From mboxrd@z Thu Jan  1 00:00:00 1970
From: helbig@Informatik.BA-Stuttgart.DE (Wolfgang Helbig)
Date: Sun, 12 Jan 2003 00:16:42 +0100 (MET)
Subject: [TUHS] unsigned division in V7's C
Message-ID: <200301120120.h0C1KrR00556@bsd.korb>

Warren replies:
>Not sure if:
>
>http://www.tuhs.org/Archive/PDP-11/Bug_Fixes/Net.v7bugs/0017
>
>is a bug fix for this.

And Dennis replies:
>Does

>  http://www.cs.bell-labs.com/who/dmr/odd.html#muldiv

>suffice to explain the behavior

Thanks for the replies, but the bug I found is about unsigned int's and
not long int's.

The V7 C code simply the DIV instruction for unsigned division. The dividend
is put in the low word of a register pair with the upper word cleared.

When the divisor is >= 2^15, DIV interprets it as a negative integer
and produces the result (N = 2^16):
	 N - a/(N-b)	instead of a/b, and
	 a%(N-b)	instead of a%b.

This explains the bug I've sent yesterday. I is easily fixed, since
when b >= N/2 and a < N, then a/b is either zero or one.

And there is another bug when using DIV for unsigned integers: If b = 1 and
a >= N/2, then a/b >= N/2, that is a signed overflow. With V7 on Bob Supniks
simulator (2.10), I get
	60000/1: 0, 60000%1: 60000
DIV did not change the dividend because of overflow.

Again a fix is easy: With unsigned ints, don't use DIV to divide by one.

In all other cases, DIV produces the correct unsigned results, if the dividend
is less N, that is its high word is zero.

I wonder why these bugs went unnoticed.

Greetings

Wolfgang