[Bug rtl-optimization/116979] New: [12 regression] fma not always used in complex product

[vincenzo.innocente at cern dot ch via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116979

            Bug ID: 116979
           Summary: [12 regression] fma not always used in complex product
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vincenzo.innocente at cern dot ch
  Target Milestone: ---

since gcc 12 fma is not always used in the product of complex numbers.

example
in "mul" no fma is emitted while is emitted for the inline product in main
(the fma result is more precise!)
so that for
gcc version 15.0.0 20240925 (experimental) [master r15-3869-g7cf85d137ba] (GCC)
c++ -O3 cplxMul.cpp -mfma
./a.out 0x1.6912cap-41f -0x1.69020ap-41f 0x1.d16ed4p-85f 0x1.6b497ep-110f
t1=(0x1.483bbap-125,-0x1.482c7ep-125)
t2=(0x1.483bbcp-125,-0x1.482c7cp-125)

assembler output in
https://godbolt.org/z/q4bYMffKd

#include <cstdio>
#include <complex>

// ./a.out 0x1.6912cap-41f -0x1.69020ap-41f 0x1.d16ed4p-85f 0x1.6b497ep-110f

std::complex<float>
__attribute__ ((noinline))
mul(std::complex<float> const & a, std::complex<float> const & b)
{
   return a*b;
}

int main(int argc, char** argv)
{
 std::complex<float> z1{atof(argv[1]),atof(argv[2])};
 std::complex<float> z2{atof(argv[3]),atof(argv[4])};
 std::complex<float> t1 = z1*z2;
 std::complex<float> t2 = mul(z1,z2);
 printf ("t1=(%a,%a)\n", t1.real(), t1.imag());
 printf ("t2=(%a,%a)\n", t2.real(), t2.imag());
 return 0;
}