* Implement Sema for `@cmpxchgWeak` and `@cmpxchgStrong`. Both runtime
and comptime codepaths are implement.
* Implement Codegen for LLVM backend and C backend.
* Add LazySrcLoc.node_offset_builtin_call_argX 3...5
* Sema: rework comptime control flow.
- `error.ComptimeReturn` is used to signal that a comptime function
call has returned a result (stored in the Inlining struct).
`analyzeCall` notices this and handles the result.
- The ZIR instructions `break_inline`, `block_inline`,
`condbr_inline` are now redundant and can be deleted. `break`,
`block`, and `condbr` function equivalently inside a comptime scope.
- The ZIR instructions `loop` and `repeat` also are modified to
directly perform comptime control flow inside a comptime scope,
skipping an unnecessary mechanism for analysis of runtime code.
This makes Zig perform closer to an interpreter when evaluating
comptime code.
* Sema: zirRetErrValue looks at Sema.ret_fn_ty rather than sema.func
for adding to the inferred error set. This fixes a bug for
inlined/comptime function calls.
* Implement ZIR printing for cmpxchg.
* stage1: make cmpxchg respect --single-threaded
- Our LLVM C++ API wrapper failed to expose this boolean flag before.
* Fix AIR printing for struct fields showing incorrect liveness data.
In some cases (such as bitcast), an operand may be the same MCValue as
the result. If that operand died and was a register, it was freed by
processDeath. We have to "re-allocate" the register.
The big change in this commit is making `semaDecl` resolve the fields if
the Decl ends up being a struct or union. It needs to do this while
the `Sema` is still in scope, because it will have the resolved AIR
instructions that the field type expressions possibly reference. We do
this after the decl is populated and set to `complete` so that a `Decl`
may reference itself.
Everything else is fixes and improvements to make the test suite pass
again after making this change.
* New AIR instruction: `ptr_elem_ptr`
- Implemented for LLVM backend
* New Type tag: `type_info` which represents `std.builtin.TypeInfo`. It
is used by AstGen for the operand type of `@Type`.
* ZIR instruction `set_float_mode` uses `coerced_ty` to avoid
superfluous `as` instruction on operand.
* ZIR instruction `Type` uses `coerced_ty` to properly handle result
location type of operand.
* Fix two instances of `enum_nonexhaustive` Value Tag not handled
properly - it should generally be handled the same as `enum_full`.
* Fix struct and union field resolution not copying Type and Value
objects into its Decl arena.
* Fix enum tag value resolution discarding the ZIR=>AIR instruction map
for the child Sema, when they still needed to be accessed.
* Fix `zirResolveInferredAlloc` use-after-free in the AIR instructions
data array.
* Fix `elemPtrArray` not respecting const/mutable attribute of pointer
in the result type.
* Fix LLVM backend crashing when `updateDeclExports` is called before
`updateDecl`/`updateFunc` (which is, according to the API, perfectly
legal for the frontend to do).
* Fix LLVM backend handling element pointer of pointer-to-array. It
needed another index in the GEP otherwise LLVM saw the wrong type.
* Fix LLVM test cases not returning 0 from main, causing test failures.
Fixes a regression introduced in
6a5094872f.
* Implement comptime shift-right.
* Implement `@Type` for integers and `@TypeInfo` for integers.
* Implement union initialization syntax.
* Implement `zirFieldType` for unions.
* Implement `elemPtrArray` for a runtime-known operand.
* Make `zirLog2IntType` support RHS of shift being `comptime_int`. In
this case it returns `comptime_int`.
The motivating test case for this commit was originally:
```zig
test "example" {
var l: List(10) = undefined;
l.array[1] = 1;
}
fn List(comptime L: usize) type {
var T = u8;
return struct {
array: [L]T,
};
}
```
However I changed it to:
```zig
test "example" {
var l: List = undefined;
l.array[1] = 1;
}
const List = blk: {
const T = [10]u8;
break :blk struct {
array: T,
};
};
```
Which ended up being a similar, smaller problem. The former test case
will require a similar solution in the implementation of comptime
function calls - checking if the result of the function call is a struct
or union, and using the child `Sema` before it is destroyed to resolve
the fields.
instead of globally storing unresolved and tentative defs,
store indices to actual symbols in the functions that are responsible
for symbol resolution.
