525b103f38swap ret_stack to a repr(C) struct for manipulation from JITNathan Braswell2024-02-05 01:41:55 -05:00
fd843bc807Swap ID to NonZeroI64 so Option<ID> is a single word and we can test for 0 in JIT for if we have an ID (for looking at ret_stack)Nathan Braswell2024-02-05 01:34:53 -05:00
a4500fed36Implement (but have not tested) all primitive functions in JIT)Nathan Braswell2024-02-05 01:26:57 -05:00
776fc7c921integrated JIT compiler with simple generated do-nothing JIT traces. Modified JIT trace interface so that it can return a trace-id and offset to jump to, allowing the trace interpreter to handle hard cases, etc. (or in this case, all cases, as it just immediatly returns the current id and offset 0).Nathan Braswell2024-02-02 01:46:13 -05:00
9cd46a31ebhad trouble getting going again today. Did create a compiled_traces map and integrated using it into execute_trace, with ideas for what the api there should be.Nathan Braswell2024-02-01 02:09:04 -05:00
0ce6d90aa9minor cleanup to wrap JIT stuff in a struct with a compile function and move to lib, in anticipation of actually JIT compiling. Not a ton of work today, was pretty tired / didn't get going.Nathan Braswell2024-01-31 01:21:57 -05:00
8e06037b4aFinished porting everything over to new packed, JIT-friendly Form, with all the tracing!Nathan Braswell2024-01-29 02:22:56 -05:00
069c9775e1Ported interpreter (without trace execution) to use new packed, JIT-friendly Form, with all needed additionsNathan Braswell2024-01-29 02:03:59 -05:00
0696ad5594Finished-ish writing the new JIT-friendly, packed Form, this time with extra repr(C) Vec implementationNathan Braswell2024-01-28 21:51:01 -05:00
a16e126aa1Started writing our new Form designed to be used from JIT with repr(C) and packed into a single word, and our own repr(C) Rc implementation to go with it.Nathan Braswell2024-01-28 02:46:08 -05:00
3002bd9237Working ExtendedBasicBlock tracing/lazy-construction! Not all constructs traceable, and no inlining, but it works!Nathan Braswell2024-01-07 00:15:44 -05:00
1e59c8ed94Small one today, getting started was hard but glad I got a tiny bit done to continue the streak. Modifed follow-on data to be stored on Ctx itself and only pass an ID around so it can be on the return stack multiple times without wasting a lot of space.Nathan Braswell2024-01-04 01:15:42 -05:00
c15e857171First barely-working trace replay! Need to clean up, finish implementing, have trace-reply restore traces, jump directly to other traces, eventually inline and concat traces, etc etcNathan Braswell2024-01-03 01:55:40 -05:00
71a2272f34mostly correct looking traces. given up on actual constant prop during tracing, only constant tracking. Will need later pass to optimize. Also, want to shift the post-call trace to after the innermost return instead of tracking with the return stack.Nathan Braswell2024-01-02 01:35:55 -05:00
88e3fa9d39Move dynamic stuff out of Cont and into two stacks so that Cont is good for resuming traces (having them depend only on code). Remove currently unneeded IDs from Symbol and Pair. Added an InlinePrim Op to tracing, and a nc: Cont to Call. IDs are probs only for trace points now (function starts and side-exits).Nathan Braswell2023-12-26 01:03:56 -05:00
82c38a32cfBasic tracing structure sketched with stubs in cases - need to add constant tracking and implement stubs. Still need some thought on tracing a closure vs tracing the code that created the closureNathan Braswell2023-12-01 21:36:37 -05:00
2787b95837Rewrite into non-recursive defunctionalized-continuations styleNathan Braswell2023-11-25 11:35:36 -05:00
6edbe7d2c8Continue moving stuff around to simplify before opt, and then start sketching out wrapped values and trace - just realized we might need all values to be wrapped tohande side effects that don't depend on input but still have effects (mutation, now ugh and assert)Nathan Braswell2023-07-09 00:24:50 -04:00
