| Age | Commit message (Collapse) | Author |
|---|
|
|
|
No need to store the case_info, all the data is reconstructible from the
context. Furthermore, this reconstruction is performed in a context where
we already access the environment, so performance is not at stake.
Hopefully this will also reduce the number of globally allocated VM values,
since the switch representation now only depends on the shape of the inductive
type.
|
|
Persistent arrays expose a functional interface but are implemented
using an imperative data structure. The OCaml implementation is based on
Jean-Christophe Filliâtre's.
Co-authored-by: Benjamin Grégoire <Benjamin.Gregoire@inria.fr>
Co-authored-by: Gaëtan Gilbert <gaetan.gilbert@skyskimmer.net>
|
|
Add headers to a few files which were missing them.
|
|
We also remove trailing whitespace.
Script used:
```bash
for i in `find . -name '*.ml' -or -name '*.mli' -or -name '*.mlg'`; do expand -i "$i" | sponge "$i"; sed -e's/[[:space:]]*$//' -i.bak "$i"; done
```
|
|
* Add a related test-suite in compare.v (generated by a bash script)
Co-authored-by: Pierre Roux <pierre.roux@onera.fr>
|
|
|
|
|
|
* This commit add float instructions to the VM, their encoding in bytecode
and the interpretation of primitive float values after the reduction.
* The flag '-std=c99' could be added to the C compiler flags to ensure
that float computation strictly follows the norm (ie. i387 80-bits
format is not used as an optimization).
Actually, we use '-fexcess-precision=standard' instead of '-std=c99'
because the latter would disable GNU asm used in the VM.
|
|
Beware of 0. = -0. issue for primitive floats
The IEEE 754 declares that 0. and -0. are treated equal but we cannot
say that this is true with Leibniz equality.
Therefore we must patch the equality and the total comparison inside the
kernel to prevent inconsistency.
|
|
|
|
This work makes it possible to take advantage of a compact
representation for integers in the entire system, as opposed to only
in some reduction machines. It is useful for heavily computational
applications, where even constructing terms is not possible without such
a representation.
Concretely, it replaces part of the retroknowledge machinery with
a primitive construction for integers in terms, and introduces a kind of
FFI which maps constants to operators (on integers). Properties of these
operators are expressed as explicit axioms, whereas they were hidden in
the retroknowledge-based approach.
This has been presented at the Coq workshop and some Coq Working Groups,
and has been used by various groups for STM trace checking,
computational analysis, etc.
Contributions by Guillaume Bertholon and Pierre Roux <Pierre.Roux@onera.fr>
Co-authored-by: Benjamin Grégoire <Benjamin.Gregoire@inria.fr>
Co-authored-by: Vincent Laporte <Vincent.Laporte@fondation-inria.fr>
|
|
|
|
|
|
|
|
The upper layers still need a mapping constant -> projection, which is
provided by Recordops.
|
|
It was actually a hack since those names are never used to represent
values, only to be passed as arguments to bytecode instructions. So
instead of reusing the structured_constant type, we follow the same
pattern as switch annotations.
|
|
We remove most of what was deprecated in `Term`. Now, `intf` and
`kernel` are almost deprecation-free, tho I am not very convinced
about the whole `Term -> Constr` renaming but I'm afraid there is no
way back.
Inconsistencies with the constructor policy (see #6440) remain along
the code-base and I'm afraid I don't see a plan to reconcile them.
The `Sorts` deprecation is hard to finalize, opening `Sorts` is not a
good idea as someone added a `List` module inside it.
|
|
This prevents the existence of a few naked pointers to C heap from the OCaml
heap. VM bytecode is represented as any block of size at least 1 whose first
field points to a C-allocated string.
This representation is compatible with the Coq VM representation of
(potentially recursive) closures, which are already specifically tailored
in the OCaml GC to be able to contain out-of-heap data.
|
|
|
|
This simplifies the representation of values, and brings it closer to
the ones of the native compiler.
|
|
|
|
Instead of using a linear representation, we simply use a table that maps
every kind of relocation to the list of positions it needs to be applied to.
|
|
The previous implementation used a list of pairs, which has size 9n where n
is the number of relocations. We instead use two arrays for a total memory
cost of 2n + 5 words.
The use of arrays may turn out to be problematic on 32-bit machines, I am unsure
if we will hit this limitation in practice.
|
|
Instead we thread a buffer.
|
|
This is actually not needed, as the only thing we do with this Bytes.t is to
pass it to a C function which will use it in a read-only way.
|
|
This shouldn't matter because the tcode_of_code function is pure, its only
effect being allocating a string and filling it with the translated bytecode.
|
|
This reduces the possibility to wreak havoc while making the API nicer.
|
|
This will allow to merge back `Names` with `API.Names`
|
|
|
|
|
|
(ProcThreeStInl.v, when the final "Defined" runs). I've verified that the
change here fixes the stack overflow there with Coq 8.5pl2.
In this version, all the recursive calls are in tail position. Instead of
taking a single list of instructions, `emit' here takes a curent list and a remaining
list of lists of instructions. That means the two calls elsewhere in the file now
add an empty list argument. The algorithm works on the current list until it's empty,
then works on the remaining lists. The most complex case is for Ksequence, where
one of the pieces becomes the new current list, and the other pieces are consed onto
the remaining sub-lists.
|
|
The `a.[i] <- x` notation is deprecated and we were getting a couple
of warnings.
|
|
The `emitcodes` string type was used in both a functional and an
imperative way, so we have to handle it with care in order to preserve
the previous optimizations and semantics.
|
|
|
|
The bytecode interpreter ensures that the stack space available at some
points is above a static threshold. However, arbitrary large stack space
can be needed between two check points, leading to segmentation faults
in some cases.
We track the use of stack space at compilation time and add
an instruction to ensure greater stack capacity when required. This is
inspired from OCaml's PR#339 and PR#7168.
Patch written with Benjamin Grégoire.
|
|
|
|
|
|
|
|
|
|
- Universes are now represented in the VM by a structured constant containing the
global levels. This constant is applied to local level variables if any.
- When reading back a universe, we perform the union of these levels and return
a [Vsort].
- Fixed a bug: structured constants could contain local universe variables in
constructor arguments, which has to be prevented.
Was showing up for instance when evaluating [cons _ list (nil _)] with
a polymorphic [list] type.
- Fixed a bug: polymorphic inductive types can have an empty stack.
Was showing up when evaluating [bool] with a polymorphic [bool] type.
- Made a few cosmetic changes.
Patch written with Benjamin Grégoire.
|
|
polymorphic definitions.
- This implementation passes universes in separate arguments and does not
eagerly instanitate polymorphic definitions.
- This means that it pays no cost on monomorphic definitions.
|
|
|
|
|
|
|
|
I was not seeing the warning due to the 10 deprecation warnings before
it...
|
|
I'm pushing this patch now because the previous treatment of such projections
in the VM was already unsound. It should however be carefully reviewed.
|
|
|
|
and 8388851 non-constant constructor.
|
|
Module substitution is made nodewise, allowing to drop it for opaque terms
in the VM. This saves a few useless substitutions.
|
