On Sat, Jan 23, 2021 at 01:03:31AM +0100, Jakub Jelinek via Gcc-patches wrote:
> The problem is that the testcase uses the
> _mm_loadl_pi
> API, and per the Intel intrinsic rules it is ok when that intrinsic
> loads from wide range of types, e.g. including pairs of integers or
> 4 shorts or 8 chars or pair of floats.
And note that this isn't specific to the _mm_loadl_pi intrinsic (and other
load and store intrinsics operating with __m64 *), user code can also do
__m64 *p = whatever;
... = *p;
and expect it to work, even if whatever doesn't point to long long or
unsigned long long, but pair of ints, ...
Jakub