b077daf12eSplit eval.rs out of ast.rs, and rename ast.rs to basic.rs. Copy it over to opt.rs, and hook up a From<> impl and a congruent(x)->bool function, and add to main.rs and test.rs. Now we actually have to make opt.rs optimize...Nathan Braswell2023-07-06 01:11:56 -04:00
f4ae4db92dMove call inside form to allow for tracing (might still need a way to get out of the end of tracing)Nathan Braswell2023-07-04 12:23:31 -04:00
ce5ab1dd99Implemented shift/reset delimited continuations using a defunctionalized double-CPS transformation! Basic tests work at least, there certinally could be bugsNathan Braswell2023-05-22 01:23:17 -04:00
4126e25463everything uses continuations now, skeleton for shift/resetNathan Braswell2023-05-21 04:14:29 -04:00
093f3c0453Basic continuation-ification of primitives, finally work out the Cursor/Cont devide. Still need to figure out which Cont's we need for evaluation (possibly a subset of) parametersNathan Braswell2023-05-16 23:31:58 -04:00
ea1516fbd1Started new kv impl with support for mutable cells, just the basic interpreter. Going to add the lazy bytecode interperter nextNathan Braswell2023-05-14 23:50:27 -04:00
b2216f7574Implemented many of Chris and Sharjeel's suggestions - still need to insert a worked example, and if I have time, benchmarks re-run with numbers and memory usageNathan Braswell2023-04-25 21:58:25 -04:00
87763fad40Tons of slide work, unfortunately depend on external MathJax - need to pull that down laterNathan Braswell2023-04-18 01:47:10 -04:00
5054441625More work including brain dumping more slides and bugfixes for memo - have found current problem, which is that a non-env se can prevent a DeriComb from pulling in valeus from a real Env frame which it's being returned from - even though this is caught, the resulting call still depends on it, marks it as such, and then it depends on a call that was real and now no longer exists.Nathan Braswell2023-04-16 00:02:29 -04:00
3dcebe5297Bunch of new slides with examples comparing Fexprs to Lisp's macros and special formsNathan Braswell2023-04-15 15:41:11 -04:00
f55dd236b6Fix one bug, or part of one - dericomb wasn't advancing if se wasn't a legal env, but was still unioning the body IDs into it's ids which could convince it that it should be able to progressNathan Braswell2023-04-15 00:52:52 -04:00
f4b99067d4More working memo that tracks used env-ids and env, fixed bugs with it too. Currently trying to track down some infinate loops caused by something that thinks it should make progress and doesn't, have an assert that notices doublesNathan Braswell2023-04-15 00:19:51 -04:00
4e7825cb2bHack in copying code between the two editors and fix recursive.css's body style overridding the slidesNathan Braswell2023-04-08 14:05:48 -04:00
6ef3392aa0prototype remarkjs slides with Kraken formatting and auto-execute code in browser on slide change! Some rough edges, including two different editors between the first slide and the reveal slideNathan Braswell2023-04-08 13:38:21 -04:00
b89b831983Website with changes made with/based on my graphic designer Mom's suggestions (thanks Mom!)Nathan Braswell2023-04-05 21:10:04 -04:00
a06c5854c8The big exponential thing was ironically Hash which was being called for memo which existed to prevent exponential behavior. Reimplmenented the caching hash, and much much faster. Still some failing tests, and bvau might be loopingNathan Braswell2023-04-04 23:40:09 -04:00
8ccdfd8ab2new invariant: when returned, each form is as evaluated as it can be. This means that union_into_tail is more rare, and only for recursion (if or deri) stopped items. If now also captures its environment if it's rec_stopped. Added memoization for legal env forms to try to deal with it, has helped some, still exponential behavior.Nathan Braswell2023-04-02 12:48:02 -04:00
3c5bb127e2Work on saving env on calls and suspended symbols, only use as env if legal. Seemingly causing exponential behavior, need to have it done incrementally... todoNathan Braswell2023-03-29 01:58:07 -04:00
