The ZIR instruction `union_init_ptr` is renamed to `union_init`.
I made it always use by-value semantics for now, not taking the time to
invest in result location semantics, in case we decide to change the
rules for unions. This way is much simpler.
There is a new AIR instruction: union_init. This is for a comptime known
tag, runtime-known field value.
vector_init is renamed to aggregate_init, which solves a TODO comment.
This implements #10113 for the self-hosted compiler only. It removes the
ability to override alignment of packed struct fields, and removes the
ability to put pointers and arrays inside packed structs.
After this commit, nearly all the behavior tests pass for the stage2 llvm
backend that involve packed structs.
I didn't implement the compile errors or compile error tests yet. I'm
waiting until we have stage2 building itself and then I want to rework
the compile error test harness with inspiration from Vexu's arocc test
harness. At that point it should be a much nicer dev experience to work
on compile errors.
The mechanism behind initializing a union's tag is a bit complicated,
depending on whether the union is initialized at runtime,
forced comptime, or implicit comptime.
`coerce_result_ptr` now does not force a block to be a runtime context;
instead of adding runtime instructions directly, it forwards analysis to
the respective functions for initializing optionals and error unions.
`validateUnionInit` now has logic to still emit a runtime
`set_union_tag` instruction even if the union pointer is comptime-known,
for the case of a pointer that is not comptime mutable, such as a
variable or the result of `@intToPtr`.
`validateStructInit` looks for a completely different pattern now; it
now handles the possibility of the corresponding AIR instruction for
the `field_ptr` to be missing or the corresponding `store` to be missing.
See the new comment added to the function for more details. An
equivalent change should probably be made to `validateArrayInit`.
`analyzeOptionalPayloadPtr` and `analyzeErrUnionPayloadPtr` functions now
emit a `optional_payload_ptr_set` or `errunion_payload_ptr_set`
instruction respectively if `initializing` is true and the pointer value
is not comptime-mutable.
`storePtr2` now tries the comptime pointer store before checking if the
element type has one possible value because the comptime pointer store
can have side effects of setting a union tag, setting an optional payload
non-null, or setting an error union to be non-error.
The LLVM backend `lowerParentPtr` function is improved to take into
account the differences in how the LLVM values are lowered depending on
the Zig type. It now handles unions correctly as well as additionally
handling optionals and error unions.
In the LLVM backend, the instructions `optional_payload_ptr_set` and
`errunion_payload_ptr_set` check liveness analysis and only do the side
effects in the case the result of the instruction is unused.
A few wasm and C backend test cases regressed, but they are due to TODOs
in lowering of constants, so this is progress.
LLVM doesn't support lowering union values, so we have to use unnamed
structs to do it, which means any type that contains a union as an
element, even if it is nested in another type, has to have a mechanism
to detect when it can't be lowered normally and has to resort itself to
an unnamed struct.
This includes arrays.
Get rid of `std.math.F80Repr`. Instead of trying to match the memory
layout of f80, we treat it as a value, same as the other floating point
types. The functions `make_f80` and `break_f80` are introduced to
compose an f80 value out of its parts, and the inverse operation.
stage2 LLVM backend: fix pointer to zero length array tripping LLVM
assertion. It now checks for when the element type is a zero-bit type
and lowers such thing the same way that pointers to other zero-bit types
are lowered.
Both stage1 and stage2 LLVM backends are adjusted so that f80 is lowered
as x86_fp80 on x86_64 and i386 architectures, and identical to a u80 on
others. LLVM constants are lowered in a less hacky way now that #10860
is fixed, by using the expression `(exp << 64) | fraction` using llvm
constants.
Sema is improved to handle c_longdouble by recursively handling it
correctly for whatever the float bit width is. In both stage1 and
stage2.
* F80Repr extern struct needs no explicit padding; let's match the
target padding.
* stage2: fix lowering of f80 constants.
