Re: [PATCH] glibc: support e5500 and e6500

[Joseph Myers <joseph at codesourcery dot com>]

On Wed, 20 Feb 2019, C.r. Guo wrote:

> From: Chunrong Guo <chunrong.guo@nxp.com>
> 
> Signed-off-by: Chunrong Guo <chunrong.guo@nxp.com>

I don't see a copyright assignment on file from NXP (though there was an 
older one from Freescale).

> diff --git a/sysdeps/powerpc/powerpc64/be/e5500/Implies b/sysdeps/powerpc/powerpc64/be/e5500/Implies
> new file mode 100644
> index 0000000..a795586
> --- /dev/null
> +++ b/sysdeps/powerpc/powerpc64/be/e5500/Implies
> @@ -0,0 +1,2 @@
> +powerpc/powerpc64/e5500
> +

No blank lines at the end of files, the commit hooks don't allow them to 
be pushed (or whitespace at end of line, or tabs-after-spaces 
indentation).

> +static const ieee_float_shape_type a_nan = {.word = 0x7fc00000 };
> +static const ieee_float_shape_type a_inf = {.word = 0x7f800000 };

Use INFINITY and NAN constants if needed, rather than hardcoding 
representations like this.

> +static const float two108 = 3.245185536584267269e+32;
> +static const float twom54 = 5.551115123125782702e-17;

Use hex floats (and there's no real need to have such variables, just 
write 0x1p108 or 0x1p-54 or corresponding float constants as needed 
directly in the code using such values).

> +#define FMADD(a_, c_, b_)                                               \
> +          ({ double __r;                                                \
> +          __asm__ ("fmadd %[r], %[a], %[c], %[b]\n"                     \
> +                   : [r] "=f" (__r) : [a] "f" (a_), [c] "f" (c_), [b] "f" (b_)); \
> +          __r;})
> +#define FNMSUB(a_, c_, b_)                                          \
> +          ({ double __r;                                                \
> +          __asm__ ("fnmsub %[r], %[a], %[c], %[b]\n"                     \
> +                   : [r] "=f" (__r) : [a] "f" (a_), [c] "f" (c_), [b] "f" (b_)); \
> +          __r;})

Use __builtin_fma / __builtin_fmaf, not asm.

-- 
Joseph S. Myers
joseph@codesourcery.com