* Value: rename `error_union` to `eu_payload` and clarify the intended
usage in the doc comments. The way error unions is represented with
Value is fixed to not have ambiguous values.
* Fix codegen for error union constants in all the backends.
* Implement the AIR instructions having to do with error unions in the
LLVM backend.
* New AIR instructions: ptr_add, ptr_sub, ptr_elem_val, ptr_ptr_elem_val
- See the doc comments for details.
* Sema: implement runtime pointer arithmetic.
* Sema: implement elem_val for many-pointers.
* Sema: support coercion from `*[N:s]T` to `[*]T`.
* Type: isIndexable handles many-pointers.
fixes a crash when lowering `a == b` and they are of type bool.
I'm not worried about floats; I think we will probably add separate AIR
instructions for floats.
After this change, the frontend and backend cooperate to keep track of
which Decls are actually emitted into the machine code. When any backend
sees a `decl_ref` Value, it must mark the corresponding Decl `alive`
field to true.
This prevents unused comptime data from spilling into the output object
files. For example, if you do an `inline for` loop, previously, any
intermediate value calculations would have gone into the object file.
Now they are garbage collected immediately after the owner Decl has its
machine code generated.
In the frontend, when it is time to send a Decl to the linker, if it has
not been marked "alive" then it is deleted instead.
Additional improvements:
* Resolve type ABI layouts after successful semantic analysis of a
Decl. This is needed so that the backend has access to struct fields.
* Sema: fix incorrect logic in resolveMaybeUndefVal. It should return
"not comptime known" instead of a compile error for global variables.
* `Value.pointerDeref` now returns `null` in the case that the pointer
deref cannot happen at compile-time. This is true for global
variables, for example. Another example is if a comptime known
pointer has a hard coded address value.
* Binary arithmetic sets the requireRuntimeBlock source location to the
lhs_src or rhs_src as appropriate instead of on the operator node.
* Fix LLVM codegen for slice_elem_val which had the wrong logic for
when the operand was not a pointer.
As noted in the comment in the implementation of deleteUnusedDecl, a
future improvement will be to rework the frontend/linker interface to
remove the frontend's responsibility of calling allocateDeclIndexes.
I discovered some issues with the plan9 linker backend that are related
to this, and worked around them for now.
* AIR no longer has a `variables` array. Instead of the `varptr`
instruction, Sema emits a constant with a `decl_ref`.
* AIR no longer has a `ref` instruction. There is no longer any
instruction that takes a value and returns a pointer to it. If this
is desired, Sema must either create an anynomous Decl and return a
constant `decl_ref`, or in the case of a runtime value, emit an
`alloc` instruction, `store` the value to it, and then return the
`alloc`.
* The `ref_val` Value Tag is eliminated. `decl_ref` should be used
instead. Also added is `eu_payload_ptr` which points to the payload
of an error union, given an error union pointer.
In general, Sema should avoid calling `analyzeRef` if it can be helped.
For example in the case of field_val and elem_val, there should never be
a reason to create a temporary (alloc or decl). Recent previous commits
made progress along that front.
There is a new abstraction in Sema, which looks like this:
var anon_decl = try block.startAnonDecl();
defer anon_decl.deinit();
// here 'anon_decl.arena()` may be used
const decl = try anon_decl.finish(ty, val);
// decl is typically now used with `decl_ref`.
This pattern is used to upgrade `ref_val` usages to `decl_ref` usages.
Additional improvements:
* Sema: fix source location resolution for calling convention
expression.
* Sema: properly report "unable to resolve comptime value" for loads of
global variables. There is now a set of functions which can be
called if the callee wants to obtain the Value even if the tag is
`variable` (indicating comptime-known address but runtime-known value).
* Sema: `coerce` resolves builtin types before checking equality.
* Sema: fix `u1_type` missing from `addType`, making this type have a
slightly more efficient representation in AIR.
* LLVM backend: fix `genTypedValue` for tags `decl_ref` and `variable`
to properly do an LLVMConstBitCast.
