Gnat/Windows float point Q

Gnat/Windows float point Q

[tmoran]

What does it mean when float'image(x) = " 47.720649 0.E+05" ?
                                                   ???

Is x a NAN or something?  It causes a Constraint_Error in
the "return x" statement in:

  subtype units is float range 0.0 .. 1.0;
  function clip(x: float) return units is
  begin
    if x < 0.0 then return 0.0;
    elsif x > 1.0 then return 1.0;
    else return x;
    end if;
  end clip;

Re: Gnat/Windows float point Q

[Eric G. Miller]

In <1MGka.57623$ug3.113346@rwcrnsc51.ops.asp.att.net>, tmora wrote:

> What does it mean when float'image(x) = " 47.720649 0.E+05" ?
>                                                    ???
> 
> Is x a NAN or something?  It causes a Constraint_Error in
> the "return x" statement in:
> 
>   subtype units is float range 0.0 .. 1.0;
>   function clip(x: float) return units is
>   begin
>     if x < 0.0 then return 0.0;
>     elsif x > 1.0 then return 1.0;
>     else return x;
>     end if;
>   end clip;

Check X'Valid if the data comes from some route like Unchecked_Conversion.
The following demonstrates the conversion error on some systems..

with Ada.Text_Io;
use  Ada.Text_Io;
with Ada.Unchecked_Conversion;

procedure Unitarians
is

   subtype Units is Float range 0.0 .. 1.0;

   function Clip(X: Float) return Units is
   begin
      if X'Valid = False then
         raise Constraint_Error;
      end if;
      if X < 0.0 then
         return 0.0;
      elsif X > 1.0 then return 1.0;
      else return X;
      end if;
   end Clip;

   type Byte4 is mod 2**32;

   function To_Float is new
     Ada.Unchecked_Conversion (Byte4, Float);

   U : Units;
   F : Float;
   UI : Byte4 := 16#FFFF_FFFF#;
begin
   Put_Line ("UI = " & Byte4'Image (UI));
   F := To_Float (UI);
   Put_Line ("F is valid? --> " & Boolean'Image(F'Valid));
   U := Clip (F);
   Put_Line ("U = " & Units'Image(U));

end Unitarians;

-- 
echo ">gra.fcw@2ztr< eryyvZ .T pveR" | rot13 | reverse

Re: Gnat/Windows float point Q

[tmoran]

>> What does it mean when float'image(x) = " 47.720649 0.E+05" ?
> Check X'Valid if the data comes from some route like Unchecked_Conversion.
In this case X comes from Ada.Numerics.Complex_Types.Abs(Y)
and Y came from Z**2, so it might well overflow.  Float'Machine_Overflows
is False, so I can't expect an exception, but is an overflowed value
on Gnat/Windows something that is always False in a compare, and
generates garbage for float'image?

Re: Gnat/Windows float point Q

[Randy Brukardt]

tmoran@acm.org wrote in message ...
>>> What does it mean when float'image(x) = " 47.720649 0.E+05" ?
>> Check X'Valid if the data comes from some route like
Unchecked_Conversion.
>In this case X comes from Ada.Numerics.Complex_Types.Abs(Y)
>and Y came from Z**2, so it might well overflow.
Float'Machine_Overflows
>is False, so I can't expect an exception, but is an overflowed value
>on Gnat/Windows something that is always False in a compare, and
>generates garbage for float'image?

I'd expect the result of such an operation to be either +infinity
or -infinity. Those compare as you expect, and probably raise
Constraint_Error when stored (if they didn't do that, the compiler would
be violating the rule about math always returning the correct result or
raising an exception).

             Randy.

Re: Gnat/Windows float point Q

[tmoran]

> I'd expect the result of such an operation to be either +infinity
> or -infinity. Those compare as you expect, and probably raise
> Constraint_Error when stored (if they didn't do that, the compiler would
> be violating the rule about math always returning the correct result or
> raising an exception).