* stage1: decide ABI size and alignment of f80 based on alignment of
u64. x86 has alignof u64 equal to 4 but arm has it as 8.
* stage2: fix Value.floatReadFromMemory to use F80Repr
replaceAllUsesWith requires the type to be unchanged. So we bitcast
the new global to the old type and use that as the thing to replace
old uses.
Fixes an LLVM assertion found while troubleshooting #10837.
LLVM union globals have to be lowered as unnamed structs if the
non-most-aligned field is the active tag. In this case it bubbles up so
that structs containing unions have the same restriction.
This fix needs to be applied to optionals and other callsites of
createNamedStruct.
The bug fixed in this commit was revealed in searching for
the cause of #10837.
Support for f128, comptime_float, and c_longdouble require improvements
to compiler_rt and will implemented in a later PR. Some of the code in
this commit could be made more generic, for instance `llvm.airSqrt`
could probably be `llvm.airUnaryMath`, but let's cross that
bridge when we get to it.
`ExternFn` will contain a maybe-lib-name if it was defined with
the `extern` keyword like so
```zig
extern "c" fn write(usize, usize, usize) usize;
```
`lib_name` will live as long as `ExternFn` decl does.
AstGen:
* rename the known_has_bits flag to known_non_opv to make it better
reflect what it actually means.
* add a known_comptime_only flag.
* make the flags take advantage of identifiers of primitives and the
fact that zig has no shadowing.
* correct the known_non_opv flag for function bodies.
Sema:
* Rename `hasCodeGenBits` to `hasRuntimeBits` to better reflect what it
does.
- This function got a bit more complicated in this commit because of
the duality of function bodies: on one hand they have runtime bits,
but on the other hand they require being comptime known.
* WipAnonDecl now takes a LazySrcDecl parameter and performs the type
resolutions that it needs during finish().
* Implement comptime `@ptrToInt`.
Codegen:
* Improved handling of lowering decl_ref; make it work for
comptime-known ptr-to-int values.
- This same change had to be made many different times; perhaps we
should look into merging the implementations of `genTypedValue`
across x86, arm, aarch64, and riscv.
This commit updates stage2 to enforce the property that the syntax
`fn()void` is a function *body* not a *pointer*. To get a pointer, the
syntax `*const fn()void` is required.
ZIR puts function alignment into the func instruction rather than the
decl because this way it makes it into function types. LLVM backend
respects function alignments.
Struct and Union have methods `fieldSrcLoc` to help look up source
locations of their fields. These trigger full loading, tokenization, and
parsing of source files, so should only be called once it is confirmed
that an error message needs to be printed.
There are some nice new error hints for explaining why a type is
required to be comptime, particularly for structs that contain function
body types.
`Type.requiresComptime` is now moved into Sema because it can fail and
might need to trigger field type resolution. Comptime pointer loading
takes into account types that do not have a well-defined memory layout
and does not try to compute a byte offset for them.
`fn()void` syntax no longer secretly makes a pointer. You get a function
body type, which requires comptime. However a pointer to a function body
can be runtime known (obviously).
Compile errors that report "expected pointer, found ..." are factored
out into convenience functions `checkPtrOperand` and `checkPtrType` and
have a note about function pointers.
Implemented `Value.hash` for functions, enum literals, and undefined values.
stage1 is not updated to this (yet?), so some workarounds and disabled
tests are needed to keep everything working. Should we update stage1 to
these new type semantics? Yes probably because I don't want to add too
much conditional compilation logic in the std lib for the different
backends.
* AIR instruction vector_init gains the ability to init arrays and
tuples in addition to vectors. This will probably also gain the
ability to initialize structs and be renamed to `aggregate_init`.
* AstGen prefers to use an `anon_array_init` ZIR instruction for
local variables when the init expr is an array literal and there is
no type.
This reverts commit d48e4245b68bf25c7f41804a5012ac157a5ee546.
I have no idea why this is failing Drone CI, but in a branch, reverting
this commit solved the problem.
