Re: [avr-gcc-list] Compiler bug ?

[Marek Michalkiewicz]

On Fri, Apr 18, 2003 at 11:10:54AM -0600, E. Weddington wrote:
> Well it looks like the bug is still there in the 20030414 snapshot of 
> GCC 3.3. I get the same output as before.
> 
> I'm willing to wait for next week's snapshot to see if this can be 
> resolved.

Commenting out the whole "*cmpqi_sign_extend" insn in avr.md fixes
the bug, but makes inefficient code (the 8-bit value is sign-extended
to 16 bits and compared with -20 as a 16-bit value, when an 8-bit
compare would be much cheaper).  The following works better:

--- avr.md      28 Sep 2002 14:14:12 -0000      1.38
+++ avr.md      18 Apr 2003 22:17:02 -0000
@@ -1636,10 +1636,10 @@
 (define_insn "*cmpqi_sign_extend"
   [(set (cc0)
         (compare (sign_extend:HI
                  (match_operand:QI 0 "register_operand"  "d"))
-                (match_operand:HI 1 "immediate_operand" "M")))]
-  ""
+                (match_operand:HI 1 "const_int_operand" "n")))]
+  "INTVAL (operands[1]) >= -128 && INTVAL (operands[1]) <= 127"
   "cpi %0,lo8(%1)"
   [(set_attr "cc" "compare")
    (set_attr "length" "1")])


I'm not committing this just yet, as Denis may suggest a better fix...

Marek