Go 1.19 started reformatting code in a way that makes it more obvious
how it will be rendered on pkg.go.dev. It gets it almost right, but not
entirely. Therefore, I had to modify some of the comments so that they
are formatted correctly.
This change implements a new "scheduler" for WebAssembly using binaryen's asyncify transform.
This is more reliable than the current "coroutines" transform, and works with non-Go code in the call stack.
runtime (js/wasm): handle scheduler nesting
If WASM calls into JS which calls back into WASM, it is possible for the scheduler to nest.
The event from the callback must be handled immediately, so the task cannot simply be deferred to the outer scheduler.
This creates a minimal scheduler loop which is used to handle such nesting.
There is no reason to specialize this per chip as it is only ever used
for JavaScript. Not only that, it is causing confusion and is yet
another quirk to learn when porting the runtime to a new
microcontroller.
This commit refactors both determining the current time and sleeping for
a given time. It also improves precision for many chips.
* The nrf chips had a long-standing TODO comment about a slightly
inaccurate clock. This should now be fixed.
* The SAM D2x/D5x chips may have a slightly more accurate clock,
although probably within the error margin of the RTC. Also, by
working with RTC ticks and converting in the least number of places,
code size is often slightly reduced (usually just a few bytes, up to
around 1kB in some cases).
* I believe the HiFive1 rev B timer was slightly wrong (32768Hz vs
30517.6Hz). Because the datasheet says the clock runs at 32768Hz,
I've used the same conversion code here as in the nrf and sam cases.
* I couldn't test both stm32 timers, so I kept them as they currently
are. It may be possible to make them more efficient by using the
native tick frequency instead of using microseconds everywhere.
This scheduler is intended to live along the (stackless) coroutine based
scheduler which is needed for WebAssembly and unsupported platforms. The
stack based scheduler is somewhat simpler in implementation as it does
not require full program transform passes and supports things like
function pointers and interface methods out of the box with no changes.
Code size is reduced in most cases, even in the case where no scheduler
scheduler is used at all. I'm not exactly sure why but these changes
likely allowed some further optimizations somewhere. Even RAM is
slightly reduced, perhaps some global was elminated in the process as
well.
Previously it would use a bitcast, which cannot directly be used on AVR
because functions live in a different address space on AVR. To fix this,
use a ptrtoint/inttoptr pair.
This allows testdata/coroutines.go to be compiled, but due to what
appears to be an LLVM bug cannot be optimized and codegen'ed:
tinygo: /home/ayke/src/github.com/tinygo-org/tinygo/llvm-project/llvm/lib/IR/Constants.cpp:1776: static llvm::Constant *llvm::ConstantExpr::getBitCast(llvm::Constant *, llvm::Type *, bool): Assertion `CastInst::castIsValid(Instruction::BitCast, C, DstTy) && "Invalid constantexpr bitcast!"' failed.
This happens as one of the function passes after the TinyGo passes and
after the module has been verified so most likely it is a bug somewhere
in LLVM.
Instead of storing the value to send/receive in the coroutine promise,
store only a pointer in the promise. This simplifies the code a lot and
allows larger value sizes to be sent across a channel.
Unfortunately, this new system has a code size impact. For example,
compiling testdata/channel.go for the BBC micro:bit, there is an
increase in code size from 4776 bytes to 4856 bytes. However, the
improved flexibility and simplicity of the code should be worth it. If
this becomes an issue, we can always refactor the code at a later time.
This is implemented as follows:
* The parent coroutine allocates space for the return value in its
frame and stores a pointer to this frame in the parent coroutine
handle.
* The child coroutine obtains the alloca from its parent using the
parent coroutine handle. It then stores the result value there.
* The parent value reads the data from the alloca on resumption.
Support for channels is not complete. The following pieces are missing:
* Channels with values bigger than int. An int in TinyGo can always
contain at least a pointer, so pointers are okay to send.
* Buffered channels.
* The select statement.
Before this commit, goroutine support was spread through the compiler.
This commit changes this support, so that the compiler itself only
generates simple intrinsics and leaves the real support to a compiler
pass that runs as one of the TinyGo-specific optimization passes.
The biggest change, that was done together with the rewrite, was support
for goroutines in WebAssembly for JavaScript. The challenge in
JavaScript is that in general no blocking operations are allowed, which
means that programs that call time.Sleep() but do not start goroutines
also have to be scheduled by the scheduler.
This reduces complexity in the compiler without affecting binary sizes
too much.
Cortex-M0: no changes
Linux x64: no changes
WebAssembly: some testcases (calls, coroutines, map) are slightly bigger
time.Sleep now compiles on all systems, so lets use that.
Additionally, do a few improvements in time unit handling for the
scheduler. This should lead to somewhat longer sleep durations without
wrapping (on some platforms).
Some examples got smaller, some got bigger. In particular, code using
the scheduler got bigger and the blinky1 example got smaller (especially
on Arduino: 380 -> 314 bytes).
