[FIX] vdiv stall

[rgiunti]

Issue

The VFU reads and writes the full VRF word regardless of vl. Two issues:

• Problem when number of bytes in src vectors < VRFWordBWidth → calculations are performed with Xs → stall for integer division FSM

• For all arithmetic instructions when number of bytes in src vectors < VRFWordBWidth —> Xs are written in the VRF along with computed bytes

Proposed Fix

• On read-side: mask the read operands in order not to propagate Xs in the datapath --> this avoids the integer divider FSM to stall

• On write-side: computation of write byte enable signal has been changed in order to write in the VRF just the bytes that come from the arithmetic instruction execution

Notes

Read and write side valid byte counts are decoupled because in narrowing instructions the source operands are twice the destination width (src SEW = dst SEW + 1)