Add compiler/testdata/go1.27.go, gated behind Go >= 1.27 (the version
that promoted generic methods out of the GenericMethods experiment),
covering both fixes from the previous two commits:
- genericMethod has a regular method and a generic method (its own
type parameter); boxing it into an interface must only include the
regular method in the runtime method set instead of panicking in
getTypeCodeName.
- onlyGenericMethod's sole method is generic, so its type code must
have hasMethodSet == false and no methodSet field at all, not an
empty one.
Verified this test panics on the pre-fix compiler/interface.go and
passes after it.
LLVM ComputeValueVTs recursively expands arrays and structs into one
value type per scalar leaf. SelectionDAG call lowering allocates data
structures proportional to this count, which makes very large values
exhaust memory or crash LLVM.
Count scalar leaves and use pointers for internal parameters and results
when the count exceeds 1024. A result pointer is the first parameter,
and aggregate parameters point to read-only memory. Exported function
types are unchanged.
Keep these SSA values in memory and copy them with memcpy when needed.
Handle calls, interfaces, maps, channels, selects, defers, goroutines,
phis, and multiple results. Update the expected compiler IR and re-enable
the native compress/flate tests.
Add compiler coverage for large aggregate parameters and results,
including function values, interfaces, maps, channels, selects, deferred
calls, goroutines, phis, and multiple results.
The interface lowering pass used a separate formatter for method
signatures. Same-named local aliases could therefore make distinct generic
methods compare equal. Reuse getTypeCodeName so interface checks preserve
type identity.
x/tools SSA names and go/types strings can preserve the source spelling of
type arguments, so aliases can make distinct instances collide. Use the
canonical type encoding for function names, synthetic local type owners,
instantiated named types, and method sets.
Add cases where same-named local aliases instantiate generic functions,
local types, and methods with float32 and float64. Record the current
collisions and incorrect results.
Go 1.27 introduced the //go:linknamestd directive, a standard-library
variant of //go:linkname that does not require importing "unsafe". The
iter package switched to it for referencing runtime.newcoro and
runtime.coroswitch, which caused "linker could not find symbol
iter.newcoro / iter.coroswitch" errors when building with Go 1.27.
Handle //go:linknamestd the same as //go:linkname, bypassing the unsafe
import requirement, and add test coverage in the pragma compiler test.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Modern golang.org/x/sys/unix (v0.36+) declares its linknames
detached from function declarations, e.g.:
func syscall_syscall(...)
//go:linkname syscall_syscall syscall.syscall
TinyGo's pragma parser only inspected function doc comments and
therefore missed these, producing link errors like:
undefined symbol: _golang.org/x/sys/unix.syscall_syscall
Extend parsePragmas to also walk the enclosing *ast.File's
free-standing comments for //go:linkname directives matching the
function's name. Function-attached directives still take
precedence. The existing 'unsafe' import gate is preserved.
Fixes#4395, #5365
Treat panicState as a bitmask so a recovered panic during Goexit
can clear the panic bit while preserving the pending Goexit bit.
This removes the separate deferFrame Goexit field and restores the
previous defer frame size.
Keep a pending Goexit separate from the active panic state so recovered
deferred panics do not cancel the original Goexit unwind. Goexit should
also ignore -panic=trap, since it is not a panic.
Wire testing.FailNow and SkipNow through runtime.Goexit, and run tests
and benchmarks in goroutines. This lets Goexit terminate the user
function without skipping test bookkeeping. Add Goexit/recover coverage
and enable crypto/ecdh in the Linux stdlib test list.
Instead of referring to an unused global, use a constant value. This is
safe even when using `-gc=none` (since no actual memory gets allocated)
which wasn't the case before. It should also reduce binary size by a few
bytes for most programs.
This means runtime.alloc isn't allocated every time but instead only
when it is actually needed. It also just cleans up the compiler code a
little bit, and makes the next commit a bit nicer.
