This finishes the work started in #14502 where atoms are owned by the
linker themselves. This now makes debug atoms fully owned by dwarf,
and no information is left stored on the decl.
The extended instructions starting with opcode `0xFC` are refactored
to make the work the same as the SIMD instructions. This means a
`Mir` instruction no longer contains a field 'secondary'. Instead,
we use the `payload` field to store the index into the extra list
which contains the extended opcode value. In case of instructions
such as 'memory.fill' which also have an immediate value, such
values will also be stored in the extra list right after the
instruction itself. This makes each `Mir` instruction smaller.
This implements `@splat` for vectors where the element type is
divisible by 8 and a power of two. This is fairly simple as we can
store the values directly within the virtual stack. But for
all other sizes, we must first shift and bitwise-or the values
before we can store them to fit them like a packed-struct, rather
than an array.
This implements `airSplat` for the native WebAssembly backend when
the features 'simd128' or 'relaxed-simd' are enabled. The commit
supports splat where the value lives in the linear memory segment,
as well as on the stack. This saves a lot of instruction cost.
When it detects the element type is not 8, 16, 32 or 64 bits,
the backend will instead use the same strategy as if the features
where disabled.
This implements loading and storing immediate values representing
a vector with exactly 128 bits. When the vector does not equal to
128 bits, or when the simd128 or relaxed-simd features are disabled
the value will be treated as an array instead.
I did not do a full audit, but I did notice a few issues which are
resolved in this commit. Probably it would be worth adding debug
infrastructure to assert that the number of reaps equals the number of
calls to resolveInst() per air lowering function.
This implements support for loading and storing where the lhs is
of pointer type with host_size != 0. e.g. when loading a specific
field from a packed struct with a non-byte alignment such as (0:1:3).
This allows the Wasm backend to construct an instance of a packed
struct during runtime. We first allocate a local, and then
shift+or each field's value into the result local. We then finally
return this result local as value.
The commit also fixes a type-issue in `airElemVal` where we used
the element type instead of a pointer type to store the value's
address into.
This also adds support loading a runtime pointer from a packed struct.
Also, this commit improves many utility functions such as `trunc` and
`intcast` to also support non-integer types such as booleans.
Simplifies the airStructFieldPtr(index) functions to only obtain the
correct struct type and field index, which is then passed into the
structFieldPtr function. This function now calculates the byte-offset
of the field's address and returns a new `WValue` with this offset.
This means we only have to do this calculation in a single function,
and no longer have to duplicate any logic. This also handles both
regular (tagged) unions and packed unions.
This implements loading a field from a packed struct, regardless of
its field's type. This means it supports pointers, floats and
integers. The commit also extracts the logic from airTrunc into its
own `trunc` function so it can be re-used.
When lowering constants of packed structs, which are smaller than 65
bits, we lower the value to an integer rather than store it in the
constant data section. This allows us to use an immediate value,
for quick loads and stores.
This is encoded as a primitive AIR instruction to resolve one corner
case: A function may include a `catch { ... }` or `else |err| { ... }`
block but not call any errorable fn. In that case, there is no error
return trace to save the index of and codegen needs to avoid
interacting with the non-existing error trace.
By using a primitive AIR op, we can depend on Liveness to mark this
unused in this corner case.
'Self' isn't a very good name to describe what it does.
This commit changes the type name into `CodeGen` and the parameter
to `func` as we're generating code for a function.
With this change, the backend's coding style is in line with the
self-hosted Wasm-linker.
When we return an operand directly as a result, we must call
`reuseOperand`. This commit ensures it's done for all currently-
implemented AIR instructions.
Rather than accepting a canonical branch and a target branch
we allow to directly merge a branch into the parent branch.
This is possible as there's no overlapping and we have infinite
registers to our availability. This makes merging a lot simpler.
When determining the type of a local (read: register), we would
previously subtract the stack locals also. However, this locals
are also within the same `locals` list, meaning the type of the
local we were retrieving was off by 2. This could create a validation
error when we re-use a local of a different type.
Upon a branch, we only allow locals to be freed which were allocated
within the same branch as where they die. This ensures that when two
or more branches target the same operand we do not try to free
it more than once. This does however not implement freeing the local
upon branch merging yet.
When reusing an operand it increases the reference count, then when
an operand dies it will only decrease the reference count. When
this reaches 0, the local will be virtually freed, meaning it can be
re-used for a new local.
By reference counting the locals, we can ensure that when we free
a local, no local will be reused while it still has references pointing
to it. This prevents misscompilations. The compiler will also panic if
we free a local more than we reference it, introducing extra safety to
ensure they match up.
This hooks reusal of locals into liveness analysis.
Meaning that when an operand dies, and is a local,
it will automatically be freed so it can be re-used
when a new local is required. The result of this, is
a lower allocation required for locals. Having less
locals means smaller binary size, as well as faster
compilation speed when loaded by the runtime.
