https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106081
--- Comment #1 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
This is an attempt to vectorize by hand, but it seems we do not generate
vpmovsxwd for the vector short->double conversion
struct pixels
{
short a __attribute__ ((vector_size(4*2)));
} *pixels;
struct dpixels
{
double a __attribute__ ((vector_size(8*4)));
};
typedef double v4df __attribute__ ((vector_size (32)));
struct dpixels
test(double *k)
{
struct dpixels results={};
for (int u=0; u<10000;u++,k--)
{
results.a += *k*__builtin_convertvector (pixels[u].a, v4df);
}
return results;
}
clang seems to do right thing here.