feat: memory access adapters, boundary chips, merkle chip E3 #1640
Conversation
…o fix a bunch of bugs
| self.control | ||
| .execute_instruction(state, &mut self.chip_complex)?; | ||
| .execute_instruction(state, &instruction.clone(), &mut self.chip_complex)?; |
There was a problem hiding this comment.
how come it's both reference and clone?
There was a problem hiding this comment.
It's because
- design-wise I prefer passing a reference
- currently it's hard to do without tricks, because this reference is inside
self.something, andself.execute_instructionrequires a mutable reference toself. That is why I did this hack and added a// TODOabove
There was a problem hiding this comment.
I meant does &instruction not work?
There was a problem hiding this comment.
No
Because it is an immutable reference to something inside &self
And execute_instruction(&mut self, ...)
| ) { | ||
| let ptr = pointer / ALIGN; | ||
| let meta = unsafe { self.meta.get_unchecked_mut(address_space) }; | ||
| meta.set( |
There was a problem hiding this comment.
is this faster or should we still do a pre-check and return early when block_size == ALIGN?
There was a problem hiding this comment.
What would we do in this case? We still want to set the meta block. Or do you think that iterating over 1..1 is slower than branch prediction in practice?
There was a problem hiding this comment.
oh hm I somehow thought we didn't need to set the meta block in this case, but I guess you need to update the timestamp regardless?
Resolves INT-3801.
Vechow it's done now,get_ftoo often (although I don't know how to avoid it normally).VmChipTestBuildernow has::default_persistent, so all tests inextensions/rv32im/circuitpass both with volatile and persistent memory interface.