forked from M-Labs/nac3
1
0
Fork 0

core: irrt add unchecked ndarray slicing

This commit is contained in:
lyken 2024-07-15 00:43:12 +08:00
parent b8c0d5836f
commit cc8103152f
6 changed files with 299 additions and 7 deletions

View File

@ -0,0 +1,137 @@
#pragma once
#include <irrt/slice.hpp>
#include <irrt/numpy/ndarray_def.hpp>
#include <irrt/numpy/ndarray_basic.hpp>
namespace {
typedef uint8_t NDSubscriptType;
extern "C" {
const NDSubscriptType INPUT_SUBSCRIPT_TYPE_INDEX = 0;
const NDSubscriptType INPUT_SUBSCRIPT_TYPE_SLICE = 1;
}
struct NDSubscript {
// A poor-man's enum variant type
NDSubscriptType type;
/*
if type == INPUT_SUBSCRIPT_TYPE_INDEX => `slice` points to a single `SizeT`
if type == INPUT_SUBSCRIPT_TYPE_SLICE => `slice` points to a single `UserRange<SizeT>`
`SizeT` is controlled by the caller: `NDSubscript` only cares about where that
slice is (the pointer), `NDSubscript` does not care/know about the actual `sizeof()`
of the slice value.
*/
uint8_t* data;
};
namespace ndarray {
namespace util {
template<typename SizeT>
SizeT deduce_ndims_after_slicing(SizeT ndims, SizeT num_subscripts, const NDSubscript* subscripts) {
irrt_assert(num_subscripts <= ndims);
SizeT final_ndims = ndims;
for (SizeT i = 0; i < num_subscripts; i++) {
if (subscripts[i].type == INPUT_SUBSCRIPT_TYPE_INDEX) {
final_ndims--; // An index demotes the rank by 1
}
}
return final_ndims;
}
}
// To support numpy "basic indexing" https://numpy.org/doc/stable/user/basics.indexing.html#basic-indexing
// "Advanced indexing" https://numpy.org/doc/stable/user/basics.indexing.html#advanced-indexing is not supported
//
// This function supports:
// - "scalar indexing",
// - "slicing and strides",
// - and "dimensional indexing tools" (TODO, but this is really easy to implement).
//
// Things assumed by this function:
// - `dst_ndarray` is allocated by the caller
// - `dst_ndarray.ndims` has the correct value (according to `ndarray::util::deduce_ndims_after_slicing`).
// - ... and `dst_ndarray.shape` and `dst_ndarray.strides` have been allocated by the caller as well
//
// Other notes:
// - `dst_ndarray->data` does not have to be set, it will be derived.
// - `dst_ndarray->itemsize` does not have to be set, it will be set to `src_ndarray->itemsize`
// - `dst_ndarray->shape` and `dst_ndarray.strides` can contain empty values
template <typename SizeT>
void subscript(SizeT num_subscripts, NDSubscript* subscripts, NDArray<SizeT>* src_ndarray, NDArray<SizeT>* dst_ndarray) {
// REFERENCE CODE (check out `_index_helper` in `__getitem__`):
// https://github.com/wadetb/tinynumpy/blob/0d23d22e07062ffab2afa287374c7b366eebdda1/tinynumpy/tinynumpy.py#L652
// irrt_assert(dst_ndarray->ndims == ndarray::util::deduce_ndims_after_slicing(src_ndarray->ndims, num_subscripts, subscripts));
dst_ndarray->data = src_ndarray->data;
dst_ndarray->itemsize = src_ndarray->itemsize;
SizeT src_axis = 0;
SizeT dst_axis = 0;
for (SizeT i = 0; i < num_subscripts; i++) {
NDSubscript *ndsubscript = &subscripts[i];
if (ndsubscript->type == INPUT_SUBSCRIPT_TYPE_INDEX) {
// Handle when the ndsubscript is just a single (possibly negative) integer
// e.g., `my_array[::2, -5, ::-1]`
// ^^------ like this
SizeT index_user = *((SizeT*) ndsubscript->data);
SizeT index = slice::resolve_index_in_length(src_ndarray->shape[src_axis], index_user);
dst_ndarray->data += index * src_ndarray->strides[src_axis]; // Add offset
// Next
src_axis++;
} else if (ndsubscript->type == INPUT_SUBSCRIPT_TYPE_SLICE) {
// Handle when the ndsubscript is a slice (represented by UserSlice in IRRT)
// e.g., `my_array[::2, -5, ::-1]`
// ^^^------^^^^----- like these
UserSlice* user_slice = (UserSlice*) ndsubscript->data;
// TODO: use checked indices
Slice slice;
user_slice->indices(src_ndarray->shape[src_axis], &slice); // To resolve negative indices and other funny stuff written by the user
// NOTE: There is no need to write special code to handle negative steps/strides.
// This simple implementation meticulously handles both positive and negative steps/strides.
// Check out the tinynumpy and IRRT's test cases if you are not convinced.
dst_ndarray->data += (SizeT) slice.start * src_ndarray->strides[src_axis]; // Add offset (NOTE: no need to `* itemsize`, strides count in # of bytes)
dst_ndarray->strides[dst_axis] = ((SizeT) slice.step) * src_ndarray->strides[src_axis]; // Determine stride
dst_ndarray->shape[dst_axis] = (SizeT) slice.len(); // Determine shape dimension
// Next
dst_axis++;
src_axis++;
} else {
__builtin_unreachable();
}
}
/*
Reference python code:
```python
dst_ndarray.shape.extend(src_ndarray.shape[src_axis:])
dst_ndarray.strides.extend(src_ndarray.strides[src_axis:])
```
*/
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];
}
}
}
}
extern "C" {
void __nac3_ndarray_subscript(int32_t num_subscripts, NDSubscript* subscripts, NDArray<int32_t>* src_ndarray, NDArray<int32_t> *dst_ndarray) {
ndarray::subscript(num_subscripts, subscripts, src_ndarray, dst_ndarray);
}
void __nac3_ndarray_subscript64(int64_t num_subscripts, NDSubscript* subscripts, NDArray<int64_t>* src_ndarray, NDArray<int64_t> *dst_ndarray) {
ndarray::subscript(num_subscripts, subscripts, src_ndarray, dst_ndarray);
}
}

