Re: Fix __ieee754_logl (-LDBL_MAX) in FE_DOWNWARD mode (bug 17022)

[Siddhesh Poyarekar <siddhesh at redhat dot com>]

On Wed, Jun 04, 2014 at 05:35:42PM +0000, Joseph S. Myers wrote:
> This patch fixes __ieee754_logl (-LDBL_MAX) on x86_64 and x86 not to
> subtract 1 from its argument and so cause spurious overflow in
> FE_DOWNWARD mode.  (For any argument strictly less than -1, it doesn't
> matter whether or not 1 is subtracted before computing log1p, as long
> as the result doesn't overflow to -Inf.)
> 
> Tested x86_64 and x86.  (This particular case lacks test coverage,
> since the testsuite doesn't cover -lieee, but it will be covered by
> tests after the following patch to test pow in all rounding modes,
> which was the context in which this bug was found.)
> 
> 2014-06-04  Joseph Myers  <joseph@codesourcery.com>
> 
> 	[BZ #17022]
> 	* sysdeps/i386/fpu/e_logl.S (__ieee754_logl): Do not subtract 1
> 	from arguments -2 or below.
> 	* sysdeps/i386/i686/fpu/e_logl.S (__ieee754_logl): Likewise.
> 	* sysdeps/x86_64/fpu/e_logl.S (__ieee754_logl): Likewise.

Looks good to me.

Siddhesh

> 
> diff --git a/sysdeps/i386/fpu/e_logl.S b/sysdeps/i386/fpu/e_logl.S
> index edae1d7..828e98a 100644
> --- a/sysdeps/i386/fpu/e_logl.S
> +++ b/sysdeps/i386/fpu/e_logl.S
> @@ -40,8 +40,11 @@ ENTRY(__ieee754_logl)
>  	fld	%st		// x : x : log(2)
>  	sahf
>  	jc	3f		// in case x is NaN or +-Inf
> +	movzwl	4+8(%esp), %eax
> +	cmpl	$0xc000, %eax
> +	jae	6f		// x <= -2, avoid overflow from -LDBL_MAX - 1.
>  4:	fsubl	MO(one)		// x-1 : x : log(2)
> -	fld	%st		// x-1 : x-1 : x : log(2)
> +6:	fld	%st		// x-1 : x-1 : x : log(2)
>  	fabs			// |x-1| : x-1 : x : log(2)
>  	fcompl	MO(limit)	// x-1 : x : log(2)
>  	fnstsw			// x-1 : x : log(2)
> diff --git a/sysdeps/i386/i686/fpu/e_logl.S b/sysdeps/i386/i686/fpu/e_logl.S
> index a0d1107..0ccc8fc 100644
> --- a/sysdeps/i386/i686/fpu/e_logl.S
> +++ b/sysdeps/i386/i686/fpu/e_logl.S
> @@ -39,8 +39,11 @@ ENTRY(__ieee754_logl)
>  	LOAD_PIC_REG (dx)
>  #endif
>  	fld	%st		// x : x : log(2)
> +	movzwl	4+8(%esp), %eax
> +	cmpl	$0xc000, %eax
> +	jae	5f		// x <= -2, avoid overflow from -LDBL_MAX - 1.
>  	fsubl	MO(one)		// x-1 : x : log(2)
> -	fld	%st		// x-1 : x-1 : x : log(2)
> +5:	fld	%st		// x-1 : x-1 : x : log(2)
>  	fabs			// |x-1| : x-1 : x : log(2)
>  	fld	MO(limit)	// 0.29 : |x-1| : x-1 : x : log(2)
>  	fcomip	%st(1)		// |x-1| : x-1 : x : log(2)
> diff --git a/sysdeps/x86_64/fpu/e_logl.S b/sysdeps/x86_64/fpu/e_logl.S
> index 315afc0..047b8db 100644
> --- a/sysdeps/x86_64/fpu/e_logl.S
> +++ b/sysdeps/x86_64/fpu/e_logl.S
> @@ -38,8 +38,11 @@ ENTRY(__ieee754_logl)
>  	fld	%st		// x : x : log(2)
>  	testb	$1, %ah
>  	jnz	3f		// in case x is NaN or +-Inf
> +	movzwl	8+8(%rsp), %eax
> +	cmpl	$0xc000, %eax
> +	jae	6f		// x <= -2, avoid overflow from -LDBL_MAX - 1.
>  4:	fsubl	MO(one)		// x-1 : x : log(2)
> -	fld	%st		// x-1 : x-1 : x : log(2)
> +6:	fld	%st		// x-1 : x-1 : x : log(2)
>  	fabs			// |x-1| : x-1 : x : log(2)
>  	fcompl	MO(limit)	// x-1 : x : log(2)
>  	fnstsw			// x-1 : x : log(2)
> 
> -- 
> Joseph S. Myers
> joseph@codesourcery.com

Attachment: pgp8wJvDxKYeh.pgp
Description: PGP signature