An i386 fp register name patch

An i386 fp register name patch

[H.J. Lu]

Here is a patch to handle spaces ina i386 FP register names. I am
enclosing a testcase here.

----
	fxch   %st( 1)
	fxch   %st( 1 )
	fxch   %st ( 1 )
	fxch   %st (1)
	fxch   %st(1)
	fxch   % st(1)
	fxch   %           st(1)
----

Thanks.

-- 
H.J. Lu (hjl@gnu.org)
--
Fri Aug 27 11:31:30 1999  H.J. Lu  (hjl@gnu.org)

	* config/tc-i386.c (i386_extract_reg): New function. Handle
	spaces in FP register names.
	(i386_is_reg): Use it.
	(parse_register): Likewise.

Index: config/tc-i386.c
===================================================================
RCS file: /work/cvs/gnu/binutils/gas/config/tc-i386.c,v
retrieving revision 1.5
diff -u -p -r1.5 tc-i386.c
--- config/tc-i386.c	1999/08/21 18:02:59	1.5
+++ config/tc-i386.c	1999/08/27 18:24:16
@@ -2507,25 +2507,57 @@ md_assemble (line)
   }
 }
 \f
+static char * i386_extract_reg PARAMS ((char *, char *, int));
+
+static char *
+i386_extract_reg (s, reg, len)
+     char *s;
+     char *reg;
+     int len;
+{
+  register char *p;
+
+  while (is_space_char (*s))
+    ++s;
+
+  p = reg;
+  while (register_chars[(unsigned char) *s] != '\0'
+	  || is_space_char (*s))
+    {
+      if (p >= &reg [2] && tolower (reg [0]) == 's'
+	  && tolower (reg [1]) == 't'
+	  && (p == &reg [2]
+	      || (reg [2] == '('
+		  && (p == &reg [3]
+		      || (*(p - 1) >= '0' && *(p - 1) <= '9')))))
+	/*
+	  1. %st (1).
+	  2. %st ( 1).
+	  3. %st ( 1 ).
+	*/
+      while (is_space_char (*s))
+	++s;
+
+      if ((*p++ = register_chars[(unsigned char) *s++]) == '\0')
+        break;
+      if (p >= reg + len)
+        return NULL;
+    }
+
+  *p++ = register_chars[(unsigned char) *s++];
+  return (p >= reg + len) ? NULL: s;
+}
+
 static int i386_is_reg PARAMS ((char *));
 
 static int
 i386_is_reg (reg_string)
      char *reg_string;
 {
-  register char *s = reg_string;
-  register char *p;
   char reg_name_given[MAX_REG_NAME_SIZE + 1];
-
-  if (is_space_char (*s))
-    ++s;
 
-  p = reg_name_given;
-  while ((*p++ = register_chars[(unsigned char) *s++]) != '\0')
-    if (p >= reg_name_given + MAX_REG_NAME_SIZE)
-      return 0;
-
-  if (!hash_find (reg_hash, reg_name_given))
+  if (!i386_extract_reg (reg_string, reg_name_given, MAX_REG_NAME_SIZE)
+      || !hash_find (reg_hash, reg_name_given))
     return 0;
   else
     return 1;
@@ -4137,7 +4169,6 @@ parse_register (reg_string, end_op)
      char **end_op;
 {
   register char *s = reg_string;
-  register char *p;
   char reg_name_given[MAX_REG_NAME_SIZE + 1];
   const reg_entry *r;
 
@@ -4145,23 +4176,16 @@ parse_register (reg_string, end_op)
   if (*s == REGISTER_PREFIX)
     ++s;
 
-  if (is_space_char (*s))
-    ++s;
-
-  p = reg_name_given;
-  while ((*p++ = register_chars[(unsigned char) *s++]) != '\0')
+  s = i386_extract_reg (s, reg_name_given, MAX_REG_NAME_SIZE);
+  if (s == NULL)
     {
-      if (p >= reg_name_given + MAX_REG_NAME_SIZE)
+      if (!allow_naked_reg)
 	{
-	  if (!allow_naked_reg)
-	    {
-	      *p = '\0';
-	      as_bad (_("bad register name `%s'"), reg_name_given);
-	    }
-	  return (const reg_entry *) NULL;
+	  reg_name_given [MAX_REG_NAME_SIZE] = '\0';
+	  as_bad (_("bad register name `%s'"), reg_name_given);
 	}
+      return (const reg_entry *) NULL;
     }
-
   *end_op = s - 1;
 
   r = (const reg_entry *) hash_find (reg_hash, reg_name_given);

Re: An i386 fp register name patch

[Alan Modra]

On Fri, 27 Aug 1999, H.J. Lu wrote:

> Here is a patch to handle spaces ina i386 FP register names. I am
> enclosing a testcase here.

