Incorrect code generated with m68k-elf-as

Incorrect code generated with m68k-elf-as

[pjak]

I have a problem when assembling m68k-elf code.
The code that causes the problem is pretty long
(about 20k), it can be found here:

http://www.laseranimation.com/peter/error6.s

When assembling this with "m68k-elf-as -a error6.s"
everything seems to be ok.
But without listing the generated code is incorrect.

A diff of a dump of the a.out files:

@@ -162,8 +162,8 @@
  20c:  33c0 0000 0000  movew %d0,0 <_a5625850_>
  212:  70ff            moveq #-1,%d0
  214:  33c0 0000 0000  movew %d0,0 <_a5625850_>
- 21a:  600c            bras 228 <_a5626516_+0x5e>
- 21c:  4879 0000 0000  pea 0 <_a5625850_>
+ 21a:  6000 4879       braw 4a95 <_a5626102_+0x3e99>
+ 21e:  0000 0000       orib #0,%d0
  222:  4eb9 0000 0000  jsr 0 <_a5625850_>
  228:  4e5e            unlk %fp
  22a:  4e75            rts
@@ -312,8 +312,7 @@
  416:  4a79 0000 0000  tstw 0 <_a5625850_>
  41c:  673c            beqs 45a <_a5627150_+0x4c>
  41e:  4a80            tstl %d0
- 420:  660a            bnes 42c <_a5627150_+0x1e>
- 422:  4200            clrb %d0
+ 420:  6600 4200       bnew 4622 <_a5626102_+0x3a26>
  424:  13c0 0000 0000  moveb %d0,0 <_a5625850_>
  42a:  6018            bras 444 <_a5627150_+0x36>
  42c:  2f00            movel %d0,%sp@-

The branches at 21a and 420 are messed (incorrect
offset 0).

Peter Jakubek
pjak@snafu.de

Re: Incorrect code generated with m68k-elf-as

[amodra]

On Sun, May 27, 2001 at 04:46:24AM +0100, pjak@snafu.de wrote:
> 
> The branches at 21a and 420 are messed (incorrect
> offset 0).

Wow, this bug predates gas-1.38.1  Fixed like this

gas/ChangeLog
	* config/tc-m68k.c (md_assemble): Ensure variable part of frag is
	allocated in the same chunk as the fixed part.

-- 
Alan Modra
 
Index: gas/config/tc-m68k.c
===================================================================
RCS file: /cvs/src/src/gas/config/tc-m68k.c,v
retrieving revision 1.22
diff -u -p -r1.22 tc-m68k.c
--- tc-m68k.c	2001/05/22 09:01:55	1.22
+++ tc-m68k.c	2001/05/27 11:38:36
@@ -3664,6 +3664,21 @@ md_assemble (str)
     }
 
   /* There's some frag hacking */
+  {
+    /* Calculate the max frag size.  */
+    int wid;
+
+    wid = 2 * the_ins.fragb[0].fragoff;
+    for (n = 1; n < the_ins.nfrag; n++)
+      wid += 2 * (the_ins.numo - the_ins.fragb[n - 1].fragoff);
+    /* frag_var part.  */
+    wid += 10;
+    /* Make sure the whole insn fits in one chunk, in particular that
+       the var part is attached, as we access one byte before the
+       variable frag for byte branches.  */
+    frag_grow (wid);
+  }
+
   for (n = 0, fromP = &the_ins.opcode[0]; n < the_ins.nfrag; n++)
     {
       int wid;