Compare commits

...

24 Commits

Author SHA1 Message Date
1531b6cc98 cargo: update dependencies 2024-12-13 19:42:01 +08:00
9bbc40bbfa flake: update dependencies 2024-12-13 19:41:52 +08:00
790e56d106 msys2: update 2024-12-13 19:39:39 +08:00
a00eb7969e [core] codegen: Implement matrix_power
Last of the functions that need to be ported over to strided-ndarray.
2024-12-13 15:23:31 +08:00
27a6f47330 [core] codegen: Implement construction of unsized ndarrays
Partially based on f731e604: core/ndstrides: add more ScalarOrNDArray
and NDArrayObject utils.
2024-12-13 15:23:31 +08:00
061747c67b [core] codegen: Implement NDArrayValue::atleast_nd
Based on 9cfa2622: core/ndstrides: add NDArrayObject::atleast_nd.
2024-12-13 15:23:31 +08:00
dc91d9e35a [core] codegen: Implement ScalarOrNDArray and use it in indexing
Based on 8f9d2d82: core/ndstrides: implement ndarray indexing.
2024-12-13 15:23:31 +08:00
438943ac6f [core] codegen: Implement indexing for NDArray
Based on 8f9d2d82: core/ndstrides: implement ndarray indexing

The functionality for `...` and `np.newaxis` is there in IRRT, but there
is no implementation of them for @kernel Python expressions because of
M-Labs/nac3#486.
2024-12-13 15:23:31 +08:00
678e56c95d [core] irrt: rename NDIndex to NDIndexInt
Unfortunately the name `NDIndex` is used in later commits. Renaming this
typedef to `NDIndexInt` to avoid amending. `NDIndexInt` will be removed
anyway when ndarray strides is completed.
2024-12-13 15:23:31 +08:00
fdfc80ca5f [core] codegen: Implement Slice{Type,Value}, RustSlice
Based on 01c96396: core/irrt: add Slice and Range and part of
8f9d2d82: core/ndstrides: implement ndarray indexing.

Needed for implementing general ndarray indexing.

Currently IRRT slice and range have nothing to do with NAC3's slice
and range. The IRRT slice and range are currently there to implement
ndarray specific features. However, in the future their definitions may
be used to replace that of NAC3's. (NAC3's range is a [i32 x 3], IRRT's
range is a proper struct. NAC3 does not have a slice struct).
2024-12-13 15:23:31 +08:00
8b3429d62a [artiq] Reimplement get_obj_value for strided ndarray
Based on 7ef93472: artiq: reimplement get_obj_value to use ndarray with
strides
2024-12-13 15:23:31 +08:00
f4c5038b95 [artiq] codegen: Reimplement polymorphic_print for strided ndarray
Based on 2a6ee503: artiq: reimplement polymorphic_print for ndarray
2024-12-13 15:23:31 +08:00
ddd16738a6 [core] codegen: implement ndarray iterator NDIter
Based on 50f960ab: core/ndstrides: implement ndarray iterator NDIter

A necessary utility to iterate through all elements in a possibly
strided ndarray.
2024-12-13 15:23:31 +08:00
44c49dc102 [artiq] codegen: Reimplement polymorphic_print for strided ndarray
Based on 2a6ee503: artiq: reimplement polymorphic_print for ndarray
2024-12-13 15:23:31 +08:00
e4bd376587 [core] codegen: Implement ContiguousNDArray
Fixes compatibility with linalg algorithms. matrix_power is missing due
to the need for indexing support.
2024-12-13 15:23:29 +08:00
44498f22f6 [core] codegen: Implement NDArray functions from a0a1f35b 2024-12-13 15:22:11 +08:00
110416d07a [core] codegen/irrt: Add IRRT functions for strided-ndarray 2024-12-13 15:22:11 +08:00
08a7d01a13 [core] Add itemsize and strides to NDArray struct
Temporarily disable linalg ndarray tests as they are not ported to work
with strided-ndarray.
2024-12-13 15:22:09 +08:00
3cd36fddc3 [core] codegen/types: Add check_struct_type_matches_fields
Shorthand for checking if a type is representable by a StructFields
instance.
2024-12-12 11:40:44 +08:00
56a7a9e03d [core] codegen: Add helper functions for create+call functions
Replacement for various FnCall methods from legacy ndstrides
implementation.
2024-12-12 11:30:36 +08:00
574ae40f97 [core] codegen: Add call_memcpy_generic_array
Replacement for Instance<Ptr>::copy_from from legacy ndstrides
implementation.
2024-12-12 11:30:36 +08:00
aa293b6bea [core] codegen: Add type_aligned_alloca 2024-12-12 11:30:35 +08:00
eb4b881690 [core] Expose {types,values}::ndarray modules
Allows better encapsulation of members in these modules rather than
allowing them to leak into types/values mod.
2024-12-12 11:30:14 +08:00
3d0a1d281c [core] Expose irrt::ndarray 2024-12-10 12:49:49 +08:00
44 changed files with 5234 additions and 1318 deletions

24
Cargo.lock generated
View File

@ -126,9 +126,9 @@ checksum = "1fd0f2584146f6f2ef48085050886acf353beff7305ebd1ae69500e27c67f64b"
[[package]]
name = "cc"
version = "1.2.3"
version = "1.2.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "27f657647bcff5394bf56c7317665bbf790a137a50eaaa5c6bfbb9e27a518f2d"
checksum = "9157bbaa6b165880c27a4293a474c91cdcf265cc68cc829bf10be0964a391caf"
dependencies = [
"shlex",
]
@ -559,9 +559,9 @@ checksum = "bbd2bcb4c963f2ddae06a2efc7e9f3591312473c50c6685e1f298068316e66fe"
[[package]]
name = "libc"
version = "0.2.167"
version = "0.2.168"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "09d6582e104315a817dff97f75133544b2e094ee22447d2acf4a74e189ba06fc"
checksum = "5aaeb2981e0606ca11d79718f8bb01164f1d6ed75080182d3abf017e6d244b6d"
[[package]]
name = "libloading"
@ -1004,9 +1004,9 @@ dependencies = [
[[package]]
name = "redox_syscall"
version = "0.5.7"
version = "0.5.8"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9b6dfecf2c74bce2466cabf93f6664d6998a69eb21e39f4207930065b27b771f"
checksum = "03a862b389f93e68874fbf580b9de08dd02facb9a788ebadaf4a3fd33cf58834"
dependencies = [
"bitflags",
]
@ -1089,24 +1089,24 @@ checksum = "94143f37725109f92c262ed2cf5e59bce7498c01bcc1502d7b9afe439a4e9f49"
[[package]]
name = "semver"
version = "1.0.23"
version = "1.0.24"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "61697e0a1c7e512e84a621326239844a24d8207b4669b41bc18b32ea5cbf988b"
checksum = "3cb6eb87a131f756572d7fb904f6e7b68633f09cca868c5df1c4b8d1a694bbba"
[[package]]
name = "serde"
version = "1.0.215"
version = "1.0.216"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "6513c1ad0b11a9376da888e3e0baa0077f1aed55c17f50e7b2397136129fb88f"
checksum = "0b9781016e935a97e8beecf0c933758c97a5520d32930e460142b4cd80c6338e"
dependencies = [
"serde_derive",
]
[[package]]
name = "serde_derive"
version = "1.0.215"
version = "1.0.216"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ad1e866f866923f252f05c889987993144fb74e722403468a4ebd70c3cd756c0"
checksum = "46f859dbbf73865c6627ed570e78961cd3ac92407a2d117204c49232485da55e"
dependencies = [
"proc-macro2",
"quote",

6
flake.lock generated
View File

@ -2,11 +2,11 @@
"nodes": {
"nixpkgs": {
"locked": {
"lastModified": 1731319897,
"narHash": "sha256-PbABj4tnbWFMfBp6OcUK5iGy1QY+/Z96ZcLpooIbuEI=",
"lastModified": 1733940404,
"narHash": "sha256-Pj39hSoUA86ZePPF/UXiYHHM7hMIkios8TYG29kQT4g=",
"owner": "NixOS",
"repo": "nixpkgs",
"rev": "dc460ec76cbff0e66e269457d7b728432263166c",
"rev": "5d67ea6b4b63378b9c13be21e2ec9d1afc921713",
"type": "github"
},
"original": {

View File

@ -12,16 +12,17 @@ use pyo3::{
PyObject, PyResult, Python,
};
use super::{symbol_resolver::InnerResolver, timeline::TimeFns};
use nac3core::{
codegen::{
expr::{destructure_range, gen_call},
irrt::call_ndarray_calc_size,
llvm_intrinsics::{call_int_smax, call_memcpy_generic, call_stackrestore, call_stacksave},
llvm_intrinsics::{call_int_smax, call_memcpy, call_stackrestore, call_stacksave},
stmt::{gen_block, gen_for_callback_incrementing, gen_if_callback, gen_with},
types::NDArrayType,
type_aligned_alloca,
types::ndarray::NDArrayType,
values::{
ArrayLikeIndexer, ArrayLikeValue, ArraySliceValue, ListValue, NDArrayValue, ProxyValue,
RangeValue, UntypedArrayLikeAccessor,
ArrayLikeIndexer, ArrayLikeValue, ArraySliceValue, ListValue, ProxyValue, RangeValue,
UntypedArrayLikeAccessor,
},
CodeGenContext, CodeGenerator,
},
@ -34,12 +35,14 @@ use nac3core::{
},
nac3parser::ast::{Expr, ExprKind, Located, Stmt, StmtKind, StrRef},
symbol_resolver::ValueEnum,
toplevel::{helper::PrimDef, numpy::unpack_ndarray_var_tys, DefinitionId, GenCall},
toplevel::{
helper::{extract_ndims, PrimDef},
numpy::unpack_ndarray_var_tys,
DefinitionId, GenCall,
},
typecheck::typedef::{iter_type_vars, FunSignature, FuncArg, Type, TypeEnum, VarMap},
};
use super::{symbol_resolver::InnerResolver, timeline::TimeFns};
/// The parallelism mode within a block.
#[derive(Copy, Clone, Eq, PartialEq)]
enum ParallelMode {
@ -458,60 +461,49 @@ fn format_rpc_arg<'ctx>(
let llvm_i1 = ctx.ctx.bool_type();
let llvm_usize = generator.get_size_type(ctx.ctx);
let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, arg_ty);
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
let llvm_arg = NDArrayValue::from_pointer_value(
arg.into_pointer_value(),
llvm_elem_ty,
llvm_usize,
None,
);
let (elem_ty, ndims) = unpack_ndarray_var_tys(&mut ctx.unifier, arg_ty);
let ndims = extract_ndims(&ctx.unifier, ndims);
let dtype = ctx.get_llvm_type(generator, elem_ty);
let ndarray = NDArrayType::new(generator, ctx.ctx, dtype, Some(ndims))
.map_value(arg.into_pointer_value(), None);
let llvm_usize_sizeof = ctx
.builder
.build_int_truncate_or_bit_cast(
llvm_arg.get_type().size_type().size_of(),
llvm_usize,
"",
)
.unwrap();
let llvm_pdata_sizeof = ctx
.builder
.build_int_truncate_or_bit_cast(
llvm_elem_ty.ptr_type(AddressSpace::default()).size_of(),
llvm_usize,
"",
)
.unwrap();
let ndims = llvm_usize.const_int(ndims, false);
let dims_buf_sz =
ctx.builder.build_int_mul(llvm_arg.load_ndims(ctx), llvm_usize_sizeof, "").unwrap();
// `ndarray.data` is possibly not contiguous, and we need it to be contiguous for
// the reader.
// Turning it into a ContiguousNDArray to get a `data` that is contiguous.
let carray = ndarray.make_contiguous_ndarray(generator, ctx);
let buffer_size =
ctx.builder.build_int_add(dims_buf_sz, llvm_pdata_sizeof, "").unwrap();
let sizeof_usize = llvm_usize.size_of();
let sizeof_usize =
ctx.builder.build_int_z_extend_or_bit_cast(sizeof_usize, llvm_usize, "").unwrap();
let buffer = ctx.builder.build_array_alloca(llvm_i8, buffer_size, "rpc.arg").unwrap();
let buffer = ArraySliceValue::from_ptr_val(buffer, buffer_size, Some("rpc.arg"));
let sizeof_pdata = dtype.ptr_type(AddressSpace::default()).size_of();
let sizeof_pdata =
ctx.builder.build_int_z_extend_or_bit_cast(sizeof_pdata, llvm_usize, "").unwrap();
call_memcpy_generic(
ctx,
buffer.base_ptr(ctx, generator),
llvm_arg.ptr_to_data(ctx),
llvm_pdata_sizeof,
llvm_i1.const_zero(),
);
let sizeof_buf_shape = ctx.builder.build_int_mul(sizeof_usize, ndims, "").unwrap();
let sizeof_buf = ctx.builder.build_int_add(sizeof_buf_shape, sizeof_pdata, "").unwrap();
let pbuffer_dims_begin =
unsafe { buffer.ptr_offset_unchecked(ctx, generator, &llvm_pdata_sizeof, None) };
call_memcpy_generic(
ctx,
pbuffer_dims_begin,
llvm_arg.shape().base_ptr(ctx, generator),
dims_buf_sz,
llvm_i1.const_zero(),
);
// buf = { data: void*, shape: [size_t; ndims]; }
let buf = ctx.builder.build_array_alloca(llvm_i8, sizeof_buf, "rpc.arg").unwrap();
let buf = ArraySliceValue::from_ptr_val(buf, sizeof_buf, Some("rpc.arg"));
let buf_data = buf.base_ptr(ctx, generator);
let buf_shape =
unsafe { buf.ptr_offset_unchecked(ctx, generator, &sizeof_pdata, None) };
buffer.base_ptr(ctx, generator)
// Write to `buf->data`
let carray_data = carray.load_data(ctx);
let carray_data = ctx.builder.build_pointer_cast(carray_data, llvm_pi8, "").unwrap();
call_memcpy(ctx, buf_data, carray_data, sizeof_pdata, llvm_i1.const_zero());
// Write to `buf->shape`
let carray_shape = ndarray.shape().base_ptr(ctx, generator);
let carray_shape_i8 =
ctx.builder.build_pointer_cast(carray_shape, llvm_pi8, "").unwrap();
call_memcpy(ctx, buf_shape, carray_shape_i8, sizeof_buf_shape, llvm_i1.const_zero());
buf.base_ptr(ctx, generator)
}
_ => {
@ -552,6 +544,8 @@ fn format_rpc_ret<'ctx>(
let llvm_i32 = ctx.ctx.i32_type();
let llvm_i8_8 = ctx.ctx.struct_type(&[llvm_i8.array_type(8).into()], false);
let llvm_pi8 = llvm_i8.ptr_type(AddressSpace::default());
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
let rpc_recv = ctx.module.get_function("rpc_recv").unwrap_or_else(|| {
ctx.module.add_function("rpc_recv", llvm_i32.fn_type(&[llvm_pi8.into()], false), None)
@ -572,8 +566,7 @@ fn format_rpc_ret<'ctx>(
let result = match &*ctx.unifier.get_ty_immutable(ret_ty) {
TypeEnum::TObj { obj_id, .. } if *obj_id == PrimDef::NDArray.id() => {
let llvm_i1 = ctx.ctx.bool_type();
let llvm_usize = generator.get_size_type(ctx.ctx);
let num_0 = llvm_usize.const_zero();
// Round `val` up to its modulo `power_of_two`
let round_up = |ctx: &mut CodeGenContext<'ctx, '_>,
@ -599,79 +592,49 @@ fn format_rpc_ret<'ctx>(
.unwrap()
};
// Setup types
let (elem_ty, ndims) = unpack_ndarray_var_tys(&mut ctx.unifier, ret_ty);
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
let llvm_ret_ty = NDArrayType::new(generator, ctx.ctx, llvm_elem_ty);
// Allocate the resulting ndarray
// A condition after format_rpc_ret ensures this will not be popped this off.
let ndarray = llvm_ret_ty.alloca(generator, ctx, Some("rpc.result"));
let (dtype, ndims) = unpack_ndarray_var_tys(&mut ctx.unifier, ret_ty);
let dtype_llvm = ctx.get_llvm_type(generator, dtype);
let ndims = extract_ndims(&ctx.unifier, ndims);
let ndarray = NDArrayType::new(generator, ctx.ctx, dtype_llvm, Some(ndims))
.construct_uninitialized(generator, ctx, None);
// Setup ndims
let ndims =
if let TypeEnum::TLiteral { values, .. } = &*ctx.unifier.get_ty_immutable(ndims) {
assert_eq!(values.len(), 1);
// NOTE: Current content of `ndarray`:
// - * `data` - **NOT YET** allocated.
// - * `itemsize` - initialized to be size_of(dtype).
// - * `ndims` - initialized.
// - * `shape` - allocated; has uninitialized values.
// - * `strides` - allocated; has uninitialized values.
u64::try_from(values[0].clone()).unwrap()
} else {
unreachable!();
};
// Set `ndarray.ndims`
ndarray.store_ndims(ctx, generator, llvm_usize.const_int(ndims, false));
// Allocate `ndarray.shape` [size_t; ndims]
ndarray.create_shape(ctx, llvm_usize, ndarray.load_ndims(ctx));
/*
ndarray now:
- .ndims: initialized
- .shape: allocated but uninitialized .shape
- .data: uninitialized
*/
let llvm_usize_sizeof = ctx
.builder
.build_int_truncate_or_bit_cast(llvm_usize.size_of(), llvm_usize, "")
.unwrap();
let llvm_pdata_sizeof = ctx
.builder
.build_int_truncate_or_bit_cast(
llvm_elem_ty.ptr_type(AddressSpace::default()).size_of(),
llvm_usize,
"",
)
.unwrap();
let llvm_elem_sizeof = ctx
.builder
.build_int_truncate_or_bit_cast(llvm_elem_ty.size_of().unwrap(), llvm_usize, "")
.unwrap();
let itemsize = ndarray.load_itemsize(ctx); // Same as doing a `ctx.get_llvm_type` on `dtype` and get its `size_of()`.
// Allocates a buffer for the initial RPC'ed object, which is guaranteed to be
// (4 + 4 * ndims) bytes with 8-byte alignment
let sizeof_dims =
ctx.builder.build_int_mul(ndarray.load_ndims(ctx), llvm_usize_sizeof, "").unwrap();
let sizeof_usize = llvm_usize.size_of();
let sizeof_usize =
ctx.builder.build_int_truncate_or_bit_cast(sizeof_usize, llvm_usize, "").unwrap();
let sizeof_ptr = llvm_i8.ptr_type(AddressSpace::default()).size_of();
let sizeof_ptr =
ctx.builder.build_int_z_extend_or_bit_cast(sizeof_ptr, llvm_usize, "").unwrap();
let sizeof_shape =
ctx.builder.build_int_mul(ndarray.load_ndims(ctx), sizeof_usize, "").unwrap();
// Size of the buffer for the initial `rpc_recv()`.
let unaligned_buffer_size =
ctx.builder.build_int_add(sizeof_dims, llvm_pdata_sizeof, "").unwrap();
let buffer_size = round_up(ctx, unaligned_buffer_size, llvm_usize.const_int(8, false));
ctx.builder.build_int_add(sizeof_ptr, sizeof_shape, "").unwrap();
let stackptr = call_stacksave(ctx, None);
// Just to be absolutely sure, alloca in [i8 x 8] slices to force 8-byte alignment
let buffer = ctx
.builder
.build_array_alloca(
llvm_i8_8,
ctx.builder
.build_int_unsigned_div(buffer_size, llvm_usize.const_int(8, false), "")
.unwrap(),
"rpc.buffer",
)
.unwrap();
let buffer = ctx
.builder
.build_bit_cast(buffer, llvm_pi8, "")
.map(BasicValueEnum::into_pointer_value)
.unwrap();
let buffer = ArraySliceValue::from_ptr_val(buffer, buffer_size, None);
let buffer = type_aligned_alloca(
generator,
ctx,
llvm_i8_8,
unaligned_buffer_size,
Some("rpc.buffer"),
);
let buffer = ArraySliceValue::from_ptr_val(buffer, unaligned_buffer_size, None);
// The first call to `rpc_recv` reads the top-level ndarray object: [pdata, shape]
//
@ -679,7 +642,7 @@ fn format_rpc_ret<'ctx>(
let ndarray_nbytes = ctx
.build_call_or_invoke(
rpc_recv,
&[buffer.base_ptr(ctx, generator).into()], // Reads [usize; ndims]. NOTE: We are allocated [size_t; ndims].
&[buffer.base_ptr(ctx, generator).into()], // Reads [usize; ndims]
"rpc.size.next",
)
.map(BasicValueEnum::into_int_value)
@ -687,16 +650,14 @@ fn format_rpc_ret<'ctx>(
// debug_assert(ndarray_nbytes > 0)
if ctx.registry.llvm_options.opt_level == OptimizationLevel::None {
let cmp = ctx
.builder
.build_int_compare(IntPredicate::UGT, ndarray_nbytes, num_0, "")
.unwrap();
ctx.make_assert(
generator,
ctx.builder
.build_int_compare(
IntPredicate::UGT,
ndarray_nbytes,
ndarray_nbytes.get_type().const_zero(),
"",
)
.unwrap(),
cmp,
"0:AssertionError",
"Unexpected RPC termination for ndarray - Expected data buffer next",
[None, None, None],
@ -705,49 +666,50 @@ fn format_rpc_ret<'ctx>(
}
// Copy shape from the buffer to `ndarray.shape`.
let pbuffer_dims =
unsafe { buffer.ptr_offset_unchecked(ctx, generator, &llvm_pdata_sizeof, None) };
// We need to skip the first `sizeof(uint8_t*)` bytes to skip the `pdata` in `[pdata, shape]`.
let pbuffer_shape =
unsafe { buffer.ptr_offset_unchecked(ctx, generator, &sizeof_ptr, None) };
let pbuffer_shape =
ctx.builder.build_pointer_cast(pbuffer_shape, llvm_pusize, "").unwrap();
// Copy shape from buffer to `ndarray.shape`
ndarray.copy_shape_from_array(generator, ctx, pbuffer_shape);
call_memcpy_generic(
ctx,
ndarray.shape().base_ptr(ctx, generator),
pbuffer_dims,
sizeof_dims,
llvm_i1.const_zero(),
);
// Restore stack from before allocation of buffer
call_stackrestore(ctx, stackptr);
// Allocate `ndarray.data`.
// `ndarray.shape` must be initialized beforehand in this implementation
// (for ndarray.create_data() to know how many elements to allocate)
let num_elements =
call_ndarray_calc_size(generator, ctx, &ndarray.shape(), (None, None));
unsafe { ndarray.create_data(generator, ctx) }; // NOTE: the strides of `ndarray` has also been set to contiguous in `create_data`.
// debug_assert(nelems * sizeof(T) >= ndarray_nbytes)
if ctx.registry.llvm_options.opt_level == OptimizationLevel::None {
let sizeof_data =
ctx.builder.build_int_mul(num_elements, llvm_elem_sizeof, "").unwrap();
let num_elements = ndarray.size(generator, ctx);
let expected_ndarray_nbytes =
ctx.builder.build_int_mul(num_elements, itemsize, "").unwrap();
let cmp = ctx
.builder
.build_int_compare(
IntPredicate::UGE,
expected_ndarray_nbytes,
ndarray_nbytes,
"",
)
.unwrap();
ctx.make_assert(
generator,
ctx.builder.build_int_compare(IntPredicate::UGE,
sizeof_data,
ndarray_nbytes,
"",
).unwrap(),
cmp,
"0:AssertionError",
"Unexpected allocation size request for ndarray data - Expected up to {0} bytes, got {1} bytes",
[Some(sizeof_data), Some(ndarray_nbytes), None],
[Some(expected_ndarray_nbytes), Some(ndarray_nbytes), None],
ctx.current_loc,
);
}
ndarray.create_data(ctx, llvm_elem_ty, num_elements);
let ndarray_data = ndarray.data().base_ptr(ctx, generator);
let ndarray_data_i8 =
ctx.builder.build_pointer_cast(ndarray_data, llvm_pi8, "").unwrap();
// NOTE: Currently on `prehead_bb`
ctx.builder.build_unconditional_branch(head_bb).unwrap();
@ -756,7 +718,7 @@ fn format_rpc_ret<'ctx>(
ctx.builder.position_at_end(head_bb);
let phi = ctx.builder.build_phi(llvm_pi8, "rpc.ptr").unwrap();
phi.add_incoming(&[(&ndarray_data_i8, prehead_bb)]);
phi.add_incoming(&[(&ndarray_data, prehead_bb)]);
let alloc_size = ctx
.build_call_or_invoke(rpc_recv, &[phi.as_basic_value()], "rpc.size.next")
@ -771,12 +733,13 @@ fn format_rpc_ret<'ctx>(
ctx.builder.position_at_end(alloc_bb);
// Align the allocation to sizeof(T)
let alloc_size = round_up(ctx, alloc_size, llvm_elem_sizeof);
let alloc_size = round_up(ctx, alloc_size, itemsize);
// TODO(Derppening): Candidate for refactor into type_aligned_alloca
let alloc_ptr = ctx
.builder
.build_array_alloca(
llvm_elem_ty,
ctx.builder.build_int_unsigned_div(alloc_size, llvm_elem_sizeof, "").unwrap(),
dtype_llvm,
ctx.builder.build_int_unsigned_div(alloc_size, itemsize, "").unwrap(),
"rpc.alloc",
)
.unwrap();
@ -1375,62 +1338,50 @@ fn polymorphic_print<'ctx>(
}
TypeEnum::TObj { obj_id, .. } if *obj_id == PrimDef::NDArray.id() => {
let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
fmt.push_str("array([");
flush(ctx, generator, &mut fmt, &mut args);
let val = NDArrayValue::from_pointer_value(
value.into_pointer_value(),
llvm_elem_ty,
llvm_usize,
None,
);
let len = call_ndarray_calc_size(generator, ctx, &val.shape(), (None, None));
let last =
ctx.builder.build_int_sub(len, llvm_usize.const_int(1, false), "").unwrap();
let (dtype, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
let ndarray = NDArrayType::from_unifier_type(generator, ctx, ty)
.map_value(value.into_pointer_value(), None);
gen_for_callback_incrementing(
generator,
ctx,
None,
llvm_usize.const_zero(),
(len, false),
|generator, ctx, _, i| {
let elem = unsafe { val.data().get_unchecked(ctx, generator, &i, None) };
let num_0 = llvm_usize.const_zero();
polymorphic_print(
ctx,
generator,
&[(elem_ty, elem.into())],
"",
None,
true,
as_rtio,
)?;
// Print `ndarray` as a flat list delimited by interspersed with ", \0"
ndarray.foreach(generator, ctx, |generator, ctx, _, hdl| {
let i = hdl.get_index(ctx);
let scalar = hdl.get_scalar(ctx);
gen_if_callback(
generator,
ctx,
|_, ctx| {
Ok(ctx
.builder
.build_int_compare(IntPredicate::ULT, i, last, "")
.unwrap())
},
|generator, ctx| {
printf(ctx, generator, ", \0".into(), Vec::default());
// if (i != 0) puts(", ");
gen_if_callback(
generator,
ctx,
|_, ctx| {
let not_first = ctx
.builder
.build_int_compare(IntPredicate::NE, i, num_0, "")
.unwrap();
Ok(not_first)
},
|generator, ctx| {
printf(ctx, generator, ", \0".into(), Vec::default());
Ok(())
},
|_, _| Ok(()),
)?;
Ok(())
},
|_, _| Ok(()),
)?;
Ok(())
},
llvm_usize.const_int(1, false),
)?;
// Print element
polymorphic_print(
ctx,
generator,
&[(dtype, scalar.into())],
"",
None,
true,
as_rtio,
)?;
Ok(())
})?;
fmt.push_str(")]");
flush(ctx, generator, &mut fmt, &mut args);

