nick-knight
commented
1 year ago You can currently do this with a sequence of calls to the "vector insertion intrinsics", e.g., __riscv_vset_v_i8m1_i8m8. Although this is more verbose than the single intrinsic you request, it is also more flexible.
camel-cdr
Hi, would it be possible to add intrinsics akin to
vuint8m8_t __riscv_vreinterprete_8xu8m1(vuint8m1_t a, vuint8m1_t b, vuint8m1_t c, vuint8m1_t d, vuint8m1_t e, vuint8m1_t f, vuint8m1_t g, vuint8m1_t h).This above would force the compiler to allocate the LMUL=1 registers a to h in a single LMUL=8 register. E.g.
a=v0, b=v1, ..., h=v7.