This PR follows @hafeidejiangyou advice to not only enable end users to
avoid reflection when calling host functions, but also use that approach
ourselves internally. The performance results are staggering and will be
noticable in high performance applications.
Before
```
BenchmarkHostCall/Call
BenchmarkHostCall/Call-16 1000000 1050 ns/op
Benchmark_EnvironGet/environGet
Benchmark_EnvironGet/environGet-16 525492 2224 ns/op
```
Now
```
BenchmarkHostCall/Call
BenchmarkHostCall/Call-16 14807203 83.22 ns/op
Benchmark_EnvironGet/environGet
Benchmark_EnvironGet/environGet-16 951690 1054 ns/op
```
To accomplish this, this PR consolidates code around host function
definition and enables a fast path for functions where the user takes
responsibility for defining its WebAssembly mappings. Existing users
will need to change their code a bit, as signatures have changed.
For example, we are now more strict that all host functions require a
context parameter zero. Also, we've replaced
`HostModuleBuilder.ExportFunction` and `ExportFunctions` with a new type
`HostFunctionBuilder` that consolidates the responsibility and the
documentation.
```diff
ctx := context.Background()
-hello := func() {
+hello := func(context.Context) {
fmt.Fprintln(stdout, "hello!")
}
-_, err := r.NewHostModuleBuilder("env").ExportFunction("hello", hello).Instantiate(ctx, r)
+_, err := r.NewHostModuleBuilder("env").
+ NewFunctionBuilder().WithFunc(hello).Export("hello").
+ Instantiate(ctx, r)
```
Power users can now use `HostFunctionBuilder` to define functions that
won't use reflection. There are two choices of interfaces to use
depending on if that function needs access to the calling module or not:
`api.GoFunction` and `api.GoModuleFunction`. Here's an example defining
one.
```go
builder.WithGoFunction(api.GoFunc(func(ctx context.Context, params []uint64) []uint64 {
x, y := uint32(params[0]), uint32(params[1])
sum := x + y
return []uint64{sum}
}, []api.ValueType{api.ValueTypeI32, api.ValueTypeI32}, []api.ValueType{api.ValueTypeI32})
```
As you'll notice and as documented, this approach is more verbose and
not for everyone. If you aren't making a low-level library, you are
likely able to afford the 1us penalty for the convenience of reflection.
However, we are happy to enable this option for foundational libraries
and those with high performance requirements (like ourselves)!
Fixes#825
Signed-off-by: Adrian Cole <adrian@tetrate.io>
While compilers should be conservative when targeting WebAssembly Core
features, runtimes should be lenient as otherwise people need to
constantly turn on all features. Currently, most examples have to turn
on 2.0 features because compilers such as AssemblyScript and TinyGo use
them by default. This matches the policy with the reality, and should
make first time use easier.
This top-levels an internal type as `api.CoreFeatures` and defaults to
2.0 as opposed to 1.0, our previous default. This is less cluttered than
the excess of `WithXXX` methods we had prior to implementing all
planned WebAssembly Core Specification 1.0 features.
Finally, this backfills rationale as flat config types were a distinct
decision even if feature set selection muddied the topic.
Signed-off-by: Adrian Cole <adrian@tetrate.io>
This simplifies the calling convention and consolidates the call frame stack
and value stack into a single stack.
As a result, the cost of function calls decreases because we now don't need
to check the boundary twice (value and call frame stacks) at each function call.
The following is the result of the benchmark for recursive Fibonacci
function in integration_test/bench/testdata/case.go, and it shows that
this actually improves the performance of function calls.
