This repository was archived by the owner on Sep 8, 2025. It is now read-only.
Merge with upstream main branch#106
Merged
Merged
Conversation
Debug into doesn't work on Pulley anyway but it's better to return a first-class error rather than a panic. This commit fills out a simple missing instruction in the Pulley backend to ensure that compilation gets far enough to the DWARF transform where there's no pulley support and an error is returned.
This commit improves the logic of detecting when to rebuild the `test-programs` artifacts used during test by parsing the `*.d` files that Cargo emits as part of its compilation and using that as the `cargo:rerun-if-changed` directive. This not only includes what was previously depended on but additionally includes features such as `path` dependencies which might temporarily be used during development.
I always forget this and it always bites us within a few months. Alas.
… in 2024 edition (#10308)
This commit updates the aarch64 tests to use the new `*-arm` images from GitHub instead of running through QEMU emulation. This should be a bit faster but primarily helps build confidence that it runs on real hardware. I'll note that release builds aren't updated at this time to run on native hardware since that's moreso about glibc compatibility and it's a bit easier to keep the setup we currently have for that. If `*-arm` machines are noticably faster than the default x64 machines then we could in theory move everything over to an aarch64-based machine as opposed to just the aarch64 build.
Signed-off-by: shenpengfeng <xinhangzhou@icloud.com>
Way back in time, this struct originally contained the stack and fuel limits. Then it also got the epoch deadline. Then it also got the exit FP/PC and entry FP. Now it is just the place where we put per-store mutable data that is accessed by JIT code and must be shared between all `VMContext`s. So it is time to rename it. This commit is purely mechanical and just renames the type and various methods and variables that use/access it.
…ace (#10311) * wasmtime-wit-bindgen: emit a definition for all types in a wit The calculation of TypeInfo only reaches types which are passed to or from a function. For types which are not reachable, default to the defining them according to the ownership setting given to bindgen. I have my doubts that `with`-reuse of bindgen types actually works properly when bindgen is set to Ownership::Borrowing but thats out of scope for this PR, which is to fix #10090 * component-macro: bless bindgen test output
* sse orpd implementation assembler integration with isle format add clippy reason, reorder avx priority in isle bless tests for orpd create separate xmm module validate function rewrite sse condition add quote from manual for sse prefix format changes move Xmm bits under Reg * use new isle constructors for sse * remove unused function * minor changes
parentheses are not balanced here
* winch(aarch64): ABI integration This commit finalizes the ABI integration between Winch and Cranelift, notably: * Updates the Cranelift ABI to ensure that all the Winch register clobbers are taken into account. * Updates the Winch ABI to treat x28 as callee-saved, since it's used as the shadow stack pointer. The alternative to treating x28 as callee-saved is to treat it as caller-saved and save it when required e.g., at call-sites, even though this approach works, it's probably more efficient to perform a store/pop once per function, to minimize the number of move instructions required. There are still some changes needed in order to fully enable running spec tests for aarch64, however, this change is one step further. If interested, you can run the call.wast test via: cargo run -- wast -Ccompiler=winch tests/spec_testsuite/call_indirect.wast * Update disas tests
* Use `alloc_zeroed` to allocate dynamic table elements This allows us to get pre-zeroed memory from the global allocator, rather than needing to manually zero-initialize the elements. * Don't ask the global allocator to allocate a block of size zero
Previously we used `AssemblerImm*` and `AssemblerSimm*` throughout the x64 ISLE to indicate the type of immediate an instruction would receive. Alex has noted previously that this is unnecessary; it does after all create more ties between the `cranelift-codegen` ISLE and `cranelift-assembler-x64` that could possibly break in the future. This change removes those ties by using Rust types in ISLE (e.g., `u8`, `i8`, `u16`, etc.) and converting to the expected assembler type in the ISLE glue layer.
Hello, A possible typo in this text: "Forward" Should be "Foreword" "Forward" means to move ahead, but the correct term for an introduction is "Foreword". Thanks.
* Expose GC refs to Wasm in `gc_alloc_raw` libcall As we are returning a GC reference to Wasm, we need to mark that GC reference as exposed to Wasm. Fixes bytecodealliance/wasmtime#9669 * miri ignore test that calls wasm and therefore can't run in miri
Co-authored-by: Wasmtime Publish <wasmtime-publish@users.noreply.github.com>
* add component-model-async/{fused|futures|streams}.wast tests
This is another piece of #9582 which I'm splitting out to make review easier.
The fused.wast test exercises fused adapter generation for various flavors of
intercomponent async->async, async->sync, and sync->async calls.
The futures.wast and streams.wast tests exercise the various intrinsics
(e.g. `stream.read`, `future.close_writable`, etc.) involving `future`s and
`stream`s.
The remaining changes fill in some TODOs to make the tests pass, plus plumbing
for a few intrinsics which aren't needed for these tests but which set the
foundation for future tests.