8fa7acbdf1Actually using the evaled param lookups with cons pairs etc - Tests failing - need to figure out how to really handle the *weird* case, and also the failing tests if the *wierd* case isn't the cause. Not sure if we need to track to prevent the dropping of the *not* redundant veval, or if the suspended symbol should hold onto the half-done env, or whatNathan Braswell2023-03-27 01:27:25 -04:00
664b336a48trac suspended param eval, with proper back conversionNathan Braswell2023-03-27 00:28:44 -04:00
70ac3e3633Add theming and a 'slick' theme based on Inter and JetBrains MonoNathan Braswell2023-03-25 16:59:24 -04:00
92bcb6f358All tests passing when run with --release. Not sure if this is really correct, or if the algorithm is being too wasteful and re-executing versions of functions taht should have been inlined or something (this may esp be true without the parameter/wrap_level stuff, and also I'm a little sus of the env chains - oh hay, it might even just be becasue only legal env chains are allowed which prevents pe-ing)
partially_broken
Nathan Braswell2023-03-20 01:06:24 -04:00
48de9517c9An unvaled Pair should have a True override for its IDsNathan Braswell2023-03-19 11:15:22 -04:00
9162b456d2Oh some were just changes in formatting. First real issue looks like it's VMAPNathan Braswell2023-03-19 02:03:41 -04:00
1df3acccd7Some tests passing again, namely all of basic and factNathan Braswell2023-03-19 01:52:18 -04:00
7b334e96dfFinally finished sketching out the new version, lets get on the compile errorsNathan Braswell2023-03-19 00:59:46 -04:00
d4920ec4b6Notes for hopefully the full simple kraken PE, with initial implementation of new NeededIds (split between head and tails (heads are ids for which progress will be made, tails might not progress but are contained), body_under and if_under).Nathan Braswell2023-03-15 01:16:37 -04:00
865fc1b4b6Update flake, split ast out into ast & pe_ast, and main into main & testNathan Braswell2023-03-14 20:14:17 -04:00
4752dde4afAdded recursive If tracking, but that still doesn't seem to have solved the problem. Have to dig deeper into the actual failed tests now...Nathan Braswell2023-02-24 01:19:55 -05:00
abddf1db06transition to fake-ids instead of rec-hashes. Still have infinate recursion, need to check on itNathan Braswell2023-02-23 13:14:10 -05:00
6357297cd1been going down withat's probs the wrong path for solving infinate recursion for bvau, did a lot on suspended IF but that doesn't seem to be the issue... Not sure if this was done in vainNathan Braswell2023-02-23 02:48:40 -05:00
bb6b15c156some bugfixes, and moving towards a reconstruct-on-error. More bugs remain though, ?uncovered? by this, including another infinate loop. Fixed a tricky one where a function would recursively call it self for real while evaluating itself fake, and create an environment that had a loop indirected through subbing itself in for SuspendedEnv (since the real and the fake had the same ID).Nathan Braswell2023-02-23 01:38:46 -05:00
e327ed7314Found the bug. All tests pass PE, except for those that overflow the stack. Those are disabled, and the remaning must be run with --release. Can maybe improve PE tail call to prevent thatNathan Braswell2023-02-22 20:16:08 -05:00
bde04e94dfStill buggy, but implemented a more tail-recursive PE with non-recursive drop_redundent_veval, special forms SuspendedEval and SuspendedIf, simplification of how primitives work, and moreNathan Braswell2023-02-22 01:22:57 -05:00
45a869b09bRemove year and weird (c) from copyright line, concensus is it's not needed these daysNathan Braswell2023-02-21 10:53:07 -05:00
21fb47470aToo much debugging to find out PE assert was broken and debugging printlns were overflowing the stack. Some of the other tests also overflow the stack if not built in release mode, and I disabled the stack-testing tests for PENathan Braswell2023-02-21 10:32:26 -05:00
7f10d111acFix attempted in SuspendedPair, a lot more tests also work PE-wise now. Many are overflowing the stack, might be recursion causing or just really bad lack of tail calls, or a bug...Nathan Braswell2023-02-20 00:22:29 -05:00
Nathan Braswell