This commit brings in a number of minor updates for rustc's support for
the wasm target which has changed in the LLVM 9 update. Notable updates
include:
* The compiler now no longer manually inserts the `producers` section,
instead relying on LLVM to do so. LLVM uses the `llvm.ident` metadata
for the `processed-by` directive (which is now emitted on the wasm
target in this PR) and it uses debuginfo to figure out what `language`
to put in the `producers` section.
* Threaded WebAssembly code now requires different flags to be passed
with LLD. In LLD we now pass:
* `--shared-memory` - required since objects are compiled with
atomics. This also means that the generated memory will be marked as
`shared`.
* `--max-memory=1GB` - required with the `--shared-memory` argument
since shared memories in WebAssembly must have a maximum size. The
1GB number is intended to be a conservative estimate for rustc, but
it should be overridable with `-C link-arg` if necessary.
* `--passive-segments` - this has become the default for multithreaded
memory, but when compiling a threaded module all data segments need
to be marked as passive to ensure they don't re-initialize memory
for each thread. This will also cause LLD to emit a synthetic
function to initialize memory which users will have to arrange to
call.
* The `__heap_base` and `__data_end` globals are explicitly exported
since they're now hidden by default due to the `--export` flags we
pass to LLD.
The commit should have changed comments as well.
At the time of writting, it passes the tidy and check tool.
Revisions asked by eddyb :
- Renamed of all the occurences of {visit/super}_mir
- Renamed test structures `CachedMir` to `Cached`
Fixing the missing import on `AggregateKind`
Avoid symbol interning in `file_metadata`.
This commit changes `created_files` so it uses strings directly as keys,
rather than symbols derived from the strings. This avoids the cost of
having to do the hash table lookups to produce the symbols from the
strings.
The commit also uses `entry` to avoid doing a repeated hash table lookup
(`get` + `insert`).
Note that PR #60467 improved this code somewhat; this is a further
improvement.
r? @davidtwco
https://github.com/rust-lang/rust/commit/cff075009 made LLVM emit
less debuginfo when compiling with "line-tables-only". The change
was essentially correct but the reduced amount of debuginfo broke
a number of tools.
This commit reverts the change so we get back the old behavior,
until we figure out how to do this properly and give external
tools to adapt to the new format.
See https://github.com/rust-lang/rust/issues/60020 for more info.
This commit changes `created_files` so it uses strings directly as keys,
rather than symbols derived from the strings. This avoids the cost of
having to do the hash table lookups to produce the symbols from the
strings.
The commit also uses `entry` to avoid doing a repeated hash table lookup
(`get` + `insert`).
Note that PR #60467 improved this code somewhat; this is a further
improvement.
Multi-variant layouts for generators
This allows generators to overlap fields using variants, but doesn't do any such overlapping yet. It creates one variant for every state of the generator (unresumed, returned, panicked, plus one for every yield), and puts every stored local in each of the yield-point variants.
Required for optimizing generator layouts (#52924).
There was quite a lot of refactoring needed for this change. I've done my best in later commits to eliminate assumptions in the code that only certain kinds of types are multi-variant, and to centralize knowledge of the inner mechanics of generators in as few places as possible.
This change also emits debuginfo about the fields contained in each variant, as well as preserving debuginfo about stored locals while running in the generator.
Also, fixes#59972.
Future work:
- Use this change for an optimization pass that actually overlaps locals within the generator struct (#52924)
- In the type layout fields, don't include locals that are uninitialized for a particular variant, so miri and UB sanitizers can check our memory (see https://github.com/rust-lang/rust/issues/59972#issuecomment-483058172)
- Preserve debuginfo scopes across generator yield points
`file_metadata_raw` interns the strings `"<unknown>"` and `""` very
frequently. This commit avoids that, which reduces the number of symbols
interned significantly and reduces instruction counts by up to 0.5% on
some workloads.
This commit doesn't actually migrate to LLVM 9, but it brings our own
C++ bindings in line with LLVM 9 and able to compile against tip of
tree. The changes made were:
* The `MainSubprogram` flag for debuginfo moved between flag types.
* Iteration of archive members was tweaked slightly and we have to
construct the two iterators before constructing the returned
`RustArchiveIterator` value.
* The `getOrInsertFunction` binding now returns a wrapper which we use
`getCallee()` on to get the value we're interested in.
By using 'impl trait', it's possible to create a self-referential
type as follows:
fn foo() -> impl Copy { foo }
This is a function which returns itself.
Normally, the signature of this function would be impossible
to write - it would look like 'fn foo() -> fn() -> fn() ...'
e.g. a function which returns a function, which returns a function...
Using 'impl trait' allows us to avoid writing this infinitely long
type. While it's useless for practical purposes, it does compile and run
However, issues arise when we try to generate llvm debuginfo for such a
type. All 'impl trait' types (e.g. ty::Opaque) are resolved when we
generate debuginfo, which can lead to us recursing back to the original
'fn' type when we try to process its return type.
To resolve this, I've modified debuginfo generation to account for these
kinds of weird types. Unfortunately, there's no 'correct' debuginfo that
we can generate - 'impl trait' does not exist in debuginfo, and this
kind of recursive type is impossible to directly represent.
To ensure that we emit *something*, this commit emits dummy
debuginfo/type names whenever it encounters a self-reference. In
practice, this should never happen - it's just to ensure that we can
emit some kind of debuginfo, even if it's not particularly meaningful
Fixes#58463
rustc_target: factor out common fields of non-Single Variants.
@tmandry and I were discussing ways to generalize the current variants/discriminant layout to allow more fields in the "`enum`" (or another multi-variant types, such as potentially generator state, in the future), shared by all variants, than just the tag/niche discriminant.
This refactor should make it easier to extend multi-variant layouts, as nothing is duplicating anymore between "tagged enums" and "niche-filling enums".
r? @oli-obk
Fix invalid DWARF for enums when using ThinLTO
We were setting the same identifier for both the DW_TAG_structure_type
and the DW_TAG_variant_part. This becomes a problem when using ThinLTO
becauses it uses the identifier as a key for a map of types that is used
to delete duplicates based on the ODR, so one of them is deleted as a
duplicate, resulting in invalid DWARF.
The DW_TAG_variant_part isn't a standalone type, so it doesn't need
an identifier. Fix by omitting its identifier.
ODR uniquing is [enabled here](f21dee2c61/src/rustllvm/PassWrapper.cpp (L1101)).
Add FromStr impl for NonZero types
This is a WIP implementation because I do have some questions regarding the solution.
Somebody should ping the lang team on this I guess.
Please see the annotations on the code for more details.
Closes#58604
We were setting the same identifier for both the DW_TAG_structure_type
and the DW_TAG_variant_part. This becomes a problem when using thinlto
becauses it uses the identifier as a key for a map of types that is used
to delete duplicates based on the ODR, so one of them is deleted as a
duplicate, resulting in invalid DWARF.
The DW_TAG_variant_part isn't a standalone type, so it doesn't need
an identifier. Fix by omitting its identifier.
