GAS support for an IEEE variation

[Richard Sandiford]

I recently needed GAS to support a variation of the IEEE standard
in which the largest exponent could be used for normal numbers instead
of NaNs.  This patch adds a target option TC_LARGEST_EXPONENT_IS_NORMAL,
that can be defined to select this kind of format.

Please install if OK.

Richard



2001-03-02  Richard Sandiford  <rsandifo@redhat.com>

	* config/atof-ieee.c (TC_LARGEST_EXPONENT_IS_NORMAL): New macro.
	(gen_to_words): Print warnings if NaNs are found and the target CPU
	does not support them.  Allow largest exponent to be used in normal
	numbers if TC_LARGEST_EXPONENT_IS_NORMAL evaluates to true.

Index: ./gas/config/atof-ieee.c
===================================================================
RCS file: /cvs/src/src/gas/config/atof-ieee.c,v
retrieving revision 1.6
diff -c -p -d -r1.6 atof-ieee.c
*** ./gas/config/atof-ieee.c	2000/08/07 19:54:34	1.6
--- ./gas/config/atof-ieee.c	2001/03/02 13:48:05
***************
*** 1,5 ****
  /* atof_ieee.c - turn a Flonum into an IEEE floating point number
!    Copyright (C) 1987, 92, 93, 94, 95, 96, 97, 98, 99, 2000
     Free Software Foundation, Inc.

     This file is part of GAS, the GNU Assembler.
--- 1,5 ----
  /* atof_ieee.c - turn a Flonum into an IEEE floating point number
!    Copyright (C) 1987, 92, 93, 94, 95, 96, 97, 98, 99, 2000, 2001
     Free Software Foundation, Inc.

     This file is part of GAS, the GNU Assembler.
***************
*** 19,24 ****
--- 19,31 ----
     Software Foundation, 59 Temple Place - Suite 330, Boston, MA
     02111-1307, USA.  */

+ /* Some float formats are based on the IEEE standard, but use the
+    largest exponent for normal numbers instead of NaNs and infinites.
+    The macro TC_LARGEST_EXPONENT_IS_NORMAL should evaluate to true
+    if the target machine uses such a format.  The macro can depend on
+    command line flags if necessary.  There is no need to define the
+    macro if it would always be 0.  */
+
  #include "as.h"

  /* Flonums returned here.  */
*************** extern const char EXP_CHARS[];
*** 40,45 ****
--- 47,56 ----
  /* Length in LittleNums of guard bits.  */
  #define GUARD (2)

+ #ifndef TC_LARGEST_EXPONENT_IS_NORMAL
+ #define TC_LARGEST_EXPONENT_IS_NORMAL 0
+ #endif
+
  static const unsigned long mask[] =
  {
    0x00000000,
*************** gen_to_words (words, precision, exponent
*** 291,296 ****
--- 302,309 ----
    /* NaN:  Do the right thing.  */
    if (generic_floating_point_number.sign == 0)
      {
+       if (TC_LARGEST_EXPONENT_IS_NORMAL)
+ 	as_warn ("NaNs are not supported by this target\n");
        if (precision == F_PRECISION)
  	{
  	  words[0] = 0x7fff;
*************** gen_to_words (words, precision, exponent
*** 328,333 ****
--- 341,349 ----
      }
    else if (generic_floating_point_number.sign == 'P')
      {
+       if (TC_LARGEST_EXPONENT_IS_NORMAL)
+ 	as_warn ("Infinities are not supported by this target\n");
+
        /* +INF:  Do the right thing.  */
        if (precision == F_PRECISION)
  	{
*************** gen_to_words (words, precision, exponent
*** 366,371 ****
--- 382,390 ----
      }
    else if (generic_floating_point_number.sign == 'N')
      {
+       if (TC_LARGEST_EXPONENT_IS_NORMAL)
+ 	as_warn ("Infinities are not supported by this target\n");
+
        /* Negative INF.  */
        if (precision == F_PRECISION)
  	{
*************** gen_to_words (words, precision, exponent
*** 578,584 ****

        return return_value;
      }
!   else if ((unsigned long) exponent_4 >= mask[exponent_bits])
      {
        /* Exponent overflow.  Lose immediately.  */

--- 597,605 ----

        return return_value;
      }
!   else if ((unsigned long) exponent_4 > mask[exponent_bits]
! 	   || (! TC_LARGEST_EXPONENT_IS_NORMAL
! 	       && (unsigned long) exponent_4 == mask[exponent_bits]))
      {
        /* Exponent overflow.  Lose immediately.  */