* Remove unused parameter from `Value.toEnum`.
After this commit, some test cases are no longer passing. This is due to
the more principled approach to comptime references causing more
anonymous decls to get sent to the linker for codegen. However, in all
these cases the decls are not actually referenced by the runtime machine
code. A future commit in this branch will implement garbage collection
of decls so that unused decls do not get sent to the linker for codegen.
This will make the tests go back to passing.
* Add AIR instruction: struct_field_val
- This is part of an effort to eliminate the AIR instruction `ref`.
- It's implemented for C backend and LLVM backend so far.
* Rename `resolvePossiblyUndefinedValue` to `resolveMaybeUndefVal` just
to save some columns on long lines.
* Sema: add `fieldVal` alongside `fieldPtr` (renamed from
`namedFieldPtr`). This is part of an effort to eliminate the AIR
instruction `ref`. The idea is to avoid unnecessary loads, stores,
stack usage, and IR instructions, by paying a DRY cost.
LLVM backend improvements:
* internal linkage vs exported linkage is implemented, along with
aliases. There is an issue with incremental updates due to missing
LLVM API for deleting aliases; see the relevant comment in this commit.
- `updateDeclExports` is hooked up to the LLVM backend now.
* Fix usage of `Type.tag() == .noreturn` rather than calling `isNoReturn()`.
* Properly mark global variables as mutable/constant.
* Fix llvm type generation of function pointers
* Fix codegen for calls of function pointers
* Implement llvm type generation of error unions and error sets.
* Implement AIR instructions: addwrap, subwrap, mul, mulwrap, div,
bit_and, bool_and, bit_or, bool_or, xor, struct_field_ptr,
struct_field_val, unwrap_errunion_err, add for floats, sub for
floats.
After this commit, `zig test` on a file with `test "example" {}`
correctly generates and executes a test binary. However the
`test_functions` slice is undefined and just happens to be going into
the .bss section, causing the length to be 0. The next step towards
`zig test` will be replacing the `test_functions` Decl Value with the
set of test function pointers, before it is sent to linker/codegen.
* There is now a main_pkg in addition to root_pkg. They are usually the
same. When using `zig test`, main_pkg is the user's source file and
root_pkg has the test runner.
* scanDecl no longer looks for test decls outside the package being
tested. honoring `--test-filter` is still TODO.
* test runner main function has a void return value rather than
`anyerror!void`
* Sema is improved to generate better AIR for for loops on slices.
* Sema: fix incorrect capacity calculation in zirBoolBr
* Sema: add compile errors for trying to use slice fields as an lvalue.
* Sema: fix type coercion for error unions
* Sema: fix analyzeVarRef generating garbage AIR
* C codegen: fix renderValue for error unions with 0 bit payload
* C codegen: implement function pointer calls
* CLI: fix usage text
Adds 4 new AIR instructions:
* slice_len, slice_ptr: to get the ptr and len fields of a slice.
* slice_elem_val, ptr_slice_elem_val: to get the element value of
a slice, and a pointer to a slice.
AstGen gains a new functionality:
* One of the unused flags of struct decls is now used to indicate
structs that are known to have non-zero size based on the AST alone.
It incorrectly did not process the death of its operand. Additionally:
* delete dead code accidentally introduced in fe14e33945
* improve AIR printing code to include liveness data for operands.
Now an exclamation point ("!") indicates the tombstone of an AIR
instruction.
* Breaking language change: inline assembly must use string literal
syntax. This is in preparation for inline assembly improvements that
involve more integration with the Zig language. This means we cannot
rely on text substitution.
* Liveness: properly handle inline assembly and function calls with
more than 3 operands.
- More than 35 operands is not yet supported. This is a low priority
to implement.
- This required implementation in codegen.zig as well.
* Liveness: fix bug causing incorrect tomb bits.
* Sema: enable switch expressions that are evaluated at compile-time.
- Runtime switch instructions still need to be reworked in this
branch. There was a TODO left here (by me) with a suggestion to do
some bigger changes as part of the AIR memory reworking. Now that
time has come and I plan to honor the suggestion in a future commit
before merging this branch.
* AIR printing: fix missing ')' on alive instructions.