Hello HJ,
  As you probably know, we already allow spaces in lots of places.  eg.
	mov % al , 5 ( % ebp , % ebx , 1 )
Notice that no spaces are allowed *in* a register name, or for that
matter in an instruction mnemonic.  eg.
	mov $1, % a l
is illegal.

So there is a philosophical question as to whether "st(1)" is really an
indivisible register name token or not.  You seem to say that the register
name is "st" with an index, so therefore we should allow spaces, eg.
"st ( 1 )".  I'm inclined to say that "st(1)" is really just a name like
any other register name, so spaces shouldn't be allowed (or if we do
allow spaces here then why not in "% a l" - something I don't like).

Where did you run into a need for spaces in fp reg names?  C asm macros?

Re: An i386 fp register name patch

[H.J. Lu]

> 
> 
> On Fri, 27 Aug 1999, H.J. Lu wrote:
> 
> > Here is a patch to handle spaces ina i386 FP register names. I am
> > enclosing a testcase here.
> 
> Hello HJ,
>   As you probably know, we already allow spaces in lots of places.  eg.
> 	mov % al , 5 ( % ebp , % ebx , 1 )
> Notice that no spaces are allowed *in* a register name, or for that
> matter in an instruction mnemonic.  eg.
> 	mov $1, % a l
> is illegal.
> 
> So there is a philosophical question as to whether "st(1)" is really an
> indivisible register name token or not.  You seem to say that the register
> name is "st" with an index, so therefore we should allow spaces, eg.
> "st ( 1 )".  I'm inclined to say that "st(1)" is really just a name like
> any other register name, so spaces shouldn't be allowed (or if we do
> allow spaces here then why not in "% a l" - something I don't like).
> 
> Where did you run into a need for spaces in fp reg names?  C asm macros?
> 
> 

I got a report from Mesa people. FWIW, assembler on Solaris &/x86 takes

        fxch   %st( 1) 
        fxch   %st( 1 )
        fxch   %st ( 1 )
        fxch   %st (1) 
        fxch   %st(1)  

But not

        fxch   % st(1) 
        fxch   %           st(1)   

I got

# cat foo.s
        fxch   %st( 1)
        fxch   %st( 1 )
        fxch   %st ( 1 )
        fxch   %st (1)
        fxch   %st(1)
        fxch   % st(1)
        fxch   %           st(1)
# /usr/ccs/bin/as -o foo.o foo.s 
Assembler: 
        aline 6 : Illegal register
        aline 6 : syntax error
        aline 7 : Illegal register
        aline 7 : syntax error

It looks like gas has some extensions, but doesn't have features
supported by other x86 assemblers. In any case, I will support it
in the Linux version of binutils.

-- 
H.J. Lu (hjl@gnu.org)

Re: An i386 fp register name patch

[Ian Lance Taylor]

   Date: Sat, 28 Aug 1999 22:34:51 +0930 (CST)
   From: Alan Modra <alan@SPRI.Levels.UniSA.Edu.Au>

     As you probably know, we already allow spaces in lots of places.  eg.
	   mov % al , 5 ( % ebp , % ebx , 1 )
   Notice that no spaces are allowed *in* a register name, or for that
   matter in an instruction mnemonic.  eg.
	   mov $1, % a l
   is illegal.

   So there is a philosophical question as to whether "st(1)" is really an
   indivisible register name token or not.  You seem to say that the register
   name is "st" with an index, so therefore we should allow spaces, eg.
   "st ( 1 )".  I'm inclined to say that "st(1)" is really just a name like
   any other register name, so spaces shouldn't be allowed (or if we do
   allow spaces here then why not in "% a l" - something I don't like).

While philosophically %st(1) is a single register name, I think that
given the format of the name we should allow spaces around the
parenthesis.  ``Be liberal in what you accept.''

On the other hand, I don't think we should accept "% a l".  That just
seems too error prone.

Ian