[Bug math/27460] New: hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[Bug math/27460] New: hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[kretz at kde dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

            Bug ID: 27460
           Summary: hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises
                    FE_INEXACT and FE_UNDERFLOW
           Product: glibc
           Version: 2.31
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: math
          Assignee: unassigned at sourceware dot org
          Reporter: kretz at kde dot org
  Target Milestone: ---

Testcase (https://godbolt.org/z/b4Ko4Y):

#include <fenv.h>
#include <math.h>

int main()
{
    volatile double a = __DBL_DENORM_MIN__;
    volatile double c = -0.;
    feclearexcept(FE_ALL_EXCEPT);
    //c = hypot(a, c); // this argument order raises no exceptions
    c = hypot(c, a);
    return fetestexcept(FE_ALL_EXCEPT);
}

This program raises FE_INEXACT and FE_UNDERFLOW. Since one argument to hypot is
0 the result is exact: the absolute value of the other argument. If you switch
the argument order the program raises no floating-point exceptions.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[kretz at kde dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

Matthias Kretz (Vir) <kretz at kde dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-pc-linux-gnu

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[adhemerval.zanella at linaro dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

Adhemerval Zanella <adhemerval.zanella at linaro dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |adhemerval.zanella at linaro dot o
                   |                            |rg

--- Comment #1 from Adhemerval Zanella <adhemerval.zanella at linaro dot org> ---
Afaik since we define __STDC_IEC_559__ POSIX allows the implementation to
return an error range [1].  This is in fact what the fix for BZ#18803 does,
where it explicit raise underflow exception when the result is tiny and
inexact.

I think we are in fact missing a underflow check for the case where second
argument is '0':

diff --git a/sysdeps/ieee754/dbl-64/e_hypot.c
b/sysdeps/ieee754/dbl-64/e_hypot.c
index 9ec4c1ced0..bcd172130e 100644
--- a/sysdeps/ieee754/dbl-64/e_hypot.c
+++ b/sysdeps/ieee754/dbl-64/e_hypot.c
@@ -100,7 +100,10 @@ __ieee754_hypot (double x, double y)
          uint32_t low;
          GET_LOW_WORD (low, b);
          if ((hb | low) == 0)
-           return a;
+           {
+             math_check_force_underflow (x);
+             return a;
+           }
          t1 = 0;
          SET_HIGH_WORD (t1, 0x7fd00000);       /* t1=2^1022 */
          b *= t1;


[1] https://pubs.opengroup.org/onlinepubs/9699919799/functions/hypot.html

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[kretz at kde dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

--- Comment #2 from Matthias Kretz (Vir) <kretz at kde dot org> ---
>From C17:

> F.10.4.3
>  The hypot functions
> — hypot(x, y), hypot(y, x), and hypot(x, −y) are equivalent.
> — hypot(x, ±0) is equivalent to fabs(x).
> — hypot(±∞, y) returns +∞, even if y is a NaN.

Combining the first two bullets leads to: hypot(±0, x) is equivalent to
fabs(x).

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[adhemerval.zanella at linaro dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

--- Comment #3 from Adhemerval Zanella <adhemerval.zanella at linaro dot org> ---
C11 already states these rules and I am not sure if the rules are transitive as
you indicate. And I would not assume the rules are transitive becuase for 'pow'
(F.10.4.4 The pow functions) the standard states all possible argument
combinations.

Not assuming they are transitive means that my previous suggestion is incorrect
and current code already follow the "hypot(x, ±0) is equivalent to fabs(x)"
rule since hypot (__DBL_DENORM_MIN__, 0) returns __DBL_DENORM_MIN__.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[kretz at kde dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

--- Comment #4 from Matthias Kretz (Vir) <kretz at kde dot org> ---
I had C17 handy...

My test case shows the current code violates "hypot(x, y), hypot(y, x), and
hypot(x, −y) are equivalent". I don't think transitivity is the correct term
here. All these items must be true for hypot to be conforming to Annex F.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[joseph at codesourcery dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

--- Comment #5 from joseph at codesourcery dot com <joseph at codesourcery dot com> ---
I think the rules in Annex F allowing spurious "inexact" and "underflow", 
"unless stated otherwise", apply here (glibc limits spurious underflow to 
cases where the returned result would be possible for an underflowing 
computation).  (I consider the rules about functions bound to IEC 60559 
operations, and about functions with qNaN argument returning qNaN with no 
exceptions, to be "stating otherwise", but not the rules stating the value 
of a result for a non-NaN argument.)

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[kretz at kde dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

--- Comment #6 from Matthias Kretz (Vir) <kretz at kde dot org> ---
F10/9 Whether or when library functions raise an undeserved "underflow"
floating-point exception is unspecified.379) Otherwise, as implied by F.8.6,
the <math.h> functions do not raise spurious floating-point exceptions
(detectable by the user), other than the "inexact" floating-point exception.

Footnote 379) It is intended that undeserved "underflow" and "inexact"
floating-point exceptions are raised only if avoiding them would be too costly.

Additionally I believe F.10.4.3 states otherwise. I.e. hypot(denorm_min, 0) is
equivalent to fabs(denorm_min) and hypot(0, denorm_min) is equivalent to
hypot(denorm_min, 0).

But I'm new to interpreting floating-point exceptions legalese. I was only
trying to verify my hypot SIMD implementation against yours and stumbled over
this inconsistency.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[adhemerval.zanella at linaro dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

--- Comment #7 from Adhemerval Zanella <adhemerval.zanella at linaro dot org> ---
Joseph, do you considere this a real issue for glibc (meaning that we should
keep this bug opened) or should we close it?

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[joseph at codesourcery dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

--- Comment #8 from joseph at codesourcery dot com <joseph at codesourcery dot com> ---
I don't think there is a bug for glibc.  Maybe there is an issue with the 
C standard wording that it should say that hypot *returns* a value given 
by fabs rather than *is equivalent* to a given call to fabs (saying a 
function not bound to an IEEE operation is equivalent to one bound to such 
an operation is very questionable).

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[zimmerma+gcc at loria dot fr]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

Paul Zimmermann <zimmerma+gcc at loria dot fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zimmerma+gcc at loria dot fr

--- Comment #9 from Paul Zimmermann <zimmerma+gcc at loria dot fr> ---
see https://mailman.oakapple.net/pipermail/cfp-interest/2021-April/001966.html:
full support for the inexact exception is not required for hypot (and
mathematical functions in general)

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/27460] hypot(__DBL_DENORM_MIN__, -0.) incorrectly raises FE_INEXACT and FE_UNDERFLOW

[adhemerval.zanella at linaro dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=27460

Adhemerval Zanella <adhemerval.zanella at linaro dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |NOTABUG

--- Comment #10 from Adhemerval Zanella <adhemerval.zanella at linaro dot org> ---
Closed based on both Joseph and Paul comments.

-- 
You are receiving this mail because:
You are on the CC list for the bug.