jets: refactors and optimizes tree-math jets #397
Conversation
This reverts commit b256c87.
|
If urbit/urbit#6819 is ever restored, 2d696e3 should be reverted. |
joemfb
force-pushed
the
jb/tree-math-jets
branch
from
c13a351 to
de74b25
Compare
joemfb
force-pushed
the
jb/tree-math-jets
branch
from
61e7c34 to
562bfac
Compare
joemfb
changed the title
pkg/noun/jets/c/mas.c
Outdated
| u3i_slab sab_u; | ||
| u3i_slab_from(&sab_u, a, 0, --b_w); | ||
|
|
||
| sab_u.buf_w[(b_w >> 5)] &= ((c3_w)1 << (b_w & 31)) - 1; |
There was a problem hiding this comment.
NB: slab_from rounds up to word boundaries in the slice specified, so we need to mask off any more significant bits than we want.
pkg/noun/jets/c/mas.c
Outdated
| sab_u.buf_w[(b_w >> 5)] &= ((c3_w)1 << (b_w & 31)) - 1; | ||
| b_w--; | ||
| sab_u.buf_w[(b_w >> 5)] |= ((c3_w)1 << (b_w & 31)); |
There was a problem hiding this comment.
This mask and store logic is wrong in the boundary condition. @joemfb is going to put masking logic for sub-word blocksizes in u3i_slab_from()
There was a problem hiding this comment.
Thanks, this was a great catch
|
I've updated I've corrected the |
Building on urbit/urbit#6592, this PR adds jets for
+pinand+hub. Additionally, it fixes some unsafec3_w->u3_atomconversions and optimizes+pegand+mas.This is a draft PR as further testing of these changes is still needed.
/cc @eamsden