Make the tuple type API for segment load store as compiler optional feature

[zakk0610]

#137 (comment)
#137 (comment)

Below is @kito-cheng's proposal for the segment load/store API:

Include both API flavor in spec, but mark tuple type style API as optional, and defined some marco to let user check __riscv_rvv_tuple_type if compiler has supported.

And non-tuple type style API / new API are mandatory, I believe implement that via tuple type style API like that is feasible, and @rjiejie has implemented this approach in the past riscv-collab/riscv-gcc#307:

void vlseg2e8_v_i8mf4 (vint8mf4_t *v0, vint8mf4_t *v1, const int8_t *base, size_t vl)
{
vint8mf4x2_t vt = vcreate_i8mf4x2 (v0, v1);
vlseg2e8_v_i8mf4x2 (vt, vase, v0);
}

[nick-knight]

We will probably need to revert #98 too, tagging these types/intrinsics appropriately.

[lhtin]

Maybe it's necessary to change the marco name __riscv_rvv_tuple_type to __riscv_v_tuple_type, because it is unified with __riscv_v_intrinsic_overloading.

[kito-cheng]

@lhtin Yeah, __riscv_v_tuple_type seems like better naming, thank for your suggestion !

[eopXD]

Multiple issues have raised request for tuple types and size-less struct and it all boils down if the compiler is able to implement them...

Redirecting other existing issues(#17 , #83, #110 , #154) here for this issue to be the placeholder to discuss on the problem after the v1.0 release.