View File

@ -10,12 +10,14 @@ use itertools::Itertools;
use parking_lot::RwLock;
use pyo3::{
types::{PyDict, PyTuple},
PyAny, PyObject, PyResult, Python,
PyAny, PyErr, PyObject, PyResult, Python,
};
use super::PrimitivePythonId;
use nac3core::{
codegen::{
types::{NDArrayType, ProxyType},
types::{ndarray::NDArrayType, ProxyType},
values::ndarray::make_contiguous_strides,
CodeGenContext, CodeGenerator,
},
inkwell::{
@ -37,8 +39,6 @@ use nac3core::{
},
};
use super::PrimitivePythonId;
pub enum PrimitiveValue {
I32(i32),
I64(i64),
@ -1085,18 +1085,19 @@ impl InnerResolver {
} else {
unreachable!("must be ndarray")
};
let (ndarray_dtype, ndarray_ndims) =
unpack_ndarray_var_tys(&mut ctx.unifier, ndarray_ty);
let (ndarray_dtype, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ndarray_ty);
let llvm_i8 = ctx.ctx.i8_type();
let llvm_pi8 = llvm_i8.ptr_type(AddressSpace::default());
let llvm_usize = generator.get_size_type(ctx.ctx);
let ndarray_dtype_llvm_ty = ctx.get_llvm_type(generator, ndarray_dtype);
let ndarray_llvm_ty = NDArrayType::new(generator, ctx.ctx, ndarray_dtype_llvm_ty);
let llvm_ndarray = NDArrayType::from_unifier_type(generator, ctx, ndarray_ty);
let dtype = llvm_ndarray.element_type();
{
if self.global_value_ids.read().contains_key(&id) {
let global = ctx.module.get_global(&id_str).unwrap_or_else(|| {
ctx.module.add_global(
ndarray_llvm_ty.as_base_type().get_element_type().into_struct_type(),
llvm_ndarray.as_base_type().get_element_type().into_struct_type(),
Some(AddressSpace::default()),
&id_str,
)
@ -1106,40 +1107,41 @@ impl InnerResolver {
self.global_value_ids.write().insert(id, obj.into());
}
let TypeEnum::TLiteral { values, .. } = &*ctx.unifier.get_ty_immutable(ndarray_ndims)
else {
unreachable!("Expected Literal for ndarray_ndims")
};
let ndarray_ndims = if values.len() == 1 {
values[0].clone()
} else {
todo!("Unpacking literal of more than one element unimplemented")
};
let Ok(ndarray_ndims) = u64::try_from(ndarray_ndims) else {
unreachable!("Expected u64 value for ndarray_ndims")
};
let ndims = llvm_ndarray.ndims().unwrap();
// Obtain the shape of the ndarray
let shape_tuple: &PyTuple = obj.getattr("shape")?.downcast()?;
assert_eq!(shape_tuple.len(), ndarray_ndims as usize);
let shape_values: Result<Option<Vec<_>>, _> = shape_tuple
assert_eq!(shape_tuple.len(), ndims as usize);
// The Rust type inferencer cannot figure this out
let shape_values = shape_tuple
.iter()
.enumerate()
.map(|(i, elem)| {
self.get_obj_value(py, elem, ctx, generator, ctx.primitives.usize()).map_err(
|e| super::CompileError::new_err(format!("Error getting element {i}: {e}")),
)
let value = self
.get_obj_value(py, elem, ctx, generator, ctx.primitives.usize())
.map_err(|e| {
super::CompileError::new_err(format!("Error getting element {i}: {e}"))
})?
.unwrap();
let value = value.into_int_value();
Ok(value)
})
.collect();
let shape_values = shape_values?.unwrap();
let shape_values = llvm_usize.const_array(
&shape_values.into_iter().map(BasicValueEnum::into_int_value).collect_vec(),
);
.collect::<Result<Vec<_>, PyErr>>()?;
// Also use this opportunity to get the constant values of `shape_values` for calculating strides.
let shape_u64s = shape_values
.iter()
.map(|dim| {
assert!(dim.is_const());
dim.get_zero_extended_constant().unwrap()
})
.collect_vec();
let shape_values = llvm_usize.const_array(&shape_values);
// create a global for ndarray.shape and initialize it using the shape
let shape_global = ctx.module.add_global(
llvm_usize.array_type(ndarray_ndims as u32),
llvm_usize.array_type(ndims as u32),
Some(AddressSpace::default()),
&(id_str.clone() + ".shape"),
);
@ -1147,17 +1149,25 @@ impl InnerResolver {
// Obtain the (flattened) elements of the ndarray
let sz: usize = obj.getattr("size")?.extract()?;
let data: Result<Option<Vec<_>>, _> = (0..sz)
let data: Vec<_> = (0..sz)
.map(|i| {
obj.getattr("flat")?.get_item(i).and_then(|elem| {
self.get_obj_value(py, elem, ctx, generator, ndarray_dtype).map_err(|e| {
super::CompileError::new_err(format!("Error getting element {i}: {e}"))
})
let value = self
.get_obj_value(py, elem, ctx, generator, ndarray_dtype)
.map_err(|e| {
super::CompileError::new_err(format!(
"Error getting element {i}: {e}"
))
})?
.unwrap();
assert_eq!(value.get_type(), dtype);
Ok(value)
})
})
.collect();
let data = data?.unwrap().into_iter();
let data = match ndarray_dtype_llvm_ty {
.try_collect()?;
let data = data.into_iter();
let data = match dtype {
BasicTypeEnum::ArrayType(ty) => {
ty.const_array(&data.map(BasicValueEnum::into_array_value).collect_vec())
}
@ -1182,38 +1192,68 @@ impl InnerResolver {
};
// create a global for ndarray.data and initialize it using the elements
//
// NOTE: NDArray's `data` is `u8*`. Here, `data_global` is an array of `dtype`.
// We will have to cast it to an `u8*` later.
let data_global = ctx.module.add_global(
ndarray_dtype_llvm_ty.array_type(sz as u32),
dtype.array_type(sz as u32),
Some(AddressSpace::default()),
&(id_str.clone() + ".data"),
);
data_global.set_initializer(&data);
// Get the constant itemsize.
let itemsize = dtype.size_of().unwrap();
let itemsize = itemsize.get_zero_extended_constant().unwrap();
// Create the strides needed for ndarray.strides
let strides = make_contiguous_strides(itemsize, ndims, &shape_u64s);
let strides =
strides.into_iter().map(|stride| llvm_usize.const_int(stride, false)).collect_vec();
let strides = llvm_usize.const_array(&strides);
// create a global for ndarray.strides and initialize it
let strides_global = ctx.module.add_global(
llvm_i8.array_type(ndims as u32),
Some(AddressSpace::default()),
&format!("${id_str}.strides"),
);
strides_global.set_initializer(&strides);
// create a global for the ndarray object and initialize it
let value = ndarray_llvm_ty
// NOTE: data_global is an array of dtype, we want a `u8*`.
let ndarray_data = data_global.as_pointer_value();
let ndarray_data = ctx.builder.build_pointer_cast(ndarray_data, llvm_pi8, "").unwrap();
let ndarray_itemsize = llvm_usize.const_int(itemsize, false);
let ndarray_ndims = llvm_usize.const_int(ndims, false);
let ndarray_shape = shape_global.as_pointer_value();
let ndarray_strides = strides_global.as_pointer_value();
let ndarray = llvm_ndarray
.as_base_type()
.get_element_type()
.into_struct_type()
.const_named_struct(&[
llvm_usize.const_int(ndarray_ndims, false).into(),
shape_global
.as_pointer_value()
.const_cast(llvm_usize.ptr_type(AddressSpace::default()))
.into(),
data_global
.as_pointer_value()
.const_cast(ndarray_dtype_llvm_ty.ptr_type(AddressSpace::default()))
.into(),
ndarray_itemsize.into(),
ndarray_ndims.into(),
ndarray_shape.into(),
ndarray_strides.into(),
ndarray_data.into(),
]);
let ndarray = ctx.module.add_global(
ndarray_llvm_ty.as_base_type().get_element_type().into_struct_type(),
let ndarray_global = ctx.module.add_global(
llvm_ndarray.as_base_type().get_element_type().into_struct_type(),
Some(AddressSpace::default()),
&id_str,
);
ndarray.set_initializer(&value);
ndarray_global.set_initializer(&ndarray);
Ok(Some(ndarray.as_pointer_value().into()))
Ok(Some(ndarray_global.as_pointer_value().into()))
} else if ty_id == self.primitive_ids.tuple {
let expected_ty_enum = ctx.unifier.get_ty_immutable(expected_ty);
let TypeEnum::TTuple { ty, is_vararg_ctx: false } = expected_ty_enum.as_ref() else {

View File

@ -2,4 +2,9 @@
#include "irrt/list.hpp"
#include "irrt/math.hpp"
#include "irrt/ndarray.hpp"
#include "irrt/range.hpp"
#include "irrt/slice.hpp"
#include "irrt/ndarray/basic.hpp"
#include "irrt/ndarray/def.hpp"
#include "irrt/ndarray/iter.hpp"
#include "irrt/ndarray/indexing.hpp"

View File

@ -22,6 +22,6 @@ using uint64_t = unsigned _ExtInt(64);
#endif
// NDArray indices are always `uint32_t`.
using NDIndex = uint32_t;
using NDIndexInt = uint32_t;
// The type of an index or a value describing the length of a range/slice is always `int32_t`.
using SliceIndex = int32_t;

View File

@ -2,6 +2,8 @@
#include "irrt/int_types.hpp"
// TODO: To be deleted since NDArray with strides is done.
namespace {
template<typename SizeT>
SizeT __nac3_ndarray_calc_size_impl(const SizeT* list_data, SizeT list_len, SizeT begin_idx, SizeT end_idx) {
@ -17,7 +19,7 @@ SizeT __nac3_ndarray_calc_size_impl(const SizeT* list_data, SizeT list_len, Size
}
template<typename SizeT>
void __nac3_ndarray_calc_nd_indices_impl(SizeT index, const SizeT* dims, SizeT num_dims, NDIndex* idxs) {
void __nac3_ndarray_calc_nd_indices_impl(SizeT index, const SizeT* dims, SizeT num_dims, NDIndexInt* idxs) {
SizeT stride = 1;
for (SizeT dim = 0; dim < num_dims; dim++) {
SizeT i = num_dims - dim - 1;
@ -28,7 +30,10 @@ void __nac3_ndarray_calc_nd_indices_impl(SizeT index, const SizeT* dims, SizeT n
}
template<typename SizeT>
SizeT __nac3_ndarray_flatten_index_impl(const SizeT* dims, SizeT num_dims, const NDIndex* indices, SizeT num_indices) {
SizeT __nac3_ndarray_flatten_index_impl(const SizeT* dims,
SizeT num_dims,
const NDIndexInt* indices,
SizeT num_indices) {
SizeT idx = 0;
SizeT stride = 1;
for (SizeT i = 0; i < num_dims; ++i) {
@ -75,8 +80,8 @@ void __nac3_ndarray_calc_broadcast_impl(const SizeT* lhs_dims,
template<typename SizeT>
void __nac3_ndarray_calc_broadcast_idx_impl(const SizeT* src_dims,
SizeT src_ndims,
const NDIndex* in_idx,
NDIndex* out_idx) {
const NDIndexInt* in_idx,
NDIndexInt* out_idx) {
for (SizeT i = 0; i < src_ndims; ++i) {
SizeT src_i = src_ndims - i - 1;
out_idx[src_i] = src_dims[src_i] == 1 ? 0 : in_idx[src_i];
@ -94,21 +99,23 @@ __nac3_ndarray_calc_size64(const uint64_t* list_data, uint64_t list_len, uint64_
return __nac3_ndarray_calc_size_impl(list_data, list_len, begin_idx, end_idx);
}
void __nac3_ndarray_calc_nd_indices(uint32_t index, const uint32_t* dims, uint32_t num_dims, NDIndex* idxs) {
void __nac3_ndarray_calc_nd_indices(uint32_t index, const uint32_t* dims, uint32_t num_dims, NDIndexInt* idxs) {
__nac3_ndarray_calc_nd_indices_impl(index, dims, num_dims, idxs);
}
void __nac3_ndarray_calc_nd_indices64(uint64_t index, const uint64_t* dims, uint64_t num_dims, NDIndex* idxs) {
void __nac3_ndarray_calc_nd_indices64(uint64_t index, const uint64_t* dims, uint64_t num_dims, NDIndexInt* idxs) {
__nac3_ndarray_calc_nd_indices_impl(index, dims, num_dims, idxs);
}
uint32_t
__nac3_ndarray_flatten_index(const uint32_t* dims, uint32_t num_dims, const NDIndex* indices, uint32_t num_indices) {
__nac3_ndarray_flatten_index(const uint32_t* dims, uint32_t num_dims, const NDIndexInt* indices, uint32_t num_indices) {
return __nac3_ndarray_flatten_index_impl(dims, num_dims, indices, num_indices);
}
uint64_t
__nac3_ndarray_flatten_index64(const uint64_t* dims, uint64_t num_dims, const NDIndex* indices, uint64_t num_indices) {
uint64_t __nac3_ndarray_flatten_index64(const uint64_t* dims,
uint64_t num_dims,
const NDIndexInt* indices,
uint64_t num_indices) {
return __nac3_ndarray_flatten_index_impl(dims, num_dims, indices, num_indices);
}
@ -130,15 +137,15 @@ void __nac3_ndarray_calc_broadcast64(const uint64_t* lhs_dims,
void __nac3_ndarray_calc_broadcast_idx(const uint32_t* src_dims,
uint32_t src_ndims,
const NDIndex* in_idx,
NDIndex* out_idx) {
const NDIndexInt* in_idx,
NDIndexInt* out_idx) {
__nac3_ndarray_calc_broadcast_idx_impl(src_dims, src_ndims, in_idx, out_idx);
}
void __nac3_ndarray_calc_broadcast_idx64(const uint64_t* src_dims,
uint64_t src_ndims,
const NDIndex* in_idx,
NDIndex* out_idx) {
const NDIndexInt* in_idx,
NDIndexInt* out_idx) {
__nac3_ndarray_calc_broadcast_idx_impl(src_dims, src_ndims, in_idx, out_idx);
}
} // namespace

View File

@ -0,0 +1,342 @@
#pragma once
#include "irrt/debug.hpp"
#include "irrt/exception.hpp"
#include "irrt/int_types.hpp"
#include "irrt/ndarray/def.hpp"
namespace {
namespace ndarray {
namespace basic {
/**
* @brief Assert that `shape` does not contain negative dimensions.
*
* @param ndims Number of dimensions in `shape`
* @param shape The shape to check on
*/
template<typename SizeT>
void assert_shape_no_negative(SizeT ndims, const SizeT* shape) {
for (SizeT axis = 0; axis < ndims; axis++) {
if (shape[axis] < 0) {
raise_exception(SizeT, EXN_VALUE_ERROR,
"negative dimensions are not allowed; axis {0} "
"has dimension {1}",
axis, shape[axis], NO_PARAM);
}
}
}
/**
* @brief Assert that two shapes are the same in the context of writing output to an ndarray.
*/
template<typename SizeT>
void assert_output_shape_same(SizeT ndarray_ndims,
const SizeT* ndarray_shape,
SizeT output_ndims,
const SizeT* output_shape) {
if (ndarray_ndims != output_ndims) {
// There is no corresponding NumPy error message like this.
raise_exception(SizeT, EXN_VALUE_ERROR, "Cannot write output of ndims {0} to an ndarray with ndims {1}",
output_ndims, ndarray_ndims, NO_PARAM);
}
for (SizeT axis = 0; axis < ndarray_ndims; axis++) {
if (ndarray_shape[axis] != output_shape[axis]) {
// There is no corresponding NumPy error message like this.
raise_exception(SizeT, EXN_VALUE_ERROR,
"Mismatched dimensions on axis {0}, output has "
"dimension {1}, but destination ndarray has dimension {2}.",
axis, output_shape[axis], ndarray_shape[axis]);
}
}
}
/**
* @brief Return the number of elements of an ndarray given its shape.
*
* @param ndims Number of dimensions in `shape`
* @param shape The shape of the ndarray
*/
template<typename SizeT>
SizeT calc_size_from_shape(SizeT ndims, const SizeT* shape) {
SizeT size = 1;
for (SizeT axis = 0; axis < ndims; axis++)
size *= shape[axis];
return size;
}
/**
* @brief Compute the array indices of the `nth` (0-based) element of an ndarray given only its shape.
*
* @param ndims Number of elements in `shape` and `indices`
* @param shape The shape of the ndarray
* @param indices The returned indices indexing the ndarray with shape `shape`.
* @param nth The index of the element of interest.
*/
template<typename SizeT>
void set_indices_by_nth(SizeT ndims, const SizeT* shape, SizeT* indices, SizeT nth) {
for (SizeT i = 0; i < ndims; i++) {
SizeT axis = ndims - i - 1;
SizeT dim = shape[axis];
indices[axis] = nth % dim;
nth /= dim;
}
}
/**
* @brief Return the number of elements of an `ndarray`
*
* This function corresponds to `<an_ndarray>.size`
*/
template<typename SizeT>
SizeT size(const NDArray<SizeT>* ndarray) {
return calc_size_from_shape(ndarray->ndims, ndarray->shape);
}
/**
* @brief Return of the number of its content of an `ndarray`.
*
* This function corresponds to `<an_ndarray>.nbytes`.
*/
template<typename SizeT>
SizeT nbytes(const NDArray<SizeT>* ndarray) {
return size(ndarray) * ndarray->itemsize;
}
/**
* @brief Get the `len()` of an ndarray, and asserts that `ndarray` is a sized object.
*
* This function corresponds to `<an_ndarray>.__len__`.
*
* @param dst_length The length.
*/
template<typename SizeT>
SizeT len(const NDArray<SizeT>* ndarray) {
if (ndarray->ndims != 0) {
return ndarray->shape[0];
}
// numpy prohibits `__len__` on unsized objects
raise_exception(SizeT, EXN_TYPE_ERROR, "len() of unsized object", NO_PARAM, NO_PARAM, NO_PARAM);
__builtin_unreachable();
}
/**
* @brief Return a boolean indicating if `ndarray` is (C-)contiguous.
*
* You may want to see ndarray's rules for C-contiguity:
* https://github.com/numpy/numpy/blob/df256d0d2f3bc6833699529824781c58f9c6e697/numpy/core/src/multiarray/flagsobject.c#L95C1-L99C45
*/
template<typename SizeT>
bool is_c_contiguous(const NDArray<SizeT>* ndarray) {
// References:
// - tinynumpy's implementation:
// https://github.com/wadetb/tinynumpy/blob/0d23d22e07062ffab2afa287374c7b366eebdda1/tinynumpy/tinynumpy.py#L102
// - ndarray's flags["C_CONTIGUOUS"]:
// https://numpy.org/doc/stable/reference/generated/numpy.ndarray.flags.html#numpy.ndarray.flags
// - ndarray's rules for C-contiguity:
// https://github.com/numpy/numpy/blob/df256d0d2f3bc6833699529824781c58f9c6e697/numpy/core/src/multiarray/flagsobject.c#L95C1-L99C45
// From
// https://github.com/numpy/numpy/blob/df256d0d2f3bc6833699529824781c58f9c6e697/numpy/core/src/multiarray/flagsobject.c#L95C1-L99C45:
//
// The traditional rule is that for an array to be flagged as C contiguous,
// the following must hold:
//
// strides[-1] == itemsize
// strides[i] == shape[i+1] * strides[i + 1]
// [...]
// According to these rules, a 0- or 1-dimensional array is either both
// C- and F-contiguous, or neither; and an array with 2+ dimensions
// can be C- or F- contiguous, or neither, but not both. Though there
// there are exceptions for arrays with zero or one item, in the first
// case the check is relaxed up to and including the first dimension
// with shape[i] == 0. In the second case `strides == itemsize` will
// can be true for all dimensions and both flags are set.
if (ndarray->ndims == 0) {
return true;
}
if (ndarray->strides[ndarray->ndims - 1] != ndarray->itemsize) {
return false;
}
for (SizeT i = 1; i < ndarray->ndims; i++) {
SizeT axis_i = ndarray->ndims - i - 1;
if (ndarray->strides[axis_i] != ndarray->shape[axis_i + 1] * ndarray->strides[axis_i + 1]) {
return false;
}
}
return true;
}
/**
* @brief Return the pointer to the element indexed by `indices` along the ndarray's axes.
*
* This function does no bound check.
*/
template<typename SizeT>
void* get_pelement_by_indices(const NDArray<SizeT>* ndarray, const SizeT* indices) {
void* element = ndarray->data;
for (SizeT dim_i = 0; dim_i < ndarray->ndims; dim_i++)
element = static_cast<uint8_t*>(element) + indices[dim_i] * ndarray->strides[dim_i];
return element;
}
/**
* @brief Return the pointer to the nth (0-based) element of `ndarray` in flattened view.
*
* This function does no bound check.
*/
template<typename SizeT>
void* get_nth_pelement(const NDArray<SizeT>* ndarray, SizeT nth) {
void* element = ndarray->data;
for (SizeT i = 0; i < ndarray->ndims; i++) {
SizeT axis = ndarray->ndims - i - 1;
SizeT dim = ndarray->shape[axis];
element = static_cast<uint8_t*>(element) + ndarray->strides[axis] * (nth % dim);
nth /= dim;
}
return element;
}
/**
* @brief Update the strides of an ndarray given an ndarray `shape` to be contiguous.
*
* You might want to read https://ajcr.net/stride-guide-part-1/.
*/
template<typename SizeT>
void set_strides_by_shape(NDArray<SizeT>* ndarray) {
SizeT stride_product = 1;
for (SizeT i = 0; i < ndarray->ndims; i++) {
SizeT axis = ndarray->ndims - i - 1;
ndarray->strides[axis] = stride_product * ndarray->itemsize;
stride_product *= ndarray->shape[axis];
}
}
/**
* @brief Set an element in `ndarray`.
*
* @param pelement Pointer to the element in `ndarray` to be set.
* @param pvalue Pointer to the value `pelement` will be set to.
*/
template<typename SizeT>
void set_pelement_value(NDArray<SizeT>* ndarray, void* pelement, const void* pvalue) {
__builtin_memcpy(pelement, pvalue, ndarray->itemsize);
}
/**
* @brief Copy data from one ndarray to another of the exact same size and itemsize.
*
* Both ndarrays will be viewed in their flatten views when copying the elements.
*/
template<typename SizeT>
void copy_data(const NDArray<SizeT>* src_ndarray, NDArray<SizeT>* dst_ndarray) {
// TODO: Make this faster with memcpy when we see a contiguous segment.
// TODO: Handle overlapping.
debug_assert_eq(SizeT, src_ndarray->itemsize, dst_ndarray->itemsize);
for (SizeT i = 0; i < size(src_ndarray); i++) {
auto src_element = ndarray::basic::get_nth_pelement(src_ndarray, i);
auto dst_element = ndarray::basic::get_nth_pelement(dst_ndarray, i);
ndarray::basic::set_pelement_value(dst_ndarray, dst_element, src_element);
}
}
} // namespace basic
} // namespace ndarray
} // namespace
extern "C" {
using namespace ndarray::basic;
void __nac3_ndarray_util_assert_shape_no_negative(int32_t ndims, int32_t* shape) {
assert_shape_no_negative(ndims, shape);
}
void __nac3_ndarray_util_assert_shape_no_negative64(int64_t ndims, int64_t* shape) {
assert_shape_no_negative(ndims, shape);
}
void __nac3_ndarray_util_assert_output_shape_same(int32_t ndarray_ndims,
const int32_t* ndarray_shape,
int32_t output_ndims,
const int32_t* output_shape) {
assert_output_shape_same(ndarray_ndims, ndarray_shape, output_ndims, output_shape);
}
void __nac3_ndarray_util_assert_output_shape_same64(int64_t ndarray_ndims,
const int64_t* ndarray_shape,
int64_t output_ndims,
const int64_t* output_shape) {
assert_output_shape_same(ndarray_ndims, ndarray_shape, output_ndims, output_shape);
}
uint32_t __nac3_ndarray_size(NDArray<int32_t>* ndarray) {
return size(ndarray);
}
uint64_t __nac3_ndarray_size64(NDArray<int64_t>* ndarray) {
return size(ndarray);
}
uint32_t __nac3_ndarray_nbytes(NDArray<int32_t>* ndarray) {
return nbytes(ndarray);
}
uint64_t __nac3_ndarray_nbytes64(NDArray<int64_t>* ndarray) {
return nbytes(ndarray);
}
int32_t __nac3_ndarray_len(NDArray<int32_t>* ndarray) {
return len(ndarray);
}
int64_t __nac3_ndarray_len64(NDArray<int64_t>* ndarray) {
return len(ndarray);
}
bool __nac3_ndarray_is_c_contiguous(NDArray<int32_t>* ndarray) {
return is_c_contiguous(ndarray);
}
bool __nac3_ndarray_is_c_contiguous64(NDArray<int64_t>* ndarray) {
return is_c_contiguous(ndarray);
}
void* __nac3_ndarray_get_nth_pelement(const NDArray<int32_t>* ndarray, int32_t nth) {
return get_nth_pelement(ndarray, nth);
}
void* __nac3_ndarray_get_nth_pelement64(const NDArray<int64_t>* ndarray, int64_t nth) {
return get_nth_pelement(ndarray, nth);
}
void* __nac3_ndarray_get_pelement_by_indices(const NDArray<int32_t>* ndarray, int32_t* indices) {
return get_pelement_by_indices(ndarray, indices);
}
void* __nac3_ndarray_get_pelement_by_indices64(const NDArray<int64_t>* ndarray, int64_t* indices) {
return get_pelement_by_indices(ndarray, indices);
}
void __nac3_ndarray_set_strides_by_shape(NDArray<int32_t>* ndarray) {
set_strides_by_shape(ndarray);
}
void __nac3_ndarray_set_strides_by_shape64(NDArray<int64_t>* ndarray) {
set_strides_by_shape(ndarray);
}
void __nac3_ndarray_copy_data(NDArray<int32_t>* src_ndarray, NDArray<int32_t>* dst_ndarray) {
copy_data(src_ndarray, dst_ndarray);
}
void __nac3_ndarray_copy_data64(NDArray<int64_t>* src_ndarray, NDArray<int64_t>* dst_ndarray) {
copy_data(src_ndarray, dst_ndarray);
}
}

View File

@ -0,0 +1,51 @@
#pragma once
#include "irrt/int_types.hpp"
namespace {
/**
* @brief The NDArray object
*
* Official numpy implementation:
* https://github.com/numpy/numpy/blob/735a477f0bc2b5b84d0e72d92f224bde78d4e069/doc/source/reference/c-api/types-and-structures.rst#pyarrayinterface
*
* Note that this implementation is based on `PyArrayInterface` rather of `PyArrayObject`. The
* difference between `PyArrayInterface` and `PyArrayObject` (relevant to our implementation) is
* that `PyArrayInterface` *has* `itemsize` and uses `void*` for its `data`, whereas `PyArrayObject`
* does not require `itemsize` (probably using `strides[-1]` instead) and uses `char*` for its
* `data`. There are also minor differences in the struct layout.
*/
template<typename SizeT>
struct NDArray {
/**
* @brief The number of bytes of a single element in `data`.
*/
SizeT itemsize;
/**
* @brief The number of dimensions of this shape.
*/
SizeT ndims;
/**
* @brief The NDArray shape, with length equal to `ndims`.
*
* Note that it may contain 0.
*/
SizeT* shape;
/**
* @brief Array strides, with length equal to `ndims`
*
* The stride values are in units of bytes, not number of elements.
*
* Note that `strides` can have negative values or contain 0.
*/
SizeT* strides;
/**
* @brief The underlying data this `ndarray` is pointing to.
*/
void* data;
};
} // namespace

View File

@ -0,0 +1,220 @@
#pragma once
#include "irrt/exception.hpp"
#include "irrt/int_types.hpp"
#include "irrt/ndarray/basic.hpp"
#include "irrt/ndarray/def.hpp"
#include "irrt/range.hpp"
#include "irrt/slice.hpp"
namespace {
typedef uint8_t NDIndexType;
/**
* @brief A single element index
*
* `data` points to a `int32_t`.
*/
const NDIndexType ND_INDEX_TYPE_SINGLE_ELEMENT = 0;
/**
* @brief A slice index
*
* `data` points to a `Slice<int32_t>`.
*/
const NDIndexType ND_INDEX_TYPE_SLICE = 1;
/**
* @brief `np.newaxis` / `None`
*
* `data` is unused.
*/
const NDIndexType ND_INDEX_TYPE_NEWAXIS = 2;
/**
* @brief `Ellipsis` / `...`
*
* `data` is unused.
*/
const NDIndexType ND_INDEX_TYPE_ELLIPSIS = 3;
/**
* @brief An index used in ndarray indexing
*
* That is:
* ```
* my_ndarray[::-1, 3, ..., np.newaxis]
* ^^^^ ^ ^^^ ^^^^^^^^^^ each of these is represented by an NDIndex.
* ```
*/
struct NDIndex {
/**
* @brief Enum tag to specify the type of index.
*
* Please see the comment of each enum constant.
*/
NDIndexType type;
/**
* @brief The accompanying data associated with `type`.
*
* Please see the comment of each enum constant.
*/
uint8_t* data;
};
} // namespace
namespace {
namespace ndarray {
namespace indexing {
/**
* @brief Perform ndarray "basic indexing" (https://numpy.org/doc/stable/user/basics.indexing.html#basic-indexing)
*
* This function is very similar to performing `dst_ndarray = src_ndarray[indices]` in Python.
*
* This function also does proper assertions on `indices` to check for out of bounds access and more.
*
* # Notes on `dst_ndarray`
* The caller is responsible for allocating space for the resulting ndarray.
* Here is what this function expects from `dst_ndarray` when called:
* - `dst_ndarray->data` does not have to be initialized.
* - `dst_ndarray->itemsize` does not have to be initialized.
* - `dst_ndarray->ndims` must be initialized, and it must be equal to the expected `ndims` of the `dst_ndarray` after
* indexing `src_ndarray` with `indices`.
* - `dst_ndarray->shape` must be allocated, through it can contain uninitialized values.
* - `dst_ndarray->strides` must be allocated, through it can contain uninitialized values.
* When this function call ends:
* - `dst_ndarray->data` is set to `src_ndarray->data`.
* - `dst_ndarray->itemsize` is set to `src_ndarray->itemsize`.
* - `dst_ndarray->ndims` is unchanged.
* - `dst_ndarray->shape` is updated according to how `src_ndarray` is indexed.
* - `dst_ndarray->strides` is updated accordingly by how ndarray indexing works.
*
* @param indices indices to index `src_ndarray`, ordered in the same way you would write them in Python.
* @param src_ndarray The NDArray to be indexed.
* @param dst_ndarray The resulting NDArray after indexing. Further details in the comments above,
*/
template<typename SizeT>
void index(SizeT num_indices, const NDIndex* indices, const NDArray<SizeT>* src_ndarray, NDArray<SizeT>* dst_ndarray) {
// Validate `indices`.
// Expected value of `dst_ndarray->ndims`.
SizeT expected_dst_ndims = src_ndarray->ndims;
// To check for "too many indices for array: array is ?-dimensional, but ? were indexed"
SizeT num_indexed = 0;
// There may be ellipsis `...` in `indices`. There can only be 0 or 1 ellipsis.
SizeT num_ellipsis = 0;
for (SizeT i = 0; i < num_indices; i++) {
if (indices[i].type == ND_INDEX_TYPE_SINGLE_ELEMENT) {
expected_dst_ndims--;
num_indexed++;
} else if (indices[i].type == ND_INDEX_TYPE_SLICE) {
num_indexed++;
} else if (indices[i].type == ND_INDEX_TYPE_NEWAXIS) {
expected_dst_ndims++;
} else if (indices[i].type == ND_INDEX_TYPE_ELLIPSIS) {
num_ellipsis++;
if (num_ellipsis > 1) {
raise_exception(SizeT, EXN_INDEX_ERROR, "an index can only have a single ellipsis ('...')", NO_PARAM,
NO_PARAM, NO_PARAM);
}
} else {
__builtin_unreachable();
}
}
debug_assert_eq(SizeT, expected_dst_ndims, dst_ndarray->ndims);
if (src_ndarray->ndims - num_indexed < 0) {
raise_exception(SizeT, EXN_INDEX_ERROR,
"too many indices for array: array is {0}-dimensional, "
"but {1} were indexed",
src_ndarray->ndims, num_indices, NO_PARAM);
}
dst_ndarray->data = src_ndarray->data;
dst_ndarray->itemsize = src_ndarray->itemsize;
// Reference code:
// https://github.com/wadetb/tinynumpy/blob/0d23d22e07062ffab2afa287374c7b366eebdda1/tinynumpy/tinynumpy.py#L652
SizeT src_axis = 0;
SizeT dst_axis = 0;
for (int32_t i = 0; i < num_indices; i++) {
const NDIndex* index = &indices[i];
if (index->type == ND_INDEX_TYPE_SINGLE_ELEMENT) {
SizeT input = (SizeT) * ((int32_t*)index->data);
SizeT k = slice::resolve_index_in_length(src_ndarray->shape[src_axis], input);
if (k == -1) {
raise_exception(SizeT, EXN_INDEX_ERROR,
"index {0} is out of bounds for axis {1} "
"with size {2}",
input, src_axis, src_ndarray->shape[src_axis]);
}
dst_ndarray->data = static_cast<uint8_t*>(dst_ndarray->data) + k * src_ndarray->strides[src_axis];
src_axis++;
} else if (index->type == ND_INDEX_TYPE_SLICE) {
Slice<int32_t>* slice = (Slice<int32_t>*)index->data;
Range<int32_t> range = slice->indices_checked<SizeT>(src_ndarray->shape[src_axis]);
dst_ndarray->data = static_cast<uint8_t*>(dst_ndarray->data) + (SizeT)range.start * src_ndarray->strides[src_axis];
dst_ndarray->strides[dst_axis] = ((SizeT)range.step) * src_ndarray->strides[src_axis];
dst_ndarray->shape[dst_axis] = (SizeT)range.len<SizeT>();
dst_axis++;
src_axis++;
} else if (index->type == ND_INDEX_TYPE_NEWAXIS) {
dst_ndarray->strides[dst_axis] = 0;
dst_ndarray->shape[dst_axis] = 1;
dst_axis++;
} else if (index->type == ND_INDEX_TYPE_ELLIPSIS) {
// The number of ':' entries this '...' implies.
SizeT ellipsis_size = src_ndarray->ndims - num_indexed;
for (SizeT j = 0; j < ellipsis_size; j++) {
dst_ndarray->strides[dst_axis] = src_ndarray->strides[src_axis];
dst_ndarray->shape[dst_axis] = src_ndarray->shape[src_axis];
dst_axis++;
src_axis++;
}
} else {
__builtin_unreachable();
}
}
for (; dst_axis < dst_ndarray->ndims; dst_axis++, src_axis++) {
dst_ndarray->shape[dst_axis] = src_ndarray->shape[src_axis];
dst_ndarray->strides[dst_axis] = src_ndarray->strides[src_axis];
}
debug_assert_eq(SizeT, src_ndarray->ndims, src_axis);
debug_assert_eq(SizeT, dst_ndarray->ndims, dst_axis);
}
} // namespace indexing
} // namespace ndarray
} // namespace
extern "C" {
using namespace ndarray::indexing;
void __nac3_ndarray_index(int32_t num_indices,
NDIndex* indices,
NDArray<int32_t>* src_ndarray,
NDArray<int32_t>* dst_ndarray) {
index(num_indices, indices, src_ndarray, dst_ndarray);
}
void __nac3_ndarray_index64(int64_t num_indices,
NDIndex* indices,
NDArray<int64_t>* src_ndarray,
NDArray<int64_t>* dst_ndarray) {
index(num_indices, indices, src_ndarray, dst_ndarray);
}
}

