This PR improves the detection of situations where we branch multiple
times on the same value in the
code generator. Previously this would only consider repeated branching
on function arguments, now on
arbitrary values.
Closes: #11018
This PR improves join point finding in the compiler through two means:
1. We now handle situations where a function `f` can only become a join
point when a function `g`
becomes a join point as well correctly.
2. We introduce a second join point finding pass after specialisation
and before the following
simplification pass, as the specialiser might have introduced new join
point opportunities for
the simplifier to exploit.
Notably in the code from #10995 we now correctly detect the missing join
point which required both
of these changes to be made.
Closes: #10995
This PR makes the eager lambda lifting heuristic more predictable by
blocking it from lifting from
any kind of inlineable function, not just `@[inline]`. It also adapts
the doc-string to describe
what is actually going on.
This PR performs more widening in ElimDeadBranches in an attempt to
improve performance in situations with a lot of local precision.
While this is not enough to make the compilation instant it pushes
compilation time from 12s to 3s for the example in #10857 and barely
introduces regressions so it seems like a good first step in this
direction.
Closes: #10857
This PR implements zero cost `BaseIO` by erasing the `IO.RealWorld`
parameter from argument lists and structures. This is a **major breaking
change for FFI**.
Concretely:
- `BaseIO` is defined in terms of `ST IO.RealWorld`
- `EIO` (and thus `IO`) is defined in terms of `EST IO.RealWorld`
- The opaque `Void` type is introduced and the trivial structure
optimization updated to account for it. Furthermore, arguments of type
`Void s` are removed from the argument lists of the C functions.
- `ST` is redefined as `Void s -> ST.Out s a` where `ST.Out` is a pair
of `Void s` and `a`
This together has the following major effects on our generated code:
- Functions that return `BaseIO`/`ST`/`EIO`/`IO`/`EST` now do not take
the dummy world parameter anymore. To account for this FFI code needs to
delete the dummy world parameter from the argument lists.
- Functions that return `BaseIO`/`ST` now return their wrapped value
directly. In particular `BaseIO UInt32` now returns a `uint32_t` instead
of a `lean_object*`. To account for this FFI code might have to change
the return type and does not need to call `lean_io_result_mk_ok` anymore
but can instead just `return` values right away (same with extracting
values from `BaseIO` computations.
- Functions that return `EIO`/`IO`/`EST` now only return the equivalent
of an `Except` node which reduces the allocation size. The
`lean_io_result_mk_ok`/`lean_io_result_mk_error` functions were updated
to account for this already so no change is required.
Besides improving performance by dropping allocation (sizes) we can now
also do fun new things such as:
```lean
@[extern "malloc"]
opaque malloc (size : USize) : BaseIO USize
```
This PR reduces the amount of symbols in our DLLs by cutting open a
linking cycle of the shape:
`Environment -> Compiler -> Meta -> Environment`
This is achieved by introducing a dynamic call to the compiler hidden
behind a `Ref` as previously
done in the pretty printer.
This PR "monomorphizes" the structure `Std.PRange shape α`, replacing it
with nine distinct structures `Std.Rcc`, `Std.Rco`, `Std.Rci` etc., one
for each possible shape of a range's bounds. This change was necessary
because the shape polymorphism is detrimental to attempts of automation.
**BREAKING CHANGE:** While range/slice notation itself is unchanged,
this essentially breaks the entire remaining (polymorphic) range and
slice API except for the dot-notation(`toList`, `iter`, ...). It is not
possible to deprecate old declarations that were formulated in a
shape-polymorphic way that is not available anymore.
This PR ensures that even if a type is marked as `irreducible` the
compiler can see through it in
order to discover functions hidden behind type aliases.
This PR adds the necessary infrastructure for recording elaboration
dependencies that may not be apparent from the resulting environment
such as notations and other metaprograms. An adapted version of `shake`
from Mathlib is added to `script/` but may be moved to another location
or repo in the future.
This PR resolves a potential bad interaction between the compiler and
the module system where references to declarations not imported are
brought into scope by inlining or specializing. We now proactively check
that declarations to be inlined/specialized only reference public
imports. The intention is to later resolve this limitation by moving out
compilation into a separate build step with its own import/incremental
system.