  Is this a legitimate output, in terms of exceptions (none), comparisons
(the last three are False in both directions) and float'image (missing
leading blanks and worse), from this program?

with ada.numerics.complex_types,
     ada.text_io;
procedure oflow is
  use ada.numerics.complex_types;
  x : complex := (4.7E+17, -1.6E+17);
  y : float;
begin
  for i in 1 .. 5 loop
    y := abs(x);
    ada.text_io.put_line("abs(x) = y =" & float'image(y)
      & " y < 0.0 " & boolean'image(y < 0.0)
      & " y >= 0.0 " & boolean'image(y >= 0.0));
    ada.text_io.put_line(float'image(re(x)) & float'image(im(x)));
    ada.text_io.put_line(float'image(re(x)) & ":" & float'image(im(x)));
    ada.text_io.new_line;
    x := x**2;
  end loop;
end oflow;

abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
 4.70000E+17-1.60000E+17
 4.70000E+17:-1.60000E+17

abs(x) = y = 2.46500E+35 y < 0.0 FALSE y >= 0.0 TRUE
 1.95300E+35-1.50400E+35
 1.95300E+35:-1.50400E+35

abs(x) = y =1.95300E+05 y < 0.0 FALSE y >= 0.0 FALSE
 +Inf*******1.50400E+05
 +Inf*******:1.50401E+05

abs(x) = y =+Inf******* y < 0.0 FALSE y >= 0.0 FALSE
3.36210E-49321.50401E+05
9.52681E+139:1.50402E+05

abs(x) = y =9.52681E+05 y < 0.0 FALSE y >= 0.0 FALSE
0.E+051.50402E+05
3.64520E-4951:1.50403E+05

Re: Gnat/Windows float point Q

[Randy Brukardt]

tmoran@acm.org wrote in message ...
>> I'd expect the result of such an operation to be either +infinity
>> or -infinity. Those compare as you expect, and probably raise
>> Constraint_Error when stored (if they didn't do that, the compiler
would
>> be violating the rule about math always returning the correct result
or
>> raising an exception).
>
>  Is this a legitimate output, in terms of exceptions (none),
comparisons
>(the last three are False in both directions) and float'image (missing
>leading blanks and worse), from this program?


Hard to say; I don't understand the rules of Complex math that well. But
I do find the results of the comparisons to be pretty weird. Have you
tried this on a different compiler?

                Randy.

Re: Gnat/Windows float point Q

[tmoran]

>>  Is this a legitimate output, in terms of exceptions (none), comparisons
>>(the last three are False in both directions) and float'image (missing
>>leading blanks and worse), from this program?
>
>Hard to say; I don't understand the rules of Complex math that well. But
>I do find the results of the comparisons to be pretty weird. Have you
>tried this on a different compiler?
  Neither of my other compilers has Ada.Numerics.Complex_Types.  However,
the complex arithmetic in this program is trivial, so I dropped that
and made q&d Float functions.  Gnat still screws up as before, so it's
a Float, not a Complex, problem.  The other two compilers raise
exceptions instead of generating garbage.

 1  with Ada.Exceptions,
 2       ada.numerics.elementary_functions,
 3       ada.text_io;
 4  procedure oflow2 is
 5    use ada.numerics.elementary_functions;
 6    type complex is record
 7      re, im : float;
 8    end record;
 9    function "**"(left:complex; right:positive) return complex is
10      result : complex := left;
11    begin
12      for i in 2 .. right loop
13        result := (re => left.re * result.re - left.im * result.im,
14                   im => left.re * result.im + left.im * result.re);
15      end loop;
16      return result;
17    end "**";
18    function "abs"(x : complex) return float is
19    begin
20      return ada.numerics.elementary_functions.sqrt(x.re**2 + x.im**2);
21    end "abs";
22    function re(x:complex) return float is
23    begin return x.re;end re;
24    function im(x:complex) return float is
25    begin return x.im;end im;
26    x : complex := (4.7E+17, -1.6E+17);
27    y : float;
28  begin
29    for i in 1 .. 5 loop
30      y := abs(x);
31      ada.text_io.put_line("abs(x) = y =" & float'image(y)
32        & " y < 0.0 " & boolean'image(y < 0.0)
33        & " y >= 0.0 " & boolean'image(y >= 0.0));
34      ada.text_io.put_line(float'image(re(x)) & float'image(im(x)));
35      ada.text_io.put_line(float'image(re(x)) & ":" & float'image(im(x)));
36      ada.text_io.new_line;
37      x := x**2;
38    end loop;
39  exception
40          when oops:others=>
41            ada.text_io.put_line(ada.exceptions.exception_information(oops));
42  end oflow2;
-----------------------
ObjectAda
abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
 4.70000E+17-1.60000E+17
 4.70000E+17:-1.60000E+17

