By: Paul DeMone (pdemone.delete@this.igs.net), August 19, 2004 6:22 am
Room: Moderated Discussions
Dresdenboy (M.Waldhauer@gmx.de) on 8/19/04 wrote:
---------------------------
>According to my idea to use microcode patching to create customized instructions,
>which use microcode to do things more efficient than it would be possible with x86
>code (e.g. MUL), I think that there are several nice candidates. We just have to
>look for vector decoded x87 or MMX codes.
Ahhh, the home fallout shelter of computer science.
Customizing and/or extending vendor instructions sets
in application specific fashion using microcode is a
very old idea in computers and this capability has been
offered by vendors several times in the past. Writing
functional microcode is very difficult and the end result
is extremely non-portable. In practice it was seldom if
ever used by customers even in the days prior to VLSI
when microarchitectures were highly self-similar and
evolved at glacial pace compared to monolithic MPUs
today. Vendors ended up spending far more developing
microcode programming tools for end users than they
ever recouped from increased sales.
So IMO it didn't make sense 30 years ago and it makes
even less sense today.
---------------------------
>According to my idea to use microcode patching to create customized instructions,
>which use microcode to do things more efficient than it would be possible with x86
>code (e.g. MUL), I think that there are several nice candidates. We just have to
>look for vector decoded x87 or MMX codes.
Ahhh, the home fallout shelter of computer science.
Customizing and/or extending vendor instructions sets
in application specific fashion using microcode is a
very old idea in computers and this capability has been
offered by vendors several times in the past. Writing
functional microcode is very difficult and the end result
is extremely non-portable. In practice it was seldom if
ever used by customers even in the days prior to VLSI
when microarchitectures were highly self-similar and
evolved at glacial pace compared to monolithic MPUs
today. Vendors ended up spending far more developing
microcode programming tools for end users than they
ever recouped from increased sales.
So IMO it didn't make sense 30 years ago and it makes
even less sense today.