I found this confusing because it includes the root item, plus the
inlined items reachable from the root item. The new formulation
separates the two parts more clearly.
Currently it overwrites all the CGUs with new CGUs. But those new CGUs
are just copies of the old CGUs, possibly with some things added. This
commit changes things so that each CGU just gets added to in place,
which makes things simpler and clearer.
It currently uses ranges, which index into `UsageMap::used_items`. This
commit changes it to just use `Vec`, which is much simpler to construct
and use. This change does result in more allocations, but it is few
enough that the perf impact is negligible.
`UsageMap` contains `used_map`, which maps from an item to the item it
uses. This commit add `user_map`, which is the inverse.
We already compute this inverse, but later on, and it is only held as a
local variable. Its simpler and nicer to put it next to `used_map`.
Currently, the code uses multiple words to describe when a mono item `f`
uses a mono item `g`, all of which have problems.
- `f` references `g`: confusing because there are multiple kinds of use,
e.g. "`f` calls `g`" is one, but "`f` takes a (`&T`-style) reference
of `g`" is another, and that's two subtly different meanings of
"reference" in play.
- `f` accesses `g`: meh, "accesses" makes me think of data, and this is
code.
- `g` is a neighbor (or neighbour) of `f`: is verbose, and doesn't
capture the directionality.
This commit changes the code to use "`f` uses `g`" everywhere. I think
it's better than the current terminology, and the consistency is
important.
Also, `InliningMap` is renamed `UsageMap` because (a) it was always
mostly about usage, and (b) the inlining information it did record was
removed in a recent commit.
We record inlining status for mono items in `MonoItems`, and then
transfer it to `InliningMap`, for later use in
`InliningMap::with_inlining_candidates`.
But we can just compute inlining status directly in
`InliningMap::with_inlining_candidates`, because the mono item is right
there. There's no need to compute it in advance.
This commit changes the code to do that, removing the need for
`MonoItems` and `InliningMap::inlines`. This does result in more calls
to `instantiation_mode` (one per static occurrence) but the performance
effect is negligible.
Remove `-Zcgu-partitioning-strategy`.
This option was introduced three years ago, but it's never been meaningfully used, and `default` is the only acceptable value.
Also, I think the `Partition` trait presents an interface that is too closely tied to the existing strategy and would probably be wrong for other strategies. (My rule of thumb is to not make something generic until there are at least two instances of it, to avoid this kind of problem.)
Also, I don't think providing multiple partitioning strategies to the user is a good idea, because the compiler already has enough obscure knobs.
This commit removes the option, along with the `Partition` trait, and the `Partitioner` and `DefaultPartitioning` types. I left the existing code in `compiler/rustc_monomorphize/src/partitioning/default.rs`, though I could be persuaded that moving it into
`compiler/rustc_monomorphize/src/partitioning/mod.rs` is better.
r? ``@wesleywiser``
This option was introduced three years ago, but it's never been
meaningfully used, and `default` is the only acceptable value.
Also, I think the `Partition` trait presents an interface that is too
closely tied to the existing strategy and would probably be wrong for
other strategies. (My rule of thumb is to not make something generic
until there are at least two instances of it, to avoid this kind of
problem.)
Also, I don't think providing multiple partitioning strategies to the
user is a good idea, because the compiler already has enough obscure
knobs.
This commit removes the option, along with the `Partition` trait, and
the `Partitioner` and `DefaultPartitioning` types. I left the existing
code in `compiler/rustc_monomorphize/src/partitioning/default.rs`,
though I could be persuaded that moving it into
`compiler/rustc_monomorphize/src/partitioning/mod.rs` is better.
I find that these structs obfuscate the code. Removing them and just
passing the individual fields around makes the `Partition` method
signatures a little longer, but makes the data flow much clearer. E.g.
- `codegen_units` is mutable all the way through.
- `codegen_units`'s length is changed by `merge_codegen_units`, but only
the individual elements are changed by `place_inlined_mono_items` and
`internalize_symbols`.
- `roots`, `internalization_candidates`, and `mono_item_placements` are
all immutable after creation, and all used by just one of the four
methods.
Three of the four methods in `DefaultPartitioning` are defined in
`default.rs`. But `merge_codegen_units` is defined in a separate module,
`merging`, even though it's less than 100 lines of code and roughly the
same size as the other three methods. (Also, the `merging` module
currently sits alongside `default`, when it should be a submodule of
`default`, adding to the confusion.)
In #74275 this explanation was given:
> I pulled this out into a separate module since it seemed like we might
> want a few different merge algorithms to choose from.
But in the three years since there have been no additional merging
algorithms, and there is no mechanism for choosing between different
merging algorithms. (There is a mechanism,
`-Zcgu-partitioning-strategy`, for choosing between different
partitioning strategies, but the merging algorithm is just one piece of
a partitioning strategy.)
This commit merges `merging` into `default`, making the code easier to
navigate and read.
- Pass a slice instead of an iterator to `debug_dump`.
- For each CGU set, print: the number of CGUs, the max and min size, and
the ratio of the max and min size (which indicates how evenly sized
they are).
- Print a `FINAL` entry, showing the absolute final results.
Run various queries from other queries instead of explicitly in phases
These are just legacy leftovers from when rustc didn't have a query system. While there are more cleanups of this sort that can be done here, I want to land them in smaller steps.
This phased order of query invocations was already a lie, as any query that looks at types (e.g. the wf checks run before) can invoke e.g. const eval which invokes borrowck, which invokes typeck, ...
Remove some uses of dynamic dispatch during monomorphization/partitioning.
This removes a few uses of dynamic dispatch and instead uses generics, as well as an enum to allow for other partitioning methods to be added later.
Add `rustc_fluent_macro` to decouple fluent from `rustc_macros`
Fluent, with all the icu4x it brings in, takes quite some time to compile. `fluent_messages!` is only needed in further downstream rustc crates, but is blocking more upstream crates like `rustc_index`. By splitting it out, we allow `rustc_macros` to be compiled earlier, which speeds up `x check compiler` by about 5 seconds (and even more after the needless dependency on `serde_json` is removed from `rustc_data_structures`).
Fluent, with all the icu4x it brings in, takes quite some time to
compile. `fluent_messages!` is only needed in further downstream rustc
crates, but is blocking more upstream crates like `rustc_index`. By
splitting it out, we allow `rustc_macros` to be compiled earlier, which
speeds up `x check compiler` by about 5 seconds (and even more after the
needless dependency on `serde_json` is removed from
`rustc_data_structures`).
Unify terminology used in unwind action and terminator, and reflect
the fact that a nounwind panic is triggered instead of an immediate
abort is triggered for this terminator.