* compiler, runtime, reflect: generate type-specific hash/equal for composite map keys
For map keys that are not trivially binary-comparable, the compiler now
generates type-specific hash and equal functions as LLVM IR instead of
going through the interface+reflection path. This covers comparable
types: strings, floats, complex numbers, interfaces, channels, and
composites containing any mix of these.
Previously, maps with composite keys containing strings or floats
converted the key to interface{}, hashed via reflection, and compared
through interface equality. Now the compiler walks struct fields and
array elements directly, dispatching to the right runtime helper for
each field type and storing keys at their actual type.
Struct keys are always handled field-by-field so padding bytes do not
affect equality or hashing. Blank fields are ignored, matching Go
equality. Generated hash/equal function names use canonical underlying
type structure so structurally identical key types can share generated
functions. Padding zeroing before map operations is no longer needed
because structs no longer use the binary key path.
Also fix reflect map iteration for interface-keyed maps: MapIter.Key
returns an interface Value for map[interface{}] keys instead of
unpacking to the concrete key kind.
* compiler: generate loops for array map key hash/equal
Previously, array key hash and equal functions were unrolled at compile
time, generating one block of IR per element. For large arrays like
[1000]int inside a struct with non-binary fields, this caused code
explosion.
Now, binary-element arrays dispatch directly to hash32/memequal for the
whole array. Non-binary-element arrays generate an LLVM IR loop. The
equal loop short-circuits on the first mismatch.
Small arrays are still unrolled instead of looping, keeping the simple
cases compact.
* reflect: fix at-runtime map issues from review, and more found locally
Maps created through reflect.MakeMap need hash/equal behavior that
matches compiler-created maps. Add hashmapMakeReflect for composite key
types, using runtime closures that reconstruct interface{} values from
raw key bytes and delegate to the interface hash and equality paths.
Interface-keyed maps are already stored as interface values, so use the
existing interface hash/equal helpers directly for those. This keeps
reflect insert, lookup, delete, and compiled lookup paths consistent.
Also fix addressable small values used as interface map keys or
interface map values. loadSmallValue puts small indirect values back in
the pointer-sized interface data field the same way valueInterfaceUnsafe
does.
* compiler, interp, reflect: fix pointer map literals; remove interface fallback
Package-level map literals with pointer keys (both *T and
unsafe.Pointer) crash the compiler: the interp pass panics when trying
to hash pointer data as raw bytes, because pointer values in the interp
memory model are symbolic identities that do not fit in a byte.
Fix this by setting a recoverable error flag instead of panicking. The
interp detects the error after each instruction and defers the map
insert to runtime init code, where real addresses are available for
hashing. This matches how the interp already handles other operations
it cannot evaluate at compile time.
With this fix, unsafe.Pointer can also be classified as a binary map
key, which was the last type requiring the interface-based fallback.
Since all comparable types now use either the binary or the
compiler-generated hash/equal path, remove the interface fallback from
the compiler and reflect packages.
* compiler, transform: always pass hash/equal function pointers to hashmapMakeGeneric
The compiler now always resolves the hash and equal functions at compile
time and passes them directly to hashmapMakeGeneric, instead of passing
an algorithm enum to hashmapMake and resolving at runtime. For string
keys, the runtime hashmapStringPtrHash/hashmapStringEqual functions are
referenced directly. For binary keys, hash32/memequal are referenced.
The old hashmapMake with alg enum is retained for reflect, which still
needs runtime resolution when creating maps dynamically.
The OptimizeMaps transform pass is updated to handle both hashmapMake
and hashmapMakeGeneric, and to recognize hashmapGenericSet in addition
to hashmapBinarySet and hashmapStringSet. The now-unused
hashmapCanGenerateHashEqual helper is removed.
* runtime: store large map keys and values indirectly
When a map key or value exceeds 128 bytes, the bucket now stores a
pointer to separately allocated memory instead of the data inline. This
matches Go's MapMaxKeyBytes/MapMaxElemBytes threshold and prevents
bucket sizes from exploding for large key/value types.
