Help - are we running low on registers in the following code snippet?
Is it generally better to choose MOVL over XCHG?
If XCHG is faster, how would one rearrange the C code to get XCHG in
the ASM code?
Many thanks in advance.



// --------------------- Incomplete code snippet
---------------------
// Logiciels/Informatique lcc-win32 version 3.8. (Dec 18 2007
19:12:54)
// Compile options  -v -S -p6 -O

125          IntSwapBuf = *IntPtrL;
126          *IntPtrL   = *IntPtrR;
127          *IntPtrR   = IntSwapBuf

	.line	125
	movl	-8(%ebp),%eax
	movl	(,%eax),%ecx
	movl	%ecx,-20(%ebp)
	.line	126
	movl	-16(%ebp),%edx
	movl	(,%edx),%edx
	movl	%edx,(,%eax)
	.line	127
	movl	-16(%ebp),%eax
	movl	%ecx,(,%eax)

MatlabMonkey@googlemail.com wrote:
> Help - are we running low on registers in the following code snippet?
> Is it generally better to choose MOVL over XCHG?
> If XCHG is faster, how would one rearrange the C code to get XCHG in
> the ASM code?
> Many thanks in advance.
> 
> 
> 
> // --------------------- Incomplete code snippet
> ---------------------
> // Logiciels/Informatique lcc-win32 version 3.8. (Dec 18 2007
> 19:12:54)
> // Compile options  -v -S -p6 -O
> 
> 125          IntSwapBuf = *IntPtrL;
> 126          *IntPtrL   = *IntPtrR;
> 127          *IntPtrR   = IntSwapBuf
> 
> 	.line	125
> 	movl	-8(%ebp),%eax
> 	movl	(,%eax),%ecx
> 	movl	%ecx,-20(%ebp)

You tell it to assign *IntPtrL to IntSwapBuf.
It does EXACTLY that.

> 	.line	126
> 	movl	-16(%ebp),%edx
> 	movl	(,%edx),%edx
> 	movl	%edx,(,%eax)

You tell it to assign *IntPtrR to *IntPtrL. It does
exactly that.

> 	.line	127
> 	movl	-16(%ebp),%eax
> 	movl	%ecx,(,%eax)

And then, you tell it to assign *IntSwapBuf to *IntPtrR.
It does that.

Now here is what MSVC with the HIGHEST level of optimization
generates (in 64 bit assembly)
; File d:\temp\tswap.c
; Line 6
         mov     rdx, QWORD PTR IntPtrL
         mov     eax, DWORD PTR [rdx]
         mov     DWORD PTR IntSwapBuf, eax
; Line 7
         mov     rax, QWORD PTR IntPtrR
         mov     ecx, DWORD PTR [rax]
         mov     DWORD PTR [rdx], ecx
; Line 8
         mov     rax, QWORD PTR IntPtrR
         mov     ecx, DWORD PTR IntSwapBuf
         mov     DWORD PTR [rax], ecx

You see?

The same code. Compilers aren't psychic to the point of realizing
that you want to swap the values


-- 
jacob navia
jacob at jacob point remcomp point fr
logiciels/informatique
http://www.cs.virginia.edu/~lcc-win32