feat: remove memory access adapters: CPU Boundary AIR, initialize/finalize memory #2318
Conversation
Commit: 60d1d19 |
Maillew
force-pushed
the
feat/remove-mem-access2
branch
from
d0e7bae to
9d0bea8
Compare
Maillew
changed the title
Maillew
changed the base branch from
develop-v1.6.0
to
feat/access-adapter-removal
Maillew
changed the title
Maillew
force-pushed
the
feat/remove-mem-access2
branch
2 times, most recently
from
f937ae0 to
3706b09
Compare
Maillew
force-pushed
the
feat/access-adapter-removal
branch
from
df72623 to
e5cca45
Compare
This reverts commit 9c30237.
Maillew
force-pushed
the
feat/remove-mem-access2
branch
from
2051624 to
a5814d4
Compare
| /// Whether access adapters are enabled. When disabled, all memory accesses must be of the | ||
| /// standard block size (ie, 4 for address spaces 1-3). | ||
| #[new(value = "true")] | ||
| pub access_adapters_enabled: bool, |
There was a problem hiding this comment.
i'm wondering if instead of storing a separate variable, we just have an access_adapters_enabled function that returns true if native address space is used and false otherwise. just so that there's a single source of truth
There was a problem hiding this comment.
there is currently a function to check if native address space is used, is_native_as_used. i kept it as a field, to force it to be true/false for testing purposes. there is an additional function, with_auto_access_adapters which automatically sets the field, depending on is_native_as_used
| use std::collections::BTreeMap; | ||
| let mut chunk_map: BTreeMap<(u32, u32), ([u32; BLOCKS_PER_CHUNK], [F; CHUNK])> = | ||
| BTreeMap::new(); |
There was a problem hiding this comment.
can we do this by just a par_iter on a vector or do we need a btreemap here? btreemaps are slow
There was a problem hiding this comment.
Resolved: 5bd398c
note, have to do it in two passes, because of race conditions if accessing the same chunk
Closes INT-5723, INT-5724, INT-5726