For example, map[[256]byte]int previously used 2128 bytes per bucket
(16 header + 256*8 keys + 8*8 values); now it uses 144 bytes per bucket
(16 header + 8*8 pointers + 8*8 values).
The indirection is fully encapsulated in the runtime via helper
functions. Store the computed key and value slot sizes on the hashmap so
all runtime and reflect paths use the same bucket layout, including
non-indirect keys and values.
Add big-key golden coverage and benchmarks. Make the benchmark vary
enough key bytes to exercise hashing.
* reflect: implement method-set based AssignableTo and Implements
Based on the design from #4376 by aykevl.
Fixes#4277, fixes#3580.
Co-authored-by: Ayke van Laethem <aykevanlaethem@gmail.com>
* builder: update expected binary sizes for reflect changes
* Make interface checks similar to invoke, allowing typeImplementsMethodSet and method info to be dropped when reflect is not present
* Add more tests that BigGo reflect tests
* Even more pruning
* Add go/token and net/url to passing tests
* Prune even further, I am less happy with this, though
* Update size test now that we are smaller
* Skip some tests
* elide method lists
* format, oops
* fix tests
* Add a panic, pull out constant to keep in sync
* Add debug info
* Remove code that was leftover from a previous refactor
---------
Co-authored-by: Ayke van Laethem <aykevanlaethem@gmail.com>
This simplifies the process of constructing and encoding layout bitmaps.
Instead of creating big integers and merging them, we can create a pre-sized bitmap and set positions within it.
This also changes the encoding logic to allow larger layouts to be encoded inline.
We would previously not encode a layout inline unless the size was less than the width of the data field.
This is overly conservative.
A layout can be encoded inline as long as:
1. The size fits within the size field.
2. All set bits in the bitmap fit into the data field.
The compiler now implements the copy builtin directly instead of calling sliceCopy.
The length is calculated with the llvm.umax.* intrinsics, and the move is performed by llvm.memmove.*.
Both of these operations are easily understood by LLVM's optimization passes.
The type's alignment is also provided to llvm.memmove.*, which is useful when rewriting the move.
Interp no longer needs to reimplement sliceCopy.
Some edge case handling was implemented by sliceCopy but not llvm.memmove.*/llvm.memcpy.*.
I copied this over, so copies of external slices should work now.
Volatile moves/copies are now run at runtime by interp.
There is a 4-byte size increase due to some confusing length logic in sendUSBPacket.
I will look at sendUSBPacket in a future PR.
This fixes two edge cases for min/max:
min/max(+0.0, -0.0) = -0.0, +0.0
min/max(number, NaN) = NaN, NaN
The compare and select method does not work here.
I switched to the llvm.minimum/llvm.maximum intrinsics which match the intended behavior.
The integer min/max were also swapped over to using intrinsics.
The compare and select path is now only used by strings.
The compiler needs to know whether a defer is in a loop to determine whether to allocate stack or heap memory.
Previously, this performed a DFS of the CFG every time a defer was found.
This resulted in time complexity jointly proportional to the number of defers and the number of blocks in the function.
Now, the compiler will instead use Tarjan's strongly connected components algorithm to find cycles in linear time.
The search is performed lazily, so this has minimal performance impact on functions without defers.
In order to implement Tarjan's SCC algorithm, additional state needed to be attached to the blocks.
I chose to merge all of the per-block state into a single slice to simplify memory management.
Strings are readonly, but the compiler doesn't always know this. Marking
them as readonly in the frontend allows the compiler to optimize based
on this knowledge.
This provides some small code size benefits. I didn't measure running
speed.
This switches the Espressif fork from LLVM 19 to LLVM 20, so we can use
the improvements made between those LLVM versions. It also better aligns
with the system-LLVM build method, which currently also defaults to LLVM
20.
