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)
New ZIR instructions:
* struct_decl_packed
* struct_decl_extern
New TZIR instruction: struct_field_ptr
Introduce `Module.Struct`. It uses `Value` to store default values and
abi alignments.
Implemented Sema.analyzeStructFieldPtr and zirStructDecl.
Some stuff I changed from `@panic("TODO")` to `log.warn("TODO")`.
It's becoming more clear that we need the lazy value mechanism soon;
Type is becoming unruly, and some of these functions have too much logic
given that they don't have any context for memory management or error
reporting.
There are some `@panic("TODO")` in there but I'm trying to get the
branch to the point where collaborators can jump in.
Next is to repair the seam between LazySrcLoc and codegen's expected
absolute file offsets.
Next up is reworking the seam between the LazySrcLoc emitted by Sema
and the byte offsets currently expected by codegen.
And then the big one: updating astgen.zig to use the new memory layout.
Add a new allocated_registers bitmap to keep track of all callee-saved
registers allocated during generation of this function.
Function(.arm).gen uses this data to generate instructions in the
function prologue and epilogue to push and pop these registers
respectively.
Previously, this would reuse an operand even if reuseOperand returned
false for both operands.
genArmBinOpCode was also changed to be more Three-address code oriented
in the process.
on the break instruction operands. This involves a new TZIR instruction,
br_block_flat, which represents a break instruction where the operand is
the result of a flat block. See the doc comments on the instructions for
more details.
How it works: when adding break instructions in semantic analysis, the
underlying allocation is slightly padded so that it is the size of a
br_block_flat instruction, which allows the break instruction to later
be converted without removing instructions inside the parent body. The
extra type coercion instructions go into the body of the br_block_flat,
and backends are responsible for dispatching the instruction correctly
(it should map to the same function calls for related instructions).
Motivating test case:
```zig
export fn _start() noreturn {
var x: u64 = 1;
var y: u32 = 2;
var thing: u32 = 1;
const result = if (thing == 1) x else y;
exit();
}
```
The main idea here is for astgen to output ideal ZIR depending on
whether or not the sub-expressions of a block consume the result
location. Here, neither `x` nor `y` consume the result location of the
conditional expression block, and so the ZIR should communicate the
result of the condbr using break instructions, not with the result
location pointer.
With this commit, this is accomplished:
```
%22 = alloc_inferred()
%23 = block({
%24 = const(TypedValue{ .ty = type, .val = bool})
%25 = deref(%18)
%26 = const(TypedValue{ .ty = comptime_int, .val = 1})
%27 = cmp_eq(%25, %26)
%28 = as(%24, %27)
%29 = condbr(%28, {
%30 = deref(%4)
< there is no longer a store instruction here >
%31 = break("label_23", %30)
}, {
%32 = deref(%11)
< there is no longer a store instruction here >
%33 = break("label_23", %32)
})
})
%34 = store_to_inferred_ptr(%22, %23) <-- the store is only here
%35 = resolve_inferred_alloc(%22)
```
However if the result location gets consumed, the break instructions
change to break_void, and the result value is communicated only by the
stores, not by the break instructions.
Implementation:
* The GenZIR scope that conditional branches uses now has an optional
result location pointer field and a count of how many times the
result location ended up being an rvalue (not consumed).
* When rvalue() is called on a result location for a block, it
increments this counter. After generating the branches of a block,
astgen for the conditional branch checks this count and if it is 2
then the store_to_block_ptr instructions are elided and it calls
rvalue() using the block result (which will account for peer type
resolution on the break operands).
astgen has many functions disabled until they can be reworked with these
new semantics. That will be done before merging the branch.
There are some new rules for astgen to follow regarding result locations
and what you are allowed/required to do depending on which one is passed
to expr(). See the updated doc comments of ResultLoc for details.
I also changed naming conventions of stuff in this commit, sorry about
that.
* fix wrong pointer const-ness when unwrapping optionals
* allow grouped expressions and orelse as lvalues
* ZIR for unwrapping optionals: no redundant deref
- add notes to please don't use rlWrapPtr, this function should be
deleted
* catch and orelse: better ZIR for non-lvalue: no redundant deref;
operate entirely on values. lvalue case still works properly.
- properly propagate the result location into the target expression
* Test harness: better output when tests fail due to compile errors.
* TZIR: add instruction variants. These allow fewer TZIR instructions to
be emitted from zir_sema. See the commit diff for per-instruction
documentation.
- is_null
- is_non_null
- is_null_ptr
- is_non_null_ptr
- is_err
- is_err_ptr
- optional_payload
- optional_payload_ptr
* TZIR: removed old naming convention instructions:
- isnonnull
- isnull
- iserr
- unwrap_optional
* ZIR: add instruction variants. These allow fewer ZIR instructions to
be emitted from astgen. See the commit diff for per-instruction
documentation.
- is_non_null
- is_null
- is_non_null_ptr
- is_null_ptr
- is_err
- is_err_ptr
- optional_payload_safe
- optional_payload_unsafe
- optional_payload_safe_ptr
- optional_payload_unsafe_ptr
- err_union_payload_safe
- err_union_payload_unsafe
- err_union_payload_safe_ptr
- err_union_payload_unsafe_ptr
- err_union_code
- err_union_code_ptr
* ZIR: removed old naming convention instructions:
- isnonnull
- isnull
- iserr
- unwrap_optional_safe
- unwrap_optional_unsafe
- unwrap_err_safe
- unwrap_err_unsafe
- unwrap_err_code