We're back to "hello world" working for the x86_64 backend.
Now the branch is compiling again, provided that one uses
`-Dskip-non-native`, but many code paths are disabled. The code paths
can now be re-enabled one at a time and updated to conform to the new
AIR memory layout.
to the link infrastructure, instead of being stored with Module.Fn. This
moves towards a strategy to make more efficient use of memory by not
storing Air or Liveness data in the Fn struct, but computing it on
demand, immediately sending it to the backend, and then immediately
freeing it.
Backends which want to defer codegen until flush() such as SPIR-V
must move the Air/Liveness data upon `updateFunc` being called and keep
track of that data in the backend implementation itself.
It's pretty compact, with each AIR instruction only taking up 4 bits,
plus a sparse table for special instructions such as conditional branch,
switch branch, and function calls with more than 2 arguments.
This commit changes the AIR file and the documentation of the memory
layout. The actual work of modifying the surrounding code (in Sema and
codegen) is not yet done.
* rename files to adhere to conventions
* remove unnecessary function / optionality
* fix merge conflict
* better panic message
* remove unnecessary TODO comment
* proper namespacing of declarations
* clean up documentation comments
* no copyright header needed for a brand new zig file that is not
copied from anywhere
AstGen had the then-else logic backwards for if expressions
on error unions. This commit fixes it.
Turns out AstGen only really needs `is_non_null` and `is_non_err`,
and does not need the `is_null` or `is_err` variants. So I removed the
`is_null{,_ptr}` and `is_err{,_ptr}` ZIR instructions (-4) and
added `is_non_err`, `is_non_err_ptr` ZIR instructions (+2) for
a total of (-2) ZIR instructions, giving us a tiny bit more headroom
within the 256 tag limit. This required swapping the order of
then/else blocks in a handful of cases, but ultimately means the
ZIR will be in the same as source order, which is convenient
when debugging.
AIR code on the other hand, gains the `is_non_err` and `is_non_err_ptr`
instructions.
Sema: fix logic in zirErrUnionCode and zirErrUnionCodePtr returning the
wrong result type.
* implement enough of ret_err_value to pass wasm tests
* only do the proper `@panic` implementation for the backends which
support it, which is currently only the C backend. The other backends
will see `@breakpoint(); unreachable;` same as before.
- I plan to do AIR memory layout reworking as a prerequisite to
fixing other backends, because that will help me put all the
constants up front, which will allow the codegen to lower to memory
without jumps.
* `@panic` is implemented using anon decls for the message. Makes it
easier on the backends. Might want to look into re-using decls for
this in the future.
* implement DWARF .debug_info for pointer-like optionals.
We can just use bitcast instead of error_to_int, int_to_error since
errorToInt and intToError do not actually do anything, just change types.
This allows us to remove 2 air ops that were the exact same as bitcast
- hash/eql functions moved into a Context object
- *Context functions pass an explicit context
- *Adapted functions pass specialized keys and contexts
- new getPtr() function returns a pointer to value
- remove functions renamed to fetchRemove
- new remove functions return bool
- removeAssertDiscard deleted, use assert(remove(...)) instead
- Keys and values are stored in separate arrays
- Entry is now {*K, *V}, the new KV is {K, V}
- BufSet/BufMap functions renamed to match other set/map types
- fixed iterating-while-modifying bug in src/link/C.zig
We've settled on the nomenclature for the artifacts the compiler
pipeline produces:
1. Tokens
2. AST (Abstract Syntax Tree)
3. ZIR (Zig Intermediate Representation)
4. AIR (Analyzed Intermediate Representation)
5. Machine Code
Renaming `ir` identifiers to `air` will come with the inevitable
air-memory-layout branch that I plan to start after the 0.8.0 release.
Conflicts:
* src/codegen/spirv.zig
* src/link/SpirV.zig
We're going to want to improve the stage2 test harness to print
the source file name when a compile error occurs otherwise std lib
contributors are going to see some confusing CI failures when they cause
stage2 AstGen compile errors.
Conflicts:
* build.zig
* src/Compilation.zig
* src/codegen/spirv/spec.zig
* src/link/SpirV.zig
* test/stage2/darwin.zig
- this one might be problematic; start.zig looks for `main` in the
root source file, not `_main`. Not sure why there is an underscore
there in master branch.
