Hi Jeffrey,
On Wed, Sep 20, 2017 at 04:49:43AM -0400, Jeffrey Walton wrote:
> test.cpp:913:42: warning: vec_lvsl is deprecated for little endian;
> use assignment for unaligned loads and stores [-Wdeprecated]
> const uint8x16_p8 perm = vec_lvsl(0, src);
>
> The warning is coming from this type of code, which was taken from the
> AltiVec PIM:
>
> // Loads a byte array, does not perform an endian conversion.
> // This function requires the subkey table is correct endianess.
> inline VectorType VectorLoadKey(const byte src[16])
> {
> const uint8x16_p8 perm = vec_lvsl(0, src);
> const uint8x16_p8 low = vec_ld(0, src);
> const uint8x16_p8 high = vec_ld(15, src);
> return (VectorType)vec_perm(low, high, perm);
> }
>
> Its not clear to me how to use assignment to load a vector or store a
> vector. I'm also not sure if it includes the automatic endian
> conversion, which is something we don't want because the byte arrays
> are big endian. (Otherwise we would use a newer function like
> vec_vsx_ld).
>
> How do does one load a vector with assignment and not invoke the
> automatic endian conversions?
Maybe you want vec_xl_be?
Segher