[amd64]
name old time/op new time/op delta
Invocation/compiler/fib_for_5-32 109ns ± 3% 81ns ± 1% -25.86% (p=0.008 n=5+5)
Invocation/compiler/fib_for_10-32 556ns ± 3% 473ns ± 3% -14.99% (p=0.008 n=5+5)
Invocation/compiler/fib_for_20-32 61.4µs ± 2% 55.9µs ± 5% -8.98% (p=0.008 n=5+5)
Invocation/compiler/fib_for_30-32 7.41ms ± 3% 6.83ms ± 3% -7.90% (p=0.008 n=5+5)
[arm64]
name old time/op new time/op delta
Invocation/compiler/fib_for_5-10 67.7ns ± 1% 60.2ns ± 1% -11.12% (p=0.000 n=9+9)
Invocation/compiler/fib_for_10-10 487ns ± 1% 460ns ± 0% -5.56% (p=0.000 n=10+9)
Invocation/compiler/fib_for_20-10 58.0µs ± 1% 54.3µs ± 1% -6.38% (p=0.000 n=10+10)
Invocation/compiler/fib_for_30-10 7.12ms ± 1% 6.67ms ± 1% -6.31% (p=0.000 n=10+9)
Signed-off-by: Takeshi Yoneda <takeshi@tetrate.io>
This commit changes wazeroir.OperationSwap to OperationSet which is the
combination of OperationSwap and Drop in the previous implementation.
Previously, OperationSwap was always followed by OperationDrop on the swapped value on top.
Because of that, we had a redundant register allocation in Swap.
As a result, we use only one register in OperationSet which is a part of translations of
local.tee and local.set.
Signed-off-by: Takeshi Yoneda takeshi@tetrate.io
Before, we allowed stubbed host functions to be defined in wasm instead
of Go. This improves performance and reduces a chance of side-effects vs
Go. In fact, any pure function was supported in wasm, provided it only
called pure functions.
This changes internals so that a wasm-defined host function can use
memory. Notably, host functions use the caller's memory, so this is
simpler to initially support in the interpreter.
This is needed to simplify and reduce performance hit of GOARCH=wasm,
GOOS=js code, which perform a lot of memory reads and do not have
idiomatic signatures.
Note: wasm-defined host functions remain internal until we gain
experience, at least conclusion of the wasm_exec host module.
Signed-off-by: Adrian Cole <adrian@tetrate.io>
* Makes CacheNumInUint64 lazy and stops crashing in assemblyscript
This makes CacheNumInUint64 lazy so that all tests for function types
don't need to handle it. This also changes the assemblyscript special
functions so they don't crash when attempting to log. Finally, this
refactors `wasm.Func` so that it can enclose the parameter names as it
is more sensible than defining them elsewhere.
Signed-off-by: Adrian Cole <adrian@tetrate.io>
This completes the implementation of SIMD proposal for both
the interpreter and compiler(amd64).
This also fixes#210 by adding the complete documentation
over all the wazeroir operations.
Signed-off-by: Takeshi Yoneda <takeshi@tetrate.io>
Co-authored-by: Crypt Keeper <64215+codefromthecrypt@users.noreply.github.com>
This drops the text format (%.wat) and renames
InstantiateModuleFromCode to InstantiateModuleFromBinary as it is no
longer ambiguous.
We decided to stop supporting the text format as it isn't typically used
in production, yet costs a lot of work to develop. Given the resources
available and the increased work added with WebAssembly 2.0 and soon
WASI 2, we can't afford to spend the time on it.
The old parser is used only internally and will eventually be moved to
its own repository named watzero, possibly towards archival.
See #59
Signed-off-by: Adrian Cole <adrian@tetrate.io>
This implements various SIMD instructions related to
load, store, and lane manipulations for all engines.
Notablely, now our engines pass the following specification tests:
* simd_address.wast
* simd_const.wast
* simd_align.wast
* simd_laod16_lane.wast
* simd_laod32_lane.wast
* simd_laod64_lane.wast
* simd_laod8_lane.wast
* simd_lane.wast
* simd_load_extend.wast
* simd_load_splat.wast
* simd_load_zero.wast
* simd_store.wast
* simd_store16_lane.wast
* simd_store32_lane.wast
* simd_store64_lane.wast
* simd_store8_lane.wast
part of #484
Signed-off-by: Takeshi Yoneda <takeshi@tetrate.io>
Co-authored-by: Adrian Cole <adrian@tetrate.io>
This commit implements the v128.const, i32x4.add and i64x2.add in
interpreter mode and this adds support for the vector value types in the
locals and globals.