View File

@ -0,0 +1,146 @@
#pragma once
#include "irrt/int_types.hpp"
#include "irrt/ndarray/def.hpp"
namespace {
/**
* @brief Helper struct to enumerate through an ndarray *efficiently*.
*
* Example usage (in pseudo-code):
* ```
* // Suppose my_ndarray has been initialized, with shape [2, 3] and dtype `double`
* NDIter nditer;
* nditer.initialize(my_ndarray);
* while (nditer.has_element()) {
* // This body is run 6 (= my_ndarray.size) times.
*
* // [0, 0] -> [0, 1] -> [0, 2] -> [1, 0] -> [1, 1] -> [1, 2] -> end
* print(nditer.indices);
*
* // 0 -> 1 -> 2 -> 3 -> 4 -> 5
* print(nditer.nth);
*
* // <1st element> -> <2nd element> -> ... -> <6th element> -> end
* print(*((double *) nditer.element))
*
* nditer.next(); // Go to next element.
* }
* ```
*
* Interesting cases:
* - If `my_ndarray.ndims` == 0, there is one iteration.
* - If `my_ndarray.shape` contains zeroes, there are no iterations.
*/
template<typename SizeT>
struct NDIter {
// Information about the ndarray being iterated over.
SizeT ndims;
SizeT* shape;
SizeT* strides;
/**
* @brief The current indices.
*
* Must be allocated by the caller.
*/
SizeT* indices;
/**
* @brief The nth (0-based) index of the current indices.
*
* Initially this is 0.
*/
SizeT nth;
/**
* @brief Pointer to the current element.
*
* Initially this points to first element of the ndarray.
*/
void* element;
/**
* @brief Cache for the product of shape.
*
* Could be 0 if `shape` has 0s in it.
*/
SizeT size;
void initialize(SizeT ndims, SizeT* shape, SizeT* strides, void* element, SizeT* indices) {
this->ndims = ndims;
this->shape = shape;
this->strides = strides;
this->indices = indices;
this->element = element;
// Compute size
this->size = 1;
for (SizeT i = 0; i < ndims; i++) {
this->size *= shape[i];
}
// `indices` starts on all 0s.
for (SizeT axis = 0; axis < ndims; axis++)
indices[axis] = 0;
nth = 0;
}
void initialize_by_ndarray(NDArray<SizeT>* ndarray, SizeT* indices) {
// NOTE: ndarray->data is pointing to the first element, and `NDIter`'s `element` should also point to the first
// element as well.
this->initialize(ndarray->ndims, ndarray->shape, ndarray->strides, ndarray->data, indices);
}
// Is the current iteration valid?
// If true, then `element`, `indices` and `nth` contain details about the current element.
bool has_element() { return nth < size; }
// Go to the next element.
void next() {
for (SizeT i = 0; i < ndims; i++) {
SizeT axis = ndims - i - 1;
indices[axis]++;
if (indices[axis] >= shape[axis]) {
indices[axis] = 0;
// TODO: There is something called backstrides to speedup iteration.
// See https://ajcr.net/stride-guide-part-1/, and
// https://docs.scipy.org/doc/numpy-1.13.0/reference/c-api.types-and-structures.html#c.PyArrayIterObject.PyArrayIterObject.backstrides.
element = static_cast<void*>(reinterpret_cast<uint8_t*>(element) - strides[axis] * (shape[axis] - 1));
} else {
element = static_cast<void*>(reinterpret_cast<uint8_t*>(element) + strides[axis]);
break;
}
}
nth++;
}
};
} // namespace
extern "C" {
void __nac3_nditer_initialize(NDIter<int32_t>* iter, NDArray<int32_t>* ndarray, int32_t* indices) {
iter->initialize_by_ndarray(ndarray, indices);
}
void __nac3_nditer_initialize64(NDIter<int64_t>* iter, NDArray<int64_t>* ndarray, int64_t* indices) {
iter->initialize_by_ndarray(ndarray, indices);
}
bool __nac3_nditer_has_element(NDIter<int32_t>* iter) {
return iter->has_element();
}
bool __nac3_nditer_has_element64(NDIter<int64_t>* iter) {
return iter->has_element();
}
void __nac3_nditer_next(NDIter<int32_t>* iter) {
iter->next();
}
void __nac3_nditer_next64(NDIter<int64_t>* iter) {
iter->next();
}
}

View File

@ -0,0 +1,47 @@
#pragma once
#include "irrt/debug.hpp"
#include "irrt/int_types.hpp"
namespace {
namespace range {
template<typename T>
T len(T start, T stop, T step) {
// Reference:
// https://github.com/python/cpython/blob/9dbd12375561a393eaec4b21ee4ac568a407cdb0/Objects/rangeobject.c#L933
if (step > 0 && start < stop)
return 1 + (stop - 1 - start) / step;
else if (step < 0 && start > stop)
return 1 + (start - 1 - stop) / (-step);
else
return 0;
}
} // namespace range
/**
* @brief A Python range.
*/
template<typename T>
struct Range {
T start;
T stop;
T step;
/**
* @brief Calculate the `len()` of this range.
*/
template<typename SizeT>
T len() {
debug_assert(SizeT, step != 0);
return range::len(start, stop, step);
}
};
} // namespace
extern "C" {
using namespace range;
SliceIndex __nac3_range_slice_len(const SliceIndex start, const SliceIndex end, const SliceIndex step) {
return len(start, end, step);
}
}

View File

@ -1,6 +1,145 @@
#pragma once
#include "irrt/debug.hpp"
#include "irrt/exception.hpp"
#include "irrt/int_types.hpp"
#include "irrt/math_util.hpp"
#include "irrt/range.hpp"
namespace {
namespace slice {
/**
* @brief Resolve a possibly negative index in a list of a known length.
*
* Returns -1 if the resolved index is out of the list's bounds.
*/
template<typename T>
T resolve_index_in_length(T length, T index) {
T resolved = index < 0 ? length + index : index;
if (0 <= resolved && resolved < length) {
return resolved;
} else {
return -1;
}
}
/**
* @brief Resolve a slice as a range.
*
* This is equivalent to `range(*slice(start, stop, step).indices(length))` in Python.
*/
template<typename T>
void indices(bool start_defined,
T start,
bool stop_defined,
T stop,
bool step_defined,
T step,
T length,
T* range_start,
T* range_stop,
T* range_step) {
// Reference: https://github.com/python/cpython/blob/main/Objects/sliceobject.c#L388
*range_step = step_defined ? step : 1;
bool step_is_negative = *range_step < 0;
T lower, upper;
if (step_is_negative) {
lower = -1;
upper = length - 1;
} else {
lower = 0;
upper = length;
}
if (start_defined) {
*range_start = start < 0 ? max(lower, start + length) : min(upper, start);
} else {
*range_start = step_is_negative ? upper : lower;
}
if (stop_defined) {
*range_stop = stop < 0 ? max(lower, stop + length) : min(upper, stop);
} else {
*range_stop = step_is_negative ? lower : upper;
}
}
} // namespace slice
/**
* @brief A Python-like slice with **unresolved** indices.
*/
template<typename T>
struct Slice {
bool start_defined;
T start;
bool stop_defined;
T stop;
bool step_defined;
T step;
Slice() { this->reset(); }
void reset() {
this->start_defined = false;
this->stop_defined = false;
this->step_defined = false;
}
void set_start(T start) {
this->start_defined = true;
this->start = start;
}
void set_stop(T stop) {
this->stop_defined = true;
this->stop = stop;
}
void set_step(T step) {
this->step_defined = true;
this->step = step;
}
/**
* @brief Resolve this slice as a range.
*
* In Python, this would be `range(*slice(start, stop, step).indices(length))`.
*/
template<typename SizeT>
Range<T> indices(T length) {
// Reference:
// https://github.com/python/cpython/blob/main/Objects/sliceobject.c#L388
debug_assert(SizeT, length >= 0);
Range<T> result;
slice::indices(start_defined, start, stop_defined, stop, step_defined, step, length, &result.start,
&result.stop, &result.step);
return result;
}
/**
* @brief Like `.indices()` but with assertions.
*/
template<typename SizeT>
Range<T> indices_checked(T length) {
// TODO: Switch to `SizeT length`
if (length < 0) {
raise_exception(SizeT, EXN_VALUE_ERROR, "length should not be negative, got {0}", length, NO_PARAM,
NO_PARAM);
}
if (this->step_defined && this->step == 0) {
raise_exception(SizeT, EXN_VALUE_ERROR, "slice step cannot be zero", NO_PARAM, NO_PARAM, NO_PARAM);
}
return this->indices<SizeT>(length);
}
};
} // namespace
extern "C" {
SliceIndex __nac3_slice_index_bound(SliceIndex i, const SliceIndex len) {
@ -14,15 +153,4 @@ SliceIndex __nac3_slice_index_bound(SliceIndex i, const SliceIndex len) {
}
return i;
}
SliceIndex __nac3_range_slice_len(const SliceIndex start, const SliceIndex end, const SliceIndex step) {
SliceIndex diff = end - start;
if (diff > 0 && step > 0) {
return ((diff - 1) / step) + 1;
} else if (diff < 0 && step < 0) {
return ((diff + 1) / step) + 1;
} else {
return 0;
}
}
} // namespace

File diff suppressed because it is too large Load Diff

View File

@ -32,9 +32,10 @@ use super::{
gen_for_callback_incrementing, gen_if_callback, gen_if_else_expr_callback, gen_raise,
gen_var,
},
types::ListType,
types::{ndarray::NDArrayType, ListType},
values::{
ArrayLikeIndexer, ArrayLikeValue, ListValue, NDArrayValue, ProxyValue, RangeValue,
ndarray::{NDArrayValue, RustNDIndex},
ArrayLikeIndexer, ArrayLikeValue, ListValue, ProxyValue, RangeValue,
TypedArrayLikeAccessor, UntypedArrayLikeAccessor,
},
CodeGenContext, CodeGenTask, CodeGenerator,
@ -42,8 +43,8 @@ use super::{
use crate::{
symbol_resolver::{SymbolValue, ValueEnum},
toplevel::{
helper::PrimDef,
numpy::{make_ndarray_ty, unpack_ndarray_var_tys},
helper::{extract_ndims, PrimDef},
numpy::unpack_ndarray_var_tys,
DefinitionId, TopLevelDef,
},
typecheck::{
@ -1553,8 +1554,6 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
} else if ty1.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id())
|| ty2.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id())
{
let llvm_usize = generator.get_size_type(ctx.ctx);
let is_ndarray1 = ty1.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
let is_ndarray2 = ty2.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id());
@ -1564,21 +1563,10 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
assert!(ctx.unifier.unioned(ndarray_dtype1, ndarray_dtype2));
let llvm_ndarray_dtype1 = ctx.get_llvm_type(generator, ndarray_dtype1);
let llvm_ndarray_dtype2 = ctx.get_llvm_type(generator, ndarray_dtype2);
let left_val = NDArrayValue::from_pointer_value(
left_val.into_pointer_value(),
llvm_ndarray_dtype1,
llvm_usize,
None,
);
let right_val = NDArrayValue::from_pointer_value(
right_val.into_pointer_value(),
llvm_ndarray_dtype2,
llvm_usize,
None,
);
let left_val = NDArrayType::from_unifier_type(generator, ctx, ty1)
.map_value(left_val.into_pointer_value(), None);
let right_val = NDArrayType::from_unifier_type(generator, ctx, ty2)
.map_value(right_val.into_pointer_value(), None);
let res = if op.base == Operator::MatMult {
// MatMult is the only binop which is not an elementwise op
@ -1627,13 +1615,12 @@ pub fn gen_binop_expr_with_values<'ctx, G: CodeGenerator>(
} else {
let (ndarray_dtype, _) =
unpack_ndarray_var_tys(&mut ctx.unifier, if is_ndarray1 { ty1 } else { ty2 });
let llvm_ndarray_dtype = ctx.get_llvm_type(generator, ndarray_dtype);
let ndarray_val = NDArrayValue::from_pointer_value(
if is_ndarray1 { left_val } else { right_val }.into_pointer_value(),
llvm_ndarray_dtype,
llvm_usize,
None,
);
let ndarray_val =
NDArrayType::from_unifier_type(generator, ctx, if is_ndarray1 { ty1 } else { ty2 })
.map_value(
if is_ndarray1 { left_val } else { right_val }.into_pointer_value(),
None,
);
let res = numpy::ndarray_elementwise_binop_impl(
generator,
ctx,
@ -1821,16 +1808,10 @@ pub fn gen_unaryop_expr_with_values<'ctx, G: CodeGenerator>(
_ => val.into(),
}
} else if ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id()) {
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_ndarray_ty = NDArrayType::from_unifier_type(generator, ctx, ty);
let (ndarray_dtype, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
let llvm_ndarray_dtype = ctx.get_llvm_type(generator, ndarray_dtype);
let val = NDArrayValue::from_pointer_value(
val.into_pointer_value(),
llvm_ndarray_dtype,
llvm_usize,
None,
);
let val = llvm_ndarray_ty.map_value(val.into_pointer_value(), None);
// ndarray uses `~` rather than `not` to perform elementwise inversion, convert it before
// passing it to the elementwise codegen function
@ -1904,8 +1885,6 @@ pub fn gen_cmpop_expr_with_values<'ctx, G: CodeGenerator>(
if left_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id())
|| right_ty.obj_id(&ctx.unifier).is_some_and(|id| id == PrimDef::NDArray.id())
{
let llvm_usize = generator.get_size_type(ctx.ctx);
let (Some(left_ty), lhs) = left else { codegen_unreachable!(ctx) };
let (Some(right_ty), rhs) = comparators[0] else { codegen_unreachable!(ctx) };
let op = ops[0];
@ -1921,14 +1900,8 @@ pub fn gen_cmpop_expr_with_values<'ctx, G: CodeGenerator>(
assert!(ctx.unifier.unioned(ndarray_dtype1, ndarray_dtype2));
let llvm_ndarray_dtype1 = ctx.get_llvm_type(generator, ndarray_dtype1);
let left_val = NDArrayValue::from_pointer_value(
lhs.into_pointer_value(),
llvm_ndarray_dtype1,
llvm_usize,
None,
);
let left_val = NDArrayType::from_unifier_type(generator, ctx, left_ty)
.map_value(lhs.into_pointer_value(), None);
let res = numpy::ndarray_elementwise_binop_impl(
generator,
ctx,
@ -2549,7 +2522,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
ty: Type,
ndims: Type,
ndims_ty: Type,
v: NDArrayValue<'ctx>,
slice: &Expr<Option<Type>>,
) -> Result<Option<ValueEnum<'ctx>>, String> {
@ -2557,7 +2530,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
let llvm_i32 = ctx.ctx.i32_type();
let llvm_usize = generator.get_size_type(ctx.ctx);
let TypeEnum::TLiteral { values, .. } = &*ctx.unifier.get_ty_immutable(ndims) else {
let TypeEnum::TLiteral { values, .. } = &*ctx.unifier.get_ty_immutable(ndims_ty) else {
codegen_unreachable!(ctx)
};
@ -2590,14 +2563,6 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
_ => 1,
};
let ndarray_ndims_ty = ctx.unifier.get_fresh_literal(
ndims.iter().map(|v| SymbolValue::U64(v - subscripted_dims)).collect(),
None,
);
let ndarray_ty =
make_ndarray_ty(&mut ctx.unifier, &ctx.primitives, Some(ty), Some(ndarray_ndims_ty));
let llvm_pndarray_t = ctx.get_llvm_type(generator, ndarray_ty).into_pointer_type();
let llvm_ndarray_t = llvm_pndarray_t.get_element_type().into_struct_type();
let llvm_ndarray_data_t = ctx.get_llvm_type(generator, ty).as_basic_type_enum();
let sizeof_elem = llvm_ndarray_data_t.size_of().unwrap();
@ -2789,31 +2754,15 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
_ => {
// Accessing an element from a multi-dimensional `ndarray`
let Some(index_addr) = make_indices_arr(generator, ctx)? else { return Ok(None) };
let num_dims = extract_ndims(&ctx.unifier, ndims_ty) - 1;
// Create a new array, remove the top dimension from the dimension-size-list, and copy the
// elements over
let subscripted_ndarray =
generator.gen_var_alloc(ctx, llvm_ndarray_t.into(), None)?;
let ndarray = NDArrayValue::from_pointer_value(
subscripted_ndarray,
llvm_ndarray_data_t,
llvm_usize,
None,
);
let num_dims = v.load_ndims(ctx);
ndarray.store_ndims(
ctx,
generator,
ctx.builder
.build_int_sub(num_dims, llvm_usize.const_int(1, false), "")
.unwrap(),
);
let ndarray_num_dims = ndarray.load_ndims(ctx);
ndarray.create_shape(ctx, llvm_usize, ndarray_num_dims);
let ndarray =
NDArrayType::new(generator, ctx.ctx, llvm_ndarray_data_t, Some(num_dims))
.construct_uninitialized(generator, ctx, None);
let ndarray_num_dims = ctx
.builder
@ -2842,7 +2791,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
llvm_i1.const_zero(),
);
let ndarray_num_elems = call_ndarray_calc_size(
let ndarray_num_elems = ndarray::call_ndarray_calc_size(
generator,
ctx,
&ndarray.shape().as_slice_value(ctx, generator),
@ -2852,7 +2801,7 @@ fn gen_ndarray_subscript_expr<'ctx, G: CodeGenerator>(
.builder
.build_int_z_extend_or_bit_cast(ndarray_num_elems, sizeof_elem.get_type(), "")
.unwrap();
ndarray.create_data(ctx, llvm_ndarray_data_t, ndarray_num_elems);
unsafe { ndarray.create_data(generator, ctx) };
let v_data_src_ptr = v.data().ptr_offset(ctx, generator, &index_addr, None);
call_memcpy_generic(
@ -3537,19 +3486,22 @@ pub fn gen_expr<'ctx, G: CodeGenerator>(
v.data().get(ctx, generator, &index, None).into()
}
}
TypeEnum::TObj { obj_id, params, .. } if *obj_id == PrimDef::NDArray.id() => {
let (ty, ndims) = params.iter().map(|(_, ty)| ty).collect_tuple().unwrap();
let llvm_ty = ctx.get_llvm_type(generator, *ty);
let v = if let Some(v) = generator.gen_expr(ctx, value)? {
v.to_basic_value_enum(ctx, generator, value.custom.unwrap())?
.into_pointer_value()
} else {
TypeEnum::TObj { obj_id, .. } if *obj_id == PrimDef::NDArray.id() => {
let Some(ndarray) = generator.gen_expr(ctx, value)? else {
return Ok(None);
};
let v = NDArrayValue::from_pointer_value(v, llvm_ty, usize, None);
return gen_ndarray_subscript_expr(generator, ctx, *ty, *ndims, v, slice);
let ndarray_ty = value.custom.unwrap();
let ndarray = ndarray.to_basic_value_enum(ctx, generator, ndarray_ty)?;
let ndarray = NDArrayType::from_unifier_type(generator, ctx, ndarray_ty)
.map_value(ndarray.into_pointer_value(), None);
let indices = RustNDIndex::from_subscript_expr(generator, ctx, slice)?;
let result = ndarray
.index(generator, ctx, &indices)
.split_unsized(generator, ctx)
.to_basic_value_enum();
return Ok(Some(ValueEnum::Dynamic(result)));
}
TypeEnum::TTuple { .. } => {
let index: u32 =
@ -3598,3 +3550,97 @@ pub fn gen_expr<'ctx, G: CodeGenerator>(
_ => unimplemented!(),
}))
}
/// Creates a function in the current module and inserts a `call` instruction into the LLVM IR.
#[allow(clippy::too_many_arguments)]
pub fn create_fn_and_call<'ctx>(
ctx: &CodeGenContext<'ctx, '_>,
fn_name: &str,
ret_type: Option<BasicTypeEnum<'ctx>>,
params: &[BasicTypeEnum<'ctx>],
args: &[BasicValueEnum<'ctx>],
is_var_args: bool,
call_value_name: Option<&str>,
configure: Option<&dyn Fn(&FunctionValue<'ctx>)>,
) -> Option<BasicValueEnum<'ctx>> {
let intrinsic_fn = ctx.module.get_function(fn_name).unwrap_or_else(|| {
let params = params.iter().copied().map(BasicTypeEnum::into).collect_vec();
let fn_type = if let Some(ret_type) = ret_type {
ret_type.fn_type(params.as_slice(), is_var_args)
} else {
ctx.ctx.void_type().fn_type(params.as_slice(), is_var_args)
};
ctx.module.add_function(fn_name, fn_type, None)
});
if let Some(configure) = configure {
configure(&intrinsic_fn);
}
let args = args.iter().copied().map(BasicValueEnum::into).collect_vec();
ctx.builder
.build_call(intrinsic_fn, args.as_slice(), call_value_name.unwrap_or_default())
.map(CallSiteValue::try_as_basic_value)
.map(Either::left)
.unwrap()
}
/// Creates a function in the current module and inserts a `call` instruction into the LLVM IR.
///
/// This is a wrapper around [`create_fn_and_call`] for non-vararg function. This function allows
/// parameters and arguments to be specified as tuples to better indicate the expected type and
/// actual value of each parameter-argument pair of the call.
pub fn create_and_call_function<'ctx>(
ctx: &CodeGenContext<'ctx, '_>,
fn_name: &str,
ret_type: Option<BasicTypeEnum<'ctx>>,
params: &[(BasicTypeEnum<'ctx>, BasicValueEnum<'ctx>)],
value_name: Option<&str>,
configure: Option<&dyn Fn(&FunctionValue<'ctx>)>,
) -> Option<BasicValueEnum<'ctx>> {
let param_tys = params.iter().map(|(ty, _)| ty).copied().map(BasicTypeEnum::into).collect_vec();
let arg_values =
params.iter().map(|(_, value)| value).copied().map(BasicValueEnum::into).collect_vec();
create_fn_and_call(
ctx,
fn_name,
ret_type,
param_tys.as_slice(),
arg_values.as_slice(),
false,
value_name,
configure,
)
}
/// Creates a function in the current module and inserts a `call` instruction into the LLVM IR.
///
/// This is a wrapper around [`create_fn_and_call`] for non-vararg function. This function allows
/// only arguments to be specified and performs inference for the parameter types of the function
/// using [`BasicValueEnum::get_type`] on the arguments.
///
/// This function is recommended if it is known that all function arguments match the parameter
/// types of the invoked function.
pub fn infer_and_call_function<'ctx>(
ctx: &CodeGenContext<'ctx, '_>,
fn_name: &str,
ret_type: Option<BasicTypeEnum<'ctx>>,
args: &[BasicValueEnum<'ctx>],
value_name: Option<&str>,
configure: Option<&dyn Fn(&FunctionValue<'ctx>)>,
) -> Option<BasicValueEnum<'ctx>> {
let param_tys = args.iter().map(BasicValueEnum::get_type).collect_vec();
create_fn_and_call(
ctx,
fn_name,
ret_type,
param_tys.as_slice(),
args,
false,
value_name,
configure,
)
}

View File

@ -13,12 +13,13 @@ use super::{CodeGenContext, CodeGenerator};
use crate::{symbol_resolver::SymbolResolver, typecheck::typedef::Type};
pub use list::*;
pub use math::*;
pub use ndarray::*;
pub use range::*;
pub use slice::*;
mod list;
mod math;
mod ndarray;
pub mod ndarray;
mod range;
mod slice;
#[must_use]
@ -60,6 +61,27 @@ pub fn load_irrt<'ctx>(ctx: &'ctx Context, symbol_resolver: &dyn SymbolResolver)
irrt_mod
}
/// Returns the name of a function which contains variants for 32-bit and 64-bit `size_t`.
///
/// - When [`TypeContext::size_type`] is 32-bits, the function name is `fn_name}`.
/// - When [`TypeContext::size_type`] is 64-bits, the function name is `{fn_name}64`.
#[must_use]
pub fn get_usize_dependent_function_name<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'_, '_>,
name: &str,
) -> String {
let mut name = name.to_owned();
match generator.get_size_type(ctx.ctx).get_bit_width() {
32 => {}
64 => name.push_str("64"),
bit_width => {
panic!("Unsupported int type bit width {bit_width}, must be either 32-bits or 64-bits")
}
}
name
}
/// NOTE: the output value of the end index of this function should be compared ***inclusively***,
/// because python allows `a[2::-1]`, whose semantic is `[a[2], a[1], a[0]]`, which is equivalent to
/// NO numeric slice in python.

View File

@ -0,0 +1,250 @@
use inkwell::{
values::{BasicValueEnum, IntValue, PointerValue},
AddressSpace,
};
use crate::codegen::{
expr::{create_and_call_function, infer_and_call_function},
irrt::get_usize_dependent_function_name,
types::ProxyType,
values::{ndarray::NDArrayValue, ProxyValue},
CodeGenContext, CodeGenerator,
};
pub fn call_nac3_ndarray_util_assert_shape_no_negative<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndims: IntValue<'ctx>,
shape: PointerValue<'ctx>,
) {
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
let name = get_usize_dependent_function_name(
generator,
ctx,
"__nac3_ndarray_util_assert_shape_no_negative",
);
create_and_call_function(
ctx,
&name,
Some(llvm_usize.into()),
&[(llvm_usize.into(), ndims.into()), (llvm_pusize.into(), shape.into())],
None,
None,
);
}
pub fn call_nac3_ndarray_util_assert_output_shape_same<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray_ndims: IntValue<'ctx>,
ndarray_shape: PointerValue<'ctx>,
output_ndims: IntValue<'ctx>,
output_shape: IntValue<'ctx>,
) {
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
let name = get_usize_dependent_function_name(
generator,
ctx,
"__nac3_ndarray_util_assert_output_shape_same",
);
create_and_call_function(
ctx,
&name,
Some(llvm_usize.into()),
&[
(llvm_usize.into(), ndarray_ndims.into()),
(llvm_pusize.into(), ndarray_shape.into()),
(llvm_usize.into(), output_ndims.into()),
(llvm_pusize.into(), output_shape.into()),
],
None,
None,
);
}
pub fn call_nac3_ndarray_size<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
) -> IntValue<'ctx> {
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_ndarray = ndarray.get_type().as_base_type();
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_size");
create_and_call_function(
ctx,
&name,
Some(llvm_usize.into()),
&[(llvm_ndarray.into(), ndarray.as_base_value().into())],
Some("size"),
None,
)
.map(BasicValueEnum::into_int_value)
.unwrap()
}
pub fn call_nac3_ndarray_nbytes<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
) -> IntValue<'ctx> {
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_ndarray = ndarray.get_type().as_base_type();
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_nbytes");
create_and_call_function(
ctx,
&name,
Some(llvm_usize.into()),
&[(llvm_ndarray.into(), ndarray.as_base_value().into())],
Some("nbytes"),
None,
)
.map(BasicValueEnum::into_int_value)
.unwrap()
}
pub fn call_nac3_ndarray_len<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
) -> IntValue<'ctx> {
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_ndarray = ndarray.get_type().as_base_type();
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_len");
create_and_call_function(
ctx,
&name,
Some(llvm_usize.into()),
&[(llvm_ndarray.into(), ndarray.as_base_value().into())],
Some("len"),
None,
)
.map(BasicValueEnum::into_int_value)
.unwrap()
}
pub fn call_nac3_ndarray_is_c_contiguous<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
) -> IntValue<'ctx> {
let llvm_i1 = ctx.ctx.bool_type();
let llvm_ndarray = ndarray.get_type().as_base_type();
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_is_c_contiguous");
create_and_call_function(
ctx,
&name,
Some(llvm_i1.into()),
&[(llvm_ndarray.into(), ndarray.as_base_value().into())],
Some("is_c_contiguous"),
None,
)
.map(BasicValueEnum::into_int_value)
.unwrap()
}
pub fn call_nac3_ndarray_get_nth_pelement<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
index: IntValue<'ctx>,
) -> PointerValue<'ctx> {
let llvm_i8 = ctx.ctx.i8_type();
let llvm_pi8 = llvm_i8.ptr_type(AddressSpace::default());
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_ndarray = ndarray.get_type().as_base_type();
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_get_nth_pelement");
create_and_call_function(
ctx,
&name,
Some(llvm_pi8.into()),
&[(llvm_ndarray.into(), ndarray.as_base_value().into()), (llvm_usize.into(), index.into())],
Some("pelement"),
None,
)
.map(BasicValueEnum::into_pointer_value)
.unwrap()
}
pub fn call_nac3_ndarray_get_pelement_by_indices<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
indices: PointerValue<'ctx>,
) -> PointerValue<'ctx> {
let llvm_i8 = ctx.ctx.i8_type();
let llvm_pi8 = llvm_i8.ptr_type(AddressSpace::default());
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
let llvm_ndarray = ndarray.get_type().as_base_type();
let name =
get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_get_pelement_by_indices");
create_and_call_function(
ctx,
&name,
Some(llvm_pi8.into()),
&[
(llvm_ndarray.into(), ndarray.as_base_value().into()),
(llvm_pusize.into(), indices.into()),
],
Some("pelement"),
None,
)
.map(BasicValueEnum::into_pointer_value)
.unwrap()
}
pub fn call_nac3_ndarray_set_strides_by_shape<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
) {
let llvm_ndarray = ndarray.get_type().as_base_type();
let name =
get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_set_strides_by_shape");
create_and_call_function(
ctx,
&name,
None,
&[(llvm_ndarray.into(), ndarray.as_base_value().into())],
None,
None,
);
}
pub fn call_nac3_ndarray_copy_data<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
src_ndarray: NDArrayValue<'ctx>,
dst_ndarray: NDArrayValue<'ctx>,
) {
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_copy_data");
infer_and_call_function(
ctx,
&name,
None,
&[src_ndarray.as_base_value().into(), dst_ndarray.as_base_value().into()],
None,
None,
);
}

View File

@ -0,0 +1,29 @@
use crate::codegen::{
expr::infer_and_call_function,
irrt::get_usize_dependent_function_name,
values::{ndarray::NDArrayValue, ArrayLikeValue, ArraySliceValue, ProxyValue},
CodeGenContext, CodeGenerator,
};
pub fn call_nac3_ndarray_index<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
indices: ArraySliceValue<'ctx>,
src_ndarray: NDArrayValue<'ctx>,
dst_ndarray: NDArrayValue<'ctx>,
) {
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_ndarray_index");
infer_and_call_function(
ctx,
&name,
None,
&[
indices.size(ctx, generator).into(),
indices.base_ptr(ctx, generator).into(),
src_ndarray.as_base_value().into(),
dst_ndarray.as_base_value().into(),
],
None,
None,
);
}

