Re: Can something similar to &x=&y be accomplished in C

3 msg	Inline assembly and local variable storage
2 msg	Help with -march and -mcpu issues
3 msg	GCC - Preserving statement expression results o...
10 msg	Solution sought to GCC 4.1.1 backtrace problem

Can something similar to &x=&y be accom...
\ Jim Stapleton (29 Aug 2007)
. \ Diego Novillo (29 Aug 2007)
. \ Tom St Denis (29 Aug 2007)
. . \ Jim Stapleton (29 Aug 2007)
. . . \ Tom St Denis (30 Aug 2007)

1 msg	GNAT/GCC 4.2 build problem on FreeBSD-CURRENT
3 msg	Local disable of specific warnings for a sectio...
3 msg	can gcc compile from memory buffer?
2 msg	Mac OS X cross compile for i386-elf
2 msg	Unsubscription
13 msg	SUSv3's 'memory location' and threads
2 msg	how to code global indirect branches ?
2 msg	Help about calculation of 'S' notation for a re...
1 msg	movaps missaligment in threads
3 msg	call stack's order of parameters
2 msg	-fno-stack-protector not ensuring binary compat...
8 msg	GCC asm block optimizations on x86_64
2 msg	.p2align
3 msg	compile errors from gcc
5 msg	Building gmp and mpfr within gcc

Subject:	Re: Can something similar to &x=&y be accomplished in C
Group:	Gcc-help
From:	Tom St Denis
Date:	30 Aug 2007

Jim Stapleton wrote:
> Thanks, I was more concerned with performance (an extra op or two each
> access), than with ease of use.
>

As the others pointed out &x references are actually implemented via
pointer deferences. C++ isn't some magical language, it's bound by the
same rules of the architecture as C. So to read/write that reference
you have to work through a pointer.

Note that GCC may alias *X to a register during it's computation. For
example,

int func(int *x, int *y)
{
*x = (*x * *y) + *y - (*y >> 2) + (*y << 4);
return *x;
}

produces this X86 code with GCC 4.1.2 (-O3 -fomit-frame-pointer)

func:
.LFB2:
movl (%rsi), %edx
movl %edx, %eax
movl %edx, %ecx
imull (%rdi), %eax
sarl $2, %ecx
leal (%rdx,%rax), %eax
sall $4, %edx
subl %ecx, %eax
addl %edx, %eax
movl %eax, (%rdi)
ret

As you can see x and y are pointed to by rsi and rdi. rsi get loaded
into edx, and reused. And now with using C++ ref's you get

_Z4funcRiS_:
.LFB2:
movl (%rsi), %edx
movl %edx, %eax
movl %edx, %ecx
imull (%rdi), %eax
sarl $2, %ecx
leal (%rdx,%rax), %eax
sall $4, %edx
subl %ecx, %eax
addl %edx, %eax
movl %eax, (%rdi)
ret

Notice some similarities?

Tom