inline assembly

inline assembly

[Kevin Diggs]

Hi,

	When doing inline assembly, is there a way to get the compiler to 
assign "extra" (one not specified for inputs and outputs) registers? In 
the following:

                 __asm__ __volatile__ (
                         "addi 5,%1,-1\n"
                         "andc 5,%1,5\n"
                         "cntlzw 5,5\n"
                         "subfic %0,5,31":
                         "=r"(j):
                         "r"(i)
                 );

Can I get the compiler to choose a register for r5?

kevin

Re: inline assembly

[Scott Wood]

Kevin Diggs wrote:
> Hi,
> 
>     When doing inline assembly, is there a way to get the compiler to 
> assign "extra" (one not specified for inputs and outputs) registers? In 
> the following:
> 
>                 __asm__ __volatile__ (
>                         "addi 5,%1,-1\n"
>                         "andc 5,%1,5\n"
>                         "cntlzw 5,5\n"
>                         "subfic %0,5,31":
>                         "=r"(j):
>                         "r"(i)
>                 );
> 
> Can I get the compiler to choose a register for r5?

Yes, like this:

int tmp;

asm volatile("addi %1, %2, -1;"
              "andc %1, %2, %1;"
              "cntlzw %1, %1;"
              "subfic %0, %1, 31" : "=r" (j), "=&r" (tmp) : "r" (i));

However, it'd be better to let the compiler do more, by just using the 
existing cntlzw() function.

-Scott

Re: inline assembly

[David Howells]

Scott Wood <scottwood@freescale.com> wrote:

> int tmp;
> 
> asm volatile("addi %1, %2, -1;"
>              "andc %1, %2, %1;"
>              "cntlzw %1, %1;"
>              "subfic %0, %1, 31" : "=r" (j), "=&r" (tmp) : "r" (i));

Registers are usually assumed to be 'long' in size, so I'd recommend using
that rather than 'int' for tmp, though I suspect it'll make little difference
(except, perhaps on x86 where you can partially use registers).

> However, it'd be better to let the compiler do more, by just using the
> existing cntlzw() function.

Look in include/asm-powerpc/bitops.h.  There are examples of the things you're
trying to do:

	static __inline__ __attribute__((const))
	int __ilog2(unsigned long x)
	{
		int lz;

		asm (PPC_CNTLZL "%0,%1" : "=r" (lz) : "r" (x));
		return BITS_PER_LONG - 1 - lz;
	}

	static __inline__ int __ffs(unsigned long x)
	{
		return __ilog2(x & -x);
	}

Where:

	asm-compat.h:79:#define PPC_CNTLZL	stringify_in_c(cntlzd)
	asm-compat.h:100:#define PPC_CNTLZL	stringify_in_c(cntlzw)

Depending on whether you're in 32-bit mode or 64-bit mode.

David

Re: inline assembly

[Scott Wood]

On Thu, Jun 05, 2008 at 11:44:51AM +0100, David Howells wrote:
> Scott Wood <scottwood@freescale.com> wrote:
> 
> > int tmp;
> > 
> > asm volatile("addi %1, %2, -1;"
> >              "andc %1, %2, %1;"
> >              "cntlzw %1, %1;"
> >              "subfic %0, %1, 31" : "=r" (j), "=&r" (tmp) : "r" (i));
> 
> Registers are usually assumed to be 'long' in size, so I'd recommend using
> that rather than 'int' for tmp, though I suspect it'll make little difference
> (except, perhaps on x86 where you can partially use registers).

I had originally written that as long, but I changed it since the asm
code is explicitly assuming 32-bit.  It's more documentation than
anything else.

-Scott