Path: utzoo!mnetor!uunet!seismo!sundc!pitstop!sun!decwrl!labrea!agate!pasteur!ames!hao!gatech!purdue!i.cc.purdue.edu!k.cc.purdue.edu!l.cc.purdue.edu!cik
From: cik@l.cc.purdue.edu (Herman Rubin)
Newsgroups: comp.arch
Subject: Re: Condition Codes in General Registers
Message-ID: <682@l.cc.purdue.edu>
Date: 18 Feb 88 20:19:03 GMT
References: <191@telesoft.UUCP> <1556@gumby.mips.COM> <208@telesoft.UUCP> <868@ut-emx.UUCP>
Organization: Purdue University Statistics Department
Lines: 50
Summary: This is not what was asked

In article <868@ut-emx.UUCP>, reeder@ut-emx.UUCP writes:
> In article <6834@sol.ARPA>, crowl@cs.rochester.edu (Lawrence Crowl) writes:
	......
> >     compare R1 with R2 place condition code in R3
> >     branch to ADDRESS if condition code in R3 is GREATER
> >     branch to ADDRESS if condition code in R3 is EQUAL
> >     branch to ADDRESS if condition code in R3 is LESS
	......
> That is exactly how you do things on a Cray, for example:
> 
>           A0      A1-A2                 Compare A1 and A2
>           JAP     addr1                 Branch to addr1 if A1 > A2
>           JAZ     addr2                 Branch to addr2 if A1 = A2
>           JAN     addr3                 Branch to addr3 if A1 < A2

This is not the same.  The first instruction computes a value and 
inserts it in A0.  The other instructions perform a test on the value
in A0 and use _that_, not the bits in a condition code, to decide
whether to perform the branch.

To clearly see the difference, suppose that several condition codes
were involved in a branch situation.  On a Cray, the appropriate values
would have to be stored in various A or S registers, and _for each
situation for which a jump is contemplated_, the appropriate value
would have to be stored in A0 or S0 and then tested.  It is not possible
to directly move A0 or S0 to another A or S register, either.  On 
machines which can do the more common type of tests, it is no faster
to compute the difference and transfer on it rather than transferring
on the comparison.

Furthermore, this does not allow the condition codes to be used for
overflow or carry.  Of course, the Cray is not too well suited for
integer multiplication or any kind of division, but it is well suited
for integer addition and shifts, and that is where overflow and carry
are important.

Let me make some additional suggestions apropos of Crowl's posting.
It should be possible to address an arbitrary byte of a register
(this is useful for other purposes).  One way of doing this is to
make a small address a reference to the register file; on virtual
machines this would be costless, and the amount of memory lost on
non-virtual machines would be negligeable.  Some machines have done
this, but I do not believe that they have done it well; either this
can only be used in certain situations, or the user's registers are
in blocks and/or separated from main memory.  If we do this, we can
keep many condition codes active for transfers.
-- 
Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907
Phone: (317)494-6054
hrubin@l.cc.purdue.edu (ARPA or UUCP) or hrubin@purccvm.bitnet