| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| |
|
|
|
|
|
| |
This replaces all uses of __tempMemory__, the old scratch space location, with calls to function imports for scratch memory access. This lets us then implement those in a way that does not use the same heap as main memory. This avoids possible bugs with scratch memory overwriting something, or just in general that it has observable side effects, which can confuse fuzzing etc.
The intrinsics are currently implemented in the glue. We could perhaps emit them inline instead (but that might limit asm.js optimizations, so I wanted to keep our options open for now - easy to change later).
Also fixes some places where we used 0 as the scratch space address.
|
| |
|
|
|
|
| |
* Don't assume function types exist in legalize-js-interface.
* Properly handle (ignore) imports in RemoveNonJSOps - do not try to recurse into them.
* Run legalize-js-interface and remove-unused-module-elements in wasm2js, the first is necessary, the last is nice to have.
|
| | |
|
| |
|
|
|
|
| |
This reverts commit cb2d63586c08a3dd194d2b733ceb3f5051c081f8.
The issues with feature validation were mostly resolved in #1993, and
this PR finishes the job by adding feature flags to wasm-as to avoid
emitting the DataCount section when bulk-memory is not enabled.
|
| |
|
|
|
|
|
|
|
|
| |
This corresponds to changes made to the initialization procedure in
the spec. It also removes all the heavy initialization work from the
external interface of the interpreter, which is a nice encapsulation
win.
Implementation of the interpretation of the remaining bulk memory
operations and more rigorous tests of that interpretation will come in
a follow-up PR.
|
| |
|
|
|
| |
cases are tricky (#2026)
leave them for later optimizers/minifiers
|
| |
|
| |
Also emit the memory growth code based on memory growth, and not whether we are "use asm" or not.
|
| |
|
| |
Split them into two i32 globals.
|
| |
|
| |
Should prevent surprises in the future.
|
| |
|
| |
Fixes #1984
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Before, we'd print
if (..) label: { .. }; else ..
But that is wrong, as it ends the if too early. After this, we print
if (..) label: { .. } else ..
The bug was we checked if the if body was a block, but not if it was a labelled block.
|
| |
|
| |
We do not actually analyze scopes properly there, see comment in the commit. For now, just demote it to a warning. Not sure if it's worth fixing it or just removing it.
|
| |
|
| |
We forgot to set them as unsigned, which meant that we were reading uninitialized memory when checking that field, which mostly worked, except if it previously contained something...
|
| |
|
|
|
| |
In the absence of the target features section or command line flags. When there are command line flags, it is an error if they do not exactly match the target features section, except if --detect-features has been provided.
Also adds a --print-features pass to print the command line flags for all enabled options and uses it to make the feature tests more rigorous.
|
| |
|
|
|
| |
* Make the memory instrumentation pass log both pointers and values.
* Use "env" as the import module - simpler to support and get working.
|
| |
|
|
|
| |
* Emit an import statement for the memory.
* Update the imported memory's buffer when we grow.
|
| |
|
|
|
|
|
|
| |
Previously we were searching for the function by name but this
doesn't work when the internal name for the function is different.
In order to repro such a case the shared.c test was converted to C++
since then binaryen's internal name is different since it comes from
the de-mangled name section.
|
| |
|
| |
Fixes #2007 #2008
|
| |
|
|
|
| |
Its presence was causing validation errors in the Emscripten test
suite. This should be reverted once the default feature set is no
longer All.
|
| |
|
|
|
| |
This replaces the multiple asm.js tables (of power-of-2 size) with a single simple table.
Also supports importing the table.
|
| |
|
|
|
|
|
|
| |
* DataCount section
Read the DataCount section and verify that it agrees with the data
section. Also emit the DataCount section when bulk-memory is enabled
and there are a nonzero number of segments. Factor out some shared
unit test code.
|
| |
|
|
|
| |
This allows us to emit a (potentially modified) target features
section and conditionally emit other sections such as the DataCount
section based on the presence of features.
|
| |
|
|
|
|
|
|
| |
* I64ToI32Lowering - don't assume address 0 is a hardcoded location for scratch memory. Import __tempMemory__ for that.
* RemoveNonJSOps - also use __tempMemory__. Oddly here the address was a hardcoded 1024 (perhaps where the rust program put a static global?).
* Support imported ints in wasm2js, coercing them as needed.
* Add "env" import support in the tests, since now we emit imports from there.
* Make wasm2js tests split out multi-module tests using split_wast which is more robust and avoids emitting multiple outputs in one file (which makes no sense for ES6 modules)
|
| |
|
|
|
| |
* Refactor memory code to share it between the two emitting modes.
