Constantine, Re: "Why would the P4 team omit something as lightweight as a simple decoder?"
If it were so simple, they wouldn't have omitted it. My guess is that an extra decoder might have:
a) limited clock frequencies
b) taken up too much die size
c) complicated certain logic
While weighing in all these things, they might have done performance estimates, and found it to be 1% or 2%, which might not have made it worth it. Keep in mind that in the far majority of the code, the trace cache is said to have a hit rate of a 8-16KB iL1 cache, which is around 85%. Optimizations can make that hit rate even larger.
But this is just a WAG. The truth, though, is that there is a lot involved in designing a new architecture - and nothing is done "just for the hell of it". There is always an educated reason behind it. That's why I consider this exercise so futile.
If, for some unknown reason, the architects *did* put a half speed decoder inside, they would have had a good reason. They wouldn't be blind to any bottleneck obvious to a bunch of forum posters like ourselves. All Kapkan and Jozef are doing is trying to prove that they are smarter than an experienced team of CPU designers. It's idiotic.
wbmw |
