On 9/11/19 2:25 AM, liuzhiwei wrote:
> + for (i = 0; i < vlmax; i++) {
> + if (i < env->vfp.vstart) {
> + continue;
> + } else if (i < vl) {
> + tmp = ~vector_mask_reg(env, rs1, width, lmul, i) &
> + vector_mask_reg(env, rs2, width, lmul, i);
> + vector_mask_result(env, rd, width, lmul, i, tmp);
> + } else {
> + vector_mask_result(env, rd, width, lmul, i, 0);
> + }
> + }
These can be processed in uint64_t units, with a mask based on width:
8: 0xffffffffffffffff
16: 0x5555555555555555
32: 0x1111111111111111
64: 0x0101010101010101
dest = ~in1 & in2 & mask;
with an additional final mask to handle vl not being a multiple of 64.
Again, I urge you not to bother with impossible vstart -- instructions like
this cannot be interrupted, and the spec allows you to not handle values of
vstart that cannot be produced by the implementation.
r~