* Get memory emitting set up in the emscripten mode.
|
| |
|
|
|
|
| |
Add a wasm2js option for the glue to be in emscripten-compatible format (as opposed to ES6). This does a few things so far:
* Emit START_FUNCTIONS, END_FUNCTIONS markers in the code, for future use in the optimizer.
* Emit the glue as a function to be called from emscripten.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Early work for #1929
* Leave core wasm module - the "asm.js function" - to Wasm2JSBuilder, and add Wasm2JSGlue which emits the code before and after that. Currently that's some ES6 code, but we may want to change that later.
* Add add AssertionEmitter class for the sole purpose of emitting modules + assertions for testing. This avoids some hacks from before like starting from index 1 (assuming the module at first position was already parsed and printed) and printing of the f32Equal etc. functions not at the very top (which was due to technical limitations before).
Logic-wise, there should be no visible change, except some whitespace and reodering, and that I made the exceptions print out the source of the assertion that failed from the wast:
-if (!check2()) fail2();
+if (!check2()) throw 'assertion failed: ( assert_return ( call add ( i32.const 1 ) ( i32.const 1 ) ) ( i32.const 2 ) )';
(fail2 etc. did not exist, and seems to just have given a unique number for each assertion?)
|
| | |
|
| |
|
| |
We handled this for the normal case, but the optimizer can also precompute a return into a value, so check the output of the precomputation as well.
|
| |
|
|
|
|
| |
In relocatable code the constant offset might be relative to
__memory_base.
|
| |
|
|
|
| |
* Only look at the sign of loads when they actually matter.
* Prepare for imported globals (just refactoring/cleanup).
|
| |
|
|
|
|
|
| |
Get fuzzer to attempt to create almost all features. Pass v8 all the flags to allow that.
Fix fuzz bugs where we read signed_ even when it was irrelevant for that type of load.
Improve wasm-reduce on fuzz testcases, try to replace a node with drops of its children, not just the children themselves.
|
| |
|
|
| |
computations (#1990)
|
| |
|
|
| |
(#1989)
|
| |
|
|
|
|
|
|
| |
Hash the contents of all of memory and log that out in random places in the fuzzer, so we are more sensitive there and can catch memory bugs.
Fix UB that was uncovered by this in the binary writing code - if a segment is empty, we should not look at &vector[0], and instead use vector.data().
Add Builder::addExport convenience method.
|
| |
|
|
|
|
|
|
| |
Turns out there was already a precedent in emscripten for using
`fp$` for these functions.
Also, improve the heuristics for guessing the stack pointer global.
There are cases where we don't use have an explicit stack pointer at
all but *do* have both imported and exported globals.
|
| |
|
|
|
|
| |
It was previously part of writing a binary, but changing the number of
segments at such a late stage would not work in the presence of bulk
memory's datacount section. Also updates the memory packing pass
to respect the web's limits on the number of data segments.
|
| |
|
|
| |
optimizing to an unreachable (#1985)
|
| |
|
|
|
| |
Adds support for the bulk memory proposal's passive segments. Uses a
new (data passive ...) s-expression syntax to mark sections as
passive.
|
| | |
|
| |
|
| |
Emscripten runs wasm-emscripten-finalize before running wasm-opt, so the target features section is stripped out before optimizations are run. One option would have been to add another wasm-opt invocation at the very end to strip the target features section, but dumping the features as metadata avoids the extra tool invocation. In the long run, it would be nice to have only as single binaryen invocation to do all the work that needs doing.
|
| |
|
|
|
|
| |
This key is used by emscripten when building with MAIN_MODULE in order
to export global variables from the main module to the side modules.
|
| |
|
|
|
|
|
| |
If the user does not supply features explicitly on the command line,
read and use the features in the target features section for
validation and passes. If the user does supply features explicitly,
error if they are not a superset of the features marked as used in the
target features section and the user does not explicitly handle this.
|
| |
|
|
|
|
|
| |
Minus multi-memory which we don't support yet.
Improve validator.
Fix some minor validation issues in our tests.
|
| |
|
|
|
|
|
|
|
| |
This allows
wasm-opt --pass-arg=KEY:VALUE
where KEY and VALUE are strings. It is then added to passOptions.arguments, where passes can read it.
This is used in ExtractFunction instead of an env var.
|
| |\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* wip [ci skip]
* works
* test output
* test update
* js build
* better location for running directize
|
| | | |
|
| | | |
|
| | | |
|
