https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116891
Bug ID: 116891
Summary: invalid optimization of -fma(-x,y,-z) when -03 and
-frounding-math are used
Product: gcc
Version: 14.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: walter.mascarenhas at gmail dot com
Target Milestone: ---
GCC generates invalid code for this statement
x = - fma( -y, z, -w )
with -frounding-math. It "optimizes" it to
x = fma( y, z, w )
but this is incorrect when the rounding mode is downwards or upwards.
In more detail, for the function
double bad_fma( double y, double z, double w) {
return std::fma( -y, z, -w );
}
with flags -O3 -mfma -frounding-math
I get this in godbolt, with gcc 14.2:
bad_fma:
vfmadd132sd xmm0, xmm2, xmm1
ret
That is, gcc incorrectly "optimizes" the code.
