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.
Technology Stocks : Intel Corporation (INTC)
INTC 47.76+1.3%10:17 AM EST

 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: Paul Engel who wrote (5428)11/18/1996 11:42:00 PM
From: Anthony Mascarenhas   of 186894
 
Paul,

<< Clearly, for some applications, this may be an unacceptable trade off, yet for others it
may be quite acceptable. We will have to wait and see what kind of MMX optimized
software is produced - and how they perform across the Pentium MMX, AMD's K6, and
Cyrix's M2.>>

For legacy applications this is a non-issue as they wouldn't use
the MMX instructions. For newly developed or modified applications
following the guidelines minimizes context switches (and the 50
clock cycles). The application writer needs to decide wether the
performance gain from using the MMX instructions is worth the performance
hit for context switching. I suspect that Intel has analyzed the
application mix, and deemed it worth to use MMX registers aliased to
the floating point registers (i.e more applications tend to benefit
from using MMX. In the worst case there would not be any performance
degradation - don't use MMX at all).

As for the second issue, Each application resides in its own process
address space and is protected from others using its state. So if
an errant program fails to save/restore the MMX state then it is
that application which would fail without causing the rest of the
system to fall over.

Well had to go into more techno babble about this, but I wouldn't
loose any sleep over it.

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