Re: [v2 PATCH 1/2] PPC64: Add libmvec SIMD single-precision power function.

[Shawn Landden <shawn at git dot icu>]

08.07.2019, 09:22, "Tulio Magno Quites Machado Filho" <tuliom@ascii.art.br>:
> Shawn Landden <shawn@git.icu> writes:
>
>>  +/* The output of log2 and thus the input of exp2 is either scaled by N
>>  + (in case of fast toint intrinsics) or not. The unscaled xd must be
>>  + in [-1021,1023], sign_bias sets the sign of the result. */
>>  +static inline
>>  +vector float
>>  +exp2_inline (vector double xdl, vector double xdr, vector unsigned sign_bias)
>>  +{
>>  + v64u kil, kir, skil, skir, sign_biasl, sign_biasr;
>>  + vector double kdl, kdr, zl, zr, rl, rr, r2l, r2r, yl, yr, sl, sr;
>>  +
>>  + vector unsigned zero = {0, 0, 0, 0};
>>  +#ifdef __LITTLE_ENDIAN__
>>  + sign_biasl = (v64u)vec_mergeh (sign_bias, zero);
>>  + sign_biasr = (v64u)vec_mergel (sign_bias, zero);
>>  +#elif __BIG_ENDIAN__
>>  + sign_biasl = (v64u)vec_mergel (sign_bias, zero);
>>  + sign_biasr = (v64u)vec_mergeh (sign_bias, zero);
>
> In big endian that's the wrong word, i.e.:
>   sign_biasl = (v64u)vec_mergel (zero, sign_bias);
>   sign_biasr = (v64u)vec_mergeh (zero, sign_bias);
>
> Fixed.
I recently got access to a Big-Endian VPS restored, so hopefully this will not happen again in the future.
>
> I'm merging both patches now.
>
> Thanks!
>
> --
> Tulio Magno

-- 
Shawn Landden