https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115161
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Trying
#include <emmintrin.h>
int
main ()
{
float f = 0x0.8p+33f;
float __attribute__((vector_size (16))) vf = { 0x0.8p+33f, 0x0.8p+33f,
0x0.8p+33f, 0x0.8p+33f };
int a = f;
int __attribute__((vector_size (16))) vi = __builtin_convertvector (vf, int
__attribute__((vector_size (16))));
int __attribute__((vector_size (16))) vi2 = (int __attribute__((vector_size
(16)))) _mm_cvttps_epi32 ((__m128) vf);
__builtin_printf ("%d\n", a);
__builtin_printf ("{%d, %d, %d, %d}\n", vi[0], vi[1], vi[2], vi[3]);
__builtin_printf ("{%d, %d, %d, %d}\n", vi2[0], vi2[1], vi2[2], vi2[3]);
}
with gcc and clang both with -O0 and -O2, this prints
-2147483648
{-2147483648, -2147483648, -2147483648, -2147483648}
{-2147483648, -2147483648, -2147483648, -2147483648}
with both compilers at -O0, clang at -O2 prints
-1720305736
{8524448, 0, 1, 135169}
{-2147483648, -2147483648, -2147483648, -2147483648}
i.e. complete garbage for the non-intrinsic cases, but what the HW does for the
intrinsic cases, while gcc at -O2 prints
2147483647
{2147483647, 2147483647, 2147483647, 2147483647}
{2147483647, 2147483647, 2147483647, 2147483647}
i.e. treats intrinsics and non-intrinsics the same.
GCC behaves this way consistently since GCC 9 (before __builtin_convertvector
hasn't been supported), but if the vi related lines are commented out, even GCC
4.7 works that way.
