Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cwjcc!tut.cis.ohio-state.edu!XEROX.COM!Leisner.Henr
From: Leisner.Henr@XEROX.COM (Marty)
Newsgroups: gnu.gcc.bug
Subject: problems when BITS_PER_UNIT == 1
Message-ID: <890511-152757-4322@Xerox>
Date: 11 May 89 22:27:49 GMT
Sender: daemon@tut.cis.ohio-state.edu
Distribution: gnu
Organization: GNUs Not Usenet
Lines: 51

I'm writing a machine description for the TI34010.

I got a minimal description together, made a cross compiler with GCC 1.35
on a sun386i.

After spending a day defining enough stuff to run the compiler, a got an
abort() in stor-layout.c/layout_type() for a REAL_TYPE.

Tracing through it, I found out the following:

My tm.h says:
/* number of bits in an addressible storage unit */
#define BITS_PER_UNIT 1		/* ti 34010 is a bit addressable processor with
variable size fields */

The following code in layout_type triggered abort():

    case REAL_TYPE:
      {
	register int prec = TYPE_PRECISION (type);
	if (prec <= GET_MODE_BITSIZE (SFmode))
	  TYPE_MODE (type) = SFmode;
	else if (prec <= GET_MODE_BITSIZE (DFmode))
	  TYPE_MODE (type) = DFmode;
	else
	  abort ();
      }

GET_MODE_BITSIZE(x) is defined in rtl.h as
/* Get the size in bits of an object of mode MODE.  */

#define GET_MODE_BITSIZE(MODE)  (BITS_PER_UNIT * mode_size[(int)(MODE)])

mode_size[MODE] returns # of bytes in MODE.

My best patch I see is to define:
#define GET_MODE_BITSIZE(MODE)  (8 * mode_size[(int)(MODE)])
since mode_size is defined in bytes.

I'm concerned there may be other assumptions in other places that
BITS_PER_UNIT == 8 (all the current machine descriptions have it).  

Am I on target?  I'm making the change right now -- does it seem
reasonable.

marty
ARPA:	leisner.henr@xerox.com
GV:  leisner.henr
NS:  martin leisner:wbst139:xerox
UUCP:	hplabs!arisia!leisner