Re: [PATCH] Fix PR target/28946

[Rask Ingemann Lambertsen <rask at sygehus dot dk>]

On Tue, Sep 05, 2006 at 01:06:58PM +0200, Uros Bizjak wrote:
> Hello!
> 
> This patch fixes PR target/28946, a regression from gcc-2.95 (!)
> 
> As desdribed in the PR, for following testcase:
> 
> int fct1 (void);
> int fct2 (void);
> 
> int
> fct (unsigned nb)
> {
>  if ((nb >> 5) != 0)
>    return fct1 ();
>  else
>    return fct2 ();
> }
> 
> i386 backend emits spurious test instructions:
> 
> fct:
>        movl 4(%esp), %eax
>        shrl $5, %eax
>        testl  %eax, %eax
>        je  .L2
>        jmp fct1
>        .p2align 4,,7
> .L2:
>        jmp fct2
> 
[snip proposed patch to combine]
> Resulting asm code is then:
> 
> fct:
>        movl 4(%esp), %eax
>        shrl $5, %eax
>        je  .L2
>        jmp fct1
>        .p2align 4,,7
> .L2:
>        jmp fct2

   This is likely because you are missing a pattern in i386.md. You should
have something like this:

(define_insn "*lshrsi3_cconly"
  [(set (reg:CC FLAGS_REG)
      (compare:CC (lshiftrt:SI (match_operand:SI 1 "nonimmediate_operand" "0")
                               (match_operand:QI 2 "nonmemory_operand" "Ci"))
                  (const_int 0)))
   (clobber (match_scratch:SI 0 "=r"))]
  ""
  "shrl %2, %0"
)

I have such a pattern in my experimental i8086 backend, and I get (for an an
i80186) with -dp:

fct:
	pushw	%bx		;# 65	*pushhi1
	movw	%sp,	%bx	;# 63	*movhi/1
	movw	4(%bx),	%ax	;# 64	*movhi/1
	shrw	$5,	%ax	;# 18	*lshrhi3_cconly_ccz/2
	je	.L2		;# 19	*beq_ccz
	call	fct1		;# 23	call_value
	popw	%bx		;# 68	*pophi1
	ret			;# 69	*return
.L2:
	call	fct2		;# 30	call_value
	popw	%bx		;# 75	*pophi1
	ret			;# 76	*return

I.e. no spurious test instruction. The same applies to all arithmetic, logic
and shift/rotate instructions.

See also <URL:http://gcc.gnu.org/ml/gcc/2004-02/msg00903.html>.

-- 
Rask Ingemann Lambertsen