Path: utzoo!utgpu!watmath!uunet!tut.cis.ohio-state.edu!GANG-OF-FOUR.STANFORD.EDU!weening
From: weening@GANG-OF-FOUR.STANFORD.EDU (Joe Weening)
Newsgroups: gnu.gcc.bug
Subject: Redundant test instructions in GCC 1.34 on m68k
Message-ID: <8903230625.AA03410@Gang-of-Four.Stanford.EDU>
Date: 23 Mar 89 06:25:29 GMT
Sender: daemon@tut.cis.ohio-state.edu
Distribution: gnu
Organization: GNUs Not Usenet
Lines: 72

I'm trying to figure out why the following program produces the assem-
bly code shown below, in GCC 1.34 on a Sun-2 running Sun OS 4.0.

--------------------------------------------------
Source code:

int i = 4;
int j = -5;
main () {
  if (i+j <= 0)
    printf ("negative\n");
  else printf ("non-negative\n");
}
--------------------------------------------------
Assembly code:

#NO_APP
gcc_compiled.:
.globl _i
.data
	.even
_i:
	.long 4
.globl _j
	.even
_j:
	.long -5
.text
LC0:
	.ascii "negative\12\0"
LC1:
	.ascii "non-negative\12\0"
	.even
.globl _main
_main:
	link a6,#0
	movel _i,d0
	addl _j,d0
	tstl d0
	jgt L2
	pea LC0
	jra L4
L2:
	pea LC1
L4:
	jbsr _printf
	unlk a6
	rts
--------------------------------------------------
The "tstl" instruction after the "addl" doesn't seem to be necessary,
because the addl will set the condition codes.

The above example arose as I was trying to answer a different
question, which is how m68k.md can get away with giving OUTPUT_JUMP a
third argument of 0 in patterns such as:

(define_insn "bgt"
  [(set (pc)
	(if_then_else (gt (cc0)
			  (const_int 0))
		      (label_ref (match_operand 0 "" ""))
		      (pc)))]
  ""
  "*
#ifdef MOTOROLA
  OUTPUT_JUMP (\"jbgt %l0\", \"fbgt %l0\", 0);
#else
  OUTPUT_JUMP (\"jgt %l0\", \"fjgt %l0\", 0);
#endif
")

Isn't the third argument sometimes going to be used?