public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug c++/113896] New: Assigning array elements in the wrong order after floating point optimization
@ 2024-02-13  0:00 noobie-iv at mail dot ru
  2024-02-13  0:05 ` [Bug tree-optimization/113896] [12 Regression] " pinskia at gcc dot gnu.org
                   ` (9 more replies)
  0 siblings, 10 replies; 11+ messages in thread
From: noobie-iv at mail dot ru @ 2024-02-13  0:00 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113896

            Bug ID: 113896
           Summary: Assigning array elements in the wrong order after
                    floating point optimization
           Product: gcc
           Version: 12.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: noobie-iv at mail dot ru
  Target Milestone: ---

Created attachment 57404
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57404&action=edit
Test program and build_and_run script

/* file: f.cpp */
extern double a1; // 1.0
extern double a2; // 1.0

void f(double K[2], bool b)
{
    double A[] = {
        b ? a1 : a2,
        0,
        0,
        0
    };

    double sum{};
    for(double  a : A) sum += a;
    for(double& a : A) a /= sum;

    if (b) {
        K[0] = A[0]; // 1.0
        K[1] = A[1]; // 0.0
    } else {
        K[0] = A[0] + A[1];
    }
}

/* file: main.cpp */
#include <iostream>

double a1 = 1.0;
double a2 = 1.0;

void f(double K[2], bool b);

int main()
{
    double K[2]{};
    f(K, true);
    std::cout << K[0] << "\t" << K[1] << "\n";
}

Bug: Returns different results when compiled with different optimization
settings:
g++ -O2 -ffast-math f.cpp main.cpp -o good
g++ -O3 -ffast-math f.cpp main.cpp -o bad
./good outputs "1 0"
./bad outputs "0 1"

The bug is reproduced in systems:
* Fedora-37-1.7 x64 with latest gcc autoupdate (12.3.1 20230508)
* Debian-12.2.0 x64 with gcc-12.3.1 build from sources

In the gcc-12 branch:
* Last good commit: d127348d7711e148e5ddd205a8c3409b37fae64c (12.2.1 20221017)
* First bad commit: fe7d74313736b8e1c30812bc49419f419bdf1c53 (12.2.1 20221017)
* Last tested bad commit: 4ced4ca95001f1583623c801c9c3642224a2c4f0 (12.3.1
20240210)

In the gcc-13 branch - There is no bug:
* Last tested good commit: c891d8dc23e1a46ad9f3e757d09e57b500d40044 (13.2.0)
In the gcc-14 branch - There is no bug:
* Last tested good commit: cff174fabd6c980c09aee95db1d9d5c22421761f (14.0.1
20240210)

Note: Function f() is a simplified version of the
XSpline::linearCombinationFor() function from the scantailor-experimental
project:
https://github.com/ImageProcessing-ElectronicPublications/scantailor-experimental/blob/main/src/math/XSpline.cpp
Now the bug has been temporarily resolved by declaring array A volatile.

^ permalink raw reply	[flat|nested] 11+ messages in thread
end of thread, other threads:[~2024-02-15  3:16 UTC | newest]

Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-02-13  0:00 [Bug c++/113896] New: Assigning array elements in the wrong order after floating point optimization noobie-iv at mail dot ru
2024-02-13  0:05 ` [Bug tree-optimization/113896] [12 Regression] " pinskia at gcc dot gnu.org
2024-02-13  0:16 ` [Bug tree-optimization/113896] [12 Regression] Assigning array elements in the wrong order after floating point optimization since r12-8841 pinskia at gcc dot gnu.org
2024-02-13  8:42 ` rguenth at gcc dot gnu.org
2024-02-13  8:43 ` rguenth at gcc dot gnu.org
2024-02-13 12:36 ` rguenth at gcc dot gnu.org
2024-02-13 12:42 ` cvs-commit at gcc dot gnu.org
2024-02-13 12:48 ` cvs-commit at gcc dot gnu.org
2024-02-14 11:43 ` cvs-commit at gcc dot gnu.org
2024-02-14 11:43 ` rguenth at gcc dot gnu.org
2024-02-15  3:16 ` noobie-iv at mail dot ru

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).