View File

@ -0,0 +1,70 @@
use inkwell::{
values::{BasicValueEnum, IntValue},
AddressSpace,
};
use crate::codegen::{
expr::{create_and_call_function, infer_and_call_function},
irrt::get_usize_dependent_function_name,
types::ProxyType,
values::{
ndarray::{NDArrayValue, NDIterValue},
ArrayLikeValue, ArraySliceValue, ProxyValue,
},
CodeGenContext, CodeGenerator,
};
pub fn call_nac3_nditer_initialize<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
iter: NDIterValue<'ctx>,
ndarray: NDArrayValue<'ctx>,
indices: ArraySliceValue<'ctx>,
) {
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_pusize = llvm_usize.ptr_type(AddressSpace::default());
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_nditer_initialize");
create_and_call_function(
ctx,
&name,
None,
&[
(iter.get_type().as_base_type().into(), iter.as_base_value().into()),
(ndarray.get_type().as_base_type().into(), ndarray.as_base_value().into()),
(llvm_pusize.into(), indices.base_ptr(ctx, generator).into()),
],
None,
None,
);
}
pub fn call_nac3_nditer_has_element<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
iter: NDIterValue<'ctx>,
) -> IntValue<'ctx> {
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_nditer_has_element");
infer_and_call_function(
ctx,
&name,
Some(ctx.ctx.bool_type().into()),
&[iter.as_base_value().into()],
None,
None,
)
.map(BasicValueEnum::into_int_value)
.unwrap()
}
pub fn call_nac3_nditer_next<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
iter: NDIterValue<'ctx>,
) {
let name = get_usize_dependent_function_name(generator, ctx, "__nac3_nditer_next");
infer_and_call_function(ctx, &name, None, &[iter.as_base_value().into()], None, None);
}

View File

@ -10,11 +10,18 @@ use crate::codegen::{
macros::codegen_unreachable,
stmt::gen_for_callback_incrementing,
values::{
ArrayLikeIndexer, ArrayLikeValue, ArraySliceValue, NDArrayValue, TypedArrayLikeAccessor,
TypedArrayLikeAdapter, UntypedArrayLikeAccessor,
ndarray::NDArrayValue, ArrayLikeIndexer, ArrayLikeValue, ArraySliceValue,
TypedArrayLikeAccessor, TypedArrayLikeAdapter, UntypedArrayLikeAccessor,
},
CodeGenContext, CodeGenerator,
};
pub use basic::*;
pub use indexing::*;
pub use iter::*;
mod basic;
mod indexing;
mod iter;
/// Generates a call to `__nac3_ndarray_calc_size`. Returns an [`IntValue`] representing the
/// calculated total size.
@ -77,7 +84,7 @@ where
/// `NDArray`.
pub fn call_ndarray_calc_nd_indices<'ctx, G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &mut CodeGenContext<'ctx, '_>,
ctx: &CodeGenContext<'ctx, '_>,
index: IntValue<'ctx>,
ndarray: NDArrayValue<'ctx>,
) -> TypedArrayLikeAdapter<'ctx, IntValue<'ctx>> {
@ -201,8 +208,8 @@ where
/// `NDArray`.
/// * `indices` - The multidimensional index to compute the flattened index for.
pub fn call_ndarray_flatten_index<'ctx, G, Index>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
indices: &Index,
) -> IntValue<'ctx>

View File

@ -0,0 +1,42 @@
use inkwell::{
values::{BasicValueEnum, CallSiteValue, IntValue},
IntPredicate,
};
use itertools::Either;
use crate::codegen::{CodeGenContext, CodeGenerator};
pub fn calculate_len_for_slice_range<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
start: IntValue<'ctx>,
end: IntValue<'ctx>,
step: IntValue<'ctx>,
) -> IntValue<'ctx> {
const SYMBOL: &str = "__nac3_range_slice_len";
let len_func = ctx.module.get_function(SYMBOL).unwrap_or_else(|| {
let i32_t = ctx.ctx.i32_type();
let fn_t = i32_t.fn_type(&[i32_t.into(), i32_t.into(), i32_t.into()], false);
ctx.module.add_function(SYMBOL, fn_t, None)
});
// assert step != 0, throw exception if not
let not_zero = ctx
.builder
.build_int_compare(IntPredicate::NE, step, step.get_type().const_zero(), "range_step_ne")
.unwrap();
ctx.make_assert(
generator,
not_zero,
"0:ValueError",
"step must not be zero",
[None, None, None],
ctx.current_loc,
);
ctx.builder
.build_call(len_func, &[start.into(), end.into(), step.into()], "calc_len")
.map(CallSiteValue::try_as_basic_value)
.map(|v| v.map_left(BasicValueEnum::into_int_value))
.map(Either::unwrap_left)
.unwrap()
}

View File

@ -1,8 +1,6 @@
use inkwell::{
values::{BasicValueEnum, CallSiteValue, IntValue},
IntPredicate,
};
use inkwell::values::{BasicValueEnum, CallSiteValue, IntValue};
use itertools::Either;
use nac3parser::ast::Expr;
use crate::{
@ -39,38 +37,3 @@ pub fn handle_slice_index_bound<'ctx, G: CodeGenerator>(
.unwrap(),
))
}
pub fn calculate_len_for_slice_range<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
start: IntValue<'ctx>,
end: IntValue<'ctx>,
step: IntValue<'ctx>,
) -> IntValue<'ctx> {
const SYMBOL: &str = "__nac3_range_slice_len";
let len_func = ctx.module.get_function(SYMBOL).unwrap_or_else(|| {
let i32_t = ctx.ctx.i32_type();
let fn_t = i32_t.fn_type(&[i32_t.into(), i32_t.into(), i32_t.into()], false);
ctx.module.add_function(SYMBOL, fn_t, None)
});
// assert step != 0, throw exception if not
let not_zero = ctx
.builder
.build_int_compare(IntPredicate::NE, step, step.get_type().const_zero(), "range_step_ne")
.unwrap();
ctx.make_assert(
generator,
not_zero,
"0:ValueError",
"step must not be zero",
[None, None, None],
ctx.current_loc,
);
ctx.builder
.build_call(len_func, &[start.into(), end.into(), step.into()], "calc_len")
.map(CallSiteValue::try_as_basic_value)
.map(|v| v.map_left(BasicValueEnum::into_int_value))
.map(Either::unwrap_left)
.unwrap()
}

View File

@ -201,6 +201,49 @@ pub fn call_memcpy_generic<'ctx>(
call_memcpy(ctx, dest, src, len, is_volatile);
}
/// Invokes the `llvm.memcpy` intrinsic.
///
/// Unlike [`call_memcpy`], this function accepts any type of pointer value. If `dest` or `src` is
/// not a pointer to an integer, the pointer(s) will be cast to `i8*` before invoking `memcpy`.
/// Moreover, `len` now refers to the number of elements to copy (rather than number of bytes to
/// copy).
pub fn call_memcpy_generic_array<'ctx>(
ctx: &CodeGenContext<'ctx, '_>,
dest: PointerValue<'ctx>,
src: PointerValue<'ctx>,
len: IntValue<'ctx>,
is_volatile: IntValue<'ctx>,
) {
let llvm_i8 = ctx.ctx.i8_type();
let llvm_p0i8 = llvm_i8.ptr_type(AddressSpace::default());
let llvm_sizeof_expr_t = llvm_i8.size_of().get_type();
let dest_elem_t = dest.get_type().get_element_type();
let src_elem_t = src.get_type().get_element_type();
let dest = if matches!(dest_elem_t, IntType(t) if t.get_bit_width() == 8) {
dest
} else {
ctx.builder
.build_bit_cast(dest, llvm_p0i8, "")
.map(BasicValueEnum::into_pointer_value)
.unwrap()
};
let src = if matches!(src_elem_t, IntType(t) if t.get_bit_width() == 8) {
src
} else {
ctx.builder
.build_bit_cast(src, llvm_p0i8, "")
.map(BasicValueEnum::into_pointer_value)
.unwrap()
};
let len = ctx.builder.build_int_z_extend_or_bit_cast(len, llvm_sizeof_expr_t, "").unwrap();
let len = ctx.builder.build_int_mul(len, src_elem_t.size_of().unwrap(), "").unwrap();
call_memcpy(ctx, dest, src, len, is_volatile);
}
/// Macro to find and generate build call for llvm intrinsic (body of llvm intrinsic function)
///
/// Arguments:
@ -343,3 +386,25 @@ pub fn call_float_powi<'ctx>(
.map(Either::unwrap_left)
.unwrap()
}
/// Invokes the [`llvm.ctpop`](https://llvm.org/docs/LangRef.html#llvm-ctpop-intrinsic) intrinsic.
pub fn call_int_ctpop<'ctx>(
ctx: &CodeGenContext<'ctx, '_>,
src: IntValue<'ctx>,
name: Option<&str>,
) -> IntValue<'ctx> {
const FN_NAME: &str = "llvm.ctpop";
let llvm_src_t = src.get_type();
let intrinsic_fn = Intrinsic::find(FN_NAME)
.and_then(|intrinsic| intrinsic.get_declaration(&ctx.module, &[llvm_src_t.into()]))
.unwrap();
ctx.builder
.build_call(intrinsic_fn, &[src.into()], name.unwrap_or_default())
.map(CallSiteValue::try_as_basic_value)
.map(|v| v.map_left(BasicValueEnum::into_int_value))
.map(Either::unwrap_left)
.unwrap()
}

View File

@ -30,7 +30,11 @@ use nac3parser::ast::{Location, Stmt, StrRef};
use crate::{
symbol_resolver::{StaticValue, SymbolResolver},
toplevel::{helper::PrimDef, numpy::unpack_ndarray_var_tys, TopLevelContext, TopLevelDef},
toplevel::{
helper::{extract_ndims, PrimDef},
numpy::unpack_ndarray_var_tys,
TopLevelContext, TopLevelDef,
},
typecheck::{
type_inferencer::{CodeLocation, PrimitiveStore},
typedef::{CallId, FuncArg, Type, TypeEnum, Unifier},
@ -38,7 +42,7 @@ use crate::{
};
use concrete_type::{ConcreteType, ConcreteTypeEnum, ConcreteTypeStore};
pub use generator::{CodeGenerator, DefaultCodeGenerator};
use types::{ListType, NDArrayType, ProxyType, RangeType};
use types::{ndarray::NDArrayType, ListType, ProxyType, RangeType};
pub mod builtin_fns;
pub mod concrete_type;
@ -510,12 +514,13 @@ fn get_llvm_type<'ctx, G: CodeGenerator + ?Sized>(
}
TObj { obj_id, .. } if *obj_id == PrimDef::NDArray.id() => {
let (dtype, _) = unpack_ndarray_var_tys(unifier, ty);
let (dtype, ndims) = unpack_ndarray_var_tys(unifier, ty);
let ndims = extract_ndims(unifier, ndims);
let element_type = get_llvm_type(
ctx, module, generator, unifier, top_level, type_cache, dtype,
);
NDArrayType::new(generator, ctx, element_type).as_base_type().into()
NDArrayType::new(generator, ctx, element_type, Some(ndims)).as_base_type().into()
}
_ => unreachable!(
@ -1119,3 +1124,106 @@ fn gen_in_range_check<'ctx>(
fn get_va_count_arg_name(arg_name: StrRef) -> StrRef {
format!("__{}_va_count", &arg_name).into()
}
/// Returns the alignment of the type.
///
/// This is necessary as `get_alignment` is not implemented as part of [`BasicType`].
pub fn get_type_alignment<'ctx>(ty: impl Into<BasicTypeEnum<'ctx>>) -> IntValue<'ctx> {
match ty.into() {
BasicTypeEnum::ArrayType(ty) => ty.get_alignment(),
BasicTypeEnum::FloatType(ty) => ty.get_alignment(),
BasicTypeEnum::IntType(ty) => ty.get_alignment(),
BasicTypeEnum::PointerType(ty) => ty.get_alignment(),
BasicTypeEnum::StructType(ty) => ty.get_alignment(),
BasicTypeEnum::VectorType(ty) => ty.get_alignment(),
}
}
/// Inserts an `alloca` instruction with allocation `size` given in bytes and the alignment of the
/// given type.
///
/// The returned [`PointerValue`] will have a type of `i8*`, a size of at least `size`, and will be
/// aligned with the alignment of `align_ty`.
pub fn type_aligned_alloca<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
align_ty: impl Into<BasicTypeEnum<'ctx>>,
size: IntValue<'ctx>,
name: Option<&str>,
) -> PointerValue<'ctx> {
/// Round `val` up to its modulo `power_of_two`.
fn round_up<'ctx>(
ctx: &CodeGenContext<'ctx, '_>,
val: IntValue<'ctx>,
power_of_two: IntValue<'ctx>,
) -> IntValue<'ctx> {
debug_assert_eq!(
val.get_type().get_bit_width(),
power_of_two.get_type().get_bit_width(),
"`val` ({}) and `power_of_two` ({}) must be the same type",
val.get_type(),
power_of_two.get_type(),
);
let llvm_val_t = val.get_type();
let max_rem =
ctx.builder.build_int_sub(power_of_two, llvm_val_t.const_int(1, false), "").unwrap();
ctx.builder
.build_and(
ctx.builder.build_int_add(val, max_rem, "").unwrap(),
ctx.builder.build_not(max_rem, "").unwrap(),
"",
)
.unwrap()
}
let llvm_i8 = ctx.ctx.i8_type();
let llvm_pi8 = llvm_i8.ptr_type(AddressSpace::default());
let llvm_usize = generator.get_size_type(ctx.ctx);
let align_ty = align_ty.into();
let size = ctx.builder.build_int_truncate_or_bit_cast(size, llvm_usize, "").unwrap();
debug_assert_eq!(
size.get_type().get_bit_width(),
llvm_usize.get_bit_width(),
"Expected size_t ({}) for parameter `size` of `aligned_alloca`, got {}",
llvm_usize,
size.get_type(),
);
let alignment = get_type_alignment(align_ty);
let alignment = ctx.builder.build_int_truncate_or_bit_cast(alignment, llvm_usize, "").unwrap();
if ctx.registry.llvm_options.opt_level == OptimizationLevel::None {
let alignment_bitcount = llvm_intrinsics::call_int_ctpop(ctx, alignment, None);
ctx.make_assert(
generator,
ctx.builder
.build_int_compare(
IntPredicate::EQ,
alignment_bitcount,
alignment_bitcount.get_type().const_int(1, false),
"",
)
.unwrap(),
"0:AssertionError",
"Expected power-of-two alignment for aligned_alloca, got {0}",
[Some(alignment), None, None],
ctx.current_loc,
);
}
let buffer_size = round_up(ctx, size, alignment);
let aligned_slices = ctx.builder.build_int_unsigned_div(buffer_size, alignment, "").unwrap();
// Just to be absolutely sure, alloca in [i8 x alignment] slices
let buffer = ctx.builder.build_array_alloca(align_ty, aligned_slices, "").unwrap();
ctx.builder
.build_bit_cast(buffer, llvm_pi8, name.unwrap_or_default())
.map(BasicValueEnum::into_pointer_value)
.unwrap()
}

View File

@ -3,21 +3,25 @@ use inkwell::{
values::{BasicValue, BasicValueEnum, IntValue, PointerValue},
AddressSpace, IntPredicate, OptimizationLevel,
};
use itertools::Itertools;
use nac3parser::ast::{Operator, StrRef};
use super::{
expr::gen_binop_expr_with_values,
irrt::{
calculate_len_for_slice_range, call_ndarray_calc_broadcast,
call_ndarray_calc_broadcast_index, call_ndarray_calc_nd_indices, call_ndarray_calc_size,
calculate_len_for_slice_range,
ndarray::{
call_ndarray_calc_broadcast, call_ndarray_calc_broadcast_index,
call_ndarray_calc_nd_indices, call_ndarray_calc_size,
},
},
llvm_intrinsics::{self, call_memcpy_generic},
macros::codegen_unreachable,
stmt::{gen_for_callback_incrementing, gen_for_range_callback, gen_if_else_expr_callback},
types::{ListType, NDArrayType, ProxyType},
types::{ndarray::NDArrayType, ListType, ProxyType},
values::{
ArrayLikeIndexer, ArrayLikeValue, ListValue, NDArrayValue, ProxyValue,
ndarray::NDArrayValue, ArrayLikeIndexer, ArrayLikeValue, ListValue, ProxyValue,
TypedArrayLikeAccessor, TypedArrayLikeAdapter, TypedArrayLikeMutator,
UntypedArrayLikeAccessor, UntypedArrayLikeMutator,
},
@ -25,39 +29,13 @@ use super::{
};
use crate::{
symbol_resolver::ValueEnum,
toplevel::{
helper::{arraylike_flatten_element_type, PrimDef},
numpy::{make_ndarray_ty, unpack_ndarray_var_tys},
DefinitionId,
},
toplevel::{helper::PrimDef, numpy::unpack_ndarray_var_tys, DefinitionId},
typecheck::{
magic_methods::Binop,
typedef::{FunSignature, Type, TypeEnum},
},
};
/// Creates an uninitialized `NDArray` instance.
fn create_ndarray_uninitialized<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
elem_ty: Type,
) -> Result<NDArrayValue<'ctx>, String> {
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
let ndarray_ty = make_ndarray_ty(&mut ctx.unifier, &ctx.primitives, Some(elem_ty), None);
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_ndarray_t = ctx
.get_llvm_type(generator, ndarray_ty)
.into_pointer_type()
.get_element_type()
.into_struct_type();
let ndarray = generator.gen_var_alloc(ctx, llvm_ndarray_t.into(), None)?;
Ok(NDArrayValue::from_pointer_value(ndarray, llvm_elem_ty, llvm_usize, None))
}
/// Creates an `NDArray` instance from a dynamic shape.
///
/// * `elem_ty` - The element type of the `NDArray`.
@ -83,6 +61,7 @@ where
) -> Result<IntValue<'ctx>, String>,
{
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
// Assert that all dimensions are non-negative
let shape_len = shape_len_fn(generator, ctx, shape)?;
@ -115,20 +94,17 @@ where
ctx.current_loc,
);
// TODO: Disallow dim_sz > u32_MAX
// TODO: Disallow shape > u32_MAX
Ok(())
},
llvm_usize.const_int(1, false),
)?;
let ndarray = create_ndarray_uninitialized(generator, ctx, elem_ty)?;
let num_dims = shape_len_fn(generator, ctx, shape)?;
ndarray.store_ndims(ctx, generator, num_dims);
let ndarray_num_dims = ndarray.load_ndims(ctx);
ndarray.create_shape(ctx, llvm_usize, ndarray_num_dims);
let ndarray = NDArrayType::new(generator, ctx.ctx, llvm_elem_ty, None)
.construct_dyn_ndims(generator, ctx, num_dims, None);
// Copy the dimension sizes from shape to ndarray.dims
let shape_len = shape_len_fn(generator, ctx, shape)?;
@ -153,7 +129,7 @@ where
llvm_usize.const_int(1, false),
)?;
let ndarray = ndarray_init_data(generator, ctx, elem_ty, ndarray);
unsafe { ndarray.create_data(generator, ctx) };
Ok(ndarray)
}
@ -186,57 +162,18 @@ pub fn create_ndarray_const_shape<'ctx, G: CodeGenerator + ?Sized>(
ctx.current_loc,
);
// TODO: Disallow dim_sz > u32_MAX
// TODO: Disallow shape > u32_MAX
}
let ndarray = create_ndarray_uninitialized(generator, ctx, elem_ty)?;
let llvm_dtype = ctx.get_llvm_type(generator, elem_ty);
let num_dims = llvm_usize.const_int(shape.len() as u64, false);
ndarray.store_ndims(ctx, generator, num_dims);
let ndarray_num_dims = ndarray.load_ndims(ctx);
ndarray.create_shape(ctx, llvm_usize, ndarray_num_dims);
for (i, &shape_dim) in shape.iter().enumerate() {
let shape_dim = ctx.builder.build_int_z_extend(shape_dim, llvm_usize, "").unwrap();
let ndarray_dim = unsafe {
ndarray.shape().ptr_offset_unchecked(
ctx,
generator,
&llvm_usize.const_int(i as u64, true),
None,
)
};
ctx.builder.build_store(ndarray_dim, shape_dim).unwrap();
}
let ndarray = ndarray_init_data(generator, ctx, elem_ty, ndarray);
let ndarray = NDArrayType::new(generator, ctx.ctx, llvm_dtype, Some(shape.len() as u64))
.construct_dyn_shape(generator, ctx, shape, None);
unsafe { ndarray.create_data(generator, ctx) };
Ok(ndarray)
}
/// Initializes the `data` field of [`NDArrayValue`] based on the `ndims` and `dim_sz` fields.
fn ndarray_init_data<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
elem_ty: Type,
ndarray: NDArrayValue<'ctx>,
) -> NDArrayValue<'ctx> {
let llvm_ndarray_data_t = ctx.get_llvm_type(generator, elem_ty).as_basic_type_enum();
assert!(llvm_ndarray_data_t.is_sized());
let ndarray_num_elems = call_ndarray_calc_size(
generator,
ctx,
&ndarray.shape().as_slice_value(ctx, generator),
(None, None),
);
ndarray.create_data(ctx, llvm_ndarray_data_t, ndarray_num_elems);
ndarray
}
fn ndarray_zero_value<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
@ -338,20 +275,24 @@ fn call_ndarray_empty_impl<'ctx, G: CodeGenerator + ?Sized>(
// Get the length/size of the tuple, which also happens to be the value of `ndims`.
let ndims = shape_tuple.get_type().count_fields();
let mut shape = Vec::with_capacity(ndims as usize);
for dim_i in 0..ndims {
let dim = ctx
.builder
.build_extract_value(shape_tuple, dim_i, format!("dim{dim_i}").as_str())
.unwrap()
.into_int_value();
let shape = (0..ndims)
.map(|dim_i| {
ctx.builder
.build_extract_value(shape_tuple, dim_i, format!("dim{dim_i}").as_str())
.map(BasicValueEnum::into_int_value)
.map(|v| {
ctx.builder.build_int_z_extend_or_bit_cast(v, llvm_usize, "").unwrap()
})
.unwrap()
})
.collect_vec();
shape.push(dim);
}
create_ndarray_const_shape(generator, ctx, elem_ty, shape.as_slice())
}
BasicValueEnum::IntValue(shape_int) => {
// 3. A scalar int; e.g., `np.empty(3)`, this is functionally equivalent to `np.empty([3])`
let shape_int =
ctx.builder.build_int_z_extend_or_bit_cast(shape_int, llvm_usize, "").unwrap();
create_ndarray_const_shape(generator, ctx, elem_ty, &[shape_int])
}
@ -474,8 +415,8 @@ fn ndarray_broadcast_fill<'ctx, 'a, G, ValueFn>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, 'a>,
res: NDArrayValue<'ctx>,
lhs: (Type, BasicValueEnum<'ctx>, bool),
rhs: (Type, BasicValueEnum<'ctx>, bool),
(lhs_ty, lhs_val, lhs_scalar): (Type, BasicValueEnum<'ctx>, bool),
(rhs_ty, rhs_val, rhs_scalar): (Type, BasicValueEnum<'ctx>, bool),
value_fn: ValueFn,
) -> Result<NDArrayValue<'ctx>, String>
where
@ -486,11 +427,6 @@ where
(BasicValueEnum<'ctx>, BasicValueEnum<'ctx>),
) -> Result<BasicValueEnum<'ctx>, String>,
{
let llvm_usize = generator.get_size_type(ctx.ctx);
let (lhs_ty, lhs_val, lhs_scalar) = lhs;
let (rhs_ty, rhs_val, rhs_scalar) = rhs;
assert!(
!(lhs_scalar && rhs_scalar),
"One of the operands must be a ndarray instance: `{}`, `{}`",
@ -500,26 +436,14 @@ where
// Assert that all ndarray operands are broadcastable to the target size
if !lhs_scalar {
let lhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, lhs_ty);
let llvm_lhs_elem_ty = ctx.get_llvm_type(generator, lhs_dtype);
let lhs_val = NDArrayValue::from_pointer_value(
lhs_val.into_pointer_value(),
llvm_lhs_elem_ty,
llvm_usize,
None,
);
let lhs_val = NDArrayType::from_unifier_type(generator, ctx, lhs_ty)
.map_value(lhs_val.into_pointer_value(), None);
ndarray_assert_is_broadcastable(generator, ctx, res, lhs_val);
}
if !rhs_scalar {
let rhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, rhs_ty);
let llvm_rhs_elem_ty = ctx.get_llvm_type(generator, rhs_dtype);
let rhs_val = NDArrayValue::from_pointer_value(
rhs_val.into_pointer_value(),
llvm_rhs_elem_ty,
llvm_usize,
None,
);
let rhs_val = NDArrayType::from_unifier_type(generator, ctx, rhs_ty)
.map_value(rhs_val.into_pointer_value(), None);
ndarray_assert_is_broadcastable(generator, ctx, res, rhs_val);
}
@ -527,14 +451,8 @@ where
let lhs_elem = if lhs_scalar {
lhs_val
} else {
let lhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, lhs_ty);
let llvm_lhs_elem_ty = ctx.get_llvm_type(generator, lhs_dtype);
let lhs = NDArrayValue::from_pointer_value(
lhs_val.into_pointer_value(),
llvm_lhs_elem_ty,
llvm_usize,
None,
);
let lhs = NDArrayType::from_unifier_type(generator, ctx, lhs_ty)
.map_value(lhs_val.into_pointer_value(), None);
let lhs_idx = call_ndarray_calc_broadcast_index(generator, ctx, lhs, idx);
unsafe { lhs.data().get_unchecked(ctx, generator, &lhs_idx, None) }
@ -543,14 +461,8 @@ where
let rhs_elem = if rhs_scalar {
rhs_val
} else {
let rhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, rhs_ty);
let llvm_rhs_elem_ty = ctx.get_llvm_type(generator, rhs_dtype);
let rhs = NDArrayValue::from_pointer_value(
rhs_val.into_pointer_value(),
llvm_rhs_elem_ty,
llvm_usize,
None,
);
let rhs = NDArrayType::from_unifier_type(generator, ctx, rhs_ty)
.map_value(rhs_val.into_pointer_value(), None);
let rhs_idx = call_ndarray_calc_broadcast_index(generator, ctx, rhs, idx);
unsafe { rhs.data().get_unchecked(ctx, generator, &rhs_idx, None) }
@ -704,9 +616,7 @@ fn llvm_arraylike_get_ndims<'ctx, G: CodeGenerator + ?Sized>(
BasicValueEnum::PointerValue(v)
if NDArrayValue::is_representable(v, llvm_usize).is_ok() =>
{
let dtype = arraylike_flatten_element_type(&mut ctx.unifier, ty);
let llvm_elem_ty = ctx.get_llvm_type(generator, dtype);
NDArrayValue::from_pointer_value(v, llvm_elem_ty, llvm_usize, None).load_ndims(ctx)
NDArrayType::from_unifier_type(generator, ctx, ty).map_value(v, None).load_ndims(ctx)
}
BasicValueEnum::PointerValue(v) if ListValue::is_representable(v, llvm_usize).is_ok() => {
@ -857,7 +767,7 @@ fn call_ndarray_array_impl<'ctx, G: CodeGenerator + ?Sized>(
// object is an NDArray instance - copy object unless copy=0 && ndmin < object.ndims
if NDArrayValue::is_representable(object, llvm_usize).is_ok() {
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
let object = NDArrayValue::from_pointer_value(object, llvm_elem_ty, llvm_usize, None);
let object = NDArrayValue::from_pointer_value(object, llvm_elem_ty, None, llvm_usize, None);
let ndarray = gen_if_else_expr_callback(
generator,
@ -933,6 +843,7 @@ fn call_ndarray_array_impl<'ctx, G: CodeGenerator + ?Sized>(
return Ok(NDArrayValue::from_pointer_value(
ndarray.map(BasicValueEnum::into_pointer_value).unwrap(),
llvm_elem_ty,
None,
llvm_usize,
None,
));
@ -1126,7 +1037,7 @@ fn call_ndarray_eye_impl<'ctx, G: CodeGenerator + ?Sized>(
/// Copies a slice of an [`NDArrayValue`] to another.
///
/// - `dst_arr`: The [`NDArrayValue`] instance of the destination array. The `ndims` and `dim_sz`
/// - `dst_arr`: The [`NDArrayValue`] instance of the destination array. The `ndims` and `shape`
/// fields should be populated before calling this function.
/// - `dst_slice_ptr`: The [`PointerValue`] to the first element of the currently processing
/// dimensional slice in the destination array.
@ -1270,85 +1181,86 @@ pub fn ndarray_sliced_copy<'ctx, G: CodeGenerator + ?Sized>(
) -> Result<NDArrayValue<'ctx>, String> {
let llvm_i32 = ctx.ctx.i32_type();
let llvm_usize = generator.get_size_type(ctx.ctx);
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
let ndarray = if slices.is_empty() {
create_ndarray_dyn_shape(
generator,
ctx,
elem_ty,
&this,
|_, ctx, shape| Ok(shape.load_ndims(ctx)),
|generator, ctx, shape, idx| unsafe {
Ok(shape.shape().get_typed_unchecked(ctx, generator, &idx, None))
},
)?
} else {
let ndarray = create_ndarray_uninitialized(generator, ctx, elem_ty)?;
ndarray.store_ndims(ctx, generator, this.load_ndims(ctx));
let ndarray =
if slices.is_empty() {
create_ndarray_dyn_shape(
generator,
ctx,
elem_ty,
&this,
|_, ctx, shape| Ok(shape.load_ndims(ctx)),
|generator, ctx, shape, idx| unsafe {
Ok(shape.shape().get_typed_unchecked(ctx, generator, &idx, None))
},
)?
} else {
let ndarray = NDArrayType::new(generator, ctx.ctx, llvm_elem_ty, None)
.construct_dyn_ndims(generator, ctx, this.load_ndims(ctx), None);
let ndims = this.load_ndims(ctx);
ndarray.create_shape(ctx, llvm_usize, ndims);
// Populate the first slices.len() dimensions by computing the size of each dim slice
for (i, (start, stop, step)) in slices.iter().enumerate() {
// HACK: workaround calculate_len_for_slice_range requiring exclusive stop
let stop = ctx
.builder
.build_select(
ctx.builder
.build_int_compare(
IntPredicate::SLT,
*step,
llvm_i32.const_zero(),
"is_neg",
)
.unwrap(),
ctx.builder
.build_int_sub(*stop, llvm_i32.const_int(1, true), "e_min_one")
.unwrap(),
ctx.builder
.build_int_add(*stop, llvm_i32.const_int(1, true), "e_add_one")
.unwrap(),
"final_e",
)
.map(BasicValueEnum::into_int_value)
.unwrap();
// Populate the first slices.len() dimensions by computing the size of each dim slice
for (i, (start, stop, step)) in slices.iter().enumerate() {
// HACK: workaround calculate_len_for_slice_range requiring exclusive stop
let stop = ctx
.builder
.build_select(
ctx.builder
.build_int_compare(
IntPredicate::SLT,
*step,
llvm_i32.const_zero(),
"is_neg",
)
.unwrap(),
ctx.builder
.build_int_sub(*stop, llvm_i32.const_int(1, true), "e_min_one")
.unwrap(),
ctx.builder
.build_int_add(*stop, llvm_i32.const_int(1, true), "e_add_one")
.unwrap(),
"final_e",
)
.map(BasicValueEnum::into_int_value)
.unwrap();
let slice_len = calculate_len_for_slice_range(generator, ctx, *start, stop, *step);
let slice_len =
ctx.builder.build_int_z_extend_or_bit_cast(slice_len, llvm_usize, "").unwrap();
let slice_len = calculate_len_for_slice_range(generator, ctx, *start, stop, *step);
let slice_len =
ctx.builder.build_int_z_extend_or_bit_cast(slice_len, llvm_usize, "").unwrap();
unsafe {
ndarray.shape().set_typed_unchecked(
ctx,
generator,
&llvm_usize.const_int(i as u64, false),
slice_len,
);
}
}
// Populate the rest by directly copying the dim size from the source array
gen_for_callback_incrementing(
generator,
ctx,
None,
llvm_usize.const_int(slices.len() as u64, false),
(this.load_ndims(ctx), false),
|generator, ctx, _, idx| {
unsafe {
let dim_sz = this.shape().get_typed_unchecked(ctx, generator, &idx, None);
ndarray.shape().set_typed_unchecked(ctx, generator, &idx, dim_sz);
ndarray.shape().set_typed_unchecked(
ctx,
generator,
&llvm_usize.const_int(i as u64, false),
slice_len,
);
}
}
Ok(())
},
llvm_usize.const_int(1, false),
)
.unwrap();
// Populate the rest by directly copying the dim size from the source array
gen_for_callback_incrementing(
generator,
ctx,
None,
llvm_usize.const_int(slices.len() as u64, false),
(this.load_ndims(ctx), false),
|generator, ctx, _, idx| {
unsafe {
let shape = this.shape().get_typed_unchecked(ctx, generator, &idx, None);
ndarray.shape().set_typed_unchecked(ctx, generator, &idx, shape);
}
ndarray_init_data(generator, ctx, elem_ty, ndarray)
};
Ok(())
},
llvm_usize.const_int(1, false),
)
.unwrap();
unsafe { ndarray.create_data(generator, ctx) };
ndarray
};
ndarray_sliced_copyto_impl(
generator,
@ -1447,8 +1359,6 @@ where
(BasicValueEnum<'ctx>, BasicValueEnum<'ctx>),
) -> Result<BasicValueEnum<'ctx>, String>,
{
let llvm_usize = generator.get_size_type(ctx.ctx);
let (lhs_ty, lhs_val, lhs_scalar) = lhs;
let (rhs_ty, rhs_val, rhs_scalar) = rhs;
@ -1461,22 +1371,10 @@ where
let ndarray = res.unwrap_or_else(|| {
if lhs_scalar && rhs_scalar {
let lhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, lhs_ty);
let llvm_lhs_elem_ty = ctx.get_llvm_type(generator, lhs_dtype);
let lhs_val = NDArrayValue::from_pointer_value(
lhs_val.into_pointer_value(),
llvm_lhs_elem_ty,
llvm_usize,
None,
);
let rhs_dtype = arraylike_flatten_element_type(&mut ctx.unifier, rhs_ty);
let llvm_rhs_elem_ty = ctx.get_llvm_type(generator, rhs_dtype);
let rhs_val = NDArrayValue::from_pointer_value(
rhs_val.into_pointer_value(),
llvm_rhs_elem_ty,
llvm_usize,
None,
);
let lhs_val = NDArrayType::from_unifier_type(generator, ctx, lhs_ty)
.map_value(lhs_val.into_pointer_value(), None);
let rhs_val = NDArrayType::from_unifier_type(generator, ctx, rhs_ty)
.map_value(rhs_val.into_pointer_value(), None);
let ndarray_dims = call_ndarray_calc_broadcast(generator, ctx, lhs_val, rhs_val);
@ -1492,17 +1390,12 @@ where
)
.unwrap()
} else {
let dtype = arraylike_flatten_element_type(
&mut ctx.unifier,
let ndarray = NDArrayType::from_unifier_type(
generator,
ctx,
if lhs_scalar { rhs_ty } else { lhs_ty },
);
let llvm_elem_ty = ctx.get_llvm_type(generator, dtype);
let ndarray = NDArrayValue::from_pointer_value(
if lhs_scalar { rhs_val } else { lhs_val }.into_pointer_value(),
llvm_elem_ty,
llvm_usize,
None,
);
)
.map_value(if lhs_scalar { rhs_val } else { lhs_val }.into_pointer_value(), None);
create_ndarray_dyn_shape(
generator,
@ -2046,25 +1939,18 @@ pub fn gen_ndarray_copy<'ctx>(
assert!(obj.is_some());
assert!(args.is_empty());
let llvm_usize = generator.get_size_type(context.ctx);
let this_ty = obj.as_ref().unwrap().0;
let (this_elem_ty, _) = unpack_ndarray_var_tys(&mut context.unifier, this_ty);
let this_arg =
obj.as_ref().unwrap().1.clone().to_basic_value_enum(context, generator, this_ty)?;
let llvm_elem_ty = context.get_llvm_type(generator, this_elem_ty);
let llvm_this_ty = NDArrayType::from_unifier_type(generator, context, this_ty);
ndarray_copy_impl(
generator,
context,
this_elem_ty,
NDArrayValue::from_pointer_value(
this_arg.into_pointer_value(),
llvm_elem_ty,
llvm_usize,
None,
),
llvm_this_ty.map_value(this_arg.into_pointer_value(), None),
)
.map(NDArrayValue::into)
}
@ -2080,10 +1966,7 @@ pub fn gen_ndarray_fill<'ctx>(
assert!(obj.is_some());
assert_eq!(args.len(), 1);
let llvm_usize = generator.get_size_type(context.ctx);
let this_ty = obj.as_ref().unwrap().0;
let this_elem_ty = arraylike_flatten_element_type(&mut context.unifier, this_ty);
let this_arg = obj
.as_ref()
.unwrap()
@ -2094,12 +1977,12 @@ pub fn gen_ndarray_fill<'ctx>(
let value_ty = fun.0.args[0].ty;
let value_arg = args[0].1.clone().to_basic_value_enum(context, generator, value_ty)?;
let llvm_elem_ty = context.get_llvm_type(generator, this_elem_ty);
let llvm_this_ty = NDArrayType::from_unifier_type(generator, context, this_ty);
ndarray_fill_flattened(
generator,
context,
NDArrayValue::from_pointer_value(this_arg, llvm_elem_ty, llvm_usize, None),
llvm_this_ty.map_value(this_arg, None),
|generator, ctx, _| {
let value = if value_arg.is_pointer_value() {
let llvm_i1 = ctx.ctx.bool_type();
@ -2132,16 +2015,16 @@ pub fn gen_ndarray_fill<'ctx>(
pub fn ndarray_transpose<'ctx, G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
x1: (Type, BasicValueEnum<'ctx>),
(x1_ty, x1): (Type, BasicValueEnum<'ctx>),
) -> Result<BasicValueEnum<'ctx>, String> {
const FN_NAME: &str = "ndarray_transpose";
let (x1_ty, x1) = x1;
let llvm_usize = generator.get_size_type(ctx.ctx);
if let BasicValueEnum::PointerValue(n1) = x1 {
let llvm_ndarray_ty = NDArrayType::from_unifier_type(generator, ctx, x1_ty);
let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, x1_ty);
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
let n1 = NDArrayValue::from_pointer_value(n1, llvm_elem_ty, llvm_usize, None);
let n1 = llvm_ndarray_ty.map_value(n1, None);
let n_sz = call_ndarray_calc_size(generator, ctx, &n1.shape(), (None, None));
// Dimensions are reversed in the transposed array
@ -2260,8 +2143,8 @@ pub fn ndarray_reshape<'ctx, G: CodeGenerator + ?Sized>(
if let BasicValueEnum::PointerValue(n1) = x1 {
let (elem_ty, _) = unpack_ndarray_var_tys(&mut ctx.unifier, x1_ty);
let llvm_elem_ty = ctx.get_llvm_type(generator, elem_ty);
let n1 = NDArrayValue::from_pointer_value(n1, llvm_elem_ty, llvm_usize, None);
let llvm_ndarray_ty = NDArrayType::from_unifier_type(generator, ctx, x1_ty);
let n1 = llvm_ndarray_ty.map_value(n1, None);
let n_sz = call_ndarray_calc_size(generator, ctx, &n1.shape(), (None, None));
let acc = generator.gen_var_alloc(ctx, llvm_usize.into(), None)?;
@ -2544,13 +2427,8 @@ pub fn ndarray_dot<'ctx, G: CodeGenerator + ?Sized>(
match (x1, x2) {
(BasicValueEnum::PointerValue(n1), BasicValueEnum::PointerValue(n2)) => {
let n1_dtype = arraylike_flatten_element_type(&mut ctx.unifier, x1_ty);
let n2_dtype = arraylike_flatten_element_type(&mut ctx.unifier, x2_ty);
let llvm_n1_data_ty = ctx.get_llvm_type(generator, n1_dtype);
let llvm_n2_data_ty = ctx.get_llvm_type(generator, n2_dtype);
let n1 = NDArrayValue::from_pointer_value(n1, llvm_n1_data_ty, llvm_usize, None);
let n2 = NDArrayValue::from_pointer_value(n2, llvm_n2_data_ty, llvm_usize, None);
let n1 = NDArrayType::from_unifier_type(generator, ctx, x1_ty).map_value(n1, None);
let n2 = NDArrayType::from_unifier_type(generator, ctx, x2_ty).map_value(n2, None);
let n1_sz = call_ndarray_calc_size(generator, ctx, &n1.shape(), (None, None));
let n2_sz = call_ndarray_calc_size(generator, ctx, &n1.shape(), (None, None));