CONSTRAINT_ERROR (numeric overflow)
Exception traceback:
   20   oflow2.abs                                 oflow2.adb
   30   oflow2                                     oflow2.adb
Exception handled at:
   40   oflow2                                     oflow2.adb
-----------------------
Janus
abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
 4.70000E+17-1.60000E+17
 4.70000E+17:-1.60000E+17

abs(x) = y = 2.46500E+35 y < 0.0 FALSE y >= 0.0 TRUE
 1.95300E+35-1.50400E+35
 1.95300E+35:-1.50400E+35

CONSTRAINT_ERROR
   Floating Point Overflow Detected
On Line Number 14 In OFLOW2.**
Called from line number 37 In OFLOW2
-----------------------
Gnat
abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
 4.70000E+17-1.60000E+17
 4.70000E+17:-1.60000E+17

abs(x) = y = +Inf******* y < 0.0 FALSE y >= 0.0 TRUE
 1.95300E+35-1.50400E+35
 1.95300E+35:-1.50400E+35

abs(x) = y = +Inf******* y < 0.0 FALSE y >= 0.0 TRUE
 +Inf*******-Inf*******
 +Inf*******:-Inf*******

abs(x) = y =0.E+05 y < 0.0 FALSE y >= 0.0 FALSE
1.19442E-4946-Inf*******
0.E+05:-Inf*******

abs(x) = y =0.E+05 y < 0.0 FALSE y >= 0.0 FALSE
4.84066E-4942Inf*******
0.E+05:0.E+05

Re: Gnat/Windows float point Q

[David C. Hoos, Sr.]

Could this paragraph from the GNAT Reference Manual have any bearing on
this problem?

9.31 GNAT.Float_Control (`g-flocon.ads') 
Provides an interface for resetting the floating-point processor into the
mode required for correct semantic operation in Ada. Some third party
library calls may cause this mode to be modified, and the Reset procedure
in this package can be used to reestablish the required mode. 

----- Original Message ----- 
From: <tmoran@acm.org>
Newsgroups: comp.lang.ada
To: <comp.lang.ada@ada.eu.org>
Sent: April 10, 2003 4:58 PM
Subject: Re: Gnat/Windows float point Q