View File

@ -0,0 +1,132 @@
#pragma once
#include <irrt/int_defs.hpp>
#include <irrt/slice.hpp>
namespace {
struct Slice {
SliceIndex start;
SliceIndex stop;
SliceIndex step;
// The length/The number of elements of the slice if it were a range,
// i.e., the value of `len(range(this->start, this->stop, this->end))`
SliceIndex len() {
SliceIndex diff = stop - 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 slice {
// "Resolve" an index value under a length in Python lists.
// If you have a `list` of length 100, `list[-1]` would resolve to `list[100-1] == list[99]`.
//
// If length == 0, this function returns 0
//
// If index is out of bounds, this function clamps the value
// (to `list[0]` or `list[-1]` in the context of a list and depending on if index is + or -)
SliceIndex resolve_index_in_length(SliceIndex length, SliceIndex index) {
if (index < 0) {
// Remember that index is negative, so do a plus here
return max<SliceIndex>(length + index, 0);
} else {
return min<SliceIndex>(length, index);
}
}
}
// A user-written Python-like slice.
//
// i.e., this slice is a triple of either an int or nothing. (e.g., `my_array[:10:2]`, `start` is None)
//
// You can "resolve" a `UserSlice` by using `UserSlice::indices(<length>)`
struct UserSlice {
// Did the user specify `start`? If 0, `start` is undefined (and contains an empty value)
bool start_defined;
SliceIndex start;
// Similar to `start_defined`
bool stop_defined;
SliceIndex stop;
// Similar to `start_defined`
bool step_defined;
SliceIndex step;
// Constructor faithfully follows Python's `slice()`.
explicit UserSlice(SliceIndex stop) {
start_defined = false;
stop_defined = true;
step_defined = false;
this->stop = stop;
}
explicit UserSlice(SliceIndex start, SliceIndex stop) {
start_defined = true;
stop_defined = true;
step_defined = false;
this->start = start;
this->stop = stop;
}
explicit UserSlice(SliceIndex start, SliceIndex stop, SliceIndex step) {
start_defined = true;
stop_defined = true;
step_defined = true;
this->start = start;
this->stop = stop;
this->step = step;
}
// Like Python's `slice(start, stop, step).indices(length)`
void indices(SliceIndex length, Slice* result) {
// NOTE: This function implements Python's `slice.indices` *FAITHFULLY*.
// SEE: https://github.com/python/cpython/blob/f62161837e68c1c77961435f1b954412dd5c2b65/Objects/sliceobject.c#L546
result->step = step_defined ? step : 1;
bool step_is_negative = result->step < 0;
if (start_defined) {
result->start = slice::resolve_index_in_length(length, start);
} else {
result->start = step_is_negative ? length - 1 : 0;
}
if (stop_defined) {
result->stop = slice::resolve_index_in_length(length, stop);
} else {
result->stop = step_is_negative ? -1 : length;
}
}
// `indices()` but asserts `this->step != 0` and `this->length >= 0`
void checked_indices(ErrorContext* errctx, SliceIndex length, Slice* result) {
if (!(length >= 0)) {
errctx->set_error(
errctx->error_ids->value_error,
"length should not be negative, got {0}", // Edited. Error message copied from python by doing `slice(0, 0, 0).indices(100)`
length
);
return;
}
if (!(this->step_defined && this->step != 0)) {
// Error message
errctx->set_error(
errctx->error_ids->value_error,
"slice step cannot be zero" // Error message copied from python by doing `slice(0, 0, 0).indices(100)`
);
return;
}
this->indices(length, result);
}
};
}

