SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Politics : Formerly About Advanced Micro Devices

 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: Petz who wrote (44934)1/5/1999 7:42:00 PM
From: Tenchusatsu  Read Replies (1) of 1571198
 
<Of course it would have to be modified somewhat to fit x86 memory access, but KNI changed registers to implement it's SIMD extensions. Atiq Raza mentioned 128 bit registers and I'm sure AMD is looking at AltiVec.>

It's not as simple as just applying an additional instruction subset or more registers and using a context switch. There is a reason why Intel chose to add only 8 128-bit registers with KNI; in the x86 instruction format, there are only 3 bits in a register field, meaning it can only address 8 registers. Of course, register renaming helps to alleviate some of the restrictions that a small number of registers impose, but it's still less powerful than just adding 32 architected registers.

KNI was designed with x86 in mind, which is why it's slightly less powerful compared to AltiVec. Sure, AMD could implement AltiVec in an x86 processor, but this would turn into a form of hardware emulation because AltiVec is so very different architecturally from x86. AMD could try to modify AltiVec to make it easier to decode in an x86 context. AMD could even try extending 3D-Now to make it more powerful than KNI or AltiVec. Yet either way requires a ton of effort to design and validate. And ROI would thus be questionable.

Tenchusatsu
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext