architeuthis/nac3
1
0
Fork 0
forked from M-Labs/nac3
Code Pull Requests Activity

core/ndstrides: implement np_array()

Browse Source 
It also checks for inconsistent dimensions if the input is a list.
e.g., rejecting `[[1.0, 2.0], [3.0]]`.

However, currently only `np_array(<input>, copy=False)` and `np_array(<input>, copy=True)` are supported. In NumPy, copy could be false, true, or None. Right now, NAC3's `np_array(<input>, copy=False)` behaves like NumPy's `np.array(<input>, copy=None)`.
This commit is contained in:
lyken 2024-08-20 15:10:39 +08:00
parent 13715dbda9
commit d222236492
No known key found for this signature in database
GPG Key ID: 3BD5FC6AC8325DD8
8 changed files with 400 additions and 46 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
nac3core/irrt/irrt.cpp
View File

@ -9,3 +9,4 @@
#include "irrt/ndarray/def.hpp"
#include "irrt/ndarray/iter.hpp"
#include "irrt/ndarray/indexing.hpp"
#include "irrt/ndarray/array.hpp"

134
nac3core/irrt/irrt/ndarray/array.hpp Normal file
View File

@ -0,0 +1,134 @@
#pragma once
#include "irrt/debug.hpp"
#include "irrt/exception.hpp"
#include "irrt/int_types.hpp"
#include "irrt/list.hpp"
#include "irrt/ndarray/basic.hpp"
#include "irrt/ndarray/def.hpp"
namespace {
namespace ndarray {
namespace array {
/**
* @brief In the context of `np.array(<list>)`, deduce the ndarray's shape produced by `<list>` and raise
* an exception if there is anything wrong with `<shape>` (e.g., inconsistent dimensions `np.array([[1.0, 2.0],
* [3.0]])`)
*
* If this function finds no issues with `<list>`, the deduced shape is written to `shape`. The caller has the
* responsibility to allocate `[SizeT; ndims]` for `shape`. The caller must also initialize `shape` with `-1`s because
* of implementation details.
*/
template<typename SizeT>
void set_and_validate_list_shape_helper(SizeT axis, List<SizeT>* list, SizeT ndims, SizeT* shape) {
if (shape[axis] == -1) {
// Dimension is unspecified. Set it.
shape[axis] = list->len;
} else {
// Dimension is specified. Check.
if (shape[axis] != list->len) {
// Mismatch, throw an error.
// NOTE: NumPy's error message is more complex and needs more PARAMS to display.
raise_exception(SizeT, EXN_VALUE_ERROR,
"The requested array has an inhomogenous shape "
"after {0} dimension(s).",
axis, shape[axis], list->len);
}
}
if (axis + 1 == ndims) {
// `list` has type `list[ItemType]`
// Do nothing
} else {
// `list` has type `list[list[...]]`
List<SizeT>** lists = (List<SizeT>**)(list->items);
for (SizeT i = 0; i < list->len; i++) {
set_and_validate_list_shape_helper<SizeT>(axis + 1, lists[i], ndims, shape);
}
}
}
/**
* @brief See `set_and_validate_list_shape_helper`.
*/
template<typename SizeT>
void set_and_validate_list_shape(List<SizeT>* list, SizeT ndims, SizeT* shape) {
for (SizeT axis = 0; axis < ndims; axis++) {
shape[axis] = -1; // Sentinel to say this dimension is unspecified.
}
set_and_validate_list_shape_helper<SizeT>(0, list, ndims, shape);
}
/**
* @brief In the context of `np.array(<list>)`, copied the contents stored in `list` to `ndarray`.
*
* `list` is assumed to be "legal". (i.e., no inconsistent dimensions)
*
* # Notes on `ndarray`
* The caller is responsible for allocating space for `ndarray`.
* Here is what this function expects from `ndarray` when called:
* - `ndarray->data` has to be allocated, contiguous, and may contain uninitialized values.
* - `ndarray->itemsize` has to be initialized.
* - `ndarray->ndims` has to be initialized.
* - `ndarray->shape` has to be initialized.
* - `ndarray->strides` is ignored, but note that `ndarray->data` is contiguous.
* When this function call ends:
* - `ndarray->data` is written with contents from `<list>`.
*/
template<typename SizeT>
void write_list_to_array_helper(SizeT axis, SizeT* index, List<SizeT>* list, NDArray<SizeT>* ndarray) {
debug_assert_eq(SizeT, list->len, ndarray->shape[axis]);
if (IRRT_DEBUG_ASSERT_BOOL) {
if (!ndarray::basic::is_c_contiguous(ndarray)) {
raise_debug_assert(SizeT, "ndarray is not C-contiguous", ndarray->strides[0], ndarray->strides[1],
NO_PARAM);
}
}
if (axis + 1 == ndarray->ndims) {
// `list` has type `list[scalar]`
// `ndarray` is contiguous, so we can do this, and this is fast.
uint8_t* dst = ndarray->data + (ndarray->itemsize * (*index));
__builtin_memcpy(dst, list->items, ndarray->itemsize * list->len);
*index += list->len;
} else {
// `list` has type `list[list[...]]`
List<SizeT>** lists = (List<SizeT>**)(list->items);
for (SizeT i = 0; i < list->len; i++) {
write_list_to_array_helper<SizeT>(axis + 1, index, lists[i], ndarray);
}
}
}
/**
* @brief See `write_list_to_array_helper`.
*/
template<typename SizeT>
void write_list_to_array(List<SizeT>* list, NDArray<SizeT>* ndarray) {
SizeT index = 0;
write_list_to_array_helper<SizeT>((SizeT)0, &index, list, ndarray);
}
} // namespace array
} // namespace ndarray
} // namespace
extern "C" {
using namespace ndarray::array;
void __nac3_ndarray_array_set_and_validate_list_shape(List<int32_t>* list, int32_t ndims, int32_t* shape) {
set_and_validate_list_shape(list, ndims, shape);
}
void __nac3_ndarray_array_set_and_validate_list_shape64(List<int64_t>* list, int64_t ndims, int64_t* shape) {
set_and_validate_list_shape(list, ndims, shape);
}
void __nac3_ndarray_array_write_list_to_array(List<int32_t>* list, NDArray<int32_t>* ndarray) {
write_list_to_array(list, ndarray);
}
void __nac3_ndarray_array_write_list_to_array64(List<int64_t>* list, NDArray<int64_t>* ndarray) {
write_list_to_array(list, ndarray);
}
}

34
nac3core/src/codegen/irrt/mod.rs
View File

@ -8,7 +8,10 @@ use super::{
llvm_intrinsics,
macros::codegen_unreachable,
model::*,
object::ndarray::{indexing::NDIndex, nditer::NDIter, NDArray},
object::{
list::List,
ndarray::{indexing::NDIndex, nditer::NDIter, NDArray},
},
stmt::gen_for_callback_incrementing,
CodeGenContext, CodeGenerator,
};
@ -1129,3 +1132,32 @@ pub fn call_nac3_ndarray_index<'ctx, G: CodeGenerator + ?Sized>(
.arg(dst_ndarray)
.returning_void();
}
pub fn call_nac3_ndarray_array_set_and_validate_list_shape<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
list: Instance<'ctx, Ptr<Struct<List<Int<Byte>>>>>,
ndims: Instance<'ctx, Int<SizeT>>,
shape: Instance<'ctx, Ptr<Int<SizeT>>>,
) {
let name = get_sizet_dependent_function_name(
generator,
ctx,
"__nac3_ndarray_array_set_and_validate_list_shape",
);
FnCall::builder(generator, ctx, &name).arg(list).arg(ndims).arg(shape).returning_void();
}
pub fn call_nac3_ndarray_array_write_list_to_array<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
list: Instance<'ctx, Ptr<Struct<List<Int<Byte>>>>>,
ndarray: Instance<'ctx, Ptr<Struct<NDArray>>>,
) {
let name = get_sizet_dependent_function_name(
generator,
ctx,
"__nac3_ndarray_array_write_list_to_array",
);
FnCall::builder(generator, ctx, &name).arg(list).arg(ndarray).returning_void();
}

9
nac3core/src/codegen/model/ptr.rs
View File

@ -182,6 +182,15 @@ impl<'ctx, Item: Model<'ctx>> Instance<'ctx, Ptr<Item>> {
Ptr(new_item).pointer_cast(generator, ctx, self.value)
}
/// Cast this pointer to `uint8_t*`
pub fn cast_to_pi8<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &CodeGenContext<'ctx, '_>,
) -> Instance<'ctx, Ptr<Int<Byte>>> {
Ptr(Int(Byte)).pointer_cast(generator, ctx, self.value)
}
/// Check if the pointer is null with [`inkwell::builder::Builder::build_is_null`].
pub fn is_null(&self, ctx: &CodeGenContext<'ctx, '_>) -> Instance<'ctx, Int<Bool>> {
let value = ctx.builder.build_is_null(self.value, "").unwrap();

57
nac3core/src/codegen/numpy.rs
View File

@ -13,6 +13,7 @@ use crate::{
},
llvm_intrinsics::{self, call_memcpy_generic},
macros::codegen_unreachable,
model::*,
object::{
any::AnyObject,
ndarray::{shape_util::parse_numpy_int_sequence, NDArrayObject},
@ -22,13 +23,13 @@ use crate::{
},
symbol_resolver::ValueEnum,
toplevel::{
helper::{extract_ndims, PrimDef},
helper::extract_ndims,
numpy::{make_ndarray_ty, unpack_ndarray_var_tys},
DefinitionId,
},
typecheck::{
magic_methods::Binop,
typedef::{FunSignature, Type, TypeEnum},
typedef::{FunSignature, Type},
},
};
use inkwell::{
@ -1840,26 +1841,6 @@ pub fn gen_ndarray_array<'ctx>(
assert!(matches!(args.len(), 1..=3));
let obj_ty = fun.0.args[0].ty;
let obj_elem_ty = match &*context.unifier.get_ty(obj_ty) {
TypeEnum::TObj { obj_id, .. } if *obj_id == PrimDef::NDArray.id() => {
unpack_ndarray_var_tys(&mut context.unifier, obj_ty).0
}
TypeEnum::TObj { obj_id, params, .. } if *obj_id == PrimDef::List.id() => {
let mut ty = *params.iter().next().unwrap().1;
while let TypeEnum::TObj { obj_id, params, .. } = &*context.unifier.get_ty_immutable(ty)
{
if *obj_id != PrimDef::List.id() {
break;
}
ty = *params.iter().next().unwrap().1;
}
ty
}
_ => obj_ty,
};
let obj_arg = args[0].1.clone().to_basic_value_enum(context, generator, obj_ty)?;
let copy_arg = if let Some(arg) =
@ -1875,28 +1856,18 @@ pub fn gen_ndarray_array<'ctx>(
)
};
let ndmin_arg = if let Some(arg) =
args.iter().find(|arg| arg.0.is_some_and(|name| name == fun.0.args[2].name))
{
let ndmin_ty = fun.0.args[2].ty;
arg.1.clone().to_basic_value_enum(context, generator, ndmin_ty)?
} else {
context.gen_symbol_val(
generator,
fun.0.args[2].default_value.as_ref().unwrap(),
fun.0.args[2].ty,
)
};
// The ndmin argument is ignored. We can simply force the ndarray's number of dimensions to be
// the `ndims` of the function return type.
let (_, ndims) = unpack_ndarray_var_tys(&mut context.unifier, fun.0.ret);
let ndims = extract_ndims(&context.unifier, ndims);
call_ndarray_array_impl(
generator,
context,
obj_elem_ty,
obj_arg,
copy_arg.into_int_value(),
ndmin_arg.into_int_value(),
)
.map(NDArrayValue::into)
let object = AnyObject { value: obj_arg, ty: obj_ty };
// NAC3 booleans are i8.
let copy = Int(Bool).truncate(generator, context, copy_arg.into_int_value());
let ndarray = NDArrayObject::make_np_array(generator, context, object, copy)
.atleast_nd(generator, context, ndims);
Ok(ndarray.instance.value)
}
/// Generates LLVM IR for `ndarray.eye`.

11
nac3core/src/codegen/object/list.rs
View File

@ -31,6 +31,17 @@ impl<'ctx, Item: Model<'ctx>> StructKind<'ctx> for List<Item> {
}
}
impl<'ctx, Item: Model<'ctx>> Instance<'ctx, Ptr<Struct<List<Item>>>> {
/// Cast the items pointer to `uint8_t*`.
pub fn with_pi8_items<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
) -> Instance<'ctx, Ptr<Struct<List<Int<Byte>>>>> {
self.pointer_cast(generator, ctx, Struct(List { item: Int(Byte) }))
}
}
/// A NAC3 Python List object.
#[derive(Debug, Clone, Copy)]
pub struct ListObject<'ctx> {

184
nac3core/src/codegen/object/ndarray/array.rs Normal file
View File

@ -0,0 +1,184 @@
use super::NDArrayObject;
use crate::{
codegen::{
irrt::{
call_nac3_ndarray_array_set_and_validate_list_shape,
call_nac3_ndarray_array_write_list_to_array,
},
model::*,
object::{any::AnyObject, list::ListObject},
stmt::gen_if_else_expr_callback,
CodeGenContext, CodeGenerator,
},
toplevel::helper::{arraylike_flatten_element_type, arraylike_get_ndims},
typecheck::typedef::{Type, TypeEnum},
};
/// Get the expected `dtype` and `ndims` of the ndarray returned by `np_array(list)`.
fn get_list_object_dtype_and_ndims<'ctx>(
ctx: &mut CodeGenContext<'ctx, '_>,
list: ListObject<'ctx>,
) -> (Type, u64) {
let dtype = arraylike_flatten_element_type(&mut ctx.unifier, list.item_type);
let ndims = arraylike_get_ndims(&mut ctx.unifier, list.item_type);
let ndims = ndims + 1; // To count `list` itself.
(dtype, ndims)
}
impl<'ctx> NDArrayObject<'ctx> {
/// Implementation of `np_array(<list>, copy=True)`
fn make_np_array_list_copy_true_impl<G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
list: ListObject<'ctx>,
) -> Self {
let (dtype, ndims_int) = get_list_object_dtype_and_ndims(ctx, list);
let list_value = list.instance.with_pi8_items(generator, ctx);
// Validate `list` has a consistent shape.
// Raise an exception if `list` is something abnormal like `[[1, 2], [3]]`.
// If `list` has a consistent shape, deduce the shape and write it to `shape`.
let ndims = Int(SizeT).const_int(generator, ctx.ctx, ndims_int, false);
let shape = Int(SizeT).array_alloca(generator, ctx, ndims.value);
call_nac3_ndarray_array_set_and_validate_list_shape(
generator, ctx, list_value, ndims, shape,
);
let ndarray = NDArrayObject::alloca(generator, ctx, dtype, ndims_int);
ndarray.copy_shape_from_array(generator, ctx, shape);
ndarray.create_data(generator, ctx);
// Copy all contents from the list.
call_nac3_ndarray_array_write_list_to_array(generator, ctx, list_value, ndarray.instance);
ndarray
}
/// Implementation of `np_array(<list>, copy=None)`
fn make_np_array_list_copy_none_impl<G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
list: ListObject<'ctx>,
) -> Self {
// np_array without copying is only possible `list` is not nested.
//
// If `list` is `list[T]`, we can create an ndarray with `data` set
// to the array pointer of `list`.
//
// If `list` is `list[list[T]]` or worse, copy.
let (dtype, ndims) = get_list_object_dtype_and_ndims(ctx, list);
if ndims == 1 {
// `list` is not nested
let ndarray = NDArrayObject::alloca(generator, ctx, dtype, 1);
// Set data
let data = list.instance.get(generator, ctx, |f| f.items).cast_to_pi8(generator, ctx);
ndarray.instance.set(ctx, |f| f.data, data);
// ndarray->shape[0] = list->len;
let shape = ndarray.instance.get(generator, ctx, |f| f.shape);
let list_len = list.instance.get(generator, ctx, |f| f.len);
shape.set_index_const(ctx, 0, list_len);