As it stands, the backend is incomplete, and there is no active contributor,
making it dead weight.
However, anyone is free to resurrect this backend at any time.
Conflicts:
* lib/std/os/linux.zig
* lib/std/os/windows/bits.zig
* src/Module.zig
* src/Sema.zig
* test/stage2/test.zig
Mainly I wanted Jakub's new macOS code for respecting stack size, since
we now depend on it for debug builds able to pass one of the test cases
for recursive comptime function calls with `@setEvalBranchQuota`.
The conflicts were all trivial.
instead of node indexes.
* AstGen: dbg_stmt instructions now have line and column indexes,
relative to the parent declaration. This allows codegen to emit debug
info without having the source bytes, tokens, or AST nodes loaded
in memory.
* ZIR: each decl has the absolute line number. This allows computing
line numbers from offsets without consulting source code bytes.
Memory management: creating a function definition does not prematurely
set the Decl arena. Instead the function is allocated with the general
purpose allocator.
Codegen no longer looks at source code bytes for any reason. They can
remain unloaded from disk.
* AstGen: add missing `break_inline` for comptime blocks.
* Module: call getTree() in byteOffset(). This generates the AST when
using cached ZIR and compile errors need to be reported.
* Scope.File: distinguish between successful ZIR generation and AIR
generation (when Decls in scope have been scanned).
- `semaFile` correctly avoids doing work twice.
* Implement first pass at `lookupInNamespace`. It has various TODOs
left, such as `usingnamespace`, and setting up Decl dependencies.
This was also an experiment to see if it were easier to implement a new
feature when using the instruction encoder.
Verdict: It's not that much easier, but I think it's certainly much more
readable, because the description of the Instruction annotates what each
field means. Right now, precise knowledge of x86_64 instructions is
still required because things like when to set the 64-bit flag, how to
read x86_64 instruction references, etc. are still not automatically
done for you.
In the future, this interface might make it sligtly easier to write an
assembler for x86_64, by abstracting the bit-fiddling aspects of
instruction encoding.
From my very cursory reading, it seems that the register manager doesn't
distinguish between registers that are physically the same but have
different sizes.
In that case, this means that during codegen, we can't rely on
`reg.size()` when determining the width of the operations we have to
perform. Instead, we must use some form of `ty.abiSize(self.target.*)`
to determine the size of the type we're operating with. If this size is
64 bits, then we should enable 64-bit operation.
This fixed a bug in the codegen for spilling instructions, which was
overwriting the previous stack entry with zeroes. See the modified test
case in this commit.
There are parts of it that I didn't modify because the byte
representation was important (e.g. we need to know what the exact
byte position where we store the address into the offset table is)
Instead of Module setting up the root_scope with the root source file,
instead, Module relies on the package table graph being set up properly,
and inside `update()`, it does the equivalent of `_ = @import("std");`.
This, in term, imports start.zig, which has the logic to call main (or
not). `Module` no longer has `root_scope` - the root source file is no
longer special, it's just in the package table mapped to "root".
I also went ahead and implemented proper detection of updated files.
mtime, inode, size, and source hash are kept in `Scope.File`.
During an update, iterate over `import_table` and stat each file to find
out which ones are updated.
The source hash is redundant with the source hash used by the struct
decl that corresponds to the file, so it should be removed in a future
commit before merging the branch.
* AstGen: add "previously declared here" notes for variables shadowing
decls.
* Parse imports as structs. Module now calls `AstGen.structDeclInner`,
which is called by `AstGen.containerDecl`.
- `importFile` is a bit kludgy with how it handles the top level Decl
that kinda gets merged into the struct decl at the end of the
function. Be on the look out for bugs related to that as well as
possibly cleaner ways to implement this.
* Module: factor out lookupDeclName into lookupIdentifier and lookupNa
* Rename `Scope.Container` to `Scope.Namespace`.
* Delete some dead code.
This branch won't work until `usingnamespace` is implemented because it
relies on `@import("builtin").OutputMode` and `OutputMode` comes from a
`usingnamespace`.