> >>  Is this a legitimate output, in terms of exceptions (none), comparisons
> >>(the last three are False in both directions) and float'image (missing
> >>leading blanks and worse), from this program?
> >
> >Hard to say; I don't understand the rules of Complex math that well. But
> >I do find the results of the comparisons to be pretty weird. Have you
> >tried this on a different compiler?
>   Neither of my other compilers has Ada.Numerics.Complex_Types.  However,
> the complex arithmetic in this program is trivial, so I dropped that
> and made q&d Float functions.  Gnat still screws up as before, so it's
> a Float, not a Complex, problem.  The other two compilers raise
> exceptions instead of generating garbage.
> 
>  1  with Ada.Exceptions,
>  2       ada.numerics.elementary_functions,
>  3       ada.text_io;
>  4  procedure oflow2 is
>  5    use ada.numerics.elementary_functions;
>  6    type complex is record
>  7      re, im : float;
>  8    end record;
>  9    function "**"(left:complex; right:positive) return complex is
> 10      result : complex := left;
> 11    begin
> 12      for i in 2 .. right loop
> 13        result := (re => left.re * result.re - left.im * result.im,
> 14                   im => left.re * result.im + left.im * result.re);
> 15      end loop;
> 16      return result;
> 17    end "**";
> 18    function "abs"(x : complex) return float is
> 19    begin
> 20      return ada.numerics.elementary_functions.sqrt(x.re**2 + x.im**2);
> 21    end "abs";
> 22    function re(x:complex) return float is
> 23    begin return x.re;end re;
> 24    function im(x:complex) return float is
> 25    begin return x.im;end im;
> 26    x : complex := (4.7E+17, -1.6E+17);
> 27    y : float;
> 28  begin
> 29    for i in 1 .. 5 loop
> 30      y := abs(x);
> 31      ada.text_io.put_line("abs(x) = y =" & float'image(y)
> 32        & " y < 0.0 " & boolean'image(y < 0.0)
> 33        & " y >= 0.0 " & boolean'image(y >= 0.0));
> 34      ada.text_io.put_line(float'image(re(x)) & float'image(im(x)));
> 35      ada.text_io.put_line(float'image(re(x)) & ":" & float'image(im(x)));
> 36      ada.text_io.new_line;
> 37      x := x**2;
> 38    end loop;
> 39  exception
> 40          when oops:others=>
> 41            ada.text_io.put_line(ada.exceptions.exception_information(oops));
> 42  end oflow2;
> -----------------------
> ObjectAda
> abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
>  4.70000E+17-1.60000E+17
>  4.70000E+17:-1.60000E+17
> 
> CONSTRAINT_ERROR (numeric overflow)
> Exception traceback:
>    20   oflow2.abs                                 oflow2.adb
>    30   oflow2                                     oflow2.adb
> Exception handled at:
>    40   oflow2                                     oflow2.adb
> -----------------------
> Janus
> abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
>  4.70000E+17-1.60000E+17
>  4.70000E+17:-1.60000E+17
> 
> abs(x) = y = 2.46500E+35 y < 0.0 FALSE y >= 0.0 TRUE
>  1.95300E+35-1.50400E+35
>  1.95300E+35:-1.50400E+35
> 
> CONSTRAINT_ERROR
>    Floating Point Overflow Detected
> On Line Number 14 In OFLOW2.**
> Called from line number 37 In OFLOW2
> -----------------------
> Gnat
> abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
>  4.70000E+17-1.60000E+17
>  4.70000E+17:-1.60000E+17
> 
> abs(x) = y = +Inf******* y < 0.0 FALSE y >= 0.0 TRUE
>  1.95300E+35-1.50400E+35
>  1.95300E+35:-1.50400E+35
> 
> abs(x) = y = +Inf******* y < 0.0 FALSE y >= 0.0 TRUE
>  +Inf*******-Inf*******
>  +Inf*******:-Inf*******
> 
> abs(x) = y =0.E+05 y < 0.0 FALSE y >= 0.0 FALSE
> 1.19442E-4946-Inf*******
> 0.E+05:-Inf*******
> 
> abs(x) = y =0.E+05 y < 0.0 FALSE y >= 0.0 FALSE
> 4.84066E-4942Inf*******
> 0.E+05:0.E+05
> _______________________________________________
> comp.lang.ada mailing list
> comp.lang.ada@ada.eu.org
> http://ada.eu.org/mailman/listinfo/comp.lang.ada
> 
>

Re: Gnat/Windows float point Q

[tmoran]

> Could this paragraph from the GNAT Reference Manual have any bearing on
> this problem?
>
> 9.31 GNAT.Float_Control (`g-flocon.ads')
I added a "with" and inserted as the first line after "begin"
  GNAT.Float_Control.Reset;
It made no difference.  I'm using Gnat 3.15p on Windows.  Perhaps
newer versions work differently?  BTW, I note that "-gnato -O2"
generates somewhat different results - still screwy in the same
ways, but different.

Re: Gnat/Windows float point Q

[Eric G. Miller]

In article <U%ola.405441$3D1.219740@sccrnsc01>, tmoran@acm.org wrote:
>> Could this paragraph from the GNAT Reference Manual have any bearing on
>> this problem?
>>
>> 9.31 GNAT.Float_Control (`g-flocon.ads')
> I added a "with" and inserted as the first line after "begin"
>   GNAT.Float_Control.Reset;
> It made no difference.  I'm using Gnat 3.15p on Windows.  Perhaps
> newer versions work differently?  BTW, I note that "-gnato -O2"
> generates somewhat different results - still screwy in the same
> ways, but different.

I tried playing around with your example, but instantiated the generic
complex type with both long and normal floats.  It seemed to make a 
difference as the generic versions caused an exception on
abs() (though Im() and Re() both worked returning infinities!).

Mind you, this is with gcc-3.2, which I guess is somewhat different from
the 3.15p release...  (I tried all the various -gnatV switches, to no
effect).


