https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98169
--- Comment #2 from denis.campredon at gmail dot com ---
This also applies to vector types.
-------
typedef float __attribute__((vector_size(8))) T;
T f(T a) {
return a != a;
}
-------
Gcc could generate:
------
f:
xorps xmm1, xmm1
cmpunordps xmm0, xmm1
ret
------
But instead generate a less optimal code:
------
f:
ucomiss xmm0, xmm0
mov edx, -1
movaps xmm1, xmm0
mov eax, 0
mov ecx, edx
shufps xmm1, xmm1, 0xe5
cmovnp ecx, eax
ucomiss xmm1, xmm1
movd xmm0, ecx
cmovp eax, edx
movd xmm2, eax
punpckldq xmm0, xmm2
ret
