== (bit_AVX | bit_CMPXCHG16B))
{
/* Intel SDM guarantees that 16-byte VMOVDQA on 16-byte aligned address
- is atomic, but so far we don't have this guarantee from AMD. */
+ is atomic, and AMD is going to do something similar soon.
+ We don't have a guarantee from vendors of other CPUs with AVX,
+ like Zhaoxin and VIA. */
unsigned int ecx2 = 0;
__get_cpuid (0, &eax, &ebx, &ecx2, &edx);
- if (ecx2 != signature_INTEL_ecx)
+ if (ecx2 != signature_INTEL_ecx && ecx2 != signature_AMD_ecx)
FEAT1_REGISTER &= ~bit_AVX;
}
#endif