View File

@ -17,7 +17,7 @@ use parking_lot::RwLock;
use super::{
concrete_type::ConcreteTypeStore,
types::{ListType, NDArrayType, ProxyType, RangeType},
types::{ndarray::NDArrayType, ListType, ProxyType, RangeType},
CodeGenContext, CodeGenLLVMOptions, CodeGenTargetMachineOptions, CodeGenTask, CodeGenerator,
DefaultCodeGenerator, WithCall, WorkerRegistry,
};
@ -471,6 +471,6 @@ fn test_classes_ndarray_type_new() {
let llvm_i32 = ctx.i32_type();
let llvm_usize = generator.get_size_type(&ctx);
let llvm_ndarray = NDArrayType::new(&generator, &ctx, llvm_i32.into());
let llvm_ndarray = NDArrayType::new(&generator, &ctx, llvm_i32.into(), None);
assert!(NDArrayType::is_representable(llvm_ndarray.as_base_type(), llvm_usize).is_ok());
}

View File

@ -23,13 +23,13 @@ use super::{
{CodeGenContext, CodeGenerator},
};
pub use list::*;
pub use ndarray::*;
pub use range::*;
mod list;
mod ndarray;
pub mod ndarray;
mod range;
pub mod structure;
pub mod utils;
/// A LLVM type that is used to represent a corresponding type in NAC3.
pub trait ProxyType<'ctx>: Into<Self::Base> {

View File

@ -0,0 +1,257 @@
use inkwell::{
context::Context,
types::{AnyTypeEnum, BasicType, BasicTypeEnum, IntType, PointerType},
values::{IntValue, PointerValue},
AddressSpace,
};
use itertools::Itertools;
use nac3core_derive::StructFields;
use crate::{
codegen::{
types::{
structure::{
check_struct_type_matches_fields, FieldIndexCounter, StructField, StructFields,
},
ProxyType,
},
values::{ndarray::ContiguousNDArrayValue, ArraySliceValue, ProxyValue},
CodeGenContext, CodeGenerator,
},
toplevel::numpy::unpack_ndarray_var_tys,
typecheck::typedef::Type,
};
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
pub struct ContiguousNDArrayType<'ctx> {
ty: PointerType<'ctx>,
item: BasicTypeEnum<'ctx>,
llvm_usize: IntType<'ctx>,
}
#[derive(PartialEq, Eq, Clone, Copy, StructFields)]
pub struct ContiguousNDArrayFields<'ctx> {
#[value_type(usize)]
pub ndims: StructField<'ctx, IntValue<'ctx>>,
#[value_type(usize.ptr_type(AddressSpace::default()))]
pub shape: StructField<'ctx, PointerValue<'ctx>>,
#[value_type(i8_type().ptr_type(AddressSpace::default()))]
pub data: StructField<'ctx, PointerValue<'ctx>>,
}
impl<'ctx> ContiguousNDArrayFields<'ctx> {
#[must_use]
pub fn new_typed(item: BasicTypeEnum<'ctx>, llvm_usize: IntType<'ctx>) -> Self {
let mut counter = FieldIndexCounter::default();
ContiguousNDArrayFields {
ndims: StructField::create(&mut counter, "ndims", llvm_usize),
shape: StructField::create(
&mut counter,
"shape",
llvm_usize.ptr_type(AddressSpace::default()),
),
data: StructField::create(&mut counter, "data", item.ptr_type(AddressSpace::default())),
}
}
}
impl<'ctx> ContiguousNDArrayType<'ctx> {
/// Checks whether `llvm_ty` represents a `ndarray` type, returning [Err] if it does not.
pub fn is_representable(
llvm_ty: PointerType<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
let ctx = llvm_ty.get_context();
let llvm_ty = llvm_ty.get_element_type();
let AnyTypeEnum::StructType(llvm_ty) = llvm_ty else {
return Err(format!(
"Expected struct type for `ContiguousNDArray` type, got {llvm_ty}"
));
};
let fields = ContiguousNDArrayFields::new(ctx, llvm_usize);
check_struct_type_matches_fields(
fields,
llvm_ty,
"ContiguousNDArray",
&[(fields.data.name(), &|ty| {
if ty.is_pointer_type() {
Ok(())
} else {
Err(format!("Expected T* for `ContiguousNDArray.data`, got {ty}"))
}
})],
)
}
/// Returns an instance of [`StructFields`] containing all field accessors for this type.
#[must_use]
fn fields(
item: BasicTypeEnum<'ctx>,
llvm_usize: IntType<'ctx>,
) -> ContiguousNDArrayFields<'ctx> {
ContiguousNDArrayFields::new_typed(item, llvm_usize)
}
/// See [`NDArrayType::fields`].
// TODO: Move this into e.g. StructProxyType
#[must_use]
pub fn get_fields(&self) -> ContiguousNDArrayFields<'ctx> {
Self::fields(self.item, self.llvm_usize)
}
/// Creates an LLVM type corresponding to the expected structure of an `NDArray`.
#[must_use]
fn llvm_type(
ctx: &'ctx Context,
item: BasicTypeEnum<'ctx>,
llvm_usize: IntType<'ctx>,
) -> PointerType<'ctx> {
let field_tys =
Self::fields(item, llvm_usize).into_iter().map(|field| field.1).collect_vec();
ctx.struct_type(&field_tys, false).ptr_type(AddressSpace::default())
}
/// Creates an instance of [`ContiguousNDArrayType`].
#[must_use]
pub fn new<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
item: BasicTypeEnum<'ctx>,
) -> Self {
let llvm_usize = generator.get_size_type(ctx);
let llvm_cndarray = Self::llvm_type(ctx, item, llvm_usize);
Self { ty: llvm_cndarray, item, llvm_usize }
}
/// Creates an [`ContiguousNDArrayType`] from a [unifier type][Type].
#[must_use]
pub fn from_unifier_type<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &mut CodeGenContext<'ctx, '_>,
ty: Type,
) -> Self {
let (dtype, _) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
let llvm_dtype = ctx.get_llvm_type(generator, dtype);
let llvm_usize = generator.get_size_type(ctx.ctx);
Self { ty: Self::llvm_type(ctx.ctx, llvm_dtype, llvm_usize), item: llvm_dtype, llvm_usize }
}
/// Creates an [`ContiguousNDArrayType`] from a [`PointerType`] representing an `NDArray`.
#[must_use]
pub fn from_type(
ptr_ty: PointerType<'ctx>,
item: BasicTypeEnum<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Self {
debug_assert!(Self::is_representable(ptr_ty, llvm_usize).is_ok());
Self { ty: ptr_ty, item, llvm_usize }
}
/// Allocates an instance of [`ContiguousNDArrayValue`] as if by calling `alloca` on the base type.
#[must_use]
pub fn alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
self.raw_alloca(generator, ctx, name),
self.item,
self.llvm_usize,
name,
)
}
/// Converts an existing value into a [`ContiguousNDArrayValue`].
#[must_use]
pub fn map_value(
&self,
value: <<Self as ProxyType<'ctx>>::Value as ProxyValue<'ctx>>::Base,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
value,
self.item,
self.llvm_usize,
name,
)
}
}
impl<'ctx> ProxyType<'ctx> for ContiguousNDArrayType<'ctx> {
type Base = PointerType<'ctx>;
type Value = ContiguousNDArrayValue<'ctx>;
fn is_type<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
llvm_ty: impl BasicType<'ctx>,
) -> Result<(), String> {
if let BasicTypeEnum::PointerType(ty) = llvm_ty.as_basic_type_enum() {
<Self as ProxyType<'ctx>>::is_representable(generator, ctx, ty)
} else {
Err(format!("Expected pointer type, got {llvm_ty:?}"))
}
}
fn is_representable<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
llvm_ty: Self::Base,
) -> Result<(), String> {
Self::is_representable(llvm_ty, generator.get_size_type(ctx))
}
fn raw_alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
name: Option<&'ctx str>,
) -> <Self::Value as ProxyValue<'ctx>>::Base {
generator
.gen_var_alloc(
ctx,
self.as_base_type().get_element_type().into_struct_type().into(),
name,
)
.unwrap()
}
fn array_alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
size: IntValue<'ctx>,
name: Option<&'ctx str>,
) -> ArraySliceValue<'ctx> {
generator
.gen_array_var_alloc(
ctx,
self.as_base_type().get_element_type().into_struct_type().into(),
size,
name,
)
.unwrap()
}
fn as_base_type(&self) -> Self::Base {
self.ty
}
}
impl<'ctx> From<ContiguousNDArrayType<'ctx>> for PointerType<'ctx> {
fn from(value: ContiguousNDArrayType<'ctx>) -> Self {
value.as_base_type()
}
}

View File

@ -0,0 +1,215 @@
use inkwell::{
context::{AsContextRef, Context},
types::{AnyTypeEnum, BasicType, BasicTypeEnum, IntType, PointerType},
values::{IntValue, PointerValue},
AddressSpace,
};
use itertools::Itertools;
use nac3core_derive::StructFields;
use crate::codegen::{
types::{
structure::{check_struct_type_matches_fields, StructField, StructFields},
ProxyType,
},
values::{
ndarray::{NDIndexValue, RustNDIndex},
ArrayLikeIndexer, ArraySliceValue, ProxyValue,
},
CodeGenContext, CodeGenerator,
};
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
pub struct NDIndexType<'ctx> {
ty: PointerType<'ctx>,
llvm_usize: IntType<'ctx>,
}
#[derive(PartialEq, Eq, Clone, Copy, StructFields)]
pub struct NDIndexStructFields<'ctx> {
#[value_type(i8_type())]
pub type_: StructField<'ctx, IntValue<'ctx>>,
#[value_type(i8_type().ptr_type(AddressSpace::default()))]
pub data: StructField<'ctx, PointerValue<'ctx>>,
}
impl<'ctx> NDIndexType<'ctx> {
/// Checks whether `llvm_ty` represents a `ndindex` type, returning [Err] if it does not.
pub fn is_representable(
llvm_ty: PointerType<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
let ctx = llvm_ty.get_context();
let llvm_ty = llvm_ty.get_element_type();
let AnyTypeEnum::StructType(llvm_ty) = llvm_ty else {
return Err(format!(
"Expected struct type for `ContiguousNDArray` type, got {llvm_ty}"
));
};
let fields = NDIndexStructFields::new(ctx, llvm_usize);
check_struct_type_matches_fields(fields, llvm_ty, "NDIndex", &[])
}
#[must_use]
fn fields(
ctx: impl AsContextRef<'ctx>,
llvm_usize: IntType<'ctx>,
) -> NDIndexStructFields<'ctx> {
NDIndexStructFields::new(ctx, llvm_usize)
}
#[must_use]
pub fn get_fields(&self) -> NDIndexStructFields<'ctx> {
Self::fields(self.ty.get_context(), self.llvm_usize)
}
#[must_use]
fn llvm_type(ctx: &'ctx Context, llvm_usize: IntType<'ctx>) -> PointerType<'ctx> {
let field_tys =
Self::fields(ctx, llvm_usize).into_iter().map(|field| field.1).collect_vec();
ctx.struct_type(&field_tys, false).ptr_type(AddressSpace::default())
}
#[must_use]
pub fn new<G: CodeGenerator + ?Sized>(generator: &G, ctx: &'ctx Context) -> Self {
let llvm_usize = generator.get_size_type(ctx);
let llvm_ndindex = Self::llvm_type(ctx, llvm_usize);
Self { ty: llvm_ndindex, llvm_usize }
}
#[must_use]
pub fn from_type(ptr_ty: PointerType<'ctx>, llvm_usize: IntType<'ctx>) -> Self {
debug_assert!(Self::is_representable(ptr_ty, llvm_usize).is_ok());
Self { ty: ptr_ty, llvm_usize }
}
#[must_use]
pub fn alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
self.raw_alloca(generator, ctx, name),
self.llvm_usize,
name,
)
}
/// Serialize a list of [`RustNDIndex`] as a newly allocated LLVM array of [`NDIndexValue`].
#[must_use]
pub fn construct_ndindices<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
in_ndindices: &[RustNDIndex<'ctx>],
) -> ArraySliceValue<'ctx> {
// Allocate the LLVM ndindices.
let num_ndindices = self.llvm_usize.const_int(in_ndindices.len() as u64, false);
let ndindices = self.array_alloca(generator, ctx, num_ndindices, None);
// Initialize all of them.
for (i, in_ndindex) in in_ndindices.iter().enumerate() {
let pndindex = unsafe {
ndindices.ptr_offset_unchecked(
ctx,
generator,
&ctx.ctx.i64_type().const_int(u64::try_from(i).unwrap(), false),
None,
)
};
in_ndindex.write_to_ndindex(
generator,
ctx,
NDIndexValue::from_pointer_value(pndindex, self.llvm_usize, None),
);
}
ndindices
}
#[must_use]
pub fn map_value(
&self,
value: <<Self as ProxyType<'ctx>>::Value as ProxyValue<'ctx>>::Base,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(value, self.llvm_usize, name)
}
}
impl<'ctx> ProxyType<'ctx> for NDIndexType<'ctx> {
type Base = PointerType<'ctx>;
type Value = NDIndexValue<'ctx>;
fn is_type<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
llvm_ty: impl BasicType<'ctx>,
) -> Result<(), String> {
if let BasicTypeEnum::PointerType(ty) = llvm_ty.as_basic_type_enum() {
<Self as ProxyType<'ctx>>::is_representable(generator, ctx, ty)
} else {
Err(format!("Expected pointer type, got {llvm_ty:?}"))
}
}
fn is_representable<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
llvm_ty: Self::Base,
) -> Result<(), String> {
Self::is_representable(llvm_ty, generator.get_size_type(ctx))
}
fn raw_alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
name: Option<&'ctx str>,
) -> <Self::Value as ProxyValue<'ctx>>::Base {
generator
.gen_var_alloc(
ctx,
self.as_base_type().get_element_type().into_struct_type().into(),
name,
)
.unwrap()
}
fn array_alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
size: IntValue<'ctx>,
name: Option<&'ctx str>,
) -> ArraySliceValue<'ctx> {
generator
.gen_array_var_alloc(
ctx,
self.as_base_type().get_element_type().into_struct_type().into(),
size,
name,
)
.unwrap()
}
fn as_base_type(&self) -> Self::Base {
self.ty
}
}
impl<'ctx> From<NDIndexType<'ctx>> for PointerType<'ctx> {
fn from(value: NDIndexType<'ctx>) -> Self {
value.as_base_type()
}
}

View File

@ -1,7 +1,7 @@
use inkwell::{
context::Context,
context::{AsContextRef, Context},
types::{AnyTypeEnum, BasicType, BasicTypeEnum, IntType, PointerType},
values::{IntValue, PointerValue},
values::{BasicValue, IntValue, PointerValue},
AddressSpace,
};
use itertools::Itertools;
@ -9,28 +9,49 @@ use itertools::Itertools;
use nac3core_derive::StructFields;
use super::{
structure::{StructField, StructFields},
structure::{check_struct_type_matches_fields, StructField, StructFields},
ProxyType,
};
use crate::codegen::{
values::{ArraySliceValue, NDArrayValue, ProxyValue},
{CodeGenContext, CodeGenerator},
use crate::{
codegen::{
values::{ndarray::NDArrayValue, ArraySliceValue, ProxyValue, TypedArrayLikeMutator},
{CodeGenContext, CodeGenerator},
},
toplevel::{helper::extract_ndims, numpy::unpack_ndarray_var_tys},
typecheck::typedef::Type,
};
pub use contiguous::*;
pub use indexing::*;
pub use nditer::*;
mod contiguous;
mod indexing;
mod nditer;
/// Proxy type for a `ndarray` type in LLVM.
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
pub struct NDArrayType<'ctx> {
ty: PointerType<'ctx>,
dtype: BasicTypeEnum<'ctx>,
ndims: Option<u64>,
llvm_usize: IntType<'ctx>,
}
#[derive(PartialEq, Eq, Clone, Copy, StructFields)]
pub struct NDArrayStructFields<'ctx> {
/// The size of each `NDArray` element in bytes.
#[value_type(usize)]
pub itemsize: StructField<'ctx, IntValue<'ctx>>,
/// Number of dimensions in the array.
#[value_type(usize)]
pub ndims: StructField<'ctx, IntValue<'ctx>>,
/// Pointer to an array containing the shape of the `NDArray`.
#[value_type(usize.ptr_type(AddressSpace::default()))]
pub shape: StructField<'ctx, PointerValue<'ctx>>,
/// Pointer to an array indicating the number of bytes between each element at a dimension
#[value_type(usize.ptr_type(AddressSpace::default()))]
pub strides: StructField<'ctx, PointerValue<'ctx>>,
/// Pointer to an array containing the array data
#[value_type(i8_type().ptr_type(AddressSpace::default()))]
pub data: StructField<'ctx, PointerValue<'ctx>>,
}
@ -41,90 +62,40 @@ impl<'ctx> NDArrayType<'ctx> {
llvm_ty: PointerType<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
let ctx = llvm_ty.get_context();
let llvm_ndarray_ty = llvm_ty.get_element_type();
let AnyTypeEnum::StructType(llvm_ndarray_ty) = llvm_ndarray_ty else {
return Err(format!("Expected struct type for `NDArray` type, got {llvm_ndarray_ty}"));
};
if llvm_ndarray_ty.count_fields() != 3 {
return Err(format!(
"Expected 3 fields in `NDArray`, got {}",
llvm_ndarray_ty.count_fields()
));
}
let ndarray_ndims_ty = llvm_ndarray_ty.get_field_type_at_index(0).unwrap();
let Ok(ndarray_ndims_ty) = IntType::try_from(ndarray_ndims_ty) else {
return Err(format!("Expected int type for `ndarray.0`, got {ndarray_ndims_ty}"));
};
if ndarray_ndims_ty.get_bit_width() != llvm_usize.get_bit_width() {
return Err(format!(
"Expected {}-bit int type for `ndarray.0`, got {}-bit int",
llvm_usize.get_bit_width(),
ndarray_ndims_ty.get_bit_width()
));
}
let ndarray_dims_ty = llvm_ndarray_ty.get_field_type_at_index(1).unwrap();
let Ok(ndarray_pdims) = PointerType::try_from(ndarray_dims_ty) else {
return Err(format!("Expected pointer type for `ndarray.1`, got {ndarray_dims_ty}"));
};
let ndarray_dims = ndarray_pdims.get_element_type();
let Ok(ndarray_dims) = IntType::try_from(ndarray_dims) else {
return Err(format!(
"Expected pointer-to-int type for `ndarray.1`, got pointer-to-{ndarray_dims}"
));
};
if ndarray_dims.get_bit_width() != llvm_usize.get_bit_width() {
return Err(format!(
"Expected pointer-to-{}-bit int type for `ndarray.1`, got pointer-to-{}-bit int",
llvm_usize.get_bit_width(),
ndarray_dims.get_bit_width()
));
}
let ndarray_data_ty = llvm_ndarray_ty.get_field_type_at_index(2).unwrap();
let Ok(ndarray_pdata) = PointerType::try_from(ndarray_data_ty) else {
return Err(format!("Expected pointer type for `ndarray.2`, got {ndarray_data_ty}"));
};
let ndarray_data = ndarray_pdata.get_element_type();
let Ok(ndarray_data) = IntType::try_from(ndarray_data) else {
return Err(format!(
"Expected pointer-to-int type for `ndarray.2`, got pointer-to-{ndarray_data}"
));
};
if ndarray_data.get_bit_width() != 8 {
return Err(format!(
"Expected pointer-to-8-bit int type for `ndarray.1`, got pointer-to-{}-bit int",
ndarray_data.get_bit_width()
));
}
Ok(())
check_struct_type_matches_fields(
Self::fields(ctx, llvm_usize),
llvm_ndarray_ty,
"NDArray",
&[],
)
}
/// Returns an instance of [`StructFields`] containing all field accessors for this type.
#[must_use]
fn fields(ctx: &'ctx Context, llvm_usize: IntType<'ctx>) -> NDArrayStructFields<'ctx> {
fn fields(
ctx: impl AsContextRef<'ctx>,
llvm_usize: IntType<'ctx>,
) -> NDArrayStructFields<'ctx> {
NDArrayStructFields::new(ctx, llvm_usize)
}
/// See [`NDArrayType::fields`].
// TODO: Move this into e.g. StructProxyType
#[must_use]
pub fn get_fields(&self, ctx: &'ctx Context) -> NDArrayStructFields<'ctx> {
pub fn get_fields(&self, ctx: impl AsContextRef<'ctx>) -> NDArrayStructFields<'ctx> {
Self::fields(ctx, self.llvm_usize)
}
/// Creates an LLVM type corresponding to the expected structure of an `NDArray`.
#[must_use]
fn llvm_type(ctx: &'ctx Context, llvm_usize: IntType<'ctx>) -> PointerType<'ctx> {
// struct NDArray { num_dims: size_t, dims: size_t*, data: i8* }
//
// * data : Pointer to an array containing the array data
// * itemsize: The size of each NDArray elements in bytes
// * ndims : Number of dimensions in the array
// * shape : Pointer to an array containing the shape of the NDArray
// * strides : Pointer to an array indicating the number of bytes between each element at a dimension
let field_tys =
Self::fields(ctx, llvm_usize).into_iter().map(|field| field.1).collect_vec();
@ -137,11 +108,46 @@ impl<'ctx> NDArrayType<'ctx> {
generator: &G,
ctx: &'ctx Context,
dtype: BasicTypeEnum<'ctx>,
ndims: Option<u64>,
) -> Self {
let llvm_usize = generator.get_size_type(ctx);
let llvm_ndarray = Self::llvm_type(ctx, llvm_usize);
NDArrayType { ty: llvm_ndarray, dtype, llvm_usize }
NDArrayType { ty: llvm_ndarray, dtype, ndims, llvm_usize }
}
/// Creates an instance of [`NDArrayType`] with `ndims` of 0.
#[must_use]
pub fn new_unsized<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
dtype: BasicTypeEnum<'ctx>,
) -> Self {
let llvm_usize = generator.get_size_type(ctx);
let llvm_ndarray = Self::llvm_type(ctx, llvm_usize);
NDArrayType { ty: llvm_ndarray, dtype, ndims: Some(0), llvm_usize }
}
/// Creates an [`NDArrayType`] from a [unifier type][Type].
#[must_use]
pub fn from_unifier_type<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &mut CodeGenContext<'ctx, '_>,
ty: Type,
) -> Self {
let (dtype, ndims) = unpack_ndarray_var_tys(&mut ctx.unifier, ty);
let llvm_dtype = ctx.get_llvm_type(generator, dtype);
let llvm_usize = generator.get_size_type(ctx.ctx);
let ndims = extract_ndims(&ctx.unifier, ndims);
NDArrayType {
ty: Self::llvm_type(ctx.ctx, llvm_usize),
dtype: llvm_dtype,
ndims: Some(ndims),
llvm_usize,
}
}
/// Creates an [`NDArrayType`] from a [`PointerType`] representing an `NDArray`.
@ -149,22 +155,18 @@ impl<'ctx> NDArrayType<'ctx> {
pub fn from_type(
ptr_ty: PointerType<'ctx>,
dtype: BasicTypeEnum<'ctx>,
ndims: Option<u64>,
llvm_usize: IntType<'ctx>,
) -> Self {
debug_assert!(Self::is_representable(ptr_ty, llvm_usize).is_ok());
NDArrayType { ty: ptr_ty, dtype, llvm_usize }
NDArrayType { ty: ptr_ty, dtype, ndims, llvm_usize }
}
/// Returns the type of the `size` field of this `ndarray` type.
#[must_use]
pub fn size_type(&self) -> IntType<'ctx> {
self.as_base_type()
.get_element_type()
.into_struct_type()
.get_field_type_at_index(0)
.map(BasicTypeEnum::into_int_type)
.unwrap()
self.llvm_usize
}
/// Returns the element type of this `ndarray` type.
@ -173,6 +175,12 @@ impl<'ctx> NDArrayType<'ctx> {
self.dtype
}
/// Returns the number of dimensions of this `ndarray` type.
#[must_use]
pub fn ndims(&self) -> Option<u64> {
self.ndims
}
/// Allocates an instance of [`NDArrayValue`] as if by calling `alloca` on the base type.
#[must_use]
pub fn alloca<G: CodeGenerator + ?Sized>(
@ -184,11 +192,198 @@ impl<'ctx> NDArrayType<'ctx> {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
self.raw_alloca(generator, ctx, name),
self.dtype,
self.ndims,
self.llvm_usize,
name,
)
}
/// Allocates an [`NDArrayValue`] on the stack and initializes all fields as follows:
///
/// - `data`: uninitialized.
/// - `itemsize`: set to the size of `self.dtype`.
/// - `ndims`: set to the value of `ndims`.
/// - `shape`: allocated on the stack with an array of length `ndims` with uninitialized values.
/// - `strides`: allocated on the stack with an array of length `ndims` with uninitialized
/// values.
#[must_use]
fn construct_impl<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
ndims: IntValue<'ctx>,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
let ndarray = self.alloca(generator, ctx, name);
let itemsize = ctx
.builder
.build_int_truncate_or_bit_cast(self.dtype.size_of().unwrap(), self.llvm_usize, "")
.unwrap();
ndarray.store_itemsize(ctx, generator, itemsize);
ndarray.store_ndims(ctx, generator, ndims);
ndarray.create_shape(ctx, self.llvm_usize, ndims);
ndarray.create_strides(ctx, self.llvm_usize, ndims);
ndarray
}
/// Allocate an [`NDArrayValue`] on the stack using `dtype` and `ndims` of this [`NDArrayType`]
/// instance.
///
/// The returned ndarray's content will be:
/// - `data`: uninitialized.
/// - `itemsize`: set to the size of `dtype`.
/// - `ndims`: set to the value of `self.ndims`.
/// - `shape`: allocated on the stack with an array of length `ndims` with uninitialized values.
/// - `strides`: allocated on the stack with an array of length `ndims` with uninitialized
/// values.
#[must_use]
pub fn construct_uninitialized<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
assert!(self.ndims.is_some(), "NDArrayType::construct can only be called on an instance with compile-time known ndims (self.ndims = Some(ndims))");
let Some(ndims) = self.ndims.map(|ndims| self.llvm_usize.const_int(ndims, false)) else {
unreachable!()
};
self.construct_impl(generator, ctx, ndims, name)
}
/// Allocate an [`NDArrayValue`] on the stack given its `ndims` and `dtype`.
///
/// `shape` and `strides` will be automatically allocated onto the stack.
///
/// The returned ndarray's content will be:
/// - `data`: uninitialized.
/// - `itemsize`: set to the size of `dtype`.
/// - `ndims`: set to the value of `ndims`.
/// - `shape`: allocated with an array of length `ndims` with uninitialized values.
/// - `strides`: allocated with an array of length `ndims` with uninitialized values.
#[deprecated = "Prefer construct_uninitialized or construct_*_shape."]
#[must_use]
pub fn construct_dyn_ndims<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
ndims: IntValue<'ctx>,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
assert!(self.ndims.is_none(), "NDArrayType::construct_dyn_ndims can only be called on an instance with compile-time unknown ndims (self.ndims = None)");
self.construct_impl(generator, ctx, ndims, name)
}
/// Convenience function. Allocate an [`NDArrayValue`] with a statically known shape.
///
/// The returned [`NDArrayValue`]'s `data` and `strides` are uninitialized.
#[must_use]
pub fn construct_const_shape<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
shape: &[u64],
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
assert!(self.ndims.is_none_or(|ndims| shape.len() as u64 == ndims));
let ndarray = Self::new(generator, ctx.ctx, self.dtype, Some(shape.len() as u64))
.construct_uninitialized(generator, ctx, name);
let llvm_usize = generator.get_size_type(ctx.ctx);
// Write shape
let ndarray_shape = ndarray.shape();
for (i, dim) in shape.iter().enumerate() {
let dim = llvm_usize.const_int(*dim, false);
unsafe {
ndarray_shape.set_typed_unchecked(
ctx,
generator,
&llvm_usize.const_int(i as u64, false),
dim,
);
}
}
ndarray
}
/// Convenience function. Allocate an [`NDArrayValue`] with a dynamically known shape.
///
/// The returned [`NDArrayValue`]'s `data` and `strides` are uninitialized.
#[must_use]
pub fn construct_dyn_shape<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
shape: &[IntValue<'ctx>],
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
assert!(self.ndims.is_none_or(|ndims| shape.len() as u64 == ndims));
let ndarray = Self::new(generator, ctx.ctx, self.dtype, Some(shape.len() as u64))
.construct_uninitialized(generator, ctx, name);
let llvm_usize = generator.get_size_type(ctx.ctx);
// Write shape
let ndarray_shape = ndarray.shape();
for (i, dim) in shape.iter().enumerate() {
assert_eq!(
dim.get_type(),
llvm_usize,
"Expected {} but got {}",
llvm_usize.print_to_string(),
dim.get_type().print_to_string()
);
unsafe {
ndarray_shape.set_typed_unchecked(
ctx,
generator,
&llvm_usize.const_int(i as u64, false),
*dim,
);
}
}
ndarray
}
/// Create an unsized ndarray to contain `value`.
#[must_use]
pub fn construct_unsized<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
value: &impl BasicValue<'ctx>,
name: Option<&'ctx str>,
) -> NDArrayValue<'ctx> {
let value = value.as_basic_value_enum();
assert_eq!(value.get_type(), self.dtype);
assert!(self.ndims.is_none_or(|ndims| ndims == 0));
// We have to put the value on the stack to get a data pointer.
let data = ctx.builder.build_alloca(value.get_type(), "construct_unsized").unwrap();
ctx.builder.build_store(data, value).unwrap();
let data = ctx
.builder
.build_pointer_cast(data, ctx.ctx.i8_type().ptr_type(AddressSpace::default()), "")
.unwrap();
let ndarray = Self::new_unsized(generator, ctx.ctx, value.get_type())
.construct_uninitialized(generator, ctx, name);
ctx.builder.build_store(ndarray.ptr_to_data(ctx), data).unwrap();
ndarray
}
/// Converts an existing value into a [`NDArrayValue`].
#[must_use]
pub fn map_value(
@ -199,6 +394,7 @@ impl<'ctx> NDArrayType<'ctx> {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
value,
self.dtype,
self.ndims,
self.llvm_usize,
name,
)

View File

@ -0,0 +1,241 @@
use inkwell::{
context::{AsContextRef, Context},
types::{AnyTypeEnum, BasicType, BasicTypeEnum, IntType, PointerType},
values::{IntValue, PointerValue},
AddressSpace,
};
use itertools::Itertools;
use nac3core_derive::StructFields;
use super::ProxyType;
use crate::codegen::{
irrt,
types::structure::{check_struct_type_matches_fields, StructField, StructFields},
values::{
ndarray::{NDArrayValue, NDIterValue},
ArraySliceValue, ProxyValue,
},
CodeGenContext, CodeGenerator,
};
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
pub struct NDIterType<'ctx> {
ty: PointerType<'ctx>,
llvm_usize: IntType<'ctx>,
}
#[derive(PartialEq, Eq, Clone, Copy, StructFields)]
pub struct NDIterStructFields<'ctx> {
#[value_type(usize)]
pub ndims: StructField<'ctx, IntValue<'ctx>>,
#[value_type(usize.ptr_type(AddressSpace::default()))]
pub shape: StructField<'ctx, PointerValue<'ctx>>,
#[value_type(usize.ptr_type(AddressSpace::default()))]
pub strides: StructField<'ctx, PointerValue<'ctx>>,
#[value_type(usize.ptr_type(AddressSpace::default()))]
pub indices: StructField<'ctx, PointerValue<'ctx>>,
#[value_type(usize)]
pub nth: StructField<'ctx, IntValue<'ctx>>,
#[value_type(i8_type().ptr_type(AddressSpace::default()))]
pub element: StructField<'ctx, PointerValue<'ctx>>,
#[value_type(usize)]
pub size: StructField<'ctx, IntValue<'ctx>>,
}
impl<'ctx> NDIterType<'ctx> {
/// Checks whether `llvm_ty` represents a `nditer` type, returning [Err] if it does not.
pub fn is_representable(
llvm_ty: PointerType<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
let ctx = llvm_ty.get_context();
let llvm_ty = llvm_ty.get_element_type();
let AnyTypeEnum::StructType(llvm_ndarray_ty) = llvm_ty else {
return Err(format!("Expected struct type for `NDIter` type, got {llvm_ty}"));
};
check_struct_type_matches_fields(
Self::fields(ctx, llvm_usize),
llvm_ndarray_ty,
"NDIter",
&[],
)
}
/// Returns an instance of [`StructFields`] containing all field accessors for this type.
#[must_use]
fn fields(ctx: impl AsContextRef<'ctx>, llvm_usize: IntType<'ctx>) -> NDIterStructFields<'ctx> {
NDIterStructFields::new(ctx, llvm_usize)
}
/// See [`NDIterType::fields`].
// TODO: Move this into e.g. StructProxyType
#[must_use]
pub fn get_fields(&self, ctx: impl AsContextRef<'ctx>) -> NDIterStructFields<'ctx> {
Self::fields(ctx, self.llvm_usize)
}
/// Creates an LLVM type corresponding to the expected structure of an `NDIter`.
#[must_use]
fn llvm_type(ctx: &'ctx Context, llvm_usize: IntType<'ctx>) -> PointerType<'ctx> {
let field_tys =
Self::fields(ctx, llvm_usize).into_iter().map(|field| field.1).collect_vec();
ctx.struct_type(&field_tys, false).ptr_type(AddressSpace::default())
}
/// Creates an instance of [`NDIter`].
#[must_use]
pub fn new<G: CodeGenerator + ?Sized>(generator: &G, ctx: &'ctx Context) -> Self {
let llvm_usize = generator.get_size_type(ctx);
let llvm_nditer = Self::llvm_type(ctx, llvm_usize);
Self { ty: llvm_nditer, llvm_usize }
}
/// Creates an [`NDIterType`] from a [`PointerType`] representing an `NDIter`.
#[must_use]
pub fn from_type(ptr_ty: PointerType<'ctx>, llvm_usize: IntType<'ctx>) -> Self {
debug_assert!(Self::is_representable(ptr_ty, llvm_usize).is_ok());
Self { ty: ptr_ty, llvm_usize }
}
/// Returns the type of the `size` field of this `nditer` type.
#[must_use]
pub fn size_type(&self) -> IntType<'ctx> {
self.llvm_usize
}
#[must_use]
pub fn alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
parent: NDArrayValue<'ctx>,
indices: ArraySliceValue<'ctx>,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
self.raw_alloca(generator, ctx, name),
parent,
indices,
self.llvm_usize,
name,
)
}
/// Allocate an [`NDIter`] that iterates through the given `ndarray`.
#[must_use]
pub fn construct<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
ndarray: NDArrayValue<'ctx>,
) -> <Self as ProxyType<'ctx>>::Value {
let nditer = self.raw_alloca(generator, ctx, None);
let ndims = ndarray.load_ndims(ctx);
// The caller has the responsibility to allocate 'indices' for `NDIter`.
let indices =
generator.gen_array_var_alloc(ctx, self.llvm_usize.into(), ndims, None).unwrap();
let nditer = <Self as ProxyType<'ctx>>::Value::from_pointer_value(
nditer,
ndarray,
indices,
self.llvm_usize,
None,
);
irrt::ndarray::call_nac3_nditer_initialize(generator, ctx, nditer, ndarray, indices);
nditer
}
#[must_use]
pub fn map_value(
&self,
value: <<Self as ProxyType<'ctx>>::Value as ProxyValue<'ctx>>::Base,
parent: NDArrayValue<'ctx>,
indices: ArraySliceValue<'ctx>,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
value,
parent,
indices,
self.llvm_usize,
name,
)
}
}
impl<'ctx> ProxyType<'ctx> for NDIterType<'ctx> {
type Base = PointerType<'ctx>;
type Value = NDIterValue<'ctx>;
fn is_type<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
llvm_ty: impl BasicType<'ctx>,
) -> Result<(), String> {
if let BasicTypeEnum::PointerType(ty) = llvm_ty.as_basic_type_enum() {
<Self as ProxyType<'ctx>>::is_representable(generator, ctx, ty)
} else {
Err(format!("Expected pointer type, got {llvm_ty:?}"))
}
}
fn is_representable<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
llvm_ty: Self::Base,
) -> Result<(), String> {
Self::is_representable(llvm_ty, generator.get_size_type(ctx))
}
fn raw_alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
name: Option<&'ctx str>,
) -> <Self::Value as ProxyValue<'ctx>>::Base {
generator
.gen_var_alloc(
ctx,
self.as_base_type().get_element_type().into_struct_type().into(),
name,
)
.unwrap()
}
fn array_alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
size: IntValue<'ctx>,
name: Option<&'ctx str>,
) -> ArraySliceValue<'ctx> {
generator
.gen_array_var_alloc(
ctx,
self.as_base_type().get_element_type().into_struct_type().into(),
size,
name,
)
.unwrap()
}
fn as_base_type(&self) -> Self::Base {
self.ty
}
}
impl<'ctx> From<NDIterType<'ctx>> for PointerType<'ctx> {
fn from(value: NDIterType<'ctx>) -> Self {
value.as_base_type()
}
}

View File

@ -2,7 +2,7 @@ use std::marker::PhantomData;
use inkwell::{
context::AsContextRef,
types::{BasicTypeEnum, IntType},
types::{BasicTypeEnum, IntType, StructType},
values::{BasicValue, BasicValueEnum, IntValue, PointerValue, StructValue},
};
@ -103,6 +103,12 @@ where
StructField { index, name, ty: ty.into(), _value_ty: PhantomData }
}
/// Returns the name of this field.
#[must_use]
pub fn name(&self) -> &'static str {
self.name
}
/// Creates a pointer to this field in an arbitrary structure by performing a `getelementptr i32
/// {idx...}, i32 {self.index}`.
pub fn ptr_by_array_gep(
@ -201,3 +207,49 @@ impl FieldIndexCounter {
v
}
}
type FieldTypeVerifier<'ctx> = dyn Fn(BasicTypeEnum<'ctx>) -> Result<(), String>;
/// Checks whether [`llvm_ty`][StructType] contains the fields described by the given
/// [`StructFields`] instance.
///
/// By default, this function will compare the type of each field in `expected_fields` against
/// `llvm_ty`. To override this behavior for individual fields, pass in overrides to
/// `custom_verifiers`, which will use the specified verifier when a field with the matching field
/// name is being checked.
pub(super) fn check_struct_type_matches_fields<'ctx>(
expected_fields: impl StructFields<'ctx>,
llvm_ty: StructType<'ctx>,
ty_name: &'static str,
custom_verifiers: &[(&str, &FieldTypeVerifier<'ctx>)],
) -> Result<(), String> {
let expected_fields = expected_fields.to_vec();
if llvm_ty.count_fields() != u32::try_from(expected_fields.len()).unwrap() {
return Err(format!(
"Expected {} fields in `{ty_name}`, got {}",
expected_fields.len(),
llvm_ty.count_fields(),
));
}
expected_fields
.into_iter()
.enumerate()
.map(|(i, (field_name, expected_ty))| {
(field_name, expected_ty, llvm_ty.get_field_type_at_index(i as u32).unwrap())
})
.try_for_each(|(field_name, expected_ty, actual_ty)| {
if let Some((_, verifier)) =
custom_verifiers.iter().find(|verifier| verifier.0 == field_name)
{
verifier(actual_ty)
} else if expected_ty == actual_ty {
Ok(())
} else {
Err(format!("Expected {expected_ty} for `{ty_name}.{field_name}`, got {actual_ty}"))
}
})?;
Ok(())
}

View File

@ -0,0 +1,3 @@
pub use slice::*;
mod slice;

View File

@ -0,0 +1,254 @@
use inkwell::{
context::{AsContextRef, Context, ContextRef},
types::{AnyTypeEnum, BasicType, BasicTypeEnum, IntType, PointerType},
values::IntValue,
AddressSpace,
};
use itertools::Itertools;
use nac3core_derive::StructFields;
use crate::codegen::{
types::{
structure::{
check_struct_type_matches_fields, FieldIndexCounter, StructField, StructFields,
},
ProxyType,
},
values::{utils::SliceValue, ArraySliceValue, ProxyValue},
CodeGenContext, CodeGenerator,
};
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
pub struct SliceType<'ctx> {
ty: PointerType<'ctx>,
int_ty: IntType<'ctx>,
llvm_usize: IntType<'ctx>,
}
#[derive(PartialEq, Eq, Clone, Copy, StructFields)]
pub struct SliceFields<'ctx> {
#[value_type(bool_type())]
pub start_defined: StructField<'ctx, IntValue<'ctx>>,
#[value_type(usize)]
pub start: StructField<'ctx, IntValue<'ctx>>,
#[value_type(bool_type())]
pub stop_defined: StructField<'ctx, IntValue<'ctx>>,
#[value_type(usize)]
pub stop: StructField<'ctx, IntValue<'ctx>>,
#[value_type(bool_type())]
pub step_defined: StructField<'ctx, IntValue<'ctx>>,
#[value_type(usize)]
pub step: StructField<'ctx, IntValue<'ctx>>,
}
impl<'ctx> SliceFields<'ctx> {
/// Creates a new instance of [`SliceFields`] with a custom integer type for its range values.
#[must_use]
pub fn new_sized(ctx: &impl AsContextRef<'ctx>, int_ty: IntType<'ctx>) -> Self {
let ctx = unsafe { ContextRef::new(ctx.as_ctx_ref()) };
let mut counter = FieldIndexCounter::default();
SliceFields {
start_defined: StructField::create(&mut counter, "start_defined", ctx.bool_type()),
start: StructField::create(&mut counter, "start", int_ty),
stop_defined: StructField::create(&mut counter, "stop_defined", ctx.bool_type()),
stop: StructField::create(&mut counter, "stop", int_ty),
step_defined: StructField::create(&mut counter, "step_defined", ctx.bool_type()),
step: StructField::create(&mut counter, "step", int_ty),
}
}
}
impl<'ctx> SliceType<'ctx> {
/// Checks whether `llvm_ty` represents a `slice` type, returning [Err] if it does not.
pub fn is_representable(
llvm_ty: PointerType<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
let ctx = llvm_ty.get_context();
let fields = SliceFields::new(ctx, llvm_usize);
let llvm_ty = llvm_ty.get_element_type();
let AnyTypeEnum::StructType(llvm_ty) = llvm_ty else {
return Err(format!("Expected struct type for `Slice` type, got {llvm_ty}"));
};
check_struct_type_matches_fields(
fields,
llvm_ty,
"Slice",
&[
(fields.start.name(), &|ty| {
if ty.is_int_type() {
Ok(())
} else {
Err(format!("Expected int type for `Slice.start`, got {ty}"))
}
}),
(fields.stop.name(), &|ty| {
if ty.is_int_type() {
Ok(())
} else {
Err(format!("Expected int type for `Slice.stop`, got {ty}"))
}
}),
(fields.step.name(), &|ty| {
if ty.is_int_type() {
Ok(())
} else {
Err(format!("Expected int type for `Slice.step`, got {ty}"))
}
}),
],
)
}
// TODO: Move this into e.g. StructProxyType
#[must_use]
pub fn get_fields(&self) -> SliceFields<'ctx> {
SliceFields::new_sized(&self.int_ty.get_context(), self.int_ty)
}
/// Creates an LLVM type corresponding to the expected structure of a `Slice`.
#[must_use]
fn llvm_type(ctx: &'ctx Context, int_ty: IntType<'ctx>) -> PointerType<'ctx> {
let field_tys = SliceFields::new_sized(&int_ty.get_context(), int_ty)
.into_iter()
.map(|field| field.1)
.collect_vec();
ctx.struct_type(&field_tys, false).ptr_type(AddressSpace::default())
}
/// Creates an instance of [`SliceType`] with `int_ty` as its backing integer type.
#[must_use]
pub fn new(ctx: &'ctx Context, int_ty: IntType<'ctx>, llvm_usize: IntType<'ctx>) -> Self {
let llvm_ty = Self::llvm_type(ctx, int_ty);
Self { ty: llvm_ty, int_ty, llvm_usize }
}
/// Creates an instance of [`SliceType`] with `usize` as its backing integer type.
#[must_use]
pub fn new_usize<G: CodeGenerator + ?Sized>(generator: &G, ctx: &'ctx Context) -> Self {
let llvm_usize = generator.get_size_type(ctx);
Self::new(ctx, llvm_usize, llvm_usize)
}
/// Creates an [`SliceType`] from a [`PointerType`] representing a `slice`.
#[must_use]
pub fn from_type(
ptr_ty: PointerType<'ctx>,
int_ty: IntType<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Self {
debug_assert!(Self::is_representable(ptr_ty, int_ty).is_ok());
Self { ty: ptr_ty, int_ty, llvm_usize }
}
#[must_use]
pub fn element_type(&self) -> IntType<'ctx> {
self.int_ty
}
/// Allocates an instance of [`ContiguousNDArrayValue`] as if by calling `alloca` on the base type.
#[must_use]
pub fn alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
self.raw_alloca(generator, ctx, name),
self.int_ty,
self.llvm_usize,
name,
)
}
/// Converts an existing value into a [`ContiguousNDArrayValue`].
#[must_use]
pub fn map_value(
&self,
value: <<Self as ProxyType<'ctx>>::Value as ProxyValue<'ctx>>::Base,
name: Option<&'ctx str>,
) -> <Self as ProxyType<'ctx>>::Value {
<Self as ProxyType<'ctx>>::Value::from_pointer_value(
value,
self.int_ty,
self.llvm_usize,
name,
)
}
}
impl<'ctx> ProxyType<'ctx> for SliceType<'ctx> {
type Base = PointerType<'ctx>;
type Value = SliceValue<'ctx>;
fn is_type<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
llvm_ty: impl BasicType<'ctx>,
) -> Result<(), String> {
if let BasicTypeEnum::PointerType(ty) = llvm_ty.as_basic_type_enum() {
<Self as ProxyType<'ctx>>::is_representable(generator, ctx, ty)
} else {
Err(format!("Expected pointer type, got {llvm_ty:?}"))
}
}
fn is_representable<G: CodeGenerator + ?Sized>(
generator: &G,
ctx: &'ctx Context,
llvm_ty: Self::Base,
) -> Result<(), String> {
Self::is_representable(llvm_ty, generator.get_size_type(ctx))
}
fn raw_alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
name: Option<&'ctx str>,
) -> <Self::Value as ProxyValue<'ctx>>::Base {
generator
.gen_var_alloc(
ctx,
self.as_base_type().get_element_type().into_struct_type().into(),
name,
)
.unwrap()
}
fn array_alloca<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
size: IntValue<'ctx>,
name: Option<&'ctx str>,
) -> ArraySliceValue<'ctx> {
generator
.gen_array_var_alloc(
ctx,
self.as_base_type().get_element_type().into_struct_type().into(),
size,
name,
)
.unwrap()
}
fn as_base_type(&self) -> Self::Base {
self.ty
}
}
impl<'ctx> From<SliceType<'ctx>> for PointerType<'ctx> {
fn from(value: SliceType<'ctx>) -> Self {
value.as_base_type()
}
}

View File

@ -207,7 +207,7 @@ pub trait TypedArrayLikeMutator<'ctx, T, Index = IntValue<'ctx>>:
/// Type alias for a function that casts a [`BasicValueEnum`] into a `T`.
type ValueDowncastFn<'ctx, T> =
Box<dyn Fn(&mut CodeGenContext<'ctx, '_>, BasicValueEnum<'ctx>) -> T>;
Box<dyn Fn(&mut CodeGenContext<'ctx, '_>, BasicValueEnum<'ctx>) -> T + 'ctx>;
/// Type alias for a function that casts a `T` into a [`BasicValueEnum`].
type ValueUpcastFn<'ctx, T> = Box<dyn Fn(&mut CodeGenContext<'ctx, '_>, T) -> BasicValueEnum<'ctx>>;

View File

@ -4,13 +4,13 @@ use super::types::ProxyType;
use crate::codegen::CodeGenerator;
pub use array::*;
pub use list::*;
pub use ndarray::*;
pub use range::*;
mod array;
mod list;
mod ndarray;
pub mod ndarray;
mod range;
pub mod utils;
/// A LLVM type that is used to represent a non-primitive value in NAC3.
pub trait ProxyValue<'ctx>: Into<Self::Base> {

View File

@ -0,0 +1,202 @@
use inkwell::{
types::{BasicType, BasicTypeEnum, IntType},
values::{IntValue, PointerValue},
AddressSpace,
};
use super::{ArrayLikeValue, NDArrayValue, ProxyValue};
use crate::codegen::{
stmt::gen_if_callback,
types::{
ndarray::{ContiguousNDArrayType, NDArrayType},
structure::StructField,
},
CodeGenContext, CodeGenerator,
};
#[derive(Copy, Clone)]
pub struct ContiguousNDArrayValue<'ctx> {
value: PointerValue<'ctx>,
item: BasicTypeEnum<'ctx>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
}
impl<'ctx> ContiguousNDArrayValue<'ctx> {
/// Checks whether `value` is an instance of `ContiguousNDArray`, returning [Err] if `value` is
/// not an instance.
pub fn is_representable(
value: PointerValue<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
<Self as ProxyValue<'ctx>>::Type::is_representable(value.get_type(), llvm_usize)
}
/// Creates an [`ContiguousNDArrayValue`] from a [`PointerValue`].
#[must_use]
pub fn from_pointer_value(
ptr: PointerValue<'ctx>,
dtype: BasicTypeEnum<'ctx>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
) -> Self {
debug_assert!(Self::is_representable(ptr, llvm_usize).is_ok());
Self { value: ptr, item: dtype, llvm_usize, name }
}
fn ndims_field(&self) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields().ndims
}
pub fn store_ndims(&self, ctx: &CodeGenContext<'ctx, '_>, value: IntValue<'ctx>) {
self.ndims_field().set(ctx, self.as_base_value(), value, self.name);
}
fn shape_field(&self) -> StructField<'ctx, PointerValue<'ctx>> {
self.get_type().get_fields().shape
}
pub fn store_shape(&self, ctx: &CodeGenContext<'ctx, '_>, value: PointerValue<'ctx>) {
self.shape_field().set(ctx, self.as_base_value(), value, self.name);
}
pub fn load_shape(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
self.shape_field().get(ctx, self.value, self.name)
}
fn data_field(&self) -> StructField<'ctx, PointerValue<'ctx>> {
self.get_type().get_fields().data
}
pub fn store_data(&self, ctx: &CodeGenContext<'ctx, '_>, value: PointerValue<'ctx>) {
self.data_field().set(ctx, self.as_base_value(), value, self.name);
}
pub fn load_data(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
self.data_field().get(ctx, self.value, self.name)
}
}
impl<'ctx> ProxyValue<'ctx> for ContiguousNDArrayValue<'ctx> {
type Base = PointerValue<'ctx>;
type Type = ContiguousNDArrayType<'ctx>;
fn get_type(&self) -> Self::Type {
<Self as ProxyValue<'ctx>>::Type::from_type(
self.as_base_value().get_type(),
self.item,
self.llvm_usize,
)
}
fn as_base_value(&self) -> Self::Base {
self.value
}
}
impl<'ctx> From<ContiguousNDArrayValue<'ctx>> for PointerValue<'ctx> {
fn from(value: ContiguousNDArrayValue<'ctx>) -> Self {
value.as_base_value()
}
}
impl<'ctx> NDArrayValue<'ctx> {
/// Create a [`ContiguousNDArrayValue`] from the contents of this ndarray.
///
/// This function may or may not be expensive depending on if this ndarray has contiguous data.
///
/// If this ndarray is not C-contiguous, this function will allocate memory on the stack for the
/// `data` field of the returned [`ContiguousNDArrayValue`] and copy contents of this ndarray to
/// there.
///
/// If this ndarray is C-contiguous, contents of this ndarray will not be copied. The created
/// [`ContiguousNDArrayValue`] will share memory with this ndarray.
pub fn make_contiguous_ndarray<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
) -> ContiguousNDArrayValue<'ctx> {
let result = ContiguousNDArrayType::new(generator, ctx.ctx, self.dtype)
.alloca(generator, ctx, self.name);
// Set ndims and shape.
let ndims = self
.ndims
.map_or_else(|| self.load_ndims(ctx), |ndims| self.llvm_usize.const_int(ndims, false));
result.store_ndims(ctx, ndims);
let shape = self.shape();
result.store_shape(ctx, shape.base_ptr(ctx, generator));
gen_if_callback(
generator,
ctx,
|generator, ctx| Ok(self.is_c_contiguous(generator, ctx)),
|_, ctx| {
// This ndarray is contiguous.
let data = self.data_field(ctx).get(ctx, self.as_base_value(), self.name);
let data = ctx
.builder
.build_pointer_cast(data, result.item.ptr_type(AddressSpace::default()), "")
.unwrap();
result.store_data(ctx, data);
Ok(())
},
|generator, ctx| {
// This ndarray is not contiguous. Do a full-copy on `data`. `make_copy` produces an
// ndarray with contiguous `data`.
let copied_ndarray = self.make_copy(generator, ctx);
let data = copied_ndarray.data().base_ptr(ctx, generator);
let data = ctx
.builder
.build_pointer_cast(data, result.item.ptr_type(AddressSpace::default()), "")
.unwrap();
result.store_data(ctx, data);
Ok(())
},
)
.unwrap();
result
}
/// Create an [`NDArrayValue`] from a [`ContiguousNDArrayValue`].
///
/// The operation is cheap. The newly created [`NDArrayValue`] will share the same memory as the
/// [`ContiguousNDArrayValue`].
///
/// `ndims` has to be provided as [`NDArrayValue`] requires a statically known `ndims` value,
/// despite the fact that the information should be contained within the
/// [`ContiguousNDArrayValue`].
pub fn from_contiguous_ndarray<G: CodeGenerator + ?Sized>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
carray: ContiguousNDArrayValue<'ctx>,
ndims: u64,
) -> Self {
// TODO: Debug assert `ndims == carray.ndims` to catch bugs.
// Allocate the resulting ndarray.
let ndarray = NDArrayType::new(generator, ctx.ctx, carray.item, Some(ndims))
.construct_uninitialized(generator, ctx, carray.name);
// Copy shape and update strides
let shape = carray.load_shape(ctx);
ndarray.copy_shape_from_array(generator, ctx, shape);
ndarray.set_strides_contiguous(generator, ctx);
// Share data
let data = carray.load_data(ctx);
ndarray.store_data(
ctx,
ctx.builder
.build_pointer_cast(data, ctx.ctx.i8_type().ptr_type(AddressSpace::default()), "")
.unwrap(),
);
ndarray
}
}

View File

@ -0,0 +1,262 @@
use inkwell::{
types::IntType,
values::{IntValue, PointerValue},
AddressSpace,
};
use itertools::Itertools;
use nac3parser::ast::{Expr, ExprKind};
use crate::{
codegen::{
irrt,
types::{
ndarray::{NDArrayType, NDIndexType},
structure::StructField,
utils::SliceType,
},
values::{ndarray::NDArrayValue, utils::RustSlice, ProxyValue},
CodeGenContext, CodeGenerator,
},
typecheck::typedef::Type,
};
/// An IRRT representation of an ndarray subscript index.
#[derive(Copy, Clone)]
pub struct NDIndexValue<'ctx> {
value: PointerValue<'ctx>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
}
impl<'ctx> NDIndexValue<'ctx> {
/// Checks whether `value` is an instance of `ndindex`, returning [Err] if `value` is not an
/// instance.
pub fn is_representable(
value: PointerValue<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
<Self as ProxyValue<'ctx>>::Type::is_representable(value.get_type(), llvm_usize)
}
/// Creates an [`NDIndexValue`] from a [`PointerValue`].
#[must_use]
pub fn from_pointer_value(
ptr: PointerValue<'ctx>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
) -> Self {
debug_assert!(Self::is_representable(ptr, llvm_usize).is_ok());
Self { value: ptr, llvm_usize, name }
}
fn type_field(&self) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields().type_
}
pub fn load_type(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.type_field().get(ctx, self.value, self.name)
}
pub fn store_type(&self, ctx: &CodeGenContext<'ctx, '_>, value: IntValue<'ctx>) {
self.type_field().set(ctx, self.value, value, self.name);
}
fn data_field(&self) -> StructField<'ctx, PointerValue<'ctx>> {
self.get_type().get_fields().data
}
pub fn load_data(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
self.data_field().get(ctx, self.value, self.name)
}
pub fn store_data(&self, ctx: &CodeGenContext<'ctx, '_>, value: PointerValue<'ctx>) {
self.data_field().set(ctx, self.value, value, self.name);
}
}
impl<'ctx> ProxyValue<'ctx> for NDIndexValue<'ctx> {
type Base = PointerValue<'ctx>;
type Type = NDIndexType<'ctx>;
fn get_type(&self) -> Self::Type {
Self::Type::from_type(self.value.get_type(), self.llvm_usize)
}
fn as_base_value(&self) -> Self::Base {
self.value
}
}
impl<'ctx> From<NDIndexValue<'ctx>> for PointerValue<'ctx> {
fn from(value: NDIndexValue<'ctx>) -> Self {
value.as_base_value()
}
}
impl<'ctx> NDArrayValue<'ctx> {
/// Get the expected `ndims` after indexing with `indices`.
#[must_use]
fn deduce_ndims_after_indexing_with(&self, indices: &[RustNDIndex<'ctx>]) -> Option<u64> {
let mut ndims = self.ndims?;
for index in indices {
match index {
RustNDIndex::SingleElement(_) => {
ndims -= 1; // Single elements decrements ndims
}
RustNDIndex::NewAxis => {
ndims += 1; // `np.newaxis` / `none` adds a new axis
}
RustNDIndex::Ellipsis | RustNDIndex::Slice(_) => {}
}
}
Some(ndims)
}
/// Index into the ndarray, and return a newly-allocated view on this ndarray.
///
/// This function behaves like NumPy's ndarray indexing, but if the indices index
/// into a single element, an unsized ndarray is returned.
#[must_use]
pub fn index<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
indices: &[RustNDIndex<'ctx>],
) -> Self {
assert!(self.ndims.is_some(), "NDArrayValue::index is only supported for instances with compile-time known ndims (self.ndims = Some(...))");
let dst_ndims = self.deduce_ndims_after_indexing_with(indices);
let dst_ndarray = NDArrayType::new(generator, ctx.ctx, self.dtype, dst_ndims)
.construct_uninitialized(generator, ctx, None);
let indices =
NDIndexType::new(generator, ctx.ctx).construct_ndindices(generator, ctx, indices);
irrt::ndarray::call_nac3_ndarray_index(generator, ctx, indices, *self, dst_ndarray);
dst_ndarray
}
}
/// A convenience enum representing a [`NDIndexValue`].
// TODO: Rename to CTConstNDIndex
#[derive(Debug, Clone)]
pub enum RustNDIndex<'ctx> {
SingleElement(IntValue<'ctx>),
Slice(RustSlice<'ctx>),
NewAxis,
Ellipsis,
}
impl<'ctx> RustNDIndex<'ctx> {
/// Generate LLVM code to transform an ndarray subscript expression to
/// its list of [`RustNDIndex`]
///
/// i.e.,
/// ```python
/// my_ndarray[::3, 1, :2:]
/// ^^^^^^^^^^^ Then these into a three `RustNDIndex`es
/// ```
pub fn from_subscript_expr<G: CodeGenerator>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
subscript: &Expr<Option<Type>>,
) -> Result<Vec<RustNDIndex<'ctx>>, String> {
// Annoying notes about `slice`
// - `my_array[5]`
// - slice is a `Constant`
// - `my_array[:5]`
// - slice is a `Slice`
// - `my_array[:]`
// - slice is a `Slice`, but lower upper step would all be `Option::None`
// - `my_array[:, :]`
// - slice is now a `Tuple` of two `Slice`-s
//
// In summary:
// - when there is a comma "," within [], `slice` will be a `Tuple` of the entries.
// - when there is not comma "," within [] (i.e., just a single entry), `slice` will be that entry itself.
//
// So we first "flatten" out the slice expression
let index_exprs = match &subscript.node {
ExprKind::Tuple { elts, .. } => elts.iter().collect_vec(),
_ => vec![subscript],
};
// Process all index expressions
let mut rust_ndindices: Vec<RustNDIndex> = Vec::with_capacity(index_exprs.len()); // Not using iterators here because `?` is used here.
for index_expr in index_exprs {
// NOTE: Currently nac3core's slices do not have an object representation,
// so the code/implementation looks awkward - we have to do pattern matching on the expression
let ndindex = if let ExprKind::Slice { lower, upper, step } = &index_expr.node {
// Handle slices
let slice = RustSlice::from_slice_expr(generator, ctx, lower, upper, step)?;
RustNDIndex::Slice(slice)
} else {
// Treat and handle everything else as a single element index.
let index = generator.gen_expr(ctx, index_expr)?.unwrap().to_basic_value_enum(
ctx,
generator,
ctx.primitives.int32, // Must be int32, this checks for illegal values
)?;
let index = index.into_int_value();
RustNDIndex::SingleElement(index)
};
rust_ndindices.push(ndindex);
}
Ok(rust_ndindices)
}
/// Get the value to set `NDIndex::type` for this variant.
#[must_use]
pub fn get_type_id(&self) -> u64 {
// Defined in IRRT, must be in sync
match self {
RustNDIndex::SingleElement(_) => 0,
RustNDIndex::Slice(_) => 1,
RustNDIndex::NewAxis => 2,
RustNDIndex::Ellipsis => 3,
}
}
/// Serialize this [`RustNDIndex`] by writing it into an LLVM [`NDIndexValue`].
pub fn write_to_ndindex<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
dst_ndindex: NDIndexValue<'ctx>,
) {
let llvm_pi8 = ctx.ctx.i8_type().ptr_type(AddressSpace::default());
// Set `dst_ndindex.type`
dst_ndindex.store_type(ctx, ctx.ctx.i8_type().const_int(self.get_type_id(), false));
// Set `dst_ndindex_ptr->data`
match self {
RustNDIndex::SingleElement(in_index) => {
let index_ptr = ctx.builder.build_alloca(ctx.ctx.i32_type(), "").unwrap();
ctx.builder.build_store(index_ptr, *in_index).unwrap();
dst_ndindex.store_data(
ctx,
ctx.builder.build_pointer_cast(index_ptr, llvm_pi8, "").unwrap(),
);
}
RustNDIndex::Slice(in_rust_slice) => {
let user_slice_ptr =
SliceType::new(ctx.ctx, ctx.ctx.i32_type(), generator.get_size_type(ctx.ctx))
.alloca(generator, ctx, None);
in_rust_slice.write_to_slice(ctx, user_slice_ptr);
dst_ndindex.store_data(
ctx,
ctx.builder.build_pointer_cast(user_slice_ptr.into(), llvm_pi8, "").unwrap(),
);
}
RustNDIndex::NewAxis | RustNDIndex::Ellipsis => {}
}
}
}

View File

@ -9,18 +9,29 @@ use super::{
UntypedArrayLikeAccessor, UntypedArrayLikeMutator,
};
use crate::codegen::{
irrt::{call_ndarray_calc_size, call_ndarray_flatten_index},
llvm_intrinsics::call_int_umin,
irrt,
llvm_intrinsics::{call_int_umin, call_memcpy_generic_array},
stmt::gen_for_callback_incrementing,
types::{structure::StructField, NDArrayType},
type_aligned_alloca,
types::{ndarray::NDArrayType, structure::StructField},
CodeGenContext, CodeGenerator,
};
pub use contiguous::*;
pub use indexing::*;
pub use nditer::*;
pub use view::*;
mod contiguous;
mod indexing;
mod nditer;
mod view;
/// Proxy type for accessing an `NDArray` value in LLVM.
#[derive(Copy, Clone)]
pub struct NDArrayValue<'ctx> {
value: PointerValue<'ctx>,
dtype: BasicTypeEnum<'ctx>,
ndims: Option<u64>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
}
@ -40,12 +51,13 @@ impl<'ctx> NDArrayValue<'ctx> {
pub fn from_pointer_value(
ptr: PointerValue<'ctx>,
dtype: BasicTypeEnum<'ctx>,
ndims: Option<u64>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
) -> Self {
debug_assert!(Self::is_representable(ptr, llvm_usize).is_ok());
NDArrayValue { value: ptr, dtype, llvm_usize, name }
NDArrayValue { value: ptr, dtype, ndims, llvm_usize, name }
}
fn ndims_field(&self, ctx: &CodeGenContext<'ctx, '_>) -> StructField<'ctx, IntValue<'ctx>> {
@ -76,6 +88,27 @@ impl<'ctx> NDArrayValue<'ctx> {
ctx.builder.build_load(pndims, "").map(BasicValueEnum::into_int_value).unwrap()
}
fn itemsize_field(&self, ctx: &CodeGenContext<'ctx, '_>) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields(ctx.ctx).itemsize
}
/// Stores the size of each element `itemsize` into this instance.
pub fn store_itemsize<G: CodeGenerator + ?Sized>(
&self,
ctx: &CodeGenContext<'ctx, '_>,
generator: &G,
itemsize: IntValue<'ctx>,
) {
debug_assert_eq!(itemsize.get_type(), generator.get_size_type(ctx.ctx));
self.itemsize_field(ctx).set(ctx, self.value, itemsize, self.name);
}
/// Returns the size of each element of this `NDArray` as a value.
pub fn load_itemsize(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.itemsize_field(ctx).get(ctx, self.value, self.name)
}
fn shape_field(&self, ctx: &CodeGenContext<'ctx, '_>) -> StructField<'ctx, PointerValue<'ctx>> {
self.get_type().get_fields(ctx.ctx).shape
}
@ -107,6 +140,40 @@ impl<'ctx> NDArrayValue<'ctx> {
NDArrayShapeProxy(self)
}
fn strides_field(
&self,
ctx: &CodeGenContext<'ctx, '_>,
) -> StructField<'ctx, PointerValue<'ctx>> {
self.get_type().get_fields(ctx.ctx).strides
}
/// Returns the double-indirection pointer to the `strides` array, as if by calling
/// `getelementptr` on the field.
fn ptr_to_strides(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
self.strides_field(ctx).ptr_by_gep(ctx, self.value, self.name)
}
/// Stores the array of stride sizes `strides` into this instance.
fn store_strides(&self, ctx: &CodeGenContext<'ctx, '_>, strides: PointerValue<'ctx>) {
self.strides_field(ctx).set(ctx, self.as_base_value(), strides, self.name);
}
/// Convenience method for creating a new array storing the stride with the given `size`.
pub fn create_strides(
&self,
ctx: &CodeGenContext<'ctx, '_>,
llvm_usize: IntType<'ctx>,
size: IntValue<'ctx>,
) {
self.store_strides(ctx, ctx.builder.build_array_alloca(llvm_usize, size, "").unwrap());
}
/// Returns a proxy object to the field storing the stride of each dimension of this `NDArray`.
#[must_use]
pub fn strides(&self) -> NDArrayStridesProxy<'ctx, '_> {
NDArrayStridesProxy(self)
}
fn data_field(&self, ctx: &CodeGenContext<'ctx, '_>) -> StructField<'ctx, PointerValue<'ctx>> {
self.get_type().get_fields(ctx.ctx).data
}
@ -128,23 +195,23 @@ impl<'ctx> NDArrayValue<'ctx> {
/// Convenience method for creating a new array storing data elements with the given element
/// type `elem_ty` and `size`.
pub fn create_data(
///
/// The data buffer will be allocated on the stack, and is considered to be owned by this ndarray instance.
///
/// # Safety
///
/// The caller must ensure that `shape` and `itemsize` of this ndarray instance is initialized.
pub unsafe fn create_data<G: CodeGenerator + ?Sized>(
&self,
ctx: &CodeGenContext<'ctx, '_>,
elem_ty: BasicTypeEnum<'ctx>,
size: IntValue<'ctx>,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
) {
let itemsize = ctx
.builder
.build_int_z_extend_or_bit_cast(elem_ty.size_of().unwrap(), size.get_type(), "")
.unwrap();
let nbytes = ctx.builder.build_int_mul(size, itemsize, "").unwrap();
let nbytes = self.nbytes(generator, ctx);
// TODO: What about alignment?
self.store_data(
ctx,
ctx.builder.build_array_alloca(ctx.ctx.i8_type(), nbytes, "").unwrap(),
);
let data = type_aligned_alloca(generator, ctx, self.dtype, nbytes, None);
self.store_data(ctx, data);
self.set_strides_contiguous(generator, ctx);
}
/// Returns a proxy object to the field storing the data of this `NDArray`.
@ -152,6 +219,219 @@ impl<'ctx> NDArrayValue<'ctx> {
pub fn data(&self) -> NDArrayDataProxy<'ctx, '_> {
NDArrayDataProxy(self)
}
/// Copy shape dimensions from an array.
pub fn copy_shape_from_array<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
shape: PointerValue<'ctx>,
) {
let num_items = self.load_ndims(ctx);
call_memcpy_generic_array(
ctx,
self.shape().base_ptr(ctx, generator),
shape,
num_items,
ctx.ctx.bool_type().const_zero(),
);
}
/// Copy shape dimensions from an ndarray.
/// Panics if `ndims` mismatches.
pub fn copy_shape_from_ndarray<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
src_ndarray: NDArrayValue<'ctx>,
) {
if self.ndims.is_some() && src_ndarray.ndims.is_some() {
assert_eq!(self.ndims, src_ndarray.ndims);
} else {
let self_ndims = self.load_ndims(ctx);
let src_ndims = src_ndarray.load_ndims(ctx);
ctx.make_assert(
generator,
ctx.builder.build_int_compare(
IntPredicate::EQ,
self_ndims,
src_ndims,
""
).unwrap(),
"0:AssertionError",
"NDArrayValue::copy_shape_from_ndarray: Expected self.ndims ({0}) == src_ndarray.ndims ({1})",
[Some(self_ndims), Some(src_ndims), None],
ctx.current_loc
);
}
let src_shape = src_ndarray.shape().base_ptr(ctx, generator);
self.copy_shape_from_array(generator, ctx, src_shape);
}
/// Copy strides dimensions from an array.
pub fn copy_strides_from_array<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
strides: PointerValue<'ctx>,
) {
let num_items = self.load_ndims(ctx);
call_memcpy_generic_array(
ctx,
self.strides().base_ptr(ctx, generator),
strides,
num_items,
ctx.ctx.bool_type().const_zero(),
);
}
/// Copy strides dimensions from an ndarray.
/// Panics if `ndims` mismatches.
pub fn copy_strides_from_ndarray<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
src_ndarray: NDArrayValue<'ctx>,
) {
if self.ndims.is_some() && src_ndarray.ndims.is_some() {
assert_eq!(self.ndims, src_ndarray.ndims);
} else {
let self_ndims = self.load_ndims(ctx);
let src_ndims = src_ndarray.load_ndims(ctx);
ctx.make_assert(
generator,
ctx.builder.build_int_compare(
IntPredicate::EQ,
self_ndims,
src_ndims,
""
).unwrap(),
"0:AssertionError",
"NDArrayValue::copy_shape_from_ndarray: Expected self.ndims ({0}) == src_ndarray.ndims ({1})",
[Some(self_ndims), Some(src_ndims), None],
ctx.current_loc
);
}
let src_strides = src_ndarray.strides().base_ptr(ctx, generator);
self.copy_strides_from_array(generator, ctx, src_strides);
}
/// Get the `np.size()` of this ndarray.
pub fn size<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
) -> IntValue<'ctx> {
irrt::ndarray::call_nac3_ndarray_size(generator, ctx, *self)
}
/// Get the `ndarray.nbytes` of this ndarray.
pub fn nbytes<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
) -> IntValue<'ctx> {
irrt::ndarray::call_nac3_ndarray_nbytes(generator, ctx, *self)
}
/// Get the `len()` of this ndarray.
pub fn len<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
) -> IntValue<'ctx> {
irrt::ndarray::call_nac3_ndarray_len(generator, ctx, *self)
}
/// Check if this ndarray is C-contiguous.
///
/// See NumPy's `flags["C_CONTIGUOUS"]`: <https://numpy.org/doc/stable/reference/generated/numpy.ndarray.flags.html#numpy.ndarray.flags>
pub fn is_c_contiguous<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
) -> IntValue<'ctx> {
irrt::ndarray::call_nac3_ndarray_is_c_contiguous(generator, ctx, *self)
}
/// Call [`call_nac3_ndarray_set_strides_by_shape`] on this ndarray to update `strides`.
///
/// Update the ndarray's strides to make the ndarray contiguous.
pub fn set_strides_contiguous<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
) {
irrt::ndarray::call_nac3_ndarray_set_strides_by_shape(generator, ctx, *self);
}
#[must_use]
pub fn make_copy<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
) -> Self {
let clone = if self.ndims.is_some() {
self.get_type().construct_uninitialized(generator, ctx, None)
} else {
self.get_type().construct_dyn_ndims(generator, ctx, self.load_ndims(ctx), None)
};
let shape = self.shape();
clone.copy_shape_from_array(generator, ctx, shape.base_ptr(ctx, generator));
unsafe { clone.create_data(generator, ctx) };
clone.copy_data_from(generator, ctx, *self);
clone
}
/// Copy data from another ndarray.
///
/// This ndarray and `src` is that their `np.size()` should be the same. Their shapes
/// do not matter. The copying order is determined by how their flattened views look.
///
/// Panics if the `dtype`s of ndarrays are different.
pub fn copy_data_from<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
src: NDArrayValue<'ctx>,
) {
assert_eq!(self.dtype, src.dtype, "self and src dtype should match");
irrt::ndarray::call_nac3_ndarray_copy_data(generator, ctx, src, *self);
}
/// Returns true if this ndarray is unsized - `ndims == 0` and only contains a scalar.
#[must_use]
pub fn is_unsized(&self) -> Option<bool> {
self.ndims.map(|ndims| ndims == 0)
}
/// If this ndarray is unsized, return its sole value as an [`AnyObject`].
/// Otherwise, do nothing and return the ndarray itself.
// TODO: Rename to get_unsized_element
pub fn split_unsized<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
) -> ScalarOrNDArray<'ctx> {
let Some(is_unsized) = self.is_unsized() else { todo!() };
if is_unsized {
// NOTE: `np.size(self) == 0` here is never possible.
let zero = generator.get_size_type(ctx.ctx).const_zero();
let value = unsafe { self.data().get_unchecked(ctx, generator, &zero, None) };
ScalarOrNDArray::Scalar(value)
} else {
ScalarOrNDArray::NDArray(*self)
}
}
}
impl<'ctx> ProxyValue<'ctx> for NDArrayValue<'ctx> {
@ -159,7 +439,12 @@ impl<'ctx> ProxyValue<'ctx> for NDArrayValue<'ctx> {
type Type = NDArrayType<'ctx>;
fn get_type(&self) -> Self::Type {
NDArrayType::from_type(self.as_base_value().get_type(), self.dtype, self.llvm_usize)
NDArrayType::from_type(
self.as_base_value().get_type(),
self.dtype,
self.ndims,
self.llvm_usize,
)
}
fn as_base_value(&self) -> Self::Base {
@ -173,7 +458,7 @@ impl<'ctx> From<NDArrayValue<'ctx>> for PointerValue<'ctx> {
}
}
/// Proxy type for accessing the `dims` array of an `NDArray` instance in LLVM.
/// Proxy type for accessing the `shape` array of an `NDArray` instance in LLVM.
#[derive(Copy, Clone)]
pub struct NDArrayShapeProxy<'ctx, 'a>(&'a NDArrayValue<'ctx>);
@ -265,6 +550,98 @@ impl<'ctx> TypedArrayLikeMutator<'ctx, IntValue<'ctx>> for NDArrayShapeProxy<'ct
}
}
/// Proxy type for accessing the `strides` array of an `NDArray` instance in LLVM.
#[derive(Copy, Clone)]
pub struct NDArrayStridesProxy<'ctx, 'a>(&'a NDArrayValue<'ctx>);
impl<'ctx> ArrayLikeValue<'ctx> for NDArrayStridesProxy<'ctx, '_> {
fn element_type<G: CodeGenerator + ?Sized>(
&self,
ctx: &CodeGenContext<'ctx, '_>,
generator: &G,
) -> AnyTypeEnum<'ctx> {
self.0.strides().base_ptr(ctx, generator).get_type().get_element_type()
}
fn base_ptr<G: CodeGenerator + ?Sized>(
&self,
ctx: &CodeGenContext<'ctx, '_>,
_: &G,
) -> PointerValue<'ctx> {
self.0.strides_field(ctx).get(ctx, self.0.as_base_value(), self.0.name)
}
fn size<G: CodeGenerator + ?Sized>(
&self,
ctx: &CodeGenContext<'ctx, '_>,
_: &G,
) -> IntValue<'ctx> {
self.0.load_ndims(ctx)
}
}
impl<'ctx> ArrayLikeIndexer<'ctx, IntValue<'ctx>> for NDArrayStridesProxy<'ctx, '_> {
unsafe fn ptr_offset_unchecked<G: CodeGenerator + ?Sized>(
&self,
ctx: &mut CodeGenContext<'ctx, '_>,
generator: &mut G,
idx: &IntValue<'ctx>,
name: Option<&str>,
) -> PointerValue<'ctx> {
let var_name = name.map(|v| format!("{v}.addr")).unwrap_or_default();
unsafe {
ctx.builder
.build_in_bounds_gep(self.base_ptr(ctx, generator), &[*idx], var_name.as_str())
.unwrap()
}
}
fn ptr_offset<G: CodeGenerator + ?Sized>(
&self,
ctx: &mut CodeGenContext<'ctx, '_>,
generator: &mut G,
idx: &IntValue<'ctx>,
name: Option<&str>,
) -> PointerValue<'ctx> {
let size = self.size(ctx, generator);
let in_range = ctx.builder.build_int_compare(IntPredicate::ULT, *idx, size, "").unwrap();
ctx.make_assert(
generator,
in_range,
"0:IndexError",
"index {0} is out of bounds for axis 0 with size {1}",
[Some(*idx), Some(self.0.load_ndims(ctx)), None],
ctx.current_loc,
);
unsafe { self.ptr_offset_unchecked(ctx, generator, idx, name) }
}
}
impl<'ctx> UntypedArrayLikeAccessor<'ctx, IntValue<'ctx>> for NDArrayStridesProxy<'ctx, '_> {}
impl<'ctx> UntypedArrayLikeMutator<'ctx, IntValue<'ctx>> for NDArrayStridesProxy<'ctx, '_> {}
impl<'ctx> TypedArrayLikeAccessor<'ctx, IntValue<'ctx>> for NDArrayStridesProxy<'ctx, '_> {
fn downcast_to_type(
&self,
_: &mut CodeGenContext<'ctx, '_>,
value: BasicValueEnum<'ctx>,
) -> IntValue<'ctx> {
value.into_int_value()
}
}
impl<'ctx> TypedArrayLikeMutator<'ctx, IntValue<'ctx>> for NDArrayStridesProxy<'ctx, '_> {
fn upcast_from_type(
&self,
_: &mut CodeGenContext<'ctx, '_>,
value: IntValue<'ctx>,
) -> BasicValueEnum<'ctx> {
value.into()
}
}
/// Proxy type for accessing the `data` array of an `NDArray` instance in LLVM.
#[derive(Copy, Clone)]
pub struct NDArrayDataProxy<'ctx, 'a>(&'a NDArrayValue<'ctx>);
@ -291,7 +668,12 @@ impl<'ctx> ArrayLikeValue<'ctx> for NDArrayDataProxy<'ctx, '_> {
ctx: &CodeGenContext<'ctx, '_>,
generator: &G,
) -> IntValue<'ctx> {
call_ndarray_calc_size(generator, ctx, &self.as_slice_value(ctx, generator), (None, None))
irrt::ndarray::call_ndarray_calc_size(
generator,
ctx,
&self.as_slice_value(ctx, generator),
(None, None),
)
}
}
@ -400,7 +782,7 @@ impl<'ctx, Index: UntypedArrayLikeAccessor<'ctx>> ArrayLikeIndexer<'ctx, Index>
indices_elem_ty.get_bit_width()
);
let index = call_ndarray_flatten_index(generator, ctx, *self.0, indices);
let index = irrt::ndarray::call_ndarray_flatten_index(generator, ctx, *self.0, indices);
let sizeof_elem = ctx
.builder
.build_int_truncate_or_bit_cast(
@ -516,3 +898,36 @@ impl<'ctx, Index: UntypedArrayLikeAccessor<'ctx>> UntypedArrayLikeMutator<'ctx,
for NDArrayDataProxy<'ctx, '_>
{
}
/// A version of [`call_nac3_ndarray_set_strides_by_shape`] in Rust.
///
/// This function is used generating strides for globally defined contiguous ndarrays.
#[must_use]
pub fn make_contiguous_strides(itemsize: u64, ndims: u64, shape: &[u64]) -> Vec<u64> {
let mut strides = Vec::with_capacity(ndims as usize);
let mut stride_product = 1u64;
for i in 0..ndims {
let axis = ndims - i - 1;
strides[axis as usize] = stride_product * itemsize;
stride_product *= shape[axis as usize];
}
strides
}
/// A convenience enum for implementing functions that acts on scalars or ndarrays or both.
#[derive(Clone, Copy)]
pub enum ScalarOrNDArray<'ctx> {
Scalar(BasicValueEnum<'ctx>),
NDArray(NDArrayValue<'ctx>),
}
impl<'ctx> ScalarOrNDArray<'ctx> {
/// Get the underlying [`BasicValueEnum<'ctx>`] of this [`ScalarOrNDArray`].
#[must_use]
pub fn to_basic_value_enum(self) -> BasicValueEnum<'ctx> {
match self {
ScalarOrNDArray::Scalar(scalar) => scalar,
ScalarOrNDArray::NDArray(ndarray) => ndarray.as_base_value().into(),
}
}
}

View File

@ -0,0 +1,176 @@
use inkwell::{
types::{BasicType, IntType},
values::{BasicValueEnum, IntValue, PointerValue},
AddressSpace,
};
use super::{NDArrayValue, ProxyValue, TypedArrayLikeAccessor, TypedArrayLikeMutator};
use crate::codegen::{
irrt,
stmt::{gen_for_callback, BreakContinueHooks},
types::{ndarray::NDIterType, structure::StructField},
values::{ArraySliceValue, TypedArrayLikeAdapter},
CodeGenContext, CodeGenerator,
};
#[derive(Copy, Clone)]
pub struct NDIterValue<'ctx> {
value: PointerValue<'ctx>,
parent: NDArrayValue<'ctx>,
indices: ArraySliceValue<'ctx>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
}
impl<'ctx> NDIterValue<'ctx> {
/// Checks whether `value` is an instance of `NDArray`, returning [Err] if `value` is not an
/// instance.
pub fn is_representable(
value: PointerValue<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
<Self as ProxyValue>::Type::is_representable(value.get_type(), llvm_usize)
}
/// Creates an [`NDArrayValue`] from a [`PointerValue`].
#[must_use]
pub fn from_pointer_value(
ptr: PointerValue<'ctx>,
parent: NDArrayValue<'ctx>,
indices: ArraySliceValue<'ctx>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
) -> Self {
debug_assert!(Self::is_representable(ptr, llvm_usize).is_ok());
Self { value: ptr, parent, indices, llvm_usize, name }
}
/// Is the current iteration valid?
///
/// If true, then `element`, `indices` and `nth` contain details about the current element.
///
/// If `ndarray` is unsized, this returns true only for the first iteration.
/// If `ndarray` is 0-sized, this always returns false.
#[must_use]
pub fn has_element<G: CodeGenerator + ?Sized>(
&self,
generator: &G,
ctx: &CodeGenContext<'ctx, '_>,
) -> IntValue<'ctx> {
irrt::ndarray::call_nac3_nditer_has_element(generator, ctx, *self)
}
/// Go to the next element. If `has_element()` is false, then this has undefined behavior.
///
/// If `ndarray` is unsized, this can only be called once.
/// If `ndarray` is 0-sized, this can never be called.
pub fn next<G: CodeGenerator + ?Sized>(&self, generator: &G, ctx: &CodeGenContext<'ctx, '_>) {
irrt::ndarray::call_nac3_nditer_next(generator, ctx, *self);
}
fn element(&self, ctx: &CodeGenContext<'ctx, '_>) -> StructField<'ctx, PointerValue<'ctx>> {
self.get_type().get_fields(ctx.ctx).element
}
/// Get pointer to the current element.
#[must_use]
pub fn get_pointer(&self, ctx: &CodeGenContext<'ctx, '_>) -> PointerValue<'ctx> {
let elem_ty = self.parent.dtype;
let p = self.element(ctx).get(ctx, self.as_base_value(), None);
ctx.builder
.build_pointer_cast(p, elem_ty.ptr_type(AddressSpace::default()), "element")
.unwrap()
}
/// Get the value of the current element.
#[must_use]
pub fn get_scalar(&self, ctx: &CodeGenContext<'ctx, '_>) -> BasicValueEnum<'ctx> {
let p = self.get_pointer(ctx);
ctx.builder.build_load(p, "value").unwrap()
}
fn nth(&self, ctx: &CodeGenContext<'ctx, '_>) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields(ctx.ctx).nth
}
/// Get the index of the current element if this ndarray were a flat ndarray.
#[must_use]
pub fn get_index(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.nth(ctx).get(ctx, self.as_base_value(), None)
}
/// Get the indices of the current element.
#[must_use]
pub fn get_indices(
&'ctx self,
) -> impl TypedArrayLikeAccessor<'ctx, IntValue<'ctx>> + TypedArrayLikeMutator<'ctx, IntValue<'ctx>>
{
TypedArrayLikeAdapter::from(
self.indices,
Box::new(|ctx, val| {
ctx.builder
.build_int_z_extend_or_bit_cast(val.into_int_value(), self.llvm_usize, "")
.unwrap()
}),
Box::new(|_, val| val.into()),
)
}
}
impl<'ctx> ProxyValue<'ctx> for NDIterValue<'ctx> {
type Base = PointerValue<'ctx>;
type Type = NDIterType<'ctx>;
fn get_type(&self) -> Self::Type {
NDIterType::from_type(self.as_base_value().get_type(), self.llvm_usize)
}
fn as_base_value(&self) -> Self::Base {
self.value
}
}
impl<'ctx> From<NDIterValue<'ctx>> for PointerValue<'ctx> {
fn from(value: NDIterValue<'ctx>) -> Self {
value.as_base_value()
}
}
impl<'ctx> NDArrayValue<'ctx> {
/// Iterate through every element in the ndarray.
///
/// `body` has access to [`BreakContinueHooks`] to short-circuit and [`NDIterValue`] to
/// get properties of the current iteration (e.g., the current element, indices, etc.)
pub fn foreach<'a, G, F>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, 'a>,
body: F,
) -> Result<(), String>
where
G: CodeGenerator + ?Sized,
F: FnOnce(
&mut G,
&mut CodeGenContext<'ctx, 'a>,
BreakContinueHooks<'ctx>,
NDIterValue<'ctx>,
) -> Result<(), String>,
{
gen_for_callback(
generator,
ctx,
Some("ndarray_foreach"),
|generator, ctx| {
Ok(NDIterType::new(generator, ctx.ctx).construct(generator, ctx, *self))
},
|generator, ctx, nditer| Ok(nditer.has_element(generator, ctx)),
|generator, ctx, hooks, nditer| body(generator, ctx, hooks, nditer),
|generator, ctx, nditer| {
nditer.next(generator, ctx);
Ok(())
},
)
}
}

