Floating point odd issue

[karlglazebrook]

I found a curious bug when trying to use gcc 14.2 from here to build GSL on an Intel Sonoma iMac.

I was able to make this standalone code that reproduces the bug on the machine:

#include <stdio.h>
#include <math.h>

/* f(x) = sin(x) */
double f_sin(double x)
{
    return sin(x);
}

// Main function: entry point for execution
int main() {

    double x,y;

    x = sin(7.0);
    printf("%20.17f \n",x);
    y = f_sin(7.0);
    printf("%20.17f \n",y);
    printf("%20.17f \n",x-y);

    return 0;
}

Here is the output:

 0.65698659871878906 
 0.65698659871878917 
-0.00000000000000011 

So calling the same function directly vs indirectly produces a different value! (Even with no optimisation -O0).

Very odd! (For those interested the equivalent of this code was executing at line 2236 of gsl-2.7.1/integration/test.c and failing the test).

Works fine on Apple Silicon.

Works fine if you change the value to 2.0 or 5.0.

Works fine if you put '7.0' in to a variable and use that!

Not sure if this is easy to fix, probably obscure and particular compiler boojums, but sharing for fun & giggles, and the public record.

% gcc -v 
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/local/gfortran/libexec/gcc/x86_64-apple-darwin23/14.2.0/lto-wrapper
Target: x86_64-apple-darwin23
Configured with: /Users/fx/gcc-14-branch/configure --prefix=/usr/local/gfortran --with-gmp=/Users/fx/build_package/deps --enable-languages=c,c++,fortran,objc,obj-c++,m2 --build=x86_64-apple-darwin23 --with-sysroot=/Library/Developer/CommandLineTools/SDKs/MacOSX14.sdk
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 14.2.0 (GCC) 

XCode 16.2 installed.

[karlglazebrook]

PS found this when I was building a complex set of libraries (incl. GSL) some with c and some with fortran hence was trying to use consistent gcc and gfortran throughout. Have used this distribution for a while - keep up the good work!