Note that this disables the machine outliner for RISC-V. It appears
there's a bug in there somewhere, with the machine outliner enabled the
crypto/elliptic package tests fail with -target=riscv-qemu.
This should ideally be investigated and reported upstream.
This ensures that calls to print/println happening in different threads
are not interleaved. It's a task.PMutex, so this should only change
things when threading is used.
This matches the Go compiler, which does the same thing:
https://godbolt.org/z/na5KzE7en
The locks are not recursive, which means that we need to be careful to
not call `print` or `println` inside a runtime.print* implementation,
inside putchar (recursively), and inside signal handlers. Making them
recursive might be useful to do in the future, but it's not really
necessary.
This rewrite simplifies the channel implementation considerably, with
34% less LOC. Perhaps the most important change is the removal of the
channel state, which made sense when we had only send and receive
operations but only makes things more compliated when multiple select
operations can be pending on a single channel.
I did this rewrite originally to make it possible to make channels
parallelism-safe. The current implementation is not parallelism-safe,
but it will be easy to make it so (the main additions will be a channel
lock, a global select lock, and an atomic compare-and-swap in
chanQueue.pop).
This only works on declarations, not definitions. This is intentional:
it follows the upstream Go implemetation.
However, we might want to loosen this requirement at some point: TinyGo
sometimes stores pointers in memory mapped I/O knowing they won't
actually escape, but the compiler doesn't know about this.
compiler: align with current wasm types proposal
https://github.com/golang/go/issues/66984
- Remove int and uint as allowed types in params, results, pointers, or struct fields
- Only allow small integers in pointers, arrays, or struct fields
- enforce structs.HostLayout usage per wasm types proposal
https://github.com/golang/go/issues/66984
- require go1.23 for structs.HostLayout
- use an interface to check if GoVersion() exists
This permits TinyGo to compile with Go 1.21.
- use goenv.Compare instead of WantGoVersion
- testdata/wasmexport: use int32 instead of int
- compiler/testdata: add structs.HostLayout
- compiler/testdata: improve tests for structs.HostLayout
The values were stored in the passed object as the values itself (not
expanded like is common in the calling convention), and read back after
assuming they were expanded. This often works for simple parameters
(int, pointer, etc), but not for more complex parameters. Especially
when there's padding.
Found this while working on `//go:wasmexport`.
This adds something like 'align 4' to the runtime.alloc calls, so that
the compiler knows the alignment of the allocation and can optimize
accordingly.
The optimization is very small, only 0.01% in my small test. However, my
plan is to read the value in the heap-to-stack transform pass to make it
more correct and let it produce slightly more optimal code.
This commit adds support for LLVM 16 and switches to it by default. That
means three LLVM versions are supported at the same time: LLVM 14, 15,
and 16.
This commit includes work by QuLogic:
* Part of this work was based on a PR by QuLogic:
https://github.com/tinygo-org/tinygo/pull/3649
But I also had parts of this already implemented in an old branch I
already made for LLVM 16.
* QuLogic also provided a CGo fix here, which is also incorporated in
this commit:
https://github.com/tinygo-org/tinygo/pull/3869
The difference with the original PR by QuLogic is that this commit is
more complete:
* It switches to LLVM 16 by default.
* It updates some things to also make it work with a self-built LLVM.
* It fixes the CGo bug in a slightly different way, and also fixes
another one not included in the original PR.
* It does not keep compiler tests passing on older LLVM versions. I
have found this to be quite burdensome and therefore don't generally
do this - the smoke tests should hopefully catch most regressions.
When a function is exported using //export, but also had a
//go:wasm-module pragma, the //export name was ignored. The
//go:wasm-module doesn't actually do anything besides breaking the
export (exported functions don't have a module name).
I've refactored and cleaned up the code, and in the process removed this
weird edge case.
Damian Gryski
Jake Bailey
Pat Whittingslow
Nia Waldvogel
deadprogram
Evan Wies
Ayke van Laethem
Randy Reddig
frenkel26
Elliott Sales de Andrade