extended asm and input parameters

[Andrew Haley]

On 07/05/2013 01:35 AM, dw wrote:
> 
>>> So, you are getting the duplicate zeroing of eax, but not
>>> the duplicate loading of rdi.
>>>
>>> I'm using 4.8.0.  You?
>> Ah, I'm on 4.7.  So this might be a regression: we should check.
> 
> Compiling with 4.7.2 and 4.7.3, I also get no duplicate loads on rdi.  
> But both 4.8.0 and 4.8.1 do.  All 4 duplicate the zeroing of eax.
> 
> What now?

First, I'd correct the asm:

   asm volatile (
      "rep stosb"
      : "+D" (Dest), "+c" (Count)
      : "a" (Data)
      : "memory");

then see if there was a real regression.

As far as I can see current GCC is (nearly) optimal:

main:
.LFB1:
	leaq	-32(%rsp), %rdi
	movl	$32, %ecx
	xorl	%eax, %eax
#APP
# 2 "z.c" 1
	rep stosb
# 0 "" 2
#NO_APP
	movl	$32, %ecx
	leaq	-32(%rsp), %rdi
#APP
# 2 "z.c" 1
	rep stosb
# 0 "" 2
#NO_APP
	xorl	%eax, %eax
	ret

Andrew.