned to ans. If I run the following code:

a = single(1.2345678);
b = single(1.3456789);
atan2(a,b)
c = atan2(a,b)

This is displayed to the workspace:

ans =

   7.4236256e-01

c =

   7.4236250e-01

Why is there a discrepancy between these two numbers?

[Walter Roberson]

Which MATLAB version are you using, on which operating system?

Also, what is your "format" set to? I do not get that format of output for any of the "format" options I try.

[per isakson]

format hex
a = single(1.2345678);
b = single(1.3456789);
atan2(a,b)
c = atan2(a,b)
version

outputs

ans =
   3f3e0b79
c =
   3f3e0b79
ans =
9.0.0.341360 (R2016a)

i.e. identical values (on Win7)

[Colby Smith]

I'm using R2014b and Windows 7. Format is longe. If I use hex:

format hex
a = single(1.2345678);
b = single(1.3456789);
atan2(a,b)
c = atan2(a,b)

The following is output:

ans =
   3f3e0b79

c =
   3f3e0b78

If I use double precision:

format hex
a = 1.2345678;
b = 1.3456789;
atan2(a,b)
c = atan2(a,b)

The answers match:

ans =
   3fe7c16f1bd9f8ff

c =
   3fe7c16f1bd9f8ff

[dpb]

Wonder how much if not all may not be related to the underlying compiler libraries...so versions used by TMW matter?

Answer 1

[dpb]

single doesn't have anything to do with it (albeit the precision of the input values is about that of single precision in decimal digits)...

>> sprintf('%.7e\n',atan2(a,b))
ans =
7.4236256e-01

>> sprintf('%.7e\n',c)
ans =
7.4236250e-01

I'm guessing OP had to have used something like the above to see the result as shown, Walter.

The answer likely has to do with Matlab displaying a result directly to the workspace from the FPP vis a vis a variable store as the difference is on order of double resolution...

>> eps(c)
ans =
5.9605e-08
>>