|
|
|
|
|
by aleph_minus_one
677 days ago
|
|
>
If they had delivered a version of Windows for ARM capable of seamlessly emulating x86 code with no compromises This is hardly possible on most ARM chips because x86 has a much stronger memory model than ARM. Also concerning "with no compromises": common x86 implementations have a very fast implementation of the SIMD instruction sets (SSE..., AVX/AVX2, perhaps AVX-512) that is much slower to emulate on ARM because their SIMD instruction set is different. The only reason why people don't realize this is that a lot of common software makes no intense use of these SIMD instructions. Then there are the subtle parts that (as far as I am aware) the ARM FPU handles multiplication of denormalized floating-point numbers slightly different than x86 (both implementations are allowed by the standard) etc. |
|