what's the vd of vcompress in case of vl =0

[Kou66666]

Hi,

what's the vd of vcompress in case of vl =0? Is old_vd to follow "vstart >= vl", vd unchanged or follow 16.5, page 87, then vd is all tail, if tail agnostic write "1", then will output all "F"

Thanks, Kou

16.5, page 87: The vector mask register speciŠed by vs1 indicates which of the Šrst vl elements of vector register group vs2 should be extracted and packed into contiguous elements at the beginning of vector register vd. The remaining elements of vd are treated as tail elements according to the current tail policy (Section Vector Tail Agnostic and Vector Mask Agnostic vta and vma).

[aswaterman]

The general "vstart >= vl" rule applies: "no elements are updated in any destination vector register group".

[TogashiHan]

The general "vstart >= vl" rule applies: "no elements are updated in any destination vector register group". I'd like to double confirm this behavior because I open one similar issue at: https://github.com/riscv/riscv-v-spec/issues/944 ,which asked about mask instructions vd result while vl=0, but got different answer. Should vd be updated with all '1?(if ta update with '1)

In 15.1: "Mask elements past vl, the tail elements, are always updated with a tail-agnostic policy."

[aswaterman]

The earlier, more general clause takes precedence.

[TogashiHan]

The earlier, more general clause takes precedence.

Thank you for the clarification.

[furuame]

The earlier, more general clause takes precedence.

Thank you for the clarification.

@TogashiHan Oops, sorry for the confusion in #944! Yes in that case, vstart >=vl shouldn't have any side effects on destinations. I think 15.1 rule only applies when vl > vstart.