Besides the T'Valid flag, how do you check specifically for or set infinity
or NaN or denormal floating point numbers?

-- 
echo ">gra.fcw@2ztr< eryyvZ .T pveR" | rot13 | reverse

Re: Gnat/Windows float point Q

[Simon Wright]

tmoran@acm.org writes:

> > I'd expect the result of such an operation to be either +infinity
> > or -infinity. Those compare as you expect, and probably raise
> > Constraint_Error when stored (if they didn't do that, the compiler would
> > be violating the rule about math always returning the correct result or
> > raising an exception).
> 
>   Is this a legitimate output, in terms of exceptions (none), comparisons
> (the last three are False in both directions) and float'image (missing
> leading blanks and worse), from this program?
> 
> with ada.numerics.complex_types,
>      ada.text_io;
> procedure oflow is
>   use ada.numerics.complex_types;
>   x : complex := (4.7E+17, -1.6E+17);
>   y : float;
> begin
>   for i in 1 .. 5 loop
>     y := abs(x);
>     ada.text_io.put_line("abs(x) = y =" & float'image(y)
>       & " y < 0.0 " & boolean'image(y < 0.0)
>       & " y >= 0.0 " & boolean'image(y >= 0.0));
>     ada.text_io.put_line(float'image(re(x)) & float'image(im(x)));
>     ada.text_io.put_line(float'image(re(x)) & ":" & float'image(im(x)));
>     ada.text_io.new_line;
>     x := x**2;
>   end loop;
> end oflow;
> 
> abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
>  4.70000E+17-1.60000E+17
>  4.70000E+17:-1.60000E+17
> 
> abs(x) = y = 2.46500E+35 y < 0.0 FALSE y >= 0.0 TRUE
>  1.95300E+35-1.50400E+35
>  1.95300E+35:-1.50400E+35
> 
> abs(x) = y =1.95300E+05 y < 0.0 FALSE y >= 0.0 FALSE
>  +Inf*******1.50400E+05
>  +Inf*******:1.50401E+05
> 
> abs(x) = y =+Inf******* y < 0.0 FALSE y >= 0.0 FALSE
> 3.36210E-49321.50401E+05
> 9.52681E+139:1.50402E+05
> 
> abs(x) = y =9.52681E+05 y < 0.0 FALSE y >= 0.0 FALSE
> 0.E+051.50402E+05
> 3.64520E-4951:1.50403E+05

With GNAT 3.16a (compiled here for Mandrake 8.2), -gnato -O2, this gives

   abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
    4.70000E+17-1.60000E+17
    4.70000E+17:-1.60000E+17

   abs(x) = y = 2.46500E+35 y < 0.0 FALSE y >= 0.0 TRUE
    1.95300E+35-1.50400E+35
    1.95300E+35:-1.50400E+35

   abs(x) = y =NaN******** y < 0.0 FALSE y >= 0.0 FALSE
    +Inf*******NaN********
    +Inf*******:NaN********

   abs(x) = y =NaN******** y < 0.0 FALSE y >= 0.0 FALSE
   NaN********NaN********
   NaN********:NaN********

   abs(x) = y =NaN******** y < 0.0 FALSE y >= 0.0 FALSE
   NaN********NaN********
   NaN********:NaN********

(at least with NaN it's correct that the comparisons are false both
ways).

With gcc-3.3-20030201 the result is

   abs(x) = y = 4.96488E+17 y < 0.0 FALSE y >= 0.0 TRUE
    4.70000E+17-1.60000E+17
    4.70000E+17:-1.60000E+17

   abs(x) = y = 2.46500E+35 y < 0.0 FALSE y >= 0.0 TRUE
    1.95300E+35-1.50400E+35
    1.95300E+35:-1.50400E+35

   abs(x) = y =0.E+05 y < 0.0 FALSE y >= 0.0 FALSE
    +Inf*******0.E+05

   Execution terminated by unhandled exception
   Exception name: CONSTRAINT_ERROR
   Message: s-imgrea.adb:235 explicit raise
   Call stack traceback locations:
   0x804a581 0x805813d 0x80587bc 0x8057d6f 0x804a29d 0x8049e47

(traceback: shows oflow.adb:14)