Minutes from 2/21 meeting
Task Group: Vector Extension
Chair: Krste Asanovic
Number of Attendees: ~13
Current issues on github: https://github.com/riscv/riscv-v-spec
Note, the Zoom meeting details have changed. Please view the member
calendar entry for updated details.
Issues discussed: #341/#377, #367, #362/#354, #381
The following issues were discussed.
#341/#377 Separate vcsr from fcsr
Last time, the agreement was to add a new vcsr but to shadow fcsr
fields in the new vcsr to reduce context swap time. Due to concerns
raised in #377, the agreement was now to keep the separate vcsr but
to not shadow fcsr fields.
This change will be made to the spec for v0.9.
#367 Tail agnostic as option
Extensive disussion continued in the group about allowing this as an
option. The group proposing this change will describe the
implementation challenges of the current tail-undisturbed scheme in
the context of a renamed vector register machine with long temporal
#362/#354 Remove non-SEW vector load/store
Discussion covered the new proposal to provide a fractional LMUL to
avoid one drawback of dropping these memory operations. The general
consensus was in favor, as the scheme also aids mixed-width
floating-point arithmetic (for which there is no equivalent to the
widening/narrowing load/stores). The proposal requires adding a new
bit to LMUL in vtype, so that a new instruction will not be required.
The detailed fractional LMUL proposal was only recently presented, so
the group will continue the discussion online before next meeting.
#381 Index size in indexed load/store and AMO
The group discussed the issue of providing XLEN-sized indices in
indexed load/stores and AMOs, if non-SEW vector load/stores are
dropped. This was also previously a concern for floating-point codes,
as there were no fixed-precision floating-point loads/stores. The two
proposals were 1) to provide only XLEN-width indices or 2) provide
XLEN and 32b indices, and whether these are only present in extended
64b encoding. Discussion to continue.