Signed-off-by: Joel Dice <joel.dice@fermyon.com>
* address review feedback
Signed-off-by: Joel Dice <joel.dice@fermyon.com>
---------
Signed-off-by: Joel Dice <joel.dice@fermyon.com>
* Update wasm-tools to the latest revision * Update component async bits for new intrinsics * Ignore options for now * I truly, and fundamentally, do not understand `cargo vet` * Fix a test
* Winch: Fix extadd implementations * Add new Wast test to excluded tests
* Remove Function argument from define_function_bytes define_function_bytes is supposed to be used when the code was not compiled by Cranelift and thus wouldn't have any associated Function. * Avoid collecting the relocations into a Vec twice
* Add Engine/Component::deserialize_raw For targets without virtual memory support, the only mechanism for load code previously was to copy bytes from a provided slice into a separate, owned, allocated buffer containing a copy of the serialized module contents. This is expensive for constrained targets and prohibits embedded runtimes like doing things like running code directly out of memory-mapped devices such as NOR flash. The tradeoff for directly using the externally owned memory is that the caller must ensure that the code memory will not be written to for the lifetime of the CodeMemory. Loading code from externally owned memory is supported for all configurations but is expected to fail to deserialize on platforms that suport virtual memory and are attempting to load modules or components that require that the memory be made executable (native code rather than pulley). bytecodealliance/wasmtime#10245 * Fix tests for s390x Previously, there was a bad assumption that pulley32/64 were the only targets but there's also the "be" variants. Use the `pulley_host` helper to better get the right pulley target for the host.
* Remove hotswapping support from cranelift-jit It was originally introduced for cg_clif. cg_clif recently removed it's use of hotswapping as the way it is implemented in cranelift-jit has various issues like leaking memory, panicking when the memory allocator decided to put two functions more than 2GB away from each other and only supporting x86_64. Better hotswapping support will likely require a fundamentally different implementation. * Fix UB when using zero-sized allocations together with pcrel relocs
…10193) * CI * Source * Remove generic.wasm from source control
* implement andpd sse instruction implement sse xorpd, addpd, subpd, andpd implement addps, andps, subps * impl xorps * orps instr * bless clif tests * remove old isle emission * reorder instructions in ascending order of size, added comment
…t` (#10349)
Additionally, support alignment 16 for `v128`s inside GC objects.
We previously allowed the alignment and minimum block size to be different
values, *and* did not round allocation sizes up to the minimum block size. This
could lead to bugs like the following sequence of events:
* free list initially has a block 0x10..MAX
* alloc(size=16) -> 0x10
* free list now has one block: 0x20..MAX
* alloc(size=16) -> 0x20
* free list now has one block: 0x30..MAX
* dealloc(0x10, size=16)
* this merges 0x10..0x20 into 0x30..MAX because the gap between 0x20 and 0x30
is smaller than the minimum block size
* dealloc(0x20, size=16)
* triggers overlapping blocks assertion failure!
If we ensure that the minimum allocation size is a multiple of our supported
alignment and we clamp requested allocations' sizes to at least the minimum
block size, then we could avoid this.
It is simpler, however, to unify our supported alignment and our minimum block
size into the same value. This PR does that. No more need for multiple concepts
which interact in subtle ways.
* Expose default value constructors for Wasm types in host API * Added doc comments
`VMRuntimeLimits` was renamed to `VMStoreContext` a little while back. Split off from bytecodealliance/wasmtime#10503
* Update wasm-tools crates to 1.228.0 * Add missing canonical builtins * Add new component model feature flags * Vet wasm-tool dependencies using wildcard * Fix tests hidden behind new CM features Several CM-async tests were failing due to upstream changes to the canonical builtins. Here we add some new Wasmtime flags for enabling these new features, `wasm_component_model_async_builtins` and `wasm_component_model_async_stackful`, and propagate that far enough to enable the features for these tests. This also involved removing all the error context details from these builtins. Co-authored-by: Alex Crichton <alex@alexcrichton.com> * Add some additional plumbing for fuzz configuration prtest:full --------- Co-authored-by: Alex Crichton <alex@alexcrichton.com>
…10508) * Analyze whether a module requires a GC heap during its compilation And avoid forcing the allocation of a GC heap for modules that will not use it. Split out from bytecodealliance/wasmtime#9350 * Simplify condition for allocating GC heap
* Guest Profiling suport for component model This change adds support for using guest profiling with the component model. In addition to the core change to support attributing stack frames to constituient modules within a component, the cli `run` and `serve` commands are also updated to support using the `--profile=guest` cli option with components. bytecodealliance/wasmtime#8773 bytecodealliance/wasmtime#7669 * fixup! Guest Profiling suport for component model
Closes bytecodealliance/wasmtime#10460 This commit ensures stack arguments greater than 8 bytes are correctly aligned. Prior to this commit this wasn't an issue since for the default calling convention, all arguments were aligned to a fixed slot size (e.g, 8 bytes). With the introduction of vector types, namely, v128, we must ensure that arguments greater than 8 bytes in size are correctly type-sized aligned.
Cranelift's Winch calling convention will never have to deal with results in more than one register since Winch doesn't make use of the types that would trigger that case, namely, `types::I128`. In all SIMD cases types are bound to `types::MxN`, which map to a single float register.
* Bump MSRV to 1.84.0 Coupled with today's release of 1.86.0 * Fix tests on windows
The current version we have pinned panics on some newer Rust syntax which was hindering debugging in the wasip3-prototyping fork recently, so update the version here to its latest.
* Add 2 more columns to the default address width. * Print newlines between inline symbols which accidentally weren't present.
This file contains hidden or 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
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
20 participants
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.
Current merge commit is an attempt, I'm probably going to have to push updates here.