// Set strides, the `data` is contiguous
ndarray.set_strides_contiguous(generator, ctx);
ndarray
} else {
// `list` is nested, copy
NDArrayObject::make_np_array_list_copy_true_impl(generator, ctx, list)
}
}
/// Implementation of `np_array(<list>, copy=copy)`
fn make_np_array_list_impl<G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
list: ListObject<'ctx>,
copy: Instance<'ctx, Int<Bool>>,
) -> Self {
let (dtype, ndims) = get_list_object_dtype_and_ndims(ctx, list);
let ndarray = gen_if_else_expr_callback(
generator,
ctx,
|_generator, _ctx| Ok(copy.value),
|generator, ctx| {
let ndarray =
NDArrayObject::make_np_array_list_copy_true_impl(generator, ctx, list);
Ok(Some(ndarray.instance.value))
},
|generator, ctx| {
let ndarray =
NDArrayObject::make_np_array_list_copy_none_impl(generator, ctx, list);
Ok(Some(ndarray.instance.value))
},
)
.unwrap()
.unwrap();
NDArrayObject::from_value_and_unpacked_types(generator, ctx, ndarray, dtype, ndims)
}
/// Implementation of `np_array(<ndarray>, copy=copy)`.
pub fn make_np_array_ndarray_impl<G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
ndarray: NDArrayObject<'ctx>,
copy: Instance<'ctx, Int<Bool>>,
) -> Self {
let ndarray_val = gen_if_else_expr_callback(
generator,
ctx,
|_generator, _ctx| Ok(copy.value),
|generator, ctx| {
let ndarray = ndarray.make_copy(generator, ctx); // Force copy
Ok(Some(ndarray.instance.value))
},
|_generator, _ctx| {
// No need to copy. Return `ndarray` itself.
Ok(Some(ndarray.instance.value))
},
)
.unwrap()
.unwrap();
NDArrayObject::from_value_and_unpacked_types(
generator,
ctx,
ndarray_val,
ndarray.dtype,
ndarray.ndims,
)
}
/// Create a new ndarray like `np.array()`.
///
/// NOTE: The `ndmin` argument is not here. You may want to
/// do [`NDArrayObject::atleast_nd`] to achieve that.
pub fn make_np_array<G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
object: AnyObject<'ctx>,
copy: Instance<'ctx, Int<Bool>>,
) -> Self {
match &*ctx.unifier.get_ty(object.ty) {
TypeEnum::TObj { obj_id, .. }
if *obj_id == ctx.primitives.list.obj_id(&ctx.unifier).unwrap() =>
{
let list = ListObject::from_object(generator, ctx, object);
NDArrayObject::make_np_array_list_impl(generator, ctx, list, copy)
}
TypeEnum::TObj { obj_id, .. }
if *obj_id == ctx.primitives.ndarray.obj_id(&ctx.unifier).unwrap() =>
{
let ndarray = NDArrayObject::from_object(generator, ctx, object);
NDArrayObject::make_np_array_ndarray_impl(generator, ctx, ndarray, copy)
}
_ => panic!("Unrecognized object type: {}", ctx.unifier.stringify(object.ty)), // Typechecker ensures this
}
}
}

14
nac3core/src/codegen/object/ndarray/mod.rs
View File

@ -1,3 +1,4 @@
pub mod array;
pub mod factory;
pub mod indexing;
pub mod nditer;
@ -74,8 +75,19 @@ impl<'ctx> NDArrayObject<'ctx> {
) -> NDArrayObject<'ctx> {
let (dtype, ndims) = unpack_ndarray_var_tys(&mut ctx.unifier, object.ty);
let ndims = extract_ndims(&ctx.unifier, ndims);
Self::from_value_and_unpacked_types(generator, ctx, object.value, dtype, ndims)
}
let value = Ptr(Struct(NDArray)).check_value(generator, ctx.ctx, object.value).unwrap();
/// Like [`NDArrayObject::from_object`] but you directly supply the ndarray's
/// `dtype` and `ndims`.
pub fn from_value_and_unpacked_types<V: BasicValue<'ctx>, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
value: V,
dtype: Type,
ndims: u64,
) -> Self {
let value = Ptr(Struct(NDArray)).check_value(generator, ctx.ctx, value).unwrap();
NDArrayObject { dtype, ndims, instance: value }
}