Re: [PATCH v2] libstdc++: ppc: conditionalize vsx-only simd intrinsics

[Segher Boessenkool <segher@kernel.crashing.org>]

On Mon, May 02, 2022 at 11:00:02PM -0300, Alexandre Oliva wrote:
> On May  2, 2022, Segher Boessenkool <segher@kernel.crashing.org> wrote:
> > Send full patches always please.
> 
> I'll try to remember that.  In case I fail, I hope you won't mind too
> much reminding me.
> 
> (You'd also asked me not to send patches as followups, but...  revised
> versions of a patch still belong in the same thread, right?)

No.  This makes it much harder to keep versions apart, even worse if you
use patchwork or similar.

The mail with the patch should be the head of a mail thread (or just
below the head if that is a cover mail), and everything below that is
discussion related to that patch.  The mail with the patch should be
ready to be committed as-is, so it should not have mangled whitespace
or encoding, should have proper commit message, etc.  This a) makes it
possible to review what will actualle be committed, and b) it makes it
possible for someone else to commit it (and that includes patch test
systems, often mistakenly called CI or CD, which are completely
different things).

> libstdc++'s bits/simd.h section for PPC (Altivec) defines various

(Spelling: PowerPC, AltiVec.  But this isn't about AltiVec really
anyway?)

> intrinsic vector types that are only available along with VSX: 64-bit
> long double, double, (un)signed long long, and 64-bit (un)signed long.

> +#if defined __VSX__ || __LONG_WIDTH__ == 32
>  _GLIBCXX_SIMD_PPC_INTRIN(signed long);
>  _GLIBCXX_SIMD_PPC_INTRIN(unsigned long);
> +#endif

Is __LONG_WIDTH__ the right macro to use here?  Nothing else in
libstdc++v3 uses it.  "__CHAR_BIT__ * __SIZEOF_LONG__" is the usual
thing to do.  Is __LONG_WIDTH__ always defined anyway?

> @@ -2450,10 +2456,11 @@ template <typename _Tp, size_t _Bytes>
>      static constexpr bool _S_is_ldouble = is_same_v<_Tp, long double>;
>      // allow _Tp == long double with -mlong-double-64
>      static_assert(!(_S_is_ldouble && sizeof(long double) > sizeof(double)),
> -		  "no __intrinsic_type support for long double on PPC");
> +		  "no __intrinsic_type support for 128-bit floating point on PPC");

You might do s/PPC/PowerPC/ at the same time.

Okay for trunk with __LONG_WIDTH__ dealt with.  Okay for the branches
a week or so after that.  Thanks!


Segher