View File

@ -0,0 +1,36 @@
use std::iter::{once, repeat_n};
use itertools::Itertools;
use crate::codegen::{
values::ndarray::{NDArrayValue, RustNDIndex},
CodeGenContext, CodeGenerator,
};
impl<'ctx> NDArrayValue<'ctx> {
/// Make sure the ndarray is at least `ndmin`-dimensional.
///
/// If this ndarray's `ndims` is less than `ndmin`, a view is created on this with 1s prepended
/// to the shape. Otherwise, this function does nothing and return this ndarray.
#[must_use]
pub fn atleast_nd<G: CodeGenerator + ?Sized>(
&self,
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
ndmin: u64,
) -> Self {
assert!(self.ndims.is_some(), "NDArrayValue::atleast_nd is only supported for instances with compile-time known ndims (self.ndims = Some(...))");
let ndims = self.ndims.unwrap();
if ndims < ndmin {
// Extend the dimensions with np.newaxis.
let indices = repeat_n(RustNDIndex::NewAxis, (ndmin - ndims) as usize)
.chain(once(RustNDIndex::Ellipsis))
.collect_vec();
self.index(generator, ctx, &indices)
} else {
*self
}
}
}

View File

@ -0,0 +1,3 @@
pub use slice::*;
mod slice;

View File

@ -0,0 +1,231 @@
use inkwell::{
types::IntType,
values::{IntValue, PointerValue},
};
use nac3parser::ast::Expr;
use crate::{
codegen::{
types::{structure::StructField, utils::SliceType},
values::ProxyValue,
CodeGenContext, CodeGenerator,
},
typecheck::typedef::Type,
};
/// An IRRT representation of an (unresolved) slice.
#[derive(Copy, Clone)]
pub struct SliceValue<'ctx> {
value: PointerValue<'ctx>,
int_ty: IntType<'ctx>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
}
impl<'ctx> SliceValue<'ctx> {
/// Checks whether `value` is an instance of `ContiguousNDArray`, returning [Err] if `value` is
/// not an instance.
pub fn is_representable(
value: PointerValue<'ctx>,
llvm_usize: IntType<'ctx>,
) -> Result<(), String> {
<Self as ProxyValue<'ctx>>::Type::is_representable(value.get_type(), llvm_usize)
}
/// Creates an [`SliceValue`] from a [`PointerValue`].
#[must_use]
pub fn from_pointer_value(
ptr: PointerValue<'ctx>,
int_ty: IntType<'ctx>,
llvm_usize: IntType<'ctx>,
name: Option<&'ctx str>,
) -> Self {
debug_assert!(Self::is_representable(ptr, llvm_usize).is_ok());
Self { value: ptr, int_ty, llvm_usize, name }
}
fn start_defined_field(&self) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields().start_defined
}
pub fn load_start_defined(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.start_defined_field().get(ctx, self.value, self.name)
}
fn start_field(&self) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields().start
}
pub fn load_start(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.start_field().get(ctx, self.value, self.name)
}
pub fn store_start(&self, ctx: &CodeGenContext<'ctx, '_>, value: Option<IntValue<'ctx>>) {
match value {
Some(start) => {
self.start_defined_field().set(
ctx,
self.value,
ctx.ctx.bool_type().const_all_ones(),
self.name,
);
self.start_field().set(ctx, self.value, start, self.name);
}
None => self.start_defined_field().set(
ctx,
self.value,
ctx.ctx.bool_type().const_zero(),
self.name,
),
}
}
fn stop_defined_field(&self) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields().stop_defined
}
pub fn load_stop_defined(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.stop_defined_field().get(ctx, self.value, self.name)
}
fn stop_field(&self) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields().stop
}
pub fn load_stop(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.stop_field().get(ctx, self.value, self.name)
}
pub fn store_stop(&self, ctx: &CodeGenContext<'ctx, '_>, value: Option<IntValue<'ctx>>) {
match value {
Some(stop) => {
self.stop_defined_field().set(
ctx,
self.value,
ctx.ctx.bool_type().const_all_ones(),
self.name,
);
self.stop_field().set(ctx, self.value, stop, self.name);
}
None => self.stop_defined_field().set(
ctx,
self.value,
ctx.ctx.bool_type().const_zero(),
self.name,
),
}
}
fn step_defined_field(&self) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields().step_defined
}
pub fn load_step_defined(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.step_defined_field().get(ctx, self.value, self.name)
}
fn step_field(&self) -> StructField<'ctx, IntValue<'ctx>> {
self.get_type().get_fields().step
}
pub fn load_step(&self, ctx: &CodeGenContext<'ctx, '_>) -> IntValue<'ctx> {
self.step_field().get(ctx, self.value, self.name)
}
pub fn store_step(&self, ctx: &CodeGenContext<'ctx, '_>, value: Option<IntValue<'ctx>>) {
match value {
Some(step) => {
self.step_defined_field().set(
ctx,
self.value,
ctx.ctx.bool_type().const_all_ones(),
self.name,
);
self.step_field().set(ctx, self.value, step, self.name);
}
None => self.step_defined_field().set(
ctx,
self.value,
ctx.ctx.bool_type().const_zero(),
self.name,
),
}
}
}
impl<'ctx> ProxyValue<'ctx> for SliceValue<'ctx> {
type Base = PointerValue<'ctx>;
type Type = SliceType<'ctx>;
fn get_type(&self) -> Self::Type {
Self::Type::from_type(self.value.get_type(), self.int_ty, self.llvm_usize)
}
fn as_base_value(&self) -> Self::Base {
self.value
}
}
impl<'ctx> From<SliceValue<'ctx>> for PointerValue<'ctx> {
fn from(value: SliceValue<'ctx>) -> Self {
value.as_base_value()
}
}
/// A slice represented in compile-time by `start`, `stop` and `step`, all held as LLVM values.
// TODO: Rename this to CTConstSlice
#[derive(Debug, Copy, Clone)]
pub struct RustSlice<'ctx> {
int_ty: IntType<'ctx>,
start: Option<IntValue<'ctx>>,
stop: Option<IntValue<'ctx>>,
step: Option<IntValue<'ctx>>,
}
impl<'ctx> RustSlice<'ctx> {
/// Generate LLVM IR for an [`ExprKind::Slice`] and convert it into a [`RustSlice`].
#[allow(clippy::type_complexity)]
pub fn from_slice_expr<G: CodeGenerator>(
generator: &mut G,
ctx: &mut CodeGenContext<'ctx, '_>,
lower: &Option<Box<Expr<Option<Type>>>>,
upper: &Option<Box<Expr<Option<Type>>>>,
step: &Option<Box<Expr<Option<Type>>>>,
) -> Result<RustSlice<'ctx>, String> {
let mut value_mapper = |value_expr: &Option<Box<Expr<Option<Type>>>>| -> Result<_, String> {
Ok(match value_expr {
None => None,
Some(value_expr) => {
let value_expr = generator
.gen_expr(ctx, value_expr)?
.map(|value| {
value.to_basic_value_enum(ctx, generator, ctx.primitives.int32)
})
.unwrap()?;
Some(value_expr.into_int_value())
}
})
};
let start = value_mapper(lower)?;
let stop = value_mapper(upper)?;
let step = value_mapper(step)?;
Ok(RustSlice { int_ty: ctx.ctx.i32_type(), start, stop, step })
}
/// Write the contents to an LLVM [`SliceValue`].
pub fn write_to_slice(&self, ctx: &CodeGenContext<'ctx, '_>, dst_slice_ptr: SliceValue<'ctx>) {
assert_eq!(self.int_ty, dst_slice_ptr.int_ty);
dst_slice_ptr.store_start(ctx, self.start);
dst_slice_ptr.store_stop(ctx, self.stop);
dst_slice_ptr.store_step(ctx, self.step);
}
}

View File

@ -1,15 +1,15 @@
{ pkgs } : [
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libunwind-18.1.8-2-any.pkg.tar.zst";
sha256 = "0f9m76dx40iy794nfks0360gvjhdg6yngb2lyhwp4xd76rn5081m";
name = "mingw-w64-clang-x86_64-libunwind-18.1.8-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libunwind-19.1.4-1-any.pkg.tar.zst";
sha256 = "0frb5k16bbxdf8g379d16vl3qrh7n9pydn83gpfxpvwf3qlvnzyl";
name = "mingw-w64-clang-x86_64-libunwind-19.1.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libc++-18.1.8-2-any.pkg.tar.zst";
sha256 = "17savj9wys9my2ji7vyba7wwqkvzdjwnkb3k4858wxrjbzbfa6lk";
name = "mingw-w64-clang-x86_64-libc++-18.1.8-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libc++-19.1.4-1-any.pkg.tar.zst";
sha256 = "0wh5km0v8j50pqz9bxb4f0w7r8zhsvssrjvc94np53iq8wjagk86";
name = "mingw-w64-clang-x86_64-libc++-19.1.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -31,9 +31,9 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-xz-5.6.2-2-any.pkg.tar.zst";
sha256 = "0phb9hwqksk1rg29yhwlc7si78zav19c2kac0i841pc7mc2n9gzx";
name = "mingw-w64-clang-x86_64-xz-5.6.2-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-xz-5.6.3-3-any.pkg.tar.zst";
sha256 = "1a7gc462gnrjy5qb0zfkr9qm8bsnnf02y6wp3c59n618dhsq7rcf";
name = "mingw-w64-clang-x86_64-xz-5.6.3-3-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -43,9 +43,9 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libxml2-2.12.9-1-any.pkg.tar.zst";
sha256 = "0cjz2vj9yz6k5xj601cp0yk631rrr0z94ciamwqrvclb0yhakf25";
name = "mingw-w64-clang-x86_64-libxml2-2.12.9-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libxml2-2.12.9-2-any.pkg.tar.zst";
sha256 = "1b1r5llgqv88id8iwhqh23qwqmn5ic9hdamdc8xzij9hmcvdmmci";
name = "mingw-w64-clang-x86_64-libxml2-2.12.9-2-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -55,75 +55,87 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-llvm-libs-18.1.8-1-any.pkg.tar.zst";
sha256 = "0rpbgvvinsqflhd3nhfxk0g0yy8j80zzw5yx6573ak0m78a9fa06";
name = "mingw-w64-clang-x86_64-llvm-libs-18.1.8-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-llvm-libs-19.1.4-1-any.pkg.tar.zst";
sha256 = "1clrbm8dk893byj8s15pgcgqqijm2zkd10zgyakamd8m354kj9q4";
name = "mingw-w64-clang-x86_64-llvm-libs-19.1.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-llvm-18.1.8-1-any.pkg.tar.zst";
sha256 = "185g5h8q3x3rav9lp2njln58ny2idh2067fd02j3nsbik6glshpf";
name = "mingw-w64-clang-x86_64-llvm-18.1.8-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-llvm-19.1.4-1-any.pkg.tar.zst";
sha256 = "1iz2c9475h8p20ydpp0znbhyb62rlrk7wr7xl7cmwbam7wkwr8rn";
name = "mingw-w64-clang-x86_64-llvm-19.1.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-clang-libs-18.1.8-1-any.pkg.tar.zst";
sha256 = "089hji3yd7wsd03v9mdfgc99l5k1dql8kg7p3hy13vrbgfsabxhc";
name = "mingw-w64-clang-x86_64-clang-libs-18.1.8-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-clang-libs-19.1.4-1-any.pkg.tar.zst";
sha256 = "1hidciwlakxrp4kyb0j2v6g4lv76nn834g6b88w1j94fk3qc765d";
name = "mingw-w64-clang-x86_64-clang-libs-19.1.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-compiler-rt-18.1.8-1-any.pkg.tar.zst";
sha256 = "1dwcxnv1k5ljim5ys4h1c3jlrdpi0054z094ynav7if65i8zjj4a";
name = "mingw-w64-clang-x86_64-compiler-rt-18.1.8-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-compiler-rt-19.1.4-1-any.pkg.tar.zst";
sha256 = "1m1yhjkgzlbk10sv966qk4yji009ga0lr25gpgj2w7mcd2wixcr3";
name = "mingw-w64-clang-x86_64-compiler-rt-19.1.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-headers-git-12.0.0.r250.gc6bf4bdf6-1-any.pkg.tar.zst";
sha256 = "0163jzjlvq7inpafy3h48pkwag3ysk6x56xm84yfcz5q52fnfzq5";
name = "mingw-w64-clang-x86_64-headers-git-12.0.0.r250.gc6bf4bdf6-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-headers-git-12.0.0.r423.g8bcd5fc1a-1-any.pkg.tar.zst";
sha256 = "08gxc7h2achckknn6fz3p6yi7gxxvbaday8fpm4j56c4sa04n0df";
name = "mingw-w64-clang-x86_64-headers-git-12.0.0.r423.g8bcd5fc1a-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-crt-git-12.0.0.r250.gc6bf4bdf6-1-any.pkg.tar.zst";
sha256 = "00cn1mi29mfys7qy4hvgnjd0smqvnkdn3ibnrr6a3wy1h2vaykgq";
name = "mingw-w64-clang-x86_64-crt-git-12.0.0.r250.gc6bf4bdf6-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-crt-git-12.0.0.r423.g8bcd5fc1a-1-any.pkg.tar.zst";
sha256 = "0fxd1pb197ki0gzw6z8gmd6wgpd9d28js6cp5d31d55kw7d1vz13";
name = "mingw-w64-clang-x86_64-crt-git-12.0.0.r423.g8bcd5fc1a-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-lld-18.1.8-1-any.pkg.tar.zst";
sha256 = "1vpij5d06m4kjy3qv8bizwlkl21gcv6fv0r2f1j9bclgm6k3144x";
name = "mingw-w64-clang-x86_64-lld-18.1.8-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-lld-19.1.4-1-any.pkg.tar.zst";
sha256 = "1a8pjyhrzpc2z3784xxwix4i7yrz03ygnsk1wv9k0yq8m8wi9nbw";
name = "mingw-w64-clang-x86_64-lld-19.1.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libwinpthread-git-12.0.0.r250.gc6bf4bdf6-1-any.pkg.tar.zst";
sha256 = "1zkzqqd31xpkv817wja3qssjjx891bsdxw07037hv2sk0qr4ffn9";
name = "mingw-w64-clang-x86_64-libwinpthread-git-12.0.0.r250.gc6bf4bdf6-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libwinpthread-git-12.0.0.r423.g8bcd5fc1a-1-any.pkg.tar.zst";
sha256 = "140m312jx1sywqjkvfij69d268m4jpdmilq5bb8khkf0ayb16036";
name = "mingw-w64-clang-x86_64-libwinpthread-git-12.0.0.r423.g8bcd5fc1a-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-winpthreads-git-12.0.0.r250.gc6bf4bdf6-1-any.pkg.tar.zst";
sha256 = "02ynia88ad3l03r08nyldmnajwqkyxcjd191lyamkbj4d6zck323";
name = "mingw-w64-clang-x86_64-winpthreads-git-12.0.0.r250.gc6bf4bdf6-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-winpthreads-git-12.0.0.r423.g8bcd5fc1a-1-any.pkg.tar.zst";
sha256 = "017j4h511wg37bacym73f8g6s0jcfgzbzabzxpc6anr3gy4kkpbg";
name = "mingw-w64-clang-x86_64-winpthreads-git-12.0.0.r423.g8bcd5fc1a-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-clang-18.1.8-1-any.pkg.tar.zst";
sha256 = "1qny934nv4g75k9gb5sf31v24bgafkg6qw7r35xv3in491w6annq";
name = "mingw-w64-clang-x86_64-clang-18.1.8-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-clang-19.1.4-1-any.pkg.tar.zst";
sha256 = "11f4i4ai2bzvq6f06vxk1ymv7056c9707vdw489f1i2bdrf0c0ii";
name = "mingw-w64-clang-x86_64-clang-19.1.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-c-ares-1.33.1-1-any.pkg.tar.zst";
sha256 = "14r6jjsvfbapbkv2zqp2yglva4vz4srzkgk7f186ri3kcafjspgq";
name = "mingw-w64-clang-x86_64-c-ares-1.33.1-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-rust-1.83.0-3-any.pkg.tar.zst";
sha256 = "0nxs571vb4f1i5vp91134p5blns9ml2r25nx6kdlg0zhd5x85kvm";
name = "mingw-w64-clang-x86_64-rust-1.83.0-3-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-brotli-1.1.0-2-any.pkg.tar.zst";
sha256 = "1q01lz9lcyrjmkhv9rddgjazmk7warlcmwhc4qkq9y6h0yfsb71n";
name = "mingw-w64-clang-x86_64-brotli-1.1.0-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-cppdap-1.65-1-any.pkg.tar.zst";
sha256 = "0phhwkcqp30dsyj5vr6w99sgm1jfm5rzg0w5x5mv9md4x7lm9lmh";
name = "mingw-w64-clang-x86_64-cppdap-1.65-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-c-ares-1.34.3-1-any.pkg.tar.zst";
sha256 = "1mpn397qsdz3l2fav6ymwjlj96ialn9m8sldii3ymbcyhranl3xx";
name = "mingw-w64-clang-x86_64-c-ares-1.34.3-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-brotli-1.1.0-4-any.pkg.tar.zst";
sha256 = "0hx9gjzibacfx3fzk11n2vzz2pmnb956babh2ig8avx3hk7vlqrg";
name = "mingw-w64-clang-x86_64-brotli-1.1.0-4-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -139,9 +151,9 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libpsl-0.21.5-2-any.pkg.tar.zst";
sha256 = "1mpx77q5g8pj45s8wgc52c4ww2r93080p6d559p56f558a3cl317";
name = "mingw-w64-clang-x86_64-libpsl-0.21.5-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libpsl-0.21.5-3-any.pkg.tar.zst";
sha256 = "0hb7wgdliic3d7fa0cvr5pj946pmwfc0apmyb0yfb5d0hc1afwsc";
name = "mingw-w64-clang-x86_64-libpsl-0.21.5-3-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -163,57 +175,45 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-openssl-3.3.2-1-any.pkg.tar.zst";
sha256 = "1djgpcz447yvhdy1yq5wh8l5d0821izxklx9afyszbw0pbr7f24y";
name = "mingw-w64-clang-x86_64-openssl-3.3.2-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-openssl-3.4.0-1-any.pkg.tar.zst";
sha256 = "0cgiqjmgdnwnv9r88z634dmqrzh06dmsfncyzymw0s16nnv2k7k2";
name = "mingw-w64-clang-x86_64-openssl-3.4.0-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libssh2-1.11.0-2-any.pkg.tar.zst";
sha256 = "0l2m823gm1rvnjmqm5ads17mxz1bhpzai5ixyhnkpzrsjxd1ygy5";
name = "mingw-w64-clang-x86_64-libssh2-1.11.0-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libssh2-1.11.1-1-any.pkg.tar.zst";
sha256 = "01l23cn5brficjzba7ldscqkdvk4rdcvvdyybd90qr2hqzligmhn";
name = "mingw-w64-clang-x86_64-libssh2-1.11.1-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-nghttp2-1.63.0-1-any.pkg.tar.zst";
sha256 = "0lfqrlmapsc7ilxjmhr7hxi578vclqlhpqimbvzq0c70c0iwk864";
name = "mingw-w64-clang-x86_64-nghttp2-1.63.0-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-nghttp2-1.64.0-1-any.pkg.tar.zst";
sha256 = "1hv8fp496l018s5dx5v8nvxc0a6rswskwk1jsrfd94rh3kbq2ilc";
name = "mingw-w64-clang-x86_64-nghttp2-1.64.0-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-nghttp3-1.5.0-1-any.pkg.tar.zst";
sha256 = "1ljl9kdasf91bxkqcmbbjchp5g00ahv8jn2zab38899z6j3x43nz";
name = "mingw-w64-clang-x86_64-nghttp3-1.5.0-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-nghttp3-1.6.0-1-any.pkg.tar.zst";
sha256 = "1p7q47fin12vzyf126v1azbbpgpa0y6ighfh6mbfdb6zcyq74kbd";
name = "mingw-w64-clang-x86_64-nghttp3-1.6.0-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-curl-8.9.1-2-any.pkg.tar.zst";
sha256 = "1zr6kgqp9i4qqrfckh3kfmz4x1cwv4xis9sfqsx7xji88priax64";
name = "mingw-w64-clang-x86_64-curl-8.9.1-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-curl-8.11.1-1-any.pkg.tar.zst";
sha256 = "16yvyqjzxyzawgv26r1g145wphvhjil2b0pyhy4nj7v5d19n6wvh";
name = "mingw-w64-clang-x86_64-curl-8.11.1-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-rust-1.80.1-1-any.pkg.tar.zst";
sha256 = "1dm4vlrfi9m6xl09zpn0yjr7qcjjr4x738z1rjfwysfnc0awq4x8";
name = "mingw-w64-clang-x86_64-rust-1.80.1-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-expat-2.6.4-1-any.pkg.tar.zst";
sha256 = "03fp2yacv7gk0g049lffz6pbj93vpjmzqxxa312d4gxczi57nqdv";
name = "mingw-w64-clang-x86_64-expat-2.6.4-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-cppdap-1.65-1-any.pkg.tar.zst";
sha256 = "0phhwkcqp30dsyj5vr6w99sgm1jfm5rzg0w5x5mv9md4x7lm9lmh";
name = "mingw-w64-clang-x86_64-cppdap-1.65-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-expat-2.6.3-1-any.pkg.tar.zst";
sha256 = "19xfl1q78q1k8j0lr5aspcf668pmfg01fgib73zq7ff7y5y5fcyi";
name = "mingw-w64-clang-x86_64-expat-2.6.3-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-jsoncpp-1.9.5-3-any.pkg.tar.zst";
sha256 = "1a8mdn4ram9pgqpx5fwxmhcmzc6bh1fq1s4m37xh0d8p6fpncv10";
name = "mingw-w64-clang-x86_64-jsoncpp-1.9.5-3-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-jsoncpp-1.9.6-3-any.pkg.tar.zst";
sha256 = "1ipilhiza17vz5dhgi61l80w2klw9f21w6jbyhi9wmfd6nxqv13c";
name = "mingw-w64-clang-x86_64-jsoncpp-1.9.6-3-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -235,27 +235,27 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libtre-git-r177.07e66d0-2-any.pkg.tar.zst";
sha256 = "0fc9hxsdks1xy5fv0rcna433hlzf6jhs77hg0hfzkzhn06f9alp4";
name = "mingw-w64-clang-x86_64-libtre-git-r177.07e66d0-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libtre-0.9.0-1-any.pkg.tar.zst";
sha256 = "1y5d2cbkpd0ngqlsv6hvz47nxb1wry33i5s1clg2k53yhgjkg8vv";
name = "mingw-w64-clang-x86_64-libtre-0.9.0-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libsystre-1.0.1-5-any.pkg.tar.zst";
sha256 = "05qsn8fkks4f93jkas43s47axqqgx5m64b45p462si3nlb8cjirq";
name = "mingw-w64-clang-x86_64-libsystre-1.0.1-5-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libsystre-1.0.1-6-any.pkg.tar.zst";
sha256 = "19c71fs5gqjrf88mv7l702fjg228xd9lfbxg0mkzm3ljvv4ljn0q";
name = "mingw-w64-clang-x86_64-libsystre-1.0.1-6-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libarchive-3.7.4-1-any.pkg.tar.zst";
sha256 = "1ykw6imllgxv6lsgwxx1miqjr4l1iryqkrj286jcbfrb8ghpzhv5";
name = "mingw-w64-clang-x86_64-libarchive-3.7.4-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libarchive-3.7.7-1-any.pkg.tar.zst";
sha256 = "01glychb0k1yd878aq4y2fn08lqh2bjydh90xmq03z5qhig66mmn";
name = "mingw-w64-clang-x86_64-libarchive-3.7.7-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libuv-1.48.0-1-any.pkg.tar.zst";
sha256 = "0kfzanvx7hg7bvy35h2z2vcfxvwn44sikd36mvzhkv6c3c6y84sn";
name = "mingw-w64-clang-x86_64-libuv-1.48.0-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-libuv-1.49.2-1-any.pkg.tar.zst";
sha256 = "1b9slshbcprxjaj2qqypaywr0f2pgajg1bgspjk83hk65sx6sklb";
name = "mingw-w64-clang-x86_64-libuv-1.49.2-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -277,9 +277,9 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-cmake-3.30.3-1-any.pkg.tar.zst";
sha256 = "0fjwf6xxzli6rcsbzr1razldmm538ibkyf5kw132lpaz5wma9bj8";
name = "mingw-w64-clang-x86_64-cmake-3.30.3-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-cmake-3.31.2-3-any.pkg.tar.zst";
sha256 = "139f91r392c68hsajm0c81690pmzkywb0p4x8ms8ms53ncxnz6gz";
name = "mingw-w64-clang-x86_64-cmake-3.31.2-3-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -289,9 +289,9 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-ncurses-6.4.20231217-1-any.pkg.tar.zst";
sha256 = "00046d52zsr8zjifl7h22jfihhh53h20ipvbqmvf9myssw2fwjza";
name = "mingw-w64-clang-x86_64-ncurses-6.4.20231217-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-ncurses-6.5.20240831-1-any.pkg.tar.zst";
sha256 = "1hlfj9g4s767s502sawwbcv4a0xd3ym3ip4jswmhq48wh5050iyb";
name = "mingw-w64-clang-x86_64-ncurses-6.5.20240831-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -307,15 +307,15 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-tcl-8.6.13-1-any.pkg.tar.zst";
sha256 = "0paaqwk0sfy2zxwlxkmxf2bqq46lyg0sx7cqgzknvazwx8xa2z4x";
name = "mingw-w64-clang-x86_64-tcl-8.6.13-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-sqlite3-3.47.2-1-any.pkg.tar.zst";
sha256 = "10pavblv9yjirlm5hix9aikpswhiamry097clba6jcvsajlx4azy";
name = "mingw-w64-clang-x86_64-sqlite3-3.47.2-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-sqlite3-3.46.1-1-any.pkg.tar.zst";
sha256 = "1axplxyjnaz411qzjjqwbj55fbrh4akq3plm2p1sx64jp844xpyq";
name = "mingw-w64-clang-x86_64-sqlite3-3.46.1-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-tcl-8.6.13-1-any.pkg.tar.zst";
sha256 = "0paaqwk0sfy2zxwlxkmxf2bqq46lyg0sx7cqgzknvazwx8xa2z4x";
name = "mingw-w64-clang-x86_64-tcl-8.6.13-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
@ -325,44 +325,38 @@
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-tzdata-2024a-1-any.pkg.tar.zst";
sha256 = "1lsfn3759cyf56zlmfvgy6ihs4iks6zhlnrbfmnq5wml02k936ji";
name = "mingw-w64-clang-x86_64-tzdata-2024a-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-tzdata-2024b-1-any.pkg.tar.zst";
sha256 = "0jihnr1i7vyzczxz60ds1x3gcm3p4ad2pq9d5vvpwjdwrxkvxmkc";
name = "mingw-w64-clang-x86_64-tzdata-2024b-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-python3.12-3.12.1-2-any.pkg.tar.zst";
sha256 = "0wmd39wl9z237w093a7c6hl5pclca9yvwxn0kiw6i2njk3sjv51a";
name = "mingw-w64-clang-x86_64-python3.12-3.12.1-2-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-python-3.12.7-3-any.pkg.tar.zst";
sha256 = "1v15j2pzy9wj4n1rjngdi2hf8h0l9z4lri3xb86yvdv1xl2msj6h";
name = "mingw-w64-clang-x86_64-python-3.12.7-3-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-python-3.11.9-1-any.pkg.tar.zst";
sha256 = "0ah1idjqxg7jc07a1gz9z766rjjd0f0c6ri4hpcsimsrbj1zjd3c";
name = "mingw-w64-clang-x86_64-python-3.11.9-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-llvm-openmp-19.1.4-2-any.pkg.tar.zst";
sha256 = "1pn1fbj74rx837s9z8gqs4b0cr7kqi5m1m2mi9ibjpw64m1aqwxv";
name = "mingw-w64-clang-x86_64-llvm-openmp-19.1.4-2-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-llvm-openmp-18.1.8-1-any.pkg.tar.zst";
sha256 = "0cy2v0l4af24j34mzj5q5nlzcqhackfajlfj1rpf6mb3rbz23qw9";
name = "mingw-w64-clang-x86_64-llvm-openmp-18.1.8-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-openblas-0.3.28-2-any.pkg.tar.zst";
sha256 = "18p1zhf7h3k3phf3bl483jg3k7y9zq375z6ww75g62158ic9lfyc";
name = "mingw-w64-clang-x86_64-openblas-0.3.28-2-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-openblas-0.3.28-1-any.pkg.tar.zst";
sha256 = "1pskcqc1lg9p8m8rk7bw3mz7mn7vw5fpl7zxa23bhjn02p5b79qq";
name = "mingw-w64-clang-x86_64-openblas-0.3.28-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-python-numpy-2.1.1-2-any.pkg.tar.zst";
sha256 = "1kiy7ail04ias47xbbhl9vpsz02g0g3f29ncgx5gcks9vgqldp6m";
name = "mingw-w64-clang-x86_64-python-numpy-2.1.1-2-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-python-numpy-2.0.1-1-any.pkg.tar.zst";
sha256 = "0ks6q8v58h4wmr2pzsjl2xm4f63g0psvfm0jwlz24mqfxp8gqfcc";
name = "mingw-w64-clang-x86_64-python-numpy-2.0.1-1-any.pkg.tar.zst";
})
(pkgs.fetchurl {
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-python-setuptools-74.0.0-1-any.pkg.tar.zst";
sha256 = "0xc95z5jzzjf5lw35bs4yn5rlwkrkmrh78yi5rranqpz5nn0wsa0";
name = "mingw-w64-clang-x86_64-python-setuptools-74.0.0-1-any.pkg.tar.zst";
url = "https://mirror.msys2.org/mingw/clang64/mingw-w64-clang-x86_64-python-setuptools-75.6.0-1-any.pkg.tar.zst";
sha256 = "03l04kjmy5p9whaw0h619gdg7yw1gxbz8phifq4pzh3c1wlw7yfd";
name = "mingw-w64-clang-x86_64-python-setuptools-75.6.0-1-any.pkg.tar.zst";
})
]