24 Oct
2005
24 Oct
'05
8:39 a.m.
On Mon, 24 Oct 2005 05:00:57 +0100 (BST), ard at p850ug1.demon.co.uk
(Tony Duell) wrote:
NO, just for fun... I doubt very much I could improve on the
performance
for running the VAX instruction set. I have not looked at the prints
carefully enough to determine how much of the instruction set is
hard-wired (if any), I wonder if it would be possible to run a
different
instruction set entirely.
-tony
Actually, with access to the micro instructions you can often
substantially improve the run-time of a given program. The Modcomp II
had the microbus accessible for use along with a number of unused
instructions. The communications instructions and floating point were
implemented using this bus.
I implemented an instruction store attached to this bus and which
allowed me to create and store new instructions. A friend working on
a CS Phd in pattern matching ginned up a program that would find
common instruction sequences in a program, deconstruct the microcode,
perform optimization on the sequence and then create a new
instruction which was substituted for the original sequence. We often
got increases of performance of 20% over the original code.
IIRC the Burroughs computers would load a different instruction set
depending on what language was being used. Tony could create a VAX
with Forth as the instruction set ;-)
CRC
The
11/730, of course, is mostly PALs (as I said) with some RAM as a
control store, 8 off 2901s as the ALU, and an 8085 (I think) to
load the
control store, etc. One day I am going to look into modifying the
microcode of that machine...
To what end? Increased performance?