Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!elroy.jpl.nasa.gov!sdd.hp.com!ucsd!ucbvax!hplabs!hpfcso!cunniff
From: cunniff@hpfcso.HP.COM (Ross Cunniff)
Newsgroups: comp.sys.amiga.programmer
Subject: Re: SAS Bugs
Message-ID: <101060005@hpfcso.HP.COM>
Date: 28 Jan 91 16:10:19 GMT
References: <91028.004246T32QC@CUNYVM.BITNET>
Organization: Hewlett-Packard, Fort Collins, CO, USA
Lines: 51

>     Has anyone encountered any bugs in 5.10 with float math calculations in a
>function and trying to pass the return values either with a global variable or
>a typedef struct of floating point values. Then passing the values to the main
>values dont truly get passed. I compile the progs same on a friends machine in
>5.04 and it works fine. I am sick of writing code and seeing it is wrong on my
>machine and its not my fault. I need help.

Well, this weekend, I just got bitten by the following nasty bug (this is
cut down from a much larger example, so no comments on programming style,
please).  The function:

	func( a, b )
	float a;
	long *b;
	{
	    *(float *)b = a;
	}

generated the code (this is from memory, and probably the wrong syntax as
well; that's what I get for using the HP assembler so much...)

	link.l	%a6,#0000
	mov.l	#000C(%a6),%a0			; Note this line; should
						; be #0010(%a6)
	movm.l	#0008(%a6),%d0-%d1
	jsr.l	convert_single_to_double	; don't remember the exact name
	mov.l	%d0,(%a0)
	unlk	%a6
	rts

It looks like the compiler realizes that it is actually a double on the
stack (hence the convert_single_to_double (or whatever nasty mnemonic they
use) call), but for the purposes of argument addressing, it thinks that
the size of the argument on the stack is *4* (not 8) bytes.  The workaround
is to rewrite the function as:

	func( a, b )
	double a;	/* Used to be float a */
	long *b;
	{
	    *(float *)b = a;
	}

Additionally, my application no longer works at all when compiled to
use the FFP math routines.  I haven't had time to track this down yet
(I have over 35k lines to search :-(), but when I do, I'll call SAS
and discuss it with them...

				Ross Cunniff
				Hewlett-Packard Colorado Language Lab
				cunniff@hpfcla.HP.COM