View File

@ -3,8 +3,10 @@
#include <irrt/core.hpp>
#include <irrt/error_context.hpp>
#include <irrt/int_defs.hpp>
#include <irrt/numpy/ndarray_def.hpp>
#include <irrt/numpy/ndarray_basic.hpp>
#include <irrt/numpy/ndarray_broadcast.hpp>
#include <irrt/numpy/ndarray_def.hpp>
#include <irrt/numpy/ndarray_fill.hpp>
#include <irrt/numpy/ndarray_subscript.hpp>
#include <irrt/slice.hpp>
#include <irrt/utils.hpp>

View File

@ -8,11 +8,13 @@
#include <irrt_everything.hpp>
#include <test/core.hpp>
#include <test/ndarray.hpp>
#include <test/test_core.hpp>
#include <test/test_ndarray.hpp>
#include <test/test_slice.hpp>
int main() {
test_int_exp();
run_all_tests_ndarray();
run_all_tests_ndarray_slice();
return 0;
}

View File

@ -1,15 +1,14 @@
#pragma once
#include <test/core.hpp>
#include <irrt/numpy/ndarray.hpp>
#include <irrt/numpy/ndarray_util.hpp>
#include <irrt_everything.hpp>
void test_calc_size_from_shape_normal() {
// Test shapes with normal values
BEGIN_TEST();
int32_t shape[4] = { 2, 3, 5, 7 };
assert_values_match(210, ndarray_util::calc_size_from_shape<int32_t>(4, shape));
assert_values_match(210, ndarray::util::calc_size_from_shape<int32_t>(4, shape));
}
void test_calc_size_from_shape_has_zero() {
@ -17,7 +16,7 @@ void test_calc_size_from_shape_has_zero() {
BEGIN_TEST();
int32_t shape[4] = { 2, 0, 5, 7 };
assert_values_match(0, ndarray_util::calc_size_from_shape<int32_t>(4, shape));
assert_values_match(0, ndarray::util::calc_size_from_shape<int32_t>(4, shape));
}
void test_set_strides_by_shape() {
@ -26,7 +25,7 @@ void test_set_strides_by_shape() {
int32_t shape[4] = { 99, 3, 5, 7 };
int32_t strides[4] = { 0 };
ndarray_util::set_strides_by_shape((int32_t) sizeof(int32_t), 4, strides, shape);
ndarray::util::set_strides_by_shape((int32_t) sizeof(int32_t), 4, strides, shape);
int32_t expected_strides[4] = {
105 * sizeof(int32_t),

View File

@ -0,0 +1,20 @@
#pragma once
#include <test/core.hpp>
#include <irrt_everything.hpp>
void test_slice_1() {
BEGIN_TEST();
UserSlice user_slice(5);
Slice slice;
user_slice.indices(100, &slice);
assert_values_match(0, slice.start);
assert_values_match(5, slice.stop);
assert_values_match(1, slice.step);
}
void run_all_tests_ndarray_slice() {
test_slice_1();
}