Question about segment load/store of rvv-v0p10-compatible-headers

[xl0714]

Hi, @eopXD :
The current form is:

#define vlseg2e8_v_i8m1(...) __riscv_vlseg2e8_v_i8m1(VA_ARGS)
#define vlseg2e8_v_i8m1_m(...) __riscv_vlseg2e8_v_i8m1_tumu(VA_ARGS)
#define vlseg2e8_v_i8m1_tu(...) __riscv_vlseg2e8_v_i8m1_tu(VA_ARGS)
#define vlseg2e8_v_i8m1_ta(...) __riscv_vlseg2e8_v_i8m1(VA_ARGS)
#define vlseg2e8_v_i8m1_tuma(...) __riscv_vlseg2e8_v_i8m1_tum(VA_ARGS)
#define vlseg2e8_v_i8m1_tumu(...) __riscv_vlseg2e8_v_i8m1_tumu(VA_ARGS)
#define vlseg2e8_v_i8m1_tama(...) __riscv_vlseg2e8_v_i8m1_m(VA_ARGS)
#define vlseg2e8_v_i8m1_tamu(...) __riscv_vlseg2e8_v_i8m1_mu(VA_ARGS)

However, gcc no longer supports non-tuple segment load/store. Should we consider mapping non-tuple to tuple?

[eopXD]

The v0p10 compatible header was intended to support transition from v0.10 to v0.11. The tuple type segment load/store was introduced in v0.12. We currently do not have a compatible header / patch in the upstream to support compatibility for v0.11 -> v0.12.

This is a duplicate issue to #255. As I mentioned in that thread, I think continue-ing the compatibility support in the upstream trunk may not be beneficial for the ecosystem.

[xl0714]

I see, thank you.