[Bug c/111741] New: gcc long double precision

[Bug c/111741] New: gcc long double precision

[bernardwidynski at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111741

            Bug ID: 111741
           Summary: gcc long double precision
           Product: gcc
           Version: 11.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: bernardwidynski at gmail dot com
  Target Milestone: ---

Created attachment 56081
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56081&action=edit
C program to compute sum of numbers 1, 2, 3, ... N

It is my understanding the long double in gcc has 80 bits precision.

I've run a simple program which shows that it is less than 80 bits precision.

The numbers 1, 2, 3, ... N are summed and compared with N*(N+1)/2

For the case where N = 2^32, the sums compare correctly.

For the case where N = 2^33, the sums are different.

2^33*(2^33-1)/2 is less than 80 bits in precision.

Why doesn't the long double have the capacity for this computation?

See attached program and output file.

This was run on Cygwin64 using gcc version 11.4.0 on an Intel Core i7-9700

[Bug c/111741] gcc long double precision

[bernardwidynski at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111741

--- Comment #1 from bernardwidynski at gmail dot com ---
Created attachment 56082
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56082&action=edit
Output file

[Bug c/111741] gcc long double precision

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111741

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

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

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
80bit is the full precission and that 80bits includes 1 bit sign bit, 64bits
for the mantissa and 15bits for the exponent.

So anything above 64bits will start to lose precission in the last digits.

[Bug c/111741] gcc long double precision

[bernardwidynski at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111741

--- Comment #3 from bernardwidynski at gmail dot com ---
Thanks for the quick response.

That explains it.

On Mon, Oct 9, 2023 at 10:20 AM pinskia at gcc dot gnu.org <
gcc-bugzilla@gcc.gnu.org> wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111741
>
> Andrew Pinski <pinskia at gcc dot gnu.org> changed:
>
>            What    |Removed                     |Added
>
> ----------------------------------------------------------------------------
>              Status|UNCONFIRMED                 |RESOLVED
>          Resolution|---                         |INVALID
>
> --- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
> 80bit is the full precission and that 80bits includes 1 bit sign bit,
> 64bits
> for the mantissa and 15bits for the exponent.
>
> So anything above 64bits will start to lose precission in the last digits.
>
> --
> You are receiving this mail because:
> You reported the bug.