It is the job of codegen backends to mark Decls that are referenced as
alive so that the frontend does not sweep them with the garbage. This
commit unifies the code between the backends with an added method on
Decl.
The implementation is more complete than before, switching on the Decl
val tag and recursing into sub-values.
As a result, two more array tests are passing.
AIR:
* `array_elem_val` is now allowed to be used with a vector as the array
type.
* New instructions: splat, vector_init
AstGen:
* The splat ZIR instruction uses coerced_ty for the ResultLoc, avoiding
an unnecessary `as` instruction, since the coercion will be performed
in Sema.
* Builtins that accept vectors now ignore the type parameter. Comment
from this commit reproduced here:
The accepted proposal #6835 tells us to remove the type parameter from
these builtins. To stay source-compatible with stage1, we still observe
the parameter here, but we do not encode it into the ZIR. To implement
this proposal in stage2, only AstGen code will need to be changed.
Sema:
* `clz` and `ctz` ZIR instructions are now handled by the same function
which accept AIR tag and comptime eval function pointer to
differentiate.
* `@typeInfo` for vectors is implemented.
* `@splat` is implemented. It takes advantage of `Value.Tag.repeated` 😎
* `elemValue` is implemented for vectors, when the index is a scalar.
Handling a vector index is still TODO.
* Element-wise coercion is implemented for vectors. It could probably
be optimized a bit, but it is at least complete & correct.
* `Type.intInfo` supports vectors, returning int info for the element.
* `Value.ctz` initial implementation. Needs work.
* `Value.eql` is implemented for arrays and vectors.
LLVM backend:
* Implement vector support when lowering `array_elem_val`.
* Implement vector support when lowering `ctz` and `clz`.
* Implement `splat` and `vector_init`.
It is possible for Zig to emit field ptr instructions to fields whos
type is zero sized. In this case llvm should return a pointer which
points to the next none zero sized parameter.
const locals now detect if the value ends up being comptime known. In
such case, it replaces the runtime AIR instructions with a decl_ref
const.
In the backends, some more sophisticated logic for marking decls as
alive was needed to prevent Decls incorrectly being garbage collected
that were indirectly referenced in such manner.
This commit fixes two problems:
* `zig build-obj` regressed from the cache-mode branch. It would crash
because it assumed that dirname on the emit bin path would not be
null. This assumption was invalid when outputting to the current
working directory - a pretty common use case for `zig build-obj`.
* When using the LLVM backend, `-fno-emit-bin` combined with any other
kind of emitting, such as `-femit-asm`, emitted nothing.
Both issues are now fixed.
Doc comments reproduced here:
This function is called by the frontend before flush(). It communicates that
`options.bin_file.emit` directory needs to be renamed from
`[zig-cache]/tmp/[random]` to `[zig-cache]/o/[digest]`.
The frontend would like to simply perform a file system rename, however,
some linker backends care about the file paths of the objects they are linking.
So this function call tells linker backends to rename the paths of object files
to observe the new directory path.
Linker backends which do not have this requirement can fall back to the simple
implementation at the bottom of this function.
This function is only called when CacheMode is `whole`.
This solves stack trace regressions on Windows and macOS because the
linker backends do not observe object file paths until flush().
Comment from this commit reproduced here:
LLVM does not allow us to change the type of globals. So we must
create a new global with the correct type, copy all its attributes,
and then update all references to point to the new global,
delete the original, and rename the new one to the old one's name.
This is necessary because LLVM does not support const bitcasting
a struct with padding bytes, which is needed to lower a const union value
to LLVM, when a field other than the most-aligned is active. Instead,
we must lower to an unnamed struct, and pointer cast at usage sites
of the global. Such an unnamed struct is the cause of the global type
mismatch, because we don't have the LLVM type until the *value* is created,
whereas the global needs to be created based on the type alone, because
lowering the value may reference the global as a pointer.
