Rather than storing it in a local and returning that,
we now keep this on the stack as all internal functions
expect it to be on the stack already and therefore were
generating extra `local.set` instructions.
When a local is no longer referenced or used, free it
so the local can be re-used by another instruction.
This means we generate less locals. Freeing this local
is a manual action and must only be used on temporaries
or where we are sure the local is not referenced by a
different AIR instruction, as that creates UB.
We now also no longer store a `WValue` when its tag is set to `none`
as those may never be referenced by any AIR instruction.
An assertion is done to make sure we never store a reference to a
`stack` value in our resolved instructions.
We internally use a lot of `load`'s that used to put
the result in a newly created local. For instance, when is considered
byRef or when we need a specific field/element/bytes from a larger type.
However, sometimes we want to directly use this value and then forget about
it, which means storing it in a local first is wasted instructions as well
as wasted locals that shouldn't be generated in the first place.
With this change it's explicit and requires the usage of `toLocal`.
This also does it for `wrapBinOp` which internally uses the already
refactored `binOp` and `wrapOperand` heavily simplifying this
function and not duplicate the logic from `binOp`
By keeping the result on the stack, we prevent codegen
from generating unneccesary locals when we have subsequent instructions
that do not have to be re-used.
Rather than always creating a new local and storing the result of
a binary operation into said local, we now leave it on top of the stack.
This allows for better codegen as we need less instructions, as well
as less total amount of locals.
When a local is no longer needed (for instance, it was used as
a temporary during arithmetic), it can be appended to one of
the typed freelists. This allows us to re-use locals and therefore
require less locals, reducing the binary size, as well as runtime
initialization.
Removed the copy of param_names inside of Fn and changed to
implementation of getParamName to fetch to parameter name from the ZIR.
The signature of getParamName was also changed to take an additional
*Module argument.
Rather than lowering float negation as `0.0 - x`.
* Add AIR instruction for float negation.
* Add compiler-rt functions for f128, f80 negation
closes#11853
`genFunctype` now accepts calling convention, param types, and return type
as part of its function signature rather than `fnData`. This means
we no longer have to create a dummy for our intrinsic call abstraction.
This also adds support for f16 division and builtins such as `@ceil` & more.
Implements `@mulAdd` for floats with bitsize 16, where it generates
a call into compiler-rt's `fmaf` function. Note that arguments
for fmaf are different in order than `@mulAdd`.
This implements binary operations and comparisons
for floats with bitsize 16. It does this by calling into
compiler-rt to first extend the float to 32 bits, perform the operation,
and then finally truncate back to 16 bits. When loading and storing the f16,
we do this as an unsigned 16bit integer.
Implements the creation of an undefined symbol for a compiler-rt intrinsic.
Also implements the building of the function call to said compiler-rt intrinsic.
Rather than storing all the shifts in temporaries, we perform the correct
shifting without temporaries. This makes the runtime code more performant
and also the backend code is simplified as we have a singular abstraction.
This does however not support floats of bitsizes
different than 32 or 64. f16, f80, f126 will require
support for compiler-rt and are out-of-scope for this commit.
Signed integers are currently not supported either.
And use it to debug a LazySrcLoc in stage2 that is set to a bogus value.
The actual fix in this commit is:
```diff
- try sema.emitBackwardBranch(&child_block, call_src);
+ try sema.emitBackwardBranch(block, call_src);
```
Split type relocs into two kinds: local and global. Global relocs
use a global type resolver and calculate offset to the existing
definition of a type abbreviation.
Local relocs use offset in the abbrev section of the containing
atom plus addend to generate a local relocation.
Implements semantic analysis for the new try/try_inline ZIR
instruction. Adds the new try/try_ptr AIR instructions and implements
them for the LLVM backend.
Fixes not calling rvalue() for tryExpr in AstGen.
This is part of an effort to implement #11772.
* `?E` where E is an error set with only one field now lowers the same
as `bool`.
* Fix implementation of errUnionErrOffset and errUnionPayloadOffset to
properly compute the offset of each field. Also name them the same
as the corresponding LLVM functions and have the same function
signature, to avoid confusion. This fixes a bug where wasm was
passing the error union type instead of the payload type.
* Fix C backend handling of optionals with zero-bit payload types.
* C backend: separate out airOptionalPayload and airOptionalPayloadPtr
which reduces branching and cleans up control flow.
* Make Type.isNoReturn return true for error sets with no fields.
* Make `?error{}` have only one possible value (null).
This also fixes the instruction for all other integer bitsizes,
as it was previously assuming to always be a bool.
128 bit substraction was also fixed as it contained a bug where it swapped
lhs with rhs.
This also implments wrapping for arbitrary integer widths between 64 and 128.
`@truncate` was fixed where the wasm types between operand and result differentiated.
We solved this by first casting and then wrapping.