This PR fixes constant folding for UIntX in the code generator. This
optimization was previously simply dead code due to the way that uint
literals are encoded.
This PR fixes a potential miscompilation when using non-exposed type
definitions using the module system by turning it into a static error. A
future revision may lift the restriction by making the compiler metadata
independent of the current module.
This PR refines and clarifies the `meta` phase distinction in the module
system.
* `meta import A` without `public` now has the clarified meaning of
"enable compile-time evaluation of declarations in or above `A` in the
current module, but not downstream". This is now checked statically by
enforcing that public meta defs, which therefore may be referenced from
outside, can only use public meta imports, and that global evaluating
attributes such as `@[term_parser]` can only be applied to public meta
defs.
* `meta def`s may no longer reference non-meta defs even when in the
same module. This clarifies the meta distinction as well as improves
locality of (new) error messages.
* parser references in `syntax` are now also properly tracked as meta
references.
* A `meta import` of an `import` now properly loads only the `.ir` of
the nested module for the purposes of execution instead of also making
its declarations available for general elaboration.
* `initialize` is now no longer being run on import under the module
system, which is now covered by `meta initialize`.
This PR changes the defeq algorithm to perform `whnf` on the `String.mk`
expression it creates for string literals.
This is currently a no-op, but will no longer be one once `String` is
redefined so that `String.mk` is a regular function instead of a
constructor.
This PR makes `IO.RealWorld` opaque. It also adds a new compiler -only
`lcRealWorld` constant to represent this type within the compiler. By
default, an opaque type definition is treated like `lcAny`, whereas we
want a more efficient representation. At the moment, this isn't a big
difference, but in the future we would like to completely erase
`IO.RealWorld` at runtime.
This PR fixes the compilation of `noConfusion` by repairing an oversight
made when porting this code from the old compiler. The old compiler only
repeatedly expanded the major for each non-`Prop` field of the inductive
under consideration, mirroring the construction of `noConfusion` itself,
whereas the new compiler erroneously counted all fields.
Fixes#9971.
This PR changes the `toMono` pass to replace decls with their `_redArg`
equivalent, which has the consequence of not considering arguments
deemed useless by the `reduceArity` pass for the purposes of the
`noncomputable` check.
This PR adds support for correctly handling computations on fields in
`casesOn` for inductive predicates that support large elimination. In
any such predicate, the only relevant fields allowed are those that are
also used as an index, in which case we can find the supplied index and
use that term instead.
This PR changes `internalizeCode` to replace all substitutions with
non-param-bound fvars in `Expr`s (which are all types) with `lcAny`,
preserving the invariant that there are no such dependencies. The
violation of this invariant across files caused test failures in a
pending PR, but it is difficult to write a direct test for it. In the
future, we should probably change the LCNF checker to detect this.
This change also speeds up some compilation-heavy benchmarks much more
than I would've expected, which is a pleasant surprise. This indicates
we might get more speedups from reducing the amount of type information
we preserve in LCNF.
This PR addresses a missing check in the module system where private
names that remain in the public environment map for technical reasons
(e.g. inductive constructors generated by the kernel and relied on by
the code generator) accidentally were accessible in the public scope.
This PR removes an error which implicitly assumes that the sort of type
dependency between erased types present in the test being added can not
occur. It would be difficult to refine the error using only the
information present in LCNF types, and it is of very little ongoing
value (I don't recall it ever finding an actual problem), so it makes
more sense to delete it.
Fixes#9692.
This PR changes the LCNF `elimDeadBranches` pass so that it considers
all non-`Nat` literal types to be `⊤`. It turns out that fixing this to
correctly handle all of these types with the current abstract value
representation is surprisingly nontrivial, and it's better to just land
the fix first.
The `isUnderBinder` check is intended to avoid inlining repeated
computations into specializations, but this doesn’t apply to local
function decls whose bodies are already delayed.
This PR consolidates common attribute-related error messages into
reusable functions and updates the wording and formatting of relevant
error messages.