Notably, the vector type values can be passed and returned by exported functions
as well as host functions via two-uint64 encodings as described in #484 (comment).
Note: implementation of these instructions on JIT will be done in subsequent PR.
part of #484
Signed-off-by: Takeshi Yoneda <takeshi@tetrate.io>
This commit completes the reference-types proposal implementation.
Notably, this adds support for
* `ref.is_null`, `ref.func`, `ref.is_null` instructions
* `table.get`, `table.set`, `table.grow`, `table.size` and `table.fill` instructions
* `Externref` and `Funcref` types (including invocation via uint64 encoding).
part of #484
Signed-off-by: Takeshi Yoneda <takeshi@tetrate.io>
This performs several changes to allow compilation config to be
centralized and scoped properly. The immediate effects are that we can
now process external types during `Runtime.CompileModule` instead of
doing so later during `Runtime.InstantiateModule`. Another nice side
effect is memory size problems can err at a source line instead of
having to be handled in several places.
There are some API effects to this, and to pay for them, some less used
APIs were removed. The "easy APIs" are left alone. For example, the APIs
to compile and instantiate a module from Go or Wasm in one step are left
alone.
Here are the changes, some of which are only for consistency. Rationale
is summarized in each point.
* ModuleBuilder.Build -> ModuleBuilder.Compile
* The result of this is similar to `CompileModule`, and pairs better
with `ModuleBuilder.Instantiate` which is like `InstantiateModule`.
* CompiledCode -> CompiledModule
* We punted on this name, the result is more than just code. This is
better I think and more consistent as it introduces less terms.
* Adds CompileConfig param to Runtime.CompileModule.
* This holds existing features and will have future ones, such as
mapping externtypes to uint64 for wasm that doesn't yet support it.
* Merges Runtime.InstantiateModuleWithConfig with Runtime.InstantiateModule
* This allows us to explain APIs in terms of implicit or explicit
compilation and config, vs implicit, kindof implicit, and explicit.
* Removes Runtime.InstantiateModuleFromCodeWithConfig
* Similar to above, this API only saves the compilation step and also
difficult to reason with from a name POV.
* RuntimeConfig.WithMemory(CapacityPages|LimitPages) -> CompileConfig.WithMemorySizer
* This allows all error handling to be attached to the source line
* This also allows someone to reduce unbounded memory while knowing
what its minimum is.
* ModuleConfig.With(Import|ImportModule) -> CompileConfig.WithImportRenamer
* This allows more types of import manipulation, also without
conflating functions with globals.
* Adds api.ExternType
* Needed for ImportRenamer and will be needed later for ExportRenamer.
Signed-off-by: Adrian Cole <adrian@tetrate.io>
This commit adds support for multiple tables per module.
Notably, if the WithFeatureReferenceTypes is enabled,
call_indirect, table.init and table.copy instructions
can reference non-zero indexed tables.
part of #484
Signed-off-by: Takeshi Yoneda <takeshi@tetrate.io>
WebAssembly Core Working Draft 1 recently came out. Before that, we had
a toe-hold feature bucked called FinishedFeatures. This replaces
`RuntimeConfig.WithFinishedFeatures` with `RuntimeConfig.WithWasmCore2`.
This also adds `WithWasmCore1` for those who want to lock into 1.0
features as opposed to relying on defaults.
This also fixes some design debt where we hadn't finished migrating
public types that require constructor functions (NewXxx) to interfaces.
By using interfaces, we prevent people from accidentally initializing
key configuration directly (via &Xxx), causing nil field refs. This also
helps prevent confusion about how to use the type (ex pointer or not) as
there's only one way (as an interface).
See https://github.com/tetratelabs/wazero/issues/516
Signed-off-by: Adrian Cole <adrian@tetrate.io>
`Runtime.WithMemoryCapacityPages` is a function that determines memory
capacity in pages (65536 bytes per page). The inputs are the min and
possibly nil max defined by the module, and the default is to return
the min.
