[X86] Bugfix for rL349334 adding a check function for two intrinsics

 _mm_mmask_i64gather_epi32 and _mm_mmask_i64gather_ps operates the
 lower 64 bits and zeroes the higher 64 bits of the return value.
 The old test suite steps 64 bits in the do_intrin_loop, which will
 overlap higher 64 bits of other data address in the last iteration.
 Especially when compiler malloc dst128_f and mask128 's address
 adjacent, this test will run fail. A new check function is added
 to check these two intrinsics.

 Differential Revision: https://reviews.llvm.org/D79158
1 file changed