3 ms·
> They are simpler to decode One of the big headaches with x86 superscalar machines is that finding the instruction boundaries is hard. The decoders, looking a
by Animats 4y ago
> They are simpler to decode
One of the big headaches with x86 superscalar machines is that finding the instruction boundaries is hard. The decoders, looking ahead, may guess at alignment and decode bytes which represent a totally bogus instruction, which causes work that gets discarded further downstream. Intel and AMD did this very differently in the early days of superscalar.
If you're going to have a variable length instruction set, it would be nice if it worked like UTF-8, where you can start anywhere in a stream and get aligned within a few bytes. x86 is not like that.