Ex. To set capacity to max when exists:
```golang
c.WithMemoryCapacityPages(func(minPages uint32, maxPages *uint32) uint32 {
if maxPages != nil {
return *maxPages
}
return minPages
})
```
Note: This applies at compile time, ModuleBuilder.Build or Runtime.CompileModule.
Fixes#500
Signed-off-by: Adrian Cole <adrian@tetrate.io>
This adds Memory.IndexByte which allows efficent scanning for a
delimiter, ex NUL(0) in null-terminated strings.
This also adds an ad-hoc test to ensure we can export memory functions
such as grow. While this is implicitly in the spectests, it is easier to
find in the ad-hoc tests.
Signed-off-by: Adrian Cole <adrian@tetrate.io>
This is an API breaking change that does a few things:
* Stop encouraging practice that can break context propagation:
* Stops caching `context.Context` in `wazero.RuntimeConfig`
* Stops caching `context.Context` in `api.Module`
* Fixes context propagation in function calls:
* Changes `api.Function`'s arg0 from `api.Module` to `context.Context`
* Adds `context.Context` parameter in instantiation (propagates to
.start)
* Allows context propagation for heavy operations like compile:
* Adds `context.Context` as the initial parameter of `CompileModule`
The design we had earlier was a good start, but this is the only way to
ensure coherence when users start correlating or tracing. While adding a
`context.Context` parameter may seem difficult, wazero is a low-level
library and WebAssembly is notoriously difficult to troubleshoot. In
other words, it will be easier to explain to users to pass (even nil) as
the context parameter vs try to figure out things without coherent
context.
Signed-off-by: Adrian Cole <adrian@tetrate.io>
This commit makes it possible for functions to be compiled before instantiation.
Notably, this adds CompileModule method on Engine interface where we pass
wasm.Module (which is the decoded module) to engines, and engines compile
all the module functions and caches them keyed on *wasm.Module.
In order to achieve that, this stops the compiled native code from embedding typeID
which is assigned for all the function types in a store.
Signed-off-by: Takeshi Yoneda <takeshi@tetrate.io>
This implements our own assertion library so that we can remove the
testify dependency. I've changed a lot of call sites so that we only
have to maintain a minimal amount of assertions.
Signed-off-by: Adrian Cole <adrian@tetrate.io>
Thanks to #454, now the compiled binary (code segment) can be reused for
multiple module instances originating from the same source (wasm.Module).
This commit introduces the caching mechanism on engine where it caches
compiled functions keyed on `wasm.Module`. As a result, this allows us to
do the fast module instantiation from the same *CompiledCode.
In order to release the cache properly, this also adds `Close` method
on CompiledCode.
Here's some bench result for instantiating multiple modules from the same CompiledCode:
```
name old time/op new time/op delta
Initialization/interpreter-32 2.84ms ± 3% 0.06ms ± 1% -97.73% (p=0.008 n=5+5)
Initialization/jit-32 10.7ms ±18% 0.1ms ± 1% -99.52% (p=0.008 n=5+5)
name old alloc/op new alloc/op delta
Initialization/interpreter-32 1.25MB ± 0% 0.15MB ± 0% -88.41% (p=0.008 n=5+5)
Initialization/jit-32 4.46MB ± 0% 0.15MB ± 0% -96.69% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Initialization/interpreter-32 35.2k ± 0% 0.3k ± 0% -99.29% (p=0.008 n=5+5)
Initialization/jit-32 94.1k ± 0% 0.2k ± 0% -99.74% (p=0.008 n=5+5)
```
Signed-off-by: Takeshi Yoneda <takeshi@tetrate.io>
This starts the process of removing all dependencies from wazero, by
isolating all assertions we use into a single file. This allows us to
port those assertions as we have time, and when twitchy is gone, the
project literally has no dependencies except go!
Signed-off-by: Adrian Cole <adrian@tetrate.io>
