Rewrite runtime, switch to tracing GC and bootstrap #1559
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
p3nGu1nZz
reviewed
Nov 25, 2020
|
Turns out that adding a shadow stack has one unfortunate consequence: It affects the |
|
Another: Putting incomplete objects onto the stack can break things if incremental GC just so happens to run in between allocating the object and properly initializing it. Looks like allocations need to zero memory to fix this, something we didn't do before to reduce overhead from zeroing memory we are going to initialize anyway. |
|
Meanwhile, the shadow stack thingy is getting together. Will prepare something for tomorrow, but for the time being, here's the bootstrap with incremental GC |
|
To not inflict too much hope here: Adjusting the GC parameters can still lead to what seems to be memory corruption or bugs in the incremental mechanism at some places, and I have concerns with a few side products like that users again have to retain/release everything externally because GC may kick in at any time in an incremental GC. Feels a bit like we'll get the cons of both reference counting and tracing as soon as this gets incremental. Will have more details to report at today's meeting. |
|
Last commit makes it so that the GC can be invoked either to perform a full collection with |
|
Could you sync this PR with master? |
|
So, here's something cool: |
|
At this point it passes all the tests, generating the exact same code the compiler compiled to JS produces, except that if one runs multiple tests in sequence there's still something surprising going on. I'm surprised. |
|
One error is that in - f32.const -7.888609052210118e-31
f32.const -0
+ f32.const -0 f64.const -8.900295434028806e-308
- f64.const -8.900295434028806e-308
+ f64.const -0 f64.const 2.225073858507201e-308
- f64.const 2.225073858507201e-308
f64.const 0
+ f64.const 0in certain test run situations. Wondering if that has something to do with float parsing, and not quite sure where these constants are in the |
This comment has been minimized.
This comment has been minimized.
1 task
|
Btw, using an optimized assemblyscript.wasm, I am seeing faster compile times compared to JS as expected.
Times for Wasm include a final full GC that's technically unnecessary when terminating right away, growing of memory that can be avoided by setting a reasonable initial memory size and overhead from interoping all the source texts into Wasm memory. |
|
FYI, Julia, the only language I know that allows triggering garbage collection manually, uses GC is a module (namespace) and gc is the function name. |
const _exports = {
rt: function () {
console.log("rt call");
}
}
_exports.rt.collect = function () {
console.log("__rt_collect call");
}usage: _exports.rt();
_exports.rt.collect(); |
|
Last commit revives the shadow stack pass and makes the
That's quite a price to pay for fully automatic GC, which was generally expected and has been extensively discussed in meetings, yet the significant code size hit is a bit surprising. From what I can tell so far:
Not much to do about that, but it's automatic. May be fine as long as it is not the default and users have options to do manual collect instead. Thoughts? |
|
Here's an unfortunate side effect. If one does for example assemblyscript.setGlobalAlias(compilerOptions,
__newString(alias),
__newString(name)
);one could expect that this just works with let aliasPtr = __pin(__newString(alias));
assemblyscript.setGlobalAlias(compilerOptions,
aliasPtr ,
__newString(name)
);
__unpin(aliasPtr);preventing premature collection through pinning, which is unfortunate. Currently evaluating how this situation can be avoided, as not having to pin arguments would be so much nicer. |
MaxGraey
reviewed
Jan 16, 2021
MaxGraey
reviewed
Jan 16, 2021
|
Last commit switches the default runtime to |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Removes the error-prone automatic reference counting implementation that has been haunting us for a long time and replaces it with a mark & sweep garbage collector. By doing so this PR also sets us up for initial experiments with Wasm GC due to introducing a very similar interface already.
__retain/__releasemechanism has been removed and replaced with a new__pin/__unpinmechanism.incremental(default),minimalandstub--exportRuntimeflag to achieve the effects of--runtime stub/fullwith any runtime (i.e. exporting__new,__pin,__unpinand__collect).