diff options
| author | Darrick Wiebe <darrickw@gmail.com> | 2025-04-12 13:40:20 -0400 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-04-12 10:40:20 -0700 |
| commit | 798bd959410eb6c2e0b9f7ae1222181ac2b295d4 (patch) | |
| tree | 94400ba51862663d02a2eac3a58b585ef749108a /chapter22/src | |
| parent | 47711b81ea915eeeb46b2ad8ebb291fb21cbaf6e (diff) | |
| download | Simple-798bd959410eb6c2e0b9f7ae1222181ac2b295d4.tar.gz Simple-798bd959410eb6c2e0b9f7ae1222181ac2b295d4.zip | |
MemStore interface (#201)
* Hook for pre-encoding work needed by M2
* Add MemStore interface
* Add both MemLoad and MemStore interfaces
* Fix NPE printing StopNode
* Attach merged mem to return when forcing a loop exit
* Made requested changes. This breaks a few tests, possibly because...
...I am not sure whether some X86 ops are loads or stores.
* Mach ops need anti-deps too
MemOp subclasses are either "Load-like" or "Store-like" based on a flag, and not on subclassing LoadNode nor StoreNode.
Mach ops pick up anti-deps, no reg mask for those.
x86 bug: do not allow "dst = [mem] + #imm", no such instruction
Empty merge-mem folds away.
x86 add-to-mem works both ways
---------
Co-authored-by: Cliff Click <cliffc@acm.org>
Diffstat (limited to 'chapter22/src')
0 files changed, 0 insertions, 0 deletions